Trait evolution and historical biogeography shape assemblages of annual killifish

Different species assemblages of annual killifish possess replicated body size distributions yet have unique sets of species in each area of endemism. Here, we use models of trait evolution and historical biogeography to discover how size variation originated and has been restructured.


| INTRODUC TI ON
The importance of evolutionary and historical biogeographic processes in community assembly is increasingly acknowledged (Gerhold, Carlucci, Procheş, & Prinzing, 2018), but they remain relatively understudied when compared to local and recent ecological processes (Mittelbach & Schemske, 2015;Warren, Cardillo, Rosauer, & Bolnick, 2014). In community phylogenetics (Webb, Ackerly, McPeek, & Donoghue, 2002), phylogenetic reconstructions are used to characterize assemblages and to determine the ecological processes at work in them. An assemblage that is phenotypically overdispersed is usually inferred to be structured by competition while phenotypically clustered assemblages are thought be shaped by environmental filtering.
However, in either case the assemblages may or may not be phylogenetically overdispersed (Emerson & Gillespie, 2008). Moreover, phylogenetic clustering methods are susceptible to interpreting historical effects on current geographic distributions as evidence for ecological and evolutionary processes (Warren et al., 2014).
To avoid this misinterpretation, one can model the process of trait evolution instead of interpreting patterns in observed traits.
Trait evolution is an essential process in the formation of species assemblages (Kraft, Cornwell, Webb, & Ackerley, 2007;Webb et al., 2002). Often, simple evolutionary models where trait differences increase with phylogenetic distance are assumed. In reality, traits can evolve in response to the different selection regimes encountered during the history of a species, and this heterogeneity is ignored under a Brownian motion (BM) model (Cavender-Bares, Kozak, Fine, & Kembel, 2009). Trait evolution models that accommodate this determine whether species traits in an assemblage tend towards different phenotypic optima, namely by identifying selection regime shifts across a phylogeny (Butler & King, 2004). By reconstructing the history of selection regimes we can improve our understanding of the role trait evolution plays in diversification and coexistence (Oke, Rolshausen, LeBlond, & Hendry, 2017).
Historical biogeography is also an intrinsic component in the development of species assemblages, but is often neglected in empirical studies (Warren et al., 2014). The geographic and ecological shifts associated with divergence events in a phylogeny provide insights into factors shaping the diversification of the lineage. In order to incorporate the geography of speciation into phylogenetic studies, ancestral range estimation (ARE) models have often been employed. Under these models, species are assigned to predefined areas that are used to estimate the frequencies of different types of 'cladogenetic events'. In this context, founder-event speciation (i.e. long-distance dispersal followed by isolation and speciation; Matzke, 2014) lies at one end of the geographic continuum and 'within-area speciation' at the other. If we are to better understand processes that form species assemblages, more detailed joint investigations of trait evolution and biogeography along phylogenetic trees are needed.
We investigated a genus of freshwater fishes (Austrolebias; Costa, 1998) with naturally replicated assemblages-i.e. different assemblages with comparable size variation but different sets of related species (Figure 1)-to better understand how trait evolution and historical biogeography have shaped them. These annual killifishes live in seasonal freshwater pools and wetlands on the South American grasslands and floodplains. Eggs persist within the soil, going through diapause to survive the dry season (Wourms, 1972).
Austrolebias are principally found in the seasonal ponds distributed throughout basins of the La Plata and Paraguay Rivers, the Negro River drainage of the Uruguay river basin and the Patos-Merin lagoon drainage system (Figure 1). These four areas of endemism (Costa, 2010) each contain a distinct assemblage and Costa (2010) hypothesized that marine transgressions during the middle and late Miocene (c. 15-11 million years) caused alternating periods of isolation and connection, driving dispersal and vicariance.
The largest Austrolebias species can grow to more than 150 mm in length while the smallest can be less than 25 mm when mature (Costa, 2006). Differently sized species are known to locally coexist in each area (Figure 1). Body size influences resource use in freshwater systems (Woodward & Hildrew, 2002) and stronger competition is expected between those species with similar body sizes because they use the same resources (MacArthur & Levins, 1967). Diet has already been shown to be linked to body size in Austrolebias (Laufer et al., 2009;Ortiz & Arim, 2016) and the largest species are known to prey upon the smaller species (Costa, 2009) indicating that differently sized species occupy distinct niches in a pond. Indeed, an ecological study within a single area found that local community structure was determined more by body size than species identity (Canavero, Hernández, Zarucki, & Arim, 2014). Considering and reconstructing body size evolution is therefore key to understanding how Austrolebias assemblages developed and will help to uncover how ecological interactions, such as competition and predation, have affected assemblages.
Austrolebias species may be adapted to different seasonal environments on the South American continent, which could affect the ability of a species to colonize a new area and therefore how species assemblages are formed. We refer to quantitative representations of the environment in which a species exists as its environmental niche. We applied a commonly used heuristic and studied changes in environmental niche using the same methods as for phenotypic traits (Münkemüller, Boucher, Thuiller, & Lavergne, 2015).
In this study we use models of trait evolution, environmental niches and historical biogeography together to disentangle the processes that led to the replicated body size distributions across Austrolebias assemblages. Niches were determined by values of environmental variables at sampling locations. Smaller scale processes, such as within-pond niche segregation (Loureiro et al., 2015), are not considered as, to date, no such data exist at the scale of this study. We tested whether inferred divergence events were associated with shifts in environmental niche traits or body size towards a new optimal value of a species relative to its ancestor. We expect that environmental niche shifts could be associated with dispersal events between areas, or might occur with dispersal within a large, heterogeneous area of endemism. To determine whether shifts coincided with migration events between areas of endemism or occurred within an area, we estimated ancestral ranges in Austrolebias.
The combination of modelling shifts in body size and environmental niche traits with ancestral range estimation may shed light on how major trait variation arose within Austrolebias and spread to compose the current species size distributions in each area of endemism.

| Phylogenetic inference
Sequences representing 63 Austrolebias individuals and two outgroups were used to build nDNA and mtDNA-based trees and a tree F I G U R E 1 A map of eastern South America depicting the four main areas of endemism in Austrolebias, inferred by drawing shapes around sampling locales for species in each region. The most northern region, in purple, is the area west of the River Paraguay (W). The basin of the La Plata river and its delta are in green (L). The drainage of the Negro River, part of the River Uruguay basin is represented in blue (N). The Patos-Merin Lagoon region is highlighted in red (P). The species used in this study are shown on the right side of this figure, grouped by area. Austrolebias apaii and Austrolebias cinereus are not shown. Body size measurements are depicted as bars alongside species names. Local patterns of species co-occurrence are shown in the pairwise table to the left of species names. Coloured dots indicate that a record of a pond with both species was found or stated in Loureiro et al. (2015). Images of fishes are approximately to scale. All fish images are of males except Austrolebias vazferreirai [Colour figure can be viewed at wileyonlinelibrary.com] constructed using nDNA and assuming a multispecies coalescent (*BEAST). Divergence times were estimated using secondary calibrations from a tree of the order Cyprinodontiformes (Helmstetter et al., 2016). Phylogenetic tree inference was conducted using BEAST2 v2.4.3 (Bouckaert et al., 2014) and led to three separate maximum clade credibility trees. Further details on our tree inference approach, including divergence time calibration, can be found in Appendix S2.
For downstream analyses we converted mtDNA and nDNA trees to species trees with the GLASS algorithm (Liu, Yu, & Pearl, 2009) using the speciesTree function in R library 'ape' (Paradis, Claude, & Strimmer, 2004). This created a pair of species trees we could use along the *BEAST species tree for further analyses.

| Phylogenetic clustering
We tested for the presence of phylogenetic clustering or overdispersion in the assemblages of each of the four areas of endemism.
Initially we calculated phylogenetic diversity (PD) for all assemblages. We then calculated mean pairwise phylogenetic distance (MPD) and mean nearest taxon distance (MNTD; Webb et al., 2002), simulating null distributions for these statistics per assemblage (ses. mntd and ses.mpd in R package 'picante'; Kembel et al., 2010). Using the same package and the 'MicEco' package (Russel, 2019) we also calculated phylogenetic beta diversity with the functions ses.comdist & ses.comdistnt (999 randomisations) to interpret the phylogenetic distances between assemblages.

| Environmental niches
We aggregated coordinates of ponds where Austrolebias have been observed using primary publications-the Global Biodiversity Information Facility (https://www.gbif.org/) and several other sources (see Table S1.3, updated until March 2017). The number of sites used for each species can be found in Table S1.4. We characterized the environmental niche of each species using variables available from curated databases and related to climate, to the height of points above the ground, soil composition and river basin characteristics at the known pond sites (see Table S1.5). Per species, we averaged standardized environmental variables over the locations where it was found and principal component analysis (PCA) transformed the outcomes into a set of traits that were not correlated (see supplementary methods for further details).

| Body size data
To prevent the comparative analyses from becoming dependent on sampled age variation and environmental variability, which cannot be standardized, we decided to focus on the maximum size possible for a species. Growth is indeterminate in fish and age information is often not available. Taking an average of size measurements and assigning it to each species would lead to underestimates because of the inclusion of younger fish, rather than our goal of a size that reflects maximum fitness for a species. Body size data were therefore obtained by taking the largest known field measurements of adult male standard length (SL) for each species from the literature and our own field records (full dataset in Table S1.6).

| Estimation of ancestral ranges
Ancestral range estimation (ARE) was conducted using the R package 'BioGeoBEARS' (Matzke, 2014). 'BioGeoBEARS' fits the Dispersal-Extinction-Cladogenesis (DEC) model (Ree & Smith, 2008) and models similar to the BayArea (Bayesian Inference of Historical Biogeography for Discrete Areas) model (Landis, Matzke, Moore, & Huelsenbeck, 2013) and the Dispersal-Vicariance (referred to as DIVA) model (Ronquist, 1997), with and without founder-event speciation (indicated by +j). We used areas of endemism (Costa, 2010) for a broad scale representation of species range. These were: the western region of the Paraguay River basin (Western Paraguay or W), the lower La Plata River basin and the middle-lower Uruguay River basin (La Plata or L), the Negro River drainage of the Uruguay River basin and the upper/ middle parts of the Jacuí, Santa Maria, Jahuarão and Quaraí river drainages (Negro River or N) and the Patos-Merin lagoon system including the southern coastal plains (Patos Lagoon or P; Figure 1). The middle section of the Iguaçu River basin is home to a single species, Austrolebias carvalhoi, for which we do not have data, so it was not included.
We ran separate sets of analyses where the maximum number of areas composing ancestral ranges was restricted to two (maximum observed) or four (maximum possible). We ran two additional sets of analyses: first, an analysis where ranges were restricted to include only adjacent areas. Second, we tested the hypothesis of Costa (2010)

| Selection regime shifts for size and environmental niche shifts
We used SURFACE (Ingram & Mahler, 2013) and l1OU (Khabbazian, Kriebel, Rohe, & Ané, 2016) to fit OU models with one or several optima and apply adapted information criterion (IC) in stepwise model comparisons. We identified selection regime shifts in body size and environmental niche shifts and then tested whether shifts converged towards the same optimum. Both methods are known to overestimate the number of shifts when relying on IC alone (Bastide, Ané, Robin, & Mariadassou, 2018;Khabbazian et al., 2016). We therefore performed simulations to test if we could reject null models of no regime shifts: an OU model with just a single optimum or a BM model (Cooper, Thomas, Venditti, Meade, & Freckleton, 2016). Univariate (size) or multivariate (environmental niche traits) BM and OU models with a single optimum were fitted to trees and trait data. Using a parametric bootstrap, datasets were resampled according to the fitted OU and BM models and the SURFACE/l1OU model selection routine was applied to each of these resampled datasets. We then compared the IC differences for the actual data with quantiles of the distributions of IC from the simulations. When actual IC differences were larger than the 95% quantile of IC change, we rejected the model on which the simulations were based.
As there were more environmental variables than species in this study, we restricted the amount of environmental niche traits for which we tried to detect shifts using a non-phylogenetic PCA on the niche traits was carried out. We then limited our analysis of regime shifts to the first two principal components (PCs) of environmental variables.
However, it is known that carrying out (phylogenetic) PC transformations can affect a subsequent analysis (Adams & Collyer, 2018;Bastide et al., 2018), so when we detected a regime shift in a PC, we tried to confirm the result for the niche trait contributing most to that PC. in 67% and 69% of nodes respectively. Support in the coalescent tree (Figure 2b) was markedly lower (38% of nodes with PP > 0.9, 58% of nodes with PP > 0.75). Estimates of the crown age were between 8.47 and 36.52 Ma across our three trees, similar to what was inferred (10-15 Ma) in a recent study of the Cynolebiini (Costa, Amorim, & Mattos, 2017). We consistently recovered three major groups within Austrolebias in all trees (Figure 2).

| Phylogenetic inference
Austrolebias bellottii and Austrolebias apaii were treated as the same species because A. apaii is a junior synonym of A. bellottii (García et al., 2012). We also decided to merge Austrolebias vazferreirai and Austrolebias cinereus for the comparative analyses in this paper, including the *BEAST analysis, in agreement with a recent proposal to consider A. vazferreirai a junior synonym of A. cinereus . Here, we refer to both as A. vazferreirai, as most individuals included were originally assigned to that species. Our mtDNA topology differed extensively from our nDNA tree (Figure 2), which had substantial effects on downstream inference. There are many reasons why the mtDNA history may not accurately reflect the species tree e.g. mitochondrial capture and lineage sorting (Ballard & Whitlock, 2004). Similarly, the poor support in the coalescent tree means that it is likely unreliable for inference. Therefore we focus on summarizing the results based on the nDNA tree.

| Phylogenetic clustering
Phylogenetic diversity was highest in Patos Lagoon and lowest in Western Paraguay (Table S1.7). We found consistently significant F I G U R E 2 Maximum clade credibility trees for Austrolebias from Bayesian analyses of (a) nDNA, (b) coalescent nDNA and (c) mtDNA. Black circles indicate a posterior probability (PP) from 0.90 to 1.00 and grey circles indicate a PP from 0.75 to 0.90. PP < 0.75 is depicted as a white node. Highlighted, colour-coded sections represent three major groups that are recovered in all trees. Relationships within groups vary among trees [Colour figure can be viewed at wileyonlinelibrary.com] phylogenetic clustering for the Negro River assemblage using MPD and MNTD (Table S1.7). For the other assemblages, results using the nDNA tree revealed no significant patterns. We examined the relatedness of assemblages using phylogenetic beta diversity. We found a trend for Western Paraguay and River Negro to be closer to each other than expected based on MPD (z = −1.42, p = 0.085) but this was not the case for MNTD (z = −0.93, p = 0.19).

| Ancestral range estimation
The best fitting ARE models were DIVA + j for all trees. The parameters and likelihoods of all models assessed are summarized in Table S1.8.
Models taking account of adjacency and marine transgressions did not significantly improve the likelihood (AICc difference < 2) when compared to the set of simpler models (Table S1.8). There is therefore no support for the hypothesis that the pattern of transgressions during the Middle and Late Miocene determined dispersal. Models with the jump dispersal parameter conferred significantly higher likelihood to the data (Table S1.9) and this event type was common (Figure 3; Figure S1.1), making up more than 50% of events in most cases ( Figure S1.2). One striking result was that Western Paraguay was colonized by Austrolebias in four (nDNA) independent instances (Figure 3). Estimated numbers of withinarea speciation events were particularly high in Patos Lagoon (an average of 4.6 events per simulation) and Negro River (2.9) while the other areas had averages below one. Vicariance events were rare, probably due to the rarity of current and ancestral ranges with more than one area.
The jump dispersal parameter has been recently criticized (Ree & Sanmartín, 2018 but see Klaus & Matzke, 2020) in regard to the overestimation of the number of jump dispersal events. Therefore, we also present results from the best fitting models without jump dispersal (DIVA in all cases; Figure S1.3) though the results generally provide the same interpretation.

| Size evolution
Our ancestral state reconstructions (Figure 4a) show substantial divergence in size. The four largest Austrolebias species form a single clade. Models consistently found a shift in selection optimum towards this clade (Group II in Figure 2) in all trees, just after the divergence of the smaller Austrolebias luteoflammulatus. (Figure 4a; Figure S1.4; SURFACE AICc decrease relative to single optimum model nDNA: 13.2, mtDNA:15.4, coalescent: 11.1). The SURFACE 5% quantile for the AICc decreases obtained from simulations of the BM null hypothesis is 11.8 for the nDNA tree. Likewise, for an OU model with a single optimum as null hypothesis, the 5% quantile equals 11.9.
The results obtained using l1OU and based on phylogenetic Bayesian Information Criteria (pBIC) were concordant (nDNA) both for the shifts ( Figure S1.5) and in having a pBIC difference exceeding the 5% quantile of the simulations. Therefore, a BM model or an OU model with a single optimum for size in the genus is implausible. We found no evidence for convergence between different selection regime shifts for size.

| Environmental niches
The first two PCs of the environmental niche variables explained 60% of species variation. Species phenotypically cluster to some degree by area of endemism (Figure 4b Simulations allowed us to reject the hypothesis of a single environmental niche optimum for the genus for the nDNA tree (5% quantile bivariate OU model: 10.6). The bivariate BM model cannot be rejected using SURFACE (5% quantile nDNA: 17.1), but the result on the actual data corresponds to the 10% quantile, which still indicates that a model with multiple optima is likely a better fit. When using l1OU, both the BM and and this might therefore be non-significant.
Differences among optima were always larger in PC2 and smaller in PC1. When we investigated whether shifts occurred for mean annual temperature (the variable with the largest weight on PC2) we found the same patterns. When comparing these shifts to ARE results we find that at least two and perhaps all four species from Western Paraguay were subject to shifts associated with their dispersal into the area (Figure 4b,c). Convergent shifts towards a new and shared environmental niche occurred independently when A.
vandenbergi and A. monstrosus colonized Western Paraguay. None of the regime shifts for environmental niche are associated with selection regime shifts for size.

| D ISCUSS I ON
Using models of trait evolution beyond BM and assessing them in the context provided by historical biogeography allowed us to identify that, despite the apparent similarity of Austrolebias assemblages, each followed different ecological and evolutionary trajectories.

| Evidence for phylogenetic clustering in one assemblage
Traditional community phylogenetic approaches concluded that only one of the four assemblages deviated from null expectations.
The Negro River assemblage was significantly more phylogenetically clustered than expected. However, there is no support for a selection regime shift increasing phenotypic variation in body size within this area. Environmental niches of species in Negro River were not strongly phenotypically clustered (Figure 4c) and we found no robust evidence for a niche regime shift here. Therefore an adaptive radiation or environmental filtering cannot be inferred from the phy-

| A single selection regime shift for body size
The processes acting in a single area of endemism have been critical for generating size variation across the genus. Our selection regime analyses consistently recovered a single body size shift towards the largest species. We found that this shift occurred in Patos Lagoon just after A. luteoflammulatus (Figures 3 and 4a) diverged. A. luteoflammulatus is a typical small species, Austrolebias cheradophilus is a large omnivore and Austrolebias prognathus is a very large piscivore. We speculate that a change in diet (see Laufer et al., 2009;Ortiz & Arim, 2016) may have been associated with the regime shift for size, and that predation by a piscivore facilitates coexistence of competitors.
A study of 48 species in the comparable African annual killifish genus Nothobranchius showed that most assemblages showed little variation in body size and only a single group of coexisting species exhibited size differences that may have been the result of ecological character displacement (Lambert, Reichard, & Pincheira-Donoso, 2019). In the La Plata and Patos Lagoon areas small, similar-sized Austrolebias species co-occur (Figure 1), along with the larger species. This is unexpected if competition between similar-sized species is strong among all annual fish in the same assemblage. Furthermore, Austrolebias can be found in ponds together with other small annual killifishes such as Trigonectes aplocheiloides, Papiliolebias bitteri and Neofundulus paraguayensis (Alonso, Calviño, Terán, & García, 2016) in Western Paraguay, and with Cynopoecilus melanotaenia (Canavero et al., 2014) in Patos Lagoon. These other species may have played a role in how species assemblages were constructed in some areas-by preventing the establishment of additional small Austrolebias species in Western Paraguay, for example.
We note that factors affecting differently sized species could similarly affect size variation within species. In phylogenetic comparative methods, within-species trait variation is seen as a source of uncertainty in species traits (Garamszegi, 2014) and thus difficult to account for in a meaningful way. Intraspecific variability in annual killifish is limited to some extent by synchronous hatching when ponds become wet (García, Loureiro, Machín, & Reichard, 2018), and interspecific competitors seem to cause little change to growth curves of a focal species (García, Smith, Machín, Loureiro, & Reichard, 2019).

| Convergent shifts for environmental niche
Environmental niches of species from Western Paraguay were markedly differentiated from the others. SURFACE analyses revealed that the colonization of Western Paraguay coincided twice with substantial and convergent shifts towards new environmental niches-potentially instrumental in the construction of the Western Paraguay assemblage. A characteristic that distinguishes Western Paraguay from other assemblages is that ponds fill in winter (Schalk, Montaña, Winemiller, & Fitzgerald, 2016) rather than in summer (Errea & Danulat, 2001) and this could explain the niche shifts in these seasonal species. Our results show that dispersal into particular areas can be mediated by environmental differences that must be overcome by colonizing species, regardless of size, thus having an effect on how assemblages are formed. However, as the environmental niche can vary without the need for evolutionary change we acknowledge that we cannot be certain that niche changes associated with dispersal involved selection regime shifts or adaptation.
A striking contrast to the environmental niche differentiation we demonstrated in South American killifish is the pattern in assemblages of five species of the genus Nothobranchius, which showed only moderate environmental niche differences (Reichard, Janáč, Polačik, Blažek, & Vrtílek, 2017). Altitude was found to be the most important factor explaining the distribution of different clades and species belonging to this African group. Environmental niche differences in Austrolebias stemmed from differences in soil characteristics, temperature and precipitation, indicating that different factors can govern geographic distributions of different annual fish clades despite their life cycle similarities.

| Repeated patterns are the result of different processes
The evolution of size variation can be explained by a specific process in three of the four assemblages. We observed a single selection regime shift in size, indicating that there was a single source of major size variation in the genus. Historical biogeographic approaches estimated that the speciation event that preceded this shift took place in the Patos Lagoon area. The species involved in this shift, (A. cheradophilus, A. luteoflammulatus and A. prognathus) currently coexist within the same ponds (Figure 1), which lends support to the idea that they coexisted during divergence. It is difficult to determine exactly what kind of interaction was behind size divergence but it may be that competition or cannibalism led to the emergence of large predator species (Van Dooren et al., 2018). Therefore, Patos Lagoon represents one extreme of size redistribution by dispersal because major size variation originated within the area. The Patos Lagoon area has apparently been less affected by marine incursions in the Miocene (Hubert & Renno, 2006), which may have limited immigration of species from other areas and thus prolonged the time interval where within-area speciation could fill available niches (Emerson & Gillespie, 2008).
Our best fitting ARE model (Figure 3) revealed two major events that were key in distributing size variation across assemblages. First, a jump dispersal (or vicarance; Figure S1.3) event around 6 Ma into La Plata led to Austrolebias elongatus, followed by an event into Western Paraguay which led to A. monstrosus. Costa (2010) suggested that marine transgressions during the Middle Miocene may have connected Western Paraguay, Negro River and Patos Lagoon. We did not find support for the hypothesis that these marine transgressions in particular determined the pattern of dispersal events in the genus.
Many important dispersal events occurred later, in particular arrivals in Western Paraguay and Negro River and the two dispersal events of piscivorous species above. This is in agreement with a recent dating of incursions in the Late Miocene (Del Rio et al., 2018).
A. vazferreirai (cinereus) is the largest species in the Negro River assemblage and is thought to be a generalist (Costa, 2006) rather than a piscivore. Negro River is the youngest assemblage, so there may not have been enough time for a large predator to colonize. Loureiro et al. (2018) discuss river capture events suggesting that these might have permitted dispersal between areas of endemism.
However, within the Negro River catchment, drainage rearrangements might also have created locally isolated populations and provoked the relatively recent within-area speciation events.

| Perspective
Understanding the processes that lead to current species assemblages is central in the study of ecology and evolution (Hutchinson, 1959), and our work adds to increasing knowledge of the importance of species interactions (Cornell & Lawton, 1992) and size differences (Simberloff & Boecklen, 1981). We find similarities between our results and those from other study groups. Leitão et al. (2016) showed, using fish, trees and birds, how rare species had a disproportionate role in the functional structure of assemblages. This may also be the case in Austrolebias, where large species are typically rarer than smaller species (Lanés, Gonçalves, & Volcan, 2014) and occupy different functional niches, therefore widening the functional richness of assemblages. Analyses of the co-occurrence of related species of hummingbirds highlighted the importance of multiple mechanisms including competition, dispersal and the environment, in shaping assemblages (Weinstein, Graham, & Parra, 2017). Although the methodological approaches were different, we also found that these mechanisms, as well as historical biogeography, were likely important in generating assemblages of Austrolebias.
The differences between assemblages suggest two specific follow-up studies. We have found niche trait selection regime shifts for species in the Western Paraguay area, but that does not demonstrate adaptation to the local environmental conditions. The species from the La Plata area only differ in a single niche PC and lab or transplant studies using species from these areas could provide support for the hypothesis that they are locally adapted. The Negro River assemblage has no local piscivore, because it has not arrived there yet or because of local conditions. There are sites without piscivores in the nearby Patos Lagoon area as well. Studies on selection pressures on body size in these areas can help determine which factors contribute to the establishment and maintenance of piscivores.

| CON CLUS ION
We have demonstrated that the simultaneous inspection of reconstructed trait evolution and historical biogeography can provide a useful perspective on ecological processes operating in assemblages. A single selection regime shift towards large body size was critical for generating the similar trait distributions observed in Austrolebias assemblages.
It occurred during a series of speciation events in a single assemblage.
Historical biogeographic reconstructions revealed that divergence in allopatry redistributed major size variation to other areas and that an environmentally distinct area was repeatedly colonised. Despite their apparent similarities, Austrolebias assemblage compositions and histories vary greatly. Our approach will be useful for other studies examining the processes that shape trait-structured assemblages.