Tempo and timing of ecological trait divergence in bird speciation

Summary paragraph

Organismal traits may evolve either gradually or in rapid pulses followed by periods of stasis, but the relative importance of these evolutionary models in generating biodiversity has proven difficult to resolve^1,2. In addition, while it is often assumed that pulses of trait evolution are associated with speciation events, few studies have explicitly examined how the tempo of trait divergence varies with respect to different geographical phases of speciation. Thus, we still know little about the trajectories of trait divergence over timescales relevant to speciation, or the extent to which these trajectories are shaped by variation in geographical isolation and overlap (sympatry) among incipient species. Here, we combine divergence time estimates, trait measurements, and geographic range data for avian sister species pairs worldwide to examine the tempo and timing of trait divergence during allopatric speciation. We show that divergence in two important ecological traits—?body mass and beak morphology—is best explained by a model including pulses of divergence and periods of relative stasis. We also infer that trait divergence pulses often precede sympatry, and that pulses leading to greater trait disparity are associated with earlier transitions to sympatry. These findings suggest that early pulses of trait divergence promote subsequent transitions to sympatry, rather than such pulses occurring after sympatry has been established, for example via character displacement³. Incorporating pulsed divergence models into allopatric speciation theory helps to resolve some apparently contradictory observations, including widespread instances of both rapid sympatry and prolonged geographical exclusion^4-6.

TEXT

Speciation in vertebrates may proceed over long and variable periods⁷. The entire process from onset to completion is often subdivided into three stages, beginning with a phase of geographic isolation (allopatry), followed by secondary contact, and finally the transition to coexistence in overlapping geographical ranges (sympatry; see Fig. 1)^7-10. In standard forms of this model, the third stage is delayed by competitive interactions¹¹ or incomplete reproductive isolation⁹, and thus only occurs when traits are sufficiently divergent to facilitate sympatry⁶. However, while the pattern of increased trait divergence in sympatric lineages is widespread among animal taxa⁴, the timing and geographical context of the process of trait divergence is often unclear. In particular, trait divergence could arise primarily by the accumulation of differences prior to secondary contact¹² or alternatively after sympatry is established³. Furthermore, the tempo and mode of trait divergence during speciation is also controversial, with some studies describing divergence as slow or gradual throughout the process^5,12, while others provide evidence for abrupt, pulse-like changes^13,14 occurring either early in allopatry^15,16 or later in sympatry^{3, 6} (Fig. 1). It can even be argued that ecological (local) adaptation in allopatry, followed by species interactions in sympatry, provide the context for multiple pulses of trait divergence over time¹⁵.

Figure 1. The speciation cycle and phenotypic trait divergence

Bird speciation typically involves a sequence of geographical states, starting with an allopatric phase (a), followed by secondary contact (b), and finally sympatry (c). Phenotypic divergence may take different pathways during this cycle: gradual models predict no pulse of divergence at any point in the cycle (d), whereas punctuated models involve stasis punctuated by pulses, which can follow the onset of coexistence (e) or precede it (f). Note that secondary contact (orange) is extended when traits are similar (e), and reduced when traits have already substantially diverged in allopatry (f).

Prolonged speciation creates problems when modeling pulsed trait evolution at a macroevolutionary scale, and hence for most studies comparing evidence for pulsed versus gradual evolution. Pulsed models typically assume that evolutionary change is concentrated at speciation², which is modelled as a single instantaneous event, potentially leading to misinterpretations about trait evolution. Moreover, the contribution of pulsed trait evolution may become obscured because stasis (or bounded evolution¹⁷) with intermittent pulses can resemble gradualism at these coarse macroevolutionary scales². Micro-evolutionary (population-level) studies conducted over short time periods confirm that stasis is prevalent¹⁸ and also that trait evolution pulses sometimes occur^4,17. Observed pulse events, however, may represent brief departures of trait values from their long-term means, followed by reversion, and may therefore not contribute strongly to patterns of trait variation at macroevolutionary scales^{4,17, 18}. Thus, although previous studies have found support for both gradual and pulsed modes of evolution^19,20, their relative importance in generating biological diversity remains unknown². An added complication is that trait divergence during the allopatric phase of speciation may be ephemeral if gene pools merge during secondary contact, whereas greater levels of divergence may lead to reproductive isolation and ultimately sympatry, a possibility that could accentuate patterns of pulsed evolution in phylogenetic approaches and the fossil record, even when divergence itself is gradual^16,21.

Disentangling these alternative divergence pathways is a key step in resolving general patterns of trait evolution and predicting which nascent species ultimately succeed, leaving daughter species, and which will perish or fail to remain distinct when changing environments re-organize geographic ranges^{16,21, 22}. However, our understanding of the rates and timing of trait divergence in relation to stages of the speciation process in vertebrates remains highly incomplete, not least because the data required to test these ideas are often lacking. In particular, although broad-scale information on ecological traits and phylogenetic history is available for some large vertebrate clades²⁰, the accompanying information on geographic ranges is not sufficiently resolved to explore divergence pathways in the context of geographical phases of speciation.

To address this issue, we examined phenotypic divergence, geographic relationships, and estimated divergence times^23-25 among 952 pairs of avian sister species. As we were interested in the tempo and timing of divergence during the speciation process, we applied a set of evolutionary models designed to span both microevolutionary and macroevolutionary processes¹⁷ to estimated trait disparities and divergence times. We used this approach because species pairs may vary in the extent to which their trait divergence is better characterized by microevolutionary or macroevolutionary processes. Focusing on two important ecological traits—body mass and beak morphology—we assessed relative support for four stochastic trait divergence models: “gradual”, “single pulse”, “multiple pulse”, and time-independent (“white noise”) models (see Methods). The first three models (i.e. all except the white noise model) incorporate a bounded evolution component to represent processes at shorter timescales. We found the strongest support for the single pulse model, in which the bounded evolution component is relatively narrow²⁶ (Supplementary Tables 10-13). Support for the single pulse model was much stronger than for the gradual model (ΔAIC 899 for body mass, ΔAICs 862, 1004, and 968 for beak PC1, PC2, and PC3, respectively; see Methods and Supplementary Tables 10-13), in agreement with phylogenetic studies reporting a strong contribution of pulses in the accumulation of phenotypic diversity^19,27.

Moreover, these results raise two further questions that we address here: 1) how does the estimated timing of divergence pulses compare to typical progressions through the geographic speciation process (allopatry, secondary contact, sympatry)? and 2) does pulsed ecological trait divergence impact transitions through this process? With respect to the first question, the expected waiting time to a pulse of divergence in the single pulse model was ∼670,000 years (95% CI from likelihood profile: 280,000 years to 1.13 My) for body mass (see also Figs. 3 and 4, Supplementary Table 10). The expected waiting times to a pulse in beak morphology divergence in single pulse models were ∼560,000 years for PC1 (95% CI: 200,000 years to 1.0 My), ∼170,000 years for PC2 (95% CI: 0 to 410,000 years), and ∼90,000 years for PC3 (95% CI: 0 to 280,000 years; see Supplementary Tables 11-13). To compare these estimates with progression to secondary contact and sympatry, we used a fine-grained geospatial database of ∼178 million species observation records and standard geographical range polygons, respectively (see Methods). Secondary contact often occurs so rapidly that the signature of allopatry is difficult to detect in our analyses of local co-occurrence (contact) and divergence time (Extended Data Fig. 4), suggesting that pulses may occur following secondary contact or even during parapatric speciation (see below, Methods and Supplementary Information). However, based on the relative timescales of trait divergence pulses and sympatry establishment shown in (Fig. 4), we also conclude that divergence pulses typically precede the establishment of sympatry, and are thus unlikely to be driven by character displacement processes³.

Figure 3. Tempo of body mass divergence for avian sister species.

Stochastic pulsed models provide better fits to patterns of body mass divergence and divergence time, with the best fit a single pulse model (ΔAIC relative to the multiple pulse model: 797). Colors denote probability density. The probability density for any time slice follows a normal distribution (most apparent in the white noise model where the probability density distribution is independent of time). Relative probability density can be assessed within each time slice but not across time. For clarity, the empirical data points are plotted only on the white noise model.

Figure 4. Timing of body mass divergence pulses and sympatry.

Comparison of timescales suggests that mass divergence tends to precede sympatry among 952 avian sister species. Yellow and orange lines are cumulative probability distributions of incurring a pulse under the single pulse model (yellow: estimated from maximum likelihood phylogenetic tree²³; orange: from 100 bootstrap trees²³). Circles are proportions of sympatric species pairs for 1-million year intervals of divergence time²³; circle sizes represent sample sizes, numbered where ≥50. For visual comparison, an exponential decay model has been fitted to the proportion of sister pairs in sympatry (blue curve; assumes sympatry is secondary).

Early pulses of ecological trait divergence theoretically reduce both competition and reproductive interference among incipient species¹⁵, potentially overcoming constraints on sympatry^10,22. To assess whether such pulses influence rates of transition through geographical stages of the speciation process, we tested whether variation in body mass and beak morphology predicted which species pairs are parapatric or sympatric. Focusing on all species pairs found to locally co-occur (n = 441, see Methods), and accounting for the effects of divergence time, dispersal ability, and latitude, we found strong evidence that sympatry is associated with greater divergence in body mass, and, to a lesser extent, beak morphology (Fig. 2, Extended Data Fig. 1, Supplementary Tables 4-5, 8-9, 18-19; see Methods). The relationship between large body mass differences and increased likelihood of sympatry was highly consistent across sensitivity analyses (Fig. 2, Extended Data Fig. 1, Supplementary Tables 4-5, 8-9, 18-19). These results are largely in agreement with previous studies showing that the transition from secondary contact to sympatry is facilitated by divergence in body mass and beak morphology^10,12, and further suggest that divergence in body mass is a more critical factor.

Figure 2. Factors associated with the establishment of secondary contact and sympatry in birds.

Results of generalized linear models assessing the relative importance of predictors of breeding range local co-occurrence (a) and sympatry (b) in 952 avian sister species. Pairs with breeding co-occurrence include both parapatric and sympatric species pairs. Relative importance is estimated as the proportion of the summed model weights for all models with ?AIC<2, and indicates the extent to which each variable predicts the probability of co-occurrence or sympatry. Pairwise interactions with relative importance >0.6 are indicated by the numbers within the bar for each variable.

We have shown that secondary contact occurs earlier in the speciation process than generally assumed under classic models of allopatric speciation (Extended Data Figs 2, 4 and 7), suggesting a potentially wider role for parapatric speciation (speciation with no stage a in Fig. 1)⁵. Under this model of speciation, divergence pulses occur despite contact, and thus the potential for gene flow, between incipient species¹⁵. Our analyses indicate that this scenario may be widespread in bird speciation. However, we also found that the observed pattern of breeding co-occurrence and divergence times could also result from purely allopatric speciation with rapid rates of transition to secondary contact (speciation with reduced stage a in Fig. 1). Stochastic models indicate than an approximate minimum rate of 0.3 transitions to secondary contact per million years is sufficient to explain the pattern of breeding co-occurrences among avian sister species pairs (Extended Data Fig. 5, Supplementary Information). Thus, trait divergence pulses may take place either during periods of allopatry or parapatry, with cases of both probably widespread.

The standard evolutionary trajectory implied by our best-fitting model—an early, pulsed divergence with constrained subsequent divergence—can potentially explain a variety of phenomena in bird speciation. Under this single pulse model, pulses vary in magnitude across species, with a fraction of species undergoing large evolutionary jumps early in the speciation process, and others incurring small-magnitude divergence¹⁷. Thus, many sister species strongly resemble each other in ecological traits whether they began to diverge recently or anciently, whereas a fraction of species pairs have undergone an early pulse of rapid divergence, and remain highly divergent regardless of their age. One interpretation of this pattern is that a species pair undergoing a small pulse of ecological trait divergence in the early stage of speciation is unlikely to undergo large pulses at later stages in the absence of other speciation events, and thus the species pair may be subject to extended periods of mutual exclusion via ecological competition, perhaps in combination with reproductive interference^5,9.

A prevailing view is that strong divergence in ecological traits between lineages typically requires long periods of time, i.e. slow-rate gradual divergence¹². Gradual divergence models can account for prolonged mutual exclusion between highly similar species, a widespread phenomenon⁵, especially when evolutionary rates of gradual evolution are low^2,10. However, gradual divergence models inadequately account for highly divergent young species pairs, unless they also incorporate brief bursts of faster gradual divergence (mimicking pulses)²⁷. While gradual models thus may adequately explain avian trait evolution at macroevolutionary scales, our findings suggest that patterns of diversification among sister pairs are better captured by models incorporating pulses of divergence, where the magnitude of pulses is independent of divergence time. In particular, the single pulse model receives strong support and can help to explain the full range of outcomes observed in nature, which not only include prolonged mutual exclusion between ecologically similar sister species⁶ but also instances of rapid sympatry following abrupt ecological trait divergence⁴.

The occurrence of early pulses of trait divergence raises the question of how such pulses arise. One possibility is that they result from the intermittent discovery by populations of unoccupied adaptive peaks, as is expected in niche-filling models of diversification⁸?a mechanism that may be particularly important for large-magnitude pulses⁴. New adaptive peaks be discovered when local adaptation drives rapid divergence after range expansion^14,22, for instance immediately following colonization of novel environments. Rapid phenotypic divergence may also result from local adaptation along environmental gradients, with or without gene flow²⁸. In such contexts, it is worth emphasizing that signals of pulsed divergence may arise from a combination of gradual local (clinal) adaptation and subsequent extinction of intermediate populations^16,29.

In combination, our results may help to resolve the longstanding question of why some nascent species survive over evolutionary time while others are ephemeral. One of the major threats to young lineages is the likelihood of fusion through swamping gene flow after secondary contact^16,21. On one hand, we have shown this risk is widespread among nascent bird species because the lag time to secondary contact is shorter than expected (Extended Data Figs 2, 4-5), supporting the view that gene flow routinely becomes possible early in the speciation process. On the other hand, our findings suggest that species pairs undergoing major early pulses of ecological trait divergence are more likely to transition rapidly to sympatry, escaping both fusion and mutual exclusion, thereby extending their lifespan as independent lineages. Conversely, if they meet at early stages in the speciation process, species pairs with minimally divergent phenotypes may incur increased hybridization rates, or increased hybrid fitness, thereby reducing their lifespan. Indeed, elevated rates of extinction in less divergent young lineages may increase the signature of large early pulses in datasets compiled from extant species. Thus, differential extinction coupled with pulses of early trait divergence may play a critical role in explaining broad-scale patterns in the longevity and macroevolutionary diversity of species, as well as their geographical distributions.

Acknowledgements

We are grateful to numerous data collectors who contributed to eBird, GenBank, and the CRC bird body mass data set (see Supplementary Information). We also thank Nico Alioravainen, Ed Braun, Samuel Jones, Rebecca Kimball, Dan Ksepka, Monte Neate-Clegg, Alex Pigot, Aaron Ragsdale and Gleb Zhelezov for data collection and technical assistance. This work was supported by the National Science Foundation (DEB-1208428 to J.G.B.), the Natural Environment Research Council (NE/I028068/1 to J.A.T.), and the Oxford Clarendon Fund and US-UK Fulbright Commission (to C.S.).

Author contributions

J.G.B. and J.P.M. conceived the study; J.G.B, J.P.M. and J.A.T. designed the conceptual framework and analyses; J.G.B. performed dating analyses and assembled phylogenetic, occurrence, and body mass information; J.A.T. and C.S. provided morphometric data; J.P.M. integrated data sets, and designed and performed statistical analyses with significant input from J.G.B.; J.P.M. produced figures and tables; J.P.M. wrote the manuscript, with significant input from all authors.

Author Information

Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. Readers are welcome to comment on the online version of this article at www.nature.com/nature. Correspondence and requests for materials should be addressed to .