Contrasting patterns of phylogenetic turnover in amphibians and reptiles are driven by environment and geography in Neotropical savannas

Cross‐taxonomic congruence in biodiversity patterns is key to understanding the main drivers of community structure, for biogeographical regionalization and to guide conservation. We aim to map the patterns of phylogenetic turnover and disentangle the geographical and environmental factors that drive the phylogenetic composition of distinct faunal assemblages.


| INTRODUC TI ON
The influence of intrinsic characteristics of distinct organisms-such as ecophysiological requirements, habitat use, dispersal ability and degree of niche conservatism-on large-scale biodiversity patterns has been a long-standing point of discussion in biogeography (De Candolle, 1820;Nelson, 1978). With increasing knowledge on the geographical distribution of multiple species and their respective evolutionary relationships, this question can be more precisely tackled (Saladin et al., 2019). It is now possible to estimate changes in lineage composition across space by measuring the replacement of phylogenetic branches among sites (phylogenetic turnover ;Rosauer et al., 2014). Phylogenetic turnover may highlight breaks in lineage distribution among sites, which is shown to be correlated with the borders of biogeographical regions (e.g. ecoregions, domains) (Daru et al., 2017). Although biogeographical regionalization is the classification of biotas and areas into distinct entities, it reflects processes that shape biotas, such as vicariance, dispersal and niche filtering (Daru et al., 2018;Vilhena & Antonelli, 2015), often correlating with climatic breaks, orographic barriers and tectonic history (Ficetola et al., 2017). Therefore, phylogenetic turnover can be used as a framework to assess how fine-scale changes in lineage composition can lead to large-scale differences in biogeographical patterns across different groups of organisms.
The importance of the evolutionary history of distinct organisms in shaping biodiversity patterns can be investigated by comparing how much phylogenetic turnover departs from taxonomic (species) turnover (Weinstein et al., 2014). For example, the rate of phylogenetic turnover is expected to be lower than the rate of taxonomic turnover when most taxa above species level are present in the majority of sites (Bryant et al., 2008). When both taxonomic and phylogenetic turnover are low, this may indicate high dispersal rates, weak historical isolation and a low degree of niche conservatism (Alexander Pyron & Burbrink, 2009;Guevara et al., 2016).
Conversely, high phylogenetic turnover is thought to be related to ancient geographical barriers, dispersal limitation, a high degree of niche conservatism or high extinction rates (Graham & Fine, 2008).
Higher phylogenetic turnover is also expected with increasing elevation, due to greater isolation and habitat heterogeneity in highlands, and along extreme and ancient environmental gradients (Bryant et al., 2008), whereas lowlands may facilitate faunal interchange , resulting in low phylogenetic turnover. If both the degree of phylogenetic turnover over evolutionary history and the responses to geography and environment are similar among taxa, the result is a common pattern of phylogenetic regionalization of the biota (Daru et al., 2017).
Squamate reptiles (lizards and snakes) and anuran amphibians (frogs and toads) generally present very distinctive habitat requirements (Vitt & Caldwell, 2013), which may influence dispersal and isolation. Amphibians generally show a lower vagility due to a stronger association to wet or humid environments (Marshall et al., 2018), which tend to be geographically more restricted in open biomes such as savannas. In contrast, reptiles occur not only in these areas but are also abundant in interfluves and drier habitats (Costa et al., 2007), with varying degrees of habitat specialization for different taxa (Nogueira et al., 2009). Consequently, water-related variables can be more critical for structuring the composition of communities and connectivity among sites for amphibians than for most reptiles, since species of several amphibian clades rely on standing water for breeding (Wells, 2007). Changes in faunal assemblage composition could therefore be more related to environmental distances for amphibians and to geographical distances for reptiles (Chen et al., 2011). Conversely, temperature is shown to be an important driver for reptile distribution due to thermoregulatory strategies (Huey & Stevenson, 1979), whereas interactions between minimum temperatures and precipitation might drive the distributions of nocturnal ectotherms, such as most amphibians (Hillman et al., 2009).
Finally, ground characteristics can be critical to driving connectivity among sites for both groups due to habitat preferences of distinct taxa, from fossorial organisms (several reptile taxa) to rock-outcrop specialists-breeding sites for several amphibian species and also preferred habitat for saxicolous reptiles (Vitt, 1993). Thus, if local habitat influences connectivity and isolation among regions for long periods, this may result in contrasting rates of phylogenetic turnover (Daru et al., 2017) between reptiles and amphibians.
In the Cerrado savannas of South America, almost 200 species of squamate reptiles and anuran amphibians have narrowly distributed ranges coincident with several relief units (Fig. S1-Azevedo et al., 2016). This coincident distribution among Cerrado endemics suggests a similar biogeographical history for both groups , which could also indicate a coincident phylogenetic regionalization across the same highlands. The comparison of the distribution patterns between species endemic or not to the Cerrado has indicated a high importance of the proximity to nearby biomes in structuring amphibian faunal assemblages in general, and with lowlands acting as corridors for faunal interchange among biomes diversity patterns of one group should ideally not be used as a surrogate to map general patterns or to understand the drivers of diversity of other co-occurring groups. Thus, conservation efforts need to be designed and implemented for each organismal group.

K E Y W O R D S
beta diversity, Cerrado, connectivity, generalized dissimilarity modelling, herpetofauna, regionalization mostly for non-endemic reptiles . Species of reptiles and amphibians endemic to the Cerrado are generally more associated with open savanna ecosystems, especially across the plateaus and highlands, better reflecting the history of the typical Cerrado biota than the non-endemic species, which generally are either more habitat-generalist, forest-dwelling and widely ranged . Therefore, a comparison of patterns between all species and Cerrado endemics only may provide relevant insights into the relative importance of topography, environment and nearby biomes in the patterns of phylogenetic turnover.
Species composition and lineage diversification of the Cerrado herpetofauna are mostly related to major biogeographical events and changes in climate since the middle Miocene in the region (Azevedo et al., 2020;Domingos et al., 2014;Vasconcellos et al., 2020). From that time, global climate cooling led savanna ecosystems to become more widespread, as evidenced by some Cerrado lineages showing signals of range expansion (Machado et al., 2014;Prado et al., 2012).
The Miocene to Pliocene uplift of the Central Brazilian Plateau, followed by its erosion by river drainages, compartmentalized the Cerrado into different mountain ranges (Ab'Sáber, 1998), which clearly delimit many ranges of numerous lineages (Guarnizo et al., 2016;Werneck et al., 2012), and is related to endemism patterns in the region (Azevedo et al., 2016;Nogueira et al., 2011). Topography also influences climate stability in the region, with plateaus and mountains harbouring a higher species richness of typical Cerrado species than in flat, lowland areas (Costa et al., 2007). More recently, Pleistocene climate fluctuations may have led to the expansion and retraction of forested biomes into the Cerrado core, fragmenting even more the ranges of some Cerrado lineages (Vasconcellos et al., 2019). These fluctuations may also explain the occurrence of lineages typical of Amazonian or Atlantic Forests, in enclaves and gallery forests of the Cerrado, and species that further specialized to living in savanna habitats (Fouquet et al., 2014;.
Altogether, climate, vegetational and topographic dynamics, and the proximity to other biomes have potentially shaped phylogenetic turnover, explaining breaks in lineage distributions, and finally, shaping regionalization patterns in the region.
In this study, we estimate phylogenetic turnover for amphibians and reptiles in the Cerrado, comparing analyses including all species with analyses including only Cerrado endemics, and addressing the following questions: 1. How much does phylogenetic turnover differ from taxonomic turnover? We expect to find a larger difference in amphibians than in reptiles due to their generally low vagility and higher association with nearby biomes (Valdujo, Silvano, et al., 2012).
For amphibians, this difference will be related to the presence of more lineages typical of neighbouring biomes restricted to areas closer to the ecotones, including either species that occur only marginally in the Cerrado or endemic species derived from lineages that have diversified mostly in other biomes (Valdujo, Silvano, et al., 2012). For reptiles, as lineages typical of neighbouring biomes are more widespread within the Cerrado , the difference between taxonomic and phylogenetic turnover should be smaller.

2.
Which environmental variables are correlated to phylogenetic turnover? Habitat-use differences between reptiles and amphibians should influence the rate of phylogenetic turnover along environmental gradients (Ferrier et al., 2007). Soil-related variables should be important for reptiles due to the large number of fossorial and semi-fossorial species in the Cerrado , and also for amphibians, as these variables may influence the formation and duration of breeding sites and shelter availability during the dry season (Vitt & Caldwell, 2013). The importance of each climatically related variable should differ between the two groups due to the differences in the thermal ecology of diurnal or active thermoregulators (mostly reptiles) and nocturnal ectotherms (most amphibians) (Huey & Stevenson, 1979).

Does the clustering of phylogenetic turnover among sites result in geographically concordant breaks in lineage composition for reptiles
and amphibians? Considering their potentially different responses to geography and environment (Question 2), we expect a distinct pattern in the clustering of phylogenetic turnover for these groups. However, as in the case of endemism patterns, a congruent pattern should emerge between both groups, considering the importance of relief units to shape the distribution of organisms in the region.

| Study area
We used the delimitation of the Cerrado as an ecoregion (Olson et al., 2001), not including the complex ecotonal ecoregion Maranhão Babaçu forests (between Caatinga and Amazonia) due to its transitional nature and scarcity of herpetofaunal data. The Cerrado is a vast savanna occupying about 1.98 million square kilometres in central South America (IBGE, 2004). The topography is characterized by plateaus, generally above 500 m, and depressions eroded by major drainage systems (Fig. S1). Gallery forests are ubiquitous at most scales, along with typical interfluvial grasslands and savannas (Ratter et al., 1997). This high water availability differentiates the Cerrado savannas from most African savannas (Cole, 1988), and is possibly the reason for the high species richness and endemism of amphibians in the region (Valdujo, Carnaval, et al., 2012). On a large scale, the texture of soils in the region varies from rocky outcrops mostly on highlands and escarpments (Ribeiro & Walter, 2008), sandy soils especially across or nearby ancient sandstones (e.g. north-western Cerrado) or alluvial deposits (e.g. Tocantins-Araguaia Basins) (Villela & Nogueira, 2011) and weathered, deep, clay-rich soils found mostly across tablelands of the Central Brazilian Shield (Ribeiro & Walter, 2008). Mean annual temperature varies within a 10° range (18-28°C), with lower mean temperatures generally towards the south, and with minimum temperatures mostly in high elevation areas (da Silva et al., 2008). Precipitation exhibits a large degree of variation in the region (~800-2400 mm per year; Bueno et al., 2018), with drier climates in the north-east and south-western Cerrado, and a mild east-to-west increase in annual precipitation.

| Distributional data
We used a previously compiled voucher-verified dataset of occurrence records for 265 species of squamate reptiles and 193 species of anuran amphibians occurring in the Cerrado (Azevedo et al., 2016;Nogueira et al., 2019;Valdujo, Silvano, et al., 2012). We then produced presence-absence matrices using a cylindrical equal-area Behrman projection grid where the width of each cell was 0.1 degrees. This choice reflects the finest resolution for a substantial part of our data (Valdujo, Silvano, et al., 2012). From this grid, we only included cells containing a minimal number of recorded species, from where we also sampled the environmental variables: (1) cells containing at least 10 recorded species for each group, as in Valdujo, Carnaval, et al., (2012); (2) sites with at least two species endemic to the Cerrado savannas; and (3) in the case of reptiles, sites with at least one species of lizard and one species of snake endemic to the Cerrado, as these two groups vary widely in detectability levels (snakes have much lower detectability; Vitt & Caldwell, 2013).
Finally, we also produced a subset of the presence-absence matrices including only species endemic to the Cerrado (100 reptiles and 92 amphibians, as defined in Nogueira et al., 2010;Valdujo, Silvano, et al., 2012;Azevedo et al., 2016). We included all selected sites in the procedure above for each group, and as very few sites coincide between reptiles and amphibians, we checked whether they represent similar distributions of environmental conditions in the region (see Modelling phylogenetic turnover).

| Phylogenetic data
To calculate phylogenetic turnover, we followed the phylogenetic hypotheses proposed by Tonini et al., (2016) and Jetz and Pyron, (2018) for reptiles and amphibians respectively. As species without molecular data were assigned to the respective genus in a posterior distribution representing different combinations of within-genus evolutionary relationships, we repeated all phylogeny-dependent analyses for a sample of 100 of these phylogenies. We then used the median values in all subsequent analyses. In the resulting phylogenies, around 60% of all species were included through assignments to genus level. We recognize that our results should be carefully interpreted at the species level. Previous sensitivity analyses with similar phylogenies indicated good accuracy and low sensitivity for mapping phylogenetic patterns . Even so, we accounted for this uncertainty in our analyses by weighting phylogenetic turnover according to the proportions of species with molecular data in each site (see the subsection Modelling phylogenetic turnover).

| Phylogenetic beta diversity
Phylogenetic turnover is one of the two phylogenetic beta diversity components and accounts for the proportion of phylogenetic branches that are replaced from one site to another. We measured phylogenetic turnover between sites with the Simpson derived pairwise phylogenetic dissimilarity index (Leprieur et al., 2012), a metric less sensitive to the total species richness, which ranges from 0 (all branches are shared between sites) to 1 (no branches are shared).
The second component is called nestedness and accounts for losses without replacement in phylogenetic branches (Baselga, 2010). We estimated the total multiple site phylogenetic beta diversity related to each of these components (Baselga, 2013). However, for the subsequent analyses, we only considered the turnover component to separate the potentially complex effects of differences in species richness among sites (Soininen et al., 2018). For the tests against the null distributions (see below), we also measured the taxonomic turnover as the Simpson pairwise dissimilarity (Koleff et al., 2003).

| Decoupling of phylogenetic and taxonomic turnover (Question 1)
To test whether phylogenetic turnover is decoupled from taxonomic turnover, we regressed their scores against each other (site by site).
We then compared this result (observed) to a null distribution of slope values of phylogenetic turnover against the taxonomic turnover. We generated this null distribution by randomly shuffling the tips of the phylogeny in 1,000 randomizations. We considered phylogenetic turnover to be decoupled from taxonomic turnover if the resulting slope was outside the interval between the 2.5 and 97.5% of the slopes resulting from the null distribution (Fig. S2).

| Environmental characterization
To characterize the climate of each site (Question 2 -Which environmental variables are correlated to phylogenetic turnover?), we downloaded 19 bioclimatic variables from the CHELSA project (http:// chels a-clima te.org/; Karger et al., 2016). We also characterized soils in each site from layers of predicted values of the proportion of sand, clay and coarse fragments (median values for the first 15 cm of the soil surface) downloaded from soilgrids.org (Hengl et al., 2014).
Additionally, we estimated the elevation, relief roughness and slope of each site with a digital elevation model (GTOPO30, 2015), and the distance to the nearest stream (derived from: Lehner et al., 2008).
However, this last variable was not included as it does not show any variation within the grid resolution used here. We measured relief roughness as the largest difference in elevation of each site with its eight closest neighbours. This variable represents environmental heterogeneity and may account for the presence of a greater variety of microhabitats within a site. We then retained only variables with no multicollinearity problems after calculating their variance inflation factor (VIF), using a threshold of 10 (Naimi et al., 2014). This selection resulted in five layers for precipitation and five for temperature, representing most of the climatic variation in the region (Table 1). Among the non-climatic variables, the slope was excluded due to high collinearity with relief roughness. Although we automatically removed variables with high VIF, we discussed all highly correlated variables (> 0.9; Dohoo et al., 1997) as groups (e.g. relief roughness and slope; Fig. S3). All groups of environmental predictors have biological justification for explaining lineage distributions ( To test which set of environmental variables best explain phylogenetic turnover, we used generalized dissimilarity modelling (GDM), a modelling technique based on regressions of nonlinear distance matrices (Ferrier et al., 2007). First, we produced GDM models with all possible combinations of variables with low VIF scores described above (five temperature, five precipitation, three soil, relief roughness, elevation and geographical distance). Then, starting from the full model, we successively eliminated the variable with the smallest contribution. In each round, we calculated the relative variable importance (percentage of explained deviance between a model with and without a variable) and significance (P-value <0.05) through matrix permutations for each variable (n = 1,000). Model selection using Akaike's Information Criterion (AIC) does not apply to techniques of regression matrices such as GDM (Wagner & Fortin, 2015). Therefore, we used an equivalent strategy by selecting the model with the highest value of explained deviance which retained only variables that were important in more than 160 rounds of permutation (following the 0.16 optimization level suggested in Heinze et al., 2018). We then visualized the most important variables using I-splines, which describe nonlinear monotonic relationships among variables and phylogenetic turnover (i.e. partial ecological distance).
We calculated the unique contributions of geographical and environmental distances by partitioning the explained deviance of GDM models including only each set of distances (geography or environment) and the model including all sets (Fitzpatrick et al., 2013). All
Percentage of sand (Percentage of clay*) Soil characteristics influence vegetation structure, locomotion strategies for reptiles and breeding sites for amphibians.

Temperature seasonality (Precipitation of driest month)
Climatic seasonality/stability in general. One of the main differences between Cerrado and the surrounding forested biomes. Encompasses a broad range of temperature characteristics inversely correlated to elevation in the Cerrado region, with the highest temperatures occurring in lower elevation areas.
Precipitation of the wettest month (Precipitation of the wettest quarter) The amount of precipitation in the rainy season influences booms of food availability and the reproduction season for both groups.
Precipitation seasonality (Precipitation of the driest quarter*) Highly seasonal areas in the Cerrado demand ecophysiological and behavioural adaptations of organisms, influencing species ranges.
Precipitation of the Warmest Quarter (Min Temp. Coldest Month*, Mean Temp. of Coldest Quarter*) These variables represent a climatic latitudinal gradient (within the Cerrado only), with coldest temperatures in the southern parts of the Cerrado. Minimum temperatures limit ranges of ectothermic organisms.
Highly correlated variables (>0.9) not included in the model selection but representing equivalent environmental gradients in the region are shown in parentheses (see also Fig. S3). Asterisks denote a negative correlation with the selected variables.
TA B L E 1 Environmental correlates of phylogenetic turnover for reptiles and amphibians selected in the best GDM model, and the respective biological interpretation GDM models were produced with the R-package gdm (Ferrier et al., 2007). As sites selected for reptiles and amphibians are generally non-overlapping, we visually inspected whether they encompass the same environmental spaces to verify the influence of site selection on our GDM modelling.
To take into consideration the variation in the phylogenies used here, we weighted the contribution of each site to the GDM models by the standard deviation in the phylogenetic diversity of a site across the sample of 100 phylogenies divided by the respective number of species at each site (PD -Faith, 1992). In this way, sites with considerable PD variation per species due to the random taxonomic assignment will have a lower contribution to the final results.

| Observed and predicted clustering (Question 3)
To address Question 3 (Does the clustering of phylogenetic turnover among sites result in geographically concordant breaks in lineage composition for reptiles and amphibians?), we compared the observed values of phylogenetic turnover between reptiles and amphibians as well as the GDM model predictions projected onto geographical space. We classified sites that are phylogenetically more similar to each other by clustering both the observed and the predicted phylogenetic turnover using UPGMA, a widely applied method for clustering dissimilarity matrices in bioregionalization analyses (Kreft & Jetz, 2010). As our intention is not to provide a definitive bioregionalization scheme, but rather assess whether breaks in the phylogenetic turnover are congruent between reptiles and amphibians, we mapped the first 12 clusters for each group in different colours.
This number was chosen to be comparable to previous regionalization schemes using species-level analyses of endemism (Azevedo et al., 2016).

| Phylogenetic beta diversity
The multiple site tests indicated that nearly all phylogenetic beta diversity for reptiles (turnover =0.94, nestedness =0.03) and amphibians (turnover =0.95, nestedness =0.02) was explained by the turnover component. This shows that the vast majority of differences in phylogenetic branches shared among sites are not related to losses in phylogenetic branches without replacement.

| Decoupling of phylogenetic and taxonomic turnover (Question 1)
For all datasets (both reptiles and amphibians with all species or only Cerrado endemics), there was a positive increase in phylogenetic turnover with increasing taxonomic turnover, which shows greater slope values for the datasets with all species compared to endemics, and with both turnover metrics more decoupled for amphibians in general (Table 2, Fig. S4 a-b).

| Correlates of phylogenetic turnover (Question 2)
In general, phylogenetic turnover was higher for amphibians than reptiles across the environmental gradient, except for precipitation seasonality (similar response) and precipitation of warmest quarter (collinear with minimum winter temperatures) with higher importance for endemic reptiles (Figure 1, Table 3). Relief roughness was significantly important for amphibians (both datasets), whereas the percentage of sand was only significant for endemic amphibians (Table 3, Figure 1). In general, the effects of geographical distance alone were more important for the dataset including all reptiles (32% of explained GDM deviance), whereas for all other datasets, purely environmental distances were more important (> 50%, Fig.   S5). Sites selected for reptiles and amphibians represented a similar environmental space (Fig. S6), thus allowing comparison between both groups. Finally, we did not find any major differences in the results using our procedure to weight phylogenetic turnover according to the number of species with molecular data available, when compared to the analyses without the weighting procedure (all results shown are derived from the weighted procedure).

| Observed and predicted clustering (Question 3)
The clustering of phylogenetic turnover indicated spatial mismatches in the direction of phylogenetic turnover between reptiles and amphibians. For both amphibians and reptiles, there was a latitudinal and a longitudinal differentiation in the phylogenetic composition of sites (Figure 2a and b). However, this differentiation was not so abrupt for reptiles, with several sites classified in different clusters occurring side by side (Figure 2a). The clustering of predicted values of phylogenetic turnover indicated a more TA B L E 2 Rate of increase in phylogenetic turnover with taxonomic turnover The numbers indicate the rate of turnover (coefficient) estimated by linear regressions. Numbers in parentheses indicate the proportion of times phylogenetic turnover rates were lower than the null distribution.
Rates of phylogenetic turnover departed more from taxonomic turnover for amphibians than for reptiles (numbers in parentheses and Fig. S4).
phylogenetically uniform biota in the western part of the Cerrado for amphibians (Figure 2d) when compared to reptiles (Figure 2c).
There was also a higher spatial mismatch between observed and predicted clusters for reptiles, indicating a lower tracking of environmental conditions for this group. Similar patterns were obtained for the endemics-only dataset (Fig. S7), with a less uniform structuring in the predicted clustering for endemic reptiles (as compared to all reptiles) and an even more extensive area with uniform lineage composition for endemic amphibians in the western part of the Cerrado (Fig. S7).

F I G U R E 1
Geographical and environmental predictors of phylogenetic turnover. I-splines were estimated from GDM models. The height of each I-spline indicates the relative contribution of each variable to the phylogenetic turnover observed along each environmental gradient (i.e. the partial ecological distance). Asterisks highlight variables with variation in additional climatic/environmental classes: *Highly collinear with elevation; **Highly collinear with minimal winter temperatures (see Table 1 and Fig. S3). Precipitation seasonality is a key variable for both reptiles and amphibians, whereas the relationship of phylogenetic turnover to all other variables differs considerably between each group [Colour figure can be viewed at wileyonlinelibrary.com]

| DISCUSS ION
We found differences between reptiles and amphibians in relation to all three questions tackled here. Firstly, in the degree of decoupling between phylogenetic and taxonomic turnover (higher for amphibians, Question 1). Secondly, the importance of environmental and geographical distances in relation to phylogenetic turnover differs between the two taxonomic groups-this is also true for the respective datasets with Cerrado endemics only, with geography alone important only for all reptiles (Question 2). Thirdly, we found that the clustering of phylogenetic turnover is largely incongruent between reptiles and amphibians (including the endemic-only datasets) in the

Cerrado savannas of South America (Question 3).
Our results contrast with the species-level geographical congruence in the distributions of reptiles and amphibians (Azevedo et al., 2016), reinforcing the importance of including information on the evolutionary relationships among species to understand biogeographical patterns (Fenker et al., 2014(Fenker et al., , 2020. These differences between major groups are likely to be the result of long-term evolutionary distinct responses to geography and environment, indicating that small-scale habitat preferences and ecophysiological differences may lead to detectable imprints in the geographical distribution of lineages. The higher decoupling of phylogenetic and taxonomic turnover in amphibians (Question 1) is in accordance with our predictions and may indicate a combination of higher dispersal limitation, higher association with nearby biomes and higher niche conservatism for this group (Myers et al., 2013;Weinstein et al., 2014). This expectation holds true even when excluding species typical of other biomes that marginally occur in the Cerrado (endemic-only dataset). The coupling between both turnover estimations for reptiles could be related to the still incomplete molecular data in the phylogenies used here, which is known to decrease the power of analyses depending on branch length estimations (Title & Rabosky, 2016). However, the reasonably complete genus-level resolution of the phylogenies used here allowed us to infer that phylogenetic turnover is not driven mostly by species from the same genus (e.g. as in Amazonian whitesand forests; Guevara et al., 2016). Also, phylogenetic turnover for all reptiles and endemic reptiles and amphibians is not predominantly driven by species from distantly related lineages (above genus level), which is more common when comparing sites from two distinct biogeographical regions (e.g. Neotropics vs. Nearctic; Antonelli, 2017).
Although we are not directly assessing the evolutionary processes that shaped phylogenetic turnover, our findings indicate distinct patterns for reptiles and amphibians and the underlying processes could potentially differ at different time scales.
Our results indicate that environmental filtering is more important in explaining phylogenetic turnover for amphibians (both datasets), whereas geographical distance is more important for all reptiles (Question 2). For amphibians in general, the presence of breeding sites (e.g. relief roughness and soil texture) and the effects of temperature on metabolism (warmer temperatures and high precipitation) (Hillman et al., 2009) are the main correlates of phylogenetic turnover, potentially indicating niche-filtering processes. -18.2** --All four models explain significantly more deviation than expected by random (p < 0.001). Asterisks indicate: *p < 0.05, **p < 0.01, ***p < 0.001. Variable importance is the percentage of change of explained deviance calculated through matrix permutations for each variable (1000×). Except for geographical distance and precipitation seasonality, significant variables for reptiles are not the same as for amphibians. a Highly correlated with Elevation.
b Highly correlated with Minimum winter temperatures.

TA B L E 3
Best GDM model for reptiles and amphibians and the respective subsets composed only of the species endemic to the Cerrado (End-) This combination of climatic variables is related to the fact that in amphibians, the permeability of the skin constrains responses to temperature, as most amphibians do not actively thermoregulate (Wells, 2007). Therefore, extremes of (especially lower) temperature in highly seasonal areas may demand additional physiological and behavioural adaptations in amphibians (Cunningham et al., 2016),

F I G U R E 2
Clustering of phylogenetic turnover for reptiles and amphibians of the Cerrado savannas. Sites with the same colour share more phylogenetic branches (= less turnover). Each terminal of the UPGMA dendrograms indicates a distinct site. (a and b) Clustering of the observed pairwise phylogenetic turnover for well-sampled sites (at least 10 species, and two Cerrado endemics). (c and d) Clustering of the GDM predictions of phylogenetic turnover with geographical distance, precipitation and temperature variables for reptiles (Table 2), plus soil and elevation variables for amphibians. This figure demonstrates strikingly contrasting patterns of phylogenetic turnover between reptiles and amphibians in both the observed and in the predicted patterns.  Valdujo, Silvano, et al., 2012) are related to relief roughness or soil texture and climatic conditions respectively.
The greater phylogenetic turnover with geographical distance for all reptiles indicates the importance of the four biomes surrounding the Cerrado in the assemblage of reptiles (i.e. sites at opposite borders will present higher phylogenetic turnover). Potential geographical barriers in the Cerrado have possibly limited dispersal of most non-endemics across the entire region . These barriers include tabletop mountains and highlands separated by peripheral depressions and river valleys (Silva, 1997), which were also suggested to have a strong impact on Cerrado endemic reptiles and amphibians by fragmenting ancestral ranges (Azevedo et al., 2016).
However, for the dataset with the endemics only, temperature and precipitation variables were more important (Figure 1, Question 2), and the importance of geographical distance alone was minor (Fig.   S5). In general, temperature is considered one of the most important determinants of the distribution of reptile species (Huey, 1982).
Distinct thermoregulatory strategies and temperature optima can vary across different phylogenetic levels in this group (Diele-Viegas et al., 2018). This may explain the importance of minimum winter temperatures for endemic reptiles (collinear with precipitation of warmest quarter, Figure 1), although this pattern could also be the result of the concentration of endemic reptiles across the larger plateaus in the south and south-east Cerrado , generally more prone to lower temperatures (da Silva et al., 2008).
Such differences in the importance of geography and environment between the two reptile datasets highlight the importance of intrinsic environmental conditions within the region for Cerrado endemics (Costa et al., 2007), when compared to the more generalist or widely ranged non-endemics.
In contrast to the pattern found for all reptiles, phylogenetic turnover does not increase considerably with distance for amphibi-  (Valdujo, Silvano, et al., 2012). This environment is mostly restricted to highlands, which are not uniformly distributed in the Cerrado, but concentrated on the central and south-eastern areas (Simon & Proença, 2000; see Fig. S1). The proximity with drier regions (Caatinga and Chaco, north-eastern and south-western borders respectively) further indicates the importance of nearby biomes in structuring amphibian assemblages. In fact, the clustering of phylogenetic turnover in amphibians was much closer to these borders than in reptiles, suggesting that the environmental similarity with neighbouring biomes is of greater importance in driving these patterns, as proposed by Valdujo, Carnaval, et al., (2012).
Contrary to our expectations, soil variables were not important in explaining phylogenetic turnover for all reptiles (Question 2), although they were slightly more important for endemics, and significantly important for endemic amphibians. As discussed earlier, this pattern is potentially due to the formation and duration of breeding sites, and also because some amphibians hide in the soil during the dry season (e.g. Nomura et al., 2009). One possible explanation is that soil variables, such as the proportion of sand and clay, vary and drive phylogenetic turnover at an even finer geographical scale than that of the available data (Figueiredo et al., 2018). It is also possible that these variables only drive phylogenetic turnover at shallow phylogenetic levels, which could be masked when analysing phylogenetic turnover in groups with such a deep evolutionary history as reptiles and amphibians (Rosauer et al., 2014). For example, several of the endemic lizard and snake species adapted to sand soils (e.g. in the Jalapão region) have closely related species occurring in other Cerrado environments (e.g. Kentropyx, Cnemidophorus, Apostolepis and Bachia) (Nogueira et al., 2010). The low level of niche conservatism and high lability of traits related to the use of soil in reptiles are patterns observed worldwide, with convergent adaptations of lizards and snakes to specific sandy or rocky substrates (Rodrigues, 1993).
At broader geographical scales, phylogenetic and taxonomic turnover are consistently congruent for most ectothermic tetrapods, especially in areas with steep relief or in biome transitions (Buckley & Jetz, 2008;Saladin et al., 2019). In contrast to the in six equally distributed regions (Valdujo, Carnaval, et al., 2012).
In contrast, our results, based on phylogenetic turnover, indicate a more homogenous phylogenetic composition of amphibian assemblages with breaks in the phylogenetic composition concentrated in the east. Once again, our results reinforce the importance of adding phylogenetic information when mapping biogeographical patterns. Historical events since the middle Miocene (climate, topographic and vegetational dynamics) are correlated to specieslevel patterns of richness and endemism (Azevedo et al., 2016;Costa et al., 2007), and the diversification at genus and species levels for many Cerrado lineages (e.g. Giugliano et al., 2013;Prado et al., 2012;Vasconcellos et al., 2020). In contrast, more ancient, large-scale distributional patterns of reptiles and amphibians in the continent-for example, larger diversity of frogs in tropical forests (Buckley & Jetz, 2007;Roll et al., 2017) -may have ultimately shaped their contrasting patterns of regionalization within the Cerrado, when considering the deep phylogenetic relationships among lineages as we found here.
Phylogenetic turnover among biological communities is driven by the history of speciation and extinction and by intrinsic clade characteristics related to dispersal limitation and niche conservatism (Daru et al., 2017). In contrast to predictions based on species diversity (Azevedo et al., 2016), our results indicate that diversity is unevenly distributed for different groups of organisms when considering their phylogenetic relationships (Question This suggests that conservation efforts need to be designed and implemented taking into consideration as many organismal groups as possible.

DATA AVA I L A B I L I T Y S TAT E M E N T
The raw distribution data used in this work were published in Azevedo et al., 2016