The following mini-essay was originally intended for the BioGeoBEARS Mistakes To Avoid page, but is a complex enough topic that it deserves its own page.
Background
Citing this page: Matzke, Nicholas J. (2016). 'BioGeoBEARS deals with "Founder-event speciation" not "founder-effect speciation."' PhyloWiki essay, 2016-08-07. http://phylo.wikidot.com/biogeobears-deals-with-founder-event-speciation-not-founder
I have recently seen confusion between "founder-effect" and "founder-event" concepts in biogeography arise a few times from different sources. Sometimes, disturbingly, this occurs in peer-reviews or in editorial comments. I have now heard of this independently a few times from different authors, who encounter the criticism from a reviewer or editor.
However, I feel that the founder-effect versus founder-event issue is directly, carefully, and adequately discussed in Matzke (2014, Systematic Biology) (1), but apparently the paper has not always been read carefully. So, I think I should reiterate my position on this question. I do so below; this is modified from a previous email I sent a correspondent, and I might put some of it into an article at some point in the future.
I encourage readers to use or cite the following discussion, as useful, in their discussions of these matters with colleagues, reviewers, and editors. In the future, I hope to write a more formal discussion along the lines of "how statistical model comparison works, and what it does and does not allow you to say" for a journal, but this may be a ways off.
This essay is still a bit rough and "off-the-cuff", so I invite further discussion, corrections, etc.
The confusion between "founder effect" and "founder event"
Here is the confusion I have seen. Consider the (common, but not universal) situation where a BioGeoBEARS analysis shows strong support for a "+J" model (such as DEC+J) over a nested sub-model (such as DEC, where j=0). This is routinely taken to be evidence that a model that includes founder-event speciation as a process explains the data better than the nested model that did not.
The criticism that sometimes arises is: founder-effect speciation is an old idea that has been much-studied over the decades. It goes back at least as far as Ernst Mayr's theory of peripatric speciation, and the idea that isolated peripheral populations will experience a population genetic founder effect, resulting in strong genetic drift, fixation of new configurations of co-adapted alleles, morphological change, speciation, etc. Mayr went so far as to propose that these could constitute "genetic revolutions", where the isolation events led to rapid genetic and morphological change. The problem is that Mayr's idea, while appealing, has not been well-supported by subsequent empirical studies. While there is a great deal of evidence for the population genetic founder-effect (where a bottleneck in population size causes reduced genetic diversity, fixation of deleterious alleles, etc.), there is not much evidence that the population-genetic founder effect typically has the downstream consequences that Mayr proposed for speciation, morphological change, etc.
The argument, then, is that because founder-effect speciation is not well-supported as an empirical matter, the founder-event model (+J) in BioGeoBEARS should be questioned. Sometimes the statement is made that the likelihood advantage of DEC+J over DEC must be some kind of mysterious artefact, although the criticism always gets quite fuzzy at this point (2).
While I agree that the BioGeoBEARS founder-event model should be questioned — all models should be questioned, in biogeography and everywhere else in phylogenetic comparative methods! — this particular criticism is without merit.
The key problem: "Founder-effect speciation" is not identical with phylogenetic/biogeographical "founder-event speciation"
The key problems are this: the "founder-effect" is not identical with "founder-events", and "founder-effect speciation" is not identical with "founder event speciation". To review the issue one more time:
- The "founder effect" operates at the level of population genetics, and refers to small, isolated populations experiencing strong genetic drift forces, leading to rapid fixation of neutral or deleterious alleles. This effect is well-known and well-documented in population genetics. However, there is a more ambitious idea, which is that the founder-effect itself promotes speciation. For example, in Ernst Mayr's classic formulation, "genetic revolutions" and rapid morphological change and speciation.
- A "founder event" or "jump dispersal event" (these are equivalent in my usage) refers merely to a rare long-distance dispersal event (3) that founds a new population that is sufficiently genetically isolated that it rapidly (4) becomes a new phylogenetic lineage. In other words, speciation and dispersal events are either literally coincident, or they coincide closely enough that they are reasonably modeled as a joint process in a phylogenetic model. The founder event *might* be associated with a classic population genetic "founder effect", but it *might not* (it would all depend on the size of colonizing population, whether it was literally one event or a few events within a short period of time, how fast the newly founded population reaches large size, etc.). Intermediate situations are also perfectly possible: it might well be the case that population genetic bottlenecks and increased genetic drift are common during founder events, but they do not have the downstream effects Mayr postulated, such as rapid morphological change.
I think "founder events" or "jump dispersal" are, a priori, likely to be important in historical biogeography, and many agree — particularly biogeographers working on island systems. But I have no strong position on the more specific questions surrounding the population-genetic founder effect and its downstream consequences. From the point of view of a molecular phylogeny and a probabilistic model of historical biogeography, these details don't matter: A genetically isolated population will coalesce independently from the parent population. Whether or not it has undergone noticable morpohological change, and whether or not the new lineage is formally classified as a "different species" by taxonomists is not a key point - all that BioGeoBEARS requires is that the tips on the phylogeny represent different lineages, i.e. populations that are coalescing independently. Generally, on a phylogeny of an appropriate age, specimens-from-the-same-coalescing-population will be quite obviously different from specimens-from-distinct-
genetically-isolated-populations, due to the dramatically shorter branch-lengths within the coalescing population.
The interesting question that the "+J" model allows us to ask has nothing to do with the population-genetic "founder-effect", and everything to do with "founder events". Namely, the +J model allows us to ask, does a biogeographical model that includes the founder-event process provide a statistically better fit to the data than a model (such as traditional DEC/Lagrange) that does not? Very often, but not always, this has been found on empirical datasets, in formal statistical comparisons.
I do think that having a biogeographical model that allows one to identify founder-events could be useful for future studies about whether or not biogeographical founder-events are usually associated with genetic, morphological, or trait-evolution consequences or by-products of population genetic founder effects, but a study of this sort is well beyond something that could be done using only BioGeoBEARS.
I said all of this before, in Matzke (2014)
Here I will quote Matzke (2014), mostly to point out that I tried to head off this confusion ahead of time:
Matzke (2014), p. 952
One well-known form of speciation that is left out of standard DEC analyses is "founder-event speciation" (Paulay and Meyer 2002; Templeton 2008), sometimes termed speciation through long-distance dispersal (Heads 2012) or allopatric mode II speciation (Wiley 1981; Maguire and Stigall 2008; Lomolino et al. 2010). In founder-event speciation, a small number of individuals, sometimes even a single individual, take part in a rare, long-distance colonization event which founds a population which is instantly genetically isolated from the ancestral population. Founder-event speciation has received extensive theoretical attention (Mayr 1954; Gould and Eldredge 1972; Carson and Templeton 1984), and it remains a controversial question whether or not the population genetic founder effect and resulting genetic bottlenecks significantly contribute to shifting of adaptive peaks (Carson and Templeton 1984; Coyne and Orr 2004) or morphological evolution. However, to many biogeographers (but not all; see Heads 2012) it seems undeniable that founder-event speciation is an important mode of lineage splitting, and of moving taxa around the planet. Founder-event speciation is particularly likely to be important in oceanic island systems (Carlquist 1974; Paulay and Meyer 2002; de Queiroz 2005; Cowie and Holland 2006; Gillespie et al. 2012). If founder-event speciation is important for explaining certain biogeographical patterns, this would remain true whether or not founder-event speciation plays any major role in production of nonbiogeographical macroevolutionary patterns.
Matzke (2014), p. 967:
SSE models. — Finally, DEC+J and DEC merit comparison to "SSE" models such as GeoSSE and ClaSSE. GeoSSE (Goldberg et al. 2011) assumes the same types of cladogenesis events as DEC, but each is treated as a macroevolutionary process that is controlled by a different rate parameter. This results in a large number of free parameters even for a two-area system, and would require many more parameters for systems with more areas, suggesting the need for large data sets for effective inference. The closest approach that GeoSSE could make to founder-event speciation would be to have a very high rate of vicariant speciation when a lineage occupies multiple ranges. This would tend to produce speciation events immediately following range expansion. Arguably, this follows the intuition that immediately after long-distance dispersal, the same species exists in two regions, and it would take some amount of time and evolutionary change before a taxonomist would split the two populations into two species.
However, such a formulation has the disadvantage of conflating two very different processes, namely, classic vicariance, wherein a geological or environmental event breaks up the population independently of the time since a lineage became widespread, and founder-event speciation, where cladogenesis is intimately connected to the original dispersal event. It is possible that both processes have occurred in the history of a particular clade, especially if different lineages have different dispersal ability, and if so, modeling the two processes with a single rate parameter would produce only an "average rate of vicariant speciation" as a compromised estimate giving the mean of what is actually a bimodal process. This consideration argues for modeling classic vicariance and founder-event speciation as distinct processes, at least until more sophisticated models are available that explicitly link traits and distance to maintenance of gene flow and probability of speciation.
In addition, while the intuition that the same species exists in two places after a rare long-distance dispersal event is appealing from the point of view of a morphological species concept, this concept may be a misleading representation of genetic and phylogenetic reality in cases where a rare colonization event establishes a genetically isolated population. Such a population would be completely isolated from its ancestral gene pool and thus be a new lineage on any mechanistic definition, whether or not it would initially be "different enough" to be named a new species. It would take some time after a founder event for genetic isolation to become detectable in genetic data, but the bottleneck effects of population genetic founder events are so strong that genetic isolation is likely to be detectable within a few generations, effectively instantaneous on historical biogeography timescales.
Conclusion
As I said above, all models need to be questioned. Certainly, there is absolutely no reason to think that the simple 2-5 parameter models we are usually using in historical biogeography capture all of the complexity of the reality. I encourage and invite researchers to try to think up new and improved models that fit data better, and to compare them to current models using standard tools of statistical model comparison. This is the whole point of doing science.
However, once a model has become published and widely accepted and used in the literature, and it often fits the data better than other models that are in wide use, I think (a) criticisms of models should carefully read the original literature proposing the model, and (b) criticisms should be published in the literature for everyone to evaluate, rather than deployed behind-the-scenes by editors or reviewers, where misconceptions apparently can develop and persist. This is particularly important if the critic is suggesting that the results of statistical model comparison should be ignored, and a model that confers much lower likelihood on the data should be preferred. If we give up on statistical model comparison, we might as well give up on the use of statistics in science, and tell whatever story we like, regardless of what the data say.
Notes
1. The "+J" model was proposed in Matzke (2014), as well as the more important (and much more generic) idea that we should be using statistical model choice to compare and test models in historical biogeography against the data.
Matzke, Nicholas J. (2014). "Model Selection in Historical Biogeography Reveals that Founder-event Speciation is a Crucial Process in Island Clades." Systematic Biology, 63(6), 951–970. http://dx.doi.org/10.1093/sysbio/syu056
Note that I might re-phrase a few things slightly, and am always open to discussion if what I have in my head is being widely misread in a different way. However, overall it still reads well to me.
2. A related problem in discussions about biogeographical models is that many discussants have only a fuzzy idea of how the likelihood calculations actually work, and they tend to rely on intuitions instead. These intuitions often come from vague verbal models, or often from from "parsimony-thinking", or from "DNA model thinking" (or "purely continuous-time model thinking"). Unfortunately, often, none of these provides a perfect analogy for what is happening in the likelihood calculations of biogeographical models, which have both anagenetic and cladogenetic processes operating, the relative importance of which can shift radically depending on the data and inferred (or pre-specified) parameters.
3. By "long-distance", all we can say is that the +J model assumes is that the dispersal event takes members of the population to a new discrete area, outside of the discrete areas occupied by the ancestor, and that this distance was long enough (or some other difference was big enough) that a new isolated genetic lineage was founded. Whether or not that distance is "long" in human terms is not something the computer thinks about, unless the researcher gives BioGeoBEARS a distance matrix, a dispersal connectivity matrix, etc.
4. "Rapidly" on a phylogenetic time scale. Typical phylogenies span millions of years, so any process that typically happens on a timescale of less than, say, 0.1 million years, can be considered "rapid" or even "effectively instantaneous" from a geological/phylogenetic point of view. Obviously this has almost nothing to do with whether or not the process would be considered "rapid" or "effectively instantaneous" on a human timescale.