wortley2015.pdf

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofitpublishers, academic institutions, research libraries, and research funders in the common goal of maximizing access tocritical research.

Evolution of Angiosperm Pollen. 1. IntroductionAuthor(s): Alexandra H. Wortley, Hong Wang, Lu Lu, De-zhu Li, and StephenBlackmoreSource: Annals of the Missouri Botanical Garden, 100(3):177-226.Published By: Missouri Botanical GardenDOI: http://dx.doi.org/10.3417/2012047URL: http://www.bioone.org/doi/full/10.3417/2012047

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in thebiological, ecological, and environmental sciences. BioOne provides a sustainableonline platform for over 170 journals and books published by nonprofit societies,associations, museums, institutions, and presses.

Your use of this PDF, the BioOne Web site, and all posted and associated contentindicates your acceptance of BioOnes Terms of Use, available at www.bioone.org/page/terms_of_use.

Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should bedirected to the individual publisher as copyright holder.
http://dx.doi.org/10.3417/2012047http://www.bioone.org/doi/full/10.3417/2012047http://www.bioone.orghttp://www.bioone.org/page/terms_of_usehttp://www.bioone.org/page/terms_of_use

EVOLUTION OF ANGIOSPERM Alexandra H. Wortley,2 Hong Wang,3 Lu Lu,3

POLLEN. 1. INTRODUCTION1 De-zhu Li,3 and Stephen Blackmore2

ABSTRACTThis paper is the first in a series that documents the diversity, distribution, and evolution of palynological characters across

angiosperms in a contemporary phylogenetic context, using modern optimization methods. The objectives of the series are: (1) todescribe the diversity of pollen morphologies across the angiosperms; (2) to estimate ancestral palynological character states,diagnostic characters, and synapomorphies for monophyletic groups; (3) to highlight and interpret inferred patterns andprocesses of evolution in palynological characters; and (4) to provide a framework for the placement of enigmatic taxa (includingfossil taxa) based on pollen morphology. This first paper examines the methods available for such a study and presents anoverview of palynological characters across angiosperms as a whole. Using a well-supported, recent, molecular phylogeny, weconsider the effects of coding strategy, method of optimization, and starting tree topology upon inference of trait evolution.Coding strategy and optimization method had significant effects upon inferred ancestral character states, the latter probably dueto the different evolutionary models applied. Phylogenetic topology had little effect upon inferred ancestral character states,because the uncertainty in topology at this level involved only nodes where few character state changes occurred. Severalpalynological characters showed consistent, structured patterns in the context of phylogeny: angiosperms are distinguished fromother seed plants by character states including supratectal elements echinate and less than 1 lm in size, and infratectumstructure columellate; eudicots, as recognized in previous studies, may be defined by globose, isopolar, radially symmetricalgrains with three equatorial apertures. We present a framework for the remainder of the series, in which the angiosperms aredivided into nine monophyletic and paraphyletic groups each having a similar level of pollen variability, and a set ofrecommendations for the analysis of these groups. The series will provide a reference for future palynological and systematicstudies and an approach that may be replicated for other character sets.Key words: Ancestral character states, diagnostic characters, evolution, flowering plants, optimization, palynology,

phylogeny, synapomorphy, trait evolution.

There is a long and illustrious history of studying angiosperms (Donoghue & Doyle, 1989), now widelypalynological characters in a broad evolutionary known as the eudicots (e.g., Doyle & Hotton, 1991;context (e.g., Wodehouse, 1935; Erdtman, 1943, APG, 1998; APG II, 2003; APG III, 2009). It is1944, 1945, 1946, 1948, 1952; Walker & Doyle, likely that further synapomorphic palynological1975; Nowicke & Skvarla, 1979). The remarkable characters remain to be discovered through ourpotential of palynological characters in systematic growing understanding of both plant phylogeny andclassification, recently discussed by Blackmore pollen development and structure.(2000), lies in the fact that pollen grains contain a The fact that pollen characters are both variableremarkable degree of information for their minute and characteristic of taxa was perhaps first noted bysize, which is in turn due to a number of factors Grew (1682). However, it was more than a centuryincluding their highly resistant (almost indestructi- later that Brown (1811) recognized they mightble) sporopollenin wall, great abundance, and therefore be useful for classification (Fig. 1A). Theincredible, heritable, variety of form (Blackmore, first attempt at a full classification of the flowering2007). The importance of pollen characters in plants based on pollen grains was made by von Mohlsystematics has been highlighted through the recog- (1835; Fig. 1B). Von Mohls work seems to have beennition of a monophyletic tricolpate clade within largely overlooked until it was rediscovered by

1 The authors gratefully acknowledge Robert K. Jansen and colleagues for permission to use their plastid genomephylogeny, and Mark Chase for useful discussions on phylogenies and the Angiosperm Phylogeny Group III classification.We also thank the herbaria of BM, E, K, KUN, and US for their help with providing pollen samples, the late John Skvarla forprovision of some SEM micrographs, and Herve Sauquet for supplying the original for Figure 1F. AHW acknowledges theSibbald Trust at the Royal Botanic Garden Edinburgh for funding to visit China during the project. This work was supportedby grants from the National Natural Science Foundation of China (Grant number 31270272), Major International JointResearch Project of the National Natural Science Foundation of China (Grant number 31320103919), and the open funds ofthe Key Laboratory of Biodiversity and Biogeography, Kunming Institute of Botany, Chinese Academy of Sciences (Grantnumber KLBB201202).

2 Royal Botanic Garden Edinburgh, 20a Inverleith Row, Edinburgh EH3 5LR, United Kingdom. Author forcorrespondence: [email protected]

3 Laboratory of Biodiversity and Biogeography, Kunming Institute of Botany, Chinese Academy of Sciences, Kunming,Yunnan, Peoples Republic of China. [email protected].

doi: 10.3417/2012047

ANN. MISSOURI BOT. GARD. 100: 177226. PUBLISHED ON 16 MARCH 2015.

Wodehouse (1935). Wodehouse himself published a The 1970s saw a gradual progression from simpleseries of beautifully illustrated works on the structure descriptions of pollen morphology to hypotheses ofof pollen grains and their use in classification (e.g., evolutionary change in pollen characters. MullerFig. 1C), focusing on Asteraceae (Wodehouse, 1926, (1970) published an evolutionary scenario showing1928a, 1928c, 1929a, 1929b, 1930, 1936). Shortly relationships between pollen grain types linked toafter this, Erdtman (1943) produced the first of his plant evolution as then understood. Walker andmany surveys of pollen and spores throughout the Doyle (1975), using some of the first SEM micro-land plants. His work, Pollen Morphology and Plant graphs (e.g., Fig. 1E), continued in this evolutionaryTaxonomy (Erdtman, 1952), remains the most vein at the Bases of Angiosperm Phylogeny sympo-comprehensive reference guide to angiosperm pollen. sium (Walker, 1975) at the 24th annual meeting ofIt not only describes and illustrates pollen grain the American Society of Plant Taxonomists, present-structure in great detail (e.g., Fig. 1D), but also draws ing similar diagrams to Muller (1970) and highlight-parallels between families based on palynological ing evolutionary trends correlated with relativecharacteristics. Sadly, only a small proportion of advancement of taxa for a wide range of pollenErdtmans work has been published, with descrip- morphological characters (Walker & Doyle, 1975).tions and illustrations of a great range of taxa Later, at the Evolutionary Significance of the Exineremaining only in the archives of the Swedish meeting (Ferguson & Muller, 1976), Van CampoMuseum of Natural History in Stockholm. (1976) explored trends of palynological evolution in

Figure 1. Selected images from the history of palynological study, using examples from the Proteaceae and other basaleudicot groups. A. Knightia excelsa R. Br. (Proteaceae), pollen plurimum auctum (mature pollen), reproduced from fig. 7 ofBrowns 1811 thesis, On the Proteaceae of Jussieu. B. Fumaria spicata L. (Papaveraceae), graine mouille (hydrated grain),reproduced from fig. 24 of von Mohls 1835 work, Sur la structure et les formes des graines de pollen. C. Platanusoccidentalis L. (Platanaceae), polar view, reproduced from plate IX, fig. 5, of Wodehouses 1935 work, Pollen Grains: TheirStructure, Identification and Significance in Science and Medicine. D. Grevillea bipinnatifida R. Br. (Proteaceae), fig. 208,reproduced with permission of the Palynological Laboratory, Swedish Museum of Natural History, Stockholm, from the 1972corrected reprint edition of Erdtmans 1952 epic, Pollen Morphology and Plant Taxonomy. E. Euplassa inaequalis Engl.(Proteaceae), triporate pollen, polar view, an early SEM micrograph (fig. 3B) reproduced with kind permission of MissouriBotanical Garden Press from Walker and Doyles 1975 study, The bases of angiosperm phylogeny: Palynology. F.Sorocephalus crassifolius Hutch. (Proteaceae), whole grain, polar view, reproduced with kind permission of the American Societyfor Plant Taxonomy from fig. 15A of Sauquet and Cantrills 2007 comprehensive evolutionary study, Pollen diversity andevolution in Proteoideae (Proteales: Proteaceae).

178 Annals of theMissouri Botanical Garden

Volume 100, Number 3 Wortley et al. 1792015 Evolution of Angiosperm Pollen. I. Introduction

various angiosperm taxa. Shortly afterward, Nowicke for instance based on molecular genetic data, theand Skvarla (1979) published a study of pollen interpretation of characters as ancestral or derived,morphology and development across 650 species synapomorphy or homoplasy can thus be seen as afocusing on the Centrospermae (roughly correspond- circular argument (for discussion see de Queiroz,ing to todays Caryophyllales). Their study deter- 1996); others argue that recognition of characters asmined diagnostic characters for higher-order group- synapomorphic or homoplastic is intrinsically, andings, albeit without explicit reference to phylogenetic rightly, bound up in the process of phylogeneticrelationships (Nowicke & Skvarla, 1979). These estimation (most famously Patterson, 1982). Second-evolutionary approaches were complemented by ly, perhaps due to the large amount of data required,Mullers ongoing documentation of angiosperm pollen analysis has been limited to the family level andin the fossil record (Muller, 1970, 1981, 1984). In below. Even had it been possible to explore further insum, advances in phylogenetic, palynological, and the taxonomic hierarchy, the results would have beenpaleobotanical research, including the relatively new erroneous because understanding of angiospermtechniques of SEM and transmission electron mi- phylogeny at the time (e.g., Takhtajan, 1980;croscopy (TEM), provided a solid foundation for Cronquist, 1981; Dahlgren, 1988; Thorne, 1992)future interpretations of pollen and spore morphology differed considerably from present-day interpreta-in a phylogenetic and ontogenetic context, provided tions. For instance, Walker and Doyle (1975) utilizedby concurrent progress in the visualization of the broadly phenetic classifications of Cronquistdevelopment and numerical (cladistic) methods for (1988) and Takhtajan (1980), which significantlyphylogenetic analysis (for discussion see Funk & influenced their inference of trait evolution.Stuessy, 1978). Since the 1990s, there has been a concerted shiftTogether, these precipitated in the early 1980s a toward classifications that prioritize monophyly. For

burst of cladistic analyses of pollen characters (alone flowering plants, the widely accepted Angiospermor in combination with other morphological charac- Phylogeny Group (APG) classifications are basedters) at familial and lower levels. Early examples largely on molecular phylogenetic studies conductedinclude the works of Blackmore (1982) on the over the last two decades (APG, 1998; APG II, 2003;Scorzonerinae (Asteraceae), Lee and Park (1982; APG III, 2009). The first APG classification (APG,Oleaceae), Donoghue (1983; Viburnum L. [Caprifo- 1998) recognized 462 families in 40 putativelyliaceae]), Kress and Stone (1983; Heliconiaceae), monophyletic orders and a few higher groups. FiveBlackmore and Cannon (1983; Morinaceae), and years on, subsequent changes in family circumscrip-Linder (1984; Restionaceae; see Appendix 1 for more tion and ordinal definition were incorporated (APG II,recent examples). In such studies, morphological 2003), giving a classification with 457 families in 45character states distributed across groups in a orders. The most recent edition (APG III, 2009)phylogeny are interpreted as homoplasies, synapo- consolidates these into 415 families among anmorphies, autapomorphies, or plesiomorphies, all of increased 58 orders. Overall, the APG classificationswhich may also be diagnostic for the purposes of have proven extremely stable. In the most recent, theclassification. These concepts enable the generation authors note that we do not see the APGof evolutionary hypotheses regarding ancestral and classification as continuing to mutate for thederived states and the direction and pattern of indefinite future. . . . We hope the classification . . .evolution in the observed traits. It may also be will not need much further change (APG III, 2009:possible to identify correlations between the evolution 106).of palynological character states, life history strate- Enabled by such advances in phylogeneticgies and syndromes, and evolutionary-ecological classification, pollen development, electron micros-factors such as the origin or diversification of copy (Blackmore, 2007), and methods for optimiza-particular pollinating animal groups. tion of character states upon phylogenies (e.g.,Notwithstanding the conceptual and factual ad- Cunningham et al., 1998; Mooers & Schluter, 1999;

vance represented by the first cladistic studies, they Omland, 1999; Pagel, 1999a, 1999b; Huelsenbeck etwere limited in two respects. Firstly, these early al., 2000; Huelsenbeck & Bollback, 2001; Nielsen,studies typically tended to combine the use of 2002; Pagel et al., 2004; Ronquist, 2004; Vander-palynological characters as evidence for the recon- poorten & Goffinet, 2006; Ekman et al., 2008), westruction of phylogenetic relationships with their can now re-evaluate the evolution of palynologicalinterpretation in the context of the resultant phylog- characters across the flowering plants within a moreeny (for an early exception see Jansen et al., 1991). In explicit phylogenetic, evolutionary, and ontogeneticthe absence of an independent phylogenetic estimate, context. Similar angiosperm-wide studies have suc-


cessfully been carried out on the evolution of floral reconstruction; these may equally be applicable tosymmetry (Jabbour et al., 2009) and secondary studies of trait evolution. The ancestral (sensuchemicals (Esteban et al., 2009). For palynological Bininda-Emonds et al., 1998) or groundplan (sensucharacters, evolutionary studies have been conducted Yeates, 1995) method uses fossils, ontogenetic data,in the Proteaceae (Sauquet & Cantrill, 2007, an SEM or prior phylogenetic analysis to estimate a primitivemicrograph from which is shown in Fig. 1F) and state in each monophyletic higher taxon. TheAsteraceae (Blackmore et al., 2009) which, with ca. democratic (Bininda-Emonds et al., 1998) or com-20,000 species (Funk et al., 2005), comprises 7% posite (Brusatte, 2010) method uses data from a10% of flowering plants. Both Sauquet and Cantrill sample of taxa within each higher taxon, which is(2007) and Blackmore et al. (2009) demonstrated that then represented by the most common state foundin the context of robust phylogenies, pollen charac- (based on the hypothesis that common equalsters can provide diagnostic and synapomorphic primitive). The exemplar method (Bininda-Emondscharacters at a range of hierarchical levels, enabling et al., 1998) employs one or more representatives ofthe formulation of adaptive hypotheses. Furthermore, each higher taxon, chosen either for their inferredthe recovery of ancestral character state combinations plesiomorphic status or simply their availability. Amay better enable the placement of fossil pollen fourth possibility, to code higher taxa as polymorphic,grains in a phylogenetic context, facilitating greater is not usually considered suitable for reconstructionunderstanding of the date and location of key events of phylogenetic trees but is, however, open toin angiosperm evolution. character studies based upon existing phylogenies.In a series of papers, of which this is the first, we Like the democratic method, it uses a sample of taxa

will survey the evolution of pollen morphologies from across the higher taxon but does not assume thatthroughout angiosperms, based on the most recent common equals primitive; the higher taxon is codedand reliable phylogenetic trees and incorporating an for all states found in its representatives. We termextensive literature review and new palynological this the comprehensive method.data. Our objectives are: (1) to describe and The choice of analytical methods for characterdocument the diversity of pollen morphologies across state optimization has received less attention in thethe angiosperms; (2) to estimate ancestral palynolog- literature than that for reconstructing phylogeniesical character states for monophyletic groups and (Pagel, 1997; Cunningham et al., 1998), and theirthereby to identify diagnostic characters and syna- implementation remains controversial (Yang, 2006;pomorphies; (3) to highlight and interpret inferred Wiens et al., 2007). As with phylogeny reconstruc-patterns and processes of evolution in palynological tion, both parsimony and model-based methods exist.characters; and (4) to provide a framework for the Maximum parsimony (MP), the most widely usedphylogenetic placement of enigmatic taxa based on method (Cunningham et al., 1998), estimates ances-pollen morphology (this is especially relevant to tral states so as to minimize the number of stateextinct taxa since pollen is often particularly well changes implied across the phylogeny as a whole.preserved in the fossil record). This first paper This method has proven informative and further hascompares the available methods for investigating trait the advantage that it is insensitive to the density ofevolution in a phylogenetic context and provides an taxon sampling within a clade for which a trait isoverview analysis of angiosperm palynological char-

fixed (D. Barker, 2011, pers. comm.). However, MPacters and their evolution at a broad scale. We hope

by definition assumes that changes of state are rare,this will stimulate similar review and analysis of other

i.e., a maximum of one per branch (Cunningham etclasses of morphological and anatomical characters.

al., 1998), which may be a weakness particularlywhen analyzing characters at higher taxonomic levels.

THE INFLUENCE OF METHODOLOGICAL FACTORS IN THE In addition, the chance of a state change occurring isANALYSIS OF TRAIT EVOLUTION considered the same on every branch regardless ofThree fundamental methodological factors may length, and MP does not provide probabilistic

influence the inference of ancestral morphological estimates of error or support for ancestral states,character states upon phylogenetic trees: coding showing only the most parsimonious even if onlystrategy, analytical (optimization) method, and start- marginally so. In contrast, probabilistic, model-baseding tree topology (i.e., phylogenetic uncertainty). approaches (including maximum likelihood [ML] andCoding morphological characters becomes non- Bayesian methods) can incorporate measures of

trivial for higher (supraspecific) taxa when more than uncertainty and/or probability in ancestral statesone character state is found therein. Three main and, where an appropriate model is selected, makecoding strategies are recognized in phylogeny use of branch length information (Cunningham et al.,


1998). However, these methods tend to be sensitive should not be considered settled until comparisonsto taxon sampling and to the accuracy of branch have been made with nuclear sequence data.lengths (Torices, 2010). Of these, ML techniquesoptimize ancestral states so as to maximize the MATERIALS AND METHODSprobability of the observed states in the terminal taxa,given a single explicit model of evolution (for STUDY TAXA

morphological data this is usually simple, such as a Study taxa were determined by the startingone-parameter Markov [Mk-1] model, in which the phylogeny (Fig. 2), which was taken from the worksingle parameter is the rate of change). ML analysis of Jansen et al. (2007), and comprised 64 taxa in 59assumes a constant evolutionary rate across the tree, genera (61 angiosperm species in 56 genera plusi.e., the likelihood of change along a branch is three gymnosperm outgroups). This phylogeny wasproportional to its length (Cunningham et al., 1998). selected from among those in the recent literatureFor phylogenies based on molecular evidence, this because it provided broad taxonomic coverage acrossmay be problematic if the morphological events of a wide range of genetic loci, was fully resolved andinterest are uncoupled from molecular change, such well supported, featured branch lengths calculatedas when morphological change is concentrated during with model-based methods, was based on the entirespeciation events. The technique sometimes known chloroplast genome (minimizing issues of rapidas empirical Bayesian (EB) inference is considered evolution at certain loci in certain taxa), and isby some authors to be superior to ML methods (e.g., highly congruent with other recent (including subse-Yang, 2006). Like ML, it uses an Mk-1 model of quent) phylogenetic estimates (e.g., Soltis et al.,transition rates derived from the data, but unlike the 2011). Pollen data were, if possible, taken from theformer it subsequently derives posterior probability same 64 species as sampled in the startingdistributions for ancestral states. Fully (hierarchical) phylogeny. However, where palynological data wereBayesian (HB) techniques have the advantage that not available for these taxa, data were obtained fromthey can incorporate multiple alternative base closely related species or genera (see Appendix 2 forphylogenies, by calculating relative probability for a list of substituted taxa).each possible character state at each node on a set oftrees sampled from a distribution (usually generated

POLLEN CHARACTERSduring Bayesian inference of phylogeny). However,some studies have suggested that the advantages of Character state information was obtained from theoptimization over multiple phylogenies are minimal literature, with taxon-specific palynological studies(Hanson-Smith et al., 2010). Possible disadvantages providing the primary reference where available. Weof Bayesian methods include their dependence on the also drew upon the broader studies of Sampsonchoice of prior probabilities, which may be somewhat (2000) for magnoliids, Zavada (1983) and Furnessarbitrary. Both ML and Bayesian methods are limited and Rudall (2003) for monocots, Punt (1984;in applicability since they can at present be Umbelliferae) and Blackmore et al. (2009; Aster-conducted only on fully resolved phylogenies and aceae), as well as Wodehouse (1935), Erdtmandatasets without polymorphisms. (1952), Walker and Doyle (1975), and Nowicke andAlthough the starting tree we employed for Skvarla (1979; cf. Appendix 1). When possible,

comparisons of coding and optimization method character state information was taken directly from(Jansen et al., 2007) was the most reliable and original published images (light microscopy [LM],suitable phylogeny available for this study, it is not SEM, and TEM) and measurements rather thanthe only possible representation of relationships written descriptions, in order to achieve a morebetween angiosperm taxa. Recent studies have found consistent description and terminology for charactera variety of patterns among some of the earliest states across all taxa. All observations pertain tobranching lineages of the angiosperm tree, and even a mature pollen grains. Differences in treatment andthorough analysis of whole genome data, aimed hydration state (for instance, of fresh materialspecifically at resolving relationships among basal compared to that taken from herbarium specimens)angiosperms, failed to reach resolution on this issue can have a significant effect upon the nature of(Moore et al., 2007). Although the majority of plastid- certain characters such as size and shape. Therefore,based studies have agreed on the topology found by acetolyzed (Erdtman, 1960) grains were used wher-Jansen et al. (2007), since topology is crucially ever possible to ensure comparability of all charactersimportant to polarizing character states we take the both within this series and with the majority ofconservative view of Soltis et al. (2008) that this palynological literature.

Figure 2. Angiosperm phylogeny redrawn from a maximum likelihood (ML) analysis of plastid genome data for 64 taxa,provided by Jansen et al. (2007), with branch lengths proportional to number of inferred nucleotide substitutions and taxonnames as in the original article. This phylogeny served as the basis for most of the analyses presented here. Modified topologiesused to test the effects of phylogenetic uncertainty in optimization of ancestral character states are described in the text. Cladesrepresented by at least two taxa in the phylogeny are labeled if mentioned in the text.



As this is a preliminary study, character selection and Orchidaceae). In some taxa, notably thefocused upon characters previously identified as Cyperaceae, pollen grains are initially borne invariable and potentially informative at intra-ordinal tetrads but reduced by abortion to single grains,level, including dispersal unit, polarity, symmetry, known as cryptotetrads or pseudomonads (Erdtman,shape, size, aperture characters, exine structure and 1952), soon after meiosis; we treat these structures assculpture, and external features such as orbicules tetrads.and viscin threads (Erdtman, 1952; Van Campo &Lugardon, 1973; Walker & Doyle, 1975; Skvarla et Polarity and symmetry. Polarity (Character 2)

al., 1978; Nowicke & Skvarla, 1979; El-Ghazaly & and symmetry (Character 3) are determined with

Chaudhary, 1993; Schols et al., 2001; Merckx et al., reference to development in the tetrad: each pollen

2008). For the purposes of data exploration, an grain has a polar axis running outward from the

inclusive approach was taken in which as many center of the tetrad, and two poles where this axis

characters as possible were coded, including those meets the surface of the grain, a proximal pole at the

that appeared to be somewhat subjective, poorly center of the tetrad and a distal pole at the outer

characterized, or potentially non-independent (note surface (Erdtman, 1952). Based on this, pollen grains

that non-independence of characters is only a may be apolar (Fig. 3A), isopolar (Fig. 3D),

concern for phylogenetic reconstruction, not for subisopolar, or heteropolar (Fig. 3E). In apolar

investigation of character evolution upon an inde- grains, such as many polyaperturate grains, orienta-

pendently derived phylogeny). For very complex tion and polarity are impossible to establish at the

features such as exine sculpture it remains difficult free microspore stage. In isopolar grains, the two

to design a perfect strategy for delimitation of hemispheres are identical, as in most tri-aperturate,

characters and character states. spheroidal, eudicot grains. Subisopolar grains are notPalynology has several complex descriptive vocab- entirely symmetrical about the equatorial plane, e.g.,

ularies, developed for the efficient portrayal of due to the addition of viscin threads on one face ofnumerous features. The terminology used here follows the grain. In this paper, we treat subisopolar grains asthat of Punt et al. (2007). One widespread shortcut isopolar, since they are usually the same inused to avoid lengthy descriptions and complex fundamental structure. In heteropolar grains, theterminology is the pollen type.; Pollen types two polar hemispheres differ distinctly in shape ordescribe the entire nature of a pollen grain (size, apertures (Erdtman, 1952).shape, apertures, exine stratification, surface sculp- When viewed from either pole, pollen grainsture, etc.) in a single phrase such as type A, or typically have either two (bilateral symmetry; Fig.Liguliflorae-type. While this approach is useful for 3F) or more than two (radial symmetry; Fig. 3G)the purpose of identifying isolated, dispersed pollen planes of symmetry running perpendicular to thegrains, as in fossil studies, to analyze pollen grain equator. The former is typical of monocolpate grains,evolution, it is more appropriate to break down pollen the latter of most tri-zono-aperturate grains (Erdtman,types and compound terms into their most basic 1952).components, each conveying a character. For exam-

Shape. The basic three-dimensional shapes ofple, instead of the compound term ana-zona-sulcatepollen grains (Character 4) are often categorized aswe would describe aperture shape (zonate), orienta-either globose or boat-shaped. Other shapes, such astion (latitudinal), structure (simple), and positionthreadlike, are restricted to a few species with(between the equator and distal pole). The palyno-reduced exines. Boat-shaped pollen, found in manylogical characters and character states investigatedgrains with a single, colpate, polar aperture, has awere as follows, organized in the sequence ofshort polar axis and unequal equatorial axes (WalkerErdtman (1952; cf. Appendix 3 and Fig. 3).& Doyle, 1975; Fig. 3F). Globose pollen, the most

Dispersal unit. The dispersal unit (Character 1) is common type, is approximately spheroidal to ellip-the arrangement in which pollen grains are found at soidal (e.g., Fig. 3A, D, E, GP). These two broadmaturity and dispersal (Punt, 1962), also known as states are included in our matrix in order to assessthe pollen unit (Walker & Doyle, 1975). States range whether the character may be informative. However,from monads (free grains; e.g., Fig. 3A) to permanent since many grains do not precisely fit such categories,dyads (fused pairs), tetrads (fused groups of four, in it has become more conventional to describe shape invarious orientations; Fig. 3B), polyads (fused in terms of two outlines: in equatorial and in polar view.defined multiples of four, to a maximum of 64; Fig. In equatorial view, grains may be defined by their3C), or pollinia (fused in larger, indeterminate shape class (Character 5; Erdtman, 1952), which isnumbers, to date only reported for Apocynaceae defined as the ratio of the lengths of their polar and

Figure 3. SEMs of pollen grains representative of some of the characters and character states used in this paper. For voucherdetails see Appendix 6. AP. Whole grains or clusters of grains. A. Plantago psyllium L. (Plantaginaceae), unknown view,illustrating dispersal unit (Character 1) as monads, pollen grains apolar (Character 2), basic shape (Character 4) globose, apertures(Character 8) many, aperture position (Character 9) global. B. Rhododendron wallichii Hook. f. (Ericaceae), dispersal unit(Character 1) as tetrads. C. Acacia nilotica (L.) Willd. ex Delile (Fabaceae), dispersal unit (Character 1) as polyads. D. Crepisnapifera (Franch.) Babc. (Asteraceae), equatorial view, pollen grains isopolar (Character 2), aperture membranes (Character 12;


arrowed) smooth, supratectal elements (Character 17) echinate. E. Acorus gramineus Sol. ex Aiton (Acoraceae), equatorial view,pollen grains heteropolar (Character 2), apertures (Character 8) one, aperture position (Character 9) distal. F. DioscoreanipponicaMakino (Dioscoreaceae), equatorial view, symmetry (Character 3) bilateral (dashed line marks single plane of symmetry),basic shape (Character 4) boat-shaped. G. Centrapalus pauciflorus (Willd.) H. Rob. (Asteraceae), polar view, symmetry(Character 3) radial, apertures (Character 8) three, aperture position (Character 9) equatorial, aperture membranes (Character 12;arrowed) granulate, supratectal elements (Character 17) echinate. H. Ulmus glabra Huds. (Ulmaceae), equatorial view, shapeclass (Character 3) oblate, tectum sculpture (Character 19) rugulate. I. Cabobanthus bullulatus (S. Moore) H. Rob. (Asteraceae),equatorial view, shape class (Character 3) subspheroidal, ectoapertures (Character 11) porate, exine (Character 29) lophate. J.Nouelia insignis Franch. (Asteraceae), equatorial view, shape class (Character 3) prolate, ectoapertures (Character 11) colpate. K. Tragopogon longifolius Heldr. & Sartori (Asteraceae), polar view, outline (Character 6) polygonal, apertures (Character 8) three,supratectal elements (Character 17) echinate. L. Oldenburgia paradoxa Less. (Asteraceae), polar view, outline (Character 6)more or less lobate, apertures (Character 8) three, tectum sculpture (Character 19) perforate. M. Poa bulbosa L. (Poaceae),oblique view, apertures (Character 8) one, aperture structure (Character 10) simple, tectum sculpture (Character 19) areolate. N.Scorzonera hispanica L. (Asteraceae), internal view of aperture, aperture structure (Character 10) compound, endoaperture(character not analyzed due to greater than 50% missing data; arrowed) lalongate, endexine (Character 23; asterisked) present. O. Illicium floridanum J. Ellis (Illiciaceae), polar view, ectoapertures (Character 11) syncolpate (reproduced with kind permissionof Elsevier from fig. 8 of Wang et al., 2009b). P. Aesculus hippocastanum L. (Sapindaceae), multiple grains, operculum(Character 13; arrowed) present. QX. Details of exine sculpture and structure. Q. Dioscorea pyrenaica Bubani & Bordere ex.Gren. (Dioscoreaceae), supratectal elements (Character 17) gemmate. R. Adenocaulon chilense Less. (Asteraceae), detail of exinesurface close to aperture, supratectal elements (Character 17) verrucate. S. Lathyrus pratensis L. (Fabaceae), tectum sculpture(Character 19) fossulate. T. Cullumia rigida DC. (Asteraceae), tectum sculpture (Character 19) reticulate. U. Croton argyratusBlume (Euphorbiaceae), tectum sculpture (Character 19) Croton-patterned. V. Croton argyratus, fractured exine in cross-section, tectum sculpture (Character 19; shown in cross-section) Croton-patterned. W. Dampiera stricta (Sm.) R. Br.(Goodeniaceae), tectum sculpture (Character 19) striate. X. Alfredia cernua (L.) Cass. (Asteraceae), fractured exine in crosssection, infratectum structure (Character 20) columellate.


equatorial axes (P/E). A P/E of less than 0.5 is Apertures. Apertures, found in the majority ofdefined as peroblate, 0.5 to 0.75 as oblate (Fig. 3H), species, have been recognized as fundamental0.75 to 1.33 as subspheroidal (Fig. 3I), 1.33 to 2.0 as features of pollen grains (Wodehouse, 1928b; Erdt-

prolate (Fig. 3J), and greater than 2.0 as perprolate man, 1952; Walker & Doyle, 1975). They comprise

(Erdtman, 1952). The widespread subspheroidal modifications of the pollen wall such as openings,

class may further be divided into four categories: thinnings, or thickenings of the exine or intine for thepurposes of interaction with the surrounding sub-suboblate (P/E 0.750.88), oblate-spheroidal (P/Estrate, including the emergence of pollen tubes0.881.0), prolate-spheroidal (P/E 1.01.14), andduring germination. Erdtman (1952) consideredsubprolate (P/E 1.141.33; Erdtman, 1952). In polarseven aspects of aperture morphology, some of whichview (amb; note that in heteropolar grains this doesmay be interdependent: number, position, structure,not necessarily coincide with the equatorial outline),shape, size, nature of the aperture membrane, and

the outline (Character 6) may be described as circularpresence or absence of an operculum. In this paper,

(Fig. 3I), elliptical, polygonal (Fig. 3K), concave-we consider all except aperture size. Note that while

polygonal (angular with concave sides), or lobateErdtman (1952) described a series of terms indicating

(curved with convex sides separated by indentations; both position and shape of apertures, includingFig. 3L). sulcus (a furrow-shaped aperture situated at the

distal pole), porus (a rounded aperture at theSize. We follow the convention of Erdtman

equator), and rugus (one of several, globally distrib-(1952) and Walker and Doyle (1975) in measuring

uted furrows), for the present analysis we prefer tosize (Character 7) as the length of the longest axis of maintain the two characters (position and shape) aseach pollen grain. As a continuous character, size can separate, for reasons described above.only arbitrarily be divided into states; furthermore, The number (Character 8) and position (Characterbecause size may vary with preparation of the grain, it 9) of apertures is correlated with developmentalhas been suggested that any states should be defined factors during and after meiosis (Blackmore & Crane,as separated at least by orders of magnitude (Walker 1988; Blackmore et al., 2007). In terms of number, in& Doyle, 1975). We adopt the following standard this study we use states zero, one (Fig. 3E, M), two,classes (Erdtman, 1952; Walker & Doyle, 1975): very three (Fig. 3G, K, L), four to six, seven to 12, andsmall (less than 10 lm), small (1024 lm), medium more than 12 (Fig. 3A). In terms of position,(2549 lm), large (5099 lm), very large (100199 apertures are defined as polar (proximal or distal;lm), and gigantic (greater than 200 lm). Fig. 3E), equatorial (Fig. 3G), or global (Fig. 3A).


Furthermore, for polygonal grains, the positioning of authors defined terms ectexine (or ektexine) andthe apertures relative to the apices of the grain may endexine based on the developmental origins andbe a useful character: in angulaperturate grains the staining properties of the layers. The endexine isapertures lie at the apices; in planaperturate grains usually single-layered, whereas the ectexine is oftenthey lie on the faces between the apices. multi-layered, comprising a tectum and infratectumAperture structure (Character 10) may be simple (which together correspond to the sexine of Erdtman

(Fig. 3M), comprising a single-layered thinning [1952]) and foot layer (which, together with the(ectoaperture) in the outer exine, or compound (Fig. endexine, comprises Erdtmans nexine). On the3N), in which there is also a thinning or opening surface of the tectum there may also be supratectal(endoaperture) in the inner exine. Occasionally, a elements.mesoaperture is present between the ecto- and Working inward from the external surface of theendoapertures; its presence or absence is treated pollen grain, we first characterize the nature of thehere as a binary character. supratectal elements, if present (Character 16). InIn terms of shape, the ectoaperture (Character 11) terms of shape (Character 17), they may be pilate

may be more or less porate (rounded; Fig. 3I), colpate (approximately cylindrical, greater in height than(furrowed; Fig. 3J), zonate (encircling the grain in a diameter), echinate (pointed, broader at base than atring), spiraperturate (encircling the grain in a spiral tip; Fig. 3D, G, K), gemmate (approximatelypattern), or syncolpate (with elongated apertures spheroidal; Fig. 3Q), or verrucate (rounded andfusing, usually at the poles; Fig. 3O). The endoa- flattened, greater in diameter than height; Fig. 3R).perture (not analyzed here due to high levels of Striate (parallel, elongated across the surface of themissing data) may be circular, endocingulate (encir- tectum) and rugulate (irregular, elongated across thecling the grain in a ring), lalongate (elliptical or surface of the tectum) supratectal elements were notrectangular with the longest axis parallel to the observed in the present study. Supratectal elementsequator; Fig. 3N), or lolongate (longest axis perpen- can also be categorized by size (Character 18), whichdicular to the equator). is here divided into two states, diameter greater thanThe thin membrane that usually covers an aperture 1 lm (corresponding to macro-elements, usually

(Character 12) may be smooth (psilate; Fig. 3D) or visible under LM) and diameter less than 1 lmbeset with granules (Fig. 3G). In certain taxa, such as (micro-elements, usually only visible under electronmany Poaceae, the aperture membranes are conspic- microscopy).uously thickened with an island of ectexinous Beneath the supratectal elements is usually amaterial toward the center, known as an operculum

sculptured tectum. In LM studies, the sculpturing(Character 13; Fig. 3P). The apertures may also each

patterns are often categorized as OL- or LO-patternbe surrounded by an annulus (Character 14), a

following Erdtman (1952). Seen under the electronconspicuous thickening of the exine, which may take

microscope, tectum sculpture forms are found to beseveral forms: aspidate (outer exine thickened and

numerous, varied, and not necessarily discrete orprotruding outward around the aperture), costate

mutually exclusive; furthermore, their interpretation(inner exine, endexine, or foot layer thickened and

depends, to some extent, on the nature of both theprotruding inward into the cytoplasm of the grain), orsupratectal elements above and the infratectum belowvestibulate (annulus subtended by a cavity).(Punt et al., 2007). Here, we categorize states as

Exine structure and sculpture. Most pollen grains follows (Character 19): perforate (with small, well-

are surrounded by an acetolysis-resistant sporopol- spaced openings less than 1 lm in diameter; Fig. 3L),

lenin exine (exceptions include many Zingiberales, foveolate (with large, circular, well-spaced holes

some aquatic angiosperms, and a few Lauraceae). The greater than 1 lm diameter), fossulate (with elongate,

exine is usually differentiated (Character 15) into irregular grooves; Fig. 3S), reticulate (with polygonal

distinct layers, for which there are two alternative openings separated by narrow muri; Fig. 3T), areolate

descriptive systems. Erdtman (1952) defined two (with raised polygonal areas separated by narrowlayers based on morphological observations only: an grooves; Fig. 3M), Croton L.-patterned (comprisingouter sexine and an inner nexine, both of which may groups of raised polygons arranged around a centralbe further subdivided into numbered layers. Howev- space; Fig. 3U, V), rugulate (with irregular, elongateer, when comparing pollen grains with differing elements greater than 1 lm long; Fig. 3H), striatenumbers of layers, this frequently results in non- (with elongate parallel elements and grooves orhomologous layers being given the same number. For spaces between; Fig. 3W), striato-reticulate (withthis reason, we adopt the less ambiguous system of elongate parallel elements and cross-links between),Fgri (1956) and Fgri and Iversen (1989). These and imperforate (without sculpture, sometimes re-


ferred to as a complete tectum sensu Erdtman COMPARISON OF METHODS FOR STUDYING TRAIT EVOLUTION[1952]). There also exists a unique tectum sculpture

In this paper, we test three aspects of thereported only for Amborella trichopoda Baill. (treated

methodology for estimating ancestral states of pollenin more detail by Lu et al., 2014, this issue), which

morphological characters upon a phylogeny of highercomprises small cupules constructed of coiled

taxa: coding strategy, optimization method, and basecylindrical strands (Sampson, 2000).

tree topology.Beneath the tectum lies the infratectum (Character

20), which is limited to one of three forms in almost Coding strategy. Because this study is based on aall seed plant species (Van Campo & Lugardon, phylogeny of higher-level (ordinal) relationships in1973): alveolate (restricted to gymnosperms and angiosperms inferred from character states (molecularextinct, non-angiosperm seed plants), columellate sequence data) for individual species, we tested three(restricted to angiosperms; Fig. 3X), and granulate methods of coding higher taxa for ancestral state(observed in some species of angiosperms, most optimization, represented in three different matrices:notably Annonaceae, and independently in some the species exemplar method (using mostly singlegymnosperms such as Gnetales). The infratectum exemplars; the possibility to select multiple exem-itself may in some taxa (e.g., many Asterales) be plars was not available as the terminal taxa were pre-distinguished into a number of layers (Character 21). selected); the comprehensive method (where possibleBelow the infratectum are the foot layer (Character coding at least six species from across the phyloge-

22) and endexine (Fig. 3N), if present (Character 23). netic and morphological range of each order, andThe endexine may differ in nature (Character 24: displaying all observed states, with taxa selected tocompact, spongy, or lamellar at maturity) and extent represent orders based on the delimitations of 2009);(Character 25: continuous, discontinuous, or solely and the democratic method (sensu Bininda-Emondsapertural). In some taxa (e.g., Asteraceae), one or et al. [1998]; a matrix created from the comprehen-more exine layers are separated by a cavity known as sive matrix but with polymorphic data points reduceda cavea, coded here as present or absent (Character to the most common state. When two states were26). Also in some taxa (most notably Asteraceae), the found to be equally prevalent, the state was coded aselements that make up the exine are themselves unknown). We did not test the ancestral methodperforated with tiny holes known as internal foramina because neither fossils, ontogenetic data, nor existing(Character 27). In other taxa, the exine is traversed by phylogenies were widely available. We also created aminute, radially orientated channels (microchannels; modified species exemplar matrix in which polymor-Character 28). The latter two characters are usually phic data points were reduced to the unknown statevisible only under TEM, and their presence or (?).absence is rarely reported in the literature. In certain Coding strategies were compared for ancestral statetaxa (including many Asteraceae such as Fig. 3I), the optimization using the MP algorithm implemented inexine may be folded into protruding ridges (lophae) Mesquite versions 2.62.72 (Maddison & Maddison,and depressions (lacunae; Character 29), forming a 2009), because this is the only method of analysispattern described as lophate (Wodehouse, 1935; Punt currently applicable to matrices containing polymor-et al., 2007). Pollen that is not fully lophate but bears phic cells, as found in two of the coding strategiessupratectal elements in a pattern as if upon lophae is tested. For consistency with tests of methodology,described as sublophate. they were tested on a single phylogeny (Jansen et al.,

2007). Consistency (ci) and retention indices (ri), asExternal structures. Two types of structures well as ensemble indices (CI, RI) were calculated for

external to the pollen grain, possibly of tapetal origin, the four matrices on this phylogeny.have been highlighted in previous studies. Viscinthreads (Character 30) are acetolysis-resistant, spo- Optimization method. We tested four methods forropollenin strands arising from the surface of the optimization of ancestral character states: MP, ML,exine, common in some Onagraceae and Ericaceae. EB, and HB. Maximum parsimony optimization wasOrbicules (Character 31) are granules of sporopol- conducted using Mesquite v.2.62.72 (Maddison &lenin found on the supratectal surface, sometimes Maddison, 2009), with unordered (Fitch) parsimonyreferred to as Ubisch bodies. Although frequently under the trace character history option, andreported from the surface of the anther locules, MacClade v.4.06 (Maddison & Maddison, 2000),usually in species with a secretory tapetum, we record using the trace all possible changes option.these only when they are found on the surface of the Maximum likelihood optimization was also conductedmature pollen grains themselves. in Mesquite, applying an Mk-1 model, with param-


eters estimated from the data and using the default data), or where all states were found to be equallysurvey interval (two optimizations at a coarseness of probable under Bayesian analysis, all states were1.0 and 10.0, respectively). Empirical and hierarchi- considered to be possible and, therefore, thecal Bayesian optimizations were carried out using optimization to be congruent with any state generatedBayesMultistate (Pagel et al., 2004) in BayesTraits by the other methods of analysis.(,http://www.evolution.rdg.ac.uk/BayesTraits.html.).Results were examined in Microsoft Excel. Since the Starting tree topology. Five alternative arrange-

three model-based optimization methods are at ments were tested, representing the most common

present applicable only to datasets without polymor- and significant variations found in recent, compre-

phic data points, to enable comparison across all four hensive phylogenetic studies (a wider range of

methods all were tested on the only two matrices that topologies will be investigated in the second paper

comply with this requirement: the democratic matrix in this series (see Lu et al., 2015), focusing on basal

and the modified species exemplar matrix (polymor- angiosperms): (1) Amborella trichopoda sister to all

phic data removed). The comprehensive matrix other angiosperms, with the Nymphaeales the next

contained so many polymorphic cells that their branching clade, Chloranthales sister to magnoliids,

removal would result in an unreasonably high and monocots sister to eudicotsthis topology was

proportion of missing data. For consistency with tests obtained, with strong support, by Jansen et al. (2007;

of coding strategy, they were tested on a single Fig. 2); (2) A. trichopoda sister to Nymphaeales,

phylogeny (Jansen et al., 2007). together sister to the remainder of angiosperms,

Testing multiple priors (see Appendix 4) indicated Chloranthales, magnoliids and eudicots forming a

that the most appropriate methodology for HB polytomy and together sister to monocotsthis

inference with our data was to implement the topology was obtained by Soltis et al. (2007); (3) A.

reversible jump hyper prior (rjhp) mechanism as trichopoda sister to Nymphaeales (sensu Soltis et al.,

recommended in BayesTraits. This was applied for all 2007), Chloranthales sister to magnoliids, and

characters, starting with a run of 100,000 iterations, monocots sister to eudicots (sensu alternative topol-

sampling every 100 generations and a burn-in of ogy in Jansen et al., 2007); (4) A. trichopoda sister to

1000 generations. Rjhp settings (prior distribution all other angiosperms (sensu Jansen et al., 2007),

type, range for uniform seeding distribution, and rate Chloranthales, magnoliids and eudicots forming a

deviation) were manipulated for each character to polytomy and together sister to monocots (sensu Soltis

ensure that the distribution of transition values et al., 2007); (5) relative positions of both A.

included the point estimates obtained from the EB trichopoda and the Nymphaeales, and Chloranthales,

analysis, and so as to obtain a mean acceptance value magnoliids, monocots, and eudicots, left unresolved.

of 20%40% (minimizing autocorrelation among Topologies were compared for ancestral state

successive states of the chain while still exploring optimization using the MP algorithm implemented

parameter space thoroughly; Pagel & Meade, 2006). in Mesquite versions 2.62.72 (Maddison & Maddi-

Tests suggested that sampling every 300 generations son, 2009). Model-based methods of analysis were

successfully avoided autocorrelation between adja- not appropriate because not all phylogenies were

cent reported results. Plots of posterior values associated with branch lengths. For consistency with

suggested that a burn-in period of anything greater tests of methodology, this was conducted using the

than 10,000 generations was ample. Thus, the final modified exemplar species matrix (polymorphic data

parameter settings for the HB analysis were to sample removed).

every 300 generations, with a burn-in period of20,000 generations for a net total of 5,000,000 RESULTS

generations (i.e., 5,020,000 including the burn-inPOLLEN CHARACTERS

period).At each node, the favored states (i.e., the most After initial documentation of 35 palynological

parsimonious, most likely or those with the greatest characters, three were removed due to extensiveposterior probability) generated by the four methods (greater than 50%) missing data (viz. aperturewere compared. Where MP optimization gave position in polar view, mesoaperture presence/multiple most parsimonious states, these were all absence, and endoaperture shape) and one due toconsidered equally supported and, thus, the method invariability (tectum presence/absence). In the re-to be congruent with other methods if any of the states sulting matrix coded according to the exemplaragreed. Similarly, where no single state was recon- method, five characters were parsimony-uninforma-structed for a node under ML (usually due to missing tive (and entirely invariant in the democratic matrix)


but were retained for their potential usefulness in of the eudicots, or between Ranunculales and the restmodel-based analyses (viz. Character 15, exine of the eudicots. With democratic coding (Fig. 4C), thedifferentiation; Character 26, cavea; Character 27, results are intermediate between these two extremes,internal foramina; Character 29, exine folding; with a clear transition from heteropolar to isopolarCharacter 30, viscin threads). The final matrices grains at the base of the eudicots, but further switches(Appendix 5) contained 31 characters (Appendix 3). to apolar and isopolar grains seen closer to the tips ofTo enable complete comparisons across coding the phylogeny.methods, all 31 characters were retained in the

Optimization method. The two matrices used forcomprehensive and democratic matrices, even whencomparison of optimization methods differed signif-they were invariant in these matrices.icantly. Of the 31 characters examined, 16 (52%)displayed different numbers of character states in the

COMPARISON OF METHODS FOR STUDYING TRAIT EVOLUTIONmatrix generated by the democratic method of coding

Coding strategy. The species exemplar matrix compared to that using exemplar species (Table 3),contained 137 cells (7%) missing data and 111 cells with the democratic matrix generally having fewer(6%) inapplicable, a total of 248 cells (13%) treated states than the exemplar species matrix. For eightas missing data by the analysis programs. Seventy-six characters, the democratic matrix contained only acells (4%) were polymorphic; when these cells were single unambiguous state. In general, the democratictreated as missing data to facilitate comparisons method of coding appeared to produce moreacross methods, the level of missing data increased to consistent results across different methods of ances-326 cells (16%). The comprehensive matrix con- tral character state optimization: for the majority oftained 42 cells (2%) missing data and 28 cells (1%) characters (23), the democratic matrix generatedinapplicable, a total of 70 cells (4%) treated as fewer total disparities between methods than themissing data by the analysis programs. Eight hundred species exemplar matrix (Table 4). The democraticand twenty cells (41%) were polymorphic. The matrix produced a greater number of differences fordemocratic matrix contained 97 cells (5%) missing only two characters (Character 2, polarity, anddata and 30 cells (2%) inapplicable, a total of 127 Character 8, aperture number), while the remainingcells (6%) treated as missing data by the analysis six characters generated the same total differencesprograms. By definition, this matrix contained no between methods with both matrices. Furthermore,polymorphic cells (see Table 1; Appendix 5). the democratic matrix produced fewer differences atAlthough the ci varied between matrices, ri and nodes for all six possible bilateral method compar-rescaled consistency indices (rc) for characters isons. Therefore, the following comparison of methodsfollowed generally similar patterns across the matri- is based on the species exemplar coding strategy, inces, with the same characters having high or low order to avoid underestimating possible discrepanciesindices in all three matrices (Table 2). Ensemble between methods.indices differed considerably, with CI 0.27, RI 0.5, Notable differences were observed for manyand RC (ensemble rescaled consistency index) 0.14 characters between the four different methods usingfor the exemplar species matrix, CI 0.51, RI 0.76, the species exemplar matrix (Tables 3, 4; Fig. 5). Byand RC 0.39 for the comprehensive matrix, and CI considering for each node the most parsimonious0.35, RI 0.72, and RC 0.25 for the democratic matrix. (from MP analysis), most likely (from ML analysis),When analyzed using MP, the differences in and most probable ancestral state (from EB and HB

inferred ancestral character state optimization be- analyses), it is possible to determine and quantify thetween the three coding strategies were often large, differences between inferred ancestral states acrossboth in terms of the location and number of character all methods (Table 4). It should be noted, however,state changes inferred (Table 1). In general, the that such a comparison is relatively crude, taking nospecies exemplar matrix invoked the greatest number account of the relative likelihood (or probability) ofof character state changes, followed by the demo- each ancestral state, which in some cases was onlycratic matrix. For example, Character 2 (pollen grain marginally greater than that for the next most likelypolarity; Fig. 4) shows a complex pattern of multiple state. The level of congruence between optimizationstate changes when coded using exemplar species methods differed among characters. For example, fourand optimized using MP (Fig. 4A). However, when characters showed identical patterns with all methodscoded comprehensively for orders, only a single state tested: Characters 3 (symmetry), 13 (operculum), 21change is inferred, from heteropolar to isopolar (Fig. (number of infratectum layers), and 29 (exine4B). The exact position of the change within the folding). Nine characters showed differences at morephylogeny is ambiguous, occurring either at the root than 15 nodes (25%): Characters 5 (shape class), 6

democratic

D,

coding;

comprehensive

C,

taxa.

higher

coding

ofmethods

alternative

three

using

generated

stud

y,this

inexam

ined

matrices

character

morph

ological

pollen

coding.

ofexem

plar

Com

parison

species

1. E,

Table

coding;

Character

12

34

56

78

910

1112

1314

1516

1718

1920

2122

2324

2526

2728

2930

31Total

Num

berof

states

C5

32

25

55

73

25

22

42

24

211

32

22

33

22

22

22

nD

13

22

24

35

32

22

23

22

42

72

22

23

32

22

12

2n

E3

32

24

45

73

24

22

42

23

211

32

22

33

22

22

22

n

Missing

data

(%cells)

C0

00

06.3

00

00

00

1.6

00

00

1414

1.6

1.6

1.6

1.6

3.1

2311

1.6

1417

00

0D

00

1.6

6.3

6.3

3.1

4.7

1.6

3.1

04.7

4.7

4.7

4.7

04.7

1917

141.6

7.8

1.6

4.7

3116

1.6

1417

00

3.1

6.4

E0

01.6

013

00

04.7

4.7

4.7

204.7

4.7

00

5050

3.1

1.6

4.7

2222

4845

1.6

3645

00

012.5

Polym

orphic

cells

(%)

C69

5845

1670

8994

9239

4881

5345

630

8364

5695

6.3

6.3

1620

7.8

9.4

6.3

7.8

06.3

3.1

3.1

41.3

D0

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

E0

1.6

06.3

7.8

7.8

2013

1.6

04.7

4.7

1.6

3.1

04.7

3.1

1.6

271.6

00

3.1

00

00

00

06.3

3.8

Num

berof

statechangesinferred

underMPcriterion

C0

22

32

51

22

10

40

21

23

25

20

10

72

00

20

00

D0

65

54

118

55

54

41

61

108

49

10

20

62

00

20

00

11E

312

55

813

1815

117

139

413

115

95

242

13

48

41

12

11

222

/a / a /a 3.6 0 53 4 0


(outline in polar view), 7 (size; Fig. 5A), 11(ectoaperture shape), 12 (aperture membrane orna-mentation), 14 (annulus; Fig. 5B), 17 (supratectalelement shape), 19 (tectum sculpture), and 24(endexine type; cf. Table 4). The greatest differences(. 35% of nodes) were seen for outline, size,annulus, supratectal element shape, and tectumsculpture. As an example, pollen size (Character 7;Fig. 5A) differed in optimization between methods atmany key, deep nodes in the phylogeny. For instance,Node 6, which links Pinus L. and the angiosperms,had an inferred ancestral state of small under EB,medium under MP and ML, and large under HB.However, at more derived nodes (for instance, theroot of the asterids and within this group), the fourmethods tended to converge on similar optimizations.In contrast, annulus type (Character 14; Fig. 5B)generated unanimous agreement between the meth-ods at many of the deepest nodes in the phylogeny,such as the roots of angiosperms, monocots, andeudicots, all inferred as lacking an annulus.In general, MP and ML tended to produce the most

congruent results (differing at only 2.2% of possiblenodes across all characters; Table 4); however, thismay in part be due simply to the high number ofequivocal optimizations generated by the ML analy-sis, which were treated as congruent with all possiblestates. The greatest incongruence was seen betweenthe EB and the MP and ML methods (differing in14.0% and 14.2% of optimizations, respectively).

Starting tree topology. For the five differenttopologies tested, using the species exemplar matrix(with polymorphic data points treated as missingdata) and characters optimized with MP, very fewchanges were apparent in terms of ancestral state.The only characters for which any differences at allwere noted were Characters 2 (polarity), 3 (symme-try), 4 (basic shape), 6 (outline in polar view), 9(aperture position), 16 (supratectal elements), 19(tectum sculpture), and 25 (endexine extent). In noneof these instances did the differences affect key nodes(i.e., those deep in the phylogeny), and the number ofstate changes inferred was the same on all trees for allcharacters. The most significant difference noted wasin Character 19 (tectum sculpture; Fig. 6). In thiscase, the state inferred at the root of the Nymphaealesdiffered depending on topology: for trees whereAmborella Baill. and Nymphaeales were positionedon successive branches (e.g., Fig. 6A), the mostparsimonious state for Nymphaeales was inferred asreticulate; for trees in which Amborella and Nym-phaeales were sister groups (e.g., Fig. 6B), the mostparsimonious state for the root of Nymphaeales wasambiguous, either reticulate or Amborella-type. This

taxa,

matrix,

higher

coded

coding

of exem

plar

0.14.

methods

the

RC

For

alternative 0.5,

matrix.

RI

chea 0.27,

ree I

for C

thfor

RC)

were

stud

y,and

RI, indices

this

(CI,

in

ensemble

exam

ined

indices

similar;

ensemble

characters

were

and

data

morph

ological

characters

missing

asall

across

treated

pollen

for

mean

were

indices

show

points

(rc)

columns data

consistency

Final

polymorph

ic(200

7).

rescaled

when

al.

etvalues

and

(ri), Jansen

retention bypoints;

provided data

(ci),

polymorph

ic

Consistency

phylogeny

the

include

on2.

calculated given

Table

values

Ensem

ble

Character

12

34

56

78

910

1112

1314

1516

1718

1920

2122

2324

2526

2728

2930

31Mean

indices

Consistency

index(ci)

C0

0.5

0.5

0.33

0.5

0.2

11

0.5

10

0.25

00.5

10.5

0.67

0.5

0.6

0.5

01

00.29

10

00.5

00

00.41

0.51

D0

0.33

0.2

0.2

0.25

0.27

0.25

0.6

0.4

0.2

0.25

0.25

10.33

10.1

0.38

0.25

0.67

10

0.5

00.33

10

00.5

00

00.33

0.35

E0.67

0.17

0.2

0.2

0.25

0.23

0.22

0.4

0.18

0.14

0.15

0.11

0.25

0.23

10.07

0.22

0.2

0.38

11

0.33

0.25

0.25

0.5

11

0.5

11

0.5

0.44

0.27

Retentionindex(ri)

C0

0.92

0.88

0.33

00.33

11

0.94

10

0.25

00.86

00

0.5

00.67

0.67

01

00.67

00

00.86

00

00.38

0.76

D0

0.84

0.69

0.2

00.47

0.25

0.92

0.86

0.79

0.7

0.75

10.64

00.68

0.58

0.5

0.82

10

0.75

00.71

00

00.86

00

00.45

0.72

E0

0.63

0.63

0.43

0.25

0.5

0.13

0.67

0.59

0.7

0.39

0.47

.67

0.47

00.5

0.42

0.2

0.38

11

00.25

0.25

00

00.67

00

0.67

0.38

0.50

Rescaledconsistencyindex(rc)

C0

0.46

0.44

0.11

00.07

11

0.47

10

0.06

00.43

00

0.33

00.4

0.33

01

00.19

00

00.43

00

00.25

0.39

D0

0.28

0.14

0.04

00.13

0.06

0.55

0.35

0.16

0.18

0.19

10.21

00.07

0.22

0.13

0.55

10

0.38

00.24

00

00.43

00

00.2

0.25

E0

0.1

0.14

0.09

0.06

0.12

0.03

0.27

0.11

0.1

0.6

0.05

0.17

0.11

00.03

0.09

0.04

0.14

11

00.06

0.06

00

00.33

00

0.33

0.14

0.14

Abb

reviations:C,comprehensive

coding;D,democratic

coding;E,exem

plar

speciescoding.


Figure 4. Comparing exemplar, comprehensive and democratic methods for coding higher taxa as terminals, using MP inferenceof ancestral states on angiosperm phylogeny, for the example of Character 2, pollen polarity: white, apolar; gray, heteropolar; black,isopolar/subisopolar. A. Coding exemplar species as terminals (12 state changes inferred). B. Coding taxonomic orderscomprehensively as terminals (two state changes inferred). C. Coding orders democratically as terminals (six state changesinferred). Branches displaying more than one shade indicate situations where the most parsimonious optimization is equivocal.


the

instates

ofnumber

the

reason

(2007).

this al.

etfor

enthat

Jans

data

note

byprovided

missing

asph

ylogeny

treated

the

oncells

calculated

(polym

orph

ictaxa,

exem

plar

higher

and

coding

democratic of

methods

the

states)

using

observed

obtained

ofmatrices

number

for

the

statistics

than

fewer

character

beof

characters

Com

parison

some

for

3.may

Table

matrix

Character

12

34

56

78

910

1112

1314

1516

1718

1920

2122

2324

2526

2728

2930

31Total

Num

berof

unam

biguous(i.e.,neith

erinapplicable,-,norun

know

n,?)states

D1

32

22

43

43

22

22

32

24

27

21

21

33

11

21

11

n/E

33

22

34

57

32

32

24

22

32

103

22

23

32

22

22

2n/

Num

berof

statechangesinferred

underMPcriterion

(figuresin

bold

indicate

greaterthan

100%

differencein

numberof

steps[M

Panalysis]betweenspeciesexem

plar

matrixanddemocratic

matrix.

Figures

inita

licsindicate

where

differencemay

bedu

eto

totalnu

mberof

characterstates

rather

than

differences

inoptim

ization)

D0

65

54

118

55

54

41

61

108

49

10

20

62

00

20

00

114

E3

125

58

1314

1511

713

94

121

159

521

21

34

84

11

211

12

122

Num

berof

internal

nodesdiffering

ininferred

statebetweenDandEun

derthefour

optim

izationmethods

(figuresin

bold

indicate

where

morethan

aqu

arterof

the63

internal

nodesshow

disagreementin

reconstructionbetweenthetwocoding

strategies)

EB

114

03

25

16

23

25

211

23

131

21

017

29

829

32

611

41

33

03

48

30HB

22

01

22

16

26

31

17

115

112

016

28

327

22

09

72

00

00

02

21ML

00

00

19

30

019

13

16

010

18

28

10

05

10

00

00

00

8MP

00

02

14

50

021

53

24

09

93

141

20

10

00

00

00

28

Total

nodesaffected

115

03

33

28

29

25

225

24

16

226

019

35

1239

32

611

103

33

03

48

390

Total

differences

136

06

4945

5728

368

4024

543

052

84

1678

76

626

123

33

03

414

704

Abb

reviations:D,democratic

matrix;E,species

exem

plar

matrix;EB,empiricalBayesianinference;HB,h

ierarchicalBayesianinference;ML,m

aximum

likelihood;

MP,maximum

parsimony.

a a 9 7 8 8


ofbetween

methods

data)

missing

reconstruction

inas

treated

cells disagreement

c(polym

orph

ishow

ednodes

exem

plar

internal

63and

the

democratic of

%25

than

the

more

using

where

obtained

indicate

matrices

bold

infor

Figures

methods

(200

7).

optim

ization

al.

etJansen

between

bydifferences

provided

node

phylogeny

observed

the

onof

calculated

Com

parison

taxa,

strategies.

4.higher

coding

Table

coding two

the

Character

12

34

56

78

910

1112

1314

1516

1718

1920

2122

2324

2526

2728

2930

31Total

Num

berof

observed

differences

(dem

ocratic

matrix)

MP-M

L0

00

00

00

00

50

10

00

10

00

00

00

10

00

00

00

MP-EB

09

08

1419

1817

85

34

015

63

70

50

00

011

20

00

00

0153

MP-H

B0

00

1016

190

11

50

30

10

33

03

00

00

63

00

00

00

ML-EB

09

011

1222

1717

62

53

014

63

70

60

00

010

20

00

00

0152

ML-H

B0

00

1214

221

11

40

20

10

43

03

00

00

20

00

00

00

EB-H

B0

90

22

219

167

25

10

156

38

16

00

00

102

00

00

00

116

Total

nodesaffected

09

015

16

23

19

178

75

40

156

610

18

00

00

17

50

00

00

01

Total

differences

027

043

5884

5552

2323

1314

046

1817

271

230

00

040

90

00

00

0573

Num

berof

observed

differences

(exemplar

speciesmatrix)

MP-M

L0

00

70

11

00

03

20

00

613

18

00

00

00

00

00

00

MP-EB

117

07

2128

266

813

1613

020

66

1910

102

06

411

33

33

04

6272

MP-H

B2

00

02

1420

22

57

50

50

617

417

10

02

50

00

00

00

116

ML-EB

119

014

1825

248

813

1613

022

611

167

152

05

411

33

23

04

5278

ML-H

B2

20

133

1219

42

59

60

80

115

211

10

02

30

00

00

00

120

EB-H

B9

70

119

149

106

812

90

176

222

617

10

62

123

33

30

46

217

Total

nodesaffected

119

014

22

18

27

118

1320

16

024

613

25

1025

20

64

18

33

33

04

63

Total

differences

3525

048

6394

9930

2644

6248

075

1842

9230

787

017

1442

99

89

012

1710538 74 7 0 91 42 24



situation is partly an artifact of removing polymorphic Character 2, pollen grain polarity (Fig. 7A),states: both Nymphaeales taxa were coded as missing showed some differences in the ancestral characterdata for this character in this matrix, due to both states inferred depending upon coding strategy, fewhaving polymorphic tectum sculpture types, which differences among analysis methods, and no differ-led to the most parsimonious state inferred at the root ence with tree topology. Optimization indicated aof the Nymphaeales depending entirely upon the single inferred transition from heteropolar (thesurrounding taxa. plesiomorphic state for angiosperms) to isopolar

grains, most likely occurring on the internal branch

OVERVIEW OF ANGIOSPERM POLLEN EVOLUTION between the stem of the monocots and the root of theeudicots (although it should be noted that further

Since topology was found to make relatively little transitions are inferred on terminal branches, notdifference to ancestral character state optimization at shown in the figure, such as to isopolar in Illicium L.,the level of angiosperms for the broad-scale pollen Chloranthus Sw., and Calycanthus L., and to apolarcharacters investigated, the following discussion is in Ipomoea L. and Musa L., among others).based on a single topology (Jansen et al., 2007), Character 3, pollen grain symmetry (Fig. 7B),which is associated with meaningful branch lengths showed an entirely congruent pattern across allderived from molecular genetic state changes, analysis methods and topologies tested, althoughenabling it to be used with both model-based and optimizations differed with coding strategy. Thismodel-free methods. Although the method of coding character is inferred to have undergone two transi-higher taxa was found significantly to impact the tions from bilateral (the plesiomorphic state) to radialresults of ancestral character state optimizations (see symmetry, once within monocots, on the branchabove), for the purposes of exploring the evolution of leading to Zingiberales (represented by Musa) andpollen characters the following discussion will be Poales, and once on the internal branch between thebased on the matrix produced using the species stem of the monocots and the root of the eudicots.exemplar method (polymorphic characters treated as Again, there are further transitions within terminalmissing data); this avoids overestimation of congru- branches (to radial symmetry in Illicium, Chloran-ence between methods. thus, and Drimys J. R. Forst. & G. Forst.). However,A number of characters showed structured patterns radial symmetry seems to be fixed within eudicots,

when optimized on the phylogeny of angiosperms with no apparent reversals to bilateral symmetryand, therefore, have potential to provide diagnostic or among the taxa studied.synapomorphic characters and hypotheses of evolu- Character 4, basic shape (Fig. 7C), differed intionary processes. Unfortunately, these tend also to reconstruction with tree topology and optimizationbe the characters that differed in optimization method to a small extent but not with coding strategy.depending on the coding strategy, optimization This character shows a clear transition from boat-method, or topology used. In general, characters that shaped to globose occurring on the branch leading towere unambiguously optimized in all contingencies Zingiberales and Poales, and another transitionwere either those for which little data were available occurring some time after the divergence of the(e.g., Character 28, microchannels), or those known Nymphaeales from the rest of the angiosperms. Withto be variable only within a single group, such as the MP and ML, this latter change of state is inferred tocavea (Character 26), internal foramina (Character have occurred before the divergence of the branch27), and lophate exine (Character 29) of certain leading to Austrobaileyales, but with both BayesianAsteraceae. methods it is inferred prior to the root of the eudicots.Nonetheless, of the 31 characters tested, only eight Depending on the method, further transitions are

(Characters 5, 17, 18, 21, 25 to 27, and 29) were invoked, e.g., to globose in the magnoliids, or arelatively uninformative. These were either variable reversal to boat-shaped in early-diverging monocots.only within a very small group (e.g., Character 29, In all cases, globose pollen appears fixed withinexine folding), lacking in data (e.g., Character 17, eudicots, with no apparent reversals to boat-shapedsupractectal element shape), or highly homoplastic at pollen. Due to high levels of missing data in basallythe present hierarchical level (e.g., Character 5, branching taxa (polymorphisms within taxa removedshape class). The remaining 23 characters were of for comparison of optimization methods), the plesio-some interest from an evolutionary or taxonomic point morphic state at the base of the angiosperms was notof view, despite variability across optimization resolved by all methods.methods. Several examples of particular interest are Character 6, outline in polar view (Fig. 7D), showsdiscussed below (Figs. 7, 8). a much more complex pattern and displayed

differences with coding strategy, optimization meth- root of the tree. This character showed no differencesod, and topology. The plesiomorphic state for in optimization between tree topologies but someangiosperms is resolved as oblate (although ambig- differences between coding strategies. The plesio-uous in the ML optimization). In the two Bayesian morphic state for angiosperms (and several subse-optimizations, there is an inferred transition to quent nodes) was inferred as mono-aperturate (MPcircular outline before the branch leading to Austro- and HB) or 4- to 6-aperturate (EB); the MLbaileyales (represented by Illicium), in which case optimization was ambiguous at this point. All fourthere is then a reversal to elliptic before the methods were congruent in inferred state for the rootsdivergence of the monocots. All methods agree that of monocots (mono-aperturate) and eudicots (tri-there is a shift to circular grains within monocots and aperturate), with multiple transitions on terminalanother before the root of the eudicots. However, in and subterminal branches to 4- to 6-, 7- to 12-, orcontrast to the previous characters, the situation poly-aperturate grains.within eudicots is highly complex and homoplastic, Character 9, aperture position (Fig. 8B), showswith little congruence between analysis methods on inferences differing slightly depending on codingthe exact placement of state transitions. strategy, optimization method, and tree topology. AllCharacter 8, aperture number (Fig. 8A), shows methods inferred the plesiomorphic state for angio-

some discrepancies between optimization methods, sperm apertures to be distal. Most methods (exceptparticularly in inferring states at nodes toward the EB) concurred on a likely transition to equatorial

Figure 5. Comparing MP, ML, EB, and HB methods for inferring ancestral character states on angiosperm phylogeny, usingspecies exemplar coding with polymorphic data points treated as missing data, for two example characters. Solid circles indicatecongruence between all methods of optimization. Pie charts indicate incongruence: lower left segment, most parsimoniousresults; lower right segment, most likely result; upper right segment, most probable result of EB analysis; upper left segment,most probable result of HB analysis. A. Character 7, pollen grain size: black, very small; yellow, small; red, medium; blue,large; green, very large. B. Character 14, annulus: black, absent; yellow, costate; red, aspidate; blue, vestibulate. In both casesgray indicates an equivocal optimization.


apertures occurring on the branch leading to the root apertures within the basally branching groups, norof eudicots. Multiple additional transitions to equa- any reversals to polar apertures within eudicots.torial apertures are inferred on terminal taxa within Optimization of Character 10, aperture structurethe more basally branching angiosperms (such as (Fig. 8C), differed slightly between optimizationIllicium and Dioscorea L.), and to global apertures methods, significantly with coding strategy, but notwithin eudicots (e.g., Gossypium L., Spinacia L.). No with tree topology. All methods concurred that thetransitions were inferred from distal to global plesiomorphic state for angiosperms is simple

Figure 6. Comparing inferences of ancestral character state based on two different topologies for angiosperms, using the MPcriterion and species exemplar coding with polymorphic data points treated as missing data, for the example of Character 19,tectum sculpture: white, imperforate tectum; dark blue, areolate; light blue, Croton-patterned; turquoise, Amborella-type; darkgreen, fossulate; light green, foveolate; yellow, perforate; orange, reticulate; red, rugulate; black, striate. Branches displayingmore than one color indicate situations where most parsimonious optimization is equivocal. A. Phylogeny as recovered byJansen et al. (2007): A. trichopoda Baill. and Nymphaeales as successive sister groups to the remaining angiosperms. B.Phylogeny as recovered by Soltis et al. (2007): A. trichopoda and Nymphaeales sister to one another. Arrows indicate groupswhere optimizations differ (see text for explanation).


Figure 7. Examples of ancestral character states inferred on angiosperm phylogeny using all four analysis methods and thespecies exemplar coding method with polymorphic data points treated as missing data. Solid circles indicate congruencebetween all methods of optimization. Pie charts indicate incongruence: lower left segment, most parsimonious results; lower rightsegment, most likely result; upper right segment, most probable result of EB analysis; upper left segment, most probable result ofHB analysis. A. Character 2, pollen grain polarity: black, apolar; yellow, heteropolar; red, isopolar/subisopolar. B.Character 3, pollen grain symmetry: black, bilateral; yellow, radial. C. Character 4, basic shape: black,

wortley2015.pdf

Documents

ancestral character

evolution of angiosperm

diagnostic characters

research libraries

research funders

method of optimization

phylogenetic topology

pollen morphology