Pleistocene climate change and the formation of regional species pools

Abstract

Although the description of bioregions dates back to the origin of biogeography, the processes originating their associated species pools have been seldom studied. Ancient historical events are thought to play a fundamental role in configuring bioregions, but the effects of more recent events on these regional biotas are largely unknown. We used a network approach to identify regional and sub-regional faunas of European Carabus beetles and developed a method to explore the relative contribution of dispersal barriers, niche similarities and phylogenetic history on their configuration. We identify a transition zone matching the limit of the ice sheets at the Last Glacial Maximum. While southern species pools are mostly separated by dispersal barriers, in the north species are mainly sorted by their environmental niches. Strikingly, most phylogenetic structuration of Carabus faunas occurred during the Pleistocene. Our results show how extreme recent historical events—such as Pleistocene climate cooling, rather than just deep-time evolutionary processes—can profoundly modify the composition and structure of geographical species pools.

1. Introduction

Naturalists have long been captivated by the geographical distribution of world biotas. Rooted in the seminal ideas of Alexander von Humboldt, this fascination has promoted a long-term research agenda aiming to delineate biogeographic regions according to their faunas and floras (e.g. [1–3]). Besides this, the large-scale eco-evolutionary processes that shape regional biotas are known to influence ecological and evolutionary dynamics at finer scales [4]. For instance, regional species pools can modulate local diversity patterns [5,6], the structure and functioning of ecosystems [7] or coevolutionary processes [8]. However, the processes that have configured regional biotas have been seldom studied, despite their fundamental importance, and most explanations of their origin and dynamics remain largely narrative [9].

Perhaps the earliest speculations about the formation of regional species pools date back to the nineteenth century (reviewed in [10]). At that time, some authors already started to emphasize historical influences as key elements determining the configuration of plant and animal regions. For instance, when Wallace [1] proposed his ground-breaking zoogeographic regions, he argued that while the distribution of ancient linages such as genera and families would probably reflect major geological and climatic changes spanning the early and mid-Cenozoic, species distributions would be more influenced by recent events such as Pleistocene glaciations [3]. These recent events could have promoted many additions and subtractions of species to regional faunas through dispersal and diversification processes. Indeed, increasing evidence suggests that Pleistocene glacial–interglacial dynamics may have driven population extinctions (e.g. [11]), allopatric speciation in glacial refugia (e.g. [12]) and post-glacial recolonization events (e.g. [13,14]). Besides shaping phylogeographic patterns (e.g. [15,16]), all these processes are likely to be underpinning diversity patterns for many taxa, particularly in the Holarctic (e.g. [17–19]). However, whether the signature of Pleistocene glaciations scales up to the configuration of regional biotas remains largely unknown.

Historical contingencies should act over the intricate interplay between ecological (i.e. environmental tolerances and dispersal) and evolutionary (i.e. diversification and adaptation to new habitats) processes underpinning the composition of regional species pools. On the one hand, niche-based processes may determine the composition of regional species pools [20], mainly throughout their effects on species distribution ranges [21]. These processes integrate responses to abiotic conditions and to local and regional biotic environments [22], which may ultimately lead to the appearance of distinct regional communities in areas of contrasted environmental conditions. Although species with similar environmental tolerances can coexist in regions of similar climate, their dispersal may be constrained by geographical barriers, which may lead to divergent species pools under similar environmental conditions. Finally, evolutionary processes also constrain all these mechanisms. For instance, environmentally driven regions may be expected if the occupancy of new areas is constrained by niche conservatism [18], which should also lead to pools of evolutionarily related species (i.e. niche conservatism generating phylogenetically related species pools; figure 1a). This pattern, however, can also be the output of biogeographical processes. Indeed, diversification of lineages within regions separated by strong dispersal barriers may also lead to phylogenetically related pools (i.e. geographically driven niche conservatism; figure 1a [8,23]). Historical contingencies may contribute to the configuration of regional pools by modifying the balance between these processes. For example, the accumulation of species due to diversification may be the predominant driver of regional species pools during climatically stable periods [24]. Yet, regions with a greater influence of climatic fluctuations such as Pleistocene glaciations may harbour pools of species mostly shaped by the joint effects of current climate and post-glacial colonization dynamics [25], as well as by species' competition during these colonization processes [26], thus eroding the signature of geographically structured diversification processes.

Figure 1.

Identifying the factors configuring regional faunas. (a) Four out of seven (see (b)) hypothetical processes that may configure regional faunas. Dotted lines depict different regions while colours correspond with different climates. The tips of the phylogeny point to the distribution of the species. (b) Workflow and potential results: (1) hypothetical results of modularity analysis over the occurrence network; (2) similarity matrix of occurrence into modules; (3) pairwise matrix of environmental niche similarities; phylogenetic distances and topographical connectivity; and (4) hypothetical results and interpretations of a partial matrix regression on species occurrence similarities as a function of niche similarities, phylogenetic distances and connectivity.

In this study, we aim to disentangle the relative importance of the processes that may contribute to the formation of regional species pools, using European Carabus (Coleoptera: Carabidae) as a model lineage. Carabus is a species-rich ground beetle genus of great popularity due to the beautiful jewel-like appearance of some species [27]. In general, Carabus species are flightless nocturnal predators of snails, earthworms and caterpillars. They hold hydrophilic adaptations and are typically associated with deciduous forests [28]. Previous evidence suggests that the richness of species from this genus in Europe is determined to a large extent by both current climatic and habitat conditions and glacial–interglacial dynamics [19]. This makes European Carabus an ideal case study to evaluate the joint effects of evolutionary, ecological and historical contingency processes as drivers of regional species pools.

Specifically, we use data on the distribution and evolutionary relationships of Carabus species, along with network and phylogenetic analyses, to evaluate six hypotheses. First, given the presumed low dispersal capacity of the species from this genus [27], we hypothesize that (H1) European Carabus species pools are mainly shaped by the main orographic barriers of the continent, but also that (H2) glacial–interglacial dynamics have led to strong differentiation between northern and southern regional species pools. If this differentiation exists, northern European Carabus faunas will be comprised of species that colonized newly vacant habitats after the retreat of the ice sheet, and hence (H3) their regional distribution will be mostly determined by the current climate. By contrast, (H4) southern faunas will be mainly shaped by the joint influence of diversification events and dispersal limitations, due to the combined effect of higher climatic stability (e.g. climatic refugia) and a more complex orography (Alps, Pyrenees, Carpathians). Therefore, (H5) species forming northern regional pools will exhibit comparatively lower levels of regional endemicity, whereas those forming southern regional pools will show comparatively higher levels of regional specificity. Finally, according to Wallace [1], the advance and retreat of the ice sheets during the Pleistocene should have determined the spatial distribution of lineages (e.g. [18,19]), eroding the effects of the former distribution of the main Carabus lineages. Therefore, (H6) we expect a temporal signal coincident with the Pleistocene in the phylogenetic structure of Carabus faunas, and no effect of deep-time events on the current geographical distribution of these lineages.

2. Material and methods

(a). Rationale and structure of the analyses

Exploring the determinants of regional faunas requires jointly analysing ecological, evolutionary and historical factors. We did so through three consecutive steps (figure 1b). First, we identified distinct regional species pools within Europe by using a network community detection algorithm. From this analysis, we derived a species pairwise similarity matrix of occurrence into different modules, each one representing different regions. Second, we assessed the relative importance of environmental, spatial and evolutionary determinants of such similarity. To do so, we constructed four pairwise matrices to describe ecological, topographical and evolutionary relationships among species; namely, a matrix of climatic-niche similarity, a matrix of habitat similarity, a matrix of spatial connectivity among distributional ranges and a phylogenetic distance matrix. Then, we used generalized partial matrix regressions to model the similarity in species occurrences as a function of these four matrices (figure 1b). We used this workflow to explore the factors involved in the configuration of Carabus faunas at both regional and sub-regional scales (i.e. through analysing species co-occurrence patterns across regions and within sub-regions, respectively). Beyond scaling effects, the use of these two scales allowed us to explore for differences among regions, and thus to delve into our hypotheses H3 and H4. Finally, we also applied ancestral range estimation analysis in order to identify the time period from which ancestral areas are estimated with less uncertainty. By doing so, we aimed to detect important historical periods contributing to the regional organization of Carabus lineages.

The interpretation of the joint and independent effects of explanatory matrices can shed light on the different processes configuring regional faunas (figure 1a). Thus, if niche similarities (i.e. represented by the climatic and habitat similarity matrices) and phylogenetic distances altogether explained the regional co-occurrence of species, then this could be interpreted as indicative of constrained niche evolution or a tendency to resemble ancestral niches in shaping regional faunas (see niche conservatism in figure 1a). However, if spatial connectivity also accounted for part of this co-occurrence, this would indicate that this niche conservatism pattern can be caused by geographical constraints (spatially driven niche conservatism in figure 1a). Further, the effects of niche similarities and spatial connectivity alone (i.e. without phylogenetic signal) can be most likely the consequence of a convergence of climatic niches due to geographical isolation, whereas the effects of connectivity and phylogeny would be indicative of a primacy of intra-regional speciation driven by geographical barriers. Niche similarities alone would point to unconstrained niche evolution shaping regional faunas (Niche convergence in figure 1a), while phylogeny alone would indicate a primacy of geographically unconstrained intra-regional speciation events. Finally, the accumulation of species in past climatic refugia (the so-called cul-de-sac effect) or a primacy of vicariant speciation events could lead to the existence of independent effects of connectivity on regional co-occurrence (vicariance or cul-de-sac in figure 1a).

(b). Identification of regional species pools

We used network community detection analysis to identify Carabus regional species pools in Europe. We first generated a bipartite network where species and grid cells constitute two disjoint sets of nodes that are connected according to the presence of species in grid cells (e.g. [8]). Species presence data comes from expert-based range maps of all Carabus species inhabiting Europe (n = 131; [27]) overlaid into a 100 × 100 km equal-area grid based on the LAEA pan-European grid system (currently available at https://www.eea.europa.eu/data-and-maps/data/eea-reference-grids-2; see [19] for details). Then, we conducted a modularity analysis using the index proposed by Barber [29] and the Louvain algorithm [30] as implemented in the Matlab function ‘Gen Louvain’, (available at http://netwiki.amath.unc.edu [31]). This analysis identified groups of grid cells, each group sharing Carabus species mainly distributed within its cells (i.e. regions and their associated faunas). The Louvain algorithm was run 500 times, and the network partition showing the highest modularity value was retained. This optimal solution was used to conduct all subsequent analyses, although all the solutions were quantitatively and qualitatively similar (electronic supplementary material, appendix S1). We evaluated the statistical significance of the modules by comparing their associated modularity to a null distribution of values (n = 100) where the original presence–absence matrix was randomized using the independent swap algorithm, a fixed–fixed null model implemented in the R package ‘picante’ [32]. Finally, to detect potential submodules (i.e. sub-regions) nested within modules, we derived a new bipartite network from each of the previously identified modules and applied the procedure described above in each case.

It is important to note that despite species and grid cells were assigned to just one module, they could also occur in other modules with different degrees of specificity. Thus, we calculated the degree of module specificity for each node (i.e. species and grid cells) as its number of links with nodes of its module divided by its total number of links. Higher module specificity would correspond to highly endemic species mainly distributed within its module, as well as to cells pertaining to well-defined regions; whereas lower module specificity would indicate widespread species and cells located in transition zones.

(c). Assessing the determinants of regional species pools

To disentangle the determinants of the current configuration of Carabus faunas in Europe, we first measured the regional co-occurrence similarity of species pairs based on the proportion of their ranges present in each module. For this, we used Schoener's index [33], which measures the proportion of overlap between pairs of species. The resultant occurrence pairwise similarity matrix was used as the dependent variable. Then, we generated four pairwise dis/similarity matrices used as explanatory variables. Two of them were used to account for environmental factors: a climatic-niche similarity matrix and a habitat similarity matrix. The remaining two considered geographical and evolutionary factors: a spatial connectivity matrix and a phylogenetic distance matrix.

(i). Climatic-niche similarity matrix

We characterized the climatic-niche of each Carabus species in the dataset following a similar approach as proposed by Broennimann et al. [34]. We selected six bioclimatic variables to account for the main water and energy aspects of climate—namely mean annual temperature, temperature of the warmest quarter, temperature of the driest quarter, total annual precipitation, total precipitation of the warmest quarter and total precipitation of the driest quarter, as well as altitudinal range to account for the effects of mesoclimatic gradients within each grid cell. These variables may be among the main determinants of the distribution of Carabus species diversity within Europe [19]. Bioclimatic variables were extracted from Worldclim (v. 1.4; available at http://www.worldclim.org [35]), whereas altitudinal data were derived from the 30-arcsecond digital elevation model GTOPO30 (available at https://lta.cr.usgs.gov/GTOPO30/). We conducted a principal component analysis on these variables to obtain a bidimensional climatic space defined by the two main axes that explained 81.4% of the variance (electronic supplementary material, figure S1). We divided this climatic space into a 100 × 100 grid system and computed the frequency of occurrence of each species in each grid cell. Finally, we measured the species overlap in the gridded space using Schoener's index (see above).

(ii). Habitat similarity matrix

The distribution of Carabus species may also be shaped by forest preferences [27]. Accordingly, we used 10 vegetation categories derived from MODIS Land Cover at 5 min resolution (evergreen broadleaf forest, deciduous needle-leaf forest, deciduous broadleaf forest, mixed forest, closed shrublands, open shrublands, woody savannahs, savannahs and grasslands; available at http://glcf.umd.edu/data/lc [36]). For each species, we computed the proportion of each category overlaying its range. With this, we computed pairwise similarities in the preference for different vegetation types using Schoener's index (see above).

(iii). Spatial connectivity matrix

To evaluate the potential influence of geophysical barriers to dispersal on the current distribution of Carabus species, the study area was divided into 1 km² grid cells. For each cell, we obtained both mean elevation—from GTOPO30 digital elevation model, and the presence of waterbodies—from Natural Earth database (available at http://www.naturalearthdata.com). Then, we derived a spatial conductance network, where grid cells were nodes linked to the eight neighbouring cells. Link weights represent the probability of transit from cell i to cell j. This probability was inversely proportional to the difference in elevation between cells, weighted by the presence of water bodies:

$$P(i\leftrightarrow j)=\left(1-\frac{\mathrm{Rise}}{\mathrm{Rise}+\mathrm{Run}}\right)\times P_{\mathrm{water}},$$

where Rise is the difference in altitude, Run the great-circle distance between the centroids of the cells and P_water the probability to transit across water bodies. We used a P_water of 0.5 to build up the conductance network, although different values provided similar results (electronic supplementary material, table S1). Then, the matrix of connectivity between the centroids of all pairs of cells at 100 × 100 km was calculated as the accumulated cost for a random walker to commute from cell i to cell j and back to cell i across the conductance network [37]. Finally, the spatial connectivity between each pair of species' distributional ranges in the dataset was estimated as the average distance among all grid cells at 100 × 100 km within the range of each species. All analyses were conducted using the ‘gdistance’ R package [37].

(iv). Phylogenetic distance matrix

To unravel the evolutionary history of Carabus lineages and assess the potential importance of evolutionary processes in determining the formation of species pools, we reconstructed a time-calibrated phylogeny including the 89 species for which we found available DNA information on 10 markers (electronic supplementary material, appendix S2 and tables S2 and S3). The final dataset was concatenated in 5603 basepairs following a total-evidence approach [38] and used to conduct Bayesian phylogenetic inference with BEAST v. 2.4.6 software ([39]; electronic supplementary material, appendix S2). Molecular dating was done with two different scenarios. In the first, the crown age of Carabus was set at 17.3 Ma, according to Deuve et al.'s [28] molecular dating. In the second, the origin of the group was set at 25.16 Ma, according to Andújar et al. [40] (electronic supplementary material, appendix S2). To account for topological and time-calibration uncertainties, we used 100 posterior phylogenies for each molecular dating scenario. In addition, we used taxonomic information and phylogenetic uncertainty methods [41] to place species lacking molecular information into the phylogeny (electronic supplementary material, appendix S2). Thus, we derived 100 different phylogenetic hypotheses from each Bayesian posterior phylogeny by randomly inserting missing species within their most derived consensus clade based on taxonomic knowledge. In total, we generated 10 000 phylogenetic hypotheses for each dating scenario, and randomly selected 1000 for subsequent analyses. Finally, we computed multiple patristic distances between species pairs using raw branch lengths and several branch-length transformations to accommodate different evolutionary models (electronic supplementary material, appendix S3). Patristic distance calculations and branch-length transformations were conducted using the R packages ‘ape’ and ‘geiger’ respectively.

We used generalized multiple regression on distance matrices and deviance partitioning to disentangle the relative importance of climatic niche, habitat preferences, dispersal barriers and evolutionary history in determining Carabus species pools in Europe. First, we conducted single regressions between the occurrence pairwise similarity matrix and each of the four explanatory matrices described above to seek for significant associations. Since the dependent variable is basically a proportion (i.e. the proportion of overlap), we set a binomial family for error distribution and a logit link function (see [8] for a similar approach). To assess for significance, we randomized the observed species per module matrix using the independent swap algorithm (see above) to derive 999 null occurrence similarity matrices. Then, we used simple regressions to relate each null similarity matrix with each one of the explanatory matrices. The relationship between an explanatory matrix and the observed species per module matrix was considered to be significant when explaining a higher proportion of the deviance than 99% of the regressions performed on the null matrices. In the case of phylogenetic distances, we repeated this procedure for each phylogenetic hypothesis and considered a relationship to be significant when 99% of phylogenetic hypotheses explained higher deviance than 99% of corresponding null matrices. Finally, we retained those variables that showed significant relationships and conducted deviance partitioning to explore for patterns of covariation among niche similarities (i.e. climatic and habitat similarity matrices), dispersal barriers and phylogenetic history. The explained deviance was computed as McFadden's pseudo-R², and we followed the partitioning approach presented by Legendre [42] (electronic supplementary material, appendix S4). We first conducted the analyses for the co-occurrence into modules (i.e. regions) and submodules (i.e. sub-regions) at a European scale. Then, we conducted independent analyses for the co-occurrence into submodules of the species grouped in each module. Regression analyses were conducted in R [43].

(d). Ancestral range estimation

To assess whether deep historical signals were eroded by Pleistocene glaciations, we used probabilistic models of geographical range evolution. We used the dispersal–extinction–cladogenesis (DEC) model of range evolution [44] implemented in the R package BioGeoBears [45]. Species ranges were coded as present/absent in each module. If the Pleistocene glacial periods had important effects on the lineage distributions, it could be expected that ancestral range estimations will increase in accuracy around the Pleistocene. To explore this, we evaluated the existence of changes in the relationship between node age and the marginal probability of the single most probable ancestral state at each internal node by fitting general additive mixed models (GAMMs), including the phylogenetic hypothesis as a random factor. We also used generalized linear mixed models (GLMMs) combined with piecewise regression to detect potential major breakpoints (i.e. temporal shifts) in the relationship between marginal probability and node age (electronic supplementary material, appendix S5). We used a binomial family and a loglink function to fit all models. General assumptions of probabilistic ancestral range estimation models may compromise subsequent interpretations [46]. Hence, we conducted additional analyses to provide further evidence on the temporal signal of Pleistocene glaciations in the phylogenetic structuration of Carabus faunas (electronic supplementary material, appendix S5). Finally, to explore for different Pleistocene effects across regions, we also calculated the probability that a phylogenetic node has all its descendant species within a given region, independently for nodes occurring either before and after the beginning of the Pleistocene (2.59 Mya; herein pre-Pleistocene and post-Pleistocene nodes).

All analyses were carried out in R [43], using the function bam of the package mcvg for GAMM [47] and the package Lme4 for GLMM analyses [48].

3. Results

(a). Identification of regional faunas

The Carabus occurrence network was significantly modular (M = 0.385, p = 0.01), dividing Europe into seven modules that group zoogeographically distinct regions with their associated faunas (i.e. different regional species pools; figure 2a; electronic supplementary material, figure S2). Furthermore, all modules but the southernmost one (SM) showed significant sub-modular structure, presenting a decrease in modularity with latitude (mean M = 0.316, ranging from 0.154 to 0.468; all p < 0.05; electronic supplementary material, table S4 and appendix S6). The transition zones between regions were clearly associated with geographical barriers such as the Pyrenees, the Alps, the Carpathian and the Ural Mountains, as well as the Turkish Straits System (figure 2b), in agreement with our first hypothesis (H1). Interestingly, we also identified a west-to-east transitional belt between southern and northern regions that closely followed the southern limits of the ice sheet at the Last Glacial Maximum (LGM). This transitional zone further suggested a link between the configuration of Carabus regional faunas and Pleistocene glacial conditions, supporting our hypothesis H2.

Figure 2.

Transition zones between regions were associated with geophysical accidents and the border of the ice sheet at LGM. European Carabus regions found by the network community detection analysis. (a) Geographical location of modules (i.e. regions) and submodules (i.e. sub-regions). Module labels correspond with: SW: southwestern module; SM: southernmost module; SE: southeastern module; CW: central-western module; CE: central-eastern module; NW: northwestern module; and NE: northeastern module. (b) Values of module specificity per grid cell; green colours (i.e. cells with low specificity) identify transition zones. The dotted black line corresponds with the southern limit of the ice sheet at LGM (extracted from [49]). The blue line depicts the breakpoint where the temperature-Carabus richness relationship changes, as found in Calatayud et al. [19].

(b). Correlates of regional co-occurrence

Matrix regressions showed that deviance of species co-occurrences across regions, across sub-regions and within each region was significantly explained (p < 0.01), primarily by spatial connectivity, and secondarily by environmental niche similarity, except for northern regions (i.e. northwestern and northeastern modules; NW and NE respectively; figure 3; electronic supplementary material, tables S1 and S5). By contrast, relationships with evolutionary relatedness were non-significant in most instances (electronic supplementary material, table S6). Moreover, in the cases where we found a significant effect (i.e. when analysing co-occurrences in modules, submodules and central-western (CW) module), the deviance explained by phylogenetic distances was rather low and mostly overlapped with that explained by spatial connectivity (figure 3). Comparing both scales, environmental niche similarity explained more deviance across sub-regions than across regions, whereas spatial connectivity did the reverse (figure 3). Comparing explained deviances between regions, the primacy of environmental niche similarity (mostly climate, see electronic supplementary material, table S5) in the northern ones (NW and NE) is consistent with the notion that northern regional pools are geographically sorted by current climate (our hypothesis H3), whereas the importance shown by spatial connectivity in the remaining regions is consistent with the more complex orography of central and southern Europe (consistent with hypothesis H4).

Figure 3.

Regional co-occurrence was mostly explained by geographical connectivity and environmental niche similarities. Results of the partial generalized matrix regression of similarity in regional co-occurrence as a function of environmental niche similarity (climate and habitat, E), topographical connectivity (C) and phylogenetic distances (P). The first and second bars correspond with the models including occurrence similarities among all modules and submodules, respectively. The remaining bars correspond with the models where the similarities in submodule occurrence were analysed independently for the species of each module. We used average results derived from phylogenies time-calibrated following Deuve et al. [28]. Both calibration scenarios provided similar results (see electronic supplementary material, table S6). Modules are labelled according to figure 2a.

(c). Ancestral range estimation

Both phylogenetic datasets (i.e. alternative calibration scenarios) yielded similar qualitative and quantitative results (electronic supplementary material, appendix S5). Thus, we only present here ancestral range estimations based on Deuve's et al. [28] calibration. GAMM results showed that node marginal probability of the most probable state increased towards younger nodes (p < 0.01, explained deviance = 7.49%, figure 4a). However, this increase showed a steep increment coinciding with the Pleistocene. Indeed, piecewise regression revealed that the relationship between marginal state probability and node age changed at 1.51 Mya (median value; with 45th and 55th percentiles at 1.24 and 1.89 Mya, respectively; p < 0.01; figure 4a; electronic supplementary material, figure S3), suggesting that most of the phylogenetic structuration of Carabus faunas began around the Pleistocene, supporting hypothesis H6. Complementary analyses yield similar results (electronic supplementary material, appendix S5).

Figure 4.

Temporal coincidence between the Pleistocene and the phylogenetic structuration of Carabus regions. (a) GAMM predictions of the marginal probability of the most probable state as a function of node age. The dashed red lines correspond with the confidence interval at 95%. The dotted black line represents the median of the breakpoint found by piecewise GLM regressions. The boxplot at the bottom represents the 45th and 55th percentile breakpoint values, whereas the whiskers depict the 25th and 75th percentiles. (b) Probability of finding a phylogenetic node having all descendant species grouped in the same region for pre-Pleistocene (pink) and post- Pleistocene nodes (blue). This probability was calculated for all modules both jointly (‘All’ in x-axis) and independently (labelled according to figure 2a in x-axis).

In agreement with these results, we found that the probability of finding phylogenetic nodes having all descendants belonging to the same region was higher for post-Pleistocene nodes (median at 0.21; 25th and 75th percentiles at 0.20 and 0.23, respectively; figure 4b; electronic supplementary material, figure S4) than for pre-Pleistocene ones (median at 0.10; 25th and 75th percentiles at 0.08 and 0.11). These low probabilities are congruent with the lack of phylogenetic signal in module co-occurrence previously found. Interestingly, the probabilities found for both types of nodes (i.e. pre- and post-Pleistocene) were higher in southern and central regions than in northern ones (figure 4b). This suggests that regions not covered by ice during the LGM can still reflect some old historical legacies (as shown by the higher pre-Pleistocene node probability) while accumulating some related lineages that diversify during and after the Pleistocene (as indicated by the higher post-Pleistocene node probability).

4. Discussion

More than 140 years ago, Wallace [1] foresaw that the influence of Pleistocene glaciations on the distribution of diversity had been strong enough to erode the imprint of previous events. Our results support Wallace's thoughts, showing a remarkable coincidence between the distribution of the ice sheets at the LGM and the current configuration and evolutionary structure of European Carabus faunas.

The first line of evidence supporting this idea comes from the close spatial relationship between the southern limits of the ice sheet at LGM and the transition zone separating the southern and northern regions. This border also coincides with the line identified by Calatayud et al. [19] where the relationship between Carabus species richness and current climate changes (figure 2). Thus, it seems that Pleistocene climate changes not only shaped phylogeographic [11–16,50] and species richness patterns [17–19], but that ice ages also left a strong imprint on the geographical structure of species composition at a regional scale. Accordingly, the species from the northwestern region (NW) show the lowest level of endemism (electronic supplementary material, figure S5), as expected for regional faunas composed of species that have recently colonized the north of Europe from southern glacial refugia [19] (our hypothesis H5). In fact, although these species show large distribution ranges in different parts of southern Europe, their ranges only overlap near the northern Carpathian Mountains (electronic supplementary material, figure S6). This area was a glacial refugium for a large and taxonomically diverse array of northern European species (e.g. [16] and references therein), including Carabus [51]. Additionally, the decrease in modularity values with latitude also points to a lesser geographical structure of northern assemblages, which can be interpreted as the result of a post-glacial colonization, together with less geographical complexity in some areas.

Besides the Pleistocene effects in the definition and geographical structure of regional species pools, we also found evidence of the imprint of this geological period on the processes configuring the distribution of Carabus faunas. The general strong relationship between regional patterns of co-occurrence and both niche similarities and spatial connectivity shows that co-occurring species tend to have similar realized environmental niches and that also tend to be geographically constrained by the same dispersal barriers. This latter result was expected given the—presumed—low dispersal capacity of Carabus species [27], which is likely to be behind the spatial coincidence of module transition zones and geographical barriers. Perhaps more unexpected is the weak effect of phylogenetic distances despite the strong relationship between regional co-occurrence and niche similarities. This implies that geographical barriers rather than climatic-niche conservatism have restricted species distributions even within regions of similar climate. These results also point to Carabus niche evolution being, to some extent, evolutionarily unconstrained, which is congruent with the generally high adaptive capacity of insects (e.g. [52]).

Whatever the origin of the relationship between species occurrence and environmental conditions, what is certainly true is that its strength changes between regions. These changes follow a latitudinal gradient in the importance of environmental niche similarities (figure 3). The occurrence into sub-regions is more strongly related to the similarity in the realized niche in the north than in the south. This might be a direct consequence of the effects of post-glacial colonization, where formerly glaciated areas show a clear sorting of species due to its environmental preferences. On the contrary, in southern regions, species are expected to have had more time to diversify and sort geographically by other factors besides climate [18]. Our findings corroborated this idea since we found strong effects of dispersal barriers in these areas. Moreover, although we did not find a significant phylogenetic signal in the sub-regional co-occurrence over these regions (except for the CW module), our analyses revealed that they hold a small but still larger number of related species compared to northern ones, supporting that more stable regions are more prone to accumulate related species.

Despite these related species of southern regions, we found a generalized lack of phylogenetic structuration of Carabus faunas. This can be the outcome of relatively recent speciation events due to vicariance and/or a cul-de-sac effect [53]. The former would imply the formation of dispersal barriers promoting the geographical split of many lineages and subsequent allopatric speciation [54]. Yet, the geophysical accidents that can be associated with the limits of the Carabus regions largely predate the origin of the genus [28,55]. On the other hand, a generalized dispersion into climatic refugia, together with a subsequent stagnancy within them (i.e. a cul-de-sac effect) may also produce the observed mixing of unrelated lineages into regions. Although it is difficult to distinguish between both processes, the latter seems more plausible, with southern regions accumulating unrelated species while acting as glacial refugia, and northern ones being recolonized by unrelated species with similar environmental niches and/or simply higher dispersal capacity [17].

Supporting the Pleistocene signature, our results showed a temporal coincidence between this geological period and the phylogenetic structuration of Carabus faunas. This result was consistent regardless of the different approaches used and across the different time-calibration scenarios. This robust temporal coincidence supports that the current regional organization of Carabus lineages is rooted at the Pleistocene, which also explains the general lack of phylogenetic structure. Our results partially contrast with ancestral range estimations for clades inhabiting areas that were never glaciated, where more ancient signals were found in the spatial sorting of lineages [56–59]. These previous findings are, nonetheless, congruent with the higher probability of holding related Carabus species of southern and more stable European regions. In sum, these results suggest that the repeated advances and retreats of ice sheets and glacial conditions that characterize the European Pleistocene produced repeated cycles of retreat to southern regions and advance towards the north of Carabus species, a hustle-and-bustle process that ultimately led to the observed mixing of unrelated lineages, with few related species inhabiting in less affected regions.

To summarize, our results provide solid arguments in favour of the importance of Pleistocene glaciations along with geographical barriers and niche-based processes in structuring the regional faunas of European Carabus. On the one hand, this group's faunas are primarily delimited by the location of the southern limit of the ice sheet at LGM, which separates two large regions that differ not only in species composition but also in the processes underlying the spatial organization of these species. On the other hand, the phylogenetic structure of these faunas coincides with the beginning of the Pleistocene. This implies that the geographical distribution of species and lineages is profoundly shaped by past climates. Moreover, our results also suggest that ecological [7,60] and evolutionary mechanisms [8,61] that rely on processes operating at regional scales can be profoundly affected by the history of Earth's climates. Hence, the study of these historical events may be essential to unravel both large and local scale diversity patterns.

Data accessibility

This article has no additional data.

Authors' contributions

J.C. and J.H. conceived the ideas. J.C. designed the study with the contributions of all authors. J.C., R.M.-V., M.L. and J.L.H. analysed the data. All the authors discussed results. J.C. wrote the paper with contributions of all authors.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the Spanish projects SCARPO (CGL2011-29317), UNITED (CGL2016-78070-P) and BIOREGIONS 2.0 (CGL2017-86926-P), funded respectively by the Spanish Ministry of Economy, Industry and Competitiveness (MINECO), the Spanish Ministry for Science, Innovation and Universities (MICINN) and AEI/FEDER, UE. J.C. was supported by an FPU-fellowship of the Spanish Ministry of Education (FPU12/00575), and M.L. and J.L.H. by MINECO FPI and Juan de la Cierva-Incorporación (ref. IJCI-2015-23618) grants, respectively.