OR I G I N A L R E S E A RCH

Part of a Special Issue on Biobased Value Chains for a Growing

Bioeconomy

Characterization of phenology physiology morphology andbiomass traits across a broad Euro‐Mediterranean ecotypicpanel of the lignocellulosic feedstock Arundo donax

Francesco Fabbrini1 | Riccardo Ludovisi1 | Omar Alasia2 | Jaume Flexas3 |

Cyril Douthe3 | Miquel Ribas Carboacute3 | Paul Robson4 | Gail Taylor56 |

Giuseppe Scarascia‐Mugnozza1 | Joost J B Keurentjes7 | Antoine Harfouche1

1Department for Innovation in BiologicalAgro‐Food and Forest SystemsUniversity of Tuscia Viterbo Italy2Alasia Franco Vivai Savigliano Italy3Research Group on Plant Biology underMediterranean Conditions Departamentde Biologia Universitat de les IllesBalears Palma de Mallorca Spain4Institute of Biological Environmental andRural Sciences (IBERS) AberystwythUniversity Aberystwyth UK5Center for Biological Sciences Universityof Southampton Southampton UK6Department of Plant Sciences Universityof California Davis California7Laboratory of GeneticsWageningenUniversity amp ResearchWageningenTheNetherlands

CorrespondenceAntoine Harfouche Department forInnovation in Biological Agro‐food andForest Systems University of TusciaViterbo ItalyEmail aharfoucheunitusit

Funding informationEuropean Commissions SeventhFramework Programme GrantAwardNumber FP72012‐2017 FP7 FoodAgriculture and Fisheries BiotechnologyGrantAward Number FP7‐311929 BrainGain Program (Rientro dei cervelli) of theItalian Ministry of Education Universityand Research

AbstractGiant reed (Arundo donax L) is a perennial rhizomatous grass which has

attracted great attention as a potential lignocellulosic feedstock for bioethanol pro-

duction due to high biomass yield in marginal land areas high polysaccharide

content and low inhibitor levels in microbial fermentations However little is

known about the trait variation that is available across a broad ecotypic panel of

A donax nor the traits that contribute most significantly to yield and growth in

drought prone environments A collection of 82 ecotypes of A donax sampled

across the Mediterranean basin was planted in a common garden experimental

field in Savigliano Italy We analysed the collection using 367 clumps represent-

ing replicate plantings of 82 ecotypes for variation in 21 traits important for bio-

mass accumulation and to identify the particular set of ecotypes with the most

promising potential for biomass production We measured morpho‐physiologicalphenological and biomass traits and analysed causal relationships between traits

and productivity characteristics assessed at leaf and canopy levels The results

identified differences among the 82 ecotypes for all studied traits those showing

the highest level of variability included stomatal resistance stem density (StN)

stem dry mass (StDM) and total biomass production (TotDM) Multiple regression

analysis revealed that leaf area index StDM StN number of nodes per stem

stem height and diameter were the most significant predictors of TotDM and the

most important early selection criteria for bioenergy production from A donax

These traits were used in a hierarchical cluster analysis to identify groups of simi-

lar ecotypes and a selection was made of promising ecotypes for multiyear and

multisite testing for biomass production Heritability estimates were significant for

all traits The potential of this ecotype collection as a resource for studies of

germplasm diversity and for the analysis of traits underpinning high productivity

of A donax is highlighted

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -This is an open access article under the terms of the Creative Commons Attribution License which permits use distribution and reproduction in any medium providedtheoriginal work is properly citedcopy 2018 The Authors GCB Bioenergy Published by John Wiley amp Sons Ltd

Received 30 April 2018 | Accepted 18 July 2018

DOI 101111gcbb12555

GCB Bioenergy 20181ndash19 wileyonlinelibrarycomjournalgcbb | 1

KEYWORD S

Arundo donax biomass ecotype variability growth traits lignocellulosic biomass multivariate

analysis perennial grasses phenology physiology

1 | INTRODUCTION

Giant reed (Arundo donax L) is a perennial rhizomatousgrass which has attracted great attention as a potential ligno-cellulosic biomass due to rapid growth high apparent pro-ductivity and adaptability to marginal land conditionscombined with low input requirements (Angelini Ceccariniamp Bonari 2005 Angelini Ceccarini Nassi o Di Nasso ampBonari 2009 Corno Pilu amp Adani 2014 LewandowskiScurlock Lindvall amp Christou 2003 Pilu Bucci Badoneamp Landoni 2012) It is distributed throughout the Mediter-ranean climate zones and has been cultivated mostly in Asiaand southern Europe After establishment and from the sec-ond year on the above‐ground biomass can be harvestedannually for almost 15ndash30 years (Ceotto et al 2013 PiluManca amp Landoni 2013) A donax lignocellulosic biomassfeedstock can be readily used as solid biofuel in direct com-bustion for heat generation or for bioethanol productionfrom alcoholic fermentation of lignocellulose material (Piluet al 2012 Silva Schirmer Maeda Barcelos amp Pereira2015) Based on a comprehensive utilization indexesobtained from an evaluation system that takes into considera-tion numerous factors deemed essential for the conversion oflignocellulosic biomass to bioethanol a recent study revealedthat A donax is a highly promising lignocellulosic feedstockfor producing bioethanol compared to corn stalks switch-grass pennisetum and silvergrass (Kou Song Zhang ampTan 2017) Moreover A donax has been used as a potentialfeedstock for thermo‐chemical conversion to biochar bio‐oiland gas for energetic purposes (Saikia Chutia Kataki ampPant 2015) A donax cropping system is important not onlybecause it provides farmers with an innovative opportunitybut also for mitigating greenhouse gas (GHG) emissionscontrolling soil erosion and phytoremediating contaminatedsoils (Cosentino Copani Scalici Scordia amp Testa 2015Fagnano Impagliazzo Mori amp Fiorentino 2015 Nsangan-wimana Marchand Douay amp Mench 2014 Zucaro et al2015) Furthermore A donax is a resilient and very produc-tive species It has been shown that the photosynthetic capac-ity of A donax in full sunlight is high compared to other C3species and comparable to C4 bioenergy grasses (RossaTuumlffers Naidoo amp von Willert 1998) Because of theseadvantages A donax is expected to play a major role in theprovision of perennial lignocellulosic biomass across muchof Europe However successful commercialization of renew-able biofuel production depends on the sustainability andbiomass yield of lignocellulosic crops

Field trials of A donax in central Italy showed an averageyield of above‐ground biomass of about 40 Mg dry matter(DM) haminus1 yrminus1 with a peak of about 50 Mg DMha at theend of the second and third years of growth over a productioncycle of 10 years (Angelini et al 2009) Similarly in a mul-ti‐year study in Alabama A donax achieved an average yieldof 355 Mg DM haminus1 yrminus1 (Ping Bransby amp van Santen2014) However biomass yield can be doubled at the end ofthe same growth cycles in northern Italy under fertirrigatedconditions (Borin et al 2013) Biomass yields in theseexperiments using small plots may be an overestimate of theyields achievable in competitive commercial planting Whileincreasing yield from the first to the third one‐year coppicerotation was often observed in A donax (Borin et al 2013Cosentino Copani Patanegrave Mantineo amp DAgosta 2008Nassi o Di Nasso Roncucci amp Bonari 2013) different con-clusions for the yield trend from the third year onward canbeen drawn (Lewandowski et al 2003) Although thereported biomass yield for A donax is high under irrigatedconditions the large‐scale plantations of this crop remain amajor limitation due to the high cost of propagation (Ceottoamp Di Candilo 2010) Biomass yields may also be consider-ably lower on marginal land For example on a sandy loamwith limited nutrient availability A donax required 3 yearsto accumulate 20 Mg DMha and build a rhizome mass of16 Mg DMha (Nassi o Di Nasso et al 2013)

Although A donax can flower and produce seed undersome conditions as yet there is no evidence that the seeds areviable (Barney amp Ditomaso 2008 Haddadchi Gross ampFatemi 2013 Lewandowski et al 2003 Mariani et al2010 Pilu et al 2012) Thus the natural variability in exist-ing populations may occur following spontaneous mutationsselected by natural pressure (Cosentino Copani DAgostaSanzone amp Mantineo 2006) Despite the low genetic diver-sity revealed in a few studies using molecular markers(Ahmad Liow Spencer amp Jasieniuk 2008 Hardion Ver-laque Baumel Juin amp Vila 2012 Hardion Verlaque Sal-tonstall Leriche amp Vila 2014 Mariani et al 2010 Piluet al 2014) heritable phenotypic differences have beendescribed in growth‐related traits (Cosentino et al 2006 Piluet al 2014) For this reason ecotype selection remainsimportant for genetic improvement of A donax (Corno et al2014 Pilu et al 2013) Further it is important that traitswhich contribute to harvestable biomass yield in grasses suchas A donax are well understood in order to drive genetic gainfor accelerated domestication as discussed for other biomass

2 | FABBRINI ET AL

crops (Robson et al 2013) A thorough understanding of theinteractions between traits both positive and negative corre-lations is required because selection for one trait can haveprofound effects on another (Farrar et al 2012)

Diversity in growth‐related traits has been reportedusing ecotypes collected from regions within the Mediter-ranean basin (Cosentino et al 2006 Saacutenchez et al 2015)including some studies of ecotypes from a wider range ofprovenances (Amaducci amp Perego 2015 Pilu et al 2014)however a comprehensive study of different parametersthat may interact with yield (ie phenology reed morphol-ogy leaf functional traits and physiology) in a large panelof A donax ecotypes collected from diverse environmentshas not been previously reported

This study aimed (a) to characterize a broad Euro‐Mediterranean panel of 82 A donax ecotypes by assaying21 biomass morphology physiology and phenology traitsin a common garden experiment and to understand howthese may correlate to each other (b) to identify traits thatmay have contributed to differences in biomass productionin field conditions and (c) to select potential high yieldingclusters through a multivariate approach This study designtherefore also serves as a key resource for parametrizingbiomass production models of A donax an emerging lig-nocellulosic crop that would ultimately be an importantcontributor to any substantive EU renewable energy strat-egy that is based on lignocellulosic biomass supply

2 | MATERIALS AND METHODS

21 | Collection of plant material

From 2006 to 2010 eighty‐two Euro‐Mediterranean eco-types of A donax were collected for use in the study To

avoid clonal ramets collection sites were at least 4 km andat most 474 km apart (mean distance = 75 km) (Figure 1)The local geography of collection sites included riverbanksand riverbeds of the Mediterranean basin uncultivatedfields and marginal areas At each site rhizomes were col-lected from a single clump that was considered to be a dis-tinct ecotype

22 | Experimental site

In May 2011 rhizomes were planted in a common gardenexperiment at the experimental field of the ldquoAlasia FrancoVivairdquo Savigliano in northern Italy (SAV 44deg36primeN 07deg38primeE 340 m above sea level) The soil is alluvial with sandyloam texture The regional climate is fully humid and warmtemperate with warm summers according to the Koumlppen‐Geiger climate classification system (Kottek Grieser BeckRudolf amp Rubel 2006) The average annual precipitationand temperature was 8482 mm and 105degC respectivelyDuring this study spring was wetter and colder and sum-mer was drier than the 5‐year seasonal average (Figure 2)During the sampling period and specifically from 19 Julyto 23 August the weather was extremely dry and warmwith a daily average maximum temperature of 294degC and12 mm of accumulated rainfall (Figure 2) as assessed by aweather station located 1 km from the common gardenexperimental field

23 | Experimental design

Each ecotype was planted in a 9 m2 plot (3 times 3 m) usingfive homogeneous rhizomes planted in 2 times 2 m spacingwith one rhizome in the centre of that space (quincunx sys-tem of planting) Plots were harvested at the end of each

FIGURE 1 Map showing the collection sites of the 82 ecotypes of the bioenergy grass Arundo donax Solid circles indicate the locationswhere samples were collected

FABBRINI ET AL | 3

growing season In 2011 and 2012 the site was fertilizedand irrigated by furrowing every 20 days while no irriga-tion water or fertilizer were applied during the 2013 grow-ing season During the 3 years no serious incidents ofinsects or diseases were observed and no pesticides orfungicides were applied

24 | Phenotypic data collection

Arundo donax ecotypes start the maturity phase by thethird one‐year coppice rotation (Angelini et al 2009)making phenotypic comparisons more meaningful from thethird growth year onwards Data were collected from Julyto November 2013 and traits and their abbreviations arelisted in Table 1 For each ecotype the main plot wasdivided into four 15 times 15 m subplots and three of thesubplots were randomly selected as biological replicates forall measurements Empty subplots where the original rhi-zomes died after establishment were excluded from theselection No centre rhizome pieces died therefore the con-tributions to the subplots were always the same Conse-quently at the beginning of the growing season the meannumber of available planted clumps per ecotype was 451(Supporting information Table S1)

241 | Phenology

Flowering phenological data were recorded in autumn2013 from 19 September [day of the year (DOY) 262] to30 October (DOY 303) The timing of five discrete pheno-logical stages was assessed daily by visual inspection fol-lowing a new scoring system In detail Phase 0 (Ph0)mdash

absence of flowers Phase 1 (Ph1)mdashonset of flowering withthe appearance of the first small inflorescence at the single‐stem level Phase 2 (Ph2)mdashthe first inflorescence hadreached at least 30 cm length Phase 3 (Ph3)mdashthe firstinflorescence was completely open and according toCosentino et al (2006) Phase 4 (Ph4)mdash50 of the reedsshowed completely open inflorescences

242 | Physiology

Mid‐day stomatal resistance (Rs scm) was measured onJuly 10th (Rsjuly) and August 22nd (Rsaugust) with adynamic diffusion porometer (AP4 Delta‐T devices Cam-bridge UK) Data were taken from the abaxial mid‐leafposition of three healthy fully mature and fully expandedleaves (5th from the top) from three randomly chosen reedsper ecotype between 1100 and 1300 hr Before each ses-sion of measurement the porometer was calibrated to fieldtemperature and humidity

The same leaf previously used for stomatal resistancemeasurement (Rsaugust) was collected oven‐dried at 75degCto constant weight and stored for subsequent carbon iso-tope discrimination (Δ permil) analysis Leaves were groundinto a homogenous fine powder Samples were combustedin an elemental analyzer (Thermo Flash EA 1112 SeriesBremen DE) CO2 and N2 were separated by chromatogra-phy and directly injected into a continuous‐flow IsotopeRatio Mass Spectrometer (Thermo‐Finnigan Delta XP Bre-men DE) for isotope analysis Peach leaf standards (NIST1547) were run every six samples 13C12C isotope ratio(δ13C) was calculated according to the definitions describedby Farquhar Ehleringer and Hubick (1989)

FIGURE 2 Changes in maximum (red dashed line) average (black solid line) and minimum (blue dashed line) air temperature and rainfall(simple bar) of (a) the experimental year (2013) in comparison with (b) medium‐term averages (2008ndash2012) at the experimental site in northern Italywhere the Euro‐Mediterranean panel of 82 Arundo donax ecotypes were studied Data were obtained in 2013 from a nearby meteorological station

4 | FABBRINI ET AL

δ13CsampleethpermilTHORN frac14 frac12ethRsample=RstandardTHORN 1 1000

where RsampleRstandard were referred to a Pee Dee Belem-nite standard

And Δ was calculated according to Farquhar et al (1989)

Δ frac14 frac12ethδ13Catm δ13CsampleTHORN=frac121thorn ethδ13Csample=1000THORNwhere the δ13Catm was assumed minus8permil

243 | Canopy and leaf morphology

Leaf area index (LAI m2m2) was indirectly calculated ineach subplot by measuring incoming photosynthetic activeradiation (PAR) below the canopy and in an open area

using the AccuPAR LP‐80 ceptometer (Decagon DevicesPullman WA USA) Three mature leaves of each ecotypewere collected from the middle part of the main axis of thelongest reed (stem) in each subplot (Hardion et al 2012)Leaves were immediately transported to the laboratory andphotographed using a digital camera Before image analy-sis pictures were corrected for geometric distortions causedby the camera lens using filters available in Adobe Photo-shop Elements 12 Leaf length (LfL cm) leaf width (LfWcm) and leaf area (LfA cm2) were then measured using theLAMINA software (Bylesjouml et al 2008) The ratiobetween LfL and LfW (LfL_LfW) was calculated and rep-resents an independent shape variable that has been usedextensively in leaf morphometric analysis (Aravanopoulos

TABLE 1 Abbreviation units and description of phenological physiological morphological and biomass and growth traits

Traits Abbreviation Units Description

Phenological traits

FloweringmdashPhase 1 Ph1a (DOY) Reflect flowering time and have large impact on the fitness and successof plants under environmental stressesFloweringmdashPhase 2 Ph2a (DOY)

FloweringmdashPhase 3 Ph3a (DOY)

FloweringmdashPhase 4 Ph4a (DOY)

Physiological traits

Stomatal resistance in July Rsjuly (scm) Reflects leaf gas exchange and adaptation to abiotic environments

Stomatal resistance in August Rsaugust (scm)

Carbon isotope discrimination Δ (permil) Reflects variation in water use efficiency

Morphological traits

Leaf area index LAI (m2m2) Descriptors of the photosynthetically active surface area of the canopyreflects the adaptation mechanisms of plants to the environmentLeaf length LfL (cm)

Leaf width LfW (cm)

Ratio between LfL and LfW LfL_LfW mdash

Leaf area LfA (cm2) Larger leaf area tends to absorb more light and use it more efficiently ingrowing biomass

Specific leaf area SLA (cm2g) Descriptor of leaf structure reflects differences in environmental stresses

Biomass and growth traits

Total biomass production TotDM (kgDMm2)

Higher biomass yield increases land use efficiency allowing morebioenergy to be generated for any given area and more carbonsequestration

Stem dry mass StDM (gDM)

Stem density StN (nm2) Higher number of stems per culm often results in more biomass

Stem nodes StNd (n) Higher number of nodes on the culm means more meristem tissues andtends to produce more biomass

Stem height H (cm) Greater stem height tends to achieve higher biomass

Stem base diameter Db (mm) Higher stem diameter tends to boost mechanical strength and achieve agreater biomass productionStem medium diameter Dm (mm)

Stem apical diameter Da (mm)

aNon‐replicated measurements bNo unit for the specific leaf trait

FABBRINI ET AL | 5

2010 Lebrija‐Trejos Peacuterez‐Garciacutea Meave Bongers ampPoorter 2010 Sun et al 2014) All leaves were thenoven‐dried at 60degC for 72 hr to constant weight and thespecific leaf area (SLA cm2g) was determined as the ratiobetween LfA and leaf dry weight

244 | Growth and biomass production

To assess growth‐related traits and biomass production allecotypes were harvested at the end of the growing period inNovember 2013 by cutting at 5 cm aboveground level Ineach subplot the weight of the above‐ground fresh biomasswas measured and the number of reeds counted to calculatestem density (StN nm) Five stems were randomly selectedand the following measurements were performed height ofthe stem (H cm) from the base‐node to the top‐node withoutthe inflorescence (if present) number of nodes per stem(StNd n) diameter of the stem in three positions above theground (a) 20 cm (Db mm) (b) at half of H (Dm mm) and(c) 20 cm below the stems apex (Da mm) and fresh weight(Cosentino et al 2006) Six sub‐samples of whole stems perecotype two for each selected subplot were weighed andoven‐dried at 75degC until a constant weight was obtained todetermine the stem dry mass (StDM g DM) and the totalbiomass production (TotDM Kg DMm)

25 | Statistical analyses

Statistical analyses of the data were performed using R soft-ware (Version 2152 A Language and Environment Copy-right 2012) Original values were tested for transformationusing the Box‐Cox procedure (Venables amp Ripley 2002) toensure homoscedasticity and normality of residuals Besidesthe determination of means and ranges the coefficient ofvariation (CV) calculated as the ratio of the traits standarddeviation to its mean value and reported as a percentage ()was used to compare the variability of traits independently ofthe units of measure To test differences among ecotypes andreplicates of each ecotype variables were analysed using theanalysis of variance (ANOVA) model as follows

Yijk frac14 μthorn Ei thorn Rj thorn ɛijk

where Yijk is the individual value μ is the general meanEi is the effect of the ith ecotype (random) Rj is the effectof the jth replicate (fixed) and εijk is the residual error term

With regard to the stomatal resistance measured at twotime‐points the following model was used to test time (T)and ecotype by time interaction (E times T) effects

Yikz frac14 μthorn Ei thorn Tk thorn Ei Tk thorn ɛikz

In this case T was the time effect considered as randomand E times T was the interaction effect considered as random

Because of the absence of a replicated block design R couldnot be corrected when significant to eliminate micro‐environ-mental variations and thus this source of variation wasremoved to make the ANOVA models simpler When theanalysis of variance showed significant E times T interaction(P le 005) the change in ranking of ecotypes was evaluatedby the Spearman rank coefficient (ρ) on ecotypic means

Pearsons correlation coefficient (r) was used to assessthe relationship among all traits using mean values foreach ecotype All statistical tests were considered signifi-cant at P le 005

Multiple linear regression analysis was used to analysetraits affecting the amount of dry biomass production andto select a sub‐set of variables to be used for cluster analy-sis Hence the analysis was performed taking into consid-eration TotDM as dependent variable and all other traits(except flowering stages for which only non‐replicatedmeasurements were obtainable) as independent variablesAs the residuals have to be normally distributed the Box‐Cox procedure (Venables amp Ripley 2002) was performedto search for the most appropriate transformation of thedependent variables of the linear regression model

Hierarchical cluster analysis was performed to group eco-types with similar characteristics on a multivariate basis andthe Euclidean distance was used to determine the differencebetween the groups The analysis was performed using stan-dardized values of all traits Wards minimum variancemethod was used to construct the clusters and the actualnumber of clusters was discerned using ldquoNbClustrdquo packagein R (Charrad Ghazzali Boiteau amp Niknafs 2013) whichprovides most of the popular indices for determining theoptimal number of clusters ANOVA model was used to testthe null hypothesis that clusters (Clu) were equal as follows

Yij frac14 μthorn Cluj thorn ɛij

where Yij is the individual value μ is the general meanCluj is the effect of the jth cluster (random) and εij is theresidual error term LSD post hoc test was used to comparesignificant differences among mean cluster values

To describe ecotypes with their interrelations and at thesame time to create a graphical display of the clusters theldquoCLUSPLOTrdquo algorithm of Pison Struyf and Rousseeuw(1999) was used Finally the dimension of the data wasreduced by principal component analysis (PCA) thenldquoCLUSPLOTrdquo displayed the bivariate plot of the objectsrelative to the first two principal components (components1 and 2) (Pison et al 1999)

26 | Broad‐sense heritability

Broad‐sense trait heritability on an individual basis (H2)was estimated as follows

6 | FABBRINI ET AL

H2 frac14 σ2G=ethσ2G thorn σ2ɛTHORNwhere σ2G and σ2ɛ are the genetic and residual variancecomponents respectively which were obtained using therestricted maximum‐likelihood (REML) procedure (Hol-land Nyquist amp Cervantes‐Martinez 2003) H2 was evalu-ated for all the studied traits except for flowering stagesfor which replicated measurements were not viable

3 | RESULTS

31 | Univariate analysis

A significant level of phenotypic variation among the 82different ecotypes of A donax was observed in all

physiological canopy growth and biomass traits measured(Table 2) Flowering phenology of all ecotypes wasrecorded using a new scoring system developed in thisstudy that covers the onset of flowering (Ph1) to the end offlower development (Ph4) Flowering was evaluated at plotscale without replication (Table 1) therefore statisticaldifferences between ecotypes and blocks for flowering phe-nology trait could not be evaluated (Table 2) Averageflowering time of Ph1 and Ph4 at ecotype level were DOY265 (22 September) and DOY 284 (11 October) respec-tively giving an average of 19 days between Ph1 and Ph4Transition between the first 3 phases was rapid for exam-ple taking an average of 4 days from Ph1 to Ph2 and fromPh2 to Ph3 while an average of 11 days were required tocomplete the last flowering phase (Figure 3 Table 2) Due

TABLE 2 General means ecotypic range of variation coefficient of variation (CV ) and level of significance of ecotype (E) and replicate(R) effects for phenological morphological physiological and growth traits in Arundo donax Broad‐sense heritability (H2) for physiologicalmorphological growth and biomass traits See Table 1 for trait abbreviations

Trait Mean Range of variation CV ()

F test

H2E R

Phenology

Ph1 (DOY) 26512 26200ndash27700 140 nd nd nd

Ph2 (DOY) 26967 26200ndash28100 207 nd nd nd

Ph3 (DOY) 27330 26200ndash29000 246 nd nd nd

Ph4 (DOY) 28403 26200ndash30300 493 nd nd nd

Physiology

Rsjuly (scm) 096 057ndash168 2820 028

Rsaugust (scm) 150 080ndash269 2776 ns 062

Δ (permil) 2098 1943ndash2223 242 ns 019

Canopy and leaf morphology

LAI (m2m2) 611 389ndash814 1523 046

LfL (cm) 5732 4825ndash6530 627 ns 017

LfW (cm) 583 461ndash673 856 ns 039

LfL_LfW 990 859ndash1162 750 ns 030

LfA (cm2) 20955 14634ndash26093 1251 ns 030

SLA (cm2g) 13336 10672ndash17640 1023 ns 025

Growth and biomass production

TotDM (kg DMm2) 920 526ndash1521 2318 032

StDM (g DM) 28640 18332ndash46782 2216 ns 039

StN (nm2) 4044 2193ndash8000 2530 054

StNd (n) 3512 2847ndash4680 1030 ns 051

H (cm) 41234 29080ndash52833 1026 ns 058

Db (cm) 2060 1678ndash2601 988 ns 051

Dm (cm) 1606 1225ndash2264 1180 ns 044

Da (cm) 1116 852ndash1391 940 ns 024

Ns non‐significant nd F test and H2 not available for phenology traitsP le 005 P le 001 P le 0001

FABBRINI ET AL | 7

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

crops (Robson et al 2013) A thorough understanding of theinteractions between traits both positive and negative corre-lations is required because selection for one trait can haveprofound effects on another (Farrar et al 2012)

Diversity in growth‐related traits has been reportedusing ecotypes collected from regions within the Mediter-ranean basin (Cosentino et al 2006 Saacutenchez et al 2015)including some studies of ecotypes from a wider range ofprovenances (Amaducci amp Perego 2015 Pilu et al 2014)however a comprehensive study of different parametersthat may interact with yield (ie phenology reed morphol-ogy leaf functional traits and physiology) in a large panelof A donax ecotypes collected from diverse environmentshas not been previously reported

This study aimed (a) to characterize a broad Euro‐Mediterranean panel of 82 A donax ecotypes by assaying21 biomass morphology physiology and phenology traitsin a common garden experiment and to understand howthese may correlate to each other (b) to identify traits thatmay have contributed to differences in biomass productionin field conditions and (c) to select potential high yieldingclusters through a multivariate approach This study designtherefore also serves as a key resource for parametrizingbiomass production models of A donax an emerging lig-nocellulosic crop that would ultimately be an importantcontributor to any substantive EU renewable energy strat-egy that is based on lignocellulosic biomass supply

2 | MATERIALS AND METHODS

21 | Collection of plant material

From 2006 to 2010 eighty‐two Euro‐Mediterranean eco-types of A donax were collected for use in the study To

avoid clonal ramets collection sites were at least 4 km andat most 474 km apart (mean distance = 75 km) (Figure 1)The local geography of collection sites included riverbanksand riverbeds of the Mediterranean basin uncultivatedfields and marginal areas At each site rhizomes were col-lected from a single clump that was considered to be a dis-tinct ecotype

22 | Experimental site

In May 2011 rhizomes were planted in a common gardenexperiment at the experimental field of the ldquoAlasia FrancoVivairdquo Savigliano in northern Italy (SAV 44deg36primeN 07deg38primeE 340 m above sea level) The soil is alluvial with sandyloam texture The regional climate is fully humid and warmtemperate with warm summers according to the Koumlppen‐Geiger climate classification system (Kottek Grieser BeckRudolf amp Rubel 2006) The average annual precipitationand temperature was 8482 mm and 105degC respectivelyDuring this study spring was wetter and colder and sum-mer was drier than the 5‐year seasonal average (Figure 2)During the sampling period and specifically from 19 Julyto 23 August the weather was extremely dry and warmwith a daily average maximum temperature of 294degC and12 mm of accumulated rainfall (Figure 2) as assessed by aweather station located 1 km from the common gardenexperimental field

23 | Experimental design

Each ecotype was planted in a 9 m2 plot (3 times 3 m) usingfive homogeneous rhizomes planted in 2 times 2 m spacingwith one rhizome in the centre of that space (quincunx sys-tem of planting) Plots were harvested at the end of each

FIGURE 1 Map showing the collection sites of the 82 ecotypes of the bioenergy grass Arundo donax Solid circles indicate the locationswhere samples were collected

FABBRINI ET AL | 3

growing season In 2011 and 2012 the site was fertilizedand irrigated by furrowing every 20 days while no irriga-tion water or fertilizer were applied during the 2013 grow-ing season During the 3 years no serious incidents ofinsects or diseases were observed and no pesticides orfungicides were applied

24 | Phenotypic data collection

Arundo donax ecotypes start the maturity phase by thethird one‐year coppice rotation (Angelini et al 2009)making phenotypic comparisons more meaningful from thethird growth year onwards Data were collected from Julyto November 2013 and traits and their abbreviations arelisted in Table 1 For each ecotype the main plot wasdivided into four 15 times 15 m subplots and three of thesubplots were randomly selected as biological replicates forall measurements Empty subplots where the original rhi-zomes died after establishment were excluded from theselection No centre rhizome pieces died therefore the con-tributions to the subplots were always the same Conse-quently at the beginning of the growing season the meannumber of available planted clumps per ecotype was 451(Supporting information Table S1)

241 | Phenology

Flowering phenological data were recorded in autumn2013 from 19 September [day of the year (DOY) 262] to30 October (DOY 303) The timing of five discrete pheno-logical stages was assessed daily by visual inspection fol-lowing a new scoring system In detail Phase 0 (Ph0)mdash

absence of flowers Phase 1 (Ph1)mdashonset of flowering withthe appearance of the first small inflorescence at the single‐stem level Phase 2 (Ph2)mdashthe first inflorescence hadreached at least 30 cm length Phase 3 (Ph3)mdashthe firstinflorescence was completely open and according toCosentino et al (2006) Phase 4 (Ph4)mdash50 of the reedsshowed completely open inflorescences

242 | Physiology

Mid‐day stomatal resistance (Rs scm) was measured onJuly 10th (Rsjuly) and August 22nd (Rsaugust) with adynamic diffusion porometer (AP4 Delta‐T devices Cam-bridge UK) Data were taken from the abaxial mid‐leafposition of three healthy fully mature and fully expandedleaves (5th from the top) from three randomly chosen reedsper ecotype between 1100 and 1300 hr Before each ses-sion of measurement the porometer was calibrated to fieldtemperature and humidity

The same leaf previously used for stomatal resistancemeasurement (Rsaugust) was collected oven‐dried at 75degCto constant weight and stored for subsequent carbon iso-tope discrimination (Δ permil) analysis Leaves were groundinto a homogenous fine powder Samples were combustedin an elemental analyzer (Thermo Flash EA 1112 SeriesBremen DE) CO2 and N2 were separated by chromatogra-phy and directly injected into a continuous‐flow IsotopeRatio Mass Spectrometer (Thermo‐Finnigan Delta XP Bre-men DE) for isotope analysis Peach leaf standards (NIST1547) were run every six samples 13C12C isotope ratio(δ13C) was calculated according to the definitions describedby Farquhar Ehleringer and Hubick (1989)

FIGURE 2 Changes in maximum (red dashed line) average (black solid line) and minimum (blue dashed line) air temperature and rainfall(simple bar) of (a) the experimental year (2013) in comparison with (b) medium‐term averages (2008ndash2012) at the experimental site in northern Italywhere the Euro‐Mediterranean panel of 82 Arundo donax ecotypes were studied Data were obtained in 2013 from a nearby meteorological station

4 | FABBRINI ET AL

δ13CsampleethpermilTHORN frac14 frac12ethRsample=RstandardTHORN 1 1000

where RsampleRstandard were referred to a Pee Dee Belem-nite standard

And Δ was calculated according to Farquhar et al (1989)

Δ frac14 frac12ethδ13Catm δ13CsampleTHORN=frac121thorn ethδ13Csample=1000THORNwhere the δ13Catm was assumed minus8permil

243 | Canopy and leaf morphology

Leaf area index (LAI m2m2) was indirectly calculated ineach subplot by measuring incoming photosynthetic activeradiation (PAR) below the canopy and in an open area

using the AccuPAR LP‐80 ceptometer (Decagon DevicesPullman WA USA) Three mature leaves of each ecotypewere collected from the middle part of the main axis of thelongest reed (stem) in each subplot (Hardion et al 2012)Leaves were immediately transported to the laboratory andphotographed using a digital camera Before image analy-sis pictures were corrected for geometric distortions causedby the camera lens using filters available in Adobe Photo-shop Elements 12 Leaf length (LfL cm) leaf width (LfWcm) and leaf area (LfA cm2) were then measured using theLAMINA software (Bylesjouml et al 2008) The ratiobetween LfL and LfW (LfL_LfW) was calculated and rep-resents an independent shape variable that has been usedextensively in leaf morphometric analysis (Aravanopoulos

TABLE 1 Abbreviation units and description of phenological physiological morphological and biomass and growth traits

Traits Abbreviation Units Description

Phenological traits

FloweringmdashPhase 1 Ph1a (DOY) Reflect flowering time and have large impact on the fitness and successof plants under environmental stressesFloweringmdashPhase 2 Ph2a (DOY)

FloweringmdashPhase 3 Ph3a (DOY)

FloweringmdashPhase 4 Ph4a (DOY)

Physiological traits

Stomatal resistance in July Rsjuly (scm) Reflects leaf gas exchange and adaptation to abiotic environments

Stomatal resistance in August Rsaugust (scm)

Carbon isotope discrimination Δ (permil) Reflects variation in water use efficiency

Morphological traits

Leaf area index LAI (m2m2) Descriptors of the photosynthetically active surface area of the canopyreflects the adaptation mechanisms of plants to the environmentLeaf length LfL (cm)

Leaf width LfW (cm)

Ratio between LfL and LfW LfL_LfW mdash

Leaf area LfA (cm2) Larger leaf area tends to absorb more light and use it more efficiently ingrowing biomass

Specific leaf area SLA (cm2g) Descriptor of leaf structure reflects differences in environmental stresses

Biomass and growth traits

Total biomass production TotDM (kgDMm2)

Higher biomass yield increases land use efficiency allowing morebioenergy to be generated for any given area and more carbonsequestration

Stem dry mass StDM (gDM)

Stem density StN (nm2) Higher number of stems per culm often results in more biomass

Stem nodes StNd (n) Higher number of nodes on the culm means more meristem tissues andtends to produce more biomass

Stem height H (cm) Greater stem height tends to achieve higher biomass

Stem base diameter Db (mm) Higher stem diameter tends to boost mechanical strength and achieve agreater biomass productionStem medium diameter Dm (mm)

Stem apical diameter Da (mm)

aNon‐replicated measurements bNo unit for the specific leaf trait

FABBRINI ET AL | 5

2010 Lebrija‐Trejos Peacuterez‐Garciacutea Meave Bongers ampPoorter 2010 Sun et al 2014) All leaves were thenoven‐dried at 60degC for 72 hr to constant weight and thespecific leaf area (SLA cm2g) was determined as the ratiobetween LfA and leaf dry weight

244 | Growth and biomass production

To assess growth‐related traits and biomass production allecotypes were harvested at the end of the growing period inNovember 2013 by cutting at 5 cm aboveground level Ineach subplot the weight of the above‐ground fresh biomasswas measured and the number of reeds counted to calculatestem density (StN nm) Five stems were randomly selectedand the following measurements were performed height ofthe stem (H cm) from the base‐node to the top‐node withoutthe inflorescence (if present) number of nodes per stem(StNd n) diameter of the stem in three positions above theground (a) 20 cm (Db mm) (b) at half of H (Dm mm) and(c) 20 cm below the stems apex (Da mm) and fresh weight(Cosentino et al 2006) Six sub‐samples of whole stems perecotype two for each selected subplot were weighed andoven‐dried at 75degC until a constant weight was obtained todetermine the stem dry mass (StDM g DM) and the totalbiomass production (TotDM Kg DMm)

25 | Statistical analyses

Statistical analyses of the data were performed using R soft-ware (Version 2152 A Language and Environment Copy-right 2012) Original values were tested for transformationusing the Box‐Cox procedure (Venables amp Ripley 2002) toensure homoscedasticity and normality of residuals Besidesthe determination of means and ranges the coefficient ofvariation (CV) calculated as the ratio of the traits standarddeviation to its mean value and reported as a percentage ()was used to compare the variability of traits independently ofthe units of measure To test differences among ecotypes andreplicates of each ecotype variables were analysed using theanalysis of variance (ANOVA) model as follows

Yijk frac14 μthorn Ei thorn Rj thorn ɛijk

where Yijk is the individual value μ is the general meanEi is the effect of the ith ecotype (random) Rj is the effectof the jth replicate (fixed) and εijk is the residual error term

With regard to the stomatal resistance measured at twotime‐points the following model was used to test time (T)and ecotype by time interaction (E times T) effects

Yikz frac14 μthorn Ei thorn Tk thorn Ei Tk thorn ɛikz

In this case T was the time effect considered as randomand E times T was the interaction effect considered as random

Because of the absence of a replicated block design R couldnot be corrected when significant to eliminate micro‐environ-mental variations and thus this source of variation wasremoved to make the ANOVA models simpler When theanalysis of variance showed significant E times T interaction(P le 005) the change in ranking of ecotypes was evaluatedby the Spearman rank coefficient (ρ) on ecotypic means

Pearsons correlation coefficient (r) was used to assessthe relationship among all traits using mean values foreach ecotype All statistical tests were considered signifi-cant at P le 005

Multiple linear regression analysis was used to analysetraits affecting the amount of dry biomass production andto select a sub‐set of variables to be used for cluster analy-sis Hence the analysis was performed taking into consid-eration TotDM as dependent variable and all other traits(except flowering stages for which only non‐replicatedmeasurements were obtainable) as independent variablesAs the residuals have to be normally distributed the Box‐Cox procedure (Venables amp Ripley 2002) was performedto search for the most appropriate transformation of thedependent variables of the linear regression model

Hierarchical cluster analysis was performed to group eco-types with similar characteristics on a multivariate basis andthe Euclidean distance was used to determine the differencebetween the groups The analysis was performed using stan-dardized values of all traits Wards minimum variancemethod was used to construct the clusters and the actualnumber of clusters was discerned using ldquoNbClustrdquo packagein R (Charrad Ghazzali Boiteau amp Niknafs 2013) whichprovides most of the popular indices for determining theoptimal number of clusters ANOVA model was used to testthe null hypothesis that clusters (Clu) were equal as follows

Yij frac14 μthorn Cluj thorn ɛij

where Yij is the individual value μ is the general meanCluj is the effect of the jth cluster (random) and εij is theresidual error term LSD post hoc test was used to comparesignificant differences among mean cluster values

To describe ecotypes with their interrelations and at thesame time to create a graphical display of the clusters theldquoCLUSPLOTrdquo algorithm of Pison Struyf and Rousseeuw(1999) was used Finally the dimension of the data wasreduced by principal component analysis (PCA) thenldquoCLUSPLOTrdquo displayed the bivariate plot of the objectsrelative to the first two principal components (components1 and 2) (Pison et al 1999)

26 | Broad‐sense heritability

Broad‐sense trait heritability on an individual basis (H2)was estimated as follows

6 | FABBRINI ET AL

H2 frac14 σ2G=ethσ2G thorn σ2ɛTHORNwhere σ2G and σ2ɛ are the genetic and residual variancecomponents respectively which were obtained using therestricted maximum‐likelihood (REML) procedure (Hol-land Nyquist amp Cervantes‐Martinez 2003) H2 was evalu-ated for all the studied traits except for flowering stagesfor which replicated measurements were not viable

3 | RESULTS

31 | Univariate analysis

A significant level of phenotypic variation among the 82different ecotypes of A donax was observed in all

physiological canopy growth and biomass traits measured(Table 2) Flowering phenology of all ecotypes wasrecorded using a new scoring system developed in thisstudy that covers the onset of flowering (Ph1) to the end offlower development (Ph4) Flowering was evaluated at plotscale without replication (Table 1) therefore statisticaldifferences between ecotypes and blocks for flowering phe-nology trait could not be evaluated (Table 2) Averageflowering time of Ph1 and Ph4 at ecotype level were DOY265 (22 September) and DOY 284 (11 October) respec-tively giving an average of 19 days between Ph1 and Ph4Transition between the first 3 phases was rapid for exam-ple taking an average of 4 days from Ph1 to Ph2 and fromPh2 to Ph3 while an average of 11 days were required tocomplete the last flowering phase (Figure 3 Table 2) Due

TABLE 2 General means ecotypic range of variation coefficient of variation (CV ) and level of significance of ecotype (E) and replicate(R) effects for phenological morphological physiological and growth traits in Arundo donax Broad‐sense heritability (H2) for physiologicalmorphological growth and biomass traits See Table 1 for trait abbreviations

Trait Mean Range of variation CV ()

F test

H2E R

Phenology

Ph1 (DOY) 26512 26200ndash27700 140 nd nd nd

Ph2 (DOY) 26967 26200ndash28100 207 nd nd nd

Ph3 (DOY) 27330 26200ndash29000 246 nd nd nd

Ph4 (DOY) 28403 26200ndash30300 493 nd nd nd

Physiology

Rsjuly (scm) 096 057ndash168 2820 028

Rsaugust (scm) 150 080ndash269 2776 ns 062

Δ (permil) 2098 1943ndash2223 242 ns 019

Canopy and leaf morphology

LAI (m2m2) 611 389ndash814 1523 046

LfL (cm) 5732 4825ndash6530 627 ns 017

LfW (cm) 583 461ndash673 856 ns 039

LfL_LfW 990 859ndash1162 750 ns 030

LfA (cm2) 20955 14634ndash26093 1251 ns 030

SLA (cm2g) 13336 10672ndash17640 1023 ns 025

Growth and biomass production

TotDM (kg DMm2) 920 526ndash1521 2318 032

StDM (g DM) 28640 18332ndash46782 2216 ns 039

StN (nm2) 4044 2193ndash8000 2530 054

StNd (n) 3512 2847ndash4680 1030 ns 051

H (cm) 41234 29080ndash52833 1026 ns 058

Db (cm) 2060 1678ndash2601 988 ns 051

Dm (cm) 1606 1225ndash2264 1180 ns 044

Da (cm) 1116 852ndash1391 940 ns 024

Ns non‐significant nd F test and H2 not available for phenology traitsP le 005 P le 001 P le 0001

FABBRINI ET AL | 7

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

growing season In 2011 and 2012 the site was fertilizedand irrigated by furrowing every 20 days while no irriga-tion water or fertilizer were applied during the 2013 grow-ing season During the 3 years no serious incidents ofinsects or diseases were observed and no pesticides orfungicides were applied

24 | Phenotypic data collection

Arundo donax ecotypes start the maturity phase by thethird one‐year coppice rotation (Angelini et al 2009)making phenotypic comparisons more meaningful from thethird growth year onwards Data were collected from Julyto November 2013 and traits and their abbreviations arelisted in Table 1 For each ecotype the main plot wasdivided into four 15 times 15 m subplots and three of thesubplots were randomly selected as biological replicates forall measurements Empty subplots where the original rhi-zomes died after establishment were excluded from theselection No centre rhizome pieces died therefore the con-tributions to the subplots were always the same Conse-quently at the beginning of the growing season the meannumber of available planted clumps per ecotype was 451(Supporting information Table S1)

241 | Phenology

Flowering phenological data were recorded in autumn2013 from 19 September [day of the year (DOY) 262] to30 October (DOY 303) The timing of five discrete pheno-logical stages was assessed daily by visual inspection fol-lowing a new scoring system In detail Phase 0 (Ph0)mdash

absence of flowers Phase 1 (Ph1)mdashonset of flowering withthe appearance of the first small inflorescence at the single‐stem level Phase 2 (Ph2)mdashthe first inflorescence hadreached at least 30 cm length Phase 3 (Ph3)mdashthe firstinflorescence was completely open and according toCosentino et al (2006) Phase 4 (Ph4)mdash50 of the reedsshowed completely open inflorescences

242 | Physiology

Mid‐day stomatal resistance (Rs scm) was measured onJuly 10th (Rsjuly) and August 22nd (Rsaugust) with adynamic diffusion porometer (AP4 Delta‐T devices Cam-bridge UK) Data were taken from the abaxial mid‐leafposition of three healthy fully mature and fully expandedleaves (5th from the top) from three randomly chosen reedsper ecotype between 1100 and 1300 hr Before each ses-sion of measurement the porometer was calibrated to fieldtemperature and humidity

The same leaf previously used for stomatal resistancemeasurement (Rsaugust) was collected oven‐dried at 75degCto constant weight and stored for subsequent carbon iso-tope discrimination (Δ permil) analysis Leaves were groundinto a homogenous fine powder Samples were combustedin an elemental analyzer (Thermo Flash EA 1112 SeriesBremen DE) CO2 and N2 were separated by chromatogra-phy and directly injected into a continuous‐flow IsotopeRatio Mass Spectrometer (Thermo‐Finnigan Delta XP Bre-men DE) for isotope analysis Peach leaf standards (NIST1547) were run every six samples 13C12C isotope ratio(δ13C) was calculated according to the definitions describedby Farquhar Ehleringer and Hubick (1989)

FIGURE 2 Changes in maximum (red dashed line) average (black solid line) and minimum (blue dashed line) air temperature and rainfall(simple bar) of (a) the experimental year (2013) in comparison with (b) medium‐term averages (2008ndash2012) at the experimental site in northern Italywhere the Euro‐Mediterranean panel of 82 Arundo donax ecotypes were studied Data were obtained in 2013 from a nearby meteorological station

4 | FABBRINI ET AL

δ13CsampleethpermilTHORN frac14 frac12ethRsample=RstandardTHORN 1 1000

where RsampleRstandard were referred to a Pee Dee Belem-nite standard

And Δ was calculated according to Farquhar et al (1989)

Δ frac14 frac12ethδ13Catm δ13CsampleTHORN=frac121thorn ethδ13Csample=1000THORNwhere the δ13Catm was assumed minus8permil

243 | Canopy and leaf morphology

Leaf area index (LAI m2m2) was indirectly calculated ineach subplot by measuring incoming photosynthetic activeradiation (PAR) below the canopy and in an open area

using the AccuPAR LP‐80 ceptometer (Decagon DevicesPullman WA USA) Three mature leaves of each ecotypewere collected from the middle part of the main axis of thelongest reed (stem) in each subplot (Hardion et al 2012)Leaves were immediately transported to the laboratory andphotographed using a digital camera Before image analy-sis pictures were corrected for geometric distortions causedby the camera lens using filters available in Adobe Photo-shop Elements 12 Leaf length (LfL cm) leaf width (LfWcm) and leaf area (LfA cm2) were then measured using theLAMINA software (Bylesjouml et al 2008) The ratiobetween LfL and LfW (LfL_LfW) was calculated and rep-resents an independent shape variable that has been usedextensively in leaf morphometric analysis (Aravanopoulos

TABLE 1 Abbreviation units and description of phenological physiological morphological and biomass and growth traits

Traits Abbreviation Units Description

Phenological traits

FloweringmdashPhase 1 Ph1a (DOY) Reflect flowering time and have large impact on the fitness and successof plants under environmental stressesFloweringmdashPhase 2 Ph2a (DOY)

FloweringmdashPhase 3 Ph3a (DOY)

FloweringmdashPhase 4 Ph4a (DOY)

Physiological traits

Stomatal resistance in July Rsjuly (scm) Reflects leaf gas exchange and adaptation to abiotic environments

Stomatal resistance in August Rsaugust (scm)

Carbon isotope discrimination Δ (permil) Reflects variation in water use efficiency

Morphological traits

Leaf area index LAI (m2m2) Descriptors of the photosynthetically active surface area of the canopyreflects the adaptation mechanisms of plants to the environmentLeaf length LfL (cm)

Leaf width LfW (cm)

Ratio between LfL and LfW LfL_LfW mdash

Leaf area LfA (cm2) Larger leaf area tends to absorb more light and use it more efficiently ingrowing biomass

Specific leaf area SLA (cm2g) Descriptor of leaf structure reflects differences in environmental stresses

Biomass and growth traits

Total biomass production TotDM (kgDMm2)

Higher biomass yield increases land use efficiency allowing morebioenergy to be generated for any given area and more carbonsequestration

Stem dry mass StDM (gDM)

Stem density StN (nm2) Higher number of stems per culm often results in more biomass

Stem nodes StNd (n) Higher number of nodes on the culm means more meristem tissues andtends to produce more biomass

Stem height H (cm) Greater stem height tends to achieve higher biomass

Stem base diameter Db (mm) Higher stem diameter tends to boost mechanical strength and achieve agreater biomass productionStem medium diameter Dm (mm)

Stem apical diameter Da (mm)

aNon‐replicated measurements bNo unit for the specific leaf trait

FABBRINI ET AL | 5

2010 Lebrija‐Trejos Peacuterez‐Garciacutea Meave Bongers ampPoorter 2010 Sun et al 2014) All leaves were thenoven‐dried at 60degC for 72 hr to constant weight and thespecific leaf area (SLA cm2g) was determined as the ratiobetween LfA and leaf dry weight

244 | Growth and biomass production

To assess growth‐related traits and biomass production allecotypes were harvested at the end of the growing period inNovember 2013 by cutting at 5 cm aboveground level Ineach subplot the weight of the above‐ground fresh biomasswas measured and the number of reeds counted to calculatestem density (StN nm) Five stems were randomly selectedand the following measurements were performed height ofthe stem (H cm) from the base‐node to the top‐node withoutthe inflorescence (if present) number of nodes per stem(StNd n) diameter of the stem in three positions above theground (a) 20 cm (Db mm) (b) at half of H (Dm mm) and(c) 20 cm below the stems apex (Da mm) and fresh weight(Cosentino et al 2006) Six sub‐samples of whole stems perecotype two for each selected subplot were weighed andoven‐dried at 75degC until a constant weight was obtained todetermine the stem dry mass (StDM g DM) and the totalbiomass production (TotDM Kg DMm)

25 | Statistical analyses

Statistical analyses of the data were performed using R soft-ware (Version 2152 A Language and Environment Copy-right 2012) Original values were tested for transformationusing the Box‐Cox procedure (Venables amp Ripley 2002) toensure homoscedasticity and normality of residuals Besidesthe determination of means and ranges the coefficient ofvariation (CV) calculated as the ratio of the traits standarddeviation to its mean value and reported as a percentage ()was used to compare the variability of traits independently ofthe units of measure To test differences among ecotypes andreplicates of each ecotype variables were analysed using theanalysis of variance (ANOVA) model as follows

Yijk frac14 μthorn Ei thorn Rj thorn ɛijk

where Yijk is the individual value μ is the general meanEi is the effect of the ith ecotype (random) Rj is the effectof the jth replicate (fixed) and εijk is the residual error term

With regard to the stomatal resistance measured at twotime‐points the following model was used to test time (T)and ecotype by time interaction (E times T) effects

Yikz frac14 μthorn Ei thorn Tk thorn Ei Tk thorn ɛikz

In this case T was the time effect considered as randomand E times T was the interaction effect considered as random

Because of the absence of a replicated block design R couldnot be corrected when significant to eliminate micro‐environ-mental variations and thus this source of variation wasremoved to make the ANOVA models simpler When theanalysis of variance showed significant E times T interaction(P le 005) the change in ranking of ecotypes was evaluatedby the Spearman rank coefficient (ρ) on ecotypic means

Pearsons correlation coefficient (r) was used to assessthe relationship among all traits using mean values foreach ecotype All statistical tests were considered signifi-cant at P le 005

Multiple linear regression analysis was used to analysetraits affecting the amount of dry biomass production andto select a sub‐set of variables to be used for cluster analy-sis Hence the analysis was performed taking into consid-eration TotDM as dependent variable and all other traits(except flowering stages for which only non‐replicatedmeasurements were obtainable) as independent variablesAs the residuals have to be normally distributed the Box‐Cox procedure (Venables amp Ripley 2002) was performedto search for the most appropriate transformation of thedependent variables of the linear regression model

Hierarchical cluster analysis was performed to group eco-types with similar characteristics on a multivariate basis andthe Euclidean distance was used to determine the differencebetween the groups The analysis was performed using stan-dardized values of all traits Wards minimum variancemethod was used to construct the clusters and the actualnumber of clusters was discerned using ldquoNbClustrdquo packagein R (Charrad Ghazzali Boiteau amp Niknafs 2013) whichprovides most of the popular indices for determining theoptimal number of clusters ANOVA model was used to testthe null hypothesis that clusters (Clu) were equal as follows

Yij frac14 μthorn Cluj thorn ɛij

where Yij is the individual value μ is the general meanCluj is the effect of the jth cluster (random) and εij is theresidual error term LSD post hoc test was used to comparesignificant differences among mean cluster values

To describe ecotypes with their interrelations and at thesame time to create a graphical display of the clusters theldquoCLUSPLOTrdquo algorithm of Pison Struyf and Rousseeuw(1999) was used Finally the dimension of the data wasreduced by principal component analysis (PCA) thenldquoCLUSPLOTrdquo displayed the bivariate plot of the objectsrelative to the first two principal components (components1 and 2) (Pison et al 1999)

26 | Broad‐sense heritability

Broad‐sense trait heritability on an individual basis (H2)was estimated as follows

6 | FABBRINI ET AL

H2 frac14 σ2G=ethσ2G thorn σ2ɛTHORNwhere σ2G and σ2ɛ are the genetic and residual variancecomponents respectively which were obtained using therestricted maximum‐likelihood (REML) procedure (Hol-land Nyquist amp Cervantes‐Martinez 2003) H2 was evalu-ated for all the studied traits except for flowering stagesfor which replicated measurements were not viable

3 | RESULTS

31 | Univariate analysis

A significant level of phenotypic variation among the 82different ecotypes of A donax was observed in all

physiological canopy growth and biomass traits measured(Table 2) Flowering phenology of all ecotypes wasrecorded using a new scoring system developed in thisstudy that covers the onset of flowering (Ph1) to the end offlower development (Ph4) Flowering was evaluated at plotscale without replication (Table 1) therefore statisticaldifferences between ecotypes and blocks for flowering phe-nology trait could not be evaluated (Table 2) Averageflowering time of Ph1 and Ph4 at ecotype level were DOY265 (22 September) and DOY 284 (11 October) respec-tively giving an average of 19 days between Ph1 and Ph4Transition between the first 3 phases was rapid for exam-ple taking an average of 4 days from Ph1 to Ph2 and fromPh2 to Ph3 while an average of 11 days were required tocomplete the last flowering phase (Figure 3 Table 2) Due

TABLE 2 General means ecotypic range of variation coefficient of variation (CV ) and level of significance of ecotype (E) and replicate(R) effects for phenological morphological physiological and growth traits in Arundo donax Broad‐sense heritability (H2) for physiologicalmorphological growth and biomass traits See Table 1 for trait abbreviations

Trait Mean Range of variation CV ()

F test

H2E R

Phenology

Ph1 (DOY) 26512 26200ndash27700 140 nd nd nd

Ph2 (DOY) 26967 26200ndash28100 207 nd nd nd

Ph3 (DOY) 27330 26200ndash29000 246 nd nd nd

Ph4 (DOY) 28403 26200ndash30300 493 nd nd nd

Physiology

Rsjuly (scm) 096 057ndash168 2820 028

Rsaugust (scm) 150 080ndash269 2776 ns 062

Δ (permil) 2098 1943ndash2223 242 ns 019

Canopy and leaf morphology

LAI (m2m2) 611 389ndash814 1523 046

LfL (cm) 5732 4825ndash6530 627 ns 017

LfW (cm) 583 461ndash673 856 ns 039

LfL_LfW 990 859ndash1162 750 ns 030

LfA (cm2) 20955 14634ndash26093 1251 ns 030

SLA (cm2g) 13336 10672ndash17640 1023 ns 025

Growth and biomass production

TotDM (kg DMm2) 920 526ndash1521 2318 032

StDM (g DM) 28640 18332ndash46782 2216 ns 039

StN (nm2) 4044 2193ndash8000 2530 054

StNd (n) 3512 2847ndash4680 1030 ns 051

H (cm) 41234 29080ndash52833 1026 ns 058

Db (cm) 2060 1678ndash2601 988 ns 051

Dm (cm) 1606 1225ndash2264 1180 ns 044

Da (cm) 1116 852ndash1391 940 ns 024

Ns non‐significant nd F test and H2 not available for phenology traitsP le 005 P le 001 P le 0001

FABBRINI ET AL | 7

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

δ13CsampleethpermilTHORN frac14 frac12ethRsample=RstandardTHORN 1 1000

where RsampleRstandard were referred to a Pee Dee Belem-nite standard

And Δ was calculated according to Farquhar et al (1989)

Δ frac14 frac12ethδ13Catm δ13CsampleTHORN=frac121thorn ethδ13Csample=1000THORNwhere the δ13Catm was assumed minus8permil

243 | Canopy and leaf morphology

Leaf area index (LAI m2m2) was indirectly calculated ineach subplot by measuring incoming photosynthetic activeradiation (PAR) below the canopy and in an open area

using the AccuPAR LP‐80 ceptometer (Decagon DevicesPullman WA USA) Three mature leaves of each ecotypewere collected from the middle part of the main axis of thelongest reed (stem) in each subplot (Hardion et al 2012)Leaves were immediately transported to the laboratory andphotographed using a digital camera Before image analy-sis pictures were corrected for geometric distortions causedby the camera lens using filters available in Adobe Photo-shop Elements 12 Leaf length (LfL cm) leaf width (LfWcm) and leaf area (LfA cm2) were then measured using theLAMINA software (Bylesjouml et al 2008) The ratiobetween LfL and LfW (LfL_LfW) was calculated and rep-resents an independent shape variable that has been usedextensively in leaf morphometric analysis (Aravanopoulos

TABLE 1 Abbreviation units and description of phenological physiological morphological and biomass and growth traits

Traits Abbreviation Units Description

Phenological traits

FloweringmdashPhase 1 Ph1a (DOY) Reflect flowering time and have large impact on the fitness and successof plants under environmental stressesFloweringmdashPhase 2 Ph2a (DOY)

FloweringmdashPhase 3 Ph3a (DOY)

FloweringmdashPhase 4 Ph4a (DOY)

Physiological traits

Stomatal resistance in July Rsjuly (scm) Reflects leaf gas exchange and adaptation to abiotic environments

Stomatal resistance in August Rsaugust (scm)

Carbon isotope discrimination Δ (permil) Reflects variation in water use efficiency

Morphological traits

Leaf area index LAI (m2m2) Descriptors of the photosynthetically active surface area of the canopyreflects the adaptation mechanisms of plants to the environmentLeaf length LfL (cm)

Leaf width LfW (cm)

Ratio between LfL and LfW LfL_LfW mdash

Leaf area LfA (cm2) Larger leaf area tends to absorb more light and use it more efficiently ingrowing biomass

Specific leaf area SLA (cm2g) Descriptor of leaf structure reflects differences in environmental stresses

Biomass and growth traits

Total biomass production TotDM (kgDMm2)

Higher biomass yield increases land use efficiency allowing morebioenergy to be generated for any given area and more carbonsequestration

Stem dry mass StDM (gDM)

Stem density StN (nm2) Higher number of stems per culm often results in more biomass

Stem nodes StNd (n) Higher number of nodes on the culm means more meristem tissues andtends to produce more biomass

Stem height H (cm) Greater stem height tends to achieve higher biomass

Stem base diameter Db (mm) Higher stem diameter tends to boost mechanical strength and achieve agreater biomass productionStem medium diameter Dm (mm)

Stem apical diameter Da (mm)

aNon‐replicated measurements bNo unit for the specific leaf trait

FABBRINI ET AL | 5

2010 Lebrija‐Trejos Peacuterez‐Garciacutea Meave Bongers ampPoorter 2010 Sun et al 2014) All leaves were thenoven‐dried at 60degC for 72 hr to constant weight and thespecific leaf area (SLA cm2g) was determined as the ratiobetween LfA and leaf dry weight

244 | Growth and biomass production

To assess growth‐related traits and biomass production allecotypes were harvested at the end of the growing period inNovember 2013 by cutting at 5 cm aboveground level Ineach subplot the weight of the above‐ground fresh biomasswas measured and the number of reeds counted to calculatestem density (StN nm) Five stems were randomly selectedand the following measurements were performed height ofthe stem (H cm) from the base‐node to the top‐node withoutthe inflorescence (if present) number of nodes per stem(StNd n) diameter of the stem in three positions above theground (a) 20 cm (Db mm) (b) at half of H (Dm mm) and(c) 20 cm below the stems apex (Da mm) and fresh weight(Cosentino et al 2006) Six sub‐samples of whole stems perecotype two for each selected subplot were weighed andoven‐dried at 75degC until a constant weight was obtained todetermine the stem dry mass (StDM g DM) and the totalbiomass production (TotDM Kg DMm)

25 | Statistical analyses

Statistical analyses of the data were performed using R soft-ware (Version 2152 A Language and Environment Copy-right 2012) Original values were tested for transformationusing the Box‐Cox procedure (Venables amp Ripley 2002) toensure homoscedasticity and normality of residuals Besidesthe determination of means and ranges the coefficient ofvariation (CV) calculated as the ratio of the traits standarddeviation to its mean value and reported as a percentage ()was used to compare the variability of traits independently ofthe units of measure To test differences among ecotypes andreplicates of each ecotype variables were analysed using theanalysis of variance (ANOVA) model as follows

Yijk frac14 μthorn Ei thorn Rj thorn ɛijk

where Yijk is the individual value μ is the general meanEi is the effect of the ith ecotype (random) Rj is the effectof the jth replicate (fixed) and εijk is the residual error term

With regard to the stomatal resistance measured at twotime‐points the following model was used to test time (T)and ecotype by time interaction (E times T) effects

Yikz frac14 μthorn Ei thorn Tk thorn Ei Tk thorn ɛikz

In this case T was the time effect considered as randomand E times T was the interaction effect considered as random

Because of the absence of a replicated block design R couldnot be corrected when significant to eliminate micro‐environ-mental variations and thus this source of variation wasremoved to make the ANOVA models simpler When theanalysis of variance showed significant E times T interaction(P le 005) the change in ranking of ecotypes was evaluatedby the Spearman rank coefficient (ρ) on ecotypic means

Pearsons correlation coefficient (r) was used to assessthe relationship among all traits using mean values foreach ecotype All statistical tests were considered signifi-cant at P le 005

Multiple linear regression analysis was used to analysetraits affecting the amount of dry biomass production andto select a sub‐set of variables to be used for cluster analy-sis Hence the analysis was performed taking into consid-eration TotDM as dependent variable and all other traits(except flowering stages for which only non‐replicatedmeasurements were obtainable) as independent variablesAs the residuals have to be normally distributed the Box‐Cox procedure (Venables amp Ripley 2002) was performedto search for the most appropriate transformation of thedependent variables of the linear regression model

Hierarchical cluster analysis was performed to group eco-types with similar characteristics on a multivariate basis andthe Euclidean distance was used to determine the differencebetween the groups The analysis was performed using stan-dardized values of all traits Wards minimum variancemethod was used to construct the clusters and the actualnumber of clusters was discerned using ldquoNbClustrdquo packagein R (Charrad Ghazzali Boiteau amp Niknafs 2013) whichprovides most of the popular indices for determining theoptimal number of clusters ANOVA model was used to testthe null hypothesis that clusters (Clu) were equal as follows

Yij frac14 μthorn Cluj thorn ɛij

where Yij is the individual value μ is the general meanCluj is the effect of the jth cluster (random) and εij is theresidual error term LSD post hoc test was used to comparesignificant differences among mean cluster values

To describe ecotypes with their interrelations and at thesame time to create a graphical display of the clusters theldquoCLUSPLOTrdquo algorithm of Pison Struyf and Rousseeuw(1999) was used Finally the dimension of the data wasreduced by principal component analysis (PCA) thenldquoCLUSPLOTrdquo displayed the bivariate plot of the objectsrelative to the first two principal components (components1 and 2) (Pison et al 1999)

26 | Broad‐sense heritability

Broad‐sense trait heritability on an individual basis (H2)was estimated as follows

6 | FABBRINI ET AL

H2 frac14 σ2G=ethσ2G thorn σ2ɛTHORNwhere σ2G and σ2ɛ are the genetic and residual variancecomponents respectively which were obtained using therestricted maximum‐likelihood (REML) procedure (Hol-land Nyquist amp Cervantes‐Martinez 2003) H2 was evalu-ated for all the studied traits except for flowering stagesfor which replicated measurements were not viable

3 | RESULTS

31 | Univariate analysis

A significant level of phenotypic variation among the 82different ecotypes of A donax was observed in all

physiological canopy growth and biomass traits measured(Table 2) Flowering phenology of all ecotypes wasrecorded using a new scoring system developed in thisstudy that covers the onset of flowering (Ph1) to the end offlower development (Ph4) Flowering was evaluated at plotscale without replication (Table 1) therefore statisticaldifferences between ecotypes and blocks for flowering phe-nology trait could not be evaluated (Table 2) Averageflowering time of Ph1 and Ph4 at ecotype level were DOY265 (22 September) and DOY 284 (11 October) respec-tively giving an average of 19 days between Ph1 and Ph4Transition between the first 3 phases was rapid for exam-ple taking an average of 4 days from Ph1 to Ph2 and fromPh2 to Ph3 while an average of 11 days were required tocomplete the last flowering phase (Figure 3 Table 2) Due

TABLE 2 General means ecotypic range of variation coefficient of variation (CV ) and level of significance of ecotype (E) and replicate(R) effects for phenological morphological physiological and growth traits in Arundo donax Broad‐sense heritability (H2) for physiologicalmorphological growth and biomass traits See Table 1 for trait abbreviations

Trait Mean Range of variation CV ()

F test

H2E R

Phenology

Ph1 (DOY) 26512 26200ndash27700 140 nd nd nd

Ph2 (DOY) 26967 26200ndash28100 207 nd nd nd

Ph3 (DOY) 27330 26200ndash29000 246 nd nd nd

Ph4 (DOY) 28403 26200ndash30300 493 nd nd nd

Physiology

Rsjuly (scm) 096 057ndash168 2820 028

Rsaugust (scm) 150 080ndash269 2776 ns 062

Δ (permil) 2098 1943ndash2223 242 ns 019

Canopy and leaf morphology

LAI (m2m2) 611 389ndash814 1523 046

LfL (cm) 5732 4825ndash6530 627 ns 017

LfW (cm) 583 461ndash673 856 ns 039

LfL_LfW 990 859ndash1162 750 ns 030

LfA (cm2) 20955 14634ndash26093 1251 ns 030

SLA (cm2g) 13336 10672ndash17640 1023 ns 025

Growth and biomass production

TotDM (kg DMm2) 920 526ndash1521 2318 032

StDM (g DM) 28640 18332ndash46782 2216 ns 039

StN (nm2) 4044 2193ndash8000 2530 054

StNd (n) 3512 2847ndash4680 1030 ns 051

H (cm) 41234 29080ndash52833 1026 ns 058

Db (cm) 2060 1678ndash2601 988 ns 051

Dm (cm) 1606 1225ndash2264 1180 ns 044

Da (cm) 1116 852ndash1391 940 ns 024

Ns non‐significant nd F test and H2 not available for phenology traitsP le 005 P le 001 P le 0001

FABBRINI ET AL | 7

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

2010 Lebrija‐Trejos Peacuterez‐Garciacutea Meave Bongers ampPoorter 2010 Sun et al 2014) All leaves were thenoven‐dried at 60degC for 72 hr to constant weight and thespecific leaf area (SLA cm2g) was determined as the ratiobetween LfA and leaf dry weight

244 | Growth and biomass production

To assess growth‐related traits and biomass production allecotypes were harvested at the end of the growing period inNovember 2013 by cutting at 5 cm aboveground level Ineach subplot the weight of the above‐ground fresh biomasswas measured and the number of reeds counted to calculatestem density (StN nm) Five stems were randomly selectedand the following measurements were performed height ofthe stem (H cm) from the base‐node to the top‐node withoutthe inflorescence (if present) number of nodes per stem(StNd n) diameter of the stem in three positions above theground (a) 20 cm (Db mm) (b) at half of H (Dm mm) and(c) 20 cm below the stems apex (Da mm) and fresh weight(Cosentino et al 2006) Six sub‐samples of whole stems perecotype two for each selected subplot were weighed andoven‐dried at 75degC until a constant weight was obtained todetermine the stem dry mass (StDM g DM) and the totalbiomass production (TotDM Kg DMm)

25 | Statistical analyses

Statistical analyses of the data were performed using R soft-ware (Version 2152 A Language and Environment Copy-right 2012) Original values were tested for transformationusing the Box‐Cox procedure (Venables amp Ripley 2002) toensure homoscedasticity and normality of residuals Besidesthe determination of means and ranges the coefficient ofvariation (CV) calculated as the ratio of the traits standarddeviation to its mean value and reported as a percentage ()was used to compare the variability of traits independently ofthe units of measure To test differences among ecotypes andreplicates of each ecotype variables were analysed using theanalysis of variance (ANOVA) model as follows

Yijk frac14 μthorn Ei thorn Rj thorn ɛijk

where Yijk is the individual value μ is the general meanEi is the effect of the ith ecotype (random) Rj is the effectof the jth replicate (fixed) and εijk is the residual error term

With regard to the stomatal resistance measured at twotime‐points the following model was used to test time (T)and ecotype by time interaction (E times T) effects

Yikz frac14 μthorn Ei thorn Tk thorn Ei Tk thorn ɛikz

In this case T was the time effect considered as randomand E times T was the interaction effect considered as random

Because of the absence of a replicated block design R couldnot be corrected when significant to eliminate micro‐environ-mental variations and thus this source of variation wasremoved to make the ANOVA models simpler When theanalysis of variance showed significant E times T interaction(P le 005) the change in ranking of ecotypes was evaluatedby the Spearman rank coefficient (ρ) on ecotypic means

Pearsons correlation coefficient (r) was used to assessthe relationship among all traits using mean values foreach ecotype All statistical tests were considered signifi-cant at P le 005

Multiple linear regression analysis was used to analysetraits affecting the amount of dry biomass production andto select a sub‐set of variables to be used for cluster analy-sis Hence the analysis was performed taking into consid-eration TotDM as dependent variable and all other traits(except flowering stages for which only non‐replicatedmeasurements were obtainable) as independent variablesAs the residuals have to be normally distributed the Box‐Cox procedure (Venables amp Ripley 2002) was performedto search for the most appropriate transformation of thedependent variables of the linear regression model

Hierarchical cluster analysis was performed to group eco-types with similar characteristics on a multivariate basis andthe Euclidean distance was used to determine the differencebetween the groups The analysis was performed using stan-dardized values of all traits Wards minimum variancemethod was used to construct the clusters and the actualnumber of clusters was discerned using ldquoNbClustrdquo packagein R (Charrad Ghazzali Boiteau amp Niknafs 2013) whichprovides most of the popular indices for determining theoptimal number of clusters ANOVA model was used to testthe null hypothesis that clusters (Clu) were equal as follows

Yij frac14 μthorn Cluj thorn ɛij

where Yij is the individual value μ is the general meanCluj is the effect of the jth cluster (random) and εij is theresidual error term LSD post hoc test was used to comparesignificant differences among mean cluster values

To describe ecotypes with their interrelations and at thesame time to create a graphical display of the clusters theldquoCLUSPLOTrdquo algorithm of Pison Struyf and Rousseeuw(1999) was used Finally the dimension of the data wasreduced by principal component analysis (PCA) thenldquoCLUSPLOTrdquo displayed the bivariate plot of the objectsrelative to the first two principal components (components1 and 2) (Pison et al 1999)

26 | Broad‐sense heritability

Broad‐sense trait heritability on an individual basis (H2)was estimated as follows

6 | FABBRINI ET AL

H2 frac14 σ2G=ethσ2G thorn σ2ɛTHORNwhere σ2G and σ2ɛ are the genetic and residual variancecomponents respectively which were obtained using therestricted maximum‐likelihood (REML) procedure (Hol-land Nyquist amp Cervantes‐Martinez 2003) H2 was evalu-ated for all the studied traits except for flowering stagesfor which replicated measurements were not viable

3 | RESULTS

31 | Univariate analysis

A significant level of phenotypic variation among the 82different ecotypes of A donax was observed in all

physiological canopy growth and biomass traits measured(Table 2) Flowering phenology of all ecotypes wasrecorded using a new scoring system developed in thisstudy that covers the onset of flowering (Ph1) to the end offlower development (Ph4) Flowering was evaluated at plotscale without replication (Table 1) therefore statisticaldifferences between ecotypes and blocks for flowering phe-nology trait could not be evaluated (Table 2) Averageflowering time of Ph1 and Ph4 at ecotype level were DOY265 (22 September) and DOY 284 (11 October) respec-tively giving an average of 19 days between Ph1 and Ph4Transition between the first 3 phases was rapid for exam-ple taking an average of 4 days from Ph1 to Ph2 and fromPh2 to Ph3 while an average of 11 days were required tocomplete the last flowering phase (Figure 3 Table 2) Due

TABLE 2 General means ecotypic range of variation coefficient of variation (CV ) and level of significance of ecotype (E) and replicate(R) effects for phenological morphological physiological and growth traits in Arundo donax Broad‐sense heritability (H2) for physiologicalmorphological growth and biomass traits See Table 1 for trait abbreviations

Trait Mean Range of variation CV ()

F test

H2E R

Phenology

Ph1 (DOY) 26512 26200ndash27700 140 nd nd nd

Ph2 (DOY) 26967 26200ndash28100 207 nd nd nd

Ph3 (DOY) 27330 26200ndash29000 246 nd nd nd

Ph4 (DOY) 28403 26200ndash30300 493 nd nd nd

Physiology

Rsjuly (scm) 096 057ndash168 2820 028

Rsaugust (scm) 150 080ndash269 2776 ns 062

Δ (permil) 2098 1943ndash2223 242 ns 019

Canopy and leaf morphology

LAI (m2m2) 611 389ndash814 1523 046

LfL (cm) 5732 4825ndash6530 627 ns 017

LfW (cm) 583 461ndash673 856 ns 039

LfL_LfW 990 859ndash1162 750 ns 030

LfA (cm2) 20955 14634ndash26093 1251 ns 030

SLA (cm2g) 13336 10672ndash17640 1023 ns 025

Growth and biomass production

TotDM (kg DMm2) 920 526ndash1521 2318 032

StDM (g DM) 28640 18332ndash46782 2216 ns 039

StN (nm2) 4044 2193ndash8000 2530 054

StNd (n) 3512 2847ndash4680 1030 ns 051

H (cm) 41234 29080ndash52833 1026 ns 058

Db (cm) 2060 1678ndash2601 988 ns 051

Dm (cm) 1606 1225ndash2264 1180 ns 044

Da (cm) 1116 852ndash1391 940 ns 024

Ns non‐significant nd F test and H2 not available for phenology traitsP le 005 P le 001 P le 0001

FABBRINI ET AL | 7

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

H2 frac14 σ2G=ethσ2G thorn σ2ɛTHORNwhere σ2G and σ2ɛ are the genetic and residual variancecomponents respectively which were obtained using therestricted maximum‐likelihood (REML) procedure (Hol-land Nyquist amp Cervantes‐Martinez 2003) H2 was evalu-ated for all the studied traits except for flowering stagesfor which replicated measurements were not viable

3 | RESULTS

31 | Univariate analysis

A significant level of phenotypic variation among the 82different ecotypes of A donax was observed in all

physiological canopy growth and biomass traits measured(Table 2) Flowering phenology of all ecotypes wasrecorded using a new scoring system developed in thisstudy that covers the onset of flowering (Ph1) to the end offlower development (Ph4) Flowering was evaluated at plotscale without replication (Table 1) therefore statisticaldifferences between ecotypes and blocks for flowering phe-nology trait could not be evaluated (Table 2) Averageflowering time of Ph1 and Ph4 at ecotype level were DOY265 (22 September) and DOY 284 (11 October) respec-tively giving an average of 19 days between Ph1 and Ph4Transition between the first 3 phases was rapid for exam-ple taking an average of 4 days from Ph1 to Ph2 and fromPh2 to Ph3 while an average of 11 days were required tocomplete the last flowering phase (Figure 3 Table 2) Due

TABLE 2 General means ecotypic range of variation coefficient of variation (CV ) and level of significance of ecotype (E) and replicate(R) effects for phenological morphological physiological and growth traits in Arundo donax Broad‐sense heritability (H2) for physiologicalmorphological growth and biomass traits See Table 1 for trait abbreviations

Trait Mean Range of variation CV ()

F test

H2E R

Phenology

Ph1 (DOY) 26512 26200ndash27700 140 nd nd nd

Ph2 (DOY) 26967 26200ndash28100 207 nd nd nd

Ph3 (DOY) 27330 26200ndash29000 246 nd nd nd

Ph4 (DOY) 28403 26200ndash30300 493 nd nd nd

Physiology

Rsjuly (scm) 096 057ndash168 2820 028

Rsaugust (scm) 150 080ndash269 2776 ns 062

Δ (permil) 2098 1943ndash2223 242 ns 019

Canopy and leaf morphology

LAI (m2m2) 611 389ndash814 1523 046

LfL (cm) 5732 4825ndash6530 627 ns 017

LfW (cm) 583 461ndash673 856 ns 039

LfL_LfW 990 859ndash1162 750 ns 030

LfA (cm2) 20955 14634ndash26093 1251 ns 030

SLA (cm2g) 13336 10672ndash17640 1023 ns 025

Growth and biomass production

TotDM (kg DMm2) 920 526ndash1521 2318 032

StDM (g DM) 28640 18332ndash46782 2216 ns 039

StN (nm2) 4044 2193ndash8000 2530 054

StNd (n) 3512 2847ndash4680 1030 ns 051

H (cm) 41234 29080ndash52833 1026 ns 058

Db (cm) 2060 1678ndash2601 988 ns 051

Dm (cm) 1606 1225ndash2264 1180 ns 044

Da (cm) 1116 852ndash1391 940 ns 024

Ns non‐significant nd F test and H2 not available for phenology traitsP le 005 P le 001 P le 0001

FABBRINI ET AL | 7

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

to high phenotypic variability there were some notableoutliers on the first day of flowering assessment two earlyflowering ecotypes were in Ph4 (ecotypes ldquo8Prime and ldquo35Prime)Conversely ecotypes ldquo14rdquo ldquo16rdquo and ldquo34rdquo did not flowerat all Of the ecotypes that did reach Ph4 the most delayedwere ldquo30rdquo ldquo37rdquo ldquo50rdquo and ldquo52rdquo (29 October)

A significant ecotypic effect was detected in all traitsexamined whereas the block effect appeared to be statisti-cally significant only for Rsjuly LAI TotDM and StN(Table 2) Traits with highest variation among ecotypeswere Rsjuly and Rsaugust (CV of 2820 and 2776respectively) ranging from 057 to 168 scm in July andbetween 08 and 269 scm in August (Table 2) NotablyRsaugust was measured under moderate drought conditions(ie 36 days of high temperatures 213degC of mean temper-ature and maximum daily average temperature of 295degC)and low rainfall amounts concentrated in 2 days (52 mmin July 29 and 32 mm in August 8) AccordinglyANOVA analysis highlighted significant differencesbetween the two time‐points (P le 0001) with ecotypesthat reacted differently to different times (P le 0001 for

E times T Figure 4ab) In addition significant changes inthe ranking of ecotypes (ρ = 0363 P le 0001) wereobserved There was an inverse relationship between Rsau-gustRsjuly ratio and absolute Rsjuly and ecotypes with thelowest Rsjuly values had higher Rsaugust values up to 3fold than Rsjuly (Figure 4c) The physiological variation inΔ found within ecotypes of our A donax collection ran-ged from 1943permil to 2223permil (Table 2) with a differenceof 28permil observed between the two most divergent eco-types ldquo11rdquo and ldquo30rdquo Moreover the HSD post hoc testrevealed significant differences among ecotypes ldquo11rdquoldquo77rdquo and ldquo82rdquo which showed the lowest Δ values andecotypes ldquo49rdquo and ldquo30rdquo which showed the highest Δ val-ues (Figure 5 Table 2)

Leaf and canopy traits showed medium to high variabil-ity (Table 2) LAI ranged from 389 to 814 m2m2 andwas the most variable of the canopy and leaf morphologygroup of traits Values of LfL ranged from 4825 to6530 cm and a CV of 627 accounting for less than halfof the variability observed for LAI Likewise LfW wasalso less variable than LAI but CV was higher than LfL(856) Leaf shape measured by LfL_LfW rangedbetween 859 and 1162 indicating leaves were approxi-mately 10 times longer than wide

At the end of the growing season differences in TotDMamong ecotypes were significant with means ranging from526 for ecotype ldquo87rdquo to 1521 kg DMm2 for ecotypeldquo49rdquo (Table 2 and Supporting information Table S1) Inaddition TotDM represented one of the most variable traitswith a CV of 2318 that was twice as large as for H(1026) and for diameter measured at different positions(approximately 10 for Db Dm and Da) Similarly highecotype differences were observed for StDM and StN witha slightly higher variation for the latter (2216 vs2530 respectively)

32 | Phenotypic correlations

The linear relationships among traits were studied usingPearsons correlation analysis (Figure 6) Significant posi-tive r values among flowering phases showed that whenthe flowering process began all the other stages followedthrough a cascade process Moreover the date of the latestflowering stage Ph4 were moderately related to LAI(r = 0705 P le 0001) and inversely related to H (r = ‐

FIGURE 3 Phenotypic variation of flowering phenology in abroad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grownin northern Italy Data are expressed in day of the year (DOY) Eachbox is based on ecotype means and represents the quartile below (Q1)and above (Q3) the median value Horizontal bars represent minimumand maximum values Dashed line represents the phenological trendbased on collection means See Table 1 for trait abbreviations

FIGURE 4 Phenotypic variation for stomatal resistance (Rs) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy Ecotypic means (black dots) plusmn SE for Rs measured in (a) July 10 (Rsjuly) and in (b) August 22 (Rsaugust) are shown togetherwith the collection mean (horizontal dashed line) (c) Pairwise trait plot between Rsjuly and RsaugustRsjuly ratio Unique identifiers of eachecotype are written below the dot except identifiers ldquo1rdquo ldquo18rdquo ldquo30rdquo ldquo36rdquo ldquo42rdquo ldquo47rdquo ldquo49rdquo ldquo55rdquo ldquo61rdquo ldquo76rdquo ldquo82rdquo that are written above thedot to avoid labels overlapping See Table 1 for trait abbreviations In (a) and (b) ecotypes were ordered by increasing Rs means Statisticallysignificant differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown above the graph (b) however no statisticalsignificance means are found in (a) In (b) for clarification for each ecotype only the first and the last letters of the group are shown

8 | FABBRINI ET AL

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

FABBRINI ET AL | 9

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

0631 P le 0001) All flowering phases were significantlyand positively correlated with Δ with Ph4 and Δ(r = 0489 P le 0001) showing the highest correlation

Regarding stomatal resistance there were few weak andnegative relationships between Rsjuly and Rsaugust and othertraits such as flowering time Δ LAI Db and Dmwhereas the two traits positively correlated with each other

LfL and LfW were positively correlated and interest-ingly leaf morphology was strongly related to StN with a

higher correlation with LfW (r = minus0610 P le 0001) thanLfL (r = minus0144 P = ns)

Among growth traits as might be expected the strongestcorrelations were detected between StDM and the diameterof the reed especially Db (r = 0851 P le 0001) TotDMon the other hand did not correlate with diameter of thesingle stem but significant correlations were found withStN (r = 0526 P le 0001) and H (r = 0500 P le 0001Figure 6)

FIGURE 5 Phenotypic variation for carbon isotope discrimination (Δ) in a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypesgrown in northern Italy Ecotypic means (black dots) plusmn SE for Δ are shown together with the collection mean (horizontal dashed line) Ecotypeswere ordered by increasing Δ means Significance of differences among ecotypes evaluated with HSD post hoc test (P le 005) are shown abovethe graph See Table 1 for trait abbreviations

FIGURE 6 Heat map showingPearsons correlation coefficients (r) for allphenological physiological morphologicalbiomass and growth traits for a broad Euro‐Mediterranean panel of 82 Arundo donaxecotypes grown in northern Italy SeeTable 1 for trait abbreviations Significantcorrelations are coloured either in red(positive) or blue (negative) hues whilecorrelations that were not significant areindicated by ldquotimesrdquo

10 | FABBRINI ET AL

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

33 | Cluster analysis

Hierarchical cluster analysis was used to group ecotypeswith similar characteristics on a multivariate basis To

reduce the number of variables to be used in the analysis asub‐set of informative traits that is LAI StDM StN StNdH Db and Dm were selected by a multiple regressionanalysis (adjusted determination coefficient R2 = 0753

TABLE 3 Statistical parameters ofmultiple regression analysis ofphysiological morphological and growthtraits on aboveground biomass productionin Arundo donax See Table 1 for traitabbreviations

Trait Estimate SE t value Pr(gt|t|) Sign

Intercept 4091E+00 1536E+00 2663 0008302

Physiology

Rsjuly minus4447E‐02 4409E‐02 minus1009 0314254 ns

Rsaugust 9529E‐04 3632E‐02 0026 0979094 ns

Δ minus3107E‐02 1022E‐02 minus1617 0107334 ns

Canopy and leaf morphology

LAI minus3117E‐02 1406E‐02 minus2217 0027622

LfL 3751E‐02 2395E‐02 1566 0118683 ns

LfW minus4273E‐01 2465E‐01 minus1734 0084343 ns

LfL_LfW minus2746E‐01 1359E‐01 minus2020 0054538 ns

LfA 2098E‐03 2033E‐03 1032 0302982 ns

SLA 1753E‐05 1001E‐03 0017 0986188 ns

Growth and biomass production

StDM 3266E‐03 4631E‐04 7053 214E‐11

StN 3412E‐02 1535E‐03 22223 lt 2E‐16

StNd 2657E‐02 6171E‐03 4305 249E‐05

H minus1444E‐03 6188E‐04 minus2334 0020469

Db 4793E‐02 1773E‐02 2703 0007404

Dm minus6908E‐02 2070E‐02 minus3337 0000989

Da minus1073E‐02 1503E‐02 minus0714 0475818 ns

ns non‐significantP le 005 P le 001 P le 0001

FIGURE 7 Dendrogram of the hierarchical cluster analysis of a broad Euro‐Mediterranean panel of 82 Arundo donax ecotypes grown innorthern Italy The dashed line indicates the stage of clustering selected for determining the number of classes

FABBRINI ET AL | 11

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

P le 0001) as the most important traits influencing biomassyield (Table 3) Wards hierarchical clustering method indi-cated that the 82 ecotypes could be divided into fivegroups (Figures 7 and 8 Table 4) In addition these clus-ters were significantly different for all the traits studied(Table 4) Principal component analysis revealed that thefirst principal component (component 1) accounted for4744 of the variation within the dataset and component2 for an additional 2179 Biometric parameters that isStDM Db and H had high positive loadings on the firstcomponent whereas the second component was weightedmost heavily by TotDM and StN (Table 5)

Clu1 consisted of 17 ecotypes and was located in thecentre of the cluster biplot (Figure 8) This cluster con-tained ecotypes classified as late in flowering (Ph4 at theend of October) medium for growth performance (biomassand morphological traits) preserved low stomatal resistancein the two data points and that showed the highest valuesof Δ (Table 4) However the biplot indicated Clu1 was anintermediate group with strong overlapping regions withClu3 and Clu4 that is areas where memberships areambiguous (Figure 8) The latter two clusters comprising25 and 19 ecotypes respectively were more divergent forTotDM and StNd Clu3 included early‐flowering ecotypes(Ph4 at the end of September) that had big leaves (LfL =5933 cm LfA = 22271 cm2) a high biomass yield(1090 kg DMm2) and a low Δ (2072 permil) Moreoverecotypes in this cluster were characterized by dense(33159 g) and knotty reeds (an average of approximately

39 nodes in each reed) On the other hand Clu4 com-prised accessions that produced few and small stemsreflecting low biomass production (683 kg DMm2)(Table 4) Clu2 and Clu5 were two distinct groups of eco-types that are more differentiated for reed dimensions andStN and as a consequence for LAI Clu2 was a smallcluster of 6 early‐flowering and high performing ecotypesthat form large and tall reeds (Db and H were 2427 mmand 49336 cm respectively) in clumps that were not par-ticularly dense (StN = ~35 per m2) together constitutinglight canopies (LAI = 511 m2m2) Clu5 included 15 late‐flowering ecotypes characterized by small reeds (Db =1803 mm and H = 39511 cm) and dense clumps (StN =~55 per m2 LAI = 625 m2m2) However both Clu2and Clu5 produced similarly high amounts of biomass thatdiffered significantly only from Clu4 Finally Clu2 andClu5 showed similar Δ (2097 permil)

34 | Heritability of leaf morphologyphysiology and biomass traits

H2values were generally highest in biomass (averageH2 = 044) and physiology (average H2 = 045) comparedto morphology traits (average H2 = 032) (Table 2) For akey physiology trait such as Rsaugust 62 of the pheno-typic variance was partitioned to genetic variance whereaslow H2 was observed for Δ (H2 = 019) Several biomasstraits (TotDM StDM StN StNd H Db Dm and Da) hadmoderate to high H2 values (024ndash058) and H had thehighest H2 value (058) whereas only LAI and LfWshowed moderate values of H2 among morphology traits(039ndash046 respectively)

4 | DISCUSSION

In this study we present a comparative analysis of a largenumber of A donax ecotypes collected from differentEuro‐Mediterranean regions throughout most of the spe-ciesrsquo native range At the end of the third one‐year coppicerotation when the species reached its maturity phase dis-tinct differences expressed in the biomass related character-istics in field conditions were demonstrated using 82ecotypes Despite the reported low genetic diversity asrevealed by molecular analysis of clonal populations col-lected in America and Europe (Ahmad et al 2008 Har-dion et al 2012 2014 Mariani et al 2010 Pilu et al2014) clonal selection can take advantage of inheritableand promising phenotypic differences that have beendescribed in traits such as number of culms culm diameterand height (Cosentino et al 2006 Pilu et al 2014)

Our results highlighted significant differences amongecotypes for all phenotypic traits studied Concerning phe-nology our collection included ecotypes with large

FIGURE 8 Representation of the principal components analysisshowing interrelations among a broad Euro‐Mediterranean panel of 82Arundo donax ecotypes grown in northern Italy The representativeclustering plot for all ecotypes showed their distribution into fivemajor clusters (circle ecotypes into cluster 1 triangle ecotypes intocluster 2 plus ecotypes into cluster 3 cross ecotypes into cluster 4diamond ecotypes into cluster 5)

12 | FABBRINI ET AL

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

variation in the late flowering stage Flowering date is con-sidered one of the most important selection criteria whenbreeding perennial grasses for cellulosic ethanol (JakobZhou amp Paterson 2009) In addition flowering phenologyand height growth patterns were shown to be significantlyassociated with functional traits such as relative growthrate leaf mass per area and hollow ratio in herbaceousgrassland species (Sun amp Frelich 2011) Previously Cosen-tino et al (2006) scored flowering in 39 ecotypes collectedin southern Italy (35 in Sicily region and 4 in Calabriaregion) and at Ph4 Surprisingly the variability observed inPh4 reported by Cosentino et al (2006) which showed a41 days difference between the two most divergent eco-types at the end of the second growing season perfectly

mirrors the variability reported here (Figure 3) Perhaps themost interesting is the finding of phenotypic correlationsthat highlighted a relationship between flowering time andsome morphological traits such as H and SLA Althoughwe compared within‐species variation a statistically signifi-cant and positive relationship among these traits has alsobeen observed in previous studies (Bolmgren amp Cowan2008 Sun amp Frelich 2011) where taller herbaceous grass-land and perennial species flowered earlier than shorterones As expected positive correlations between floweringphenology phases and SLA were observed Both earlyflowering time and reduced SLA are associated with plantshighly adapted to arid environments (Kooyers 2015) In Adonax vegetative and reproductive phases often end at the

TABLE 4 Mean cluster values and coefficient of variation (CV ) of phenological physiological morphological and biomass and growthtraits in Arundo donax The number of ecotypes in each cluster (Clu) is enclosed by curly brackets CV was calculated as the ratio of thestandard deviation among ecotypic values in the cluster to the mean cluster value reported as a percentage within round bracket For each traitsignificant differences among clusters (F test) are indicated by asterisks Different letters among clusters are related to LSD post hoc test(P le 005) See Table 1 for trait abbreviations

Trait F test Clu1 17 Clu2 6 Clu3 25 Clu4 19 Clu5 15

Phenology

Ph1 (DOY) ndash 26671 (156) 26200 (0) 26345 (096) 26457 (104) 26623 (164)

Ph2 (DOY) ndash 27259 (152) 26520 (067) 26633 (186) 26828 (179) 27300 (239)

Ph3 (DOY) ndash 27800 (228) 26833 (056) 26841 (180) 27174 (202) 27900 (151)

Ph4 (DOY) ndash 29635 (311) 27160 (129) 27338 (384) 28458 (514) 28986 (297)

Physiology

Rsjuly (scm) 076 (1756)b 094 (2501)ab 096 (2744)a 107 (3176)a 106 (2001)a

Rsaugust (scm) 124 (1834)c 129 (1297)bc 163 (2620)a 162 (2939)a 149 (2690)ab

Δ (permil) 2126 (194)a 2097 (282)abc 2072 (272)c 2105 (245)ab 2097 (203)bc

Canopy and leaf morphology

LAI (m2m2) 701 (598)a 511 (1234)d 571 (1556)cd 599 (1546)bc 625 (1026)b

LfL (cm) 5705 (699)b 5637 (338)b 5933 (493)a 5585 (701)b 5652 (519)b

LfW (cm) 605 (681)a 597 (630)a 597 (775)a 587 (749)a 530 (811)b

LfL_LfW 945 (400)c 952 (633)c 1003 (689)b 956 (714)c 1072 (485)a

LfA (cm2) 21368 (1306)ab 21392 (795)ab 22271 (1008)a 20301 (1250)bc 19112 (1193)c

SLA (cm2g) 13818 (836)c 12141 (573)a 12935 (1067)ab 13449 (1036)bc 13696 (1029)c

Growth and biomass production

TotDM (kg DMm2) 883 (1093)b 972 (1225)ab 1090 (2068)a 683 (1452)c 986 (1601)ab

StDM (g DM) 28953 (1039)c 40621 (859)a 33159 (1272)b 24594 (1154)d 21584 (1029)e

StN (nm2) 3884 (1564)b 3565 (1160)bc 3890 (2046)b 3365 (1420)c 5510 (1840)a

StNd (n) 3478 (763)b 3499 (487)b 3895 (847)a 3333 (715)bc 3209 (775)c

H (cm) 39286 (385)c 49336 (492)a 44552 (591)b 37801 (875)d 39511 (500)c

Db (cm) 2159 (486)b 2427 (496)a 2151 (564)b 1947 (523)c 1803 (557)d

Dm (cm) 1703 (559)d 2021 (648)e 1615 (832)c 1542 (601)b 1396 (658)a

Da (cm) 1189 (678)a 1201 (685)a 1102 (977)b 1117 (657)b 1029 (927)c

ns non‐significantP le 005 P le 001 P le 0001

FABBRINI ET AL | 13

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

same period (from early to late fall Pilu et al 20122013) therefore late flowering together with high SLAmay be indicative of improved growth capacity Likewisethere were strong positive correlations between floweringtime and LAI reflecting greater canopy soil coverage forthe ecotypes that had longer growing seasons Ecotypeswith high LAI had longer vegetative periods indicatingthat late flowering may be advantageous under continuousmild water deficit as it allows stress acclimatization overtime but not under severe drought as it may imply thatthey could not finish their phenological cycle

Gradual water reduction in the soil is common in natu-ral ecosystems and one of the earliest responses to thisdepletion is by stomata (Bogeat‐Triboulot et al 2007Chaves Flexas amp Pinheiro 2009 Harfouche Meilan ampAltman 2014 Plomion et al 2006) There is also strongevidence that stomatal limitation is preponderant at mildlevels of abiotic stresses like drought (reviewed by Cornicamp Fresneau 2002) We have shown a significant variationamong ecotypes in mid‐day abaxial stomatal resistance anda clear ecotype‐dependent response to a period of watershortage as highlighted by the E times T interaction analysisIndeed ecotypes presenting the lowest values of Rsjuly (ieusing more water through transpiration) had much higherRsaugust (as they might have exhausted the available water)than others and thus showed a higher RsaugustRsjuly forwhich a negative relationship between both parameters isapparent (Figure 4) However this response varies amongecotypes Ideally the perfect ecotype could be one showinglow Rsjuly (allowing high rate of photosynthesis during theseason when there is still some water in the soil) and eitherhaving largely increased Rsaugust (saving water when it ismuch less available) or keeping low Rsaugust (meaning thatit is still able to uptake water probably due to an efficientroot system) It would therefore be interesting in furtherstudies to check drought tolerance of ecotypes positioned

at different places along the RsaugustRsjuly versus Rsjulyrelationship The fact that Δ is only marginally related tostomatal resistance suggests that water use efficiency inArundo ecotypes is more dependent on variability in photo-synthetic capacity rather than on stomatal physiology(OLeary 1988) for which further studies on the variabilityand H2 of photosynthetic capacity would be also of interestfor selection purposes We also observed a negative rela-tionship between stomatal activity and LAI as already seenin various broadleaved and coniferous forest stands underdifferent climates (Granier Loustau amp Breacuteda 2000)

The selection of suitable ecotypes for sustainable ligno-cellulosic biomass production has to be based on key char-acteristics among which carbon isotope composition couldbe a relevant trait Mean Δ observed in this study waslower than the one observed by Nackley Vogt and Kim(2014) (Δ = ~24permil) in a growth chamber pot experimentIn our study we observed a difference of 28permil betweenthe two most divergent ecotypes in terms of Δ In a collec-tion of chickpea (Cicer arietinum) ranges of variation of31permil and 27permil were observed among genotypes sub-jected to terminal drought condition (Krishnamurthy et al2013) In cocksfoot (Dactylis glomerata) tall fescue (Fes-tuca arundinacea) and reed canary grass (Phalaris arundi-nacea) grown under water and water‐limited conditions anoverall difference of 25permil was observed between the twotreatments (Maringrtensson et al 2017) Plants of sunflower(Helianthus annuus) subjected to two drought treatments(ie progressive water‐stress and unvaried stressful condi-tion) showed a significant difference between treatmentswith a Δ of 286permil and 271permil in stressed and no‐stressedplants respectively (Adiredjo et al 2014) As supportedby these papers the range of variation observed in ourstudy is of high interest and could be the result of the highvariability in the ecology of native sites where A donaxecotypes were collected The small CV observed in Δ notexceeding 25 was several times smaller than the CVhere estimated for other physiological morphological andgrowth related traits However caution should be takenwhen interpreting significances of low Δ variations Indeeda supposed low CV value of 242 that referred to a gen-eral mean value of 2098permil would imply a large variationof Δ of about 28permil Such a variation is physiologicallyrelevant suggesting good levels of adaptation for Δ in ourA donax panel Correlation analysis found weak and nega-tive relationship between Δ and TotDM moreover clusterdifferentiation highlighted that more productive ecotypesshowed lower Δ values The consistently negative associa-tions between Δ and WUE might suggest that crop yieldand Δ might also be negatively related (Condon RichardsRebetzke amp Farquhar 2004) However large variance inflowering time and reed height as here observed couldaffect the strength and direction of correlation between Δ

TABLE 5 Loadings of the sub‐set of informative traits in the firsttwo principal components of the PCA accounting for 6923 of theecotype variability in Arundo donax See Table 1 for traitabbreviations

Component 1 Component 2Coefficients Coefficients

StDM 0494 0031

Db 0457 0233

H 0429 minus0268

Dm 0377 0322

StNd 0317 minus0216

TotDM 0226 minus0601

LAI minus0173 0189

StN minus0190 minus0569

14 | FABBRINI ET AL

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

and total yield (Condon et al 2004) Positive correlationsamong flowering phases and Δ were found in our workand this result reflects the antagonist behaviours betweendrought escape (ie early flowering) and avoidance (iehigher WUE ie lower Δ) strategies (Kooyers 2015)Finally negative correlations between Rs and Δ hereobserved resulted as expected results probably due tolower CO2 inside leaf concentration in response to greaterstomatal resistance (Farquhar et al 1989)

Canopy architecture affects growth performance of treespecies used in SRC plantations (Dillen Marron SabattiCeulemans amp Bastien 2009) therefore LAI was measuredto evaluate the variation in canopy density among the largeset of ecotypes analysed here Other leaf traits affectingcanopy structure and light capture have also been charac-terized in the perennial biomass crop genus Miscanthus(Robson et al 2013) The range of LAI values observed inthe current study is consistent with recent findings Theaverage value of 611 m2m2 is similar to that reported byCosentino Scordia Sanzone Testa and Copani (2014) ina trial carried out in southern Italy using a local ecotype atthe third year of growth under moderate water supply(690 m2m2) Likewise the maximum value of 814 m2m2

is in agreement with the previous recorded LAI of about9 m2m2 for a local ecotype at its third year of growth innorthern Italy (Ceotto et al 2013) Finally the value of389 m2m2 measured in the ecotype with the lightestcanopy is comparable with the one estimated by Nassi oDi Nasso Roncucci Triana Tozzini and Bonari (2011) incentral Italy but in a stand at the seventh one‐year coppicerotation Notably except for Cosentino et al (2014) thecomparison of LAI values was carried out during the sameperiod of time that is the middle of July because LAIchanges through the season and in A donax it reaches themaximum value in early August and tends to decline untilthe end of the season (Ceotto et al 2013) Leaf dimen-sions described by LfL and LfW were in the range ofthose reported by Hardion et al (2012) where A donaxplant material from 16 sites across the Mediterranean basinwas investigated Finally the observed average value ofSLA was consistent with data collected in wild populationsof A donax along moisture gradients in southern USA(Watts amp Moore 2011) Interestingly SLA of A donaxwas suggested as an indicator of a poor adaptation to fre-quent or prolonged submerged conditions (Mommer Lens-sen Huber Visser amp De Kroon 2006)

Due to its high biomass yields combined with lowinput requirements A donax has been proposed as anherbaceous model lignocellulosic crop for Mediterraneanenvironments (Angelini et al 2009 Borin et al 2013Nassi o Di Nasso et al 2011 2013) However the bio-mass data from A donax in Italy differ considerably In astudy by Cosentino et al (2006) in southern Italy a mean

yield of 11 Mg DMha at the end of the first one‐year cop-pice rotation and of 22 Mg DMha in the second year wererecorded with a peak of 342 Mg DMha reached by themost productive clone at the end of the second growingseason In central‐southern Italy Fagnano et al (2015)obtained a mean yield of about 15 Mg DMha in a 9‐yeartrial carried out in an experimental site characterized bysevere climatic conditions and a clayey soil textureWhereas in northern Italy a 12‐year field study conductedby Angelini et al (2009) highlighted an average yield of38 Mg DM haminus1 yrminus1 with a peak of about 50 Mg DMhaat the end of the second and third years of growth undernon‐limiting conditions of water and nutrient availabilitySimilarly in another experimental site in northern Italywith optimal water and N inputs Borin et al (2013)showed a high biomass yield of about 100 Mg DMha atthe second and third year of growth being the highest Adonax biomass yield ever recorded in Italy although thebiomass yields from small‐scale experimental plots may behigher than those achievable at commercial scale Thisresult was supported by our data the highest yielding eco-types in our collection were more comparable with theyields of 150 Mg DMha reported in wild fields in Califor-nia and India (Giessow Casanova Lecler MacArthur ampFleming 2011 Sharma Kushwaha amp Gopal 1998 Spen-cer Liow Chan Ksander amp Getsinger 2006) Howeverour yield estimates are likely to be overestimates becauseour results were obtained from single plots in a commongarden collection and because we were unable to avoid thebias of increased yields often seen in plants growing at theedges of plots This is also not unexpected as biomassyields in commercial‐scale fields might well be lower thanthose obtained in small research plots (Ping et al 2014Verhulst Cox amp Govaerts 2013) For our purpose thehomogeneous plot design of the common garden experi-ment was successful Comparing biomass components thevariability of H measured in the current study was perfectlycomparable to the range for A donax of between approxi-mately 340 and 600 cm including panicle (20ndash70 cm)described by Hardion et al (2012) The mean H valuereported here was higher than those observed in southernand central Italy of 339 cm and 286 cm in trials at the sec-ond and seventh year respectively (Cosentino et al 2006Nassi o Di Nasso et al 2011) As compared to the experi-ment carried out in Sicily (Cosentino et al 2006) our col-lection had a slightly lower average D in the basal middleand upper part of the stem (Db 2060 vs 2140 mm Dm1606 vs 1880 mm Da 1116 vs 1540 mm in SAV andSicily respectively) while double stem density (4044 vs2140 per m2) and stem weight (28640 vs 12760 per m2)were registered

Correlation analyses in A donax indicated that high drybiomass yield was largely due to plant apical growth and

FABBRINI ET AL | 15

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

stem density (Angelini et al 2009 Cosentino et al 2006)The low but statistically significant correlations observedbetween biomass production and various growth traits (ieStDM StN StNd and H) highlighted biomass yield as acomplex trait made up of multiple simple traits (Clifton‐Brown Robson amp Allison 2008)

H2 assessments in this study showed that biomass traitshad moderate to high H2 estimates whereas morphologyand ecophysiology trait H2 estimates ranged from low tohigh Low H2 estimates were observed primarily in physi-ology traits (Rs Δ) which is most likely related to somephenotypic plasticity and reflecting the environment inwhich the trait was sampled Notably the wide change ofH2 estimates for Rsjuly (H2 = 028) and Rsaugust(H2 = 062) is to be expected because stomatal resistanceis an instantaneously measured physiological trait related toleaf gas exchange (Ackerly et al 2000) Other studies inA donax have also observed moderate H2 in biomass andmorphology traits (Cosentino et al 2006 Pilu et al2014) Individual traits with higher H2 (morphology bio-mass) tended to be intercorrelated Our H2 results for bio-mass traits (032 le H2 le 039) are slightly lower thanother perennial grasses such as switchgrass(045 le H2 le 050 Rose Das amp Taliaferro 2008 Jiang etal 2014) and miscanthus (034 le H2 le 055 Slavov et al2013 Gifford Chae Swaminathan Moose amp Juvik2015)

Our multivariate analysis demonstrated that biomass andgrowth traits StDM StN StNd H Db and Dm as well asLAI were the most indicative of dry biomass production inour panel of 82 ecotypes These growth‐related traitsallowed us to identify similar groups and to select promis-ing individual ecotypes Cluster analysis highlighted clus-ters 2 3 and 5 as the most interesting for biomass yieldpotential and that the high biomass producing ecotypesadopted different growth strategies For example ecotypesin Clu3 deployed heavy and large leaves in a lightercanopy Ecotypes in the lower side of the biplot were char-acterized by a higher number of reeds and higher biomassproduction compared with individuals localized in theupper part (Figure 8) The two most divergent groupsClu2 and Clu5 contained ecotypes that exhibited signifi-cant differences in reed diameter and number and as aconsequence in LAI High‐biomass ecotypes with largeand tall stems organized into less‐dense and light clumpswere grouped in Clu2 Conversely ecotypes grouped inClu5 had small reeds forming dense clumps Clusters 2 3and 5 correspond to 56 of our collection (46 ecotypes)thus an extensive multi‐year multi‐location examination ofphenotypic variation of these 46 ecotypes among many dif-ferent types of traits could provide considerable insight intokey traits favouring the production of biomass yield in dif-ferent cropping environments Finally the most promising

ecotypes belong to these three clusters that showed the bestcombinations of biomass determinants such as H StNStNd and LAI Although Δ is not a significant biomassdeterminant Clusters 2 3 and 5 showed lower Δ than theother two and this underlines its importance In somegrowing areas in semi‐arid environments biomass potentialmay not be achieved due to water shortage and hencewater use efficiency may be a more relevant trait Thesetraits should be taken into consideration for both clonalselection and systems and synthetic biology approachesWhile selection for these traits is warranted targetedgenetic improvement of A donax bioenergy feedstock inthese critical morphological and biomass traits is of greatimportance However to maximize the efficiency of bio-mass conversion and reduce biomass recalcitrance in Adonax future breeding efforts should also target cell wallproperties optimized for saccharification and fermentation(Scordia Cosentino Lee amp Jeffries 2011 2012)

Creating a successful bioenergy program from lignocel-lulosic feedstocks is a major challenge and will require sig-nificant research effort to maximize plant biomassproduction Our study provides an essential stepping stonetowards this goal The results presented here demonstratehigh variation in phenotypic traits among a large collectionof A donax ecotypes spanning most of the speciesrsquo naturalrange The assembled panel also represents a suitableresource to screen for improved bioconversion efficiencyusing a similar approach to improve the already favourablecomprehensive utilization index for A donax as discussedby Kou et al (2017) The study also highlights the avail-ability of different ideotypes (described by Donald (1968)as a biological model which is expected to perform orbehave in a predictable manner within a defined environ-ment) of A donax that achieve high yield through differentcombinations of traits and it will be interesting to see ifthese ideotypes differ in their ability to generate high yieldacross a range of environments Extensive assessments oftraits underscored the consistency observed in phenotypictraits and to our knowledge present the most comprehen-sive assessments of biomass traits in A donax to dateEfforts are under way to exploit genomic tools in A donaxto identify loci responsible for the observed trait variation

ACKNOWLEDGEMENTS

This work was supported by grants from the EuropeanCommissions Seventh Framework Programme (FP72012‐2017) under the Grant Agreement Number FP7‐311929(WATBIO) and the Brain Gain Program (Rientro dei cer-velli) of the Italian Ministry of Education University andResearch (AH) For the help with the field work andassistance with data entry we thank Muriel Gaudet andPaolo Latini

16 | FABBRINI ET AL

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

CONFLICT OF INTEREST

The authors declare no competing business interests orrelationships that could be viewed as a conflict of interest

ORCID

Paul Robson httporcidorg0000-0003-1841-3594Antoine Harfouche httporcidorg0000-0003-3998-0694

REFERENCES

Ackerly D D Dudley S A Sultan S E Schmitt J Coleman JS Linder C R hellip Lechowicz M J (2000) The evolution ofplant ecophysiological traits Recent advances and future direc-tions BioScience 11 979ndash995httpsdoiorg1016410006-3568(2000)050[0979TEOPET]20CO2

Adiredjo A L Navaud O Muntildeos S Langlade N B LamazeT amp Grieu P (2014) Genetic control of water use efficiencyand leaf carbon isotope discrimination in sunflower (Helianthusannuus L) subjected to two drought scenarios PLoS ONE 9e101218 httpsdoiorg101371journalpone0101218

Ahmad R Liow P S Spencer D F amp Jasieniuk M (2008)Molecular evidence for a single genetic clone of invasive Arundodonax in the United States Aquatic Botany 88 113ndash120 httpsdoiorg101016jaquabot200708015

Amaducci S amp Perego A (2015) Field evaluation of Arundo donaxclones for bioenergy production Industrial Crops and Products75 122ndash128 httpsdoiorg101016jindcrop201504044

Angelini L G Ceccarini L amp Bonari E (2005) Biomass yieldand energy balance of giant reed (Arundo donax L) cropped incentral Italy as related to different management practices Euro-pean Journal of Agronomy 22 375ndash389 httpsdoiorg101016jeja200405004

Angelini L G Ceccarini L Nassi o Di Nasso N amp Bonari E(2009) Comparison of Arundo donax L and Miscanthus times gigan-teus in a long‐term field experiment in central Italy Analysis ofproductive characteristics and energy balance Biomass and Bioen-ergy 33 635ndash643 httpsdoiorg101016jbiombioe200810005

Aravanopoulos F A (2010) Contribution of leaf morphometrics inthe study of genetic entries in Salix L Electronic Journal of PlantBreeding 1 1320ndash1328

Barney J N amp Ditomaso J M (2008) Nonnative species andbioenergy Are we cultivating the next invader BioScience 5864ndash70 httpsdoiorg101641B580111

Bogeat‐Triboulot M B Brosche M Renaut J Jouve L LeThiec D Fayyaz P hellip Dreyer E (2007) Gradual soil waterdepletion results in reversible changes of gene expression proteinprofiles ecophysiology and growth performance in Populuseuphratica a poplar growing in arid regions Plant Physiology143 876ndash892

Bolmgren K amp Cowan P D (2008) Time‐size tradeoffs A phylo-genetic comparative study of flowering time plant height and seedmass in a north‐temperate flora Oikos 117 424ndash429 httpsdoiorg101111j20070030-129916142x

Borin M Barbera A C Milani M Molari G Zimbone S M ampToscano A (2013) Biomass production and N balance of giant

reed (Arundo donax L) under high water and N input in Mediter-ranean environments European Journal of Agronomy 51 117ndash119 httpsdoiorg101016jeja201307005

Bylesjouml M Segura V Soolanayakanahally R Y Rae A MTrygg J Gustafsson P hellip Street N R (2008) LAMINA Atool for rapid quantification of leaf size and shape parametersBMC Plant Biology 8 82 httpsdoiorg1011861471-2229-8-82

Ceotto E amp Di Candilo M (2010) Shoot cuttings propagation ofgiant reed (Arundo donax L) in water and moist soil The pathforward Biomass and Bioenergy 34 1614ndash1623 httpsdoiorg101016jbiombioe201006002

Ceotto E Di Candilo M Castelli F Badeck F‐W Rizza FSoave C hellip Marletto V (2013) Comparing solar radiationinterception and use efficiency for the energy crops giant reed(Arundo donax L) and sweet sorghum (Sorghum bicolor LMoench) Field Crops Research 149 159ndash166 httpsdoiorg101016jfcr201305002

Charrad M Ghazzali N Boiteau V amp Niknafs A (2013)NbClust An examination of indices for determining the Numberof Clusters R package version 14

Chaves M M Flexas J amp Pinheiro C (2009) Photosynthesisunder drought and salt stress Regulation mechanisms from wholeplant to cell Annals of Botany 103 551ndash560 httpsdoiorg101093aobmcn125

Clifton‐Brown J C Robson P R amp Allison G G (2008) Mis-canthus Breeding our way to a better future Aspects of AppliedBiology 90 199ndash206

Condon A G Richards R A Rebetzke G J amp Farquhar G D(2004) Breeding for high water‐use efficiency Journal of Experi-mental Botany 55 2447ndash2460 httpsdoiorg101093jxberh277

Cornic G amp Fresneau C (2002) Photosynthetic carbon reductionand carbon oxidation are the main electron sinks for photosystemII activity during a mild drought Annals of Botany 89 887ndash894httpsdoiorg101093aobmcf064

Corno L Pilu R amp Adani F (2014) Arundo donax L A non‐food crop for bioenergy and bio‐compound production Biotech-nology Advances 32 1535ndash1549 httpsdoiorg101016jbiotechadv201410006

Cosentino S L Copani V DAgosta G M Sanzone E amp Manti-neo M (2006) First results on evaluation of Arundo donax Lclones collected in Southern Italy Industrial Crops and Products23 212ndash222 httpsdoiorg101016jindcrop200506004

Cosentino S L Copani V Patanegrave C Mantineo M amp DAgostaG M (2008) Agronomic energetic and environmental aspects ofbiomass energy crops suitable for Italian environments ItalianJournal of Agronomy 2 81ndash95 httpsdoiorg104081ija200881

Cosentino S L Copani V Scalici G Scordia D amp Testa G(2015) Soil erosion mitigation by perennial species underMediterranean environment Bioenergy Research 8 1538ndash1547httpsdoiorg101007s12155-015-9690-2

Cosentino S L Scordia D Sanzone E Testa G amp Copani V(2014) Response of giant reed (Arundo donax L) to nitrogen fer-tilization and soil water availability in semi‐arid Mediterraneanenvironment European Journal of Agronomy 60 22ndash32 httpsdoiorg101016jeja201407003

Dillen S Y Marron N Sabatti M Ceulemans R amp Bastien C(2009) Relationships among productivity determinants in twohybrid poplar families grown during three years at two contrasting

FABBRINI ET AL | 17

sites Tree Physiology 29 975ndash987 httpsdoiorg101093treephystpp036

Donald C M (1968) The breeding of crop ideotypes Euphytica 17385ndash403 httpsdoiorg101007BF00056241

Fagnano M Impagliazzo A Mori M amp Fiorentino N (2015)Agronomic and environmental impacts of giant reed (Arundodonax L) Results from a long‐term field experiment in hilly areassubject to soil erosion BioEnergy Research 8 415ndash422 httpsdoiorg101007s12155-014-9532-7

Farquhar G D Ehleringer J R amp Hubick K T (1989) Carbonisotope discrimination and photosynthesis Annual Review of PlantPhysiology and Plant Molecular Biology 40 503ndash537 httpsdoiorg101146annurevpp40060189002443

Farrar K Bryant D N Turner L Gallagher J A Thomas AFarrell M hellip Donnison I S (2012) Breeding for bio‐ethanolproduction in Lolium perenne L Association of allelic variationwith high water‐soluble carbohydrate content BioenergyResearch 5 149ndash157 httpsdoiorg101007s12155-011-9156-0

Giessow J Casanova J Lecler R MacArthur R amp Fleming G(2011) Arundo donax ndash distribution and impact report BerkeleyCA California Invasive Plant Council 240 pp

Gifford J M Chae W B Swaminathan K Moose S P amp JuvikJ A (2015) Mapping the genome of Miscanthus sinensis forQTL associated with biomass productivity Global Change Biol-ogy Bioenergy 7 797ndash810 httpsdoiorg101111gcbb12201

Granier A Loustau D amp Breacuteda N (2000) A generic model offorest canopy conductance dependent on climate soil water avail-ability and leaf area index Annals of Forest Science 57 755ndash765 httpsdoiorg101051forest2000158

Haddadchi A Gross C L amp Fatemi M (2013) The expansion ofsterile Arundo donax (Poaceae) in southeastern Australia isaccompanied by genotypic variation Aquatic Botany 104 153ndash161 httpsdoiorg101016jaquabot201207006

Hardion L Verlaque R Baumel A Juin M amp Vila B (2012)Revised systematics of Mediterranean Arundo (Poaceae) based onAFLP fingerprints and morphology Taxon 61 1217ndash1226

Hardion L Verlaque R Saltonstall K Leriche A amp Vila B(2014) Origin of the invasive Arundo donax (Poaceae) A trans‐Asian expedition in herbaria Annals of Botany 114 455ndash462httpsdoiorg101093aobmcu143

Harfouche A Meilan R amp Altman A (2014) Molecular and phys-iological responses to abiotic stress in forest trees and their rele-vance to tree improvement Tree Physiology 34 1181ndash1198httpsdoiorg101093treephystpu012

Holland J B Nyquist W E amp Cervantes‐Martinez C T (2003)Estimating and interpreting heritability for plant breeding Anupdate Plant Breeding Reviews 22 9ndash112

Jakob K Zhou F S amp Paterson A (2009) Genetic improvementof C4 grasses as cellulosic biofuel feedstocks In Vitro Cellularand Developmental Biology‐Plant 45 291ndash305 httpsdoiorg101007s11627-009-9214-x

Jiang Q Webb S L Yesudas C R Bhandari H S Narasimhamoor-thy B Bouton J H amp Saha M C (2014) Variance componentsand heritability of biomass yield in switchgrass(Panicum virgatumL) grown in the Southern Great Plains Field Crops Research 168148ndash155 httpsdoiorg101016jfcr201407016

Kooyers N J (2015) The evolution of drought escape and avoidancein natural herbaceous populations Plant Science 234 155ndash162httpsdoiorg101016jplantsci201502012

Kottek M Grieser J Beck C Rudolf B amp Rubel F (2006)World map of the Koumlppen‐Geiger climate classification updatedMeteorologische Zeitschrift 15 259ndash263 httpsdoiorg1011270941-294820060130

Kou L Song Y Zhang X amp Tan T (2017) Comparison of fourtypes of energy grasses as lignocellulosic feedstock for the pro-duction of bio‐ethanol Bioresource Technology 241 424ndash429httpsdoiorg101016jbiortech201704078

Krishnamurthy L Kashiwagi J Tobita S Ito O Upadhyaya HD Gowda C L L hellip Varshney R K (2013) Variation in car-bon isotope discrimination and its relationship with harvest indexin the reference collection of chickpea germplasm FunctionalPlant Biology 40 1350ndash1361 httpsdoiorg101071FP13088

Lebrija‐Trejos E Peacuterez‐Garciacutea E A Meave J A Bongers F ampPoorter L (2010) Functional traits and environmental filteringdrive community assembly in a species‐rich tropical system Ecol-ogy 91 386ndash398 httpsdoiorg10189008-14491

Lewandowski I Scurlock J M O Lindvall E amp Christou M(2003) The development and current status of perennial rhizoma-tous grasses as energy crops in the US and Europe Biomass andBioenergy 25 335ndash361 httpsdoiorg101016S0961-9534(03)00030-8

Mariani C Cabrini R Danin A Piffanelli P Fricano A Gomar-asca S hellip Soave C (2010) Origin diffusion and reproductionof the giant reed (Arundo donax L) A promising weedy energycrop Annals of Applied Biology 157 191ndash202 httpsdoiorg101111j1744-7348201000419x

Maringrtensson L‐M Carlsson G Prade T Koslashrup K Laeligrke P Eamp Jensen E S (2017) Water use efficiency and shoot biomassproduction under water limitation is negatively correlated to 13Cdiscrimination in the C3 grasses Dactylis glomerata Festuca arun-dinacea and Phalaris arundinacea Plant Physiology and Bio-chemistry 113 1ndash5 httpsdoiorg101016jplaphy201701021

Mommer L Lenssen J P M Huber H Visser E J W amp DeKroon H (2006) Ecophysiological determinants of plant perfor-mance under flooding A comparative study of seven plant fami-lies Journal of Ecology 94 1117ndash1129 httpsdoiorg101111j1365-2745200601175x

Nackley L L Vogt K A amp Kim S H (2014) Arundo donaxwater use and photosynthetic responses to drought and elevatedCO2 Agricultural Water Management 136 13ndash22 httpsdoiorg101016jagwat201401004

Nassi o Di Nasso N Roncucci N amp Bonari E (2013) Seasonaldynamics of aboveground and belowground biomass and nutrientaccumulation and remobilization in giant reed (Arundo donax L)A three‐year study on marginal land Bioenergy Research 6 725ndash736 httpsdoiorg101007s12155-012-9289-9

Nassi o Di Nasso N Roncucci N Triana F Tozzini C ampBonari E (2011) Productivity of giant reed (Arundo donax L)and miscanthus (Miscanthus times giganteus Greef et Deuter) asenergy crops Growth analysis Italian Journal of Agronomy 6e22 httpsdoiorg104081ija2011e22

Nsanganwimana F Marchand L Douay F amp Mench M (2014)Arundo donax L a candidate for phytomanaging water and soilscontaminated by trace elements and producing plant‐based feed-stock International Journal of Phytoremediation 16 982ndash1017httpsdoiorg101080152265142013810580

OLeary M H (1988) Carbon isotopes in photosynthesis BioS-cience 38 328ndash336 httpsdoiorg1023071310735

18 | FABBRINI ET AL

Pilu R Bucci A Badone F C amp Landoni M (2012) Giant reed(Arundo donax L) A weed plant or a promising energy cropAfrican Journal of Biotechnology 11 9163ndash9174

Pilu R Cassani E Landoni M Cerino Badone F Passera ACantaluppi E hellip Adani F (2014) Genetic characterization ofItalian giant reed (Arundo donax L) clones collection Exploitingclonal selection Euphytica 196 169ndash181 httpsdoiorg101007s10681-013-1022-z

Pilu R Manca A amp Landoni M (2013) Arundo donax as anenergy crop Pros and cons of the utilization of this perennialplant Maydica 58 1ndash6

Ping H Bransby D I amp van Santen E (2014) Long‐term biomassyields of giant reed mimosa and switchgrass in Alabama Biofu-els Bioproducts amp Biorefining 8 59ndash66 httpsdoiorg101002bbb1437

Pison G Struyf A amp Rousseeuw P J (1999) Displaying a clus-tering with CLUSPLOT Computational Statistics amp Data Analy-sis 30 381ndash392 httpsdoiorg101016S0167-9473(98)00102-9

Plomion C Lalanne C Claverol S Meddour H Kohler ABogeat‐Triboulot M B hellip Bonneu M (2006) Mapping the pro-teome of poplar and application to the discovery of drought‐stressresponsive proteins Proteomics 6 6509ndash6527 httpsdoiorg101002(ISSN)1615-9861

Robson P Jensen E Hawkins S White S R Kenobi K Clif-ton‐Brown J hellip Farrar K (2013) Accelerating the domestica-tion of a bioenergy crop Identifying and modelling morphologicaltargets for sustainable yield increase in Miscanthus Journal ofExperimental Botany 64 4143ndash4155 httpsdoiorg101093jxbert225

Rose L W IV Das M K amp Taliaferro C M (2008) Estimationof genetic variability and heritability for biofuel feedstock yield inseveral populations of switchgrass Annals of Applied Biology152 11ndash17 httpsdoiorg101111j1744-7348200700186x

Rossa B Tuumlffers A V Naidoo G amp von Willert D J (1998)Arundo donax L (Poaceae) ndash a C3 species with unusually highphotosynthetic capacity Plant Biology 111 216ndash221

Saikia R Chutia R S Kataki R amp Pant K K (2015) Perennialgrass (Arundo donax L) as a feedstock for thermo‐chemical con-version to energy and materials Bioresource Technology 188265ndash272 httpsdoiorg101016jbiortech201501089

Saacutenchez E Scordia D Lino G Arias C Cosentino S L ampNogueacutes S (2015) Salinity and water stress effects on biomass pro-duction in different Arundo donax L clones BioEnergy Research8 1461ndash1479 httpsdoiorg101007s12155-015-9652-8

Scordia D Cosentino S L Lee J amp Jeffries T W (2011) Diluteoxalic acid pretreatment for biorefining giant reed (Arundo donaxL) Biomass and Bioenergy 35 3018ndash3024 httpsdoiorg101016jbiombioe201103046

Scordia D Cosentino S L Lee J amp Jeffries T W (2012) Bio-conversion of giant reed (Arundo donax L) hemicellulose hydro-lysate to ethanol by Scheffersomyces stipitis CBS6054 Biomassand Bioenergy 39 296ndash305 httpsdoiorg101016jbiombioe201201023

Sharma K P Kushwaha S P amp Gopal B (1998) A comparativestudy of stand structure and standing crops of two wetland spe-cies Arundo donax and Phragmites karka and primary produc-tion in Arundo donax with observations on the effect of clippingJournal of Tropical Ecology 39 3ndash14

Silva C F L Schirmer M A Maeda R N Barcelos C A ampPereira N Jr (2015) Potential of giant reed (Arundo donax L)for second generation ethanol production Electronic Journal ofBiotechnology 18 10ndash15 httpsdoiorg101016jejbt201411002

Slavov G Robson P Jensen E Hodgson E Farrar K AllisonG hellip Donnison I (2013) Contrasting geographic patterns ofgenetic variation for molecular markers vs phenotypic traits in theenergy grass Miscanthus sinensis Global Change Biology Bioen-ergy 5 562ndash571 httpsdoiorg101111gcbb12025

Spencer D F Liow P S Chan W K Ksander G G amp Get-singer K D (2006) Estimating Arundo donax shoot biomassAquatic Botany 84 272ndash276 httpsdoiorg101016jaquabot200511004

Sun S amp Frelich L E (2011) Flowering phenology and heightgrow pattern are associated with maximum plant height relativegrowth rate and stem tissue mass density in herbaceous grasslandspecies Journal of Ecology 99 991ndash1000 httpsdoiorg101111j1365-2745201101830x

Sun L Wang Y Yan X Cheng T Ma K Yang W hellip ZhangQ (2014) Genetic control of juvenile growth and botanical archi-tecture in an ornamental woody plant Prunus mume Sieb etZucc as revealed by a high‐density linkage map BMC Genetics15(Suppl 1) S1 httpsdoiorg1011861471-2156-15-S1-S1

Venables W N amp Ripley B D (2002) Modern applied statisticswith S (4th ed p 495) New York NY Springer‐Verlaghttpsdoiorg101007978-0-387-21706-2

Verhulst N Cox R amp Govaerts B (2013) Yield and yield compo-nents A practical guide for comparing crop management prac-tices Mexico DF CIMMYT International Maize and WheatImprovement Center 23 pp

Watts D A amp Moore G W (2011) Water‐Use Dynamics of anInvasive Reed Arundo donax from Leaf to Stand Wetlands 31725ndash734 httpsdoiorg101007s13157-011-0188-1

Zucaro A Forte A Fagnano M Bastianoni S Basosi R ampFierro A (2015) Comparative attributional life cycle assessmentof annual and perennial lignocellulosic feedstocks productionunder Mediterranean climate for biorefinery framework IntegratedEnvironmental Assessment Management 11 397ndash403 httpsdoiorg101002ieam1604

SUPPORTING INFORMATION

Additional supporting information may be found online inthe Supporting Information section at the end of the article

How to cite this article Fabbrini F Ludovisi RAlasia O et al Characterization of phenologyphysiology morphology and biomass traits across abroad Euro‐Mediterranean ecotypic panel of thelignocellulosic feedstock Arundo donax GCBBioenergy 2018001ndash19 httpsdoiorg101111gcbb12555

FABBRINI ET AL | 19