“When I use a word,” Humpty Dumpty said, in rather a scornful tone, “it means just what I choose it to mean – neither more nor less.”
“The question is,” said Alice, “whether you can make words mean so many different things.”
- Lewis Carroll, Through the Looking Glass.
The literature on speciation has expanded dramatically in recent years, catalyzed by the emergence of new conceptual frameworks, new theoretical approaches, and new methods for characterizing pattern and inferring process. As a consequence, the language used to describe the speciation process has become more complex. Increasing complexity may be an accurate reflection of current thinking with respect to how phenotypic differences limit gene flow, how selection results in the evolution of reproductive isolation, and genetic changes that contribute to speciation. However, increased language complexity has come at a cost; old definitions have been reconfigured and new terms have been introduced. In some instances, the introduction of new terminology has failed to recognize historical usage, leading to unnecessary ambiguity and redundancy. Although the writings of Mayr and Dobzhansky remain a reference point in the language of speciation, the last decades of the 20th century saw substantial changes in our thinking about the speciation process. During that period, the language of speciation remained relatively stable. In contrast, the first decade of the 21st century has witnessed a remarkable expansion of the language of speciation. Here, the origin and evolution of ideas about speciation are viewed through the lens of changing language use.
The words we choose to describe concepts, models, patterns, and processes often reflect a particular outlook or point of view. Consequently, the language we use can intentionally or inadvertently direct and constrain our thinking (and the thinking of others). Although differences in usage and meaning are often subtle, these differences can have important consequences for how results of experiments or observations are interpreted. In some cases, terms in common usage lack precise definitions, or have different implications and are used in different contexts by different people. As a science matures, novel ways of thinking naturally give rise to new terms and new concepts. But, new terms also arise even when they are not needed; perfectly useful old terminology is abandoned or forgotten, and links in the great chain of ideas are broken.
The literature on species and speciation has expanded rapidly in recent years, and the language of speciation has become correspondingly more complex. Although 21st century evolutionary biologists confront many of the same questions asked by evolutionary biologists 50 years ago, opportunities to answer those questions have increased dramatically. Abundant data and detailed analyses from new model systems, each with unique features, provide opportunities for drawing subtle distinctions and generating new terms. Examining the current language of speciation and its evolution helps to clarify how views are changing with respect to the geography and genetics of speciation, the importance of different barriers to gene flow, and the role of divergent natural selection in speciation. Here, these issues are viewed through the lens of current language use and in the context of the origin, elaboration, and evolution of both language and ideas.
Much of the original thinking about speciation (and much of the original language of speciation) can be found in the writings of Mayr and Dobzhansky (see Mayr 1963 and Dobzhansky 1970); their views still serve as points of reference for much of the current speciation literature. For readers who did not grow up in the shadow of these giants, it is important to understand their views on the science of speciation, and also the potential biases and prejudices that were introduced because of the ways in which they thought about and described the evolutionary process. Their influence was profound, and many current debates in the speciation literature can be viewed as attempts to revise or replace the “old” ways of thinking that Mayr and Dobzhansky represent. The emergence, over the past decade, of “ecological speciation” as a major theme in speciation research could be interpreted as one such attempt. Ecological speciation emphasizes the central role of divergent natural selection in speciation, often in the context of divergence with gene flow. Less emphasis is placed on allopatric divergence and the role of genetic drift (major themes in Mayr's writing) and on the importance of intrinsic postzygotic barriers (a focus for Dobzhansky and many of his academic descendents).
Although the writings of Mayr and Dobzhansky are still often cited, the last decades of the 20th century ushered in major innovations in the way we think about speciation and barriers to gene exchange. The study of hybrid zones, as “laboratories” for studying speciation, contributed new data, new insights, and new methods of analysis (Harrison 1993). Many well-studied hybrid zones represented interactions between recently diverged lineages that were reproductively isolated, often by multiple barriers. During the same period, a cadre of Drosophila researchers took advantage of tools available for this model organism to focus on the genetics of postzygotic barriers. Their efforts resulted in an elaboration of the Dobzhansky–Muller model for genic incompatibilities and explanations for Haldane's Rule (Coyne and Orr 2004). Sympatric speciation, so vigorously opposed by Mayr, came to be accepted (sometimes grudgingly) as a possible mode of speciation (Coyne and Orr 2004; Gavrilets 2004). Neither natural selection nor sexual selection was ignored during this period; indeed, a major role for genetic drift in speciation, popular with Mayr, had few adherents after the mid-1980s. In spite of new ideas and new data, the language of speciation remained relatively stable. New terms appeared, but few had much staying power. In contrast, the language of speciation has seen dramatic changes in the 21st century, including new definitions for old terms and the introduction of new terms. Why has this happened and what are the implications for our understanding of the speciation process?
The Current Speciation Literature: Words and Ideas
This Perspective is not intended to be a comprehensive review of all terms in the speciation literature. Nor does it include any discussion of the persistent controversy over the “language of species,” (what they are, how they can be defined, etc.; see Harrison 1998, de Queiroz 2007). The focus is on four major discussions or debates in the current speciation literature: (1) the importance of geography and/or gene flow as the context in which speciation proceeds; (2) the classification of barriers to gene exchange and how the recent focus on “ecological speciation” has influenced our thinking about such barriers; (3) the idea of the genome as a mosaic of different evolutionary histories and the nature of that mosaic; and (4) the importance of speciation via hybridization (see Table 1 for a summary of terms discussed in this paper).
|Term||Equivalent/overlapping terms||Related terms, specific examples|
|Geography||allopatric speciation||geographic speciation, m=0||peripatric, centrifugal, founder effect|
|parapatric speciation||semi-geographic speciation, 0<m<0.5, divergence with gene flow||divergence along environmental gradients|
|sympatric speciation||m=0.5, divergence with gene flow||mosaic sympatry, hybrid speciation|
|Gene Flow||potential gene flow||dispersal, migration|
|realized gene flow||effective gene flow, introgression|
|barriers to gene flow||physiological/geographic, intrinsic/extrinsic|
|intrinsic barriers||reproductive isolation, isolating mechanisms||pre-mating, post-mating and pre-zygotic, post-zygotic|
|post-zygotic barriers||reduced viability or fertility||environment-dependent or –independent|
|ecological isolation||potential mates do not meet||temporal isolation, habitat/resource isolation|
|habitat isolation||differential adaptation, habitat preference||ecogeographic isolation, immigrant inviability, isolation by adaptation|
|Genetics||genetic architecture||number, distribution, effect size of genes||speciation gene, barrier gene|
|genic view||gene as unit of evolution||genealogical discordance, FST outlier, differential introgression, semi-permeable species boundary|
|islands of speciation||genomic islands of divergence, genetic mosaic, mosaic genome, heterogeneous genome divergence||divergence hitchhiking|
|Hybridization||hybrid||individual of mixed ancestry||F1 hybrid, backcross|
|hybrid zone||secondary contact, primary intergradation||mosaic hybrid zone|
|hybrid speciation||mosaic genome speciation, hybrid trait speciation||allopolyploidy, homoploid recombinational speciation, introgressive hybridization|
GEOGRAPHY AND GENE FLOW
The names associated with models of speciation have historically reflected spatial distributions and the role of geography. In the beginning, there was geographic (allopatric) speciation, semigeographic speciation, and nongeographic (sympatric) speciation (Mayr 1942). Over the subsequent decades, more refined or taxon-specific models emerged, describing speciation as peripatric (Mayr 1982), centrifugal (Brown 1957), stasipatric (White 1969, 1978), microallopatric (Smith 1965; Paulay 1985; see Fitzpatrick et al. 2008), or allosympatric (Coyne and Orr 2004; Mallet 2005). Like Mayr's original triumverate, these terms focus on the geographic context in which populations become differentiated. From the outset, however, it was evident that geography was a proxy for gene flow. Allopatric speciation was recognized as a default model, because geographically isolated populations are not connected by gene flow, and both genetic drift and natural selection will inevitably result in divergence. Semigeographic speciation (origin of species gaps in zones of intergradation) represented what we now call parapatric speciation or differentiation along an environmental gradient (also called primary intergradation). Mayr (1942) was doubtful that this type of speciation was likely, because “continuous gene flow in the zone of intergradation” would prevent the development of reproductive isolation. Instantaneous sympatric speciation (e.g., hybridization producing polyploid or parthenogenetic lineages) was acknowledged to occur, but gradual divergence in sympatry was considered unlikely (or impossible) because it would be prevented by the “disturbing inflow of alien genes” (Mayr 1963).
In recent years, models based on geographic context have gradually morphed into models based on the extent of gene flow. One motivation for change is that a classification based on geography “divides a continuum into discrete categories” (Butlin et al. 2008), thereby placing undue emphasis on the extremes and potentially diverting attention from more general models of divergence with gene flow (Rice and Hostert 1993; Fitzpatrick et al. 2008, 2009). Allopatric and sympatric speciation can be seen as ends of a continuum from m= 0 to m= 0.5, where m is the fraction of individuals exchanged between two populations each generation. m= 0 represents no gene exchange (allopatry), whereas m= 0.5 represents panmixia or random mating (sympatry). Given an increased reliance on coalescent models for inferring the recent history of divergence with or without gene flow (Wakeley and Hey 1997; Hey and Nielsen 2004; Wakeley 2008, 2010; Hey 2010) it makes good sense to frame models of speciation in terms of gene flow rather than spatial distribution. However, the translation of spatial models into gene flow models is not always straightforward (Fitzpatrick et al. 2009); one consequence is that there can be considerable awkwardness in how terms are now used.
The term “parapatric” provides a good example. Parapatric initially (and correctly) implied a particular spatial distribution of two diverging/divergent lineages; it implied that distributions abut, that the taxa are contiguous and not broadly overlapping, although some modest overlap at the boundary is possible (Smith 1965; Endler 1977). As description of speciation models has shifted from geography to gene flow, parapatric has come to represent 0 < m < 0.5, that is, it occupies the entire continuum between the allopatric and sympatric endpoints (Gavrilets 2003, 2004; Butlin et al. 2008; Mallet et al. 2009). This usage is inappropriate, because many different spatial distributions can result in intermediate levels of gene flow (0 < m < 0.5). A solution to this problem is to jettison parapatric as a descriptor of speciation (except for those circumstances in which divergence actually occurs in a contiguous series of populations along an environmental gradient), and to recognize that speciation can occur when m= 0 (divergence without gene flow) and that speciation can also occur (although with less certainty) when m > 0 (divergence with gene flow). Fitzpatrick et al. (2008) are strong advocates of this approach, suggesting that divergence-with-gene flow is the most general model (which it is) and perhaps the most common process (although divergence in the absence of gene flow may well represent the majority of speciation events).
A number of theoretical models predict that under certain circumstances, divergence and speciation can occur when initial conditions include m= 0.5 (see Coyne and Orr 2004; Gavrilets 2004). Whether the assumptions of such models are often met, and whether divergence in the face of random mixing occurs, remains controversial. The debate about the reality of sympatric speciation continues, and there is a steady stream of papers that invoke sympatric speciation or imply that it is a likely scenario to explain observed patterns of variation (e.g., Barluenga et al. 2006; Savolainen et al. 2006; Herder et al. 2008; Forbes et al. 2009; Crow et al. 2010; Michel et al. 2010).
But even the term “sympatric” can be ambiguous, especially when one contrasts biogeographical and population genetic concepts of sympatry (Fitzpatrick et al. 2008; Coyne 2011). The argument that an appropriate classification of speciation modes might be based solely on gene flow has engendered considerable unhappiness among proponents of sympatric speciation, who suggest that “the classical argument about whether sympatric speciation (in its original, spatial sense) is common in nature is inadequately addressed by this non-spatial, demic definition” (Mallet et al. 2009; see Harrison 2011). But it is germane to ask why the “classical argument” should be resolved rather than abandoned. I can understand why those who work on systems that have historically been portrayed as models of sympatric speciation are invested in seeing their models vindicated. The struggle against the Mayrian dogma of NO sympatric speciation has been long and arduous; if the framework for discussing modes of speciation is fundamentally altered, then the debate becomes focused not on endpoints (sympatric vs allopatric), but on amounts of gene flow during divergence. Perhaps recognizing the dilemma, Mallet et al. (2009) redefine sympatry (broaden its meaning) to include situations in which populations overlap in spatial distribution, but specialize (within the area of overlap) on different resources or different habitats. They call this phenomenon “mosaic sympatry” in contrast to “pure sympatry,” only the latter representing cases where m= 0.5. This redefinition raises several concerns. First, it implies that sympatric speciation includes cases for which the initial conditions are m < 0.5, violating an assumption that many evolutionary biologists have made about what must be the starting point for sympatric speciation. Second, there is no explanation for how (in what context) populations have come to specialize on different resources or habitats; if mosaic sympatry is the starting point for sympatric speciation, then did specialization arise in allopatry or are there obvious trajectories by which populations move directly (without a phase during which gene flow is reduced) from pure to mosaic sympatry? Only in the latter case are we really observing sympatric speciation (sensu stricto).
The notion of mosaic sympatry is not new; indeed, the pattern is clearly represented in the phenomenon of mosaic hybrid zones (Harrison 1986, 2011; Howard 1986; Harrison and Rand 1989). The latter term has been frequently used to describe situations in which diverged/diverging lineages co-occur and exhibit a patchy or mosaic distribution. In most cases, the patchy distribution is not a chance event, but is determined by a habitat or resource template. (Patchy distributions can arise by chance, through a combination of random colonization and hybrid unfitness; such situations have been termed “mottled hybrid zones”[Hauffe and Searle 1998].) Mosaic sympatry may well be a common phenomenon in nature, but it is frequently the result of divergence in allopatry (adaptation to different resources or habitats) followed by secondary contact in a patchy environment. It surely cannot be argued that mosaic sympatry is evidence of sympatric speciation. For example, the mosaic distribution of the host races of pea aphid (on clover and alfalfa), often cited as a possible case of sympatric divergence (Via 1999, 2001), would appear to represent a secondary mosaic hybrid zone in North America, where both host races, already differentiated, have been introduced (Harrison 2011). It is certainly possible that divergence is ongoing (in sympatry), but affiliation with a particular host was already present in the European ancestors of the clover and alfalfa races of pea aphid (Peccoud et al. 2009). Therefore, using a strict definition of sympatric speciation, it is probably not possible to say anything about whether initial divergence was allopatric or sympatric. Indeed Via now seems to play down the geography of host plant divergence, arguing that the “conditions of the initial split are far less important to the study of speciation than is the fact that divergent selection currently maintains genetically based phenotypic differentiation” (Via and West 2008). For many evolutionary biologists, however, the conditions that allow initial divergence remain the focus of speciation research.
Of course, reliance on population genetic (rather than spatial) models of speciation does not guarantee freedom from ambiguity. Coyne and Orr (2004) suggest that the “central problem of speciation is understanding the origin of those isolating barriers that actually or potentially prevent gene flow in sympatry” (p. 57). Studying speciation is all about understanding gene flow and how properties of genomes, organisms, and environments lead to restrictions in gene flow. And yet the term “gene flow” itself has more than one definition. In some texts, gene flow is simply defined as the movement of alleles from one population or place to another (Freeman and Heron 2007). But other definitions suggest that gene flow must involve “incorporation” of alleles from one population into the gene pool of another population (Endler 1977; Futuyma 2009). Endler (1977, p. 20) makes very clear that gene flow is not equivalent to dispersal or migration, because neither of these “necessarily leads to entry of genes or gene arrangements in a given gene pool.” These definitions emphasize the important distinction between potential gene flow (arrival of an individual from a different population) and realized gene flow (incorporation of immigrant alleles into the recipient gene pool).
If the migration rate m is defined as the fraction of individuals exchanged per generation, then m is a measure of potential gene flow (= dispersal), not realized gene flow. This definition of m is silent about whether individuals survive and reproduce when they arrive in the new population. However, in a purely neutral model, m is also a measure of realized gene flow, because immigrants will have a probability of successful mating/reproduction identical to that of the resident individuals. When m is estimated from genetic data, we are, in fact, estimating historical realized gene flow, factoring in selection as well as all other barriers to gene exchange between members of the two populations. In examining patterns of variation for a single trait or genetic marker, we are also estimating realized gene flow for a particular genome region, and estimates will almost certainly vary across the genome (see below).
Realized gene flow is equivalent to “introgression,” a term first introduced to describe patterns of variation seen in plant hybridization. Introgression (or introgressive hybridization) was defined as “the infiltration of germ plasm from one species into another through repeated backcrossing of the hybrids to the parental species” (Anderson and Hubricht 1938). Patterns of introgression have been much discussed in the botanical literature, as has the significance of introgressive hybridization for the origin of adaptations and new species (more on this below). Patterns of differential introgression across hybrid zones have also been a subject of intense interest, because such patterns provide insights into the genetic architecture of reproductive isolation (more below; Harrison 1990; Payseur 2010). Introgression is more often used when realized gene flow is between species or subspecies, but the term is appropriate regardless of the taxonomic rank of hybridizing entities (Anderson 1949; Rieseberg and Wendel 1993).
Finally, it is important to note that spatial distributions can change over time; the geographic (and gene flow) context in which populations diverge is likely to be dynamic. Elsewhere (Harrison 2011), I have proposed that we should think about the amount of gene exchange at the time divergence begins (before any intrinsic barriers arise) and about how extrinsic barriers to gene flow subsequently change over time. In this context, we can then consider “pure” models (in which extrinsic barrier strength remains constant), but also compound models in which extrinsic barrier strength varies temporally. Divergence in allopatry followed by secondary contact would represent such a model; more complex versions are probably realistic, for example, populations of many taxa in North Temperate regions have probably experienced repeated cycles of isolation and contact (Hewitt 2000, 2001, 2011).
BARRIERS TO GENE FLOW
Closely related species often exhibit multiple “axes of differentiation,” and identification of “speciation phenotypes” (traits for which divergence results in intrinsic barriers to gene exchange) must be the starting point for studies of speciation (Shaw and Mullen 2011). Quantifying the importance of individual barriers and the order in which they arose remains a central (and often elusive) goal of speciation research (see Ramsey et al. 2003; Dopman et al. 2010, for examples). Barriers to gene exchange may rarely be selected directly (the process of reinforcement is a clear exception), but often are an indirect result of divergence between populations, driven either by selection or drift (Shaw and Mullen 2011).
The basic terminology associated with classifying barriers to gene exchange derives (as does so much in the speciation literature) from Dobzhansky (1937a,b) and Mayr (1942). Dobzhansky (1937b) introduced the term “isolating mechanism,” which he subdivided into “two large categories, the geographical and the physiological.” He then further subdivided physiological isolating mechanisms into (1) mechanisms that prevent hybrid zygote production or prevent hybrids from reaching reproductive age and (2) hybrid sterility. This is not the usual prezygotic versus postzygotic division, and presumably reflects Dobzhansky's deep interest in hybrid sterility (In Dobzhansky (1937a), the long chapter on “Isolating Mechanisms” is followed by an even longer chapter on “Hybrid Sterility.”) Mechanisms that prevent formation or development of hybrids of reproductive age are yet again divided into those in which parental forms do not meet (habitat and temporal isolation) and those in which parental forms occur together, but do not mate or, if they do, the hybrids produced are not viable (sexual [behavioral] isolation, mechanical isolation, gametic isolation, and hybrid inviability).
Mayr (1942) agreed with most previous authors, including Dobzhansky (1937a), that barriers could be classified as geographic or physiological (or biological). But he proposed a somewhat different classification, in which isolation could be achieved “because potential mates do not meet … , or because they do not mate even if they meet … , or because they do not produce their normal quota of offspring even though they mate.” That potential mates do not meet could be explained by external factors (geographic isolation) or by ecological differences in situations where ranges overlap. Failure to mate, in spite of meeting, could be due to ethological (behavioral) or mechanical differences, and failure to produce the "normal quota” of offspring could result from gametic incompatibilities, reduced fertility in the parental cross, or reduced hybrid viability. In contrast to Dobzhansky, Mayr (1942) does not even mention hybrid sterility.
By the 1960s, Mayr (1963) and Dobzhansky (1970) had converged on a classification of intrinsic barriers that focused initially on whether barriers are premating (prezygotic) or postmating (postzygotic). This distinction remains the first division in a current classification of reproductive isolating barriers (e.g., Coyne and Orr 2004), although it is now recognized that mating (or spawning) and zygote formation can be separated in time, and that it is also important to recognize barriers that occur after mating but before zygote formation (postmating prezygotic barriers). Thus, gamete isolation, which Dobzhansky (1970) classified as premating or prezygotic, would now be considered a postmating, prezygotic barrier. Premating barriers either prevent potential mates from meeting (habitat and temporal isolation—both considered types of ecological isolation) or prevent successful mating even when encounters occur (behavioral/ethological and mechanical isolation). Postzygotic barriers include reduced fitness (viability or fertility) of hybrids or individuals of mixed ancestry. Although many different terms can be applied to the same phenomenon, the language used in classifying barriers to gene exchange has been, for the most part, uncontroversial. A clear exception is use of the term “isolating mechanism,” which has been much discussed and criticized because the term would seem to imply that barriers arise to isolate, which many do not. If barriers arise as by-products of divergence in allopatry, then “isolating mechanisms” is certainly not an appropriate descriptor, whereas it might be in cases of barriers that arise through reinforcement. Those who doubt the reality of reinforcement have been among the most vocal critics of the term “isolating mechanisms” (see Paterson 1985; Harrison 1998).
Geographic isolation is typically viewed as an extrinsic barrier, often a consequence of historical factors (e.g., a recent vicariance event). However, current distributions reflect ecological as well as historical factors, and thus geographic isolation may reflect intrinsic ecological differences between taxa (Endler 1982; Kozak and Wiens 2006; Sobel et al. 2010). This is not a new idea. Although Dobzhansky (1937a, p. 231) distinguished between two types of isolation, geographic and physiological, he qualified the distinction by suggesting that geographic isolation may be a consequence of “each species [being] attached to the environment (climate, etc.) available in one but not the other region. In this case, we are dealing however with a kind of physiological isolation which is expressed in geographical terms.” Geographic isolation due to adaptation to different habitats or resources, which themselves are geographically separate, has been termed “ecogeographic isolation,” and Sobel et al. (2010) provide a detailed discussion of this term. Ecogeographic isolation is a special form of ecological or habitat isolation, because unlike most barriers, which are studied in sympatry (where they prevent individuals from meeting or mating), ecogeographic isolation characterizes allopatric taxa.
The past decade has witnessed reinterpretation and renaming of barriers that arise from differences in the ability or tendency of organisms to use habitats and resources. The recent proliferation of terms is a result of increased attention devoted to “ecological speciation” and extensive discussion of issues surrounding the role of divergent natural selection in the speciation process (Schluter 2000, 2001, 2009; Rundle and Nosil 2005; Sobel et al. 2010; Harrison 2011).
The introduction of many new terms in the rapidly expanding literature on “ecological speciation” has been motivated by the fundamental observations that an organism well adapted to use particular resources or habitats may perform relatively poorly in other contexts and that hybrids, on average, perform less well than parentals. The result is environment-dependent fitness. Some new terms predict patterns in nature that can be examined empirically. For example, “isolation by adaptation” (Nosil et al. 2008) has been suggested as an analogue of “isolation by distance.” The implication is that increasing ecological distance (however measured) will, on average, result in greater genetic distance, presumably because the more divergent are two habitats or resources, the greater the limitations on gene flow between organisms that occupy those habitats.
Other new terms revise the traditional classification of intrinsic barriers to gene exchange, refining, or subdividing already existing terms. An important modification of the language used in the barrier literature is recognition that there are two types of postzygotic barriers; those that are environment independent (in terms of fitness consequences) and those in which the reduced fitness of hybrids is a function of environment. The former represent what Mayr and Dobzhansky seemed to have in mind, namely the reduced survivorship and/or fertility of hybrid individuals independent of the environment (e.g., disruption of normal somatic or germ line development when two distinct genomes compete to give instructions to the developing organism). Alternatively, the performance of hybrids might be compromised because parentals occupy and are adapted to distinct environments and hybrid individuals perform less well in either of these environments. This phenomenon has been termed environment-dependent postzygotic isolation (Rice and Hostert 1993) or ecologically dependent postzygotic isolation (Rundle and Whitlock 2001). Coyne and Orr (2004) refer to this as extrinsic postzygotic isolation. The distinction between the two forms of postzygotic barriers highlights a perhaps underappreciated role of the environment in determining barrier strength. Environment-dependent postzygotic isolation is easy to envision when relevant environmental factors are distributed as discrete patch types rather than as a continuous gradient. If an environmental variable is continuously distributed, hybrids may be disadvantaged in some environments but perhaps not across the full spectrum of available environments. Indeed, in some environments, hybrids may be favored (the bounded hybrid superiority model; see Moore 1977).
Another recent addition to the classification of reproductive barriers is the term “immigrant inviability” (Nosil et al. 2005), which refers to the “reduced survival of immigrants upon reaching foreign habitats that are ecologically divergent from their native habitat.”Nosil et al. (2005) suggest that immigrant inviability represents a “major source of reproductive isolation” that has previously been overlooked. They appropriately emphasize that habitat isolation can have two components (preference and performance) and that these should be measured independently. They also suggest that “habitat isolation” has historically been used to refer to isolation that results from habitat choice or preference, and that the reduced fitness of immigrants that find themselves in the “wrong” habitat has received little attention. It is true that Dobzhansky (1937a) and Mayr (1963) (who often seem to play the role of straw men in discussions of the importance of ecology and divergent selection in speciation) did not provide definitions of habitat isolation that clearly distinguish between performance and preference as causative agents.
However, the literature on mosaic hybrid zones (which Nosil et al.  use as one source of data to demonstrate the role of immigrant inviability) includes explicit discussion of preference and performance as alternative explanations for habitat isolation. Thus, analysis of a field cricket hybrid zone, in which two cricket species are associated with different soil types, contrasted “crickets selecting habitats” and “habitats selecting crickets” (Harrison 1986). And in a subsequent review of hybrid zones (Harrison 1990), it was argued that an observed habitat association could be due to intrinsic differences in behavior (habitat preference model) or could occur when genotypes distribute themselves at random but have differential survival and/or reproductive success within the two (or more) habitat or resource patches. The latter appears to describe immigrant inviability. More recently, Coyne and Orr (2004) defined habitat isolation in terms of a genetic or biological propensity to occupy different habitats, and they clearly stated that isolation can arise as consequence of differential adaptation, differential preference, and/or competition.
Differential adaptation implies that individuals of two types each perform better in their home habitat and have reduced survival when they move to or are transplanted to an away habitat. Reciprocal transplant experiments appear to be direct tests of “immigrant inviability,” and their prominent role in the history of evolutionary biology suggests that the barrier described by immigrant inviability was not, in fact, overlooked in the evolutionary biology literature (Sobel et al. 2010). Nosil et al. (2005) use data from reciprocal transplant experiments to provide evidence that habitat association observed for pairs of close relatives is often due to differential adaptation (immigrant inviability). But this evidence for the relative importance of “immigrant inviability” (compared with other barriers) comes from studies selected because they include reciprocal transplant experiments. Such experiments are both difficult and time consuming, and few evolutionary biologists would contemplate doing transplant experiments unless there was already indirect evidence that environment-dependent fitness differences existed. Therefore, the studies cited by Nosil et al. (2005) may not be an unbiased sample for assessing the importance of immigrant inviability.
The phenomenon described by the term “immigrant inviability” is an important component of reproductive isolation in many taxa. For phytophagous insects and for pathogens (fungi, viruses, etc.), a shift to (accidental arrival on) a novel host may result in greatly reduced survivorship, limiting or preventing genetic exchange with residents on that novel host (Nosil 2004; Giraud 2006). The question is not whether immigrant inviability is a real phenomenon. But should this be the term we use to describe the component of “habitat isolation” due to differential adaptation? And what are the implications of introducing this particular term into the classification of barriers to gene exchange?
As mentioned above, Nosil et al. (2005) equate “habitat isolation” with habitat preference and introduce “immigrant inviability” as a complementary term to describe the consequences of differential adaptation. One limitation of this dichotomy is that it contrasts habitat isolation, which is an observed pattern (organisms are associated with different habitats), with habitat preference, which together with differential survivorship (immigrant inviability) is an explanation for that pattern. Furthermore, if “immigrant inviability” is the sole descriptor of differential adaptation to habitats and resources, a number of additional issues arise: (1) In many situations it will be difficult to define exactly what we mean by an immigrant. When environmental variables change continuously (temperature, salinity, depth, moisture, etc.) rather than discretely (patchy resources or habitats), how far along a gradient must an individual move to qualify as an immigrant? And given the possibility that organisms can acclimate, might the consequences of movement between habitats depend on how steep are the gradients and how far individuals move? Individuals transplanted between extremes of a gradient may exhibit low survival, but if gradually moved along the gradient, survivorship may be relatively high. (2) Differential adaptation implies that immigrants have reduced survivorship/reproductive success compared to natives, but immigrants are not necessarily inviable. Differential adaptation is by no means an all-or-none phenomenon. (3) If immigrants have reduced (but not zero) survivorship, then hybridization and introgression may occur; in such cases, we must ponder the fate of immigrant alleles, rather than the fate of immigrant individuals or genotypes? (4) Finally, it is possible that immigrants have reduced survivorship and/or reproductive success, not because of their genotype, but simply because there are physiological costs to dispersing that impact subsequent survivorship and reproductive success. In this case, the relationship between potential and realized gene flow may depend on extrinsic factors (e.g., distance traveled between populations). This is a distinctly different sort of immigrant inviability than that imagined by Nosil et al. (2005).
I would suggest that Coyne and Orr (2004) speak for many evolutionary biologists when they join habitat-associated fitness differences with habitat preference traits as contributors to habitat isolation, a barrier that is itself only one component of ecological isolation. In the context of reproductive barriers, association with a defined habitat or resource is the barrier to gene flow, regardless of whether that association is a result of differences in survivorship within habitats or differences in preferences for habitats. Differential adaptation is central to understanding speciation (as proponents of “ecological speciation” would argue), but the term “immigrant inviability” fails to convey the complexities of how divergent selection may result in habitat or resource isolation.
SPECIES BOUNDARIES AS SEMIPERMEABLE, THE GENOME AS A MOSAIC
In lineages with sexual reproduction, recombination shuffles genome regions among individuals within an interbreeding population. As a consequence, adjacent genome regions often represent different evolutionary histories, a product of ancestral polymorphism, mutation, and random and/or selective lineage sorting. Local selective sweeps lead to reduced variation within populations and increased differentiation among populations. The size of the differentiated region depends on the strength of selection (the rate at which an advantageous variant is swept to fixation) and the rate of recombination (which depends both on genome processes and on frequency of mating between individuals from different populations). Gene flow between populations or species also varies across the genome, a consequence of variation in the strength of divergent selection or the strength of reproductive barriers. In the face of hybridization, some regions become homogenized, whereas others remain distinct.
The notion of the genome as a mosaic is not new; indeed, the classic concept of differential introgression implies a genomic mosaic. In the speciation literature, two papers from the late 1960s introduced this concept. Key (1968), discussing proposed models of stasipatric speciation in Morabine grasshoppers, suggested that tension zones are semipermeable, allowing free passage of some genes or gene arrangements, but constituting a barrier to movement of others. Bazykin (1969), commenting on models of sympatric speciation, introduced the concept of “isolation for part of the gene pool.” These ideas relatively quickly became established in the hybrid zone literature. Invoking Bazykin (1969), Barton and Hewitt (1981) suggested that biological species (defined by interbreeding within and reproductive isolation between) might have to be defined gene by gene. They wrote: “Strict application of the biological species concept (BSC) might lead to different results for different loci; perhaps one can only define “groups of actually or potentially interbreeding natural populations … at the gene level” (p. 119). Harrison (1986, 1990) emphasized that because selection varies across the genome, the barrier strength of a hybrid zone will not be constant for all loci. Species boundaries can be (as in Key 1968) described as semipermeable, a term borrowed from cell biology where it is used to refer to membranes that are selectively permeable, that allow only certain molecules or ions to pass through. Thus, semipermeable applied to the boundary between species was meant to imply that alleles at some loci could move across the boundary, whereas alleles at other loci could not, that is that there is differential introgression. (Note that the term “porous” has also been applied to hybrid zones [e.g., Wu 2001]. Porous simply means permeable, and does not directly convey the notion of heterogeneity in rate across the genome.)
Hampton Carson also referred to heterogeneity in the extent to which genomes could be disrupted by gene flow (Carson 1975). He described what he termed the “open” and “closed” systems of variability, the latter held together by epistatic interactions (and perhaps by physical linkage or inversions). With respect to possible introgression upon secondary contact, he proposed that “the closed variability systems of the two hybridizing species or subspecies may remain intact as the open systems are undergoing recombination” (Carson 1975).
Patterns of differential introgression have frequently been documented for hybrid zones (e.g., Harrison 1990; Rieseberg et al. 1999; Payseur et al. 2004; Payseur and Nachman 2005; Teeter et al. 2008, 2010; Maroja et al. 2009; Macholan et al. 2011), and these patterns provide direct evidence that gene exchange cannot be viewed as a property of the genome as a whole. The early literature relied on morphological, allozyme, and organelle DNA data; multilocus DNA sequence data were not yet available. In many animal hybrid zones, mtDNA appeared to introgress more readily than other markers (Ferris et al. 1983; Powell 1983; Harrison et al. 1987), consistent with the mitochondrial genome segregating independently of all genes in the nuclear genome, where most genes affecting local adaptation and reproductive isolation presumably reside (Barton and Jones 1983). Similarly, evidence from hybridizing plant species provided evidence of “chloroplast capture,” the differential introgression of cpDNA from one species into another (Rieseberg and Soltis 1991; Tsitrone et al. 2003).
With the availability of faster and more efficient technologies for generating DNA sequence and genotype data, the nature of species boundaries and variation in patterns of isolation or differentiation across the genome have received increased attention. A fundamental assumption is that regions that are (and remain) differentiated are candidate regions for including genes that contribute to local adaptation and/or reproductive isolation. Such regions are most obvious when divergence is recent or when ongoing gene flow can homogenize regions that are neutral with respect to adaptation and reproductive isolation.
Multilocus sequence data have provided repeated examples of discordance among gene genealogies of closely related species or subspecies (e.g., Ting et al. 2000, Beltrán et al. 2002; Machado and Hey 2003; Dopman et al. 2005; Putnam et al. 2007; Andres et al 2008; Maroja et al. 2009; White et al. 2009; Lassance et al. 2011). Genealogical discordance often appears to be the result of ancestral polymorphism and lineage sorting, but in other cases it would seem to be a consequence of recent hybridization and differential introgression (Shaw 2002; Ohshima and Yoshikawa 2010). Genome scans using AFLPs, polymorphic microsatellite markers, or SNPs have also been used to document heterogeneity in divergence (e.g., Campbell and Bernatchez 2004; Emelianov et al. 2004; Grahame et al. 2006; Wood et al. 2008; Via and West 2008; Kulathinal et al. 2009; Manel et al. 2009; Galindo et al. 2010; Neafsey et al. 2010; Renaut et al. 2011). One consequence of these new technologies and the data they produce has been a proliferation of terminology to describe both concepts and observed patterns. New terms may be needed to describe the complexities seen in patterns of genome divergence, but some terminology has emerged without appropriate recognition that the ideas represented are not entirely new.
Wu (2001) suggested that evolutionary biologists need to adopt a “genic view” of species and speciation. He argued (contra Barton and Hewitt 1981) that the BSC requires that we view reproductive isolation at the level of the whole organism (or whole genome) and that the literature on speciation had failed to incorporate the perspective of the gene as the unit of evolution. He further suggested that the whole genome view was a consequence of thinking about speciation in terms of the break-up of coadapted gene complexes (e.g., Mayr 1963), with the implication that many genes may be involved. Although late in his review, Wu (2001) acknowledged that the hybrid zone literature “provides some of the best evidence of gene sharing across race or species boundary,” he never explicitly recognized that this literature had, in fact, long ago incorporated the genic view that he espoused. For example, Harrison (1990) had written: “Genetic isolation must be considered as a property of individual genes (or chromosome segments), not as a characteristic of entire genomes” (p. 99).
Data from a variety of taxa now strongly support the idea that the genome is a mosaic, that there may be a limited number of gene regions that are differentiated between recently diverged lineages, and that these regions may be characterized by low recombination (Noor et al. 2001; Rieseberg 2001; Navarro and Barton 2003; Butlin 2005). Genome regions that have become differentiated or remain differentiated in the face of potential gene flow may harbor so-called “speciation genes” or “barrier genes,” and identification of such regions can be a first step in characterizing the genetic basis of speciation phenotypes.
Turner et al. (2005) used microarrays to estimate genome-wide patterns of differentiation between two forms of the African malaria mosquito, Anopheles gambiae. They found three small regions for which the two forms were significantly differentiated, and described these regions as “genomic islands of speciation.” Subsequent work has raised questions about the pattern of differentiation for these mosquitoes (Lawniczak et al. 2010; Neafsey et al. 2010; Turner and Hahn 2010; White et al. 2010); the new data suggest widespread (but heterogeneous) genome divergence, and appear inconsistent with substantial ongoing gene flow. Regardless of the true pattern in A. gambiae, the “islands of speciation” terminology has attracted considerable attention, and the geographic/topographic imagery has been adopted and extended by other authors (e.g., Nosil et al. 2009; Nosil and Feder 2012). Some suggest that a more common pattern may be “archipelagoes” or “continents” of speciation (Michel et al. 2010). This argument follows upon discovery of widespread genomic divergence between a pair of taxa (the host races of Rhagoletis pomonella) that are thought to be examples of recent divergence with gene flow. The debate about island size has also been influenced by discussion of the extent to which recombination will reduce linkage disequilibrium and has led to other new terms—“divergence hitchhiking” (Via and West 2008; see also Nosil et al. 2009) and “genome hitchhiking” (Feder et al. 2012). These terms describe the fact that, in the presence of divergent selection (e.g., in different habitats or on different resources), rates of recombination between diverged populations (either in the vicinity of single genes or across the entire genome) will be less than rates within those populations (e.g., estimated based on map distance/chromosome location and an assumption of random mating). Essentially, linkage disequilibrium will persist longer and over a greater genomic extent because opportunities for recombination (in individuals heterozygous for population-specific markers) are limited by nonrandom mating and/or negative selection (Charlesworth et al. 1997; Via and West 2008; Smadja et al. 2008). Again, the terms are new, but the underlying concepts are not.
The genetic mosaic (Via and West 2008), the mosaic genome (Wang-Sattler et al. 2007), genomic islands of divergence (Nosil et al. 2009), and heterogeneous genome divergence (Nosil et al. 2009) are recent additions to the terminology used to describe the observation that the amount of genetic divergence between close relatives varies across the genome. The causes of such patterns are various, and include the possibility that regions with high levels of divergence represent regions in which embedded genes contribute to reproductive isolation. But, ancestral polymorphism and lineage sorting can lead to heterogeneity in divergence (and higher divergence in regions of restricted recombination), even in the absence of gene flow. Therefore, islands of divergence may have nothing to do with speciation and might be considered “mirages” (Noor and Bennett 2009). In the absence of evidence for differential realized gene flow (differential introgression) it is probably better to avoid “islands of speciation” and rely on more neutral terminology. Certainly not all islands of divergence will turn out to be of interest, but where such islands are identified, they do deserve further characterization. As discussed above, the island metaphor has been widely adopted, and debate has now shifted to issues about island size and island topography, where elevation above some “sea level” reflects the amount of genetic divergence (Nosil et al. 2009; Michel et al. 2010; Nosil and Feder 2012). It is not yet clear that employing detailed landscape imagery to represent genomic patterns will prove instructive.
Heterogeneity in genome divergence can certainly be exploited to identify gene regions that contribute to reproductive isolation. An ultimate goal of such studies is to identify genes that encode phenotypes that are responsible for pre- or postzygotic barriers to gene flow and to understand the speciation process at the molecular level. Recent studies (mostly, but not exclusively, in Drosophila) have been successful in this regard, and the genes thus identified have often been referred to as “speciation genes” (Butlin and Ritchie 2001; Orr et al. 2004). Nosil and Schluter (2011) restrict this term to genes for which new variants contribute to an increase in the strength of the overall barrier to gene flow (i.e., speciation cannot yet be complete when the new variant arises). This strict interpretation is an accurate reflection of what we want to know about the history of differentiation and its impact on gene exchange. But it means that identifying speciation genes requires knowing the order in which speciation phenotypes arose.
Although the term “speciation gene” is a convenient (and appealing) notation, it can be misleading if applied too liberally. A gene known to contribute to a current barrier is not a “speciation gene” if the barrier (phenotype) was not involved in initial divergence and reproductive isolation between two lineages (Shaw and Mullen 2011). Such genes can be viewed as “candidate speciation genes,” but are perhaps more accurately described as “barrier genes” (Noor and Feder 2006), because they determine phenotypic variation that contributes to current reproductive barriers. The term “speciation gene” is also misleading because it seems to imply that the identified gene may contribute to reproductive isolation in many taxa (that the gene is of general interest for speciation rather than limited in its effect to the focal taxon) (Shaw and Mullen 2011). Thus, what might be viewed as a debate about fine points of nomenclature reflects very real difficulties associated with identifying relevant phenotypes and inferring the history of lineages from observations of current patterns of variation.
The role of hybridization in the evolutionary process, and especially its role in the origin of species, has been a persistent source of disagreement within the evolutionary biology community. Historically, botanists and zoologists have had conflicting views. Botanists noted that plant hybrids are common and viewed hybridization as playing an important role in evolution (Anderson and Stebbins 1954; Stebbins 1959; Grant 1981). In contrast, zoologists often thought of hybridization as relatively rare in nature and with limited impact on the evolutionary process. For many zoologists, hybridization represented a “breakdown” of reproductive isolation, although hybrid zones could be sites where barriers to gene flow would be perfected (reinforcement). More recently, however, some evolutionary biologists have promoted the role of hybridization in speciation and diversification, arguing that hybridization is both more common and more constructive than we have acknowledged (e.g., Arnold 1997, 2004; Mallet 2007).
What is a hybrid? Some would restrict the term hybrid to the product of interspecific crosses; others see virtually all crosses as “hybrid” because, in most sexually reproducing populations, all individuals have unique genotypes. Indeed, it has long been recognized that there is a clear continuum of “hybrid” crosses. A hundred years ago, Mitchell (1910) wrote that “Hybridism therefore grades into mongrelism, mongrelism into cross-breeding, and cross-breeding into normal pairing, and we can say little more than that the success of the union is more unlikely the further apart the parents are in natural affinity.”
Hybridization events can result in the immediate reproductive isolation of the hybrid progeny (what some would call “saltational” speciation); examples include polyploidy in plants (backcrossing of F1× parental produces sterile offspring) and parthenogenesis in lizards (hybrids are asexual). But in the absence of immediate barriers, what makes hybrids of interest is that they are heterozygous at two or more loci, thereby providing a platform for recombination to give rise to novel allelic combinations, which may be the source of novel traits (and by extrapolation, new species). In this case, the focus is on “recombinational” homoploid speciation (the origin of new species through the production of a population of stabilized recombinants, in the absence of the immediate barrier provided by polyploidy). A “hybrid species” must be a coherent entity, deriving genetic material from two (or more) parents, and reproductively isolated from the parental species. Hybrid populations are indeed common in animals as well as plants, but in many cases these populations reflect recent or ongoing gene exchange between two species, without evidence that these hybrid populations are reproductively isolated. In fact, from patterns of variation alone, it is very difficult to distinguish between hybridization resulting in a clinal or mosaic hybrid zone, and hybridization producing an entity that is reproductively isolated from both “parents.”
In plants, homoploid hybrid speciation may be relatively rare compared to allopolyploid hybrid speciation, but it is still considered to be an important mechanism giving rise to new species (Mallet 2007; Abbot et al. 2010). Sunflowers (Helianthus) provide classic examples of homoploid hybrid speciation; three different xeric-adapted hybrids are ecologically isolated from their mesic-adapted parents (Rieseberg et al. 2003). In animals, the frequency of homoploid hybrid species is not yet well defined, but several recent studies have identified “entities” that are cited as the result of homoploid hybrid speciation (e.g., Schwarz et al. 2005; Gompert et al 2006; Mavarez et al. 2006; Nolte et al. 2009; Nolte and Tautz 2009). In the case of Heliconius butterflies, a new species is described as having arisen by “hybrid trait speciation,” in contrast to “mosaic genome speciation” (Jiggins et al. 2008; Salazar et al. 2010). The distinction rests, in part, on defining in what proportion genomes are mixed to produce a hybrid species. In hybrid trait speciation, a few genes (gene regions) have introgressed, resulting in phenotypically distinct individuals that are reproductively isolated from either parent (Jiggins et al. 2008). In what has traditionally been viewed as homoploid hybrid speciation (mosaic genome speciation), and so elegantly demonstrated in Helianthus, the two genomes are more evenly mixed (e.g., Rieseberg et al. 2003; Gross and Rieseberg 2005). The proportion of the “hybrid” genome derived from each of the two parents may vary widely (Mallet 2007), as may the patterns of mating that have given rise to the hybrid species. Homoploid hybrid species may “condense” out of a hybrid swarm, or they may result from introgression of a small number of genome regions. In the first case, the two genomes are thoroughly mixed in a hybrid population, and a “stable” recombinant type presumably emerges from the many possible genotypes that are formed. In the latter case, hybridization may be far more limited, and backcrossing to one of the parental forms results in selective introgression of a small number of genes or gene regions. So hybrid trait speciation and mosaic genome speciation may represent rather different outcomes within a continuum of possible recombinant genomes.
Scientific language naturally evolves greater complexity and specificity as scientific disciplines expand and mature. It is no surprise that the language of speciation has become increasingly complex as our understanding of the behavioral, ecological, physiological, and genetic basis of the speciation process has become more sophisticated. The introduction of new terms often reflects an improved understanding of pattern and process, but new terms must be carefully defined relative to existing vocabulary. Proponents of new terminology must not only be absolutely clear about how and in what context terms are to be used, but they also must provide a convincing rationale for why existing terminology is not sufficient. Sometimes new terms are accurate descriptors of observed patterns, but what they imply about underlying process may differ in subtle (or not so subtle) ways from existing terminology that describes the same pattern. Modifying the definitions of old terms can also be a logical consequence of new perspectives on the evolutionary process, but such revisions have the potential to confuse as well as enlighten.
The term “mosaic sympatry,” describes a spatial/ecological pattern that some have called microallopatry and others have recognized as mosaic hybrid zones. Each of these terms may describe the same pattern; the difference in terminology reflects different interpretations of how such patterns arise and how spatial scale relates to gene flow. Other new terms have recently appeared in the speciation literature to subdivide older, broader categories (e.g., the distinction between environment-dependent and environment-independent postzygotic barriers). This can be a legitimate exercise, but requires that essential differences between the new terms be clearly defined. Some new terms appear to reflect already existing ideas, which may not have been blessed with a catchy name. This would seem to be true for the “genic view” of species and speciation, a view that preceded by several decades the use of the term. Similarly, the term “immigrant inviability” appears to describe a component of habitat isolation (differential adaptation) that has long been recognized as important (Sobel et al. 2010). Immigrant inviability is an apt descriptor of barriers operating in some systems (especially in systems where alternative specialists are adapted to different hosts), but it seems inappropriate as a general term to describe differential adaptation.
Many concepts recently given new names have been in the hybrid zone literature for decades (see Bierne et al. 2011). For example, students of hybrid zones have often described patterns of differential introgression and semipermeable species boundaries; these terms both clearly imply a genic view of species and recognize that the genome is likely to be a mosaic of different histories. Similarly, studies of mosaic hybrid zones have focused attention on the importance of a heterogeneous habitat or resource template and on the role of divergent selection implied by such an environment. And, of course, hybrid zones are in fact locales where divergence with gene flow may be occurring. Ideas about mosaic genomes, differential adaptation, and divergence with gene flow are now much discussed in the speciation literature, but without sufficient recognition that they have a long history. Many evolutionary biologists still seem to use the writings of Mayr and Dobzhansky as the reference with which 21st century views of speciation should be contrasted.
Proponents of ecological speciation and divergence with gene flow often study pairs of races, strains, or species that differ in only one or a few ecological traits, traits that almost certainly vary as a result of environmental heterogeneity and divergent natural selection. Observed trait differences directly or indirectly result in reproductive isolation. The relative simplicity of these systems combined with new opportunities to unravel the genetics and genomics of trait differences is providing important new insights into the role of divergent natural selection as a cause of speciation. But, like Mayr and Dobzhansky before them, members of the current “schools” tend to see the world through the eyes of the organisms they work on. Their language reflects their point of view. As new data and new insights accumulate, we would all do well to acknowledge that our views on speciation, as well as the language we use to describe the process, can be substantially biased by our experience. We should also recognize that the speciation process, the source of remarkable biodiversity, is probably as diverse in its characteristics as the lineages it produces.
Associate Editor: T. Craig
I thank current and past members of the Harrison lab for many discussions that helped to clarify my thoughts and for allowing me to see the evolutionary process through the eyes of many different organisms. Comments on an earlier version of the manuscript from current members of the Harrison lab, the Kerry Shaw lab, three colleagues (M. Noor, M. Ritchie, and D. Schemske), and two reviewers forced me to reexamine some of my views. S. Mullen and M. Kronforst provided input that greatly improved the current version. However, the opinions expressed here are distinctly my own. National Science Foundation (NSF) and United States Department of Agriculture (USDA) have generously supported my research on field crickets and European corn borer, model systems that have been central to the development of my ideas.