|Home | About | Journals | Submit | Contact Us | Français|
Major contrasts in the climatic preferences of grass subfamilies have been noted for more than half a century (Hartley 1950), but their significance was only recognized following the discovery of C4 photosynthesis. Tropical and subtropical grasslands are dominated by the predominantly C4 Panicoideae and Chloridoideae (Hartley 1958a, b; Hartley & Slater 1960), which together account for more than half of the world's grass species (Fig. 1; Linder & Rudall 2005) and ~20% of gross terrestrial carbon fixation (Lloyd & Farquhar 1994). Their carbon-concentrating mechanism suppresses the energetically wasteful process of photorespiration that plagues C3 grasses at high temperatures, and significantly raises the efficiency of photosynthesis in warm climate regions. In contrast, C3 grass subfamilies such as the Pooideae are largely confined to temperate climates, where photorespiration is naturally limited by lower temperatures (Hartley 1961, 1973).
The evolutionary history of these patterns was elucidated only in the past two decades, following the realization that C4 photosynthesis imparts a distinctive carbon isotope signature to plant materials and trophic pathways. By analysing the isotopic composition of fossilized soils (palaeosols) and the teeth of herbivores, geochemists uncovered a surprise; the domination of low-latitude ecosystems by C4 grasses is a recent phenomenon in geological terms, occurring only 4–8 Myr ago (Ma) at the Miocene–Pliocene boundary, when C4 grasslands expanded across at least four continents (Fig. 2; Quade et al. 1989; Cerling et al. 1997). The factors behind this evolutionary phenomenon have remained controversial since its discovery.
At first, debate focused on the relative merits of CO2 as a driver of C4 grassland expansion (e.g. Cerling et al. 1994; Morgan et al. 1994a, b). The case for CO2 was persuasive, and based on the premise that selective and competitive advantages of C4 photosynthesis result from the energetic benefits of eliminating photorespiration at high temperatures and low atmospheric CO2 (Ehleringer et al. 1991, 1997; Cerling et al. 1997). These advantages over the ancestral C3 condition are reversed when photorespiration is naturally suppressed by low temperatures or high CO2, because the C4 carbon-concentrating mechanism requires energy. This leads to critical thresholds of temperature (around 20–25 °C at today's atmospheric CO2 concentration) and CO2 (around 500 p.p.m. in tropical environments) where C3 and C4 photosynthesis have equal energy requirements and, by extension, equal competitive and selective advantages (Cerling et al. 1997). The proponents of this elegant hypothesis noted the close correspondence between the theoretical temperature threshold and mean growing season value in modern regions of equal C3 and C4 grass species richness, and postulated that declining CO2 crossed an equivalent threshold at the Miocene–Pliocene boundary (Ehleringer et al. 1991, 1997; Cerling et al. 1997). Their ideas were supported by a geochemical model of atmospheric CO2, which indicated the necessary decline during the Cenozoic (Berner 1998).
By the late 1990s, the ideas underpinning the CO2 starvation hypothesis were widely accepted, but still awaited the crucial test provided by palaeo-CO2 reconstructions. This soon came with the publication of three data sets (Pagani et al. 1999; Pearson & Palmer 2000; Royer et al. 2001), each using an independent proxy for CO2, and each showing a long period of stasis in the level of atmospheric CO2 during the expansion of C4 grasslands (Fig. 3b; reviewed by Royer 2006). Unless these palaeo-CO2 proxy records are challenged on technical or theoretical grounds, the geological evidence therefore stands firmly against the CO2 starvation hypothesis for C4 grassland expansion, and new mechanisms must be sought. Instead, the latest evidence suggests that atmospheric CO2 dropped sharply through the C3–C4‘crossover threshold’ at 25–30 Ma during the Oligocene, when it initiated our modern ‘icehouse’ era of advancing and retreating polar ice sheets (Fig. 3; Pagani et al. 2005; Royer 2006). The emerging picture of palaeoenvironmental change therefore lends credence to an alternative hypothesis, proposing declining CO2 concentration as a key selection pressure for the evolutionary origins of C4 photosynthesis in the grasses, rather than C4 grassland expansion (Ehleringer et al. 1991; Pagani et al. 2005).
Here, I review two major issues currently facing geologists, physiological ecologists and ecosystem scientists with interests in this field. I first evaluate the proposed role of CO2 in the evolutionary origins of C4 photosynthesis, focusing on the question of when the pathway first arose in the grasses. However, my principal focus is on the current debate surrounding the causes of C4 grassland expansion, particularly the hypothesized effects of climate change and fire. I argue that a deeper understanding of these proposed abiotic drivers could be achieved by explicitly considering their contrasting interactions within independent groups of C4 grasses.
A crucial test of the CO2 hypothesis for C4 grass origins is to establish whether the first appearance of the pathway coincides with the Oligocene drop in atmospheric CO2. Using a molecular clock, Gaut & Doebley (1997) calculated that divergence of two major C4 grass tribes, the Paniceae and Andropogoneae (Fig. 1), occurred 25–32 Ma, leading later authors to argue that origins of C4 photosynthesis must therefore be at least this old (e.g. Kellogg 1999). However, subsequent phylogenetic analysis cast doubt on this idea, suggesting that C4 origins may have post-dated the divergence event, evolving in up to eight independent subgroups of the Paniceae and Andropogoneae (Giussani et al. 2001; Aliscioni et al. 2003; Sage 2004). Dating the nodes of the grass phylogenetic tree (Fig. 1) has proved difficult because it is becoming increasingly clear that variation in the mutation rate among branches generates large errors in molecular clocks (Pulquiéro & Nichols 2007). Attempts at calibration using the fossil record have yielded dates that vary hugely because of the low sample sizes used (Linder & Rudall 2005). More precise dating of evolutionary events within the grass family therefore depends crucially on a better resolved fossil record.
Fossils only provide minimum dates for nodes of the phylogenetic tree because the geological record is incomplete. This problem is particularly acute for grasses of open habitats, where the potential for fossilization is extremely low (Cerling 1999). The oldest C4 grass macrofossils date to only 12.5 Ma in the Middle Miocene (reviewed by Osborne & Beerling 2006). However, the pollen record for grasses overall stretches back more than 65 Myr to the Late Cretaceous (Jacobs et al. 1999), indicating the potential for much earlier C4 origins. Grass pollen cannot be identified below the family level, but phytolith traits map onto the grass phylogeny at the subfamilial scale (Piperno & Sues 2005; Prasad et al. 2005). Phytoliths form, to a varying degree, in all groups of living vascular plants (Piperno 2006). However, grasses have a higher silica production than most other plants (Hodson et al. 2005), and phytoliths in this family are more diagnostic than for most other plant groups. The interpretation of fossil phytolith assemblages is complicated by multiplicity (each species produces more than one phytolith morphotype), and redundancy (different species produce similar morphotypes) (Piperno 1988). Nevertheless, the comparison of fossil assemblages with modern reference collections has allowed robust statistical inferences about the phylogenetic affinities and ecology of extinct grass communities (Strömberg 2004, 2005).
Phytolith analysis has been used to identify grasses with affinities to the PACCMAD crown group (Fig. 1) from 65- to 67-Myr-old dinosaur dung in India (Prasad et al. 2005), and from the predominantly C4 subfamily Chloridoideae (Fig. 1) in 19-Myr-old Great Plains sediments (Strömberg 2005). Although this technique cannot establish when C4 photosynthesis originated in each clade, complementary methods have been developed for analysing the carbon isotopic signature of phytolith assemblages (Smith & White 2004). Interpretation of these data is difficult because of interspecific variation and differences in the biochemical nature of carbon compounds between C3 and C4 grass phytoliths (reviewed by Smith & White 2004). However, preliminary data suggest that up to 50% of Great Plains grasses may have used the C4 pathway by 12 Ma (Smith 2001).
Geochemical analyses of herbivore teeth, palaeosols and the molecular markers of plant cuticles trace the C4 carbon isotope signal back to 16–18 Ma in the Early Miocene (reviewed in Tipple & Pagani 2007). These data are consistent with the presence of C4 grasses comprising up to 30% of the total biomass in tropical and subtropical ecosystems throughout the Miocene (Fox & Koch 2003; Pagani et al. 2005; Tipple & Pagani 2007). However, the use of bulk isotope analyses to estimate C4 biomass in extinct plant communities is not a precise science. Critical uncertainties remain about the isotopic composition of atmospheric CO2through geological time, variation in the background signal caused by the water relationships of C3 plants and the taxonomic identity of the C4 plants (Tipple & Pagani 2007).
Partial resolution of these issues may come from a new analytical technique for measuring the carbon isotope composition of tiny samples, which currently allows C3 or C4 pathways to be identified from individual pollen grains with > 85% reliability (Nelson et al. 2007). This emerging method should enable researchers to establish directly the photosynthetic pathway of parent plants using grass pollen recovered from sediments dating to the Early Miocene and Oligocene. Unlike analyses of palaeosol or tooth carbon, this technique does not require abundant C4 plant biomass to resolve a C4 signal from the C3 background and offers the potential for identifying rare C4 plants in a predominantly C3 community, in addition to allowing identification of taxa to at least the family level. Furthermore, because it contrasts ‘pure C3’ with ‘pure C4’ samples, the technique is not compromised by variations in the isotopic signature of atmospheric CO2. Pollen-based evidence may therefore bridge the gap between postulated C4 origins and our oldest current evidence for C4 plants. However, questions will still remain over the precise taxonomic identity of these plants, and pinning down the earliest C4 grasses is only part of the challenge confronting geologists in this field. Plummeting CO2 concentrations were correlated with a whole suite of climatic changes during the Oligocene, including falling temperature and increasing aridity (Fig. 3; Zachos et al. 2001; Dupont-Nivet et al. 2007), and may not have been the only selection pressure for C4 photosynthesis.
Geological evidence therefore raises two linked questions. If CO2 was not the driving force, what caused the Miocene–Pliocene expansion of C4 grasslands? And, if C4 grasses were present from the Early Miocene onwards, why did they not dominate ecosystems earlier in their evolutionary history? The key to answering both of these questions may lie in fossil evidence that documents an abundance of woody C3 plants throughout the Miocene, forming forests, woodlands or savannas in regions that subsequently became C4 grasslands (reviewed by Osborne & Beerling 2006). Woody plant cover exerts a major limitation on C4 grass abundance, because trees and shrubs rapidly overtop herbaceous plants in the absence of disturbance, and most C4 species are intolerant of shading (Sage et al. 1999; Sage & Pearcy 2000). Explanations of C4 grassland expansion in the geological record therefore evoke a combination of climatic and disturbance factors that reduce tree cover, focusing primarily on changes in rainfall patterns and fire regime.
A number of alternative palaeoclimate hypotheses have been developed. The first is supported by the oxygen isotope signature of palaeosols in South Asia, and proposes that the Miocene replacement of C3 woody vegetation by C4 grasslands was driven by a strong increase in rainfall seasonality, caused by abrupt intensification of monsoon systems (Table 1; reviewed by Osborne & Beerling 2006). A second hypothesis based on oxygen isotope data from freshwater bivalve shells in Nepal suggests a decrease in the total amount of rainfall with no significant change in seasonality (Dettman et al. 2001), and a similar mechanism is evoked for East Africa, rooted in model simulations for the region (Table 1; Sepulchre et al. 2006). Both of these climatic drying hypotheses propose that an increased frequency and intensity of drought events killed trees and allowed the incursion of C4 grasses, to produce more open savanna or grassland vegetation (Table 1). Both are linked ultimately to tectonic events, including episodes of mountain building such as the uplift of the Tibetan Plateau and East African Rift system, or changes in ocean circulation triggered, for example, by closure of the seaway between the Americas (reviewed by Osborne & Beerling 2006).
An alternative viewpoint has emerged from analyses of palaeosol structure, which use the soil horizon structure to infer a vegetation type of desert, grassland or woodland, and the depth of the calcic horizon to estimate rainfall amounts (Retallack 2001). Application of this approach to geological sediments suggests that spatial gradients in rainfall generated parallel regional variation in vegetation (Fig. 4). Ecosystems varied from desert grassland to dry woodland throughout the Early and Middle Miocene in South Asia, the central and western United States and East Africa (Fig. 4; reviewed by Retallack 2001). Palaeobotanical evidence from East Africa also indicates savanna and woodland vegetation during the Early and Middle Miocene (Jacobs et al. 1999). However, the inference of dry climates is not supported by plant macrofossils, phytoliths and pollen from the Great Plains, which indicate productive savanna or woodland vegetation with a significant C3 grass component (Jacobs et al. 1999; Strömberg 2004, 2005). The presence of palms, gingers and bamboos (Strömberg 2004), woody dicots confined to moister climates today (Axelrod 1985), and giant tortoises and alligators (Hutchinson 1982) suggests a relatively humid, rather than arid, climate. Palaeosol data suggest a major change in the ecology of these ecosystems during the Late Miocene, with C4 grasses displacing woodland communities in mesic regions and shifting the grassland–woodland ecotone to higher rainfall areas in North America, Africa and Asia (Table 1, Fig. 4; Retallack 2001). The causes of this ecotone shift are unknown, but have been attributed to the coevolution of grasses and grazers (Retallack 2001), and significant changes in fire regime resulting from shifts in the seasonal distribution, but not total amount, of rainfall (Keeley & Rundel 2005).
Evidence supporting a linkage between fire and the expansion of C4 grasslands comes from black carbon (BC) abundance in marine sediments, a geological proxy for fire occurrence. BC increases by 100- to 1000-fold in Pacific Ocean localities downwind of the Indian subcontinent (Herring 1985; Keeley & Rundel 2003), and 5-fold in the South China Sea (Jia et al. 2003), during the Miocene–Pliocene interval of major C4 grassland expansion in these regions. Sediments off the Atlantic coast of West Africa also show significant increases in charred grass cuticle and pollen abundance during the same period (Morley & Richards 1993). In the last 5 years, a new and compelling hypothesis has therefore added ecosystem-scale feedbacks between fire and vegetation to previous ideas about palaeoclimate (Table 1; Keeley & Rundel 2003, 2005).
Fire sustains C4 grasslands by killing woody plants and entrains a positive feedback because the dead foliage of grasses provides abundant fuel for fires, thereby increasing tree mortality and promoting the further spread of grasses (Keeley & Rundel 2003, 2005). The fire hypothesis therefore proposes that increasing climatic seasonality during the Late Miocene raised fire frequency by supporting rapid biomass production and the development of a high fuel load during wet summer conditions. Intensifying winter drought promoted drying, and increased the likelihood that this material would ignite, with ignition itself provided by lightening strikes at the end of the dry season. Further feedbacks on the fire regime have been proposed via the hydrological cycle (Beerling & Osborne 2006) and wind strength (Tipple & Pagani 2007), but these remain speculative at present.
Ecosystem fire regimes show an important interaction with climate. Fuel remains too moist to support frequent fires in wet, aseasonal environments, whilst the low productivity of dry climate regions produces too little fuel to carry significant fires (Keeley & Rundel 2005). However, given sufficient periods of dry weather, fire has the potential to displace mesic forests in favour of grasslands. This is demonstrated in modern ecosystems by complementary data from: model simulations showing that fire reduces forest cover by ~50% at the global scale (Bond et al. 2005); analysis of woody plant cover across Africa indicating that significant areas of savanna are maintained by disturbance in regions with sufficient rainfall to support forest (Sankaran et al. 2005); fire exclusion experiments which allow the establishment of fire-sensitive trees in mesic savanna ecosystems (Bond et al. 2003, 2005); and the fire-mediated replacement of native forests on Pacific Islands by invasive grasses (D’Antonio & Vitousek 1992). These observations each provide indirect support for the hypothesized mechanism of vegetation change at the Miocene–Pliocene boundary.
An emerging consensus therefore evokes interactions between palaeoclimate change, vegetation–climate relationships and fire frequency to explain the Miocene expansion of C4 grasslands (Table 1; Keeley & Rundel 2005). This ‘consensus scenario’ suggests that Early and Middle Miocene landscapes were dominated by C3 forests or woodlands, with C4 grasses occupying open ground between the patches of woody plants. Increasing seasonality during the Late Miocene concentrated rainfall into a hot growing season, creating a fire regime that removed woody vegetation and shifted the woodland–grassland ecotone to wetter areas (Table 1). Formulated in this way, the hypothesis asserts that climates capable of supporting frequently burning mesic C4 grasslands at the expense of woodland and forests were absent during the Early and Middle Miocene.
However, application of the hypothesis to North and South America is problematic in two important respects. First, patterns of climatic change in these regions at the Miocene–Pliocene boundary are less clear than those in Asia, and an increase in rainfall seasonality has yet to be demonstrated (Passey et al. 2002; Fox & Koch 2004; Osborne & Beerling 2006). Secondly, pollen records, phytolith assemblages and mammalian tooth morphology suggest that a transition from closed forest to open woody vegetation with a significant C3 grass component may have occurred as early as 25 Ma (Late Oligocene to Early Miocene) in the Great Plains (Strömberg 2004, 2005) and 32 Ma (Early Oligocene) in South America (Jacobs et al. 1999). The adaptive radiation of specialist grazers (Strömberg 2006) and an extremely high browser diversity in these Great Plains ecosystems resulted in species-rich mammalian herbivore communities by 15 Ma (Janis et al. 2000). Carbon isotope data demonstrate that some grazers showed a dietary shift in the Late Miocene, subsisting almost exclusively on C4 plants for the 2 Myr before C4 grassland expansion, with presumably significant, but currently unknown, effects on grass–tree dynamics (Fox & Koch 2004; Beerling & Osborne 2006).
I argue that our overall understanding and ability to test climatic and ecological hypotheses regarding Miocene ecosystems would be improved by an explicit consideration of C4 grass phylogeny. To make this case, I first present biogeographical data indicating significant contrasts in the abundance of different C4 grass clades along rainfall gradients, and show how this information might be used in hypothesis testing. Secondly, I re-analyse a published data set in the same context, suggesting that changes in the species richness of independent C4 grass lineages track the length of the rainy season. Finally, I present intriguing preliminary evidence of fire controls on the abundance of particular C4 clades.
The subfamilies Panicoideae and Chloridoideae together account for the majority of modern C4 grass species (Fig. 1; Linder & Rudall 2005), but analyses of regional biodiversity patterns suggest a crucial difference in the climate relationships of these major clades. For the United States and Argentina, the percentage of C4 species belonging to the Panicoideae shows a positive correlation with annual rainfall, whereas the same relationship for the Chloridoideae is negative (Fig. 5; Taub 2000; Cabido et al. 2007).
The mechanisms underpinning these relationships are unclear at present, but may be linked to the close association between biochemical subtypes of C4 photosynthesis and specific phylogenetic groups; the majority of Panicoideae species utilize the NADP-malic enzyme (NADP-ME) pathway whereas most members of the Chloridoideae use the NAD-malic enzyme (NAD-ME) pathway (Taub 2000). As a consequence, contrasting correlations with rainfall are also recognized for different C4 subtypes (Fig. 5b,c; e.g. Hattersley 1992; Taub 2000; Cabido et al. 2007). Experimental investigations suggest that these alternative forms of the pathway may be differentially adaptive in relation to water availability (Ghannoum et al. 2002), or a correlated environmental variable such as soil nutrient status (Ghannoum et al. 2005). However, none of these experiments have explicitly controlled for phylogeny, and the issue remains unresolved. Further questions remain about how these species richness patterns translate into plant abundance along rainfall gradients.
Despite these uncertainties, the geographical distributions of modern C4 grasses indicate that changing rainfall patterns will drive significant shifts in the phylogenetic composition of C4 grass communities. This observation has important implications for palaeoclimate hypotheses about the Miocene–Pliocene expansion of C4 grasslands, because it allows predictions about the clades of C4 grasses that are involved. First, the range of Middle Miocene ecosystems along inferred rainfall gradients (Retallack 2001) should contain varying proportions of C4 grass subfamilies, following the qualitative patterns in Fig. 5(a). Secondly, a shift in the woodland–grassland ecotone to higher rainfall areas should increase the proportion of C4 species belonging to the Panicoideae relative to the Chloridoideae. Finally, a decrease in the total amount of rainfall should drive the reverse pattern, favouring species of the Chloridoideae relative to the Panicoideae.
These opposing predictions are especially critical for regions such as East Africa and South Asia, where alternative hypotheses are postulated. Crucially, they can be tested by using emerging techniques for quantifying the relative abundance of different grass subfamilies within fossil phytolith assemblages (e.g. Prasad et al. 2005), providing a new means of evaluating alternative hypotheses using geological evidence. Conversely, the use of these techniques to reconstruct the phylogenetic make-up of Miocene grassland communities in the Great Plains could offer vital clues about the environmental drivers of C4 grassland expansion in this region. Finally, quantifying shifts in the abundance of different grass subfamilies could aid in the interpretation of carbon isotope signals. For example, a decrease in the proportion of Chloridoideae (NAD-ME) relative to Panicoideae (NADP-ME) species would be expected to ‘amplify’ the C4 signal, because discrimination against 13C is stronger in NAD-ME than NADP-ME grasses (Hattersley 1982), i.e. the NAD-ME type is slightly more ‘C3-like’. The power of these approaches would be increased greatly by a better understanding of how C4 grass distributions vary in relation to rainfall seasonality and fire frequency.
Previous studies suggest that species richness of the tribes Andropogoneae and Paniceae (Fig. 1; Panicoideae) may change significantly in response to the intensity and seasonality of rainfall. For southern Africa, Gibbs-Russell (1988) used a qualitative comparison of floristic and climatic data to show that Andropogoneae species account for the largest fraction of the Panicoideae in monsoonal summer rainfall areas where mean annual precipitation (MAP) > 500 mm. The opposite pattern occurs in the Paniceae, which account for the largest proportion of Panicoideae species in more arid summer-rainfall regions (MAP < 500 mm).
Hartley (1950, 1958a, b) carried out a similar analysis at the global scale, but expressed species numbers in each tribe as a fraction of the total grass flora, rather than restricting the analysis to the subfamily Panicoideae. Based on qualitative descriptions of maps, Hartley hypothesized that centres of the highest species diversity in the Andropogoneae are typically located in subtropical monsoon climates with a short rainy season (Hartley 1958a), whereas the highest diversity in the Paniceae occurs in less seasonal wet equatorial climates (Hartley 1958b). To evaluate this hypothesis further, I have carried out a quantitative analysis of the same species richness data sets (Hartley 1950, 1958a) by using climate data recorded at nearby meteorological stations (Müller 1982). A series of generalized linear models (GLIMs) were fitted to the data, using a quasibinomial distribution in place of the binomial distribution when data were underdispersed [glm(), R version 2.4.1, The R Foundation for Statistical Computing]. I used a forward stepwise approach, testing first the simple additive hypotheses posed in the literature, and then adding more variables and interactions.
Numerous previous studies have demonstrated that temperature is the primary global control on species richness for C4 grasses (reviewed by Sage et al. 1999), and the Andropogoneae and Paniceae tribes followed this general pattern (Table 2a,b). GLIMs for Eurasia–Africa and the Americas showed highly significant additive effects of mean annual temperature (MAT) and tribe on relative species richness (Fig. 6a,b, Table 2a,b). Relative diversity was lower in the Andropogoneae than the Paniceae (Table 2a,b), and this difference was more pronounced in the Americas than Eurasia and Africa (Fig. 6a,b).
I then added rainy season length (RSL) to the model, restricting the analysis to tropical and subtropical localities (latitude < 35°N or S). Because the latitude criterion excluded most of the data from the Americas, this analysis focused on Eurasia and Africa. RSL was estimated for each locality using a biologically based definition of the growing season for C4 grasses; as the number of consecutive months in which the mean minimum temperature > 10 °C and total precipitation > 30 mm (Collatz et al. 1998). The temperature threshold for growth was based upon two lines of complementary evidence: (i) experimental observations of chilling-mediated photoinhibition and impairment of leaf extension in C4 grasses (Long 1983); and (ii) the minimum summer temperature required for C4 species to persist in a grass flora (Sage et al. 1999), a threshold that was observed in this data set. These criteria defined a reasonable (non-zero) rainy season for all localities except those in desert regions (e.g. Syria and Yemen), where plants are probably associated with bodies of water or infrequent rains, or at high elevation (e.g. Lesotho), where C4 species may show an unusual resistance to low temperature extremes (Márquez et al. 2007). These localities with a zero RSL were excluded from further analysis.
The minimum adequate GLIM showed highly significant additive effects of MAT, RSL and tribe, but not the hypothesized interaction between MAT and RSL (Table 2c). Species richness increased with RSL in both tribes, but showed a differing nonlinear response: values for the Andropogoneae were low when 0 < RSL ≤ 3 months, and high when 4 ≤ RSL ≤ 12 months (Fig. 6c); in contrast, values for the Paniceae increased in an approximately linear response to reach a maximum at RSL ≥ 9 months (Fig. 6d). I therefore constructed a new GLIM attempting to account for these apparent nonlinearities by grouping RSL data into three temporal categories: 1–3, 4–7 and 9–12 months (there were no localities with RSL = 8 months). This model showed an additive effect of MAT, and an interaction between RSL and tribe (Table 2d). The interaction was caused by equal species richness in climates with a rainy season of 7 months or less, and greater species richness in the Panicoideae than Andropogoneae in aseasonal climates with rainy seasons of 9 months or more (Fig. 7).
This quantitative analysis supports the hypothesis that species richness in the Paniceae is highest in moist, aseasonal climates of the subtropics and tropics, but fails to show a similar association between maximal Andropogoneae diversity and a short rainy season. Instead, species richness in the Andropogoneae remains constant across localities with rainy seasons of 4–12 months (Figs 6c and and7).7). The analysis therefore highlights a phylogenetic contrast between the tribes that could prove useful for hypothesis testing in the geological record: a climatic change from moist, aseasonal conditions to a strongly seasonal climate is expected to drive a shift in the grass flora from a predominance of Paniceae species towards a more equal representation of the Paniceae and Andropogoneae. Further work is required before this expectation can be tested in the fossil record, because techniques for analysing phytolith assemblages cannot yet distinguish tribes of the Panicoideae. However, archaeologists already use phytolith-based diagnostics for identifying individual crop species of the Andropogoneae (Piperno 2006), indicating good promise for increasing the taxonomic resolution of this technique.
The value of these observed patterns of species richness would be increased greatly by an understanding of the underlying mechanisms. One possibility is suggested by Bond et al. (2003), who hypothesize that the Andropogoneae are key to the fire-mediated displacement of forests by grasslands in mesic climates of southern Africa. These authors note the dominance of this group in fire-maintained and nutrient-poor mesic savannas, the obligate dependence of Andropogoneae species such as Themeda triandra on frequent defoliation, and their decline in the absence of fires. Frequent fires are promoted by the rapid growth of Andropogoneae species during summer, the accumulation of tannin-like compounds in their foliage, and the low nutritional quality of leaves during winter (reviewed by Bond et al. 2003). These latter traits retard decomposition and result in a low palatability to grazers during the dry season, leading to the build up of a high fuel load for fires.
The evolution of such fire-promoting traits in one or more clades of C4 grasses during the Miocene therefore provides a possible explanation for how the grassland–woodland ecotone shifted to more mesic areas. But what might have selected for these characters? One idea is suggested by the Miocene radiation of grazers, and the specialization of some animals on a C4 diet prior to grassland expansion. Recent studies indicate that modern grasses have evolved a suite of characteristics to resist grazing, including the accumulation of phenolic compounds (e.g. tannins), a high silica content, and tough fibrous leaves with a low palatability and nutritional quality (Burt-Smith et al. 2003; Massey et al. 2007). Ecological theory proposes that selection for these traits is strongest in nutrient-poor habitats (Grime 2001), and empirical evidence shows that grazer-driven selection for unpalatable species occurs in dry ecosystems with a long history of herbivory (Díaz et al. 2007). Speculative links may therefore be drawn between high grazing pressure in nutrient-poor savannas during the Miocene, and selection for grazing-resistant (but fire-promoting) traits in C4 grasses. As these traits may also occur in C3 grasses (Grigulis et al. 2005), this hypothesis provides an ecological explanation for Miocene C4 grassland expansion that is not directly linked to photosynthetic pathway.
The origin of C4 photosynthesis and expansion of C4 grasslands were major events in Earth history with significant consequences for tropical and subtropical ecology. Newly uncovered geological evidence has bolstered support for decreasing CO2 as a selection agent for the pathway, leading to the expectation that C4 grasses first evolved during the Oligocene. However, these same data cast serious doubt on the role of CO2 in C4 grassland expansion. Instead, current ecological understanding of vegetation–fire dynamics in savannas has generated hypotheses integrating the roles of climate change and fire. I propose that the contrasting modern diversity patterns shown by independent C4 grass clades along climatic gradients generate a further expectation: that palaeoclimate change forced significant shifts in the phylogenetic composition of Miocene grass communities. New techniques for the geochemical analysis of pollen and the extraction of phylogenetic information from phytolith assemblages provide the necessary tools for testing these ideas, and will help to bring the evolutionary history of C4 grasses into sharper focus.
I thank The Royal Society for funding through a University Research Fellowship, and Andrew Beckerman, David Beerling, Jeremy Midgley, Gareth Phoenix, Brad Ripley, Caroline Strömberg and Vernon Visser for stimulating discussions during the preparation of this work, and critical comments on the manuscript.