What drives the evolution of body size in ectotherms? A global analysis across the amphibian tree of life

Aim: The emergence of large-scale patterns of animal body size is the central expectation of a wide range of (macro)ecological and evolutionary hypotheses. The drivers shaping these patterns include climate (e.g. Bergmann's rule), resource availability (e.g. ‘resource rule’), biogeographic settings and niche partitioning (e.g. adaptive radiation). However, these hypotheses often


| INTRODUC TI ON
Body size influences most ecological, evolutionary and demographic processes underlying patterns of animal biodiversity from local to global scales (Calder, 1984;Peters, 1983;Smith & Lyons, 2013). On the one hand, the evolution of traits involved in ecological, sexual, reproductive and physiological performance is generally mediated by body size (Andersson, 1994;Brown et al., 2004). On the other hand, species predisposition to demographic collapses that lead to extinctions is often associated with their body sizes (Cardillo et al., 2005;Dirzo et al., 2014;Pincheira-Donoso & Hodgson, 2018;Ripple et al., 2017). Therefore, given the influence of body size on most components of an organism's fitness (Andersson, 1994;Roff, 2002), elucidating the drivers behind the evolution of this complex trait has been a central focus of research for centuries (Peters, 1983;Smith & Lyons, 2013).
The evolution of body size diversity is also influenced by selection from ecological resources. Central to macroecology, the 'resource rule' posits that constraints on upper body size limits relax as resource abundance increases, promoting larger size in more productive environments (Geist, 1987;Huston & Wolverton, 2011;McNab, 2010;Rosenzweig, 1968;Yom-Tov & Geffen, 2006). The underlying mechanisms can involve increasing energy investment into body growth without the constraining effects of trade-offs, or selection for smaller body mass to reduce resource requirements as productivity declines (McNab, 2010). Similarly, the 'seasonality hypothesis' suggests that accumulation of nutritional reserves confers fitness advantages, promoting larger body mass towards more seasonal regions (Boyce, 1979;Calder, 1984;Lindsey, 1966).
Insularity has also played a central role in elucidating the evolution of body size diversity (Losos, 2009;MacArthur & Wilson, 1967).
The unique ecological features that prevail on islands (e.g. isolation, low predation and character release) have been hypothesized to drive species gigantism or dwarfism in size relative to their mainland close relatives (Benítez-López et al., 2021;Foster, 1964;Meiri, 2008;Van Valen, 1973). A stream of large-scale tests have both supported (Benítez-López et al., 2021;Clegg & Owens, 2002;Lomolino, 2005;Meiri, 2007;Raia et al., 2003) and challenged (Meiri et al., 2004(Meiri et al., , 2006 this prediction. Collectively, theories and evidence have significantly advanced our understanding of the drivers and implications of body size diversity. However, (i) support for some predictions (e.g. Bergmann's rule) is inherently weak to justify their validity; (ii) some hypotheses (e.g. resource rule vs water conservation hypothesis) make opposite discuss our findings in the context of the emerging hypothesis that climate change can drive body size shifts.

K E Y W O R D S
adaptive radiation, amphibians, Bergmann's rule, body size, ectotherms, island rule, resource availability predictions based on the same sources of selection; (iii) most studies test one or a few of these hypotheses separately, without integrating them; and (iv) large-scale studies are available for some taxa such as lizards and birds (e.g. Benítez-López et al., 2021;Meiri, 2008;Olson et al., 2009;Slavenko et al., 2019) but not for others, for example amphibians, which lack comprehensive large-scale studies despite existing efforts (e.g. Adams & Church, 2008;Benítez-López et al., 2021;Slavenko & Meiri, 2015). Here, based on the most comprehensive global dataset of body size for amphibians compiled to date (spanning >87% of their known species diversity), we address the role that a range of key hypotheses (Table 1) attribute to climatic factors, ecology and biogeographic settings as the drivers shaping large-scale patterns of body size diversity in these organisms. Amphibians offer ideal model systems for testing these predictions at macroecological scale given their remarkable diversity in body size; their distribution across the full worldwide spectrum of environments (except the poles); the contrasting body plans across their three extant orders (Anura or frogs, Caudata or salamanders, and Gymnophiona or caecilians); and their unparalleled range of genome size diversity relative to tetrapods as a whole (Liedtke et al., 2018;Pincheira-Donoso et al., 2023). In addition, amphibians are nature's most threatened animals (IUCN, 2021)-a phenomenon that has directly or indirectly been linked to their diversity in body size Pincheira-Donoso & Hodgson, 2018;Ripple et al., 2017).

| Body size data
We created a global dataset on maximum body size (largest available record for a species regardless of sex) for 7270 amphibian species spanning all three living orders (Anura, Caudata and Gymnophiona). These data were collected from the primary literature, including articles and books, and from direct observation of specimens both in museums and in the field. Given that snout-vent length (SVL) is the most common proxy for body size in anurans and salamanders (Amado et al., 2021;Pincheira-Donoso, Harvey, Grattarola, et al., 2021;Wells, 2007), whereas total body length is used for caecilians (Pincheira-Donoso et al., 2019), we used species body mass as our measure for body size to make analyses comparable. Actual measures of body mass are considerably scarcer than body length measures both in the literature and in museum specimens. Therefore, we employed the approach presented by Pough (1980) for conversion of maximum SVL and maximum total length available for each species into body mass using orderspecific allometric formulas (Pincheira-Donoso & Hodgson, 2018;Ripple et al., 2017). This dataset (Table S1) is part of the Global Amphibian Biodiversity Project (GABiP) initiative (www.amphi bianb iodiv ersity.org).

| Climatic, geographic and ecological predictors
To address the hypotheses that spatial gradients in climate drive the evolution of body size, we obtained extent of geographic occurrence range maps for all amphibian species from the International Union for Conservation of Nature (IUCN) archive (www.iucnr edlist.org) for species for which body size data are available (Table S1). From the distributional data, we then created a dataset spanning a range of variables that have widely been as- ArcGIS v.10.2. Finally, we added geographic data to the climatic database from the same sources described above. These data consist of elevational, latitudinal and insular distributions. For the elevational data, we obtained the minimum and maximum elevational records available for each species, from which we calculated both elevational midpoint (between maximum and minimum elevation) and elevational range (the range of metres above sea level contained between maximum and minimum known elevations). Latitude was extracted as the centroid for each species from the IUCN distributional maps described above. Insular data were obtained from the IUCN archive and from the literature (the full dataset is available in Table S1).
To test the hypotheses on the role of habitat structures as drivers of body size evolution, from the same sources described above, we created a dataset consisting of two variables that capture habitat use by species as 'perching' sites ('microhabitats') and microhabitat structures for egg deposition ('nesting sites'). Each species was assigned to one of four perching microhabitat structures: aquatic (species that depend on direct contact with water bodies, including strictly aquatic species and species whose subsistence depends on permanent contact with water bodies), terrestrial (ground-dwellers that do not depend on permanent contact with water), vegetation (bush dwellers and arboreal species) and fossorial (species that, except for the breeding seasons, have underground lifestyles). For nesting site data, we assigned each species to one of five categories depending on whether parents lay their eggs in water (e.g. streams, lakes and seasonal pools, but not on vegetation, e.g. pitcher plants), the ground (terrestrial sites separated from water), burrows (enclosed nests in the ground, or in caves), on vegetation (bushes and trees, including those that use small accumulations of water within flowers and pitchers, or between leaves) or in the body of the parents in gastric brooder or skin-brooder species (Table S1). To test the role of diel activity for influencing body size, we also collated data from the same sources described above, categorizing species as either diurnal, nocturnal, cathemeral or crepuscular. Species for which different literature sources provided conflicting data on the use of habitat structures were removed from the analyses.

| Statistical analyses
Phylogenetic MCMC generalized linear mixed models, MCMCglmm hereafter (Hadfield, 2021), run through R 4.1.0 (R Development Core Team, 2021) were used to predict body mass from environmental variables using a cross-species approach. Further, the simplicity of Bayesian regressions considering phylogenetic nonindependence compared with phylogenetic path analyses is better suited to highlight the patterns and hypotheses we are testing (Table 1). Comparatively, an alternative method such as an assembly-based approach would not be as insightful for our study, given we are testing >7000 species of amphibians across the entire globe which is ideally suited to cross-species phylogenetic methods In analyses, this phylogeny was implemented as a random effect where the inverse of the sum matrix of phylogenetic correlation was calculated (Garamszegi, 2014). Environmental predictors were assessed for collinearity using a conservative threshold of 0.65 to exclude variables from multivariate models owing to collinearity. None of the environmental predictors were collinear, so were all included in our models. A multiple regression model of the four climatic predictors (temperature, precipitation, temperature seasonality and precipitation seasonality) and net primary production (NPP) was performed to predict body mass. We then performed separate models with latitude as a predictor of body size variation.
The reason behind the use of latitude separately is that, although it is widely used as a key predictor of geographic biodiversity clines in macroecology (Womack & Bell, 2020), latitude is in fact a 'catchall' proxy for most environmental factors that vary geographically, rather than an environmental source of selection that could be implicated as the mechanism driving the evolution of those gradients . Therefore, latitude was not incorporated into the same models with the environmental predictors that covary strongly with it. Measures of elevation (elevation range and elevation midpoint) were co-linear, so multiple univariate models were implemented to predict body mass with these elevation measures as predictors. All predictors were  (Garamszegi, 2014).
All models were run for a minimum of 20,000 iterations and 5000 discarded as burn-in. Iterations and burns were increased if the effective sample size of any model was below 1000 (Hadfield, 2021).
Trace plots of the fixed and random effects were visually assessed for convergence (Hadfield, 2021).
Phylogenetic ANOVAs using the 'aov.phylo' function from the 'geiger' package (Harmon et al., 2008) were performed to statistically analyse differences in body mass between groups (i.e. insularity, microhabitat, activity and nesting site). The phylogenetic signal, estimated based on Pagel's lambda (Pagel, 1999), of group effects on body size was calculated using the 'phylosig' function from the 'phytools' package (Revell, 2012). The phylogenetic tree shown in Figure 2 was plotted using the 'ggtree' and 'ggtreeextra' packages implemented in R (Xu et al., 2021;Yu, 2020;Yu et al., 2017Yu et al., , 2018.

| RE SULTS
Body size diversity spans several orders of magnitude across extant amphibians (Figures 1 and 2), ranging from 8 mm (0.03 g) of total body length in the frog Paedophryne amauensis to >1600 mm (1593 g) in the caecilian Caecilia guntheri, and 1500 mm (10,793 g) in the giant salamander Andrias japonicus (  Figure 1).

| Climatic drivers of body size
Climatic analyses consistently reject Bergmann's rule. Geographic gradients in environmental temperature failed to predict variations of body size in any direction for each order separately (Table 2). In contrast, our analyses reveal that body size in anurans decreases predictably towards environments with increasing seasonality in precipitation and with decreasing thermal seasonality (Table 2; Figure 2). In caecilians, body size decreases towards environments with increasing levels of annual precipitation, but seasonality plays no role (Table 2; Figure 2). All our climatic predictors failed to significantly explain body size variation in salamanders, with seasonality in precipitation having a borderline effect in the opposite direction observed for anurans (Table 2; Figure 2). The resource rule was consistently rejected by all our analyses as NPP failed to predict body size variation in any direction for each group separately (Table 2; Figure 2). All analyses incorporating amphibians of all three orders combined yielded results that consistently matched the findings observed in anurans ( Figure 2).

| Geographic and ecological drivers of body size
Our models consistently failed to identify a role for latitude as a predictor of spatial variation in body size among each of the three amphibian orders separately (Table 2), revealing the lack of a latitudinal gradient of body size (Figure 3). Our analyses addressing the influence of elevation on body size variation revealed a significant increase in caecilian body mass towards higher elevations (Table 2).
No such relationship was observed among anurans or salamanders, although a weak, nonsignificant tendency for anuran body size to decrease with increasing elevation was observed (Table 2). Elevational range, in contrast, showed a strong positive relationship with body size in anurans and caecilians, but not in salamanders (Table 2).
Analyses testing the role of ecological predictors consistently failed to identify any effect of diel activity, microhabitat use or nesting site choice as drivers of body size variation across all three amphibian orders (Table 3; Figure S1).

F I G U R E 1 Probability density distribution of log(body mass)
for each of the three orders separately, as shown by the colour scheme.

| DISCUSS ION
Our study provides a global-scale empirical overview addressing a range of classic hypotheses on the drivers behind the evolution of animal body size, using amphibians-the vertebrate class for which evidence remains more scattered-as model systems. By using amphibians as a model system, our study covers these classic questions  (Silver et al., 1999;Wells, 2007).
Given that ~45% of oxygen has been recorded to be taken through the skin in some caecilians (Smits & Flanagin, 1994), a reduction in size (and thus, a corresponding increase in surface area relative to mass) is likely to increase overall efficiency of O 2 consumption in hypoxic environments (Wells, 2007). Elevation was found to be an equally strong predictor of body size gradients. Whereas increasing body mass is associated with increasing tolerance to a wider elevational range in both anurans and caecilians, body mass in caecilians increases in species occurring at higher elevations. Remarkably, none of these climatic or elevational factors showed any relationship with body size diversity in salamanders.
Finally, our analyses failed to identify an effect of ecological factors (diel activity and use of habitat structures) and insularity on the predictable evolution of body size patterns. Combined, these findings suggest that climatic gradients are the dominant drivers of body size patterns at large scales. However, even these dominant effects are inconsistent, with different factors driving body size clines in different clades, or not driving them at all. It remains possible that studies at smaller spatial scales may identify factors that operate at regional scale or within specific clades. Our study, however, has F I G U R E 2 Relationship between body mass in amphibians and a range of environmental and geographic predictors. The plot to the left shows the distribution of temperature seasonality (inner ring of light grey bars), mean annual precipitation (second ring of medium grey bars), precipitation seasonality (third ring of medium-dark grey bars) and log-body mass (outer ring of dark grey bars) over a phylogenetic tree for 5554 amphibian species (orders shown with the pictures placed over the external blue ring). The node tips represent the mainland (green), insular (blue) or mainland+insular (red) distribution of the species. To the right, model coefficients from the MCMCglmm phylogenetic analyses for tested predictors of log(body mass) variation across amphibians. Coloured dots represent the mean coefficient, while whiskers show the 95% credible intervals (CI). Models were run for all amphibians combined and for each order separately. Predictors are considered significant when the 95% CI does not cross zero (dashed grey line).
aimed to identify the underlying factors that drive body size evolution more 'universally'.

| The climatic macroecology of body size: Why are body size rules often exceptions?
The study of macroecological patterns of body size across the animal tree of life has been guided for over 150 years by Bergmann's rule (Bergmann, 1847;James, 1970), which simultaneously combines the statuses of being both a landmark and one of the most discredited ecogeographic predictions (Adams & Church, 2008; Olalla- Tarraga TA B L E 2 Results of phylogenetic analyses of body mass variation as a function of multiple climatic, environmental and geographic (latitude and elevation) predictors. Amphibian orders are tested separately (results of models combining all amphibian orders were qualitatively identical to results from anuran models). An effect is considered to exist if the 95% credible intervals (CI) do not cross zero. Significant results in boldface. Our study, the largest in both geographic and taxonomic scale conducted in these organisms, rejects the role of mean environmental temperature predicted by both Bergmann's rule and its alternative, the 'heat balance' hypothesis. Therefore, the findings we present add to a history of conflicting evidence, where both the underlying cause and the predicted outcome of Bergmann's rule are regularly discredited. In fact, a similar rationale extends to most ecogeographic rules on body size, for which support is conflicting and conditioned by multiple exceptions, such that they may apply depending on geographic region, lineage or size itself (Krizmanic et al., 2005;Olalla-Tarraga & Rodriguez, 2007;Pincheira-Donoso et al., 2008).
Why do predictions from ecogeographic rules about body size fail to find empirical support so often? A straightforward answer is the unrealistic logic of such predictions. Body size, like other complex phenotypes, is shaped by the additive effects of a wide array of selection pressures operating simultaneously on ecological (e.g. predation avoidance, climatic resistance and habitat use), sexual (e.g. mate competition) and reproductive (e.g. fecundity) performance (Andersson, 1994;Calder, 1984;Losos, 2009;Lynch & Walsh, 1998;Peters, 1983;Pincheira-Donoso & Hunt, 2017;Roff, 2002;Rosenthal, 2017;Smith & Lyons, 2013), in addition to the effects of shared ancestry (as demonstrated by the consistently high phylogenetic signal in all our models; see  , 2008). Therefore, body size is often the outcome of trade-offs between the demands of multiple selection pressures, thus making the expectation that large-scale patterns of body size variation across organisms (e.g. endotherms and ectotherms) will be neatly explained by one single factor unrealistic.
Rather than aiming to make predictions about specific spatial clines of body size evolution as a function of single selection pressures (e.g.
larger size with decreasing temperature), the elucidation of mechanisms underlying macroecological processes may benefit from a focus on how patterns of body size are shaped by the constraints that environmental pressures impose to set (maximum) thresholds beyond which the trait becomes inviable. For example, as climates get progressively colder, larger ectothermic body sizes are likely to be less viable (e.g. Ashton & Feldman, 2003;Pincheira-Donoso et al., 2008). As expected, no truly large ectotherms exist towards temperate regions (e.g. Pough et al., 2015;Sindaco & Jeremcenko, 2008;Wells, 2007). However, these constraints are likely to be relaxed, rather than reversed, towards lower latitudes as temperatures become warmer. As a result, tropical latitudes are more likely to accommodate a higher variance in body sizes across species. Therefore, while colder climates are expected to set a 'maximum viable body size' at any given latitudinal point (regardless of whether sexual or fecundity selection favours body masses larger than the maximum viable), the relaxation of this threshold towards tropical latitudes is more likely to allow the evolution of a wider variation of body sizes, from very small to very large. This is, again, the case-tropical climates host the smallest as well as the largest sized amphibians.

| A 'third universal response' to climate change?
The influence of the temperature-body size 'rules' led to the emergence of the hypothesis that anthropogenic global warming should

TA B L E 3
Results of phylogenetic analyses of variance of body mass variation as a function of multiple ecological predictors. Amphibian orders are tested separately (results of models combining all amphibian orders were qualitatively identical to results from anuran models). See Figure S1 for violin plots from these analyses.
drive widespread changes (mostly reductions) in animal body size (Gardner et al., 2011;Sheridan & Bickford, 2011). This phenomenon has been deemed a potential 'third universal response' to global warming alongside phenological and geographic range shifts (Daufresne et al., 2009;Gardner et al., 2011). The lack of consistent evidence for the temperature-body size relationship (e.g. Riemer et al., 2018;Slavenko et al., 2019) has led to questioning the significance of this warming hypothesis (Riemer et al., 2018). However, rising temperatures are associated with severe alterations in other climatic components, particularly changes in precipitation (Fischer & Knutti, 2016;Krauss, 2021;Masson-Delmotte et al., 2021), and in its seasonality (Easterling et al., 2000;Feng et al., 2013;Ning et al., 1999). In concordance with our analyses, previous studies reveal a widespread role for precipitation as a driver of body size variation in amphibians, where body mass in both anurans and caecilians decreases towards environments with increasing rainfall (Olalla-Tarraga et al., 2009;Pincheira-Donoso et al., 2019). Therefore, based on the rationale of the warming hypothesis (Gardner et al., 2011), strong natural selection on body mass may arise from rapid alterations in precipitation regimes, leading to rapid size shifts. The directions in body size shifts are expected to be spatially heterogeneous given the geographic heterogeneity in the predicted changes in rainfall regimes (Chadwick et al., 2016;Feng et al., 2013;Marvel & Bonfils, 2013;Masson-Delmotte et al., 2021). Yet, the tendencies for extreme climatic events to aggravate under climate change, with intensification of rainfall, lead to the expectation that average body size in amphibians is likely to decrease. In addition to the wider impacts predicted to result from widespread reductions in animal body sizes (Daufresne et al., 2009;Gardner et al., 2011;Sheridan & Bickford, 2011), emerging evidence has shown how smaller size in amphibians is linked to higher extinction risk (Pincheira-Donoso & Hodgson, 2018) given the detrimental demographic effects of lower reproductive output .
Collectively, we argue that this 'third universal response' should remain as a next frontier for empirical interrogation.

ACK N O WLE D G E M ENTS
DPD is indebted to Queen's University Belfast and the School of Biological Sciences for a joint start-up grant that played a key role in the completion of this study. JVJ and LEBG are grateful for funding from the Department for Economy (UK). CF and JG are supported by PhD funding from a Natural Environment Research Council (NERC, UK) DTP programme and LPH by funding provided by Nottingham Trent University.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare no conflict.

DATA AVA I L A B I L I T Y S TAT E M E N T
All the datasets used in this study are available as supplementary material to the main article, and will be made open-access at the GABiP initiative's repository at http://www.amphi bianb iodiv ersity.
org upon publication.