On the completion of speciation

Ever since Darwin's seminal work [1], the origin of species has been a central problem in our field. Yet, speciation is not so difficult. Populations will inevitably evolve in different ways and, hence, become incompatible: hybrid genotypes will almost never have been tested by natural selection, and so tend to be unfit. Moreover, there are many ways for organisms to make a living, and so they will tend to diversify into a multitude of distinct ecological niches: ‘One may say there is a force like a hundred thousand wedges trying [to] force every kind of adapted structure into the gaps in the œconomy of Nature’ [2, p. 135e] Seen in this light, the main problem is why speciation should be so slow and incomplete: populations that have diverged for millions of years can still interbreed [3,4], and genomics has now revealed slight, yet biologically significant, gene flow between distinct species [5,6].

Conversely, excessive speciation would be disastrous, since it leads to ever smaller and more specialized populations. Regular sexual reproduction is maintained in eukaryotes, most likely because it facilitates adaptation, which is most efficient in large and freely recombining populations [7]. Long before the development of evolutionary genetics, Darwin noted that species with large numbers and geographic range tended to be at a competitive advantage [1, ch. 11]. Random drift imposes a robust limit to the number of traits and the amount of information that can be evolved and maintained [8], and Lynch [9] has argued that in the long term, complex molecular adaptations require very large population sizes. Some reproductive isolation is required to allow adaptation to local conditions and coexistence of specialists in sympatric niches, but it need be by no means complete: the adaptive advantages of large population size can be retained despite partial speciation. This is, of course, not to argue that species will evolve to some optimal degree of reproductive isolation. Nevertheless, over-specialized species may be doomed, both by the limited genetic diversity available to them, and more directly, by their precarious niche.

We can envisage biological species as collections of locally adapted populations, connected by a trickle of gene flow, yet able to coexist in sympatry, if sufficiently reproductively isolated; this has been termed a ‘syngameon’ [10,11]. There will be a constant tension between specialization to particular niches and openness to gene flow. Fragmentation may be limited not only by the several risks of extinction to a small and specialized population, but also by the difficulty and costs of evolving reproductive isolation. Many of the model systems discussed in this theme issue can be seen in this way––for example, the cichlid and stickleback radiations [12,13], and on a smaller scale, the ecotypes of Littorina snails [14,15]. An especially clear example, and one of the first to show how distinct ‘species’ can coexist in sympatry, and yet largely share the same genetic diversity, is in Drosophila persimilis and Drosophila pseudoobscura, which diverge only at a few inversions [16]. Similarly, most divergence between the Littorina ecotypes is associated with inversions [15]. Host races in Rhagoletis also involve inversions, though here, there is extensive divergence across the genome [17,18]. By reducing recombination, chromosomal rearrangements facilitate divergence despite gene flow [19,20], but allow high diversity to be retained across the rest of the genome. However, note that inversions may not be necessary for divergence and could arise only after the initial divergence [21].

Very low levels of gene flow may have significant evolutionary consequences. Much attention has been given to the swamping of local adaptation by gene flow. However, this requires a high rate of gene flow, comparable with selective strength (m ≈ s) [22]. In two dimensions, even free diffusion of genes, σ², will not prevent adaptation, as long as selection is sustained over a large enough spatial area (approx. σ²/s) [23]. Thus, the low levels of gene flow inferred from sequence data [5,6] may present a negligible obstacle to selection. By contrast, just a small number of migrants can inhibit neutral divergence (Nm ≈ l), and even lower migration rates can introduce more variation than mutation (Nm ≈ Nµ ≪ 1 per site). Even this comparison underplays the possible role of occasional gene flow. If selection builds up complex alleles, consisting of multiple mutations that (for example) tune gene expression, then extremely low rates of gene flow would introduce these far faster than multiple mutations could reconstruct them. Thus, selection might easily maintain distinct locally adapted populations, and gene flow might even be low enough to allow neutral divergence––and yet adaptive variation could nevertheless arrive more often by migration than by mutation. Such a situation is familiar among bacteria, where even though recombination is very rare, the effective population size is so large that recombination is an important source of variation, often involving co-adapted sets of genes [24].

It is natural to suppose that the spread of selectively favoured alleles throughout a biological species is beneficial. However, such global sweeps interfere with local adaptation at linked loci, imposing a load on the population; conversely, local adaptation interferes with global adaptations [25,26]. Such Hill–Robertson interference can be alleviated by recombination and thus leads to indirect selection to maintain sexual reproduction [7]. Another cost to the spread of globally favoured alleles may be more serious. The spread of a selfish element is deleterious for the species as a whole, though the spread of a suppressor in response would be beneficial. Coughlan & Matute [27] review the role of intrinsic incompatibilities and point out that these often arise through some kind of conflict; such incompatibilities arise when different populations resolve conflicts in different ways.

The scenario outlined here, where locally adapted populations remain connected by gene flow, is reminiscent of Wright's ‘shifting balance’ [28], in which local demes drift to different adaptive peaks and then compete with each other. Seen more generally, local demes may reach different ‘adaptive peaks’ through drift (as Wright envisaged), or through the indirect effects of fluctuating extrinsic selection. The scenario discussed in the present paper places more emphasis on adaptation to local environments and, in addition, allows partially isolated populations to become sympatric. There have been extensive surveys of genetic differentiation across space, and more recently, ‘genome scans’, which seek loci with excess geographic differentiation (e.g. [13,14,29] in this issue). Yet, we still know very little about how much differentiation between local populations is selected, and how much of that differentiation is extrinsic versus intrinsic: that is, a direct response to local environment, versus divergence to alternative combinations of alleles (adaptive peaks). Either kind of divergence will contribute to reproductive isolation.

A spatially extended population may contain substantial divergence––including adaptations to diverse environments, and alternative ways of adapting to the same environment. Yet, this divergence may not be at all obvious, if there is gradual ‘isolation by distance’, with no sharp boundaries, and if there are very many different and uncorrelated environmental variables. It is only when divergence involving different loci and traits is brought together that we see obvious ‘species’ [30]. Such ‘coupling’ of diverse isolating factors may happen within a local population, through linkage disequilibrium [31], or across the geographic range, through the coming together of scattered clines into a strong hybrid zone [3]. The contrast between gradual and cryptic divergence, and sharp disjunction is seen most clearly in ‘ring species’ [32]. The coupling together of different reproductive barriers is a reshuffling of existing variation, and so may occur relatively fast, as a completion of the speciation process.

Several articles in this issue discuss the possibility that with sufficient divergence, a ‘tipping point’ may be reached [33], after which speciation is rapidly completed [13,34–36]. The idea is that once the barrier to gene flow is sufficiently strong, further divergence is facilitated, causing a positive feedback that leads to complete isolation. This feedback may occur because a strengthening barrier allows a wider set of locally favoured alleles to establish, despite gene flow, because reinforcement occurs, or because, through epistasis, more alleles are favoured only on one genetic background. The coupling of barriers within a single population, via linkage disequilibrium, discussed above, may also be very rapid [31].

The feedback that causes ‘tipping points’ can be seen directly in 'stepped’ clines, in which linkage disequilibrium in the centre of a hybrid zone generates a strong barrier, so that individual alleles introgress at a reduced rate [37]. However, there are few clear examples of such ‘spatial phase transitions' (Mus [38], Bombina [39], Podisma [40]). This may be because dense spatial sampling is needed to identify a step, but more likely is because the genetic map is typically long enough that selection does not often maintain a strong barrier. A separate question is whether there is a sharp tipping point at which sympatry becomes possible, when ecological divergence outweighs the cost of hybridization; there is a suggestion that this may be the case in Bombina, where in some places there is a sharp hybrid zone, but in others, the species are found in different habitats within a sympatric mosaic [41]. However, sympatry may not require extensive divergence; in this issue, Osborne et al. [29] argue that divergent flowering time allows coexistence in different habitats within a small island.

As argued in the Introduction to this issue [34], speciation does not progress through a regular series of stages. Although overall, reproductive isolation tends to increase through time, it may sometimes break down. Under a Dobzhansky–Muller scenario [42], incompatible populations are connected via a chain of fit ancestral genotypes, allowing free gene flow even between populations that would seem incompatible. Virdee & Hewitt [43] describe an example where F1 hybrid male grasshoppers are sterile, and yet no such sterility is seen in a natural hybrid zone, where fit recombinants have been regenerated. Similarly, F1 hybrids between shrews that differ by multiple chromosomal fusions are sterile, but in nature, acrocentric chromosomes (the ancestral state) arise that greatly reduce the incompatibility [44]. Thus, (contra [27]), Dobzhansky–Muller incompatibilities may be fragile, at least until F2 and backcross hybrids are completely inviable. Incompatibilities driven by ecological selection may be similarly vulnerable to loss when environments change (for example, when turbid waters weaken visual mate preferences in cichlids [12]).

DNA sequences are shaped by demographic history and by selection. Reflecting the current state of the field, most of the empirical papers in this issue attempt to use sequence data to infer the joint effects of population history and selection on the process of speciation. There was early enthusiasm for this approach when sequence data first emerged in the 1980s, which greatly strengthened in recent years, as full genomes became available. However, it has proved extremely challenging to convincingly disentangle the many processes involved. It is now clear that sequence diversity is substantially influenced by linked selection, as witnessed most directly by the correlation between diversity and recombination rate [45,46]. Typically, sequence data are analysed by separate application of multiple statistical methods––each of which has specific assumptions that cannot all be satisfied; it is rare to see a synthesis of the separate approaches. We are only beginning to understand how to allow for the joint effect of demography and selection, and do not yet know how far they can be separated, even in principle.

The questions raised in the introduction to this issue [34, box 1] mainly concern interactions between processes, which may alter the dynamics of speciation––for example, whether there is a ‘snowball effect’, or ‘tipping points’, which accelerate the completion of reproductive isolation. Such questions are inherently hard to address, since we cannot follow speciation through time (except in highly artificial experiments). Comparative analysis may help and has been very successful, for example, in showing how pre- and post-zygotic isolation increase with genetic distance [47], and in elucidating Haldane's Rule [48]; however, such meta-analyses rest on an enormous number of individual studies. In this commentary, I draw attention to a different, and perhaps more tractable, question: how far do low levels of gene flow assist adaptation, by allowing populations to draw on a wider pool of genetic diversity? Coalitions of distinctly adapted populations may gain by not completing speciation.

Data accessibility

This article has no additional data.

Competing interests

I declare I have no competing interests.

Funding

I received no funding for this study.