Goldilocks Meets Santa Rosalia: An Ephemeral Speciation Model Explains Patterns of Diversification Across Time Scales

1How often do new species form? Studies of plant and animal speciation rates have focused on different species, over different time scales, using different methods (e.g., Simpson; Givnish; Coyne and Orr ). Here we review our current understanding of speciation rates, focusing on reconciling what we know from studies of speciation based on the fossil record, phylogenies, and mathematical models. We find that speciation rates estimated from these different studies can be contradictory, with some rates clearly much faster than others. Given that some rates seem “too slow” and some seem “too fast”, is there a single framework that predicts rates that are “just right”? To reconcile results from different approaches, we highlight an “ephemeral speciation model”, under which new species form frequently but rarely persist.

2The most common way to estimate rates of speciation is to use data from the fossil record. Paleontological studies estimate how many new species formed over a given time interval (the per lineage speciation rate; e.g., Stanley; Van Valen; Jablonski; Hulbert; Sepkoski ). Many paleontological estimates of speciation rates have been calculated from groups with reasonably complete fossil records like marine invertebrates (e.g., Raup and Sepkoski; Peters ). For groups with incomplete fossil records, sophisticated statistical analyses can correct for incomplete sampling (e.g., Alroy et al.; Foote; Ezard et al. ). Paleontological approaches for inferring speciation rates use direct information about species that lived in the past, whereas other methods (discussed below) must infer the past dynamics indirectly. The primary limitation of paleontological studies is the uncertainty that arises from gaps in the fossil record (e.g., limited specimen material, uneven sampling effort, and/or taphonomical biases; Raup; Sepkoski; Alroy et al. ). Furthermore, paleontological studies generally rely on higher taxa (Valentine ), and how these taxonomic groups are defined can have a profound influence on inferred speciation rates (Ezard et al. in press).

3Paleontological studies suggest that speciation rates vary widely both through time and across lineages. Most paleontological estimates of per-lineage speciation rates range from 0.01 to 10 speciation events per lineage per million years (Sepkoski ); Sepkoski suggests a “canonical” estimate of 0.3 speciation events per lineage per million years. Even though this rate varies tremendously both across taxa and through time (Sepkoski ), it can serve as a rough but useful “benchmark” for comparisons across datasets.

4Speciation rates can also be inferred from phylogenetic studies of extant species. This approach requires phylogenies whose branch lengths have been scaled to time. The simplest way to calculate speciation rates from these trees is to compare species richness to clade age, which provides a minimum bound on the rate of speciation, assumed to be constant (Magallon and Sanderson; but see Rabosky, who warns against this approach). Another approach is to fit models (i.e., birth–death models) to phylogenetic branch lengths and estimate rates of speciation and potentially extinction (reviewed in Nee ). Using phylogenies of extant species takes advantage of the wealth of data from the tree of life (Hedges and Kumar ). However, phylogenetic trees are estimated with error, do not include direct information about extinct species, and suffer from a number of biases related to sampling, all of which can affect speciation rate estimates (e.g., Revell et al.; Phillimore and Price; Rabosky; Cusimano and Renner; Brock et al. ).

5Despite these potential caveats, most phylogenetic studies of speciation rates recover estimates that are of the same order of magnitude as the fossil record. One meta-analysis of speciation rates estimated from phylogenies found rates that ranged from 0.01 to 10 speciation events per lineage per million years under a pure birth model (McPeek and Brown ). An approach that estimated both speciation and extinction rates simultaneously recovered speciation rates of the same order of magnitude (e.g., Alfaro et al. ).

6In contrast to the studies discussed above, many studies of young evolutionary radiations estimate very high rates of speciation. These estimates often come from adaptive radiations in insular habitats (e.g., islands and lakes; see Losos and Ricklefs,, and chapters therein). For example, one of the best-known examples of rapid diversification is the cichlid fishes in lakes of the African Rift Valley (Seehausen ). Estimated speciation rates for cichlid species flocks are quite high. An extreme example occurs in Lake Victoria, where speciation rates might be as high as 400 speciation events per lineage per million years (i.e., the formation of ~450 species in ~15,000 years, Johnson et al.; Genner et al.; but see Day et al. ). Other studies suggest that similarly high speciation rates may occur in other systems [e.g., Hawaiian drosophila (Coyne and Orr ), silverswords (Baldwin and Sanderson )]. In fact, there are several well-known adaptive radiations that have occurred so quickly, it is extremely difficult to infer the true phylogenetic relationships among species [e.g., Galapagos finches (Sato et al.; Petren et al.; Burns et al.; Grant and Grant ), three-spined sticklebacks (reviewed in Schluter )].

7Concordant with studies of young evolutionary radiations, mathematical models of speciation also suggest that speciation rates can be quite high. Mathematical models of speciation take a number of forms, including models of sympatric speciation (e.g., Maynard-Smith; Felsenstein; Dieckmann and Doebeli; Doebeli and Dieckmann ), models of divergence with gene flow (e.g., Wu; Hausdorf ), models of speciation in allopatry (e.g., Gavrilets ), and metapopulation models of adaptive radiations (e.g., Gavrilets and Vose ). Modeling approaches allow for a mechanistic understanding of how specific parameters influence the process of speciation, but are obviously limited by their simplifying assumptions. For example, models of speciation make specific assumptions about population structure, genetic architecture, and the strength and type of selection. Additionally, mathematical models must necessarily assume a relatively simplistic species concept. Model assumptions and parameter values can have dramatic impacts on the dynamics of speciation models (reviewed in Gavrilets ).

8Mathematical models of speciation almost never focus on speciation rate per se, and instead generally ask whether or not speciation occurs and, if so, how long it takes from start to finish (i.e., the speciation “transition time”, Coyne and Orr ). Transition times estimated from mathematical models are typically very fast (e.g., Doebeli and Dieckmann, but see Orr and Turelli ). However, transition times are not directly related to speciation rates, which describe the time from one speciation event to the next (i.e., the speciation “waiting time”). One modeling study that focused explicitly on speciation rates found that when speciation occurred, the waiting time for speciation varied between 5,000 to more than 200,000 generations depending on model parameters (Gavrilets ). For organisms with a generation time of 1 year, this corresponds to speciation rates of 2–200 speciation events per lineage per million years. These rates are comparable to the highest speciation rates observed empirically in the young evolutionary radiations discussed above, and likely represent an upper bound for speciation rates. It is difficult to define a lower bound for speciation rates from mathematical model because these models typically do not focus on parameter values where speciation never happens.

9Speciation rates estimated from studies at different time scales suggest a contradiction. Mathematical models of speciation and studies of young evolutionary radiations find that new species can form quickly and often. However, phylogenetic studies over longer time scales and paleontological studies find that new species usually form more slowly. What explains this apparent discrepancy in speciation rates across different types of studies?

10Here we call attention to an explanation that may help unify what we know about speciation rates from paleontological, phylogenetic, and mathematical studies: the “ephemeral speciation model”. It is possible that speciation is very common and very rapid, but that the new species produced almost never persist. Therefore we suggest that some approaches (e.g., studies of speciation in action and mathematical models) actually focus on the formation of ephemeral species while others (e.g., phylogenetic studies over longer time scales and paleontological studies) focus on the persistence of these ephemeral species. Instead of being a contradiction, these differences in speciation rates reflect two aspects of the same underlying model: the formation and the persistence of ephemeral species.

11The concept of fragile new species has deep historical roots. The idea that many more incipient species form than persist traces back to Mayr. Stanley also discussed this phenomenon, referring to these failed incipient species as “aborted species” (Almon ). Levin proposed a related idea for plants where many incipient species form and have differential survival (i.e., “isolate selection”). Other authors have discussed characteristics necessary for “successful speciation” (e.g., geographic range expansion and ecological niche differentiation (Price; Rundell and Price ). Hubbell’s neutral theory of ecology also exhibits high turnover of young “incipient” species (Hubbell; Rosindell et al.; Etienne and Rosindell ). Our suggested name—“the ephemeral speciation model”—takes inspiration from Futuyma’s ephemeral diversification model (which focuses primarily on trait change: Futuyma, ).

12Under an ephemeral speciation model, new incipient species are constantly forming at a high rate (Fig. 1). Recent research on mechanisms of speciation suggest that speciation can occur via a plurality of interacting mechanisms (e.g., geography, selection, genomic rearrangements; see Coyne and Orr; Gavrilets ). Thus the conditions for some mode of speciation may often be met in natural populations. Although speciation occurs frequently in the ephemeral speciation model, persistence of incipient species is rare. The lack of persistence could be due to extinction or “reabsorption” via hybridization of the incipient species (Seehausen et al.; Seehausen; Taylor et al.; Richmond and Jockush; Behm et al. ). For example suppose that in some clades speciation typically happens in small allopatric populations at the edge of a species range (Mayr ). These new species will be very vulnerable to extinction and to changes in the conditions that maintain reproductive isolation (Mayr ). New species are also likely to be fragile under other non-allopatric modes of speciation as well [e.g., new polyploid species have very small population sizes (Holloway et al. ) and new ecological species require continued divergent selection until other forms of reproductive isolation evolve (Nosil and Sandoval )]. Therefore failed speciation is common because speciation takes time to complete and because conditions change over time.

13The ephemeral speciation model is consistent with two key empirical observations. First, what taxonomists recognize as species are often comprised of many incipient forms. Species typically show extensive genetic and phenotypic variation, and this variation is usually hierarchically structured (Avise; Bickford et al.; Mallet ). These incipient forms are recognized taxonomically by a variety of names (e.g., geographic races, subspecies, incipient species; Simpson; Mayr ). Here we follow the general lineage concept of species, which encompasses many different species concepts as points along a continuum and which is consistent with the idea that species themselves can be a collection of distinct lineages (De Queiroz ). Second, new species can be unstable. When incipient species form, they can go extinct or cause their parental forms to go extinct (e.g., Hegde et al. ). Additionally, incipient species can collapse and be reabsorbed by their parental form (Mallet ). One recent example is the collapse of a stickleback species pair in Enos Lake (Taylor et al.; Behm et al. ), but similar collapses of incipient species have been observed in cichlids (Seehausen ); trout (Bettles et al. ), Galapagos finches (Grant and Grant, ); skinks (Richmond and Jockusch ) and house spiders (Croucher et al. ).

14One way to model ephemeral speciation over long time scales is to use a simple high turnover birth–death model with high speciation and high extinction rates (e.g., Alfaro et al. ). These high turnover models do a good job of predicting species diversity in clades of intermediate age [e.g., jawed vertebrates (Alfaro et al. ) and ray-finned fishes (Santini et al. )]. But high turnover birth–death models are inconsistent with empirical data in three ways. First, high turnover models predict that species deep in the tree should accumulate exponentially through time (i.e., linear lineage through time plot, Nee et al. 1992), but this pattern is not commonly seen in phylogenetic trees (Rüber et al.; Phillimore and Price ). Second, high turnover models suggest that the persistence of incipient species should be random, but empirical work on persistence of populations across a species range suggests that population survival is not typically random with respect to both biotic and abiotic factors (e.g., Levin,; Owens et al.; Jones et al.; Cardillo et al. ). Third, in high turnover models, species fail to persist only because they go extinct, but incipient forms can fail to persist because they are reabsorbed. Therefore, more elaborate models of ephemeral speciation are needed.

15We suggest that ephemeral speciation models should be hierarchical (Gould ) where incipient forms are constantly forming and dissolving inside larger entities (i.e., species, Fig. 1a). There are several existing models that consider a hierarchical process where “incipient” forms arise and go extinct at a higher rate than “full” species (Cadena et al.; Pons et al.; Phillimore et al.; Pigot et al. ). The most general example is Phillimore et al. who find that the rate of bird “subspeciation” is between 30 and 40 times higher than the rate of speciation (see also Martin and Tewksbury ).

16There are several important consequences of hierarchical models of ephemeral speciation (Fig. 1). First, species go extinct only when all incipient forms are lost. Therefore the extinction rate is no longer a property of a species but depends on the number of incipient forms within the species and their extinction rate. Second, like other hierarchical models (e.g., Wakeley; Hubbell ), the ephemeral speciation model predicts an uneven distribution of incipient form within species (Fig. 1b): a few species will contain many incipient forms while most species will contain few (consistent with the observation that rare species are common; Lim et al. ). Third, this uneven distribution will lead to differences in effective speciation and extinction rates across species. Species with many incipient forms will have high speciation and low extinction rates compared to species with few incipient forms (see also Kisel and Barraclough ). The unevenness of speciation rates across taxa is consistent with the fact that phylogenetic trees tend to be more unbalanced than birth–death models predict (Mooers and Heard ). Finally, under the ephemeral speciation model it is very likely for entire clades of newly formed species to go extinct. The clades that survive to the present day are disproportionately likely to have undergone a burst of speciation early in their history (Phillimore and Price ), which could be an explanation for observed slowdowns in lineage accumulation through time (see also Pigot et al., Fig. 1c).

17Finally, the conceptual link between the ephemeral speciation model and the ephemeral divergence model (Futuyma; Futuyma ) reflects a common pattern for rates of trait evolution and rates of speciation. New incipient forms are constantly arising within species. Similarly, traits are constantly changing in response to local selection pressures and/or drift. In both cases these changes rarely persist over long time scales. It is important to note that we are not arguing for a model of punctuated equilibrium; trait change may or may not be associated with the formation of incipient species (Bokma ). The important point is that most of the change that occurs over short time periods does not last. Therefore a fundamental shift suggested by both ephemeral models is that evolutionary studies of diversity patterns should focus on not only the formation but also the persistence of new traits and species [e.g., why do some some species persist and others perish quickly? (Levin,; Weir and Schluter; Martin and Tewksbury; Stanley )].

18Although some discussion of the fragile nature of species has occurred in the evolutionary biology literature over the last 50 years, we suggest that the idea of ephemeral speciation has not been deeply incorporated in the way scientists think about speciation. Following Hutchinson, evolutionary biologists have often referred to Santa Rosalia when asking why are there so many or so few species on Earth (e.g., Felsenstein ). We suggest that the ephemeral speciation model provides a resolution to the Goldilocks paradox of species diversity: the balance between rapid species formation and rare persistence can explain why the number of species on Earth is “just right”.