Read the pdf and answer the questions

Phylogenetic analyses reveal the shady history of C4 grasses Erika J. Edwardsa,1 and Stephen A. Smithb

aDepartment of Ecology and Evolutionary Biology, Brown University, Providence, RI 02912; and bNational Evolutionary Synthesis Center, Durham, NC 27705

Edited by Michael J. Donoghue, Yale University, New Haven, CT, and approved December 31, 2009 (received for review August 24, 2009)

Grasslands cover more than 20% of the Earth's terrestrial surface, and their rise to dominance is one of the most dramatic events of biome evolution in Earth history. Grasses possess two main photo- synthetic pathways: the C3 pathway that is typical of most plants and a specialized C4 pathway that minimizes photorespiration and thus increases photosynthetic performance in high-temperature and/or low-CO2 environments. C4 grasses dominate tropical and subtropical grasslands and savannas, and C3 grasses dominate the world's cooler temperate grassland regions. This striking pattern has been attributed to C4 physiology, with the implication that the evolution of the pathway enabled C4 grasses to persist in warmer climates than their C3 relatives. We combined geospatial and molecular sequence data from two public archives to produce a 1,230-taxon phylogeny of the grasses with accompanying climate data for all species, extracted from more than 1.1 million herba- rium specimens. Here we show that grasses are ancestrally a warm-adapted clade and that C4 evolution was not correlated with shifts between temperate and tropical biomes. Instead, 18 of 20 inferred C4 origins were correlated with marked reductions in mean annual precipitation. These changes are consistent with a shift out of tropical forest environments and into tropical wood- land/savanna systems. We conclude that C4 evolution in grasses coincided largely with migration out of the understory and into open-canopy environments. Furthermore, we argue that the evo- lution of cold tolerance in certain C3 lineages is an overlooked innovation that has profoundly influenced the patterning of grass- land communities across the globe.

C4 photosynthesis | climate niche evolution | cold tolerance | phylogeny

The term “C4 photosynthesis” refers to a suite of biochemical andanatomical modifications to the standard plant C3 photosynthetic pathway that work to concentrate CO2 around the carbon-fixing enzyme Rubisco. The C4 pathway greatly improves photosynthetic performance in situations that promote photorespiration, typically high-temperature and low-CO2 environments (1). The pathway also promotes more efficient photosynthetic water use, because the CO2 concentration mechanism allows C4 plants to maintain a lower sto- matalconductanceforagivenphotosyntheticrate.C4photosynthesisis estimatedtohaveevolvedatleast50timesinterrestrialplants(2,3)and is most prominent in grasses, where roughly half the species (∼5,000) are C4 (4), including economically important species such as maize, sugarcane, sorghum, and switchgrass. C4 grasses currently dominate wideregionsoftheEarth,anditisestimatedthattheyaccountforupto 25% of global annual terrestrial primary production (5). Recent studies suggest that C4 photosynthesis is a relatively

recent innovation in plants, with the earliest appearances coin- ciding with plummeting atmospheric CO2 levels during the mid- Oligocene and many origins occurring much later (4, 6). In grasses, the evolutionary history of C4 photosynthesis is complex, with multiple origins, probable reversals, and a general lag-time between the evolution of the pathway and the formation of C4- dominated ecosystems (3, 6–8). Although low atmospheric CO2 certainly was a prerequisite for C4 evolution, it is thought that multiple stressors worked in concert to promote the pathway. C4 plants are notably prevalent in arid, high-light, saline, and dis- turbed environments, and it is largely accepted that water stress

has provided a strong selection pressure for C4 evolution in eudicots (4). Grasses have long been viewed as an interesting exception to this pattern (9). Significant positive correlations between C4 grass abundance and growing season temperature have been documented at both continental and regional scales (10–13); C4 grasses dominate tropical grasslands and savannas but are virtually absent from cool-temperate grasslands and steppes. Furthermore, both experimental measurements of photosynthetic light use efficiency (termed “quantum yield”), and predictions of leaf models of C3 and C4 photosynthesis provide strong evidence that C4 grasses outperform C3 grasses at higher temperatures (5, 14–16). Most studies concerning C4 grasses and precipitation have focused on C3/C4 mixed temper- ate grassland systems, where the timing of C4 growth often is restricted to periods with significant rainfall (12, 13, 17, 18). The C4 pathway thus has been largely dismissed as an adap-

tation to water stress in grasses, with all data indicating that C4 evolution allowed grasses to invade and diversify successfully into hot climates. However, few studies have compared C4 grasses with their closest living C3 relatives. C4 origins are not distributed uniformly across Poaceae but are clustered in one major grass lineage, informally named the “PACMAD” clade (3, 6). Most of the C3 grasses that dominate cool-climate grasslands belong to the Pooideae, a lineage that last shared a common ancestor with PACMAD grasses ≈65–50 Mya (19). It thus is possible that differences between Pooideae and PACMAD grasses that have nothing to do with photosynthetic pathway variation are driving the apparently strong sorting of C3 and C4 species along temperature gradients. We employed an explicitly phylogenetic approach to assess the

evolutionary history of climate niche space for grasses on a worldwide scale. We used two public archives of data to build the most inclusive phylogeny for Poaceae that permitted analysis of a climate dataset for all taxa (Materials and Methods). This analysis resulted in a 1,230 taxon tree with broad coverage of all of the major Poaceae lineages and, importantly, good sampling of known C3 and C4 transitions. Our phylogenetic tree includes roughly 10% of all Poaceae species and is the largest grass phylogeny built to date. We identified 21 nodes representing evolutionary transitions between photosynthetic types and used these nodes to generate phylogenetically independent C3/C4 pairwise comparisons (Tables S1 and S2) (20). Nearly all iden- tified photosynthetic transitions were reconstructed as C4 origins, with one purported reversal. Fifteen of the 21 transitions occurred within the Panicoideae, a major PACMAD lineage containing more than 3,000 species. To account for topological uncertainty in this area of the phylogeny, we performed Bayesian analyses for a 299-taxon Panicoideae dataset and ran all diver-

Author contributions: E.J.E. and S.A.S. designed research, performed research, analyzed data, and wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. 1To whom correspondence should be addressed. E-mail: [email protected].

This article contains supporting information online at www.pnas.org/cgi/content/full/ 0909672107/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.0909672107 PNAS Early Edition | 1 of 6

EV O LU

TI O N

gence analyses across the Bayesian posterior distribution of trees. We also used this reduced dataset to reconstruct envi- ronmental niche evolution in the Panicoideae and tested whether shifts between photosynthetic types corresponded with significant changes in temperature and precipitation niche optima under a stabilizing selection model of evolution.

Results Climate data extracted from all available geo-referenced her- barium material provided clear evidence that certain grass line- ages have specialized in certain habitats (Fig. 1). Importantly, two strictly C3 grass lineages, the Pooideae and the Dantho- nioideae, stood apart as inhabiting much cooler environments, measured either by mean annual temperature (MAT) (Fig. 1) or temperature of the wettest, coldest, or warmest month (Fig. S1). All other C3 lineages were indistinguishable from C4 lineages with respect to their temperature profiles. Distinctive sorting of precipitation variables was less apparent, although Pooideae occupied the drier end of the spectrum alongside the C4 lineages Aristidoideae and Chloridoideae (Fig. S2). Our phylogenetic analyses concurred with these general tem-

perature and precipitation profiles for the different grass clades, particularly the observation that, with the exception of Pooideae and Danthonioideae, grasses are warm-climate specialists (Fig. 2). The evolution of C4 photosynthesis appears to have had little influence on gross temperature niche: Only 10 of 21 photo- synthetic shifts resulted in increases in MAT in the C4 lineage, with a mean divergence of −0.13 °C. Likewise, there was no significant relationship between photosynthetic pathway and mean temperature of the hottest, coldest, or wettest months (Table 1). In sharp contrast, there were large and predictable shifts in mean annual precipitation (MAP) between C3 and C4 sister taxa; 18 of the 21 divergences resulted in C4 taxa living in areas with lower MAP (P < 0.01), with a mean difference of 546 mm year−1. Fifteen of those divergences also resulted in shifts toward increased seasonality of precipitation. Results were statistically similar using a more conservative, reduced dataset generated by removing three focal taxa with questionable placement in the phylogeny and recoding a C3-C4 intermediate species as C3 (Table 1) and also were robust to testing against a wide range of alternative tree topologies within Panicoideae (Table S3). These results are highly consistent with an earlier

analysis of the Hawaiian grass flora that used a similar approach but was based on very limited geographic and phylogenetic sampling (21). Analyses of niche evolution within Panicoideae provided fur-

ther support that C4 evolution was associated with shifts into drier, but not warmer, environments. For most climate variables, two-optimum stabilizing selection Ornstein-Uhlenbeck (OU) models were preferred over Brownian motion or single-optimum models, implying that C3 and C4 lineages have experienced divergent selection over time (Table 2). However, the mean C3 and C4 temperature optima were not largely different, and in all cases, the model inferred lower temperature optima for C4 Panicoideae lineages than for C3 Panicoideae lineages. MAP received the strongest support for a two-optimum model, with the C4 optimum inferred to be drier than the C3 optimum by more than 500 mm year−1.

Discussion These analyses provide clear evidence that C4 origins in grasses coincided with ecological shifts into drier environments. How- ever, the actual values of precipitation divergences across our datasets largely occurred squarely within what generally is con- sidered a mesophytic moisture niche (22). No reconstructed C4 ancestral focal node had a MAP of less than 500 mm year−1; across all C3/C4 contrasts, the C3 MAP average was 1,839 mm year−1, and the C4 average was 1,229 mm year

−1. In the tropics, such differences in annual precipitation values often lead to major changes in biome structure, because a MAP of 1,500 mm year−1 generally is considered the minimum amount of rainfall required to support a closed-canopy forest (23).Considering the relative values of MAT, MAP, and precipitation seasonality, one-third of our focal nodes present climate profiles that are highly consistent with a shift from a closed-canopy tropical moist forest to an open tropical woodland/savanna biome (Fig. 3) (24). This finding also is consistent with the unexpected and significant increase in temperature seasonality in C4 lineages, suggesting geographic movements north and south from the equator and into the savanna belt. This general picture of a shift from closed to open canopies mirrors results of an independent dataset that focused on qualitative ecological assessments for 117 genera of grasses (22) and suggests a direct link between C4 evolution and the establishment of the savanna biome.

Fig. 1. Species accumulation curves for mean annual temperature and precipitation, sorted by major grass lineage. These data represent 1,584,351 inde- pendent collection localities spread across 10,469 taxa. Each point in the curve is a species’ mean value.

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.0909672107 Edwards and Smith

Although our data provide strong evidence that C4 origins were not correlated with movements between cooler and warmer climates at a macroecological scale, these results do not preclude hightemperatureasapotential factorindrivingtheevolutionofthe pathway. Ultimately, C4 photosynthesis is an evolutionary response to debilitating levels of photorespiration (4), so any hypotheses regarding the ecological drivers of C4 evolution must include an explanation of how that new environment translates into increased photorespiration at the leaf level. In the case of a shift from tropical moist forests to tropical open-canopy systems, we propose two alternatives. First, water limitation caused by lower rainfall could promote photorespiration directly via increased stomatal resist- ance and lowered internal leaf CO2 levels. However, it is possible that theseinitial reductionsin precipitation had little effect onplant water status, particularly if species shifted their timing of growth to coincide with the rainy season (as most C4 grasses do today). Instead, changes in precipitation may have had an indirect effect on C4 evolution by limiting canopy growth, thereby creating a high-

irradiance environment that simultaneously would raise leaf tem- peratures and provide the high light needed to drive the CO2- concentrating mechanism. Thus, high temperatures still could be an important driving force of C4 evolution at the leaf level. Dis- cerning between these two hypotheses requires more information than can be gleaned from the macroecological approach presented here. In fact, this uncertainty is a fine illustration of the limitations inherent in interpreting ecological function from geographical distribution data; what we ultimately need are more field studies comparing important eco-physiological parameters of different organisms. In this case, we currently lack critical information about the seasonal water status, gas exchange, and phenology of grass species spanning these C3/C4 focal nodes. Finally, it is clear that the C4 pathway, although a complex and

ecologically important innovation, is not an evolutionarily diffi- cult feat for grasses. Certain aspects of grass leaf anatomy and genetics probably precondition these species toward developing the syndrome whenever the right environmental conditions arise.

Pooideae

Bambusoideae

Ehrhartoideae

Panicoideae

Centothecoideae Micrairioideae

Chloridoideae

Danthonioideae

Aristidoideae

Photosynthetic pathway

C3 C4

A Pooideae

Bambusoideae

Ehrhartoideae

Panicoideae

Centothecoideae Micrairioideae

Chloridoideae

Danthonioideae

Aristidoideae

Photosynthetic pathway

C3 C4

MAT (˚C) -16 to 10 10 to 13 13 to 16 16 to 29

Pooideae

Bambusoideae

Ehrhartoideae

Panicoideae

Centothecoideae Micrairioideae

Chloridoideae

Danthonioideae

Aristidoideae

B

Fig. 2. The evolution of photosynthetic pathway and temperature niche in grasses. (A) Green lines indicate C3 photosynthesis; black lines indicate C4 photosynthesis. Maximum likelihood methods reconstructed 20 origins of C4 photosynthesis and one reversal to C3 photosyn- thesis. (B) Maximum likelihood reconstructions of mean annual temperature (MAT), using species’ mean values that were generated from 1,146,612 geo-referenced herbarium specimens.

Edwards and Smith PNAS Early Edition | 3 of 6

EV O LU

TI O N

These factors have been reviewed extensively elsewhere (25, 26) and include traits such as high leaf vein densities, enlarged bundle sheaths, and a propensity for whole-genome duplications. Diversification into cold climates, on the other hand, has hap- pened much more rarely. Thus grasses lend further support to the “tropical conservatism” hypothesis, generated by the obser- vation that relatively few lineages have diversified successfully outside of the tropics, perhaps because low temperature is one of the more difficult environmental barriers for organisms to overcome (27). The Pooideae stand out as an exceptional lineage of grasses in that they occupy both the coldest and the driest climate space in Poaceae. They should not be thought of as typical C3 grasses nor as a starting point for comparative C3/C4 physiology and ecology. Traits that promote chilling and frost tolerance in Pooideae have received attention with regard to understanding the physiology of winter crop cereals (28–30) but have not really been considered in the context of global grassland ecology. It is likely that the evolution of cold tolerance in the Pooideae has been just as relevant as C4 photosynthesis in shaping current global patterns of grass distribution.

Materials and Methods Climate Dataset. We extracted all geo-referenced herbarium specimens housed in herbaria and natural history collections that have been made available via the Global Biodiversity Information Facility (GBIF) web portal (http://www.gbif.org/). For each set of coordinates we extracted monthly temperature and precipitation values from the Climate Research Unit 10′ global gridded climate layers (31). After we purged duplicate records and outliers, our final climate dataset consisted of 1,584,351 independent col-

lection points spanning 10,469 taxa, including subspecies, varieties, and hybrids. From these collection points, we excluded all hybrid taxa as well as taxa that were represented by fewer than 10 independent localities, reducing the dataset to 1,146,612 collection points spanning 4,309 taxa. The mean number of collections per species was 932, although this value was heavily influenced by several taxa (e.g., Nardus stricta, Danthonia decum- bens, Phragmites australis, Poa pratensis) with very large numbers of col- lections. After the 5% most heavily collected taxa were removed, the mean number of collections per species for the remaining 95% was 89.

Estimating geographical ranges from herbarium specimens is prone to inherent biases in sampling; for instance, in our dataset, Europe and North America are far more heavily sampled than tropical regions (Fig. S3). How- ever, most of this imbalance results from repeated collecting of a small handful of widespread taxa that represent a very small number of tips on our phylogeny. In general, with the exception of India, the GBIF coverage of the major tropical and subtropical grassland areas (e.g., Eastern Africa, Northern Australia, Northern South America) is quite good.

Phylogenetic Tree Construction. We assembled a 1,230-taxon DNA sequence matrix that consisted of chloroplast regions atpB (59 taxa), matK (266 taxa), ndhF (437 taxa), rbcL (251 taxa), rpl16 (176 taxa), and trnL-trnF (810 taxa), and nuclear regions phyB (93 taxa) and the internal transcribed spacer (ITS; 753 taxa), using the Phylogeny Assembly with Databases (PHLAWD) tool (http:// code.google.com/p/phlawd) (32). Before matrix building, we filtered all sequences retrieved from the National Center for Biotechnology Information (NCBI) through our 4,309-taxon list to ensure a complete climate dataset across the tips of the tree. All sequence alignments were conducted using MUSCLE (v. 3.6) (33). The phylogeny was constructed using RAxML 7.1.0 (34) with all genes partitioned, allowing each gene region to have independent parameter estimates for molecular rate matrices. We employed aGTRGAMMA model ofnucleotide substitution, with a GTR substitution model and aΓ model of among-site rate heterogeneity. All matrices and trees can be obtained from

Table 2. Mean modeled selection optima for temperature and precipitation variables in Panicoideae, enforcing separate optima for C3 and C4 lineages

Variable OU attraction strength (± SD) C3 mean (± SD) C4 mean (± SD) ΔAICc*

MAT (°C) 5.94 (2.29) 21.97 (0.14) 20.12 (0.11) 5.46 Tmax (°C) 4.66 (0.68) 25.45 (0.14) 25.46 (0.16) −0.05 Tmin (°C) 6.08 (2.50) 18.11 (0.15) 14.27 (0.12) 7.91 Twettest (°C) 5.31 (0.93) 23.27 (0.10) 22.68 (0.10) 1.14 Tseasonality 6.23 (2.63) 2.60 (0.04) 4.02 (0.03) 11.65 MAP (mm year−1) 5.33 (1.00) 1781 (25.51) 1269 (16.94) 34.06 Pseasonality 5.49 (1.18) 0.55 (0.004) 0.58 (0.004) 0.95

*Larger ΔAICc numbers indicate stronger support for a two-optimum model.

Table 1. Independent contrast analyses for temperature and precipitation variables

Variable Contrast (C4 relative to C3) Number of positive contrasts

Full dataset (n = 21) MAT (°C) −0.13 10 Tmax (°C) 1.24 14 Tmin (°C) −1.71 11 Twettest (°C) 0.90 11 Tseasonality 1.07 16* MAP (mm year−1) −546** 3** Pseasonality 0.20** 15

Reduced dataset (n = 17) MAT (°C) 0.60 9 Tmax (°C) 1.30 11 Tmin (°C) −0.202 10 Twettest (°C) 1.14 9 Tseasonality 0.53 13* MAP (mm year−1) −541** 2** Pseasonality 0.21** 13*

MAP, mean annual precipitation; MAT, mean annual temperature; Pseasonality, coefficient of variation of monthly average precipitation; Tmax, average temperature of the warmest month; Tmin, average temperature of the coldest month; Tseasonality, SD of monthly average temperature; Twettest, average temperature of the wettest month; *, P < 0.05; **, P < 0.01.

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.0909672107 Edwards and Smith

the authors. We rooted Poaceae by including known close relatives Flagellaria indica, Ecdeiocolea monostachya, and Joinvillea ascendens, which sub- sequently were pruned from the tree for character analyses. Rapid bootstrap analyses were conducted on the dataset before a final maximum likelihood analysis, which used 10 random bootstrap results as starting trees. The Pan- icoideae were found to be monophyletic in the constructed phylogeny and therefore were used for further analyses. This dataset was excised from the larger dataset and resulted in 299 taxa and 6128 sites. A second phylogenetic analysis on Panicoideae was conducted using MrBayes v. 3.1.2 (35) with a GTR +I+Γ. The MrBayes analysis was conducted with two runs of four chains each run for 10,000,000 generations. Convergence and burn-in was determined by examining time-series plots of likelihood scores and parameter estimates, as well as by examining the effective sample size. Trees were used after removing 3,000,000 generations of burn-in.

Dating. Methods for creating a time-calibrated phylogeny for Poaceae are limited because of the size of the dataset. To overcome this problem, we first constructed a pruned Poaceae tree of 300 tips and ran the nonparametric rate smoothing procedure as implemented in r8s (36). We used the Powell algorithm and restarted three times to verify convergence. We extracted the estimated ages for all internal nodes in the reduced dataset and used these ages as calibrations for the full dataset. To date the remaining nodes in the full dataset, we used the nonparametric dating method PATHd8 (37). For the initial r8s run, we fixed the age of the node subtending crown Poaceae (BEP + PACMAD lineages) to 65 Mya (19). We also set minimum stem ages of 34 Mya for Stipeae and Chusquea (7) and set a minimum stem age of Dican- thelium to 8 Mya (38). Using fossil phytolith data to date the Poaceae tree is considered controversial by some authors (3, 6); however, it is relative ages, rather than absolute ages, that are important for our analyses. Furthermore, our dates of C4 origins are largely consistent with those of other recent studies (3, 6), with most occurring between 30 and 18 Mya.

To conduct comparative analyses on the rates of niche evolution and to ensure higher accuracy, we separately obtained time-calibrated phylogenies for the Panicoideae. To accommodate for phylogenetic uncertainty, we dated a sample of 100 trees from the post–burn-in posterior distribution of phy- logenetic trees from the MrBayes analysis (see above). Each tree was dated with r8s (36) using the nonparametric rate-smoothing algorithm (39) with the Powell algorithm and restarting three times to verify convergence. Because absolute ages were not necessary for our analyses, we set the root to age 1 to obtain relative ages.

Character Evolution Analyses. We used the Analysis of Traits module in Phy- locom (40) to identify 21 phylogenetically independent contrasts between C3 and C4 taxa. We also used LASRdisc (v. 1.0) (41) to reconstruct the evolution of photosynthetic shifts using maximum likelihood. Both approaches converged on the same focal nodes. Despite missing several known evolutionary tran- sitions from our dataset (e.g., Sartidia, a C3 lineage in C4 Aristidoideae;

Merxmuellera rangei, the C3 sister taxon of C4 Centropodia; and Neurachne munroi, which represents a recent C4 origin within the C3 Neurachne) (3), we have recovered more shifts in pathway than previously had been reported for grasses. Of the 21 nodes, 20 were reconstructed as a C4 origin rather than as a reversal to C3; the one exception is Homopholis proluta, which now stands as a putative reversal to C3 alongside Eragrostis walteri (42) and Alloteropsis semialata subsp. ecksonii (43) (neither sampled here). Three of our focal nodes result from the addition of taxa that have not been included in previous analyses: Homopholis proluta, Panicum amarum, and Panicum decom- positum. These taxa were represented by archived sequence data from NCBI that is not published, and there is no way to confirm that the samples were identified correctly. Another potentially misleading node is the divergence between Steinchisma laxa (C3) and Steinchisma hians (a C3/C4 intermediate). In the 21-focal node analysis we coded S. hians as C4. To account for these uncertainties, we ran analyses on a reduced dataset, removing H. proluta, P. amarum, and P. decompositum and recoding S. hians as C3. This revision resulted in 17 rather than 21 C3/C4 divergences. We tested that the magnitude of change in climate variables at focal nodes was significantly different from zero by using a one-sample t test and assessed whether the overall direc- tionality of change was significantly positive or negative against the binomial expectation, as recommended in (40).

To determine whether C3/C4 species have experienced divergent selection for temperature and precipitation variables, we tested the fit of multiple models of trait evolution using the noncensored approach in BROWNIE (v. 2.1) (44). The noncensored approach requires ancestral reconstructions for C3/C4 at internal nodes, so we used a procedure in BROWNIE that recon- structs the most likely state at each internal node before testing for model fit. We then reconstructed an OU model (45), allowing the C4 branches and the C3 branches to have optimal mean values for climate variables from which values could deviate according to the attraction parameter (OU2). To test for the significance of these results, we compared the Akaike Infor- mation Criterion scores corrected for small sample size (AICc), comparing the OU2 model described above with a model with a global mean value (OU1) and a Brownian motion model with a single rate of evolution and with no attraction parameter (BM). We made this comparison for each of the 100 dated trees drawn from the Bayesian posterior distribution and calculated the mean and standard deviation of each estimated parameter. A difference <2 between the AICc of the OU2 model and the OU1 model was taken as evidence for the OU1 model, whereas a difference >4 suggested consid- erable evidence for the OU2 model. In all cases, the difference between AICc of the OU2 model and the BM model was >200, suggesting that Brownian motion is a very poor model of trait evolution in this case.

ACKNOWLEDGMENTS. We thank D. Ackerly, J. Beaulieu, A. Bergland, C. Dunn, R. Early, and B. O'Meara for help and advice with analyses and M. Arakaki, M. Donoghue, M. Ogburn, C. Osborne, K. Schmandt, S. Schmerler, E. Spriggs, and A. Williard for helpful discussions and comments on the manuscript.

1.21.00.80.60.40.2 CV of intra-annual precipitation

tropical forest zone savanna zonesavanna zone

C3 C4

tropical forest zone

2500 2000 1500 1000 500 mean annual precipitation

Fig. 3. Maximum likelihood reconstructions of ancestral C3 and C4 precipitation niches for 21 C3/C4 evolutionary divergences. White dots indicate C3 value; black dots indicate C4 value. Background shading indicates gross climate delineations between a closed-canopy tropical forest (white) and open woodland/ savanna system (gray); transition occurs around MAP ∼1500 mm year−1, coefficient of variation of precipitation ∼0.75. Gray bars highlight C3/C4 transitions that are consistent with a movement of the C4 lineage from tropical forest into open woodland/savanna.

Edwards and Smith PNAS Early Edition | 5 of 6

EV O LU

TI O N

1. Hatch MD (1971) Photosynthesis and Photorespiration eds Hatch MD, Osmond CB, Slayter RO (Academic, San Diego), pp 139–152.

2. Muhaidat R, Sage RF, Dengler NG (2007) Diversity of Kranz anatomy and biochemistry in C4 eudicots. Am J Bot 94:362–381.

3. Christin PA, et al. (2008) Oligocene CO2 decline promoted C-4 photosynthesis in grasses. Curr Biol 18:37–43.

4. Sage RF (2004) The evolution of C-4 photosynthesis. New Phytol 161:341–370. 5. Still CJ, Berry JA, Collatz GJ, DeFries RS (2003) Global distribution of C-3 and C-4

vegetation: Carbon cycle implications. Global Biogeochem Cycles 17:1006. 6. Vicentini A, Barber JC, Aliscioni SS, Giussani LM, Kellogg EA (2008) The age of the

grasses and clusters of origins of C-4 photosynthesis. Glob Change Biol 14:2963–2977. 7. Stromberg CAE (2005) Decoupled taxonomic radiation and ecological expansion of

open-habitat grasses in the Cenozoic of North America. Proc Natl Acad Sci USA 102: 11980–11984.

8. Cerling TE, et al. (1997) Global vegetation change through the Miocene/Pliocene boundary. Nature 389:153–158.

9. Ehleringer JR, Cerling TE, Helliker BR (1997) C-4 photosynthesis, atmospheric CO2 and climate. Oecologia 112:285–299.

10. Teeri JA, Stowe LG (1976) Climatic patterns and distribution of C4 grasses in North- America. Oecologia 23:1–12.

11. Tieszen LL, Senyimba MM, Imbamba SK, Troughton JH (1979) Distribution of C3-grass and C4-grass and carbon isotope discrimination along an altitudinal and moisture gradient in Kenya. Oecologia 37:337–350.

12. Vogel JC, Fuls A, Danin A (1986) Geographical and environmental distribution of C-3 and C-4 grasses in the Sinai, Negev, and Judean deserts. Oecologia 70:258–265.

13. Hattersley PW (1983) The distribution of C3-grass and C4-grasses in Australia in relation to climate. Oecologia 57:113–128.

14. Ehleringer JR (1978) Implications of quantum yield differences on distributions of C3 and C4 grasses. Oecologia 31:255–267.

15. Collatz GJ, Berry JA, Clark JS (1998) Effects of climate and atmospheric CO2 partial pressure on the global distribution of C-4 grasses: Present, past, and future. Oecologia 114:441–454.

16. Ehleringer J, Bjorkman O (1977) Quantum yields for CO2 uptake in C-3 and C-4 plants - dependence on temperature, CO2, and O2 concentration. Plant Physiol 59:86–90.

17. Epstein HE, et al. (2002) The relative abundance of three plant functional types in temperate grasslands and shrublands of North and South America: Effects of projected climate change. J Biogeogr 29:875–888.

18. Paruelo JM, Lauenroth WK (1996) Relative abundance of plant functional types in grasslands and shrublands of North America. Ecol Appl 6:1212–1224.

19. Prasad V, Stromberg CAE, Alimohammadian H, Sahni A (2005) Dinosaur coprolites and the early evolution of grasses and grazers. Science 310:1177–1180.

20. Maddison WP (2000) Testing character correlation using pairwise comparisons on a phylogeny. J Theor Biol 202:195–204.

21. Edwards EJ, Still CJ (2008) Climate, phylogeny, and the ecological distribution of C4 grasses. Ecol Lett 11:266–276.

22. Osborne CP, Freckleton RP (2009) Ecological selection pressures for C4 photosynthesis in the grasses. Proc Roy Soc B Biol Sci 276:1753–1760.

23. Holdridge LR (1967) Life Zone Ecology (Tropical Science Center, San Jose, Costa Rica). 24. Nix HA (1983) Ecosystems of the World 13: Tropical Savannas, ed Bourliere F (Elsevier

Scientific, Amsterdam), pp 37–62.

25. Monson RK (1999) C4 Plant Biology, eds Sage RF, Monson RK (Academic, San Diego),

pp 377–410. 26. Sage RF (2001) Environmental and evolutionary preconditions for the origin and

diversification of the C-4 photosynthetic syndrome. Plant Biol 3:202–213. 27. Donoghue MJ (2008) A phylogenetic perspective on the distribution of plant

diversity. Proc Natl Acad Sci USA 105:11549–11555. 28. Badawi M, Danyluk J, Boucho B, Houde M, Sarhan F (2007) The CBF gene family in

hexaploid wheat and its relationship to the phylogenetic complexity of cereal CBFs.

Mol Genet Genomics 277:533–554. 29. Sowinski P, Rudzinska-Langwald A, Dalbiak A, Sowinska A (2001) Assimilate export

from leaves of chilling-treated seedlings of maize. The path to vein. Plant Physiol

Biochem 39:881–889. 30. del Viso F, et al. (2009) Molecular characterization of a putative sucrose:fructan 6-

fructosyltransferase (6-SFT) of the cold-resistant patagonian grass Bromus pictus

associated with fructan accumulation under low temperatures. Plant Cell Physiol 50:

489–503. 31. New M, Lister D, Hulme M, Makin I (2002) A high-resolution data set of surface

climate over global land areas. Clim Res 21:1–25. 32. Smith SA, Beaulieu JM, Donoghue MJ (2009) Mega-phylogenies for comparative

biology: An alternative to supertree and supermatrix approaches. BMC Evol Biol 9:37. 33. Edgar RC (2004) MUSCLE: Multiple sequence alignment with high accuracy and high

throughput. Nucleic Acids Res 32:1792–1797. 34. Stamatakis A (2006) RaxML-VI-HPC: Maximum likelihood-based phylogenetic analyses

with thousands of taxa and mixed models. Bioinformatics 22:2688–2690. 35. Ronquist F, Huelsenbeck JP (2003) MrBayes3: Bayesian phylogenetic inference under

mixed models. Bioinformatics 19:1572–1574. 36. Sanderson MJ (2003) R8s: Inferring absolute rates of evolution and divergence times

in the absence of a molecular clock. Bioinformatics 19:301–302. 37. Britton T, Anderson CL, Jacquet D, Lundqvist S, Bremer K (2007) Estimating

divergence times in large phylogenetic trees. Syst Biol 56:741–752. 38. Thomasson JR (1978) Observations on the characteristics of the lemma and palea of

the late Cenozoic grass Panicum elegans. Am J Bot 65:34–39. 39. Sanderson MJ (1997) A nonparametric approach to estimating divergence times in

the absence of rate constancy. Mol Biol Evol 14:1218–1231. 40. Webb CO, Ackerly DD, Kembel SW (2008) Phylocom: Software for the analysis of

phylogenetic community structure and trait evolution. Bioinformatics 24:2098–2100. 41. Jackson VK (2004) LASRDisc: Likelihood Ancestral State Reconstruction for Discrete

Characters (ed. 1.0). 42. Ellis RP (1984) Eragrostis-Walteri—a 1st record of non-Kranz leaf anatomy in the sub-

family Chloridoideae (Poaceae). S Afr J Bot 3:380–386. 43. Ibrahim DG, Burke T, Ripley BS, Osborne CP (2009) A molecular phylogeny of the genus

Alloteropsis (Panicoideae, Poaceae) suggests an evolutionary reversion from C-4 to C-3

photosynthesis. Ann Bot (Lond) 103:127–136. 44. O'Meara BC, Ane C, Sanderson MJ, Wainwright PC (2006) Testing for different rates

of continuous trait evolution using likelihood. Evolution 60:922–933. 45. Butler MA, King AA (2004) Phylogenetic comparative analysis: A modeling approach

for adaptive evolution. Am Nat 164:683–695.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.0909672107 Edwards and Smith

Supporting Information Edwards and Smith 10.1073/pnas.0909672107

−10 0 10 20 30 40

0.0 2.0

4.0 6 .0

8.0 0. 1

mean temperature of wettest month

0 10 20 30 40

0.0 2.0

4.0 6.0

8.0 0. 1

mean temperature of hottest month

−30 −20 −10 0 10 20 30 40 0.0

2.0 4.0

6. 0 8.0

0 .1 mean temperature of coldest month

0 5 10 15 20

0.0 2.0

4.0 6.0

8.0 0.1

temperature seasonality

0.0 0.5 1.0 1.5 2.0 2.5 3.0

0.0 2.0

4.0 6.0

8.0 0. 1

CV of intra−annual precipitation

Pooideae (C3)

Danthonioideae (C3)

Panicoideae (C3 + C4)

Chloridoideae (~ C4)

Aristidoideae (~ C4)

Micrairioideae (C3 + C4) Ehrhartoideae (C3)

Bambusoideae (C3)

Centothecoideae (C3)

l at

ot f o

n oi tc

ar f

l at

ot f o

n oitc

arf

l at

ot f o

n oitc

arf

l at

ot f o

n oitc

arf

l at

ot f o

n oitc

arf

Fig. S1. Species accumulation curves for climate variables, sorted by major grass lineage. These data represent 1,584,351 independent collection localities spread across 10,469 taxa. Each point in the curve is a species’ mean value.

MAP (mm) > 2000 1500-2000 1000-1500 < 1000

Pooideae

Bambusoideae

Ehrhartoideae

Panicoideae

Centothecoideae Micrairioideae

Chloridoideae

Danthonioideae

Aristidoideae

Fig. S2. Maximum likelihood reconstructions of mean annual precipitation (MAP), using species’ mean values generated from 1,146,612 geo-referenced herbarium specimens.

Edwards and Smith www.pnas.org/cgi/content/short/0909672107 1 of 4

Fig. S3. Global distribution of 1,584,351 geo-referenced grass collections, accessed via the Global Biodiversity Information Facility web portal (http://www. gbif.org/).

Edwards and Smith www.pnas.org/cgi/content/short/0909672107 2 of 4

Table S1. Node descriptions, bootstrap support, and climatic ranges for descendant species of C3/C4 nodes

Node description ML bootstrap support values

# C3 species

# C4 species

C3 range MAT

C4 range MAT

C3 range Tmax

C4 range Tmax

C3 range Tmin

C4 range Tmin

C3 range Twettes t

C4 range Twettest

Danthoniodeae/ Chloridoideae

61 71 148 4–19 8–28 10–29 16–34 −7–18 −7–25 2–22 5–29

Danth+Chloridoid/ Aristidoideae

83 219 43 4–28 12–28 10–34 21–32 −7–25 −2–24 2–29 12–29

C3 Centothecoideae/ Tristachya avenaceae

37 4 1 17–25 20–20 24–26 23–23 11–24 17–17 12–26 22–22

C3 Centothecoideae/ Loudetia simplex

39 3 1 15–18 25–25 26–27 28–28 3–7 22–22 18–20 24–24

Streptostachys asperifolia/Axonopus

51* 1 2 26–26 20–25 27–27 25–26 25–25 13–24 26–26 22–24

Ichnanthus/Axonopus +Paspalum

69* 7 28 20–26 16–25 22–27 18–28 13–25 6–23 21–26 17–26

Steinchisma laxa/ Panicum decompositum

85* 1 1 21–21 24–24 27–27 26–26 13–13 22–22 25–25 25–25

Steinchisma/ Steinchisma hians

96* 3 1 21–24 18–18 25–27 27–27 13–21 9–9 24–25 21–21

Panicum pilosum/ Panicum amarum

70* 1 1 16–16 25–25 26–26 26–26 25–25 25–25 23–23 25–25

Steinchisma- Hymanachne/ Leptocoryphium lanatum

67* 11 1 16–25 24–24 25–27 26–26 6–23 22–22 21–26 25–25

Panicum mertensii/ Panicum caricoides

96* 1 1 26–26 26–26 27–27 28–28 25–25 24–24 26–26 26–26

Panicum mertensii- caricoides/Panicum stenodes

100* 2 1 26–26 26–26 27–28 27–27 25–25 24–24 26–26 26–26

Apochloa chnoodes/ Panicum prionits clade

94* 1 4 21–21 15–23 22–22 25–27 20–20 3–18 21–21 19–25

Homolepis etc./ Cyphonanthus discrepans

0* 11 1 15–26 26–26 22–28 28–28 3–25 25–25 19–26 26–26

X = 10 Paniceae/ Andropogoneae

46* 60 103 15–26 9–27 18–28 19–31 3–25 −4–25 17–26 10–30

Homopholis proluta/ Panicum repens

100* 1 1 17–17 20–20 24–24 27–27 9–9 13–13 16–16 20–20

Panicum sp./ Echinochloa

0* 4 11 20–22 10–28 23–26 18–32 16–19 1–23 20–22 12–29

Acroceras-Ottochloa/ Alloteropsis

0* 2 2 24–24 23–24 26–26 27–27 21–22 18–20 25–26 26–26

Forest shade clade/ Digitaria-Cenchrus- Setaria

0* 47 85 10–28 7–29 18–32 15–34 −2–24 −2–24 12–29 13–29

Centothecoideae +Panicoideae/ Loudetiopsis chrysothrix

94 314 1 7–29 23–23 16–34 25–25 −4–25 20–20 10–30 24–24

Isachne/Eriachne 100 2 5 21–21 22–27 22–23 26–31 17–19 14–23 21–22 26–28

Because of high topological uncertainty within Panicoideae, this region of the tree was subjected to additional Bayesian sensitivity analyses. Bold lettering indicates the C4 component of the sister taxa. Shaded pairs were removed for the 17-node analyses because of uncertainties either in character coding (Steinchisma hians) or in taxon identification from National Center for Biotechnology Information entries (remaining three). Temperature values are °C; precipitation values are mm −year. MAT, mean annual temperature; ML, maximum likelihood; Tmax, average temperature of the warmest month; Tmin, average temperature of the coldest month; Twettest, average temperature of the wettest month. *The node is found within Panicoideae.

Edwards and Smith www.pnas.org/cgi/content/short/0909672107 3 of 4

Table S2. Node descriptions, bootstrap support, and climatic ranges for descendant species of C3/C4 nodes

Node description

ML bootstrap support value

# C3 species

# C4 species

C3 range Tseasonality

C4 range Tseasonality

C3 range MAP

C4 range MAP

C3 range Pseasonality

C4 range Pseasonality

Danthoniodeae/Chloridoideae 61 71 148 0.3–9.4 0.8–10.8 237–3540 70–2601 0.14–0.75 0.20–1.4 Danth+Chloridoid/Aristidoideae 83 219 43 0.3–10.8 0.7–9.5 70–3540 138–2293 0.14–1.42 0.19–1.5 C3 Centothecoideae/Tristachya avenaceae

37 4 1 0.7–4.7 2.3–2.3 425–2906 1140–1140 0.34–0.62 1.13–1.13

C3 Centothecoideae/Loudetia simplex

39 3 1 7.4–8.4 1.9–1.9 1187–1337 1019–1019 0.20–0.22 1.01–1.01

Streptostachys asperifolia/Axonopus

51* 1 2 0.8–0.8 0.6–4.4 2335–2335 1606–2288 0.38–0.65 0.63–0.63

Ichnanthus/Axonopus+ Paspalum

69* 7 28 0.5–4.4 0.7–7.3 1606–2823 499–3054 0.38–0.65 0.24–1.1

Steinchisma laxa/ Panicum decompositum

85* 1 1 5.2–5.2 1.6–1.6 936–936 1979–1979 0.79–0.79 0.54–0.54

Steinchisma/Steinchisma hians 96* 3 1 1.6–5.2 6.5–6.5 936–1979 1239–1239 0.46–0.79 0.31–0.31 Panicum pilosum/ Panicum amarum

70* 1 1 7.4–7.4 1.1–1.1 813–813 2496–2496 0.79–0.79 0.52–0.52

Steinchisma-Hymanachne/Leptocoryphium lanatum

67* 11 1 1.1–7.4 1.4–1.4 813–2529 1931–1931 0.23–0.95 0.61–0.61

Panicum mertensii/Panicum caricoides

96* 1 1 0.9–0.9 1.0–1.0 1183–1183 914–914 0.65–0.65 1.2–1.2

Panicum mertensii-caricoides/Panicum

stenodes

100* 2 1 0.9–1.0 1.0–1.0 914–1183 1844–1844 0.65–1.2 0.69–0.69

Apochloa chnoodes/Panicum prionits clade

94** 1 4 0.5–0.5 3.6–8.3 1061–1061 1198–1919 0.46–0.46 0.20–0.73

Homolepis etc./Cyphonanthus discrepans

0* 11 1 0.5–8.3 0.9–0.9 914–2520 1319–1319 0.20–1.2 0.42–0.42

X = 10 Paniceae/ Andropogoneae

46* 60 103 0.5–8.3 0.6–9.4 499–3054 409–2571 0.20–1.2 0.19–1.3

Homopholis proluta/Panicum repens

100* 1 1 5.6–5.6 5.1–5.1 486–486 1017–1017 0.24–0.24 0.64–0.64

Panicum sp./ Echinochloa 0* 4 11 1.8–3.8 3.0–8.9 1500–1803 429–1448 0.33–0.99 0.23–1.3 Acroceras-Ottochloa/Alloteropsis 0* 2 2 1.3–2.1 2.5–3.4 1806–2414 1051–1530 0.47–0.61 0.72–0.82 Forest shade clade/ Digitaria-Cenchrus-Setaria

0* 47 85 0.8–9.1 0.8–9.1 364–2417 258–3180 0.18–1.32 0.11–1.39

Centothecoideae+Panicoideae/ Loudetiopsis chrysothrix

94 314 1 0.5–9.4 1.8–1.8 258–3180 1296–1296 0.11–1.39 0.70–0.70

Isachne/Eriachne 100 2 5 0.8–2.1 1.9–5.5 1615–2527 491–1508 0.61–0.73 0.64–1.17

Because of high topological uncertainty within Panicoideae, this region of the tree was subjected to additional Bayesian sensitivity analyses. Bold lettering indicates the C4 component of the sister taxa. Shaded pairs were removed for the 17-node analyses because of uncertainties either in character coding (Steinchisma hians) or in taxon identification from National Center for Biotechnology Information entries (remaining three). Temperature values are in °C; precipitation values are mm −year. MAP, mean annual precipitation; ML, maximum likelihood; Pseasonality, coefficient of variation of monthly average precip- itation;Tseasonality, standard deviation of monthly average temperature. *The node is found within Panicoideae.

Table S3. Independent contrast analyses across Bayesian posterior distribution of 7,001 Panicoideae topologies

Variable Mean value of C3 node (±1 SE)

Mean value of C4 node (±1 SE)

Mean magnitude of contrast (C4 relative to C3)

# trees that were positive (P < 0.05)

# trees that were negative

(P < 0.05)

# trees that were not different from zero (P < 0.05)

# trees with significant sign

tests

MAT (°C) 22.3 (± 0.04) 21.3 (± 0.04) −1.00 0 2 6,999 29 (−) Tmax (°C) 25.4 (±0.02) 26.0 (±0.02) 0.64 120 0 6,881 36 (+) Tmin (°C) 18.9 (±0.07) 16.0 (±0.07) −2.83 0 320 6,681 33 (−) Twettest (°C) 23.1 (±0.04) 23.4 (±0.03) 0.26 0 0 7,001 0− Tseasonality 2.35 (±0.03) 3.61 (±0.03) 1.27 768 0 6,233 1,256 (+) MAP (mm year−1) 1762 (±7.3) 1361 (±4.5) −400 0 5,442 1,559 1,961 (−) Pseasonality 0.50 (±0.00) 0.62 (±0.00) 0.11 4,441 0 2,590 331 (+)

MAP, mean annual precipitation; MAT, mean annual temperature; Pseasonality, coefficient of variation of monthly average precipitation; Tmax, average temperature of the warmest month; Tmin, average temperature of the coldest month; Tseasonality, SD of monthly average temperature; Twettest, average temperature of the wettest month.

Edwards and Smith www.pnas.org/cgi/content/short/0909672107 4 of 4