integration of system phenotypes in microbiome networks to ... · 12/12/2019 · 112 phona first...

1

Integration of system phenotypes in microbiome networks to identify 1

candidate synthetic communities: a study of the grafted tomato rhizobiome 2

3

Ravin Poudel1234#, Ari Jumpponen5, Megan Kennelly4, Cary Rivard6, Lorena Gomez-4

Montano1234, and Karen A. Garrett1234#. 5

1Department of Plant Pathology, University of Florida, Gainesville, FL 32611 6

2Institute for Sustainable Food Systems, University of Florida, Gainesville, FL 32611 7

3Emerging Pathogens Institute, University of Florida, Gainesville, FL 32611 8

4Department of Plant Pathology, Kansas State University, Manhattan, KS 66506 9

5Division of Biology and Ecological Genomics Institute, Kansas State University, Manhattan, KS 10

66506 11

6Department of Horticulture and Natural Resources, Kansas State University, Olathe, KS 66061 12

13

#Correspondence: Karen A. Garrett and Ravin Poudel 14

E-mail addresses: [email protected] and [email protected] 15

16

Running title: Phenotype-based network for microbial consortia design 17

.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 13, 2019. . https://doi.org/10.1101/2019.12.12.874966doi: bioRxiv preprint

https://doi.org/10.1101/2019.12.12.874966

http://creativecommons.org/licenses/by-nc-nd/4.0/

2

ABSTRACT Understanding factors influencing microbial interactions, and designing methods 18

to identify key taxa, are complex challenges for achieving microbiome-based agriculture. Here 19

we use a grafted tomato system to study how grafting and the choice of rootstock influence root-20

associated fungal communities. We then construct a phenotype-OTU network analysis (PhONA) 21

using random forest and network models of the observed sequence-based fungal Operational 22

Taxonomic Units (OTUs) and associated tomato yield data. PhONA provides a framework to 23

select a testable and manageable number of OTUs to support microbiome-based agriculture. We 24

studied three tomato rootstocks (BHN589, RST-04-106 and Maxifort) grafted to a BHN589 25

scion and profiled the associated endosphere and rhizosphere fungal communities by sequencing 26

the fungal Internal Transcribed Spacer (ITS2). The data provided evidence for a rootstock effect 27

(explaining ~2% of the total captured variation, p < 0.01) on the fungal community. Moreover, 28

the most productive rootstock, Maxifort, supported higher fungal species richness than the other 29

rootstocks or controls. Differentially abundant OTUs (DAOTUs) specific to each rootstock were 30

identified for both endosphere and rhizosphere compartments. PhONA integrates the information 31

about individual OTU obtained via differential abundance tests and role analyses. It identified 32

OTUs that were directly important for predicting tomato yield, and other OTUs that were 33

indirectly linked to yield through their links to these OTUs. Fungal OTUs that are directly or 34

indirectly linked with tomato yield represent candidate components of synthetic communities for 35

testing in agricultural systems. 36

37

IMPORTANCE The realized benefits of microbiome analyses for plant health and disease 38

management are often limited by the lack of methods to select manageable and testable synthetic 39

microbiomes. We evaluated the composition and diversity of root-associated fungal communities 40


https://doi.org/10.1101/2019.12.12.874966


3

from grafted tomatoes. We then constructed a phenotype-OTU network analysis (PhONA) using 41

random forest and network models. By incorporating yield data in the network, PhONA 42

identified OTUs that were directly predictive of tomato yield, and other OTUs that were 43

indirectly linked to yield through their links to these OTUs. Follow-up functional studies of taxa 44

associated with effective rootstocks identified using PhONA could support the design of 45

synthetic fungal communities for microbiome-based crop production and disease management. 46

The PhONA framework is flexible for incorporation of other phenotypic data and the underlying 47

models can readily be generalized to accommodate other microbiome or genomics data. 48

49

KEYWORDS fungi, phenotypes, microbiome networks, model integration, tomato 50


https://doi.org/10.1101/2019.12.12.874966


4

Interactions are key to defining system behaviors, structures, and outcomes. In microbial 51

systems, interactions among microbes define their distribution, assemblies, and ecosystem 52

functions. In addition to microbe-microbe interactions, microbes interact with their hosts, and are 53

essential to host health and performance (1-6). In agriculture, plant–microbe interactions 54

improve plant productivity by providing access to nutrients (7-9), infection of plant pathogens (5, 55

10), triggering plant growth promoting factors (11, 12), and enhancing plant resistance (13, 14) 56

and tolerance to abiotic stresses (15-17). Although the importance of microbes and host-microbe 57

interactions to host health and ecological processes is known, interaction-based approaches to 58

manage crop-production remain a scientific frontier. Past attempts to translate information about 59

microbes to design biocontrol agents or biofertilizers have often had limited efficacy and 60

durability (18, 19). Most microbes have been applied as single species, often selected based on 61

pairwise relations of microbes with a pathogen or the host. Interactions among microbes as well 62

as with the host are important, and the net outcome of these complex interactions defines host 63

health and ecosystem functions. Thus, it is critical to understand the ecology of microbes 64

selected for biological applications, and systems approaches centered on host-microbe 65

interaction can help guide the selection of microbes for engineering synthetic communities. 66

Among tools to better understand microbial ecology, network models of microbial 67

communities, and studies of network structures and key groups, have proven popular for 68

generating hypotheses about how to engineer microbial consortia. In such network models of 69

microbiomes, a node represents an OTU, and a link exists between two OTUs if their sequence 70

proportions are significantly associated across samples. When evaluated with other conventional 71

measures of microbial community, such as diversity indices, network models can be used to 72

identify hub taxa that may be key to maintaining microbial assemblages and diversity (20), or to 73


https://doi.org/10.1101/2019.12.12.874966


5

evaluate changes in community complexity and interactions in response to experimental 74

treatments (21). The use of microbiome networks to describe general community structures and 75

key properties is valuable and often the most practical option when additional information is 76

missing or the goal is to compare across studies with different types of data (22, 23). The utility 77

of network analysis for identifying candidate assemblages for biocontrol can be enhanced by 78

incorporating nodes that represent more types of features (24, 25). For instance, a novel 79

association of host metadata with the microbiome was revealed in an integrated microbiome-80

metadata network (26), where a feature strongly associated with hub microbes can serve as a 81

marker to measure host performance. In agriculture, plant yield or other phenotypic traits can be 82

integrated in microbiome networks, with the potential to identify microbial consortia that are 83

predictive of host phenotypes. Because such models include host phenotypes, they facilitate 84

finding candidate sets of OTUs that may directly or indirectly affect key OTUs or phenotypic 85

traits. Visualization of networks is often valuable for this purpose; however, the real value of 86

phenotype-based network models is their ability to identify candidate taxa or consortia. The 87

hypothesized beneficial sets of OTUs may represent targets for pure culturing efforts or, if 88

cultures exist, the sets can be evaluated in lab or field studies. 89

While phenotype-based network models have the potential to predict key taxa, 90

application of such models should be integrated with findings from other community analyses so 91

that the inference about key taxa is biologically and ecologically meaningful. For example, plant 92

microbiome studies indicate that a small but consistent proportion of variation in microbial 93

communities is often explained by host genotype (27-32), indicating the potential for genotype-94

based modulation of microbial communities in crop-production on a broader scale. These results 95

support the idea of host-specific microbial community selection (33). Many of the microbes 96


https://doi.org/10.1101/2019.12.12.874966


6

selected may be taxa that are evolutionarily essential for the survival and function of plants (34, 97

35). In addition, the extent of host genotype filtering of microbes differs across the rhizosphere, 98

rhizoplane, and endosphere, and varies from one host species to another (36-38). Results that 99

indicate microbial filtering by different crop hosts, plant compartments, geographic locations, 100

and environmental factors (39, 40)are promising for minimizing the search space, or necessary 101

sample numbers, to identify candidate taxa for synthetic communities. For instance, factors that 102

explain greater variation in microbial community composition, but that are outside the control of 103

management, can be treated as blocks in experimental designs, so that host- or compartment-104

specific effects on the microbial community can be searched to identify candidate taxa. 105

In our current study, building on previously described agricultural experiments in grafted 106

tomato systems (41, 42), we characterized the root-associated fungal (RAF) communities and 107

implemented a new approach to select potential candidate fungal taxa that are predictive of 108

tomato yield and/or that are in significant association with such fungal taxa. The new phenotype-109

OTU network analysis (PhONA) is a network framework to support the selection of candidate 110

taxa and to integrate system traits (such as host yield) in microbe-microbe association networks. 111

PhONA first identifies the OTUs predictive of phenotype using a random forest model and 112

indicates positive or negative association (Pearson's correlation coefficient) of the predicted 113

OTUs with the host phenotype, and then combines the OTUs associated with the phenotype with 114

other OTUs in a network model (Fig. 1). Random forest, a non-parametric ensemble learning 115

method, has often been used in microbiome studies (43, 44) for classification of sample states 116

and is suited for non-linear data types involving complex interactions (45). 117

Phenotype-based selection of microbial consortia is an effective approach to select 118

simplified yet representative microbial taxa and could support the design of microbiome-based 119


https://doi.org/10.1101/2019.12.12.874966


7

products. Changes in abundance (46), successive selection over multiple generations (3), or 120

analyses of binary host-microbe relationships (47) are some the recent phenotype-based 121

applications to select candidate taxa for biological applications. In contrast to these approaches, 122

first PhONA utilizes a statistical model to select OTUs predictive of host phenotype, then further 123

integrates the predicted OTUs in the network framework. PhONA provides an interaction-based 124

understanding of microbial communities to support microbiome studies. In addition, we 125

contrasted the RAF community’s diversity and interactions among the productive rootstocks and 126

the controls, for endosphere and rhizosphere compartments. Based on our yield data, and 127

previous studies of microbial interactions (21), we expected a greater number of fungal OTUs 128

and a greater number of microbial associations for more productive rootstocks. Moreover, in our 129

previous studies of bacterial communities in the tomato rhizobiome, we observed compartment-130

specific (endosphere vs rhizosphere) effects of grafting and rootstocks on bacterial community 131

composition and diversity (42), and expected similar effects on RAF diversity and composition. 132

133

METHODS 134

Experimental Plots, Rootstocks, and Study Sites. We studied grafted tomato plants in 135

high tunnels in an experimental design similar to the one described by Poudel et al. (42). Tomato 136

plants were grafted following a tube-grafting protocol described in Meyer et al. (41). Our study 137

included three rootstocks (BHN589; RST-04-106, and Maxifort) in four treatments: 1) 138

nongrafted BHN589 plants; 2) selfgrafted BHN589 plants (plants grafted to their own rootstock); 139

3) BHN589 grafted to RST-04-106; and 4) BHN589 grafted to Maxifort. We chose BHN589 as 140

scion primarily based on its popularity due to high yield and high-quality fruit with a long shelf 141


https://doi.org/10.1101/2019.12.12.874966


8

life. For rootstocks, we selected Maxifort because it is a productive and popular rootstock, and 142

RST-04-106 as a new rootstock variety based on tomato breeders’ recommendations. 143

Our study included two sites: Olathe Horticulture Research and Extension Center 144

(OHREC) and Common Harvest Farm, a farm managed by a collaborating farmer. At each study 145

site, the four graft treatments were assigned to four plots per block in a randomized complete 146

block design. Each plot consisted of 5-8 plants, and one middle plant per plot was sampled 147

during the peak growth stage. There were six blocks at OHREC, and four blocks at Common 148

Harvest Farm, such that for each year, a given rootstock was replicated 10 times. The experiment 149

was repeated for two years (2014 and 2015) with a similar design; however, the blocks and 150

rootstocks were randomized in each year. More information about each study site is provided in 151

Table S1. 152

Sample Preparation, DNA Extraction, and Amplicon Generation. To compare the 153

fungal communities, we selected a center plant from each plot and carefully dug it out such that 154

the majority of the roots remained intact. Endosphere and rhizosphere samples were prepared as 155

previously described (42) and the total genomic DNA was extracted using a DNA extraction kit 156

(MoBio UltraClean Soil DNA Isolation Kit; MoBio, Carlsbad, CA, USA) as per manufacturer’s 157

protocol, with slight modification during the homogenization step (42). To recover high genetic 158

diversity, we opted for the two-step PCR approach. The primary PCR amplicons were generated 159

in 50 μL reactions under the following conditions: 1 μM forward and reverse primers, 10 ng 160

template DNA, 200 μM of each dioxynucleotide, 1.5 mM MgCl2, 10 μL 5x Phusion Green HF 161

buffer (Finnzymes,Vantaa, Finland), 24.5 μL molecular biology grade water, and 1 unit (0.5 μL) 162

Phusion Hot Start II DNA Polymerase (Finnzymes, Vantaa, Finland). PCR cycle parameters 163

consisted of a 98° C initial denaturing step for 30 seconds, followed by 30 cycles at 98° C for 10 164


https://doi.org/10.1101/2019.12.12.874966


9

seconds, 57° C annealing temperature for 30 seconds, and 72° C extension step for 30 seconds, 165

followed by a final extension step at 72° C for 10 minutes. All samples were PCR-amplified in 166

triplicate to minimize stochasticity, pooled, and cleaned using Diffinity RapidTip (Diffinity 167

Genomics, West Chester, PA, USA). In this PCR, we amplified the entire ITS region of fungal 168

rRNA genes using primers ITS1F-CTTGGTCATTTAGAGGAAGTAA and ITS4-169

TCCTCCGCTTATTGATATGC, (e.g. 48). The average amplicon length of the ITS region in 170

fungi is about 600 bp and could not be fully covered with the Illumina MiSeq platform (v.3-171

chemistry) in a single read. Thus, in the following nested PCR, only ITS2 out of the whole ITS 172

region was amplified using fITS7-ITS4 primers (49) incorporating unique Molecular Identifier 173

Tags (MIDs) at the 5' end of the reverse primer (ITS4). For the nested PCR, we used similar 174

reagents and PCR conditions as in the primary PCR, with some modifications: amplification 175

cycles were reduced to ten, total reaction volume was reduced to 25 ul, and 5 ul of cleaned PCR 176

product from the first PCR amplification was used as the DNA template. Similarly, the nested 177

PCR was run in triplicate, pooled by experimental unit, and cleaned with an Agencourt AmPure 178

cleanup kit using a SPRIplate 96-ring magnet (Beckman Coulter, Beverly, MA, USA) as per the 179

manufacturer’s protocol. Then, 200 ng of cleaned, barcoded PCR amplicons were combined per 180

experimental unit, and the final pool was cleaned again using an Agencourt AmPure cleanup kit 181

as above. Illumina MiSeq adaptors were ligated to the library and paired-end sequenced on a 182

MiSeq Personal Sequencing System (Illumina, San Diego, CA, USA) using MiSeq Reagent Kit 183

V3 with 600 cycles. The endosphere and the rhizosphere amplicon libraries were sequenced 184

separately in two runs. Adaptor ligation and sequencing were performed at the Integrated 185

Genomics Facility at Kansas State University. All sequence data generated in this study were 186

deposited in the NCBI Sequence Read Archive depository (BioProject:……). 187


https://doi.org/10.1101/2019.12.12.874966


10

Bioinformatics and OTU Designation. The sequence library of fastq files was curated 188

using the MOTHUR pipeline (Version 1.33.3; (50)) following steps modified from the MiSeq 189

Standard Operating Protocol (SOP; www.mothur.org/wiki/MiSeq_SOP). Briefly, the forward 190

and the reverse reads were assembled into contigs using the default alignment algorithm. Any 191

sequences shorter than 250 base pairs or containing an ambiguous base call or more than eight 192

homopolymers or missing MIDs were removed from the library. Barcoded sequences were 193

assigned to experimental units, and fasta and group files for the endosphere and rhizosphere 194

libraries were merged and processed together for the remaining steps in the MOTHUR pipeline. 195

The pairwise distance matrix based on the filtered sequences was created and data clustered into 196

OTUs at 97% sequence similarity using the nearest-neighbor joining algorithm. The clustered 197

OTUs were assigned to a putative taxonomic identity using a Bayesian classifier (51) referencing 198

the UNITE plus INSD non-redundant ITS database (52). To minimize the bias resulting from 199

unequal sequence counts per sample, samples were rarified at the sequence frequency of the 200

sample yielding the lowest count (6,777). The final curated OTU database included 1,084,281 201

total sequences representing 16,151 fungal OTUs, including singletons (5376). 202

Statistical analyses. We evaluated the network of associations among fungal OTUs to 203

better understand the community composition and the interactions therein. The observed OTU 204

database was divided into eight subsets, each combination of the four rootstocks and two 205

compartments (endosphere and rhizosphere), such that we constructed eight networks in total. In 206

our network models, a node represents an OTU and a link exists between a pair of OTUs if there 207

is strong evidence (p < 0.01) that their frequencies are correlated across samples. Reducing false 208

associations due to compositional bias in network modeling of microbiome data is important for 209

clearer interpretation (53). Thus, we used a Sparse Correlations for Compositional data (SparCC) 210


https://doi.org/10.1101/2019.12.12.874966


11

method to evaluate the pairwise associations (53), designed to minimize the compositional bias 211

effect due to normalization. In our analyses, associations were defined in 20 iterations, and the 212

significance of a pairwise association was determined from 100 bootstrapped datasets. Once the 213

matrix defining all the pairwise associations was derived, we selected only those OTUs for 214

which the absolute value of the association was greater than 0.5 (and p < 0.01) in the network 215

analyses for each of the rootstock genotypes. 216

To identify the OTUs associated with tomato yield in each rootstock, a regression-based 217

random forest model was fitted to the observed data. Marketable tomato yield data reported by 218

Meyer et al. (41) was the response variable and fungal OTUs were potential predictors. In order 219

to minimize the influence of rare and transient OTUs, only those OTUs that occurred in more 220

than 20% of the samples were filtered then transformed using Z-scores. The transformed data 221

were used for training the model. OTUs in the resulting model were ranked by their importance 222

based on increased mean square error (%IncMSE) of the model. Random forest uses out-of-bag 223

(OOB) samples (set of samples not used for constructing the tree) to estimate the MSE error for 224

each decision tree. The average MSE across the decision trees is then evaluated with the error 225

due to the permutation of a variable (predictor) in the model, where an increase in error 226

(%IncMSE) represents feature importance. From the list of OTUs ranked by their importance to 227

the model, we selected all the OTUs with %IncMSE > 0 for constructing PhONA. PhONA 228

integrates the results from the random forest model for yield with the OTU-OTU association 229

network. We then plotted the resulting network using the igraph package (54) in R. To evaluate 230

the role of nodes in the network, we placed each node in one of four categories – peripherals, 231

module hubs, network hubs, and connectors – based on the within-module degree and among-232

module connectivity (24, 55). Categorization of individual OTUs based on differential 233


https://doi.org/10.1101/2019.12.12.874966


12

abundance tests or role analyses were indicated visually as a component of the PhONA, to 234

provide an analysis integrating information from multiple analyses. 235

To evaluate the effects of rootstocks on fungal diversity, Shannon entropy and species 236

richness were evaluated using the vegan package (56) implemented as part of the phyloseq 237

package (57) in R (58). Differences in diversity across the rootstocks were compared using a 238

mixed model ANOVA in the lme4 package in R (59). Study site and sampling year were treated 239

as random factors, blocked by study sites, whereas the rootstock and compartments were treated 240

as fixed factors. Changes in fungal community composition across the samples were estimated 241

based on a Bray-Curtis dissimilarity matrix and visualized in non-metric multidimensional 242

scaling (NMDS) plots. The contribution of factors to the observed variation in fungal 243

composition was estimated in a permutational multivariate analysis of variance (PERMANOVA, 244

using 1000 permutations) using the adonis function in the vegan package (56). To identify OTUs 245

that were sensitive to the rootstock treatments, the observed frequency (proportion) of each OTU 246

was evaluated by fitting a generalized linear model (GLM) with negative binomial distribution, 247

to identify depleted or enriched OTUs (Differentially Abundant OTUs – DAOTUs). Likelihood 248

ratio tests and contrast analyses (between the hybrid rootstock and controls) were performed for 249

the fitted GLM to identify the DAOTUs. We used OTU frequencies from selfgrafts and 250

nongrafts as controls, in comparisons with other rootstocks using contrasts. All tests were 251

adjusted to control the false discovery rate (FDR, p < 0.01) using the Benjamini-Hochberg 252

method (60). A differential abundance test was performed within the controls (selfgraft vs. 253

nongraft) to identify the OTUs responsive to the grafting procedure itself. 254


https://doi.org/10.1101/2019.12.12.874966


13

RESULTS 255

General Root-Associated Fungal Community in the Grafted Tomato System. Once 256

rare taxa (<10 sequence counts, which accounted for more than 90% of the observed OTUs) 257

were removed, the community consisted of 1586 OTUs and 1,063,017 sequences. Of these 258

sequences, 4.8% remained unclassified at the phylum level (Fig. S1). The classified sequences 259

represented Ascomycota (52.5%), Basidiomycota (25.6%), Zygomycota (11.5%), 260

Chytridiomycota (3.6%), Glomeromycota (1.7%), and Rozellomycota (0.07%) (Fig. 1). At the 261

class level, Pezizomycetes, Agaricomycetes, and Dothideomycetes were the most abundant 262

across all the rootstocks. At the order level, the communities were dominated by Pezizales, 263

Pleosporales, Cantharellales, Mortierellales, and Hypocreales. Analyses at the family level 264

revealed that Pyronemataceae, Mortierellaceae, Ceratobasidiaceae, and Pleosporaceae were the 265

most common taxa in the overall community. At the genus level, Mortierella spp., 266

Thanatephorus spp., and Alternaria spp. were the most abundant genera. 267

Effects of Grafting and Rootstock on α-Diversity. There was strong evidence for an 268

effect of rootstocks on OTU richness (Fdf, 3 = 8.6, p < 0.001) and Shannon entropy (Fdf, 3 = 3.2, p 269

= 0.02) for tomato RAF communities. Mean species richness was higher in both the endosphere 270

(p = 0.01) and rhizosphere (p = 0.001) of one of the hybrid rootstocks, Maxifort, compared to the 271

nongrafted control. Shannon entropy followed trends similar to richness with a higher estimate 272

for Maxifort; however, there was only evidence for higher Shannon entropy in Maxifort for the 273

rhizosphere (p = 0.004), but not for the endosphere (p = 0.6) (Fig. 2). Both species richness and 274

Shannon entropy were higher (p < 0.001) in the rhizosphere than in the endosphere across all the 275

rootstock genotypes (Fig. S2). 276


https://doi.org/10.1101/2019.12.12.874966


14

Effects of Grafting and Rootstock on RAF Composition. Based on previous studies of 277

the plant genotype effect on the rhizobiome (42), we expected a significant effect of rootstocks 278

on the community compositions. Rootstock explained 2% of the variation in the RAF community 279

composition (PERMANOVA; Fdf, 3 = 1.39, R2 = 0.02, p = 0.02), while compartment, study site, 280

and year explained a greater proportion of the variation than rootstock (Fig. S3 and Table S2). 281

Endosphere-rhizosphere compartments accounted for 8.9% of the variation. Study site and 282

interannual variation explained 8.4% and 5.4% of the total variation, respectively (Table S2). 283

Comparison of DAOTUs. The analysis of differential abundance found effects of 284

rootstock genotype at the individual OTU level. While analyses of alpha diversity indicated 285

higher diversity in the rhizosphere than in the endosphere, we observed the opposite in the 286

analysis of DAOTUs, with nearly twice as many DAOTUs in the endosphere (n = 135) as in the 287

rhizosphere (n = 72) (Figs. 2 and S4). Comparison across rootstocks indicated a greater number 288

of DAOTUs in Maxifort (n = 92) than in RST-04-106 (n = 64) and the selfgraft control (n = 51). 289

Compared to the hybrid rootstocks, the number of depleted taxa was greater in the selfgraft 290

control (n = 27). Among the enriched OTUs in Maxifort, 31 OTUs belonged to Basidiomycota, 291

23 to Ascomycota, and 12 to Glomeromycota, whereas four basidiomycete, three ascomycete, 292

and two zygomycete OTUs were depleted in Maxifort. In the case of RST-04-106, enriched taxa 293

included 21 OTUs in Basidiomycota, 25 OTUs in Ascomycota and four OTUs in 294

Glomeromycota, whereas the depleted OTUs included five in Zygomycota. Between the self- 295

and nongraft controls, seven OTUs in Ascomycota, eight OTUs in Basidiomycota, and five 296

OTUs in Zygomycota were enriched, whereas 10 OTUs in Basidiomycota, six OTUs in 297

Ascomycota, four OTUs in Zygomycota, and four OTUs in Glomeromycota were depleted. 298


https://doi.org/10.1101/2019.12.12.874966


15

Network Analysis/ General Network Structures. Fungal community complexity, 299

defined in terms of mean node degree and community structures/motifs, varied among the 300

rootstocks in both the endosphere and the rhizosphere, with a greater mean node degree for one 301

of the hybrid rootstocks, Maxifort, compared to both controls and RST-04-106 (Figs. 3, S5, S6, 302

and Table S3). Comparing compartments, complexity was higher in the rhizosphere than in the 303

endosphere (Figs. 3, S5, S6, and Table S3). In addition to the total number of links, the link type 304

(either positive or negative) differed among the rootstocks in both compartments (Table S3), 305

with a higher ratio of negative to positive links in Maxifort in both the endosphere and the 306

rhizosphere compartments. Rhizosphere fungal communities had a higher ratio of negative to 307

positive links as compared to the endosphere, for all rootstocks. Although we observed rootstock 308

specific or compartment specific effects on the node degree and ratio of negative to positive 309

links, the number of modules defined using a simulated annealing (SA) algorithm, were similar 310

in both the endosphere and rhizosphere compartments and across the rootstocks (Table S3). Our 311

analyses of node types divided the observed nodes in the association-network into four 312

categories: peripherals, module hubs, network hubs, and connectors. More taxa in the 313

rhizosphere were identified as key nodes than in the endosphere, and key nodes were more 314

common in the hybrid rootstocks than in the non- and selfgrafted controls (Figs. 4 and 5). 315

Random Forest and PhONA. Using random forest models, we identified the 316

compartment-specific OTUs useful for predicting tomato yield in each rootstock. The number of 317

OTUs with positive %IncMSE (importance) differed between the compartments and among the 318

rootstocks. The number of OTUs with positive %IncMSE was higher in the rhizosphere (54.3 319

17. 1; mean sd) than in the endosphere (28 6.5), with the highest number for Maxifort. This 320

might indicate that rhizosphere OTUs are more important for determining host-specific traits. 321


https://doi.org/10.1101/2019.12.12.874966


16

For each treatment, the observed PhONA provided a graphical representation that can be used to 322

select candidate taxa based on the network structure, phenotype, and node attributes evaluated in 323

differential abundance tests and role analyses. The rational selection of taxa could even be 324

improved with a priori knowledge about potential biological and ecological functions of the 325

observed OTUs. 326

327

DISCUSSION 328

This study demonstrated the effect of rootstocks on RAF community composition and structure. 329

General diversity-based analyses indicated the effect of rootstocks on RAF community 330

composition. The more productive hybrid rootstock, Maxifort, supported higher fungal richness 331

and Shannon entropy, as well as a greater number of DAOTUs than the controls, consistent with 332

the expectation that higher diversity and a higher number of responsive taxa (DAOTUs) would 333

be associated with a more productive genotype. Also consistent with our expectations, we 334

observed higher microbial diversity and fewer responsive taxa (DAOTUs) in the rhizosphere 335

than in the endosphere. The integrated host phenotype and OTU network in the PhONA 336

indicated potential candidate taxa for each rootstock, and community structures in the 337

endosphere and rhizosphere compartments. The general network analysis found more 338

interactions and complex structures among fungal taxa in Maxifort, consistent with our 339

expectation that increased community complexity would be associated with the more productive 340

rootstock. Community complexity, when defined in terms of mean node degree, also differed 341

between the compartments: the endosphere community was less complex than the rhizosphere 342

community in all the rootstocks. Overall, our study i) showed that rootstocks and grafting are 343

significant drivers of RAF community composition, diversity and structure, and ii) provided 344


https://doi.org/10.1101/2019.12.12.874966


17

PhONA as an analytical framework to select potential candidate taxa for microbiome-based 345

agriculture, where the taxa were predictive of yield and in significant association with other 346

OTUs in the community. From a practical standpoint, our results indicate the potential for using 347

plant genotypes and agricultural practices to modulate plant-associated microbial communities, 348

and the potential for the PhONA illustrated here to improve identification of candidate taxa to 349

support microbiome-based crop production. 350

In contrast to network models that portray only microbe-microbe interactions, PhONA 351

integrates the results of random forest models of microbe association with phenotypic traits, to 352

support inference about candidate taxa and predictive microbiome analyses. Thus, candidate taxa 353

can be selected not only because they have a direct association with the host response variable(s), 354

but also because they have an indirect association with the host response variable through their 355

associations with other taxa in the community that have direct associations with host traits. For 356

instance, a node that has a positive association with the system phenotype node (in our case, 357

yield) might have negative or positive association with other OTUs. Such OTUs with indirect 358

positive associations with the desired phenotype might also be included in designing a 359

consortium for biofertilizers. Using a PhONA, a rational consortium can be selected based on the 360

phenotype of interest. Applying PhONA for the case of disease or pathogen-based phenotypes 361

could be useful for designing a rational consortium for biocontrol. We also observed some OTUs 362

with direct negative associations with the yield node. For instance, g_Thanatephorus is one of 363

the nodes identified as depleted in a differential abundance test (Fig. 2), and had a negative 364

association with yield for Maxifort (Fig. 3). Some members of g_Thanatephorus (OTU00001) 365

are known plant pathogens. Similarly, the g_Thanatephorus (OTU00012 ) node was observed to 366

be a connector and had a negative association with Maxifort yield (Fig. 3). Effort to control this 367


https://doi.org/10.1101/2019.12.12.874966


18

particular taxon, as well as the taxa that are have positive associations with g_Thanatephorus, 368

might contribute to maximizing yield. Although we did not observe any disease symptoms in our 369

experiments, OTUs negatively associated with the yield node might represent a case of 370

asymptomatic yield loss. Efforts to explore negatively associated OTUs might provide 371

opportunity to minimize asymptomatic yield loss in crops. The main goal of PhONA is to 372

provide a systems framework to generate hypotheses about the role of microbiome components 373

in host function and performance, and to support the potential for mechanistic/predictive models 374

to understand host-microbiome interactions. In planta experiments with fungal cultures are 375

essential to test the hypotheses generated by these models, to help to differentiate between 376

associations that are based on consistent biological interactions and not simply based on shared 377

(or opposing) niche preferences. It is important to be cautious in attributing biological 378

interactions to the key structures in network attributes, because the links in the PhONA may or 379

may not depict biological interactions. That is, many links may represent correlative 380

relationships rather than causal relationships(24). For instance, the hub node in the network is 381

often regarded as a key node, but the high number of links with the hub node in the association 382

network could be due to shared niches, biological interactions, or a mixture thereof. If the 383

associations are mostly due to shared niches, removing such a hub node will have a more limited 384

effect, whereas removal of a hub node involved in many biological interactions could lead to 385

fragmentation of the microbial community. 386

RAF community composition, diversity, and interactions differed between the 387

endosphere and rhizosphere compartments. Although the endosphere and the rhizosphere are 388

continuous, they are distinct in community composition and diversity. Compartment specificity 389

in community composition and diversity has been reported for other plant species, in both natural 390


https://doi.org/10.1101/2019.12.12.874966


19

and agricultural settings (27, 61, 62). Usually, bulk soil is considered a source of microbial 391

communities, a subset of which is selected for in the rhizosphere (27, 63), mainly as a function of 392

root exudates and rhizodeposits (38, 64-66). Selection of the rhizosphere microbiome could be 393

specific (e.g., antagonistic to plant pathogens) (67) or more general with less influence of host 394

genotype. In comparison, the endosphere of host plants often supports lower microbial diversity 395

compared to the rhizosphere (62, 63). Host tissues and immune systems act as a biotic filter (2). 396

As a result, the microbiome is more specialized in the endosphere than in the rhizosphere. The 397

RAF compartment specificity may also be an important consideration for microbe-based disease 398

management strategies – especially for the management of pathogens or pests that are 399

compartment specific, such as endoparasites and ectoparasites. 400

RAF community composition and diversity also differed among the rootstock genotypes. 401

Plant genotypes can structure root-associated fungal communities (27, 62, 68). The commercial 402

rootstocks in our study have been bred for defined purposes, specifically to provide resistance 403

against specific soil-borne pathogens and pests. Small variations in the host genotype could alter 404

the physiological and immunological responses in the root systems, thereby selecting genotype-405

specific root-associated fungal communities. For example, root exudates and metabolites could 406

be specific to plant host (69-71), and thus could provide specific control of microbial 407

communities (72, 73). In some cases, the host genotype effect can be directly attributed to root 408

anatomy (74). Effective and efficient root types and architectures are desired agronomic traits to 409

cope with biotic and abiotic stresses (75), and root systems vary among plant species and 410

genotypes (74). Moreover, the control of plant genotypes on microbial communities in the root 411

system is linked to nutrient positive feedbacks between the scion and rootstock. Plant genotypes 412

supporting higher yield and biomass may support higher microbial diversity by excreting a 413


https://doi.org/10.1101/2019.12.12.874966


20

greater volume of photosynthates as root exudates and metabolites. Although we did not evaluate 414

root exudates, our study is consistent with a role for higher yield and biomass (as for the 415

Maxifort rootstock) associated with higher fungal diversity. Additionally, we observed an effect 416

of rootstocks on the RAF community composition. Collectively, the results support our 417

expectations of rootstock-specific control of the RAF community. 418

While the effects of plant genotype and compartment are evident, our current study of 419

fungi and previous study of bacteria (42) in the same system indicate that the magnitude of the 420

effect may vary between fungi and bacteria. The estimated effect of rootstocks on bacterial 421

communities (3%) was slightly greater than for fungal communities (2%), although the effects of 422

study sites and compartments were far greater in both the bacterial and fungal communities. As 423

in the RAF communities, compartment (endosphere vs. rhizosphere) and study sites played 424

major roles in structuring bacterial community composition (42); however, bacterial 425

communities in the endosphere and the rhizosphere were much more distinct than the fungal 426

endosphere and rhizosphere - possibly a result of their distinct growth forms (predominantly 427

mycelial or unicellular). Compartment and study site effects on the bacterial communities (42) 428

were greater than on the fungal communities in the same experiment: the estimated compartment 429

effect was three-fold higher (about 25-28%) for bacteria than fungi, and the estimated study site 430

effect was two-fold higher (about 17-18%). In another example plant system, Agave species, the 431

bacterial community was most strongly influenced by the plant compartment whereas the fungal 432

community was more strongly influenced by the study site (61). Lack of a strong study-site 433

effect on the fungal community in our study might indicate a lesser influence of edaphic factors 434

on fungal compared to bacterial communities, or the presence of a wider environmental range for 435

optimal fungal growth (76). Despite the differences in the magnitude of the effect for bacteria 436


https://doi.org/10.1101/2019.12.12.874966


21

versus fungi, in both studies there was strong evidence for the effect of plant genotypes, 437

compartments, and study sites on the microbial community. 438

Our definition of complexity is based on interactions in networks, using a definition 439

similar to that used in other microbiome network analyses (21, 30, 77). However, an increased 440

number of interactions and complex network structures/motifs would tend to be observed 441

whenever there are more nodes in these association networks. The higher number of OTUs 442

associated with Maxifort would tend to result in higher complexity by this measure. Another 443

potential measure of complexity is network density, the proportion of links observed in a 444

network relative to the total number of possible links. In the case of the endosphere community, 445

Maxifort did not have higher density estimates than the other rootstocks, so by that measure of 446

complexity its network was not more complex. Statistical methods comparable to rarefaction, 447

designed to equalize the number of nodes across networks or methods to balance OTU richness 448

for sampling efforts(78), will be a valuable future effort for understanding how network 449

complexity responds to treatments and for making comparisons across studies. In addition, 450

methods to optimize and automate the selection of level of association to maximize the 451

ecosystem functions represent gaps and opportunities for microbiome network analyses. For 452

graphics in the figures in this analysis, we selected a level of association such that an 453

interpretable number of OTUs were depicted. Studies directly applied to identify potential 454

assemblages for agricultural applications could benefit from exploring results for a range of 455

thresholds. 456

The value of information depends on its ability to reduce uncertainty and maximize the 457

utility of predictions. Given the unprecedented complexity of microbial communities, it’s 458

difficult to make a rational selection of candidate taxa for biological applications. A 459


https://doi.org/10.1101/2019.12.12.874966


22

system/community level analytical scheme to incorporate information gained via multiple 460

sources or statistical analyses is imperative. Combining results obtained from different analyses 461

in the PhONA framework supports selection of a rational microbial consortium, enhancing the 462

quality of decision-making. However, careful interpretation is needed, because often two or more 463

OTUs with the same name/taxonomy may have totally different roles in the network. This can be 464

due in part to the low resolving power of the current reference databases. Improvement of 465

reference databases and strain level information will enhance PhONA in the future. 466

Our study indicates the rootstock-genotype specific effect on RAF diversity, composition, 467

and interactions, and also shows how to integrate system phenotypes such as plant yield in a 468

network-based model to support selection of candidate taxa for biological use. However, in 469

sequence-based studies such as ours, the biological and functional significance of the candidate 470

OTUs is commonly unknown. Follow-up experiments with fungal cultures will be necessary to 471

determine the biological roles of the candidate OTUs, and to differentiate causal associations 472

from correlations based on niche preference. Similarly, further development of PhONA to 473

incorporate temporal microbiome data and Bayesian learning and inference methodologies (79, 474

80) has the potential to support causal inference, including understanding of directionality in 475

microbiome networks. PhONA utilizes a random forest model to link OTUs with a system 476

phenotype, although any other models such as generalized linear models or regression could also 477

be used. Given the nature of microbiome data with a high number of features (p) and relatively 478

small number of samples (n), other models to address the n x p problem can readily be applied in 479

PhONA when they improve on random forest models. Additionally, with a larger sample size, 480

more rigorous approaches such as model validation by splitting data into training and test sets 481

could also be used to enhance PhONA. Models to integrate multiomics data to predict the 482


https://doi.org/10.1101/2019.12.12.874966


23

structure of microbial assemblages and provide mechanistic understanding are another domain 483

where an integrated network model like PhONA may offer a new perspective. As lab-based 484

technologies and computational resources become less expensive, and are combined with 485

improved predictive analytics such as PhONA, microbial community analyses can go beyond 486

simple analyses of diversity studies to help make microbiome-based agriculture a reality. 487

488

Acknowledgements 489

We appreciate support from the Ceres Trust, a USDA NCR SARE Research and Education 490

Grant (LNC13-355), the Kansas Agricultural Experiment Station, the National Institute for 491

Mathematical and Biological Synthesis (NIMBioS), and the University of Florida. We thank 492

members of Garrett lab for their comments on the early version of the manuscript. K.A.G., A.J., 493

M.M.K., C.L.R., and R.P. designed the study. R.P. and L.G.M. collected and processed the 494

samples. R.P. analyzed and interpreted the data. R.P., K.A.G., and A.J. wrote the manuscript. 495

R.P., K.A.G., A.J., M.M.K., C.L.R., and L.G.M. revised the manuscript. All authors read and 496

approved the final manuscript. 497


https://doi.org/10.1101/2019.12.12.874966


24

REFERENCES 498

1. Berendsen RL, Pieterse CM, Bakker PA. 2012. The rhizosphere microbiome and plant 499

health. Trends Plant Sci 17:478-86. 500

2. Lebeis SL, Paredes SH, Lundberg DS, Breakfield N, Gehring J, McDonald M, Malfatti S, 501

del Rio TG, Jones CD, Tringe SG, Dangl JL. 2015. Salicylic acid modulates colonization 502

of the root microbiome by specific bacterial taxa. Science 349:860-864. 503

3. Panke-Buisse K, Poole AC, Goodrich JK, Ley RE, Kao-Kniffin J. 2015. Selection on soil 504

microbiomes reveals reproducible impacts on plant function. ISME J 9:980-9. 505

4. Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL, Duncan A, Ley RE, Sogin ML, 506

Jones WJ, Roe BA, Affourtit JP, Egholm M, Henrissat B, Heath AC, Knight R, Gordon 507

JI. 2009. A core gut microbiome in obese and lean twins. Nature 457:480-4. 508

5. Xue C, Penton CR, Shen Z, Zhang R, Huang Q, Li R, Ruan Y, Shen Q. 2015. 509

Manipulating the banana rhizosphere microbiome for biological control of Panama 510

disease. Sci Rep 5:11124. 511

6. Gould AL, Zhang VV, Lamberti L, Jones EW, Obadia B, Korasidis N, Gavryushkin A, 512

Carlson JM, Beerenwinkel N, Ludington WB. 2018. Microbiome interactions shape host 513

fitness. Proc Natl Acad Sci USA 115:E11951-E11960. 514

7. Jin CW, Ye YQ, Zheng SJ. 2014. An underground tale: Contribution of microbial activity 515

to plant iron acquisition via ecological processes. Ann Bot 113:7-18. 516

8. Koide RT. 1991. Nutrient supply, nutrient demand and plant-response to mycorrhizal 517

infection. New Phytol 117:365-386. 518

9. Mishra PK, Bisht SC, Mishra S, Selvakumar G, Bisht JK, Gupta HS. 2012. Coinoculation 519

of Rhizobium leguminosarum-pr1 with a cold tolerant Pseudomonas sp. improves iron 520


https://doi.org/10.1101/2019.12.12.874966


25

acquisition, nutrient uptake and growth of field pea (Pisum sativum l.). J Plant Nutr 521

35:243-256. 522

10. Mendes R, Kruijt M, de Bruijn I, Dekkers E, van der Voort M, Schneider JHM, Piceno 523

YM, DeSantis TZ, Andersen GL, Bakker PAHM, Raaijmakers JM. 2011. Deciphering 524

the rhizosphere microbiome for disease-suppressive bacteria. Science 332:1097-1100. 525

11. Gaiero JR, McCall CA, Thompson KA, Day NJ, Best AS, Dunfield KE. 2013. Inside the 526

root microbiome: Bacterial root endophytes and plant growth promotion. Am J Bot 527

100:1738-50. 528

12. Glick BR. 1995. The enhancement of plant-growth by free-living bacteria. Can J 529

Microbiol 41:109-117. 530

13. Conrath U. 2009. Priming of induced plant defense responses. Plant Innate Immunity 531

51:361-395. 532

14. Pieterse CMJ, Zamioudis C, Berendsen RL, Weller DM, Van Wees SCM, Bakker 533

PAHM. 2014. Induced systemic resistance by beneficial microbes. Annu Rev 534

Phytopathol 52:347-375. 535

15. Marasco R, Rolli E, Ettoumi B, Vigani G, Mapelli F, Borin S, Abou-Hadid AF, El-536

Behairy UA, Sorlini C, Cherif A, Zocchi G, Daffonchio D. 2012. A drought resistance-537

promoting microbiome is selected by root system under desert farming. PLoS One 7(10): 538

e48479. 539

16. Yang J, Kloepper JW, Ryu CM. 2009. Rhizosphere bacteria help plants tolerate abiotic 540

stress. Trends Plant Sci 14:1-4. 541


https://doi.org/10.1101/2019.12.12.874966


26

17. Yuan ZL, Druzhinina IS, Labbe J, Redman R, Qin Y, Rodriguez R, Zhang CL, Tuskan 542

GA, Lin FC. 2016. Specialized microbiome of a halophyte and its role in helping non-543

host plants to withstand salinity. Sci Rep 6:32467. 544

18. Mazzola M, Freilich S. 2017. Prospects for biological soilborne disease control: 545

Application of indigenous versus synthetic microbiomes. Phytopathology 107:256-263. 546

19. Xu XM, Jeffries P, Pautasso M, Jeger MJ. 2011. Combined use of biocontrol agents to 547

manage plant diseases in theory and practice. Phytopathology 101:1024-1031. 548

20. Agler MT, Ruhe J, Kroll S, Morhenn C, Kim ST, Weigel D, Kemen EM. 2016. Microbial 549

hub taxa link host and abiotic factors to plant microbiome variation. PLoS Biol 14(1): 550

e1002352. 551

21. Shi S, Nuccio EE, Shi ZJ, He Z, Zhou J, Firestone MK. 2016. The interconnected 552

rhizosphere: High network complexity dominates rhizosphere assemblages. Ecol Lett 553

19:926-36. 554

22. Ascunce MS, Shin K, Huguet-Tapia JC, Poudel R, Garrett KA, van Bruggen AHC, Goss 555

EM. 2018. Penicillin trunk injection affects bacterial community structure in citrus trees. 556

Microb Ecol 78:457–469. 557

23. Garrett KA, Alcala-Briseno RI, Andersen KF, Buddenhagen CE, Choudhury RA, Fulton 558

JC, Hernandez Nopsa JF, Poudel R, Xing Y. 2018. Network analysis: A systems 559

framework to address grand challenges in plant pathology. Annu Rev Phytopathol 560

56:559-580. 561

24. Poudel R, Jumpponen A, Schlatter DC, Paulitz TC, Gardener BBM, Kinkel LL, Garrett 562

KA. 2016. Microbiome networks: A systems framework for identifying candidate 563

microbial assemblages for disease management. Phytopathology 106:1083-1096. 564


https://doi.org/10.1101/2019.12.12.874966


27

25. Rottjers L, Faust K. 2018. From hairballs to hypotheses-biological insights from 565

microbial networks. Fems Microbiology Reviews 42:761-780. 566

26. Wang Z, Bafadhel M, Haldar K, Spivak A, Mayhew D, Miller BE, Tal-Singer R, 567

Johnston SL, Ramsheh MY, Barer MR, Brightling CE, Brown JR. 2016. Lung 568

microbiome dynamics in copd exacerbations. European Respiratory Journal 47:1082-569

1092. 570

27. Cregger MA, Veach AM, Yang ZK, Crouch MJ, Vilgalys R, Tuskan GA, Schadt CW. 571

2018. The Populus holobiont: Dissecting the effects of plant niches and genotype on the 572

microbiome. Microbiome 6:31. 573

28. Edwards JA, Santos-Medellin CM, Liechty ZS, Nguyen B, Lurie E, Eason S, Phillips G, 574

Sundaresan V. 2018. Compositional shifts in root-associated bacterial and archaeal 575

microbiota track the plant life cycle in field-grown rice. PLoS Biol 16:e2003862. 576

29. Fitzpatrick CR, Copeland J, Wang PW, Guttman DS, Kotanen PM, Johnson MTJ. 2018. 577

Assembly and ecological function of the root microbiome across angiosperm plant 578

species. Proc Natl Acad Sci USA 115:E1157-E1165. 579

30. Marasco R, Rolli E, Fusi M, Michoud G, Daffonchio D. 2018. Grapevine rootstocks 580

shape underground bacterial microbiome and networking but not potential functionality. 581

Microbiome 6:3. 582

31. Wagner MR, Lundberg DS, del Rio TG, Tringe SG, Dangl JL, Mitchell-Olds T. 2016. 583

Host genotype and age shape the leaf and root microbiomes of a wild perennial plant. Nat 584

Commun 7:12151. 585


https://doi.org/10.1101/2019.12.12.874966


28

32. Herrera J, Poudel R, Bokati D. 2013. Assessment of root-associated fungal communities 586

colonizing two species of tropical grasses reveals incongruence to fungal communities of 587

north american native grasses. Fungal Ecol 6:65-69. 588

33. Bulgarelli D, Schlaeppi K, Spaepen S, Ver Loren van Themaat E, Schulze-Lefert P. 589

2013. Structure and functions of the bacterial microbiota of plants. Annu Rev Plant Biol 590

64:807-38. 591

34. Hol WHG, Bezemer TM, Biere A. 2013. Getting the ecology into interactions between 592

plants and the plant growth-promoting bacterium Pseudomonas fluorescens. Front Plant 593

Sci 4:81. 594

35. Ortiz-Castro R, Contreras-Cornejo HA, Macias-Rodriguez L, Lopez-Bucio J. 2009. The 595

role of microbial signals in plant growth and development. Plant Signal Behav 4:701-12. 596

36. Edwards J, Johnson C, Santos-Medellín C, Lurie E, Podishetty NK, Bhatnagar S, Eisen 597

JA, Sundaresan V. 2015. Structure, variation, and assembly of the root-associated 598

microbiomes of rice. Proc Natl Acad Sci USA 112:E911-E920. 599

37. Lundberg DS, Lebeis SL, Paredes SH, Yourstone S, Gehring J, Malfatti S, Tremblay J, 600

Engelbrektson A, Kunin V, del Rio TG, Edgar RC, Eickhorst T, Ley RE, Hugenholtz P, 601

Tringe SG, Dangl JL. 2012. Defining the core Arabidopsis thaliana root microbiome. 602

Nature 488:86-90. 603

38. Wang P, Marsh EL, Ainsworth EA, Leakey ADB, Sheflin AM, Schachtman DP. 2017. 604

Shifts in microbial communities in soil, rhizosphere and roots of two major crop systems 605

under elevated CO2 and O3. Sci Rep 7:15019. 606


https://doi.org/10.1101/2019.12.12.874966


29

39. Herrera J, Poudel R, Nebel KA, Collins SL. 2011. Precipitation increases the abundance 607

of some groups of root-associated fungal endophytes in a semiarid grassland. Ecosphere 608

2(4):art50. 609

40. Bokati D, Herrera J, Poudel R. 2016. Soil influences colonization of root-associated 610

fungal endophyte communities of maize, wheat, and their progenitors. J Mycol 2016:1-9. 611

41. Meyer L. 2016. Grafting to increase high tunnel tomato productivity in the central united 612

states. Master Thesis. http://hdl.handle.net/2097/32736. 613

42. Poudel R, Jumpponen A, Kennelly MM, Rivard CL, Gomez-Montano L, Garrett KA. 614

2019. Rootstocks shape the rhizobiome: Rhizosphere and endosphere bacterial 615

communities in the grafted tomato system. Appl Environ Microbiol 85:01765–01718. 616

43. Roguet A, Eren AM, Newton RJ, McLellan SL. 2018. Fecal source identification using 617

random forest. Microbiome 6:185. 618

44. Belk A, Xu ZZ, Carter DO, Lynne A, Bucheli S, Knight R, Metcalf JL. 2018. 619

Microbiome data accurately predicts the postmortem interval using random forest 620

regression models. Genes 9(2):104. 621

45. Cutler DR, Edwards TC, Beard KH, Cutler A, Hess KT. 2007. Random forests for 622

classification in ecology. Ecology 88:2783-2792. 623

46. Niu B, Paulson JN, Zheng XQ, Kolter R. 2017. Simplified and representative bacterial 624

community of maize roots. Proc Natl Acad Sci USA 114:E2450-E2459. 625

47. Paredes SH, Gao TX, Law TF, Finkel OM, Mucyn T, Teixeira PJPL, Gonzalez IS, 626

Feltcher ME, Powers MJ, Shank EA, Jones CD, Jojic V, Dangl JL, Castrillo G. 2018. 627

Design of synthetic bacterial communities for predictable plant phenotypes. Plos Biology 628

16. 629


https://doi.org/10.1101/2019.12.12.874966


30

48. Gardes M, Bruns TD. 1993. ITS primers with enhanced specificity for basidiomycetes--630

application to the identification of mycorrhizae and rusts. Mol Ecol 2:113-8. 631

49. Ihrmark K, Bodeker ITM, Cruz-Martinez K, Friberg H, Kubartova A, Schenck J, Strid Y, 632

Stenlid J, Brandstrom-Durling M, Clemmensen KE, Lindahl BD. 2012. New primers to 633

amplify the fungal ITS2 region - evaluation by 454-sequencing of artificial and natural 634

communities. FEMS Microbiol Ecol 82:666-677. 635

50. Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, Hollister EB, Lesniewski RA, 636

Oakley BB, Parks DH, Robinson CJ, Sahl JW, Stres B, Thallinger GG, Van Horn DJ, 637

Weber CF. 2009. Introducing mothur: Open-source, platform-independent, community-638

supported software for describing and comparing microbial communities. Appl Environ 639

Microbiol 75:7537-41. 640

51. Wang Q, Garrity GM, Tiedje JM, Cole JR. 2007. Naive bayesian classifier for rapid 641

assignment of rRNA sequences into the new bacterial taxonomy. Appl Environ Microbiol 642

73:5261-5267. 643

52. Koljalg U, Nilsson RH, Abarenkov K, Tedersoo L, Taylor AFS, Bahram M, Bates ST, 644

Bruns TD, Bengtsson-Palme J, Callaghan TM, Douglas B, Drenkhan T, Eberhardt U, 645

Duenas M, Grebenc T, Griffith GW, Hartmann M, Kirk PM, Kohout P, Larsson E, 646

Lindahl BD, Luecking R, Martin MP, Matheny PB, Nguyen NH, Niskanen T, Oja J, Peay 647

KG, Peintner U, Peterson M, Poldmaa K, Saag L, Saar I, Schuessler A, Scott JA, Senes 648

C, Smith ME, Suija A, Taylor DL, Telleria MT, Weiss M, Larsson KH. 2013. Towards a 649

unified paradigm for sequence-based identification of fungi. Mol Ecol 22:5271-5277. 650

53. Friedman J, Alm EJ. 2012. Inferring correlation networks from genomic survey data. 651

PLoS Comput Biol 8:e1002687. 652


https://doi.org/10.1101/2019.12.12.874966


31

54. Csardi G, Nepusz T. 2006. The igraph software package for complex network research. 653

InterJournal Complex Systems:1695. 654

55. Guimera R, Amaral LAN. 2005. Functional cartography of complex metabolic networks. 655

Nature 433:895-900. 656

56. Oksanen J, Blanchet F, Friendly M, Kindt R, Legendre P, McGlinn D, Minchin P, O'Hara 657

R, Simpson G, Solymos P, Stevens M, Szoecs E, Wagner H. 2018. Vegan: Community 658

ecology package. R package version 2.4-6.:39pp. 659

57. McMurdie PJ, Holmes S. 2013. phyloseq: An r package for reproducible interactive 660

analysis and graphics of microbiome census data. PLoS One 8(4): e61217. 661

58. R Core Team. 2018. R: A language and environment for statistical computing. R 662

foundation for statistical computing, vienna, austria. URL https://www.R-project.Org. 663

59. Bates D, Machler M, Bolker BM, Walker SC. 2015. Fitting linear mixed-effects models 664

using lme4. J Stat Softw 67:1-48. 665

60. Ferreira JA. 2007. The Benjamini-Hochberg method in the case of discrete test statistics. 666

Int J Biostat 3:Article 11. 667

61. Coleman-Derr D, Desgarennes D, Fonseca-Garcia C, Gross S, Clingenpeel S, Woyke T, 668

North G, Visel A, Partida-Martinez LP, Tringe SG. 2016. Plant compartment and 669

biogeography affect microbiome composition in cultivated and native Agave species. 670

New Phytol 209:798-811. 671

62. Poli A, Lazzari A, Prigione V, Voyron S, Spadaro D, Varese GC. 2016. Influence of 672

plant genotype on the cultivable fungi associated to tomato rhizosphere and roots in 673

different soils. Fungal Biol 120:862-872. 674


https://doi.org/10.1101/2019.12.12.874966


32

63. Gottel NR, Castro HF, Kerley M, Yang ZM, Pelletier DA, Podar M, Karpinets T, 675

Uberbacher E, Tuskan GA, Vilgalys R, Doktycz MJ, Schadt CW. 2011. Distinct 676

microbial communities within the endosphere and rhizosphere of Populus deltoides roots 677

across contrasting soil types. Appl Environ Microbiol 77:5934-5944. 678

64. Badri DV, Vivanco JM. 2009. Regulation and function of root exudates. Plant Cell 679

Environ 32:666-681. 680

65. Dennis PG, Miller AJ, Hirsch PR. 2010. Are root exudates more important than other 681

sources of rhizodeposits in structuring rhizosphere bacterial communities? FEMS 682

Microbiol Ecol 72:313-327. 683

66. van der Heijden MG, Schlaeppi K. 2015. Root surface as a frontier for plant microbiome 684

research. Proc Natl Acad Sci USA 112:2299-300. 685

67. Berg G, Opelt K, Zachow C, Lottmann J, Gotz M, Costa R, Smalla K. 2006. The 686

rhizosphere effect on bacteria antagonistic towards the pathogenic fungus Verticillium 687

differs depending on plant species and site. FEMS Microbiol Ecol 56:250-261. 688

68. Reazin C, Baird R, Clark S, Jumpponen A. 2019. Chestnuts bred for blight resistance 689

depart nursery with distinct fungal rhizobiomes. Mycorrhiza 29:313-324. 690

69. Nasholm T, Huss-Danell K, Hogberg P. 2000. Uptake of organic nitrogen in the field by 691

four agriculturally important plant species. Ecology 81:1155-1161. 692

70. Reeve JR, Smith JL, Carpenter-Boggs L, Reganold JP. 2008. Soil-based cycling and 693

differential uptake of amino acids by three species of strawberry (Fragaria spp.) plants. 694

Soil Biol Biochem 40:2547-2552. 695


https://doi.org/10.1101/2019.12.12.874966


33

71. U'Ren N. 2007. Types, amounts, and possible functions of compounds released into the 696

rhizosphere by soil-grown plants, p 1-21, The rhizosphere 697

doi:10.1201/9781420005585.ch1. CRC Press. 698

72. Beattie GA. 2018. Metabolic coupling on roots. Nat Microbiol 3:396-397. 699

73. Zhalnina K, Louie KB, Hao Z, Mansoori N, da Rocha UN, Shi SJ, Cho HJ, Karaoz U, 700

Loque D, Bowen BP, Firestone MK, Northen TR, Brodie EL. 2018. Dynamic root 701

exudate chemistry and microbial substrate preferences drive patterns in rhizosphere 702

microbial community assembly. Nat Microbiol 3:470-480. 703

74. Warschefsky EJ, Klein LL, Frank MH, Chitwood DH, Londo JP, von Wettberg EJB, 704

Miller AJ. 2016. Rootstocks: Diversity, domestication, and impacts on shoot phenotypes. 705

Trends Plant Sci 21:418-437. 706

75. Malamy JE. 2005. Intrinsic and environmental response pathways that regulate root 707

system architecture. Plant Cell Environ 28:67-77. 708

76. Rousk J, Baath E, Brookes PC, Lauber CL, Lozupone C, Caporaso JG, Knight R, Fierer 709

N. 2010. Soil bacterial and fungal communities across a pH gradient in an arable soil. 710

ISME J 4:1340-1351. 711

77. Cao X, Zhao D, Xu H, Huang R, Zeng J, Yu Z. 2018. Heterogeneity of interactions of 712

microbial communities in regions of Taihu Lake with different nutrient loadings: A 713

network analysis. Sci Rep 8:8890. 714

78. Carlson CJ, Zipfel CM, Garnier R, Bansal S. 2019. Global estimates of mammalian viral 715

diversity accounting for host sharing. Nat Ecol Evol 3:1070-1075. 716


https://doi.org/10.1101/2019.12.12.874966


34

79. McGeachie MJ, Sordillo JE, Gibson T, Weinstock GM, Liu YY, Gold DR, Weiss ST, 717

Litonjua A. 2016. Longitudinal prediction of the infant gut microbiome with dynamic 718

bayesian networks. Sci Rep 6:20359. 719

80. Penalver Bernabe B, Cralle L, Gilbert JA. 2018. Systems biology of the human 720

microbiome. Curr Opin Biotechnol 51:146-153. 721


https://doi.org/10.1101/2019.12.12.874966


35

FIG 1 A systematic diagram representing the Phenotype-OTU network analysis (PhONA), 722

where an OTU-OTU association network (A) was combined with the nodes selected based on a 723

random forest model (B) to create PhONA (C). 724

725

FIG 2 Enriched and depleted OTUs across tomato rootstock genotypes (nongraft BHN589, 726

selfgraft BHN589, and BHN589 grafted on two hybrid rootstocks (RST-04-106 and Maxifort)) 727

evaluated for the rhizosphere (A) and the endosphere (B), using OTU counts from self-grafts and 728

non-grafts as controls. All the tests were adjusted to control the false discovery rate (FDR, p < 729

0.01) using the Benjamini-Hochberg method. Each point represents an OTU labeled at the genus 730

level and colored based on phylum, and the position along the x axis represents the abundance 731

fold-change contrast with controls (except for the selfgraft vs. nongraft comparison, where the 732

nongraft treatment was used as a control for the contrast). 733

734

FIG 3 Phenotype-OTU network analysis (PhONA) of endosphere fungal taxa for BHN589 735

grafted on Maxifort. Node color indicates the phylum, except that the tomato-color node 736

represents yield associated with the rootstock. Nodes connected to the rootstock yield node with 737

black links are taxa that were predictive of rootstock yield, where dotted and solid lines indicate 738

negative and positive associations (Pearson's correlation coefficient) with the yield node, 739

respectively. Red and blue links represent negative and positive associations, respectively, 740

between OTUs. Nodes are labeled with the lowest available taxonomic information. Square-741

shaped and star-shaped nodes represent DAOTU and key nodes detected in role analyses, 742

respectively. 743


https://doi.org/10.1101/2019.12.12.874966


36

FIG 4 Partitioning of endosphere fungal OTUs according to their network roles. Nodes were 744

divided into four categories based on within-module degree and among-module connectivity. 745

The blue dashed line represents a threshold value (0.62) for among-module connectivity, and the 746

red dashed line represents a threshold value (2.5) for within-module degree. Nodes were 747

categorized as peripherals, connectors, module hubs, and network hubs. Node color indicates 748

rootstock treatment. 749

750

FIG 5 Partitioning of rhizosphere fungal OTUs according to their network roles. Nodes were 751






757

FIG S1 Relative abundance of endosphere and rhizosphere fungi at the phylum level recovered 758

from four tomato rootstock treatments: nongraft BHN589, selfgraft BHN589, and BHN589 759

grafted on two hybrid rootstocks (RST-04-106 and Maxifort). Each individual bar represents a 760

rootstock treatment, and the colored area within the bar represents the relative abundance of the 761

corresponding phylum. 762

763

FIG S2 Comparison of overall fungal diversity (A) and richness (B) associated with tomato 764

rootstock genotypes and controls, evaluated in the endosphere and rhizosphere. The plot is 765

divided by the four tomato rootstock treatments: nongraft BHN589, selfgraft BHN589, and 766


https://doi.org/10.1101/2019.12.12.874966


37

BHN589 grafted on two hybrid rootstocks (RST-04-106 and Maxifort). Shannon entropy and 767

species richness, measures of community diversity, were both higher for Maxifort (p < 0.005) 768

compared to the self-graft and RST-04-106 in the rhizosphere, while there was not evidence for a 769

difference in Shannon entropy in the endosphere (p = 0.634). Treatment means were separated 770

using the "difflsmeans" function as specified in the lmerTest package in R. Tests for boxplots 771

sharing a letter or letter case type had p > 0.05. 772

773

FIG S3 Non-metric multidimensional scaling (NMDS) ordination plot of samples labeled by 774

tomato rootstock (nongraft BHN589, selfgraft BHN589, and BHN589 grafted on two hybrid 775

rootstocks (RST-04-106 and Maxifort)), compartment (endosphere or rhizosphere), and study 776

site, based on the Bray-Curtis dissimilarity distance matrix of fungal OTUs. Color indicates 777

rootstock treatment, shape indicates study site, and size indicates compartment. Ellipses 778

surrounding the samples indicate the 95% CI of the endosphere and rhizosphere sample 779

centroids. 780

781

FIG S4 Number of DAOTUs in a contrast analysis, evaluated for the endosphere and 782

rhizosphere compartments for four tomato rootstock treatments: nongraft and selfgraft BHN589, 783

and BHN589 grafted on two hybrid rootstocks (Maxifort and RST-04-106). The green color in 784

each bar represents the number of enriched taxa, and the red color represents the number of 785

depleted taxa. The number of differentially changed taxa was greater for the endosphere than for 786

the rhizosphere. Among the contrast pairs, hybrid rootstocks had a greater number of enriched 787

taxa compared to depleted taxa. However, the number of depleted taxa was higher compared to 788


https://doi.org/10.1101/2019.12.12.874966


38

enriched taxa in the controls. Among the treatments, Maxifort had the highest number of 789

DAOTUs in both compartments. 790

791

FIG S5 Phenotype-OTU network analysis (PhONA) of endosphere fungal taxa for tomato 792

rootstock treatments: (A) nongraft and (B) selfgraft BHN589, and BHN589 grafted on (C) RST-793

04-106. Node color indicates the phylum, except that the tomato-color node represents yield 794

associated with the rootstock. Nodes connected to the rootstock yield node with black links are 795

taxa that were predictive of rootstock yield, where dotted and solid lines indicate negative and 796

positive associations with the yield node, respectively. Red and blue links represent negative and 797

positive associations, respectively, between OTUs. Nodes are labeled with the lowest available 798

taxonomic information. Square-shaped and star-shaped nodes represent DAOTUs and key nodes 799

detected in role analyses, respectively. 800

801

FIG S6 Phenotype-OTU network analysis (PhONA) of rhizosphere fungal taxa for tomato 802

rootstock treatments: (A) nongraft and (B) selfgraft BHN589, and BHN589 grafted on two 803

hybrid rootstocks ((C) Maxifort and (D) RST-04-106). Node color indicates the phylum, except 804

that the tomato-color node represents yield associated with the rootstock. Nodes connected to the 805

rootstock yield node with black links are taxa that were predictive of rootstock yield, where 806

dotted and solid lines indicate negative and positive associations with the yield node, 807



shaped and star-shaped nodes represent DAOTUs and key nodes detected in role analyses, 810

respectively. 811


https://doi.org/10.1101/2019.12.12.874966


39

TABLE S1 Sites included in the study, their soil type, and geographic coordinates. 812

813

TABLE S2 Results of the multivariate permutational analysis of variance (PERMANOVA) for 814

fungal taxon abundance data. Permutation was based on the Bray-Curtis distance matrix 815

generated for root associated fungal communities at the OTU level from four tomato rootstock 816

treatments: nongraft and selfgraft BHN589, and BHN589 grafted on two hybrid rootstocks 817

(Maxifort and RST-04-106) (1000 permutations). P values < 0.05 are in bold. 818

819

TABLE S3 Network attributes and links observed in the fungal association networks for four 820

tomato rootstock treatments: nongraft and selfgraft BHN589, and BHN589 grafted on two hybrid 821

rootstocks (Maxifort and RST-04-106) in each of rhizosphere and endosphere compartments. 822


https://doi.org/10.1101/2019.12.12.874966


1

1

2 FIG 1 A systematic diagram representing the Phenotype-OTU network analysis (PhONA), 3

where an OTU-OTU association network (A) was combined with the nodes selected based on 4

random forest model (B) to create PhONA (C). 5

6

Phenotype

OTU12

OTU5

OTU9

OTU16

OTU7OTU1

OTU2

OTU4

OTU5

OTU6

OTU7

OTU8OTU9

OTU10

OTU11OTU12

OTU13OTU14

OTU15OTU16

OTU17

OTU18OTU19

Phenotype

OTU12 OTU5

OTU9

OTU16

OTU7

OTU1

OTU4

OTU6

OTU10

OTU13

OTU14

OTU15

OTU17

OTU18

OTU19+ =

(A) OTU-OTU association network based on SparCC

(B) OTUs predictive of host phenotype, based on random forest model

(C) PhONA predicting candidate taxa based on association network (A) and phenotype-predictive OTUs (B)


https://doi.org/10.1101/2019.12.12.874966


2

(A) 7

8 9


https://doi.org/10.1101/2019.12.12.874966


3

(B) 10

11 FIG 2 Enriched and depleted OTUs across tomato rootstock genotypes (nongraft BHN589, 12

selfgraft BHN589, and BHN589 grafted on two hybrid rootstocks (RST-04-106 and Maxifort)) 13

evaluated for the rhizosphere (A) and the endosphere (B), using OTU counts from self-grafts and 14

non-grafts as controls. All the tests were adjusted to control the false discovery rate (FDR, p < 15

0.01) using the Benjamini-Hochberg method. Each point represents an OTU labeled at the genus 16

level and colored based on phylum, and the position along the x axis represents the abundance 17

fold change contrast with controls (except for the selfgraft vs. nongraft comparison, where the 18

nongraft treatment was used as a control for the contrast). 19

20


https://doi.org/10.1101/2019.12.12.874966


4

21

FIG 3 Phenotype-OTU network analysis (PhONA) of endosphere fungal taxa for BHN589 22

grafted on Maxifort. Node color indicates the phylum, except that the tomato-color node 23

represents yield associated with the rootstock. Nodes connected to the rootstock yield node with 24

black links are taxa that were predictive of rootstock yield, where dotted and solid lines indicate 25

negative and positive associations (Pearson's correlation coefficient) with the yield node, 26




respectively. 30

Maxifortg__Mortierella

g__Mortierellap__Ascomycota_unclassified

g__Haematonectria

g__Gibberella

g__Rhizophagus

g__Rhizophagus

g__unclassified_Cercozoa

f__Pyronemataceae_unclassified

g__Thanatephorus

g__unclassified_Hypocreales

k__Fungi_unclassified

o__Pleosporales_unclassified

g__unclassified_Ustilaginales

p__Ascomycota_unclassified

g__unclassified_Cercozoa g__Thanatephorus

o__Pezizales_unclassified


g__Monosporascus g__Mortierella


g__Alternaria

g__Olpidium

g__Olpidium

g__Monographella

g__Wallemia

g__Mortierella

g__Alternaria

f__Dipodascaceae_unclassified

g__Mortierella

g__Crinipellis


g__Guehomyces

g__Podospora

g__Exophiala

g__Acremonium

g__Phoma

g__Rhizophagus

c__Agaricomycetes_unclassified






g__Hydnomerulius

c__Dothideomycetes_unclassified

g__Occultifur

g__Gibberella

g__Ceratobasidium


g__unclassified_Lycoperdaceae

f__Pleosporaceae_unclassified

g__Hydnomerulius


g__unclassified_Orbiliaceae

g__Rhizophagus



g__Cryptococcus


g__Mortierella

g__Conocybe

g__Septoglomusg__Glomus


c__Zygomycota_class_Incertae_sedis_unclassified

g__unclassified_Auriculariales


g__Conocybeg__Rhizophagus

g__Rhizophagus

g__Coprinopsis


g__Coprinopsis

g__Fusarium

f__Glomeraceae_unclassified

g__Malassezia

g__Rhizophagus

g__Septoglomus


g__unclassified_Auriculariales


g__Glomus


g__Glomus

g__Glomus


g__Glomus



g__Rhizophagus

c__Sordariomycetes_unclassified

MaxifortAscomycotaBasidiomycotaCercozoaChytridiomycotaGlomeromycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


5

31

32 33 FIG 4 Partitioning of endosphere fungal OTUs according to their network roles. Nodes were 34






40

o__Mortierellales

o__Pezizales

o__Saccharomycetales

o__Olpidiales

o__Pleosporales

o__Chaetothyriales


o__Cantharellales

k__Fungi_unclassifiedo__PezizalesModule hubs Network hubs

Peripherals Connectors-2

-1

0

1

2

3

0.00 0.25 0.50 0.75 1.00Among-module connectivity

With

in-m

odul

e de

gree

RootstocksNongraftSelfgraftRST-04-106Maxifort


https://doi.org/10.1101/2019.12.12.874966


6

41 42 43

44 45 FIG 5 Partitioning of rhizosphere fungal OTUs according to their network roles. Nodes were 46






52


o__Mortierellales

o__Mortierellales

o__Hypocreales


o__Tremellales

o__Capnodiales

o__Xylariales

o__Ustilaginales

o__Pleosporales

o__Hypocreales

o__Pezizales


o__Wallemiales

o__Tremellales

o__Olpidiales


o__Hypocreales

o__Trichosporonales


o__Phallales


o__Hypocrealeso__Pleosporales


o__Mortierellales

o__Pezizales

o__Basidiobolales

o__Dothideomycetes_order_Incertae_sedis

k__Fungi_unclassifiedModule hubs Network hubs

Peripherals Connectors-2

0

2

0.00 0.25 0.50 0.75 1.00Among-module connectivity

With

in-m

odul

e de

gree

RootstocksNongraftSelfgraftRST-04-106Maxifort


https://doi.org/10.1101/2019.12.12.874966


7

53

54 FIG S1 Relative abundance of endosphere and rhizosphere fungi at the phylum level recovered 55

from four tomato rootstock treatments: nongraft BHN589, selfgraft BHN589, and BHN589 56

grafted on two hybrid rootstocks (RST-04-106 and Maxifort). Each individual bar represents a 57

rootstock treatment, and the colored area within the bar represents the relative abundance of the 58

corresponding phylum. 59

60


https://doi.org/10.1101/2019.12.12.874966


8

(A) 61

62 63


https://doi.org/10.1101/2019.12.12.874966


9

64

(B) 65

66 FIG S2 Comparison of overall fungal diversity (A) and richness (B) associated with tomato 67

rootstock genotypes and controls, evaluated in the endosphere and rhizosphere. The plot is 68

divided by the four tomato rootstock treatments: nongraft BHN589, selfgraft BHN589, and 69

BHN589 grafted on two hybrid rootstocks (RST-04-106 and Maxifort). Shannon entropy and 70

species richness, measures of community diversity, were both higher for Maxifort (p < 0.005) 71

compared to the self-graft and RST-04-106 in the rhizosphere, while there was not evidence for a 72

difference in Shannon entropy in the endosphere (p = 0.634). Treatment means were separated 73

using the "difflsmeans" function as specified in the lmerTest package in R. Tests for boxplots 74

sharing a letter or letter case type had p > 0.05. 75

76


https://doi.org/10.1101/2019.12.12.874966


10

77

78 FIG S3 Non-metric multidimensional scaling (NMDS) ordination plot of samples labeled by 79

tomato rootstock (nongraft BHN589, selfgraft BHN589, and BHN589 grafted on two hybrid 80

rootstocks (RST-04-106 and Maxifort)), compartment (endosphere or rhizosphere), and study 81

site, based on the Bray-Curtis dissimilarity distance matrix of fungal OTUs. Color indicates 82

rootstock treatment, shape indicates study site, and size indicates compartment. Ellipses 83

surrounding the samples indicate the 95% CI of the endosphere and rhizosphere sample 84

centroids. 85

86


https://doi.org/10.1101/2019.12.12.874966


11

87 FIG S4 Number of DAOTUs in a contrast analysis, evaluated for the endosphere and 88

rhizosphere compartments for four tomato rootstock treatments: nongraft and selfgraft BHN589, 89

and BHN589 grafted on two hybrid rootstocks (Maxifort and RST-04-106). The green color in 90

each bar represents the number of enriched taxa, and the red color represents the number of 91

depleted taxa. The number of differentially changed taxa was greater for the endosphere than for 92

the rhizosphere. Among the contrast pairs, hybrid rootstocks had a greater number of enriched 93

taxa compared to depleted taxa. However, the number of depleted taxa was higher compared to 94

enriched taxa in the controls. Among the treatments, Maxifort had the highest number of 95

DAOTUs in both compartments. 96

97


https://doi.org/10.1101/2019.12.12.874966


12

A)98

99 100

Nongraft


g__Gibberella




g__Ramicandelaberk__Fungi_unclassified



g__Mortierellag__Phallus




g__Alternariag__Phoma

g__Mortierella

g__Olpidium

g__Haematonectria

g__Thanatephorus

g__Thanatephorus

g__Monographella

g__Mortierella

g__Mortierella

g__Mortierella

g__Alternaria

g__Mortierella

g__Mortierella

g__unclassified_Ceratobasidiaceae


g__Mortierella

g__Podospora

g__Botryodontia

g__Coprinus



k__Fungi_unclassifiedg__Rhizophagus


g__Mortierella

g__Mortierella



g__Fusarium

g__Blastobotrys



g__Mortierella

g__Rhizophagus

g__unclassified_Bionectriaceae



g__Rhizophagus


g__Ramicandelaber




g__Rhizophagus

g__Rhizophagus


g__Mortierella

g__Fusarium

g__Rhizophagus

g__Rhizophagus

g__Rhizophagus



c__Eurotiomycetes_unclassified


p__Basidiomycota_unclassified


g__Cylindrocarpon

g__Thanatephorus




g__Thanatephorus



NongraftAscomycotaBasidiomycotaCercozoaChytridiomycotaGlomeromycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


13

B)101

102

Selfgraft

g__Gibberella

g__Mortierella

g__Malassezia

g__Olpidium

g__Mortierella

g__Mortierella

g__Fusarium

g__Scedosporium

k__Fungi_unclassifiedg__Thanatephorus



g__Alternaria

g__Monosporascus


g__Thanatephorusp__Basidiomycota_unclassified

g__Acremonium





g__Alternaria

g__Phoma

g__Mortierella

g__Haematonectria

g__Thanatephorus

g__Thanatephorus

g__Monographella

g__Mortierella




g__Mortierella


g__Monographella


g__Guehomyces

g__Mortierella

g__Davidiella

g__Exophiala

g__Botryodontia


g__Ramicandelaber




g__Gymnoascus

g__Phoma

g__Rhizophagus

g__Fusarium


g__Wallemia

g__Spizellomyces


g__unclassified_Orbiliaceaeg__Debaryomyces

g__Preussia


g__Edenia

g__Mortierella



g__Rhizophagus

g__Ceratobasidiumk__Fungi_unclassified

g__Pochonia

g__Mortierella

f__Phaeosphaeriaceae_unclassified

c__Zygomycota_class_Incertae_sedis_unclassifiedg__Rhizophagus

g__Leptospora


f__Stephanosporaceae_unclassified


g__Pulvinula

g__unclassified_Pleosporales


p__Glomeromycota_unclassified

g__unclassified_Spizellomycetaceae

g__Rhizophagus

g__Bipolaris

g__Cyphellophora

SelfgraftAscomycotaBasidiomycotaCercozoaChytridiomycotaGlomeromycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


14

C)103

104 105 FIG S5 Phenotype-OTU network analysis (PhONA) of endosphere fungal taxa for tomato 106

rootstock treatments: (A) nongraft and (B) selfgraft BHN589, and BHN589 grafted on (C) RST-107

04-106.Node color indicates the phylum, except that the tomato-color node represents yield 108

associated with the rootstock. Nodes connected to the rootstock yield node with black links are 109

taxa that were predictive of rootstock yield, where dotted and solid lines indicate negative and 110

positive associations with the yield node, respectively. Red and blue links represent negative and 111

positive associations, respectively, between OTUs. Nodes are labeled with the lowest available 112

taxonomic information. Square-shaped and star-shaped nodes represent DAOTUs and key nodes 113

detected in role analyses, respectively. 114

115

RST-04-106


g__Alternaria

g__Alternaria g__unclassified_Hypocreales



f__Nectriaceae_unclassified


g__Phoma

g__Mortierella

g__Olpidium

g__Wallemia

g__Thanatephorus

g__Thanatephorus

g__Olpidium

g__Mortierella


g__Gibberella



g__Mortierella

g__Monographella



g__Mortierella

g__Exophiala

g__Botryodontia

g__Candida

g__Cordyceps




g__Mortierella

g__Paurocotylis


g__Cryptococcus

g__Septoglomus


g__Coprinopsis



g__Mortierella




g__Septoglomus

c__Zygomycota_class_Incertae_sedis_unclassified


g__Cyathus


g__Ceratobasidium


g__Mortierella

g__Malassezia

g__Septoglomus






g__Thanatephorus

f__Polyporaceae_unclassified

g__Piriformospora

g__Capronia

g__Rhizophlyctis


g__Mortierella

g__Hyphodontia

RST-04-106AscomycotaBasidiomycotaCercozoaChytridiomycotaGlomeromycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


15

A)116

117 118

Nongraft

g__Mortierella

g__Wallemia

g__Mortierella

g__Alternaria

g__Scedosporium



g__Blastobotrys


g__Cryptococcusf__Dipodascaceae_unclassified



g__Lysurus


g__Phallus

g__Cryptococcus

g__Mortierella

g__unclassified_Microascales

g__Gibberella



g__Phoma

g__Olpidium

g__Thanatephorus

g__Thanatephorus

g__Cryptococcus

g__Olpidium

g__Monographella

g__Mortierella

g__Mortierella


g__Mortierella


o__Capnodiales_unclassified

g__Alternaria

g__Mortierella

g__Wallemia

g__Mortierella



g__Guehomyces

g__Davidiella

g__Bipolaris

g__Exophiala

g__Botryodontia

g__Coprinus

g__Ramicandelaber




g__Acremonium

g__Mortierella

g__Acremonium


g__Alternaria

g__Ustilago

g__Mortierella


g__Melampsora

g__Wallemia


g__Sporobolomyces


g__Occultifur

g__Trichosporon

g__Mortierella

g__unclassified_Stephanosporaceae


g__Gibberella

g__Clonostachys

g__Scedosporium


g__Lysurus



g__unclassified_Pyronemataceae

g__Pseudogymnoascus





g__Basidiobolus

g__unclassified_Sporidiobolales

g__Rhodosporidium

k__Fungi_unclassifiedg__Rhodotorula



g__Bullera

g__Mortierella

g__Myrothecium

g__Mortierella


g__Ramicandelaber

g__Occultifur

g__Rhizophagus


g__Bipolaris

f__Sporidiobolales_family_Incertae_sedis_unclassified

g__Mortierella

g__unclassified_Nectriaceae

g__Fusarium


g__Lysurus

g__Cryptococcus

g__Cystofilobasidium

g__Parasola

g__Lysurus


g__Septoglomus

g__Lachancea

g__unclassified_Rozellomycota


g__Mortierella

g__Parasola

g__Candida

f__Chionosphaeraceae_unclassified

g__Coprinopsis


o__Hypocreales_unclassified

g__Coprinopsis

g__Scedosporium

g__Acremonium


g__Minimedusa

g__Hypoxylon


p__Ascomycota_unclassifiedg__Mortierella

g__Hyphoderma

g__Hypocrea

g__Occultifur

g__unclassified_Polyporales


g__Funneliformis


NongraftAscomycotaBasidiomycotaCercozoaChytridiomycotaGlomeromycotaRozellomycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


16

B)119

120 121

Selfgraft


g__Guehomyces


g__Scedosporium

g__Bipolaris

g__Mortierella

g__Rhodosporidium

g__Pseudogymnoascus

g__Fusarium

g__Trichoderma


g__Phoma

g__Cryptococcusg__Scedosporium






g__Alternaria

g__Phoma

g__Olpidium

g__Haematonectria

g__Wallemia

g__Thanatephorus

g__Cryptococcus

g__Olpidium

g__Monographella

g__Mortierella

g__Mortierella

c__Sordariomycetes_unclassified c__Sordariomycetes_unclassified

g__Mortierella

g__Cryptococcus

g__Acremonium



g__Alternariag__Mortierella



g__Mortierellag__Davidiella

g__Bipolaris

g__Acremonium

g__Exophiala

g__Botryodontia

g__Metarhizium


g__Ramicandelaber


g__Acremonium

g__Gymnoascus


g__Acremonium

g__Mortierella

g__Acremonium

g__Rhodotorula

g__Mortierella

g__Alternaria

g__Ustilago


g__Fusarium

f__Phaeosphaeriaceae_unclassified

g__Rhodotorula


g__Fusarium

g__Blastobotrys

g__Occultifur g__Trichosporon

g__Gibberella

g__Peziza

g__Scedosporium


g__Lysurus

g__unclassified_Bionectriaceaeg__unclassified_Pyronemataceae

g__Mortierella




f__Microascaceae_unclassified

g__Myrothecium

g__Botryodontia


g__Occultifur

g__Phallus

g__Bipolaris

g__Ceratobasidium

f__Sporidiobolales_family_Incertae_sedis_unclassified

g__Candida

g__Sporobolomyces

g__Preussiag__unclassified_Cercozoa

g__Stachybotrys

g__unclassified_Gymnoascaceae


g__Umbelopsis

g__Olpidium

g__unclassified_Zygomycota


g__Trichosporon


g__unclassified_Zygomycota


g__unclassified_Pleosporaceae


g__Lysurus

g__Mortierellag__Candida

g__Occultifur

g__Hypocrea


g__Mortierella

g__Tomentella

g__Cryptococcus

g__Mortierella

g__Lycoperdon


SelfgraftAscomycotaBasidiomycotaCercozoaChytridiomycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


17

C)122

123 124

RST-04-106

g__Cryptococcus

g__Olpidium

g__Pseudozyma

g__Mortierella

g__Phallus

g__Sporobolomyces

g__Wallemia

g__Phoma

g__Pseudozyma

g__Clonostachys


g__Mortierella

g__Bipolaris

g__Cryptococcus

g__Mortierella

g__Fusarium


g__Cryptococcus


g__Ramicandelaber

g__Ramicandelaber g__Cryptococcus



g__Cordyceps

g__Mortierella




g__Alternaria

g__Phoma

g__Olpidium

g__Haematonectria

g__Wallemia

g__Thanatephorus

g__Thanatephorus

g__Monographella



g__Mortierella

g__Acremonium



g__Alternaria

g__Mortierella


g__Wallemia

g__Mortierella

g__unclassified_Microascales

g__Guehomyces

g__Davidiella

g__Bipolaris

g__Acremonium

g__Exophiala

g__Botryodontia

g__Candida

g__Metarhizium



g__Gymnoascus

g__Acremonium

g__Mortierella

g__Phallusg__Acremonium

g__Rhodotorula

g__Mortierella

g__Alternaria


g__Cryptococcus

g__Wallemia

g__Rhodotorula


g__Occultifur

g__Trichosporon

g__Olpidium



g__Septoglomus

g__Gibberella

g__Scedosporium


g__Rhizophagus

g__Lysurus

g__Trichosporon

g__Pseudogymnoascus


g__Mortierella

g__Wallemia

g__Rhizophagus


g__Wallemia


g__Debaryomyces



g__Wallemia

g__Mortierella

g__Septoglomus

g__Mortierella


g__Myrothecium


g__Stagonosporag__Scedosporium


c__Agaricomycetes_unclassifiedg__unclassified_Sporidiobolales

g__Preussia


g__Stachybotrys

g__Mortierella

g__Rhizophagus

g__Lysurus

g__Phallus

g__Wallemia

g__Lysurus

g__Umbelopsisg__Septoglomus

g__Ascobolus

g__Mortierella


g__Spizellomyces


g__Candida

g__Sterigmatomyces

g__Lysurus


g__Entoloma


g__Wallemia



g__Rhizophlyctis

g__Capronia

f__Stictidaceae_unclassified



g__unclassified_Chaetothyriales


g__Arachnion

g__Wallemia

g__Sporisorium

RST-04-106AscomycotaBasidiomycotaCercozoaChytridiomycotaGlomeromycotaRozellomycotaUnclassifiedZygomycota


https://doi.org/10.1101/2019.12.12.874966


18

D)125

126 FIG S6 Phenotype-OTU network analysis (PhONA) of rhizosphere fungal taxa for tomato 127

rootstock treatments: (A) nongraft and (B) selfgraft BHN589, and BHN589 grafted on two 128

hybrid rootstocks ((C) Maxifort and (D) RST-04-106). Node color indicates the phylum, except 129

that the tomato-color node represents yield associated with the rootstock. Nodes connected to the 130

rootstock yield node with black links are taxa that were predictive of rootstock yield, where 131

dotted and solid lines indicate negative and positive associations with the yield node, 132




respectively. 136

137

Maxifort



g__Mortierella

g__Phoma

g__Alternaria

g__Acremonium

g__Scedosporium


p__Basidiomycota_unclassifiedf__Phaeosphaeriaceae_unclassified


g__Wallemia


g__Mortierella

g__Mortierella

g__Mortierella



g__Cryptococcus

g__Thanatephorus

g__Stachybotrys

g__unclassified_Nectriaceae

g__Tetracladium

g__Mortierella

g__Scedosporium

g__Septoglomus

g__Cryptococcus

g__Rhodotorula

g__Thanatephorus

g__Galactomyces

g__Phallus

g__Bullera

g__Rhodotorulag__Westerdykella


g__Candidag__Acremonium

g__Cystofilobasidium g__Mortierella


g__Bipolaris




g__Alternaria

g__Mortierella

g__Olpidium

g__Haematonectria

g__Cryptococcus

g__Fusarium

g__Olpidium

g__Monographella


g__Mortierella

g__Mortierella


g__Mortierella



g__Mortierella


g__Guehomyces

g__Davidiella

g__Bipolaris

g__Acremonium

g__Exophialag__Metarhizium

g__unclassified_Ustilaginalesg__Acremonium


g__Coprinellus


g__Acremonium

g__Phoma

g__Acremonium

g__Rhodotorula

g__Rhizophagus


g__Ustilago

g__Mortierella

g__Mortierella

g__Cryptococcus

g__Fusarium k__Fungi_unclassifiedk__Fungi_unclassified

g__Fusarium


g__Hydnomeruliusk__Fungi_unclassified

g__Occultifur

g__Trichosporon



g__Pseudaleuria

g__Ceratobasidium

g__Wallemia

g__Scedosporium

g__Pseudozyma


g__Lysurus



g__Trichosporon

g__Pseudogymnoascus


g__Hydnomerulius


g__Mortierella


g__Basidiobolus


g__Rhodosporidium





g__Pseudaleuria

g__Trichoderma

g__Septoglomusg__Mortierella

g__Myrothecium

g__Mortierella



g__Phallus

g__Stagonospora

g__Coprinopsis

g__Sporobolomyces



g__Coprinopsis


g__Fusarium

g__unclassified_Sporidiobolalesf__Nectriaceae_unclassified

g__Lysurus

g__Cryptococcus

g__Phallus

g__unclassified_Gymnoascaceae



g__Lysurus

g__Umbelopsis

g__Septoglomus





g__Preussia

g__Sporobolomyces

g__Lysurus

g__Malassezia



o__Agaricales_unclassified


g__Coprinopsis

g__Glomus

g__Stagonospora



g__Glomus

g__Ganoderma


g__Coprinopsis



g__Aureobasidium


g__Pyrenochaetopsis

g__Wallemia



g__Acremonium

o__Polyporales_unclassified

f__Hypocreaceae_unclassified

g__Glomusg__Hypocrea


g__Glomus



Maxifort

Ascomycota

Basidiomycota

Cercozoa

Chytridiomycota

Glomeromycota

Rozellomycota

Unclassified

Zygomycota


https://doi.org/10.1101/2019.12.12.874966


19

138 139 TABLE S1 Sites included in the study, their soil type, and geographic coordinates. 140 141

Study sites Location Latitude Longitude Soil type

Olathe Horticulture Research and Extension Center (OHREC) Johnson County, KS 38.88N 94.99W Chase silt loam

Common Harvest Douglas County, KS 38.96N 95.20W Eudora-Kimo complex

142


https://doi.org/10.1101/2019.12.12.874966


20

TABLE S2 Results of the multivariate permutational analysis of variance (PERMANOVA) for 143

fungal taxon abundance data. Permutation was based on the Bray-Curtis distance matrix 144

generated for root associated fungal communities at the OTU level from four tomato rootstock 145

treatments: nongraft and selfgraft BHN589, and BHN589 grafted on two hybrid rootstocks 146

(Maxifort and RST-04-106) (1000 permutations). P values < 0.05 are in bold. 147

Factor Sum of Squares % Explained P value

Rootstocks 1.24 2.07 < 0.01

Compartment 5.36 8.92 < 0.001

Study_Site 5.01 8.34 < 0.001

Year 3.23 5.38 < 0.001

Rootstocks:Compartment 0.58 0.97 0.972

Rootstocks:Study_Site 1.39 2.32 < 0.01

Compartment:Study_Site 1.59 2.65 < 0.001

Rootstocks:Year 1.00 1.66 0.111

Compartment:Year 1.00 1.66 < 0.001

Study_Site:Year 1.46 2.43 < 0.001

Rootstocks:Compartment:Study_Site 0.55 0.91 0.998

Rootstocks:Compartment:Year 0.49 0.82 1.000

Rootstocks:Study_Site:Year 1.22 2.03 < 0.01

Compartment:Study_Site:Year 0.60 1.00 < 0.01

Rootstocks:Compartment:Study_Site:Year 0.60 1.00 0.989

Residuals 34.75 57.86

Total 60.07 100.01 148


https://doi.org/10.1101/2019.12.12.874966


21

TABLE S3 Network attributes and links observed in the fungal association networks for four tomato rootstock treatments: nongraft 149

and selfgraft BHN589, and BHN589 grafted on two hybrid rootstocks (Maxifort and RST-04-106) in each of rhizosphere and 150

endosphere compartments. 151

152

153

Compartment Rootstocks Node Edge Node Degree Density Modules

(SA) Negative

Edge Positive

Edge Negative/Positive

Edge

Endosphere

Nongraft 605 155 0.26 0.0004 18 5 150 0.03 Selfgraft 600 145 0.24 0.0004 20 9 136 0.07

RST-04-106 584 156 0.27 0.0005 17 10 146 0.07 Maxifort 700 221 0.32 0.0005 20 36 185 0.19

Rhizosphere

Nongraft 950 381 0.40 0.0004 17 82 299 0.27 Selfgraft 995 312 0.31 0.0003 19 67 245 0.27

RST-04-106 991 397 0.40 0.0004 23 58 339 0.17 Maxifort 1175 785 0.67 0.0006 18 289 496 0.58


https://doi.org/10.1101/2019.12.12.874966


integration of system phenotypes in microbiome networks to ... · 12/12/2019 · 112 phona first...

Documents