Annals of Botany 112: 1613-1628, 2013 r w doi:10.1093/aob/mct027, available online at www.aob.oxfordjournals.org I— 1 I Y A J ______Founded 1887 PART OF A SPECIAL ISSUE ON INFLORESCENCES
Macroevolution of panicoid inflorescences: a history of contingency and order of trait acquisition
R. Reinheimer1,2’*, A. C. Vegetti1 and G. H. Rua3 1Morfologia Vegetal, Facultad de Ciencias Agracias, Instituto de Agrobiotecnologia del Litoral (UNL-CONICET), Kreder 2805, S3080HOF Esperanza, Santa Fe, Argentina, 2Instituto de Botdnica Darwinian, Casilla de Correo 22, B1642HYD San Isidro, Buenos Aires, Argentina and 3Cdtedra de Botdnica Agricola, Facultad de Agrononria, Universidad de Buenos Aires, C1417DSE, Capital Federal, Buenos Aires, Argentina * For correspondence. E-mail [email protected] or [email protected]
Received: 28 September 2012 Revision requested: 30 October 2012 Accepted: 14 December 2012 Published electronically: 10 March 2013
• Background and Aims Inflorescence forms of panicoid grasses (Panicoideae s.s.) are remarkably diverse and they look very labile to human eyes; however, when performing a close inspection one can identify just a small subset of inflorescence types among a huge morphospace of possibilities. Consequently, some evolutionary constraints have restricted, to some extent, the diversification of their inflorescence. Developmental and genetic mechanisms, the photosynthetic type and plant longevity have been postulated as candidate constraints for angiosperms and panicoids in particular; however, it is not clear how these factors operate and which of these have played a key role during the grass inflorescence evolution. To gain insight into this matter the macroevo- lutionary aspects of panicoid inflorescences are investigated. • Methods The inflorescence aspect (lax versus condensed), homogenization, truncation of the terminal spikelet, plant longevity and photosynthetic type were the traits selected for this study. Maximum likelihood and Bayesian Markov chain Monte Carlo methods were used to test different models of evolution and to evaluate the existence of evolutionary correlation among the traits. Both, models and evolutionary correlation were tested and analysed in a phylogenetic context by plotting the characters on a series of trees. For those cases in which the correlation was confirmed, test of contingency and order of trait acquisition were preformed to explore further the patterns of such co-evolution. • Key Results The data reject the independent model of inflorescence trait evolution and confirmed the existence of evolutionary contingency. The results support the general trend of homogenization being a prerequisite for the loss of the terminal spikelet of the main axis. There was no evidence for temporal order in the gain of homogen ization and condensation; consequently, the homogenization and condensation could occur simultaneously. The correlation between inflorescence traits with plant longevity and photosynthetic type is not confirmed. • Conclusions The findings indicate that the lability of the panicoid inflorescence is apparent, not real. The results indicate that the history of the panicoids inflorescence is a combination of inflorescence trait contingency and order of character acquisition. These indicate that developmental and genetic mechanisms may be important con straints that have limited the diversification of the inflorescence form in panicoid grasses.
Key words: Inflorescence, morphology, evolution, panicoids, Panicoideae, Poaceae.
INTRODUCTION Paspaleae, and Paniceae s.s. tribes; Morrone et al., 2011) is characterized by distinct patterns of inflorescence character Inflorescence architecture in the grasses varies enormously state transitions. These results indicated that the evolution of even among close related species (Kellogg, 2000; Doust and the inflorescence of panicoids may follow certain patterns Kellogg, 2002; Doust et a i, 2005; Liu et a i, 2005, 2007; and cannot evolve freely. Reinheimer and Vegetti 2008; Reinheimer et a i, 2009). In It has been postulated that limitation in biological diversity fact, the evolution of the grass inflorescence seems to be may be due to developmental and genetic constraints as well as very complex and the identification of clear trends appears selection at other levels (Prusinkiewicz et al., 2007). In fact, it almost impossible. Recently, we have partially investigated has been empirically corroborated that the extensive variation this issue using the panicoid grasses as a model group (also of inflorescence architecture among angiosperms is somewhat known as Panicoideae s.s.\ Sanchez-Ken and Clark, 2010) correlated with season length, plant longevity, pollination (Reinheimer et a i, 2012). We have found that within the the system and seed dispersal mode (Friedman and Harder, oretical morphospace comprising all possible combinations of 2004, 2005; Prusinkiewicz et al., 2007). Whether, why and inflorescence traits, only a few were actually observed in the how this presumed co-evolution has happened in the grasses group, and among them some inflorescence architectures pre is an intriguing question that remains open for further investi dominate over others. In addition, the inflorescence of each gations. Interestingly, Morrone et al. (2011) recently reported of the major clades (Arundinelleae ,v.,v., Andropogoneae, that qualitative inflorescence character changes (such as
(C) The Author 2013. Published by Oxford University Press on behalf of the Annals of Botany Company. All rights reserved. For Permissions, please email: [email protected] 1614 Reinheimer et al. — Macroevolution of panicoid inflorescences abortion of the fertile apex of main axes and rachises, lack of Amndinelleae s.s., Paspaleae and Paniceae ,v.,v. tribes which branching beyond first order, and modification of the disarticu include most of the species diversity in panicoids. lation zone of the diaspores) in Paniceae s. I. (Paspaleae + We estimated phylogenetic relationships using Bayesian Paniceae s.s. tribes) predominates at the C4 clades, whereas Markov chain Monte Carlo (MCMC) analysis as implemented the inflorescence of the C3 clades has not changed much in MrBayes version v31.2 (Huelsenbeck and Ronquist, 2001). during evolution. Why this happens is still unclear, although The best-fit model (GTR + G + 1) was inferred with the ancestral character evolution analyses suggested that MrModeltest v.2-3 (Nylander, 2004) based on the Akaike cri changes in qualitative traits of the Paniceae s. I. inflorescence terion. Two Metropolis-coupled Markov chains with an incre may represent a delayed response to the change in photosyn mental heating temperature of 0-2 were run for 10 million thetic pathway (Morrone et al., 2011). generations and sampled every 1000th generation. The ana To get new insights on the underlying bases of the evolution lysis was repeated twice, starting from random trees. The con of the grass inflorescence, we have used the molecular phyl- vergence and the effective sample size (>100) of each ogeny of panicoid grasses and maximum likelihood (ML) replicate were checked using Tracer v. 1-5 (Rambaut and and Bayesian Markov chain Monte Carlo (MCMC) methods Drummond, 2007). A majority-mle consensus tree was then to (a) test models of evolution, (b) identify directional reconstructed after applying a burn-in of 2-5 million genera trends, (c) test the existence of correlated evolution among in tions (thus considering a total of 15 002 trees). florescence traits, and (d) evaluate the association between in florescence traits with plant longevity and photosynthetic type. Ancestral character state reconstruction Character states were estimated for nodes with a posterior MATERIALS AND METHODS probability equal to/higher than 0-95. We have inferred ancestral Inflorescence morphology dataset character states using ML and MCMC methods. For the MCMC analyses, we used the ‘multistate' module in The morphological dataset was based on the following charac BayesTraits (Pagel et al., 2004; program available at ters: (a) inflorescence aspect (0, condensed; 1, lax; 2, lax to con www.evolution.rdg.ac.uk), and the 15 002 trees previously cal densed); (b) degree of homogenization (0, non-homogenized; 1, culated. An exponential prior seed from a uniform prior model partially homogenized; 2, fully homogenized); and (c) presence/ was chosen. To set the range of the hyperprior, preliminary absence of terminal spikelet (0, non-truncated; 1, truncated). We chains were run under ML (Pagel et a l, 2004). Given these followed the character definitions described in Rua and results, the range of the hyperprior was set to 0-30 for all Weberling (1998) and Reinheimer et al. (2012). The reconstruc cases. In addition, several exploratory chains were also run tion of the homogenization character resulted in ambiguity at adjusting the value of the rate coefficient proposals (the deep nodes when the trait was codified as a multistate charac ratedev parameter) until an approximate acceptance rate of ter. For this reason, we have explored the reconstruction of the 20-40 % was achieved (Pagel et a l, 2004). Final values of homogenization of the inflorescence as a binary character (0, ratedev were 0-1 for inflorescence appearance and homogen absence; 1, presence). Character-state data were compiled ization degree, from 0-01 to 0-05 for presence/absence of hom from published reports (reviewed in Reinheimer et a l, 2012). ogenization, and 2 for presence/absence of terminal spikelet. The dataset used in this study is presented in Supplementary Ancestral states were estimated using the most recent common Data Table SI. ancestor (‘MRCA') of selected taxa command. Once these para To calculate the incidence of the different inflorescence meters were set, two independent analyses were run for 10 types within panicoids we have calculated the proportion of million generations and sampled every 1000th generation to taxa with a given inflorescence type. ensure independence. The first 1 million generations were dis carded as bum-in (convergence and ESS were checked with Tracer vl-5; Rambaut and Drummond, 2007) and the rest of Phytogeny the samples from the two replicates were combined (18 002 Phylogenetic reconstruction and analysis of character evolu samples). Hypothesized character states at internal nodes were tion were based on ndhF nucleotide sequences and alignment tested by estimating Bayes factor (BF), comparing MCMC previously published by Aliscioni et al. (2003), Sanchez-Ken runs in which the node in question was constrained to one and Clark (2010), Zuloaga et al. (2010) and Morrone et al. state versus the other using the ‘fossil' command (Pagel et al., (2011). Recent studies have suggested that panicoids should 2004). The BF approach was based on smoothed estimates of be divided into six well-supported tribes (Sanckez-Ken and marginal likelihood analysed with Tracer vl-5 (Rambaut and Clark, 2010; Zuloaga et a l, 2010; Morrone et al., 2011): Drummond, 2007), which applies the method used by Newton Steyermarkochloeae p.p., Tristachyideae, Amndinelleae ,v.,v., and Raftery (1994) with modifications by Suchard et al. Andropogoneae, Paspaleae (also known as the Paniceae x = (2001). A value between 2 and 5 indicates ‘positive' support, 9 clade; Giussani et a l, 2001) and Paniceae ,v.,v. (also referred between 5 and 10 ‘strong' support, and any value >10 means in the literature as the Paniceae x = 10 clade; Giussani et al., ‘very strong' support. Similar tendencies were recovered when 2001). The phylogenetic position of Steyermarkochloa angu- additional analyses of ancestral state reconstruction were run stifolia (Spreng.) Judz. and Tristachyideae is not well resolved using a uniform prior (0-100). (Sanckez-Ken and Clark, 2010; Morrone et al., 2011). For conflicting nodes, we have also inferred the ancestral Consequently, we will focus our studies on the three state reconstruction using the ML method. Maximum likeli major, well-supported panicoid lineages: Andropogoneae + hood reconstructions were conducted using the ‘multistate' Reinheimer et al. — Macroevolution of panicoid inflorescences 1615 module included in the computer package BayesTraits (Pagel inflorescence (state 1), non-truncated inflorescence (state 0) et al. 2004), and the 15 002 trees previously calculated. and truncated inflorescence (state 1). For the ML test, each Ancestral states were estimated using the most recent pair of characters was analysed under two different models common ancestor (‘MRCA') of selected taxa command, and (Pagel and Meade, 2006). The first model (the independent hypothesized character states at internal nodes were then model I) assumed that the characters vary independently of tested by estimating the differences of the log likelihood in each other. The second model (the dependent model D) which the node in question was constrained to one state allowed one character to vary based on the other. We determined versus the other using the ‘fossil' command (Pagel et a l. the likelihood ratio (LR) using the equation LR = -2[L(D) - 2004). We applied the general rule that two log likelihood L(I)] where L(D) and L(I) are the likelihood estimates of the units (using the mean and median of 15 002 tree likelihoods) dependent and independent models, respectively. We tested the constitute a significant difference (Pagel et al.. 2004). There likelihood ratio against distribution, with four degrees of was no appreciable difference between the mean and the freedom (Pagel and Meade, 2006). P < 0-05 is taken as positive median of the 15 002 tree likelihoods. Consequently, we evidence that the dependent model is favoured, P < 0 01 means only calculated the mean. strong evidence, and P < 0-001 is very strong evidence. When For simplicity, we report ancestral state reconstructions on the MCMC algorithm was implemented, we used a uniform the majority rule consensus topology of 15 002 trees. prior for the independent model and an exponential hyperprior for the model of dependent evolution (Pagel and Meade, 2006). Repeating the analyses applying a uniform hyperprior for the de Models o f inflorescence evolution pendent model does not change the results. Each model, was run To test hypotheses on inflorescence evolution within pani using the MCMC settings and the approach to assess the conver coids, two models of character evolution were assessed. The gence of the analysis as described above. When necessary, the first model assumes all character state transitions are unordered number of generations was doubled until the effective size of (i.e. direct transitions are possible among all character states), parameters exceeds 200 (checked with Tracer vl-5; Rambaut whereas the second model assumes ordered character state and Drummond, 2007). In all cases, the range of hyperprior transitions as follows: transitions between lax inflorescence was set to 0-30, and ratedev was set to 0-1. Then, we compared to condensed inflorescence are constrained to pass through a the model that assumed the independent evolution of the two lax to condensed inflorescence, transitions between non binary characters being compared with that of a model that homogenized inflorescence and fully homogenized inflores allowed for correlated evolution between these characters (de cence are constrained to pass through a partially homogenized pendent model) using BF as described above. Results >2 are inflorescence, and the transitions from non-truncated to trun taken as positive evidence that the dependent model is favoured, cated inflorescences is irreversible. Models were tested on >5 is strong evidence, and > 10 represents very strong evidence the 15 002 set of trees by ML and MCMC approaches using (Pagel and Meade, 2006). the ‘restrict' command at the ‘multistate' module of the com Several factors (i.e. season length, plant longevity and puter package BayesTraits (Pagel et al., 2004). For the ML photosynthesis type) are thought to be correlated with inflores model test we applied the general rule that two log likelihood cence architecture (Friedman and Harder, 2004, 2005; units constitute a significant difference as explained above Prusinkiewicz, et al., 2007; Morrone et al., 2011). To (Pagel et al. 2004). In the MCMC analysis, we used the explore better the strength of such correlations in the panicoids MCMC settings and the approach to assess the convergence we statistically assessed parallelism between plant longevity of the analysis as described above. The models were compared (0, annual; 1, perennial) and photosynthetic pathway (0, C3 using approximate BF as explained before. or C3-C4 intermediates; 1, full-C4) with the inflorescence In addition, we calculated the global rate of the inflorescence traits studied here. We compiled the plant longevity and photo aspect processes, and the homogenization processes as: synthetic type datasets from published reports and on-line ^condensation ^1,2 Cj 1,0 -)- (flf) VerSUS ^/dc-condcnsalion q2,\ databases (GrassBase, http://www.kew.org/data/grasses- <70,1 + qQ,2 for the inverse process, and ^homogenization = <70,1 + db.html; Watson and Dallwitz, 1992; Zuloaga et al., 2000; q\,2 + q0,2 versus ^e-homogenization = ql.O + q2,l + q2,0 for the Giussani et al., 2001; Morrone et al., 2011). The dataset is pre inverse process, where the parameter qi, j is the transition rate sented in Supplementary Data Table SI. For this analysis, we from state 7" to state The statistical differences among pairs used the ML and the MCMC algorithms in the ‘discrete' of rates were studied using the non-parametric Mann-Whitney module implemented by BayesTraits (Pagel et al., 2004), fol U-test in the program InfoStat version 2009 (Grupo InfoStat, lowing the same methodology described above. FCA, Universidad Nacional de Cordoba, Argentina). When we confirmed that two traits were correlated, we tested hypotheses about conditional evolution and the temporal order of trait acquisition by ML and MCMC analyses con Analysis of correlated evolution ducted in the ‘discrete' module of BayesTraits (Pagel et al. We explored correlations between inflorescence characters 2004). For the ML method, we compared the likelihood esti across the evolutionary tree by using ML and the MCMC algo mates of two correlated evolution models: one in which all rithms in the ‘discrete' module implemented by BayesTraits the transition rates were allowed to vary, to one that con (Pagel et al., 2004). For this purpose, each trait was coded as a strained the two transition rates to be equal. The likelihood discrete character: condensed and lax to condensed inflores ratio of the two models is tested against a x2 distribution cence (state 0), lax inflorescence (state 1), non-homogenized with one degree of freedom (Pagel and Meade, 2006). P < inflorescence (state 0), partially and fully homogenized 0-05 is taken as positive evidence that the dependent model 1616 Reinheimer et al. — Macroevolution of panicoid inflorescences is favoured, P < 0-01 means strong evidence, and P < 0-001 is was from lax inflorescences to lax to condensed inflorescences very strong evidence. When MCMC was implemented, we (q\,2 = 2-94 + 0-78), and from condensed inflorescences to analysed the posterior probability distributions of the values lax to condensed inflorescences (<70,2 = 2-97 + 0-83), of the transition parameters estimated under the dependent whereas the lowest rate of change was from lax to condensed model. For this analysis we calculated the Z-scores and inflorescences iq 1,0 = 0-41 + 0-77; Fig. 2). Slightly different average of the transition parameters (Gonzalez-Voyer et a i, results were obtained under ML methods in which the highest 2008). MCMC results are summarized in flow diagrams. rate of change was from condensed inflorescences to lax to con densed inflorescences (<70,2 = 3-7 + 0-65), and the lowest rates were to lax to condensed inflorescence as well as lax to con RESULTS densed to condensed inflorescences (<71,0 = 0-66 + 0-09, and Phylogenetic and ancestral state reconstniction analyses q2.Q = 0-65 + 0-21; Supplementary Data Fig. S2). The statistic al differences among pairs of rates, associated with both evolu The topology of the Bayesian majority rule consensus (from tionary processes of de-condensation versus condensation, 15 002 trees) is similar to that recovered using parsimony showed that the former process was favoured over the latter one methods by Morrone et al. (2011). The Panicoids are divided independently of the method used (Bayesian MCMC inference: in three major and well-supported lineages: Arundinelleae *7de-condensation 2-4 + 1-29 VS. ^condensation 1‘89 + 1-46, ML ,v.,v. + Andropogoneae, Paspaleae and Paniceae ,v.,v. tribes. inference. <7de-condensation 2-58 ± 1-45 VS. ^condensation Andropogoneae and Arundinelleae ,v.,v. tribes form a well- 0-99 + 0-58; Fig. 2 and Supplementary Data Fig. S2). supported clade, which is sister to the Paspaleae tribe. In addition, In terms of homogenization, the restricted model fits the Paniceae ,v.,v. is sister to the (Andropogoneae + Arundinelleae dataset better, although there is no significant difference s.s.) + Paspaleae clade. between the two models (InBF = 200; log-likelihood All reconstruction methods found that the ancestor of pani difference = 0-42). The reconstruction of the transition rates coid grasses had a lax, partially or fully homogenized, and based on the Bayesian MCMC method and the unordered non-truncated inflorescence (Fig. 1). Lax to condensed inflor model of evolution showed that the highest rate of change was escences or condensed inflorescences, as well as truncated from partially homogenized to non-homogenized inflorescences inflorescences evolved later at the base of the Arundinelleae iq 1,0 = 8-61 + 3-45), whereas the lowest rate of change was from ,v.,v and Andropogoneae tribes, and after the divergence of fully homogenized to non-homogenized inflorescences (<72,0 = Paspaleae and Paniceae s. s. Reversal to the ancestral inflores 0-13 + 0-37). The statistical differences among pairs of rates, cence aspect is observed at the Arthropogoninae and associated to both evolutionary processes of homogenization Melinidinae subtribes; however, reversal to the ancestral non versus de-homogenization, showed that the second process was truncated inflorescence was not documented. In terms of hom favoured over the former (<7de-homogemzation = 4-59 + 4-55 vs. ogenization, ML and MCMC methods suggest that the ancestor ^homogenization = 3-5 + 4-24). The maximum likelihood method of the panicoids had a partially or fully homogenized inflores has produced similar tendencies (Supplementary Data Fig. S2). cence (Fig. 1), with 66-93 % probability and InBF between 1 Finally, we found that for truncation of the terminal spikelet and 3-7, depending on the strategy applied. Non-homogenized the unordered model fit the dataset better than the restricted inflorescences evolved only once in the Andropogoneae, and model (InBF = 2-67; log-likelihood difference = 7-28). The several independent times during the diversification of reconstruction of the transition rates based on the Bayesian Paspaleae and Paniceae s.s. Reversals to the ancestral homoge MCMC and ML methods and the unordered model of evolution nized inflorescence occurred in the Arthropogoninae, showed that the highest rate of change was from non-truncated Otachyriinae and Melinidinae sub tribes, and possibly in the to truncated inflorescences (MCMC method: <70,1 = 1-06 + Cenchrinae and Boivinellinae. 0-34 vs. 1-04 + 0-36; ML method: <70,1 = 0-97 + 0-09 vs. Eight basic inflorescence types are identified when the three ql.Q = 0-96 + 0-13; Fig. 2 and Supplementary Data Fig. S2). characters and their states are combined (presence/absence of branch condensation + presence/absence of homogenization + presence/absence of the terminal spikelet of the main axis; Evolutionaiy correlations Supplementary Data Fig. SI). Lax, non-homogenized, non- Bayesian MCMC and ML methods indicate, with positive truncated inflorescences are the most common inflorescence statistical evidence, that there is correlation between homogen type observed in the panicoids. In contrast lax, non-homogenized, ization and truncation (lnBF= 4-60; LR = 12-5, P < 0-01; truncated inflorescences are rare, and lax, homogenized, truncated Table 2). Also, there is strong evidence for a correlation inflorescences were not observed in the group. between the evolution of condensed inflorescences (condensed or lax to condensed inflorescences) the homogenization process (InBF = 22-91; LR = 20-71, P < 0 001), and trunca Models o f inflorescence evolution tion (InBF = 1516; LR = 34-92, P < 0 001). In addition, Both, MCMC and ML methods showed similar results MCMC and ML methods suggested a weak correlation (Table 1). The free model allowing all the possible transitions between photosynthetic type and inflorescence aspect among the states of inflorescence aspect fit the dataset better (InBF = 2 02; LR = 7-9, P = 0 09). None of the inflorescence than the restricted model (InBF = 3-88; log-likelihood characters studied show positive correlation with the plant difference = 4-21). The reconstruction of the transition rates longevity. based on the Bayesian MCMC method and the unordered Having confirmed the correlation between inflorescence model of evolution, showed that the highest rate of change traits we tested hypotheses about conditional evolution and Reinheimer et al. — Macroevolution of panicoid inflorescences 1617
Zeugites pittieri Chasmanthium latifolium Chasmanthium laxum inflorescence aspect Thysnaoiaena maxima Arundinella hirta 1 (0) Condensed Apiuda mutica ('■:P/spp p PpP inivus I l(6 Lax D ppPiPPPPs sec:vi’ui Phaceiurus digitatus Microstegium nudum | (2) Lax to condensed Miscanthus japonicus ( 'ccsincline sorghoides Sorghum bicolor „ Coix aquatica Cymbopogon flexuosus Heteropogon contortus Capiiiipedtum parviflorum Bothriochioa biadhii Dichanthium aristatum Andropogon gerardii Hyparrnenia hirta Schizachyrium scoparium Eiionurus muticus Tripsacum dactyioides Zea mays Oncorachis ramosa Opiismenopsis najada Mesosetum chaseae Tatianyx arnacttes „, Arthropogon yillosus Altoparadisium chapadense
n gymnocarpon a Tanceolata Apochloa euprepes s ubtiramulosa Coleataenia petersonn Coleataenia prionitis Coleataenia stenodes Coleataenia caricoides Coleataenia anceps Coleataenia longifolia Coleataenia tenera Anthaenantia ianata Panicum hylaeicum
Hymdnachne perrfambucensis Palgiantha tenella Oil/Cpi/PPP! VPPPCPPP Steinchisma iaxa Steinchisma decipiens pilPPs/SsPSl PP1PS Steinchisma spatheiiosa Ichnanthus paiiens Hopia obtusa Paspaium conjugatum Paspaium conspersum Paspaium malacophyllum i ’/ sptcnm iccccccccc: Paspaium remotum Paspaium vaginatum Puspninm hunmnpii Paspaium arundineiium Paspaium virgatum i ’ilSPilium PCI/Cl/C i CIS PC CSC IPP-IP; PC! Paspaium vaiidum Anthaenantiopsis rojasiana Panicum tuerckheimii ppppPisPicpys pspppipce Axonopus fissifolius Axonopus anceps •7 .• i .-V fa hydro tit hi cl Echinoiaena infiexa Oceiiochiba piauiensis Ocellochloa chapadensis Digitaria ciriaris Digitaria radicosa Digitaria setigera I cl PCS CP pi/p p /p cp p s Panicum gouinii Panicum repens Panicum bergii Panicum miliaceum Panicum stramineum Panicum fauriei Panicum nepheiophiium Panicum cervicatum Panicum rudgei i ciPCPCC PiysiciscPis Panicum olyroides Panicum chloroleucum Panicum racemosum Panicum trichoiaenoides Panicum virgatum Panicum aquaticum Panicum dichotomiflorum i ClPCPIP! pop:/SOPH Rupichioa acuminata Moorochioa eruciformis Melinis repens Chaetium bromoides Eriochloa punctata Urochioa plantaginea Urochloa maxima ; cipcpccUrochioa pptcipycp mutica, Zuioagaea bulbosa Setaria iachnea Setaria viridis Setaria geminata Stenotaphrum secundatum pei/ltCl PPCPPSiclCPVP Setaria sphaceiata Setaria palmifolia Setaria parviflora Setaria verticillata Cenchrus compressus Cenchrus ciiiaris Cenchrus mutiiatus Cenchrus sefaceus Panicum aden Panicum Dichanthelium cumbucana Dichanthelium clandestinum Dichanthelium koolauense Dichanthelium sabuiorum Dichanthelium acuminatum Acroceras zizanioides Echinochloa colona Echinochloa frumentacea OPPSPIPPPS PPi/SPIS Lasiacis sorghoidea Pseudechinolaena polystachya Parodiophyllochloa cordovensis Paroaiophyiiochioa missiona i ’ppippippy/pcppip pvppippi Parodiophyllochloa penicillata I '/iPlCPPl UPCPPPilPPP! i ’ppppip; ppppppipp Panicum sellbwii Panicum verrucosum Sacciolepis indica Trichanthecium cyanescens Trichanthecium wettsteinii Trichanthecium parvifoiium •PCPPPtPPCPIP! SOPP/POKPPPPP! Danthoniopsis dinteri Danthoniopsis petiolata
Fig. 1. Reconstructed ancestral character states on the Bayesian MCMC majority rule consensus tree. Branch shading indicates maximum parsimony recon struction (according to Reinheimer et al., 2012). Pie charts indicate Bayesian ancestral characters posterior probabilities at selected nodes. Numbers in parenthesis indicate the state with the highest likelihood based on the Bayes factor (BF) results. Two or more numbers in parenthesis indicates an ambiguous assignation of the ancestral character state. *, BF between 2 and 5 (positive support); **, BF between 5 and 10 (strong support). 1618 Reinheimer et al. — Macroevolution of panicoid inflorescences
Arundineiia hirta Homogenization degree Cnionacnne xooatatt Ischaemum afrum Apluda muiica I | (0) Non Cnn/ccpcucn iuivu.s Coelorachis selloana Jhacelurus digitatus I | (1) Partially ucrostegium nudum Miscantrfus japonicus Cleistachne sorghoides Sorghum bicolor (2) Fully Coix aquatica Cymbopogon flexuosus Heteropogon contortus Capillipedium parviflorum Boihriochloa bladhii Dichanthium aristatum /incucpcpcn aouuan „ , Hypkrmenia hirta . Schizachyrium scoparium Eiionurus muticus Tripsacum dactyloides Onccmcn::-,Z eam ays femcxi Opiismenopsis najada
Arthropogon villosus Aitoparadisium chapadense Altoparadisium chapadense b ! icmciepic OiPPPiPm Homolepts isocalycia Stephostachys mertensii Phanopyrum gymnocarpon Canastra lanceolata
Cciuaiaupp'i pncpiiis Coleataenia stenodes Coieataema caricoides Coleataenia anceps Coleataenia longifolia Coleataenia tenera Anthaenantia lanata i p p ic p p i im teicum ,, Panicum p , . . . Hymenachne donacifoha Hymenachne grumosa : ivmenacppe penmmpiicen.si.s Palgiantha tenella Oiacn/num ver.sicoio; Sfeinchisma iaxa Sioincnismi-i aecipiens Sieipcpieppi puiPc Steinchisma spathellosa Ichnanthus pallens „ Hgpia obtusa Paspaium con/ugatum Paspaium conspersum Paspaium malacophyllum Paspaium paniculatum Paspaium remotum Paspaium vaginatum Paspaium haumanii Paspaium arundinellum Paspaium virgatum I pisptiipp; aia/icvn I ppppipp; Pppicppp: Paspaium validum Anthaenantiopsis rojasiana _ Panicum tuerckheimii Streptostachys Axonopus fissirolius Axonopus anceps i hydrolithica I PPP!CPPPPP tPPPKP QcellocNoa stolonifera Ocellochaoa piauiensis Ocellochloa chapadensis Digitaria cifiaris Digitaria radicosa „ Digitaria, seiiqera i 'PPPipP! ippppppipjpp Panicum gouinii Panicum repens „ Panicum nergii Panicum mihaceum Panicum stramineum Panicum fauriei Panicum nephelophilum Panicum cervicatum „ Panicum rudqei i 'nninpp; miz~PiP:p:p, Panicum olyroides Panicum chloroleucum Panicum racemosum Panicum tricholaenoides Panicum virgatum Panicum aquaticum i flnicum dichotomiflorum Panicum pedersenii Rupichloa acuminata Moorpchloa eruciformis Meitnts repens Chaetium bromoides Eriochloa punctata Urpchloa piantaginea Urochioa maxima Urochloa mutica Puucum aniiaoiato Zuloagaea bulbosa Setaria lachnea Setaria viridis Seta.ria geminata Stenotaphrum secundatum Setaria macrostachya Setaria sphacelata Setaria palmifoha Setaria parviflora Setaria verticillata Cenchrus compressus Cenchrus ciliaris Cenchrus mutilatus „ Cenchrus setaceus Panicum adenophorum Panicum claytonii Dicip-tnii}eimm cumbucana Dicnanineimm cianaesiinum Dichan thelium koolauense Dichanthelium sabulorum \croceras z______Echinochloa colona Echinochloa frumentacea Qplismenus hirtellus Lasiacis sorghotdea Pseudechinoiaena polystachya Parodiophyllochloa cordovensis Parodiophyllochloa missiona / a:cci;cpav::cr:n;ca ovimkaa Parodiophyllochloa penicillata / amcnm incnaninum i flnicum mmegrana Panicum seffowii Panicum, verrucosum Sacctoiepts mdica Trichanthecium cyanescens Trichanthecium wettsteinii Trichanthecium parvHoiium Trichanthecium schwackeanum Danthoniopsis dinteri Subclade B Boivinellinae Subclade A Cenchrinae Melinidinae Panicinae Anthephorinae Paspalinae Otachyriinae Arthropogoninae Danthoniopsis petiolata
Fig. 1 Continued Reinheimer et al. — Macroevolution of panicoid inflorescences 1619
Presence/absence of terminal spikelet | | (0) Presence | (1) Absence
scoparium
rachis famosa
%toL°lufBu%sa iataeriia petersonii Ieataenia priomtis aa taenia stenodes eataenia. caricoides I eataenia anceps
oilosum donacifolia icensis
conjugatum
im haumani. i arundinellL
ltiopsis rojasiana n tuerckheimii i cnys asperifoha
oa stolonifera i oa piauiensis a.cnapadensis
dichotomifiorvm jm peders.enn iloa acuminata iloa eruciformis teiS&a
Panicum icana ttinum iense irum latum
■ofdovensis missiona
escens iteinii keanum Subclade B Boivineilinae Subclade A Cenchrinae Melinidinae Panicinae Anthephorinae Paspalinae Otachyriinae Arthropogoninae mthoniopsis
Fig. 1 Continued 1620 Reinheimer et al. — Macroevolution of panicoid inflorescences
Ta b l e 1. Results of the unordered versus an ordered model of character evolution using MCMC and ML methods
MCMCML Character Model In /'(model | data) s.e. InBF -log likelihood Difference in -log likelihood
Inflorescence aspect Free -102 17 0-047 3-88 -97 962 006 4-21 Restricted -106 052 0-078 -102 179 487 Homogenization Free -93 989 0-048 2-00 -8 8 086 746 0-42 Restricted -91 985 0-046 -8 8 509 734 Terminal spikelet truncation Free -4 0 959 0-026 2-67 -39 320407 7-28 Restricted -43 632 0-035 -4 6 608 380
3-00 r 300 3-00 r 3-00 r 2-50 r
2-25 2-25 2-25 2-25 1-88 Cl) a 1-50 1-50 □c 1-50 1-50 1-25
0-75 0-75 0-75 0-75 0-63
g0,1 qr1,0 <71,2 qr2,1 q 0,2 q 2,0 g1,2 q 0,2 7concl Tje-cond Character state transition
* 9-00 - 9-00 - * 0-30 900 500
7-20 - 7-20 - 0-22 6-75 3-75 a> 5-40 - 5-40 - ra 0-15 4-50 2-50 □C 3-60 - 3-60 - 007 2-25 1-25 1-80 - 1-80 -
0 i ------1 0 i 0 0 0
<70,1 qr1,0 <71,2 q 2,1 q 0,2 q 2,0 <71,0 qr1,2 ^hom ^de-hom Character state transition
Character state transitions u 106 Mean(1) Mean(2) s.d.(1) s.d.(2) W P (D (2)
0-80 <70,1 c/1,0 2-20 0-41 1-27 0-77 439503952 <0-0001* c/1,2 c/2,1 2-94 2-01 0-78 1-47 382477680 <0-0001* a) Inflorescence appearance c/0,2 c/2,0 2-97 2-30 0-83 1-31 364927854 <0-0001* ra 0-53 c/0,2 c/1,2 2-97 2-94 0-83 0-78 326610335 0-0103 □C Tond ^de-con d 1-89 2-40 1-46 1-29 660572475 <0-0001* 0-27 c/0,1 c/1,0 1-71 8-61 1-39 3-45 164652694 <0-0001* c/1,2 c/2,1 8-08 4-55 3-41 2-43 423964335 <0-0001* c/0,2 c/2,0 0-28 0-13 0-59 0-37 347953303 <0-0001* Homogenization qr0,1 qr1,0 c/1,0 c/1,2 8-61 8-08 3-45 3-41 336634520 <0-0001* 4-59 3-50 4-55 4-24 773532335 <0-0001* Character state transition ^de-hom ^hom
Truncation c/0,1 c/1,0 1-06 1-04 0-34 0-36 327174653 0-0017*
Fig. 2. Rates of change between the different states of inflorescence aspect (A), homogenization (B) and truncation (C). obtained using an unordered model of character evolution and MCMC method. (D) Results of the statistical differences among pairs of rates using the non-parametric Mann-Whitney C'-LesL. qi. j indicates the transition from the state i to the state /. An asterisk indicates significant differences under the non-parametric Mann-Whitney C'-LesL.
the temporal order of trait acquisition with the contingency and is homogenized (<73,1 < <74,2; Table 3 and Supplementary ordered tests using ML and MCMC methods. The contingency Data Fig. S3). However, the test could not infer with confi test, using ML methods, indicated, with positive evidence, that dence, whether the homogenization happened before or after non-homogenized inflorescences are more likely to evolve the change on the condensation aspect of the inflorescence when the inflorescence is lax (<72,1 < <74,3), and condensed ()7l,2 ~ <71,3; <74,2 ~ <74,3). We found positive evidence for inflorescences are more likely to evolve when the inflorescence the evolution of non-truncated inflorescences when the Reinheimer et al. — Macroevolution of panicoid inflorescences 1621
T a b l e 2 . Bayes factor and likelihood ratio results o f the BayesTraits Discrete analyses using MCMC and ML methods
MCMCML Character Model In /'(model | data) s.e. InBF -log likelihood LR
Truncation with homogenization Independent -104 107 0-101 4-60 -1 0 1 94 12-5** Dependent -9 9 502 0-079 -9 5 69 Inflorescence condensation with homogenization Independent -133 763 0-105 22-91 -1 3 3 40 20-71*** Dependent -1 1 0 85 0-134 -1 2 3 04 Inflorescence condensation with truncation Independent -106 54 0-094 15-16 -104 51 34.92*** Dependent -9 1 376 0-116 - 8 7 05 Inflorescence aspect with plant habit Independent -127 484 0-041 1-13 - 1 0 1 80 1-16 Dependent -1 2 8 617 0-068 - 1 0 1 22 Homogenization with plant habit Independent -1 1 8 054 0-054 0-09 -1 1 8 92 1-7 Dependent -118 144 0-059 -1 1 8 07 Truncation with plant habit Independent -103 229 0-053 0-15 -101 14 1-58 Dependent -1 0 3 588 0-05 - 9 9 56 Inflorescence aspect with photosynthesis Independent -102 714 0-087 2-02 - 9 9 79 7-9*? Dependent -1 0 0 694 0-11 -9 5 84 Homogenization with photosynthesis Independent -92 662 0-091 0-25 - 8 9 75 3-08 Dependent - 9 2 408 0-104 - 8 8 21 Truncation with photosynthesis Independent -77 251 0-085 1-82 -73 90 7-38 Dependent -7 5 426 0-097 - 7 0 21
Likelihood values were tested over a distribution. *? P = 0-09. * P < 0 05. ** P < 0-01. *** P < 0-001. inflorescence is lax (<72,1 < <74,3). However, there was no evi <0-1 % of the sampled Markov chains (Fig. 3). The alternative dence for temporal order in the gain of these traits (Table 3 and route which involves first a transition towards truncated inflor Supplementary Data Fig. S3). The analyses indicated very escences (<73,4 ~ 0) followed by a transition to condensed strong evidence for conditional evolution between homogen inflorescences (<74,2 > 0) is not supported (Fig. 4). The transi ization and truncation events (Table 3 and Supplementary tion that involves first the truncation process was assigned a Data Fig. S3). The homogenization process occurred before value of zero in over 96 % of the sampled Markov chains the truncation of the terminal spikelet (q 1,2 > q 1,3), and trun (Fig. 3). The inverse path from the most derived state of con cation occurred if the inflorescence is homogenized (q 1,3 < densed and truncated inflorescences toward the most ancestral <72,4; Table 3 and Supplementary Data Fig. S3). Similar state of lax and non-truncated inflorescences is most likely to trends were inferred based on MCMC methods (Fig. 3). occur via an initial gain of lax aspect (<72,4 > 0) followed Ancestral state reconstruction indicated that the ancestor of by the gain of a terminal spikelet (<74,3 > 0; Fig. 4). These Panicoids had a lax, homogenized and non-truncated inflores transitions were assigned a value of zero in <0-8 % of the cence. The most likely evolutionary pathway from the ances sampled Markov chains (Fig. 3). The alternative inverse tral to the derived state of the traits can be inferred from the route may not be possible given a restriction to gain a terminal posterior probabilities of the transition rate parameters spikelet when the inflorescence is condensed (<72,1 ~ 0; (Fig. 3). The posterior distribution shows that the most likely Fig. 4). This transition was assigned a value of zero in 93 % path from the ancestral state of lax and homogenized inflores of the sampled Markov chains (Fig. 3). Finally, the posterior cences to the derived state of condensed and non-homogenized distribution shows that the most likely path from the ancestral inflorescences may involve (a) a transition toward condensed state of non-truncated and homogenized inflorescences to trun inflorescence (<74,2 > 0) followed by a transition toward non cated and non-homogenized inflorescences involved first a homogenized inflorescence (<72,1 > 0), or (h) a transition truncation process (q 1,2 > 0) followed by the loss of hom toward non-homogenized inflorescence (q4,3 > 0) followed ogenization (<72,4 > 0; Fig. 4). None of these transitions in by a transition toward condensed inflorescence (<73,1 > 0; this evolutionary path were assigned a value of zero in the Fig. 4). These transitions were assigned a value of zero in sampled Markov chains (Fig. 4). The alternative route that <2 % of the sampled Markov chains. The inverse path from involves first the loss of homogenization (q 1,3 ~ 0) followed the most derived state of condensed and non-homogenized by truncation (<73,4 > 0) may not be possible (Fig. 4). The inflorescences toward the ancestral state of lax and homoge transition <71,3 was assigned a value of zero in 80 % of nized inflorescences is not likely to occur (<73,4 ~ 0; q2,4 ~ the sampled Markov chains (Fig. 3). The inverse path from 0; Fig. 4). Both transitions were assigned a value of zero in the derived state of non-homogenized and truncated inflores about 60 % of the sampled Markov chains (Fig. 3). In addition, cences toward the ancestral state of homogenized and non- the posterior distribution shows that the most likely path from truncated inflorescences is most likely to occur via a loss of the ancestral state of lax and non-truncated inflorescences to the terminal spikelet of the main axis first (<74,3 > 0) followed condensed and truncated inflorescences involved first a transi by the gain of homogenization (<73,1 > 0; Fig. 4). These tran tion toward condensed (<73,1 > 0), followed by a transition sitions were assigned a value of zero in <0-1 % of the sampled toward truncated inflorescences (<71,2 > 0; Fig. 4). Both transi Markov chains (Fig. 3). The alternative inverse route may not tions in this evolutionary path were assigned a value of zero in be possible given a restriction to gain homogenization when 1622
T a b l e 3. Likelihood ratio values for test of contingent evolution and temporal order between inflorescence characters n c -f O O P 3 § ’§ 3 3 >3 33 g b-o o cp > 9, a B Reinheimer P ’ 43 44 9 &-3 ■9 & h ctj E-i > K D U ID • H fl M ^ l f O O £ l f O 3 p *3 CD 3 T 33'S t O 0) O ctj 0 0 e i l T3 _fc) _33 £ . £ ^ .3 3 <2 3 .3 ^ .g £ o o o-i CJ 4 ^ 73 jS43 2 D CD CD CD > O l f <0-0 43 CD CD § 2 o 0 33 ° P 2 H o £ "EL 2 $ fe ctj CD a l a fl 1s c _ CD O i s s 45 ■fl ^ ' g •2 S & « D >•> CD cti
c £ ^ s § 's = e a. — al. et o a _ bJD , W> a £ 33 s e 9 o o 9 .3 5 3 . ! S Cfl 1 £ : ! fi 2 ^ .2 X) a Macroevolution o f panicoid inflorescences f panicoid o Macroevolution 4 4 o -S E E 4D &H &H V V S 3 33 c prerequisite which allows the later loss of the terminal spikelet spikelet terminal the of loss later the the allows which prerequisite confirm results The evolution. of model pendent the inflores that found non-homogenized a had method, have may reconstruction ancestor panicoid parsimony the using ter a of (Reinheimer gain reports and homogenization condensation, of processes unordered an following evolved inflorescence the panicoids, presented results inflorescence, panicoid the of lability parent inflorescence panicoid the of reversals and origins pendent ceae, Gesneriaceae, among others; Weberling 1965, 1985; Rua Rua al.. 1985; et 1965, Scrophularia- Weberling others; Lamiaceae, among Resedaceae, Gesneriaceae, Podostemaceae, Primulaceae, ceae, Brassicaceae, Malvaceae, Commelinaceae, Fabaceae, Amaranthaceae, Cyperaceae, Chenopodiaceae, (i.e. homo as described fam been genized many also In have 1985). inflorescences (1977, truncated ilies Weberling by suggested originally condensa and simultaneously. and occurred have homogenization homogenization of could that tion gain the suggesting in evi no order was condensation, there temporal for Interestingly, axis. dence main inflorescence the of terminal the of loss the and inflores the among homogenization aspect, contingency cence evolutionary an of existence and (1998). (1996) Rua by Weberling and in hypothesized intermediate Rua an previously homogenized was as partially requires and not florescence that is condensed to lax process from and inflorescence transition two-step a inflorescences homogenized to condensed necessarily to In lax non-homogenized from from de-homogenization. evolution of the that process the homogeniza over of process favoured the is tion reconstruction, parsimony meth the Under in Reinheimer by differences to used due is odology result conflicting This cence. previous with agree partially results the These over spikelet. favoured minal appear de-condensation, of truncation all and processes among the possible which are de-homogenization in states) transitions of (direct character origin evolution the of After model inflorescence. homo lax, a had non-truncated grasses and panicoid the genized of ancestor the that cate evolution general some of existence the support work ap work the this despite in any previous in However, another, pattern. to with specific state a one without from agreement switch direction, easily can in characters here, studied inde traits many show methods MCMC and ML of the Reconstructions of % 56 in zero of value a This 4). assigned Fig. 0; (<74,2 was ~ tmncated transition already is inflorescence the spikelet of the main axis. Overall, the data presented here here a being presented data homogenization of the trend Overall, general the axis. support main strongly the of spikelet inde the reject strongly analyses MCMC and ML time, ary Reinheimer addition, below. discussed are which patterns, and trends ary (Reinheimer 3). (Fig. chains Markov sampled s lwr o siees o te nisem nlrsec may inflorescence angiosperm the of spikelets) or flowers as 1999; Acosta Acosta 1999; Despite the lability of inflorescence characters in evolution in characters inflorescence of lability the Despite indi reconstructions state ancestral ML and MCMC Both iia ted my cu i ohr nisem aiis as families, angiosperm other in occur may trends Similar 2012). In these cases, the loss of the apical organs (such (such organs apical the of loss the cases, these In 2012). t i, a et t al.. et 2012). This suggests that inflorescence inflorescence that suggests This 2012). t i, a et t al. et 2009; Panigo Panigo 2009; SCUSSI N IO S S U C IS D (2012) suggest, as do our results, results, our do as suggest, (2012) 2012). Reinheimer Reinheimer 2012). et al. et (2012) and those used here. here. used those and (2012) t al.. et 2011; Reutemann Reutemann 2011; t al. et (2012), (2012), 450 0 350 0 700 0 6 00 0 400 0 - 2 -6 0 + 1-63 3-29 + 1-78 _ 1-82 ±0-91 0-58 + 0-92 300 0 600 0 5000 350 0 Z = 0-03 Z = 0-01 Z= 0-02 ■ Z = 0-61 250 0 - 5000 - 300 0 - 400 0 g 250 0 2000 400 0 - 1 S' 300 0 a- 2000 1500 300 0 - 1 1500 200 0 1000 - 200 0 - 1 2 1000 - 1000 500 500 1000 - 1 il li , , , r-bo 0 0 n , , , 0 0 ■ ,i i...... , Fig. 3. Posterior probability distribution of the rate coefficients values of the correlated evolution model using the MCMC method. (A) Inflorescence aspect and homogenization. (B) inflorescence aspect and truncation, and (C) homogenization and truncation, q i.j indicates the transition from the state i to the state j. Z-values represent the proportion of the sampled runs from the Markov chain in which the parameter was assigned a value of zero. Values are the average and the standard deviation of the parameter values from the sampled runs. 0to \ uo 0\ K> 4- B 10000 10000 900 0 10000 9000 - — 2 -85 ± 0 - 7 9 900 0 - — 2-96 ± 1-08 800 0 ■ 0 -09 ± 0 - 4 2 9000 - — 2-71 + 0-79 >3 8000 - Z = 0-000 800 0 - Z = 0-002 700 0 I Z = 0-93 8000 - Z = 0-000 7000 - 700 0 - 600 0 ■ 7000 - S' 6000 600 0 6000 - - - 5000 I 5000 5000 5000 - ~ - 400 0 4000 400 0 4000 - 3000 _ 300 0 _ 300 0 I 3000 - 2000 _ 200 0 _ 200 0 I 2000 - 1000 - 1000 - 1000 I 1000 - 0 ...... 0 i i i i i i i i i i 0 “ I I...... 0 ------1------1------1------1------1------1------1------1------1------1------1_ . ,-. S' - S' „ .V — S’ — S’ _ S ' — S’ — S’ - S ’ _. S’ s 1Ts « Ts 0S\fNf NfN%\? ^ 1 <£ r§> tip Fig. 3 Continued 8000 p n3000 3000 >3 1-39 ±0-71 70001 0-13 ±0-33 2-28 ±0-66 2-07 ±0-68 2500 2500 2000 Z = 0-000 Z = 0-80 Z = 0-000 Z = 0-000 6000 I S' 2000 2000 5000 I ffc c0) 1500 3 4000 I 1500 1500 S’ 1000 3000 I 1000 1000 LL 2000 I 500 500 500 1000 I Q Inn r CL a T kT kS kS K P K P kS kS kS '~)<3N\ c 1500 1500 1500 ^5 1000 53 g 2 ' 500 500 500 0 0O n 0 II rim ll III— II - 0 I III— lU <£> # ; ^ '-VQ \ \ Fig. 3 Continued C\ro ■j\ 1626 Reinheimer et al. — Macro evolution o f panicoid inflorescences Inflorescence aspect vs. Homogenization require a previous organization or specialization of the axillary branches below the apex, such as simplification of the branding pattern and the presence of long and short branches (Rua and Condensed V <71,2 Non-homogenized Weberling, 1998; Rua, 1999; Reinheimer et al., 2012). Homogenization and truncation may represent evolutionary 1'3 <72,1 \ advantages for angiosperm taxa, and in particular for the grasses, given the reiterative pattern of acquisition of such traits Lax Condensed for many species, or by entire angiosperm families. Non-homogenized Homogenized The data presented here show that the evolutionary pathway from the ancestral inflorescence type of panicoids to the most- derived inflorescence type could have been accomplished either by a five- or a three-step process, indicating that the con densed, non-homogenized and truncated inflorescence may q4,3 have resulted from one of two different evolutionary scenarios (Fig. 5). Our results also suggest that reversion from the derived inflorescence type to the ancestral condition is a three- B Inflorescence aspect vs. Truncation step process, but one that cannot be fully completed given the constraint of a lax inflorescence to become homogenized (Fig. 5). Under this tentative scenario, the lax, non- homogenized non-truncated inflorescence type, which is the most abundant type among panicoids (Reinheimer et al., 2012; this work), may represent a first step from the ancestral to the derived type, or, in several cases, an incomplete rever sion to the ancestral morphology (Fig. 5). This may also explain why the lax, homogenized non-truncated inflorescence is probably one of the less abundant types of panicoid inflor escences (Reinheimer et al., 2012; this work). It has been postulated that developmental and genetic mechanisms available for selection, as well as ecological- physiological factors could impose constraints to the biological diversity of angiosperms (Kellogg, 2000; Doust and Kellogg 2002; Friedman and Harder 2004, 2005; Prusinkiewicz et al., 2007; Morrone et al., 2011).The present work had Homogenization vs. Truncation demonstrated that in panicoids some inflorescence character- state transitions could occur frequently, some are rare, and some others are unlikely to happen. The fact that inflorescence character-state transitions occur with different rates indicates that, though such transitions are developmentally and genetic ally possible, their specific frequency may reflect different in tensities of selection. In contrast, those character-state transitions that are unlikely to happen or that have never been documented may have such high morphogenetic costs that there is a restriction on the availability of these pathways for selection. It is also possible that, because of the high number of convergences and reversals observed, inflorescence diversification has occurred as a response to various ecological- physiological pressures. However, so far neither plant lon gevity nor the photosynthetic type have been shown to be Fig. 4. Flow diagram showing the most probable evolutionary pathway from strongly correlated with the evolution of the inflorescence (A) the ancestral state of lax and homogenized inflorescence to the derived characters analysed here. state of condensed and non-homogenized inflorescence, (B) the ancestral The present study clearly reveals some comprehensive state of lax and non-truncated inflorescence to the derived state of condensed and truncated inflorescence, and (C) the ancestral state of homogenized and macroevolutionary trends on the evolution of the panicoid in non-truncated inflorescence to the derived state of non-homogenized and trun florescence. Overall, the differences among the frequencies of cated inflorescence, qi, j indicates the transition from the state i to the state j. character-state transitions and the dependent evolution of in Thick black arrows highlight transitions that have a low posterior probability of florescence traits suggest a complex of constraints imposing being zero (< 5 %'), thin arrows denote transitions that have higher probability restrictions to the diversification of the panicoid inflorescence. of being zero (56-64 %') and dotted arrows indicate transitions that have a very high probability of being zero (> 8 0 %). The grey shaded boxes indicate the Among such constraints, morphogenetic cost (the cost to es inferred ancestral state. tablish new developmental and genetic mechanisms versus Reinheimer et al. — Macroevolution of panicoid inflorescences 1627 / f l •s V V T Lax Condensed i t , Non-homoaenized ------f c . Non-homogenized Non-truncated Non-truncated Lax Condensed Condensed Condensed Homogenized Homogenized Homogenized Non-homogenized Non-truncated Non-truncated Truncated Truncated • n / • i 3 Lax Homogenized Non-truncated Fig. 5. Diagram summarizing our results concerning the macroevolutionary patterns of panicoid inflorescence evolution. The white box represents the ancestral inflorescence type. Grey boxes indicate intermediate inflorescence types. The black box represents the most derived inflorescence type. The dotted arrow indicates an unlikely transition according to MCMC methods. Numbers in the top left corner of the boxes indicate the inflorescence type (see Supplementary Data Fig. SI). Black ovals represent a spikelet. The asterisks represent the missing terminal spikelets of the inflorescence. (A) The evolutionary pathway from the ancestral inflorescence type to the most derived inflorescence type may have occurred via either a five- (3 -1 -5 -7 -8 -6 ) or three-step (3—7—8—6) process. (B) The most likely evolutionary pathway from the most derived inflorescence type to the ancestral inflorescence type. the cost of using developmental and genetic mechanisms ACKNOWLEDGEMENTS already established) may be a key factor. We thank Diego Salariato and Gustavo D. Marino for assistance More detailed analyses of particular clades of panicoids, with BayesTraits and non-parametric analysis, respectively. We involving a more exhaustive taxon sampling, as well as a are also grateful to three anonymous reviewers for critically larger set of inflorescence traits and environmental para reading the manuscript. Funding was provided by the Agencia meters, could reveal local evolutionary pathways that do Nacional de Promocion Cientifica y Tecnologica Argentina not necessarily agree with the general trends described (grant no PICT-2010-1067) to R.R. and the Universidad here. It would also be interesting to investigate whether Nacional del Litoral (grant no UNL-CAID + D 2009) to R.R. similar trends can be found in other grass or angiosperm lineages. Such investigations would reveal to what extent the models of inflorescence evolution described here can be extended to other groups. LITERATURE CITED Acosta JM, Perreta M, Amsler A, Vegetti AC. 2009. The flowering unit in the synflorescences of Amaranthaceae. Botanical Review 75: SUPPLEMENTARY DATA 365-376. Aliscioni SS, Giussani LM, Zuloaga FO, Kellogg EA. 2003. A molecular Supplementary data are available online at www.aob.oxford- phylogeny of Panicum (Poaceae: Paniceae): test of monophyly and phylo joumals.org and consist of the following. Table SI: dataset genetic placement within the Panicoideae. American Journal o f Botany used in this study. Figure SI: incidence of a given inflores 90: 796-821. Doust AN, Kellogg EA. 2002. Inflorescence diversification in the Panicoid cence type among panicoids main lineages. Figure S2: rates bristle grass clade (Paniceae, Poaceae): evidence from molecular phylo- of change between the different states of inflorescence genies and developmental morphology. American Journal of Botany aspect, homogenization and truncation obtained using an un 89: 1203-1222. ordered model of character evolution and the maximum likeli Doust AN, Devos KM, Gadberry MD, Gale MD, Kellogg EA. 2005. The genetic basis for inflorescence variation between foxtail and green hood method. Figure S3: posterior probability distribution of millet (Poaceae). Genetics 169: 1659-1672. the rate coefficients values of the correlated evolution model Friedman J, Harder LD. 2004. Inflorescence architecture and wind pollin using the maximum likelihood method. ation in six grass species. Functional Ecology 18: 851-860. 1628 Reinheimer et al. — Macroevolution of panicoid inflorescences Friedman J, Harder LD. 2005. Functional associations of floret and inflores Reinheimer R, Vegetti AC. 2008. Inflorescence diversity and evolution in the cence traits among grass species. American Journal of Botany 92: PCK clade (Poaceae: Panicoideae: Paniceae). Plant Systematics and 1862-1870. Evolution 215: 133-167. Giussani LM, Cota-Sanchez JH, Zuloaga FO, Kellogg EA. 2001. A mo Reinheimer R, Zuloaga FO, Vegetti AC, Pozner R. 2009. Diversification of lecular phylogeny of the grass subfamily Panicoideae (Poaceae) shows inflorescence development in the PCK Clade (Poaceae: Panicoideae: multiple origins of C4 photosynthesis. American Journal of Botany 88: Paniceae). American Journal of Botany 96: 549-564. 1935-1944. Reinheimer R, Amsler A, Vegetti AC. 2012. Insight into panicoid inflores Gonzalez-Voyer A, Fitzpatrick JL, Kolm N. 2008. Sexual selection deter cence evolution. Organisms Diversity and Evolution 12: el 10. http:// mines parental care patterns in cichlid fishes. Evolution 62: 2015-2026. dx.doi.org/10.1007/s 13127-012-0110-6. Huelsenbeck JP, Ronquist F. 2001. MrBayes: Bayesian inference of phylo Reutemann A, Lucero L, Guarise N, Vegetti AC. 2012. Structure of the genetic trees. Bioinformatics 17: 754-755. Cyperaceae inflorescence. Botanical Review 78: 184-204. Kellogg EA. 2000. Molecular and morphological evolution in the Rua GH. 1996. The inflorescences of Paspaium L. (Poaceae, Paniceae): the Andropogoneae. In: Wilson L, Morrison DA. eds. Monocots: systematics Quadrifaria group and the evolutionary pathway towards the fully homo and evolution. Melbourne: CSIRO, 149-158. genized, truncated common type. Plant Systematics and Evolution 201: Liu Q, Zhao NX, Hao G. 2005. Inflorescence structures and evolution in sub 199-209. family Chloridoideae (Gramineae). Plant Systematics and Evolution 251: Rua GH. 1999. Inflorescencias: bases teoricas para su andlisis. Buenos Aires: 183-198. Sociedad Argentina de Botanica. Liu Q, Peterson PM, Columbus JT, Zhao N, Hao G, Zhang D. 2007. Rua GH, Weberling F. 1998. Growth form and inflorescence structure of Inflorescence diversification in the ‘finger millet clade” (Chloridoideae, Paspaium L. (Poaceae: Paniceae): a comparative morphological approach. Poaceae): a comparison of molecular phylogeny and developmental Beitrdge zur Biologie der Pflanzen 69: 363-431. morphology. American Journal of Botany 94: 1230-1247. Sanchez-Ken GJ, Clark LG. 2010. Phylogeny and a new tribal classification Morrone O, Aagesen L, Scataglini MA, et al. 2011. Phylogeny of the of the Panicoideae s.I. (Poaceae) based on plastid and nuclear Paniceae (Poaceale: Panicoideae): integrating plastid DNA sequences sequence data and structural data. American Journal of Botany 97: and morphology into a new classification. Cladistic 28: 333-356. Newton MA, Raftery AE. 1994. Approximate Bayesian inference with the 1732-1748. Weighted Likelihood Bootstrap. Journal of the Royal Statistical Society Suchard MA, Weiss RE, Sinsheimer JS. 2001. Bayesian selection of con 56: 3-48. tinuous time Markov chain evolutionary models. Molecular Biology Nylander JAA. 2004. MrModeltest v2. Program distributed by the author. and Evolution 18: 1001-1013. Evolutionary Biology Centre, Uppsala University, http://www.abc.se/ Watson L, Dallwitz MJ. 1992. The grass genera o f the world. Wallingford, ~nylander/mrmodeltest2/mrmodeltest2 .html. UK: CAB International. Pagel M, Meade A. 2006. Bayesian analysis of correlated evolution of discrete Weberling F. 1965. Typology of inflorescences. Botanical Journal Linnaean characters by reversible-jump Markov Chain Monte Carlo. The American Society 59: 15-221. Naturalist 167: 808-825. Weberling F. 1985. Aspectos modernos de la morfologia de las inflorescen Pagel M, Meade A, Barker D. 2004. Bayesian estimation of ancestral char cias. Boletm de la Sociedad Argentina de Botanica 24: 1-28. acter states on phylogenies. Systematic Biology 53: 673-684. Zuloaga F, Morrone O, Giussani LM. 2000. A cladistic analysis of the Panigo E, Ramos J, Lucero L, Perreta M, Vegetti AC. 2011. The inflores Paniceae: a preliminary approach. In: Jacobs SWL, Everett J. eds. cence in Commelinaceae. Flora 206: 294-299. Grasses: systematics and evolution. Melbourne: CSIRO, 123-135. Prusinkiewicz P, Erasmus Y, Lane B, Harder LD, Coen E. 2007. Evolution Zuloaga FO, Scataglini MA, Morrone O. 2010. A phylogenetic evaluation and development of inflorescence architectures. Science 316: 1452-1456. of Panicum sects. Agrostoidea, Megista, Prionita and Tenera Rambaut A, Drummond AJ. 2007. Tracer vl-5. http://beast.bio.ed.ac.uk/ (Panicoideae, Poaceae): two new genera, Stephostachys and Sorengia. Tracer. Taxon 59: 1535-1546. ^ ^ ^ v$' v!V +' tg>' S' ,+ ,+ 'S'- ^ + 'S-' Transition Transition Transition Transition :1 ; parameter value q^ ,2 parameter value q^ ,3 parameter value g2,1 parameter value q2,4 2 600 0 600 0 5000 2-02+1-02 0-53 ±0-84 4-98 + 1-81 450 0 4 -7 7 + 1 -9 2 5000 5000 Z = 0-000 Z = 0-64 Z = 0-000 400 0 Z = 0-000 ' 400 0 400 0 350 0 300 0 _ 300 0 300 0 250 0 2000 2000 2000 ^5 1500 53 1000 1000 1000 g 500 2 ' 0 _i_ _i_ _i_ 1 0 0 ^ 0 <& 4°