Temperature is considered an important determinant of biodiversity distribution patterns. Grasses (Poaceae) occupy among the warmest and coldest environments on earth but the role of cold tolerance evolution in generating this distribution is understudied.
We studied cold tolerance of Danthonioideae (c. 280 species), a major constituent of the austral temperate grass flora. We determined differences in cold tolerance among species from different continents grown in a common winter garden and assessed the relationship between measured cold tolerance and that predicted by species ranges. We then used temperatures in current ranges and a phylogeny of 81% of the species to study the timing and mode of cold tolerance evolution across the subfamily.
Species ranges generally underestimate cold tolerance but are still a meaningful representation of differences in cold tolerance among species. We infer cold tolerance evolution to have commenced at the onset of danthonioid diversification, subsequently increasing in both pace and extent in certain lineages. Interspecific variation in cold tolerance is better accounted for by spatial than phylogenetic distance.
Contrary to expectations, temperature – low temperature in particular – appears not to limit the distribution of this temperate clade. Competition, time or dispersal limitation could explain its relative absence from northern temperate regions.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Climate, and in particular temperature, has long been considered a principal determinant of biotic distribution patterns (Humboldt & Bonpland, 1807; Merriam, 1894). For example, temperature is one factor frequently put forward to explain one of the most striking distribution patterns on earth: the paucity of lineages in colder latitudes compared with warmer latitudes (Hillebrand, 2004; Wiens & Donoghue, 2004). Because many lineages restricted to the tropics are both old and geographically widespread, it has been suggested that neither lack of time nor dispersal ability can explain this pattern (Wiens & Donoghue, 2004; Fine & Ree, 2006; Mittelbach et al., 2007). Instead, it has been attributed to the assumed difficulty with which organisms evolve the necessary physiology to survive low temperatures, and freezing in particular (Sakai & Larcher, 1987; Latham & Ricklefs, 2008; Wiens & Donoghue, 2004; Donoghue, 2008).
Grasses (Poaceae) have a near global distribution, with well over 10 000 described species (Clayton et al., 2006 (onwards)), dominating several ecosystems from savannahs to tundras and covering over 30% of earth's terrestrial surface (31–43%; Gibson, 2009). Despite this richness and distribution, and although some grasses are among the most cold tolerant of plants (Bannister, 2007), only a small proportion of lineages have made the transition into the temperate zone since their origin in the warm, wet forests of Gondwana (Kellogg, 2001; Bouchenak-Khelladi et al., 2010). These are the Pooideae, to which many economically important crop species belong, the Danthonioideae, to which the horticulturally famous pampas grasses belong, the temperate bamboos and some species of Chloridoideae. Of these, the Pooideae, the dominant Northern Hemisphere temperate grasses, and their southern counterpart, Danthonioideae, stand out as occupying the coldest habitats overall (Edwards & Smith, 2010).
Cold tolerance in grasses has primarily been studied from the perspective of the genetic mechanisms involved in frost tolerance of cereals and forage grasses of northern origin (in Pooideae; Antikainen & Griffith, 1997; Sandve et al., 2011). Pooideae have evolved unique gene families to cope with cold stress (Sidebottom et al., 2000; Sandve et al., 2008; Sandve & Fjellheim, 2010), avoiding damage by ice by preventing ice formation in the cells (Sidebottom et al., 2000; Zhang et al., 2010). By contrast, the only two species of Danthonioideae that have been studied revealed a different strategy, involving control of the site of ice formation (restricted to the leaf surfaces) rather than whether it happens at all (Wharton et al., 2010). These findings, along with the fact that the tropical lineages in which each of these two clades is embedded diverged from each other before recruitment of novel gene families in Pooideae (GPWG, 2001), indicate independent evolutionary routes into freezing habitats in these lineages. However, the species within each of these lineages occupy a range of habitats, spanning a range of temperature regimes, and the details of the evolution of cold tolerance and its potential role in determining global distribution patterns remain poorly understood.
Southern Hemisphere plants are generally considered less cold tolerant than their northern equivalents, being spared the extreme temperatures endured by northern plants thanks to the buffering afforded by the expansive southern oceans (Bannister, 2007). It is possible that the relative absence of Danthonioideae in the Northern Hemisphere is related to the strategy of frost tolerance employed, rendering them less cold tolerant than the northern pooids. Here we tested this idea by combining common garden experimentation and comparative analyses in a phylogenetic framework to investigate to what extent the distribution of the 281 species (Linder et al., 2010) of C3 grasses in Danthonioideae could be limited by (low) temperature. We tested, for 10 species from several continents, whether current ranges reflect inherent cold tolerances. We then analysed, across the entire clade, when cold tolerance is likely to have evolved and what the mode of evolution over time and among lineages has been, to infer whether the range of Danthonioideae is at equilibrium, with respect to cold tolerance.
Materials and Methods
Plant material used in the common garden plots
Seeds for as many species as possible were collected in the field in Australia and New Zealand during the 2005/2006 austral summer or obtained from the United States Department of Agriculture (USDA) or Silvehill Seeds, Cape Town, South Africa (Table 1). One hundred seeds per species were sown in spring 2006 in equal parts of topsoil, compost and peat, with a small amount of river and pumice sand. Field-collected material was collected from multiple individuals from across a single population for each species. Plants were cultivated in a glasshouse for 18 months, by which time at least one flowering had taken place, and then re-potted before the commencement of the common garden experiment in October 2007. Soil quality and root development are not considered limiting. Species from several continents and temperate climate regions (ranging from the lowlands of southeastern Australia and North America to the highlands of New Zealand and southern Africa), and for which several individual adult plants were available, were selected for the common garden experiment.
Table 1. Species and numbers of individuals used in the common winter garden experiment
‘a’, those plants that had already spent two winters in the plots (hardened); ‘b’, plants for which this was their first winter outside (nonhardened). USDA, United States Department of Agriculture.
USDA, number W619122
Field. Voucher Humphreys 23
Field. Voucher Humphreys 120
Field. Voucher Humphreys 51
Cult. Voucher Humphreys 152
Field. Voucher Pirie 308
USDA, number PI202162
Field. Voucher Humphreys 145
Silverhill Seeds, South Africa
Silverhill Seeds, South Africa
Number of plants per plot
The common (winter) garden experiment
The common garden experiment was carried out on 10 species in total (Table 1) and at two plots that differ in aspect and therefore temperature regime in the Botanic Gardens in Zurich, during three consecutive winters. However, because of limited availability or death of plants during the first or second winter, not all species were present in both plots in all years. We therefore treat each year and plot as a separate experiment.
Winter 1 (2007/2008)
In October 2007, 10 individuals for each of eight species (Table 1) were selected at random and placed outside in a sheltered spot to acclimatize to cooler but frost-free conditions, before being transferred to the experimental plot 1 month later. Plants were placed in rows, with one individual per species per row, randomized within each row to remove a systematic effect from neighbouring plants or microclimatic variation across the plot. Plants were placed in the ground in their pots, arranged so that each pot came into contact with the neighbouring pot. A temperature logger (HOBO U23-003 Pro v2; Tempcon Instrumentation, Arundel, UK) was placed with one reader at ground level and the other c. 1 m above-ground to capture temperature variation caused by wind and snow cover. Plants were removed from the experimental site after the last spring frost (spring 2008) and placed outside the glasshouse. One month later, frost tolerance was scored as the proportion of individuals per species showing new growth (Bannister, 2007). Species with ≥ 50% surviving individuals were considered tolerant of winter conditions.
Winter 2 (2008/2009)
All plants survived the first winter and were therefore included in a repeat experiment the following winter. To increase the temperature range to which plants were exposed, 10 additional plants per species (where possible) plus seven plants for one additional species (Table 1) were selected from the plants in the glasshouses and placed in a second plot (Table 1), having first been placed outside for 1 month to acclimatize to ambient conditions. Temperature conditions and survival rates were recorded as before.
Winter 3 (2009/2010)
All plants that survived the previous winter were placed back in their respective plots for a third and final round of the experiment in autumn 2009. These plants are referred to as ‘hardened’ as they had spent two (plot 1) or one (plot 2) winter in the experimental plots already. In addition, to gain more understanding of the differences in cold tolerance within and among species, 10 ‘nonhardened’ plants for each of five species were placed in each of the two plots (Table 1). These species were again selected based on availability of plant material and spanned three geographical regions. Temperature conditions and survival rates were recorded as before.
To provide a measure of cold tolerance, species were ranked according to their survival rates in each plot and year. This is appropriate for field experiments where determination of absolute tolerances may be complicated by leaf temperatures falling below air temperatures as a consequence of thermal radiation during clear-skied nights (Körner, 2003; Bannister, 2007). Furthermore, we did not consider possible moisture differences between the two plots, nor measure precipitation or soil temperature. We assume any moisture differences experienced in the two common garden plots to have been smaller than those among the species' native ranges and do not consider lack of soil temperature information limiting, as the equivalent information will not be carried in the BioClim variables (see later), to which we compare the experimental results. Finally, we note that our measurement of cold tolerance is based on survival of adult plants, not on seedling establishment or survival. While it is clear that all life stages must be able to survive to maintain a viable population, most seed germination will not occur in winter (A. M. Humphreys, pers. obs.), meaning that adult survival is a vital component of frost tolerance of these species.
Temperature conditions in the observed ranges
To estimate the realized temperature niches of the species in the common garden experiment, georeferenced occurrence data were obtained from herbarium records (Southern Hemisphere) and from the Global Biodiversity Information Facility (GBIF, www.gbif.org; for North American records; see R. O. Wüest et al., unpublished; Linder et al., 2013). Temperature data for these localities were extracted from the WorldClim database (BIO1–BIO11; Hijmans et al., 2005) using ArcGIS 9.2 (ESRI, 2008). Correlation between the mean, minimum (95%) and maximum (95%) of each variable (Supporting Information Table S1) was assessed for each species with Kendall's tau. All uncorrelated temperature values were used to represent temperature conditions in the observed range of each species (17 parameters; Table S2). The main axes of variation among the parameters were summarized with a phylogenetic principal component analysis (PPCA) so that expected covariances given phylogeny were taken into account (Felsenstein, 1985). The ‘vcv’ function in the R (R Development Core Team, 2011) package ape (Paradis et al., 2004) was used to generate a variance-covariance matrix among the species and the ‘phyl_pca’ function of Revell (2009) was used to carry out the PPCA. Phylogenetic information was taken from the rate smoothed maximum clade credibility (MCC) tree of Antonelli et al. (2011), pruned to include only the experimental species.
Comparing measured (fundamental) and modelled (realized) cold tolerance
Because field measurements cannot provide reliable estimates of absolute temperature tolerances (Bannister, 2007), we investigated the relationship between relative survival rates (percentage survival) and temperature niches estimated from species' observed ranges. To do this, while correcting for possible autocorrelation among the species resulting from phylogenetic relatedness, we carried out phylogenetic generalized least squares (PGLS) regressions between survival rates and each of the first two principal component (PC) axes using the ‘pgls’ function of the R package caper (Orme et al., 2011). The phylogenetic signal in the data and its potential influence on model fit were assessed by repeating regressions: allowing the phylogenetic branch length transformation parameter λ (Pagel, 1999) to be estimated as part of the model fitting procedure; fixing λ = 1, which implies that the trait varies perfectly with the phylogeny, equivalent here to Brownian motion (BM; Schluter et al., 1997); or fixing λ = 0, which implies that there is no phylogenetic signal in the trait data. The best-fitting models based on their Akaike Information Criterion, corrected for small sample sizes, (AICc) scores were used to infer the slope and intercept of the regressions. Regressions were carried out separately for each plot and year.
Timing of cold tolerance evolution
To analyse the evolution of cold tolerance across the whole subfamily, we extracted temperature-related BioClim data for all available danthonioid occurrence records, using 20 800 georeferenced herbarium records for the Southern Hemisphere, representing all major austral herbaria, and from 27 000 GBIF entries, representing North American and European distributions (www.gbif.org; see R. O. Wüest et al., unpublished). Data representing possible inaccurate identification and georeferencing (e.g. data points in oceans or weedy occurrences) and species not included in the phylogeny were removed to yield a data set of 41 607 georeferenced occurrence points (average 186 per species). These data provide a good representation of the clade's true distribution (see Linder et al., 2013).
First we confirmed that the variation among temperature variables for the experimental species reflects variation across Danthonioideae. We repeated the PPCA across all mean, minimum and maximum values for BIO1–11 and on the 17 independent parameters used previously, for all species for which both phylogenetic and georeferenced occurrence data were available. The chronogram of Antonelli et al. (2011) was pruned to include only those species for which climate data were available, retaining multiple accessions for species whose position differed significantly between nuclear and plastid trees, using the taxon duplication technique of Pirie et al. (2008, 2009). Thus, a set of trees with 253 accessions representing 224 species was generated as in Humphreys et al. (2011).
Climate data for four species of the sister clade Chloridoideae were included to assess when during the evolutionary history of Danthonioideae cold tolerance is likely to have evolved. Two species of Centropodia (Chloridoideae) already form outgroup taxa in the phylogeny, and climate data based on the occurrence of Centropodia glauca and Centropodia forskalii, from temperate Africa and Asia, respectively, were included for these taxa. The other two outgroup taxa in the phylogeny were used as placeholders for two more chloridoid ‘summary taxa’, based on all available GBIF occurrence data for the widespread genus Zoysia. BioClim temperature variables (BIO1–BIO11) were extracted for all occurrence records of Zoysia, representing 10 species, and split into two groups depending on their distribution: a tropical (four species from Asia and the Pacific) and a temperate (six species from Australia, New Zealand and Japan) group. For each BioClim parameter, mean values were then used together with those for Centropodia to represent the realized temperature niche of Chloridoideae. This is sparse representation of a widespread group but, taken together with the earlier demonstration that the chloridoids occupy an overall warmer niche than Danthonioideae (Edwards & Smith, 2010), we consider our inferences based thereon robust.
The trend in change over time in the cold tolerance axis (PC1), indicative of any departure from BM (Freckleton & Harvey, 2006), was visualized by plotting absolute values of phylogenetically independent contrasts (Felsenstein, 1985) at each node against median node age. Contrasts were extracted using the ‘pic’ function in the R package ape. Then, change in PC1 was modelled with BM and with two derivations of BM that include a branch length transformation parameter (lambda (λ) or delta (δ); Pagel, 1999) to allow for departure from the linear accumulation of trait variance with time, assumed under BM. Maximum likelihood (ML) estimates of λ and δ allow assessment of the phylo-genetic signal in the data and whether most trait change has occurred deeper in the tree or towards the tips. BM, λ and δ models were fitted using the ‘transformPhylo.ML’ functions in the R package motmot (Thomas & Freckleton, 2011) on the MCC tree and, to assess robustness of results to phylogenetic uncertainty, across 1000 trees randomly sampled from the posterior distribution of trees.
Ancestral states were estimated under different models of trait change using the ‘ace’ function in ape. To estimate ancestral states under the λ or δ models, the MCC tree was first transformed with the ML estimates of these two parameters as implemented in Geiger (Harmon et al., 2008). Models were fitted with restricted maximum likelihood (REML) to reduce bias in ancestral state estimation (Paradis et al., 2004).
Lineage-specific differences in cold tolerance evolution
Under BM, rates of change are assumed to be the same along all branches. Under the λ and δ models, rates of change may vary over time but are assumed to remain constant among lineages. To gain a more complete picture of how cold tolerance has evolved in Danthonioideae, we explored the possibility of departure from this assumption in several ways. First, change in the cold tolerance axis (PC1) was inferred for each of the two major sister clades separately (referred to as clade 1 and clade 2; see Fig. 4a), comparing model fit and parameter estimates in motmot as before. Clade 1 comprises the South African Cape genus Pentameris (95 tips) and clade 2 comprises several genera distributed on all continents (169 tips). Ancestral state reconstructions (see previous section) suggest possible differences in cold tolerance evolution between these two lineages.
Next, we tested for lineage-specific differences in cold tolerance evolution more generally by fitting a derivative of BM that looks for significant shifts in the rate of change of a continuous trait (the differential rates analysis of Thomas & Freckleton, 2011). We allowed up to 10 rate shifts, within or along stem branches leading to clades of at least 10 species, retaining only shifts supported by an AIC value ≥ 5. This analysis was carried out using the ‘transformPhylo.ML’ function and the MCC tree as described in the previous section.
Having identified significant rate differences among different lineages, we asked whether different rates operate in lineages tending towards different temperature optima. We delimited clades in which to test for different optima based on the positions of the rate shifts identified. These were Danthonia, the Pentameris lima clade, ‘core’ Pentameris, Rytidosperma, core Tribolium and ‘core’ Danthonioideae. Chionochloa, a genus from New Zealand, contains species occupying some of the coldest danthonioid habitats, but no exceptional rates of cold tolerance evolution were inferred for this clade. It was delimited as a separate clade to allow inference of the temperature regime occupied by this genus, independently of the rest of the subfamily. We tested whether Danthonioideae as a whole is evolving towards a single optimum, described as an Ornstein–Uhlenbeck (OU) process in which a trait changes with a constant pull towards an optimum value (Hansen, 1997; Butler & King, 2004). Then we compared the fit of this model to one where the mean optimum is allowed to differ among groups, while also allowing the rate of change to vary among groups (Beaulieu et al., 2012), as implemented in the R package OUwie (Beaulieu & O'Meara, 2012). OUwie requires a tree with node labels denoting along which branches the different optima may be acting. We generated such a tree by estimating ‘ancestral states’ of clade membership under ML and the equal rates (ER) model in ape, labelling internal nodes with the inferred ‘ancestral states’. Inspection of the labelled tree verified this approach. Model fit was assessed with AICc.
Separating the effects of phylogeny and geographical proximity on cold tolerance variation among species
Because closely related species may occur in closer geographical proximity to each other compared with more distantly related species, an apparent phylogenetic signal in trait data may in fact be geographical (Freckleton & Jetz, 2009). To further understand how interspecific differences in cold tolerance have evolved, we disentangled the relative roles of phylogenetic and spatial distance in generating interspecific differences in cold tolerance. We fitted several variations of BM, which included not only an estimate of the phylogenetic signal (λ, as before) but also an estimate of the relative contribution of spatial proximity (φ) to interspecific differences. Spatial data were generated by calculating the average latitude and longitude a species occupies (Freckleton & Jetz, 2009; Cooper et al., 2011). This may be a crude representation of the centroid of a species' range, but manual inspection of the resulting latitude–longitude coordinates confirmed that it provides a good approximation of how geographically close species are, especially in a group such as this where clades of closely related species tend to be geographically confined. Using these data, we compared fit of: a model in which λ assumes its ML value (no spatial term); a model that includes no influence of phylogeny but a high spatial signal (λ = 0; φ = 1); and a model in which both λ and φ are estimated during model fitting. A high value of φ (approaching 1) indicates a large influence of geographical proximity on interspecific variation. The relative contribution of phylogeny, λ′, is calculated from ML estimates of λ and φ, as (1 − φ) × (1 − λ) (Freckleton & Jetz, 2009). If the relative contributions of spatial and phylogenetic distances do not sum to 1, then the remainder of the interspecific variation is considered independent of either geographical proximity or phylogenetic relatedness (denoted by γ). Models were fitted in R using the ‘cpglm’ functions of Freckleton & Jetz (2009), repeated for the entire subfamily and across clades 1 and 2 separately (see previous section). Model fit was assessed with AICc.
Because all plants survived the first winter, we consider only the results of winters 2 (winter 2008/2009) and 3 (2009/2010) of the common garden experiment. Temperatures differed among plots and years, with plot 2 being overall colder than plot 1 (Table 2). Consistently with these differences, survival rates differed among plots and years (Fig. 1). Winter survival ranged from 0 to 100% and, crucially, all species showed the same trend in both winters: those species that were the most tolerant in 2008/2009 were also the most tolerant in 2009/2010 (Fig. 1). Similarly, differences in survival rates were consistent among plots, regardless of whether plants were ‘hardened’ or not (‘nonhardened’). For instance, Rytidosperma unarede was present in all replicates and always showed higher survival rates in plot 1 than plot 2.
Minimum, maximum and mean temperatures (°C) measured by each of the two temperature readers and time (h) for which each plot experienced freezing temperatures, expressed as hours at or below 0, −5 and −10°C.
Minimum T (ºC)
Maximum T (ºC)
Mean T (ºC)
Time (h) ≤ 0°C
Time (h) ≤ −5°C
Time (h) ≤ −10°C
Realized temperature niches
The realized niches of the experimental species were calculated from temperature conditions at 4806 georeferenced occurrence points. Over half of these were for Rytidosperma setaceum (1510) and Danthonia spicata (1369) and only one was for Rytidosperma paschale. The plants of R. paschale derive from a single cultivated individual, in turn deriving from a single collection from Easter Island. We believe that the low survival rates found for this species (Fig. 1) reflect adaptation to the mesic, oceanic conditions on Easter Island. However, lack of genetic variation and occurrence data for this species, along with more recent concerns that this taxon may not merit specific status (Humphreys, 2010), led us to exclude it from further analyses relating to the common garden experiment. The mean number of records per species excluding R. paschale was 534. The North American D. spicata experiences the coldest temperatures overall in its native range (Fig. 2; Table S1). The rest of the species, all from the Southern Hemisphere, occupy much more similar climates, although Rytidosperma buchananii from New Zealand and Tenaxia disticha from southern Africa have relatively cold cool seasons and the Australian Rytidosperma richardsonii and R. setaceum experience overall milder cool seasons.
The PPCA partitioned 83% of the variation into two axes (Table S3). PC1 (42.3%) separates high from low temperatures during all seasons and is most strongly (negatively) correlated with mean annual temperature and temperatures during the wettest and driest quarter. PC2 (40.3%) separates high isothermality and increasing temperatures during cold seasons (upper end of axis) from high seasonality and increasing temperatures during hot seasons (lower end of axis).
Comparison of measured (fundamental) and modelled (realized) cold tolerance
The lowest temperatures tolerated in the garden plots (Table 2) compared with minimum temperatures in species' observed ranges (Fig. 2; Table S1) suggest that realized niches underestimate winter survival ability for seven species. Four species are able to survive more extreme cold than would be predicted from their distribution (R. buchananii, Rytidosperma laeve, R. unarede and T. disticha). Three species can tentatively be assumed to be able to do so, being tolerant (≥ 50% survival) of conditions only in plot 1 (R. richardsonii, R. setaceum and Tenaxia stricta). Conditions in the observed range of D. spicata overlap with conditions experienced in the winter gardens and Rytidosperma pilosum did not survive in either plot (Fig. 1; grey boxes, Fig 2).
The PGLS regressions (Table S4) revealed a consistently positive, but nonsignificant relationship between relative survival rates and PC1 (Fig. 3) but not PC2 (Fig. S1). We interpret PC1 as being a meaningful representation of relative cold tolerance differences among species: the highest survival rates were found in species that experience the lowest temperatures in their observed ranges and vice versa. Lack of significance for these results may be a consequence of the low number of data points (five to seven) rather than implying ecological nonsignificance (r2 values ranged from 0.3 to 0.7). The repeatability of results among plots and years strengthens the validity of this interpretation.
Timing of cold tolerance evolution
The PPCA of the temperature-related BioClim variables for 224 danthonioid species confirms that the variation among the nine experimental species reflects variation across the whole subfamily (Table S5) and that the variation in PC1 is suitable for understanding evolution of cold tolerance across the subfamily. The plot of the absolute contrasts of PC1 suggests that most change in cold tolerance has occurred towards the present as opposed to deeper in the evolutionary past (absolute contrasts increase with node age; Fig. 4b). Accordingly, the best-fitting model for ancestral state reconstruction was one in which δ > 1, that is, with most change occurring towards the tips (δ = 3.1, log likelihood (LL) = 1364.6, compared with BM where LL = 1138.6 and λ where LL = 1336.3; λ = 0.75). The resulting ancestral state reconstruction reveals that Danthonioideae were probably cold tolerant at the onset of the crown group diversification, at least compared with the much lower cold tolerance inferred for the sister group Chloridoideae (Fig. 4a,c). However, extreme cold tolerance has only evolved more recently (Fig. 4a,d).
These results are consistent with those inferred for the ingroup alone, for the MCC tree, as well as over the sample of 1000 trees (Table 3). For the MCC tree, λ is inferred as the best model (λ = 0.75, LL = 96.1). Across the sample of trees, both λ and δ are significantly better than both BM and λ = 0 (ΔAICc ≥ 13), with λ being favoured for 80% of the trees and δ being favoured for 7%. In 13% of the trees the two models are statistically indistinguishable (ΔAICc < 3). Parameter estimates for the MCC tree fall in the range of estimates for the sample of trees (Table 3). Thus, these results are robust to phylogenetic uncertainty and suggest that most change in cold tolerance has occurred towards the tips, mostly as described by λ and in a small proportion of trees as described by δ.
Table 3. Models of trait change fitted across the maximum clade credibility (MCC) tree, across clades 1 and 2 (see Fig. 4a) separately and across 1000 treesa
Lineage-specific differences in cold tolerance evolution
The most extreme levels of cold tolerance have evolved only in one of the two major sister clades: clade 2 (Fig. 4c). The apparent phylogenetic signal of cold tolerance is also higher in this clade (λ = 0.75 compared with λ = 0.54 in clade 1).
An initial differential rates analysis identified a shift in a small clade including R. paschale (see Fig. 4a). Because inferences concerning this taxon need to be treated with caution (see section ‘Realised temperature niches’), we repeated the analysis without it. Other findings remained unaffected by whether or not R. paschale was included, but we report the results of analyses without this taxon. Thus, the best model describes six significant rate shifts (LL = 133.3, significantly better than BM: P <0.001, χ2 test with 6 df; Table S6). This demonstrates that the 10 rate shifts allowed were sufficient to capture the rate heterogeneity among lineages. There is a rate shift deep in crown Danthonioideae (‘core’ Danthonioideae; Fig. 5a, excluding Capeochloa, Geochloa and Merxmuellera). Several rate increases are found in lineages currently occupying the coldest areas: very high rates in Danthonia and moderate rates in Rytidosperma and within Pentameris, in a clade comprising southern African highland species (P. lima clade). There is also a moderate increase in rates in ‘core’ Pentameris – support for this shift is the weakest (ΔAIC = 5) – and finally, a significant decrease in rates within Tribolium, a small clade of semi-desert species.
The best model inferred from the OU analysis is when both temperature optima and rates of change are allowed to differ among groups (OUMV model; ΔAICc = 36.6 compared with the second best model; Tables 4, S7). This model provides clear evidence that species are under different selective regimes with regard to temperature (Fig. 5b). Chionochloa (New Zealand), Danthonia (the Americas), the P. lima clade (southern African highlands) and Rytidosperma (Australasia and South America) are inferred as tending towards significantly colder temperature regimes than the other groups (core Danthonioideae, core Pentameris and core Tribolium, the latter tending towards the warmest optimum overall). Thus, selection for increased cold tolerance has occurred in Danthonioideae independently of lineage (Chionochloa, Danthonia, P. lima clade and Rytidosperma) and continent (clades occupying all Southern Hemisphere continents and North America).
Table 4. Selective optima in cold tolerance (the first principal component (PC1)) inferred from the best Ornstein–Uhlenbeck (OU) model (OUMVl Table Supporting Information S7)
Danthonia, mean ± SE
Rytidosperma, mean ± SE
P. lima clade, mean ± SE
Core Tribolium, mean ± SE
Chionochloa, mean ± SE
Core Pentameris, mean ± SE
High PC1 values correspond to high cold tolerance.
Phylogenetic versus spatial signals in cold tolerance data
The addition of a spatial term to models explaining interspecific differences in cold tolerance significantly improves model fit (Table 5). The best model is one in which both λ and φ are estimated during model fitting (LL = 117.3, ΔAICc = 32 compared with the second best model). When λ and φ are estimated simultaneously, the ML estimate of λ drops from 0.75 to 0.39, meaning that much apparent phylogenetic signal in the data is in fact spatial. Accordingly, the ML estimate of φ is high (0.88) and the relative contribution of phylogeny in accounting for the variation is negligible (λ′ = 0.05). Similar results are found when clades 1 and 2 are analysed separately (Table 5). However, in clade 1 both phylogenetic and spatial distances contribute less to interspecific differences in cold tolerance (λ′ = 0.002; φ = 0.77), meaning that a larger proportion of the variation is independent of both phylogeny and geography (γ = 0.23). In clade 2 the relative contribution of spatial distance is very high (φ = 0.91), with phylogeny accounting for only a small proportion of the variation (λ′ = 0.05).
Table 5. Estimates of the relative contribution of spatial and phylogenetic distances to interspecific differences in cold tolerancea
The relative contribution of spatial distance (φ) is estimated during model fitting. The relative contribution of phylogenetic distance (λ′) and of effects independent of both phylogenetic and spatial distances (γ) are calculated from maximum likelihood (ML) estimates of parameters λ and φ as in Freckleton & Jetz (2009).
We found that seven out of the eight Southern Hemisphere species tested survived the Northern Hemisphere winter, enduring temperatures down to −9 to −13°C (Table 2). These species survived colder temperatures than would be predicted from minimum temperatures in their native ranges (Fig. 2). Although field experiments cannot be interpreted as revealing absolute climate tolerances (Bannister, 2007), we show that relative winter survival rates are consistently, albeit nonsignificantly, related to relative realized temperature niches extracted from climate data from across species' ranges (PC1 axis; Fig 3): species that experience colder temperatures in their native ranges generally showed higher survival rates. These findings are consistent with realized niches being mostly contained within, while accurately predicting, fundamental niches (Hutchinson, 1957; Soberón, 2007). Species distribution modelling relies on inferences based on the realized niche (Guisan & Zimmermann, 2000). There are surprisingly few experimental tests of how these inferences hold with respect to predicting distribution patterns based on the fundamental niche (but see Vetaas, 2002; Cavender-Bares et al., 2011). Our study is rare in providing some insight into the relationship between the two, and our findings validate temperature tolerances obtained by niche modelling. Because experimentation is arguably less straightforward in some systems (e.g. mammals; Olalla-Tárraga et al., 2011) an experimental approach in amenable systems becomes all the more important (see also Vetaas, 2002; Cavender-Bares et al., 2011). Further experimentation is needed to establish more generally the strength and nature of the relationship between realized and fundamental niches, to bolster findings from the vast niche modelling literature.
Evolution of cold tolerance in danthonioid grasses
Using relative cold tolerance, expressed as position along the PC1 axis of temperature-related climate variables, to study the evolution of cold tolerance across the subfamily, we found that Danthonioideae made the transition into the temperate zone at the onset of its diversification in Africa in the late Oligocene (Bouchenak-Khelladi et al., 2010; Pirie et al., 2012; Fig. 4a,c). Thus, cold tolerance evolved more recently in Danthonioideae than the recruitment of novel gene families involved in cold tolerance in Pooideae (Sidebottom et al., 2000; Sandve & Fjellheim, 2010), but broadly coincidentally with a major period of global cooling and expansion of the Antarctic ice sheet (Zachos et al., 2001). However, evolution of cold tolerance has been constant neither over time nor among lineages. Most change is inferred to have occurred towards the present (Table 3, Fig 4b) and it has apparently not yet plateaued (Fig. 4d). The most extreme levels of cold tolerance thus evolved only relatively recently and in certain lineages only: in Chionochloa, Danthonia and Rytidosperma in New Zealand and the Americas (Fig. 4a), which together with a small clade of southern African highland species (P. lima clade) tend towards significantly colder climate regimes than the other lineages (Fig. 5b). Most extreme levels of cold tolerance are found in lineages that are inferred to have experienced elevated rates of cold tolerance evolution during the Miocene (Figs 4d, 5a). Because these lineages are geographically confined, it is possible that increased rates were triggered by expansion into new regions. Rytidosperma arrived in New Zealand in the late Miocene, around the time when mountain building commenced there (Humphreys et al., 2010). During the same period Danthonia diverged from its less cold-tolerant South American ancestors and started migrating into North America (Pirie et al., 2012). By contrast, extreme cold tolerance in Chionochloa is associated with neither a shift in rates nor arrival in a new region: Chionochloa was probably present in New Zealand much earlier (Pirie et al., 2010, 2012) and therefore probably responded to the cooling climate in situ. These apparently independent entries into the coldest habitats occupied by danthonioids would have involved either fine-tuning a genetic mechanism already in place, or evolution of extreme cold tolerance de novo. Distinguishing these two possibilities will require establishing whether the strategy of frost tolerance employed by Chionochloa (Wharton et al., 2010) is shared by all danthonioid lineages.
The spatial signal of cold tolerance and the difference between Cape lineages and those elsewhere
Geographical proximity of closely related species can account for almost all interspecific differences in cold tolerance in Danthonioideae (Table 5). The apparently high phylogenetic signal in the cold tolerance data (Tables 3, 5) therefore cannot be explained by an inability to evolve climate tolerances not expressed by ancestral populations (phylogenetic niche conservatism sensu Wiens & Graham, 2005). Rather, our results suggest that closely related species occupy similar temperature regimes because they have not dispersed from their ancestral ranges and are therefore under similar selection pressures (Freckleton & Jetz, 2009). Evidence for phylogenetic niche conservatism in temperature in other organisms is mixed: there is no evidence for this in mammals (Freckleton & Jetz, 2009; Cooper et al., 2011) whereas Drosophila (Kellermann et al., 2012) and amphibians (Olalla-Tárraga et al., 2011) apparently show a degree of conservatism.
Within Danthonioideae, the degree to which spatial proximity can explain variation in cold tolerance differs among the two major sister clades. Clade 1 comprises all species of Pentameris, the largest group of grasses in the Cape flora of South Africa, which extends also into sub-Saharan Africa and the Afromontane/Afroalpine regions (Linder & Ellis, 1990; Galley & Linder, 2007). Sister species in this clade generally differ in their realized temperature niche (Fig. 4a) and, accordingly, a considerable proportion of the variation in cold tolerance cannot be accounted for by phylogeny or spatial proximity (Table 5). A plausible explanation for this in a Cape lineage is that variation in temperature tolerances has arisen through divergent selection (Goldblatt, 1978; Schluter, 2001; Linder, 2003), as diverging populations adapt to their own narrow portion of the topographically heterogeneous Cape terrain. By contrast, clade 2 occurs on all continents except the polar regions, with closely related species generally occupying geographically confined areas on a single continent. Interspecific temperature differences in this clade are almost entirely accounted for by spatial distance between species (Table 5), suggesting that local adaptation to climatic conditions has occurred across lineages and continents. Together with the findings of increasing rates of evolution towards more extreme cold tolerance, occurring independently of lineage and region, these results provide strong evidence against temperature tolerances limiting the current distribution of Danthonioideae.
The current range of danthonioid grasses does not appear limited by (low) temperature
In conclusion, by unravelling the details of the evolution of cold tolerance in danthonioid grasses, drawing on both experimental and comparative approaches, we have demonstrated that, although Danthonioideae are an entirely temperate clade, they exhibit significant divergence in temperature tolerances and no evidence to suggest that their current distribution is generally limited by their tolerance of low temperatures. Cold tolerance is inferred to be increasing, both in rate of change and in extent, and species from several continents are able to tolerate lower temperatures than predicted from their observed ranges. This seems reasonable for species belonging to an already cold-adapted clade. From a molecular perspective, the levels of flexibility in cold tolerance among grasses, possibly involving altering the strength or duration of the cold tolerance response, remain poorly understood (Sandve et al., 2011). From a macroevolutionary perspective, it is possible that these results are a signature of climates no longer present, an adaptive relict of increasingly cold glacial periods during the Pleistocene. Such an explanation implies that the current realized climate niche of Danthonioideae is truncated (R. O. Wüest et al., unpublished), possibly by climatic warming since the last glacial maximum. In support of this, danthonioid grasses are often dominant in the south, extending their ranges to the southernmost continental edges. In turn, this supports the conclusion that degree of cold tolerance is unlikely to be a major factor limiting expansion in the Northern Hemisphere. Instead, competition, possibly from the analogous pooid grasses, the time available for colonization of northern continents (see also Linder et al., 2013) and/or dispersal limitation, supported here by the strong spatial autocorrelation in cold tolerance data, provide alternative explanations. There is mounting evidence that one or all of these factors are of much greater importance for explaining broad-scale biodiversity patterns than is generally appreciated (e.g. Salisbury et al., 2012), possibly also providing a more general explanation for the limited southern expansion of north temperate, species-rich, herbaceous lineages such as the pooid grasses, buttercups (Ranunculaceae) and sedges (Carex).
We thank Markus Meierhofer and Melanie Ranft for practical assistance and Mike Pirie, Bryan and Pam Simon and Neville Walsh for support in the field. Luke Harmon, Michael Kessler, Lynsey McInnes, Paul Page, Ally Phillimore, Mike Pirie, Rafi Wüest and five anonymous reviewers provided valuable discussion and/or comments on earlier versions of this paper. Financial support by the Swiss NSF (grant 3100A0-107927 to H.P.L. and grant PBZHP3-133420 to A.M.H.) and the Claraz-Schenkung Stiftung is acknowledged.