Species invasions and the phylogenetic signal in geographical range size

Aim : Accelerating rates of anthropogenic introductions are leading to a dramatic restructuring of species distributions globally. However, the extent to which invasions alter the imprint of evolutionary history in species geographical ranges remains unclear. Here, we provide a global assessment of how the introduction, establishment and spread of alien species alters the phylogenetic signal in geographical range size using birds as a model system. Location : Global. Time period Contemporaneous. Taxa Birds. Methods : We compare the phylogenetic signal in alien range size with that of native distributions of species globally ( n = 9,993) We compare the phylogenetic signal in the native and alien ranges of established species with the native ranges of both introduced species n = and the entire global n We then use stochastic simulations to identify the stage(s) in the invasion pathway at which significant differences in phylogenetic signal arise, and the taxonomic and geographical biases causing these differences. Overall, our results demonstrate that the process of species invasion decouples variation in range size from species evolutionary ancestry but that this phenomenon is detectable only after accounting for biases in the history of species introductions. The null models identified the transition from the global pool to introduced species as a key stage in the invasion pathway gener-ating differences in phylogenetic patterns (i.e., null model 1). To test the causes of this finding, we compared the phylogenetic signal in the native ranges of introduced species with that expected under a suite of introduction scenarios, specifically testing the following hypotheses: Differences in λ arise from the tendency preferentially to introduce species: (a) With larger ranges ((cid:147)range size-dependent scenario(cid:148)), (b) from certain regions ((cid:147)region-de-pendent scenario(cid:148)) and (c) from particular clades ((cid:147)clade-depend-ent scenario(cid:148)). We parameterized these models by fitting a series of generalized linear mixed effects models with a binomial error structure, predicting whether species from the global avifauna ( n = 9,993) have been introduced (one) or not (zero). We variously included native range size as a fixed effect ((cid:147)range size-depend-ent scenario(cid:148)), and taxonomic family ((cid:147)clade-dependent scenario(cid:148), n = 194 families) and biogeographical realm ((cid:147)region-dependent scenario(cid:148), n = 9 realms) as random effects. We used the parameter estimates from these models to determine the probability of each species being selected for introduction in our stochastic models. As differences in λ may arise from a combination of factors, we additionally implemented three synthetic models combining the effects of range size and either avian family or biogeographi-ca(cid:1140) rea(cid:1140)m(cid:314)


| INTRODUCTION
The geographical distributions of species are shaped both by the current environment and by their evolutionary history (Gaston, 2003).
Over recent centuries, the introduction of species to novel locations beyond their natural geographical range has become an increasingly important force shaping the distribution of life on Earth (Seebens et al., 2017). This anthropogenic dismantling of biogeographical barriers is leading to the mixing of previously distinct evolutionary biotas, driving some species to extinction (Bellard, Cassey, & Blackburn, 2016) and fundamentally altering the phylogenetic structure of species assemb ages Capinha Ess Seebens Moser Pereira Nonetheless, how and to what extent the signature of evolutionary history in the size and location of species geographical distributions is altered by anthropogenic introductions remains unclear.
The size of the geographical range of species often exhibits a moderate but detectable phylogenetic signal, whereby range sizes are more similar amongst closely related species than amongst distant relatives (Abellán & Ribera, 2011;Gaston, 1998; Hunt, Roy, & Jab onski Machac Zrzav Storch Webb Gaston 2005). This phylogenetic signal in range size is often explained as a result of heritable intrinsic traits, such as dispersal ability or niche breadth, that determine the potential geographical range that a species can maintain (Jablonski, 1987). In addition, range sizes are expected to vary predictably with evolutionary relatedness because closely related species tend to arise in the same geographical region and are thus subject to the same environmental and biogeographica barriers Freck eton Jetz Whi e heritab e intrinsic traits and spatial location should promote phylogenetic signal in range size, random dispersal events and speciation may decouple variation in range size from evolutionary relatedness (Pigot, Phillimore, Owens, & Orme, 2010;Waldron, 2007). In particular, the isolation of peripheral populations during speciation can result in daughter species initially having very different range sizes (Pigot, Phillimore, et al., 2010). This asymmetry is expected to diminish over time, either as species with small geographical ranges go extinct or as species expand their distributions to reach the limits imposed by the environment and their intrinsic traits (Waldron, 2007). The finding that phylogenetic signal in range size is stronger than expected under null models of speciation (Waldron, 2007) supports the controversial idea that geographical range size might be a heritable property of species, with important implications for understanding the past and future dynamics of biodiversity (Jablonski, 1987).
How anthropogenic species invasions alter the phylogenetic signal in range size remains unclear because different aspects of the invasion process may have potentially contrasting effects on alien range size and how it varies with evolutionary relatedness. In the case of birds, there is evidence that the phylogenetic signal in alien range size may be substantially weaker than is typical of avian species in their native distributions (Dyer et al., 2016). Such a pattern may be expected, because the process of human introduction bears some resemblance (but see Wilson et al., 2016) to the natural process of speciation that tends to weaken phylogenetic signal in native range size. In particular, most invasions are initiated by small founding populations; therefore, a strong phylogenetic signal in alien range size might emerge only amongst species that were introduced long ago and that have had sufficient time to expand their distributions to the limits imposed by their intrinsic traits or the environment (Byers et al., 2015;Wilson et al., 2007). However, differences in species residence times or other aspects of the anthropogenic introduction process could also have a positive effect on phylogenetic signal depending on their relative phylogenetic patterning. The number of introduction attempts is known to be an important determinant of alien range size (Dyer et al., 2016;Lockwood, Cassey, & Blackburn, 2005;Williamson et al., 2009). To the extent that closely related species tend to share characteristics (e.g., life history traits) that make them more or less likely to be introduced (Allen, Street, & Capellini, 2017), a phylogenetic signal in introduction effort would act to promote a phylogenetic signal in the range size attained by alien species.
Alternatively, if differences in range size are primarily determined by geographica ocation Machac et a then the consequences of invasion will depend on the phylogenetic clustering of species introductions in space. If the introduction of closely related species occurs to widely scattered regions across the globe, then this would further decouple variation in alien range size from evolutionary ancestry regardless of the time available for their dispersal.
The examination of how invasions alter the phylogenetic signal in species range size is complicated by the fact that invasion is a multistage process, and differences in phylogenetic patterns can therefore arise through a variety of different routes (Blackburn et al., 2011;Figure a In particu ar a though differences in phy ogenetic signa between the native and alien ranges of established species must reflect processes operating post-invasion, broader comparisons of patterns in phylogenetic signal (e.g., between alien and nonintroduced native species) require accounting for the fact that introduced species and/or those that successfully establish represent a nonrandom subset of species in terms of their traits, evolutionary history or geographical origin (Allen et al., 2017;Duncan, Blackburn, & Sol, 2003; van K eunen et a Figure a For instance in the case of birds certain taxonomic families (e.g., pheasants, ducks) and geographical regions (e.g., Palearctic, Nearctic) have disproportionately been sources of introductions (Dyer, Cassey, et al., 2017), whereas successful establishment is known to depend on a variety of intrinsic life history traits (Sol et al., 2012), and thus is also nonrandom with respect to phylogeny. These biases in introduction and establishment could in theory either amplify or dampen differences in phylogenetic signal between alien and native species. Another potential source of bias is that introductions are more likely to involve widespread species than those that are geographically restricted (Blackburn & Duncan, 2001b;Blackburn, Lockwood, & Cassey, 2009;Pyšek et al., 2009). If small range size is a symptom of recent speciation, and speciation tends to decouple variation in range size from evolutionary relatedness (Pigot, Phillimore, et al., 2010), then by selecting more widespread species the process of human introductions may impart an anomalously strong phylogenetic signal to the range size of established species.
Separating these alternative explanations is challenging because it requires information not only on the range size of species that have successfully established, but also on the range size of those introduced species that fai ed to estab ish Figure a Unfortunate y information on these failed invasions is rarely available.
Here, we overcome this challenge by using a unique database, the Global Avian Invasion Atlas, which contains records of all known avian introductions and the geographical distributions of all established alien species (Dyer, Redding, & Blackburn, 2017). When combined

FIGURE
Phylogenetic signal in species geographical range size across the invasion pathway. (a) Cartoon phylogenies showing how evolutionary ancestry relates to the range size (circle size) of all species in a clade (all native), those species that have been introduced to new locations by human activity (introduced species) and those species that have established alien populations in these new locations (established species). Established species potentially have a distinct phylogenetic signal in their native (blue) and alien (orange) ranges. (b and c) Amongst established species, the symmetry in range size between sister pairs (i.e., species that are each other's closest relatives) may differ between native and alien distributions. Range size symmetry (area of smaller species range/area of larger species range) varies between zero and one, with higher values indicating ranges that are more similar in size.  (established, breeding, unsuccessful, died out, extirpated and unknown). Here, we focused on established species (i.e., those with self-sustaining populations) for which information on their geographical distribution was available (n = 359). We overlaid species ranges onto a recent biogeographical regionalization for birds (Holt et al., 2013) and assigned species to the realm in which the majority of their distribution falls (n = 9 realms).

| Quantifying phy ogenetic signa in range size
We quantified phylogenetic signal in range size using Pagel's λ (Pagel, estimated in the R package MOTMOT Thomas Freck eton 2012). The parameter λ represents a multiplier applied to the offdiagonal elements of the phylogenetic variance-covariance matrix and varies from zero, where the trait is independent of phylogeny, to one, where variation is consistent with a Brownian motion model of evolution. We estimated λ separately for the native and alien ranges of established species (λ Established native and λ Established alien , n = 359) and for the native ranges of all bird species (λ All native , n = 9,993) and those that have been introduced (λ Introduced native , n = 965). In the phylogeny of Jetz et al. (2012), species lacking genetic sequence data were inserted according to taxonomic constraints. Any resulting error in inferred evolutionary relationships may lead to biased estimates of λ.
To test whether this influenced our results, we recalculated λ for only those species represented by genetic data: All species (n = 6,670), introduced species (n = 859) and established species (n = 329). Native and alien range sizes were strongly right skewed and were log 10 transformed before analysis.

| Quantifying symmetry in range size
In addition to λ, we calculated the symmetry in range size within pairs of closely related established species (hereafter "sister species", n = pairs Wa dron Webb Gaston Figure b and c). Although not true sister species, these pairs represent lineages that are each other's closest relatives amongst the set of established species. Symmetry was calculated separately for native and alien ranges, as the area of the smaller species range divided by the area of the larger species range. According to this metric, a stronger phylogenetic signal should be reflected in sister species having more similar (i.e., symmetric) range sizes. Here, we compare range symmetry only between the native and alien ranges of established species, rather than across different stages in the invasion pathway.
Estimates of range symmetry are not comparable across these different subsets because "sister species" would differ greatly in their average phy ogenetic separation a birds Myr introduced species Myr estab ished species Myr

| Testing for differences in phy ogenetic signa and symmetry between native and a ien ranges
To test whether there was a significant difference in λ between native and alien distributions of species, we compared the fit of a model in which the value of λ could differ between groups (n = 2 parameters) with a null model assuming a single global λ (n = 1 parameter). Relative model fit was assessed using the Akaike information criterion (AIC), where an AIC difference (ΔAIC indicates substantial support for the more complex model. In addition, we also report AIC weights (AICW), which quantify the relative probability that each model is correct given the set of models being compared.
We tested for a significant difference in range size symmetry | Testing for differences in the spatia and phy ogenetic components of native and a ien ranges In addition to the effects of phylogenetically conserved traits, a phylogenetic signal in native range size is expected because c ose y re ated species are genera y c ustered in space Freck eton & Jetz, 2009). Consequently, if species are introduced to different locations at random with respect to phylogeny, this would weaken the phylogenetic signal in alien range size. We examined this possibi ity in two ways First we tested the hypothesis that introduction locations are random with respect to phylogeny by calculating the spatial overlap between sister species in both their native and a ien distributions For each group we compared the frequency of spatial overlap with that expected under a null model in which species ranges were randomly reassigned to species (1,000 replicate simulations). Overlap scores were calculated as the area of overlap divided by the area of the smaller species range (Pigot, Tobias, & Jetz, 2016), as follows: where A1 and A2 are the range sizes of the two species. Second, we jointly quantified the variation in range size that is uniquely structured according to either space (Φ) or phylogeny (λ or that is independent of both components (γ), using the approach of Freck eton and Jetz Within this framework Φ quantifies the proportion of the variance in range size attributable to spatial location (0 = no spatial effect, 1 = pure spatial effect). This spatial effect was modelled assuming that the variance in range size between species increases linearly with the great circle distance between species geographical range centroids. The parameter λ Φ) λ, is a spatially corrected version of λ that quantifies the proportion of the variance in range size uniquely attributable to phylogenetic relatedness (0 = no phylogenetic effect, 1 = pure phy ogenetic effect Fina y γ Φ λ) describes the proportion of the variance in range size that is independent of either space or phylogeny. We used maximum likelihood simultaneously to estimate Φ and λ separately for both the native and the alien range size of introduced and established species. Code to fit this mode was kind y provided by R Freck eton We predict that if spatial proximity is the primary determinant of phylogenetic signal in native range size then accounting for space should lead to a weaker phylogenetic signal in native range size (i.e., λ λ). In contrast, if spatial proximity is decoupled from phylogenetic similarity amongst species alien ranges, estimates of phylogenetic signal should be similar regardless of whether we account for space (λ or not λ).

| Testing for differences in phy ogenetic signa across the invasion pathway
To test whether differences in phylogenetic signal arise from nonrandom patterns of introduction and establishment, we conducted a series of stochastic simu ations First treating the g oba avifauna as the species pool (n = 9,993, "global pool"), we randomly sampled 965 species, equivalent to the number introduced (null model 1). Second, we randomly sampled 359 species from the global pool, equivalent to the number of established species (null model 2). This latter null model assumes that established species are a random sample of the global avifauna. However, species can establish only if they have first been introduced. We therefore implemented a third null model (null model 3) in which 359 species were randomly sampled from the pool of species that have been introduced (n = introduced poo For each nu mode we performed 10,000 trials (i.e., 100 replicates for each of 100 phylogenetic trees For each tria we estimated λ for the simulated data and tested whether this differed significantly from the observed value of λ by comparing the AIC of a model with a single global λ (n = 1 parameter) with a model in which the value of λ could differ between groups (n = 2 parameters). Through these null models, we aimed to identify the stage s in the invasion pathway g oba introduced estab ished during which any potentia differences in phylogenetic patterns arise.

| Stochastic mode s of species introductions
The null models identified the transition from the global pool to introduced species as a key stage in the invasion pathway generating differences in phylogenetic patterns (i.e., null model 1). To test the causes of this finding, we compared the phylogenetic signal in the native ranges of introduced species with that expected under a suite of introduction scenarios, specifically testing the following hypotheses: Differences in λ arise from the tendency preferentially to introduce species: (a) With larger ranges ("range size-dependent scenario"), (b) from certain regions ("region-dependent scenario") and (c) from particular clades ("clade-dependent scenario"). We parameterized these models by fitting a series of generalized linear mixed effects models with a binomial error structure, predicting whether species from the global avifauna (n = 9,993) have been introduced (one) or not (zero). We variously included native range size as a fixed effect ("range size-dependent scenario"), and taxonomic family ("clade-dependent scenario", n = 194 families) and biogeographical realm ("region-dependent scenario", n = 9 realms) as random effects. We used the parameter estimates from these models to determine the probability of each species being selected for introduction in our stochastic models.
As differences in λ may arise from a combination of factors, we additionally implemented three synthetic models combining the effects of range size and either avian family or biogeographica rea m For mode s inc uding random effects we compared a model including random slopes or random intercepts and used the mode with the ower AIC Fina y we fitted a mode containing a three variables. In this case, a model with random slopes for both avian family and biogeographical realm could not be estimated; therefore, only models including random slopes for either family or rea m were considered For each scenario we used the mode  Note. Phylogenetic signal was estimated in isolation (λ) or having accounted for spatial effects (λ In the atter case the unique components of range size variation attributable to phylogeny (λ space Φ) or that is independent of either space or phylogeny (γ) are reported. Values of λ are shown for all species and those represented by genetic data. Values are maximum likelihood estimates (and 95% confidence interval). The parameters Φ, λ and γ can each vary continuously between zero and one (summing to one), corresponding to scenarios in which none (zero) or all (one) of the variation is associated with space (Φ), phylogeny (λ or neither space or phy ogeny γ). In accordance with the patterns observed in λ, we found that overall range size symmetry was significantly higher for the native (mean symmetry = 0.34) compared with the alien (mean symmetry = 0.24) distributions of established species [effect = 1.06 ± 0.37 (SE), p n = 115 pairs; Supporting Information Table S1)].

waxbill (Estrilda coerulescens) and grey waxbill (Estrilda
There was a significant interaction between maximum range size and geographical origin (i.e., alien vs. native) in explaining range size symmetry (effect = 0.64 ± 0.17 SE, p

Figure a Supporting
Information Table S1). Specifically, the symmetry in native range size was independent of maximum range size, whereas the symmetry in alien range size decreased strongly with maximum alien range size All these results were qualitatively unchanged when restricting our analysis to only those pairs represented by genetic data (n = 106 pairs; Supporting Information Table S1).

| The effects of introduction ocation on the phylogenetic signal in range size
We found that 56% of established sister species co-occur across at least part of their native geographical range (mean overlap of co-occurring pairs Supporting Information Figure   alien range sizes appears to be driven largely by the tendency for closely related species to occur in the same geographical locations rather than because of phylogenetically conserved traits. The phylogenetic signal in the native range size of established species (λ Established native = 0.5, n = 359 species) is similar to that of introduced species (λ Introduced native = 0.61, n = 965 species), which in turn is similar to that of the global avifauna (λ All native = 0.54, n = 9,993 species). These similarities in λ are robust to whether estimates are made across all species or only those represented by genetic data (Table 1) and appear to suggest that as species pass through the various stages in the invasion pathway the phylogenetic signal in native range size remains largely unaltered. In accordance with this, we found that the phylogenetic signal in the native range size of es-   (Table 1). Thus, the process of introduction selects for species with a high phylogenetic signal in range size, whereas the process of establishment and spread appears subsequently to erase, albeit not entirely, the imprint of evolutionary history on species range size.  (Table 1). Likewise, although the phylogenetic signal in the native range sizes of introduced species appears similar to that of the global avifauna (Table 1), null model simulations revealed that this can be explained only by highly nonrandom patterns of species introduction Figure  If introductions had occurred randomly with respect to species identity, then the phylogenetic signal in the native range size of introduced species would be expected to be significantly weaker than is observed Figure a Why the phylogenetic signal in the native range size of introduced species should be particularly strong is unclear because, to our knowledge, this pattern has not previously been documented. Our results, however, suggest that this pattern can be explained by the nonrandom process of avian introductions, which has been biased towards a few geographical regions and taxonomic families, and species with arge geographica ranges Supporting Information Figure   S3; Blackburn et al., 2009). The effects of these biases are consistent with the predictions from theoretical models of range size evolution.
In particular, speciation is expected to lead to closely related species with highly asymmetric range sizes, and phylogenetic signal will thus increase over time as species with small ranges either expand their distributions or are "filtered out" by the process of extinction (Pigot, Phillimore, et al., 2010;Waldron, 2007). In a similar way, by preferentially selecting species with larger geographical ranges (Blackburn et al., 2009;Pyšek et al., 2009), the process of human introduction may impart a higher phylogenetic signal to the range size of introduced species than expected by chance Furthermore we found  Tests of differences in phylogenetic signal (λ) in range size between native and alien distributions and across different stages of the invasion pathway that the preferential sourcing of introduced species from a subset of geographical regions (particularly the Palearctic and Nearctic) also contributed to a higher phylogenetic signal in the range size of introduced species Figure a This makes sense because our resu ts show that spatial proximity between closely related species is the major driver of phylogenetic signal in native range size, with no independent effect of evolutionary relatedness.
In contrast to the intermediate phylogenetic signal in the native range size of established birds, variation in alien range size exhibits a much weaker phylogenetic signal (Table 1). Indeed, even very closely related species often had highly asymmetric alien range sizes Figure a One exp anation for this pattern is that upon introduction, species will initially be uniformly rare, thus decoupling alien range size from phylogenetic ancestry. A stronger phylogenetic signal, comparable to that of native distributions, may only be expected to emerge over time, as alien range size expands to the limits imposed by species intrinsic traits (Byers et al., 2015). However, our results show that although the first introduction times of closely related alien species were in some cases separated by centuries, differences in range size were unrelated to differences in residence times.
Thus, the phylogenetic signal in alien range size does not appear to be limited by a lack of time for dispersal.
An alternative possibility is that the weaker phylogenetic signal in alien range size is a consequence of the particular way in which  Table S2 for the parameters used in each stochastic model humans have redistributed species across the p anet For native species, the phylogenetic signal in range size is almost entirely attributed to the tendency for closely related species to occur in close geographica proximity Tab e Freck eton Jetz  This effect of space presumably arises because species occurring in the same region will be subject to the same environmental and geographical barriers to range expansion Machac et a Pigot Owens Orme, 2010). As a result, if closely related species tend to be transported to widely scattered locations then this may decouple species phylogenetic relatedness from spatial proximity, thus eroding the key mechanism promoting the phylogenetic signal in range size.
Our results, however, do not support this idea either. We found that closely related alien species tend to be introduced to the same locations more often than expected by chance Figure b Furthermore as with native ranges, spatial proximity accounted for most of the variation in alien range size (Table 1). Thus, the positive association between phylogenetic and spatial distance observed in native species appears to be largely maintained in the alien distributions of species, despite the very different processes involved in determining the spatial location of their geographical ranges.
A lack of time for dispersal and the spatial patterns of species introduction therefore appear unlikely to explain the weak phylo- suggest that phylogenetically conserved traits may be unlikely to explain the subsequent extent of spread. This weak predictive ability of phylogeny is highlighted by the fact that alien range sizes are highly asymmetrical even amongst the most closely related species, and these asymmetries could not be explained by differences in the length of time available for dispersal. Where apparent associations exist between phylogenetic relatedness and the range size of invasive species, our results suggest that this is likely to be attibutable simply to closely related species being introduced to the same regions and with similar effort.

| CONCLUSIONS
Whether the geographical extent attained by alien species is predictable on the basis of their evolutionary ancestry has not previously been thoroughly explored. Here, we show, for an entire class of organisms, that the phylogenetic signal in alien range size is weak compared with native species distributions. This result implies that heritable intrinsic traits have relatively little effect on the range size of alien species and that phylogenetic relatedness is unlikely to provide a robust approach for predicting the spread of invaders.
Importantly, we demonstrate that these patterns are detectable only when accounting for the taxonomic and geographical biases in species introductions and are not evident when using the naïve assumption that avian introductions represent a random sample of species. Predictions of the potential spread and impact of introduced species should therefore account for historical biases in species introductions.