When I mentioned to Jonathan Losos that I had applied a new biogeographic model to anoles, he gave me a copy of Lizards in an Evolutionary Tree. What could I do but take it as both a generous gift and gentle suggestion? This is all to say that I’m no herpetologist. I know this, and so should you. I write humbly to the AA readers as a twig anole standing on the shoulders of crown-giants.
With that said, several close colleagues (Ignacio Quintero, Martha Muñoz, Felipe Zapata, Michael Donoghue) and I recently had a paper published that introduces a new modeling framework that allows regional features of geography to inform phylogenetic rates of dispersal, extinction, and speciation – called FIG (Feature-Informed GeoSSE). We then applied FIG to the phylogenetic and biogeographic Anolis dataset published by Steven Poe and co-authors. This piece summarizes that work and subjects its readers to a few lame jokes.
Cartoon of FIG model behavior.
The first aim of our project was to propose a new framework for modeling historical biogeography using GeoSSE, a diversification model introduced by Goldberg et al. (2011). Under GeoSSE models, all species occupy different sets of regions in a shared geographical arena. As time advances, species stochastically disperse, go extinct, and speciate. Speciation itself might occur within a region or between regions (e.g., in a manner consistent with allopatric speciation).
Thinking realistically, when and where these events happen should somehow depend on the set of regions each species occupies and the geographical features of the relevant regions. Unless instructed otherwise, however, GeoSSE estimates every possible event rate from phylogenetic and biogeographic data alone. Regional features of geography aren’t used. This leaves historical biogeographers with at least one technical problem and one missed opportunity.
First, concerning the technical problem, the number of event rates explodes with the number of regions. A geographical system with two regions (A and B) requires seven rate parameters: two within-region speciation rates for regions A and B, two extinction rates for A and B, two dispersal rates for A into B and B into A, and one between-region speciation rate for splitting the widespread ancestral species with range A+B into two daughter lineages with ranges A against B. Three regions requires 18 rates, four regions requires 45 rates, and so on. Between-region (“allopatric”) speciation rates and, to a lesser degree, the dispersal rates fuel this explosion in rate parameters. Somehow the number of rates must be reduced if we hope to estimate any of them at all.
In FIG, regional features (size, distance, barriers) can inform event rates. Different features are free to have different relationships with each process. For example, region size may increase within-region speciation rates while decreasing local extinction rates. Distance may decrease dispersal rates but increase between-region speciation rates. FIG also allows for barriers and distances to inform the rate and way in which a widespread range splits following between-region speciation.
Second, concerning the missed opportunity, biogeographers often hypothesize how geographical features should influence evolutionary outcomes – for example, that a single species should “split” in two more rapidly when its range is subdivided by a barrier – but GeoSSE has not really been used to test such hypotheses, in large part because of the technical issue mentioned above.
For hypothesis testing, FIG analyses estimate the probability that each geographical feature has an effect on a corresponding class of evolutionary rates. This is done with Bayesian reversible-jump Markov chain Monte Carlo to turn “on” and “off” different relationships between features and rates. For example, if FIG estimates that the effect of distance on dispersal was “on” in 97% of MCMC samples, you might make the bold claim that distance influences dispersal rates between regions. Or, if the effect of size on extinction was “on” in only 31% of samples, you might say no conclusive relationship or non-relationship was detected.
The other aim of our work was to see if FIG produced any biologically interesting inferences. If not, why bother? Anolis led the pack when selecting a clade to feature for empirical analysis. Anoles are well-known for their distribution throughout the Caribbean and the neotropical mainland, with some expectation of moderate, but not rampant, dispersal. Because anole phylogeny and biogeography has been studied in such detail over the decades, the clade is also ideal for test driving phylogenetic models.
Geographical distribution of Anolis.
We adapted the Poe et al. (2019) dataset into a nine-region biogeographical system with five continental and four insular regions of varying sizes and distances. There are ~380 anoles in this dataset, with nearly as many insular as continental species, but nearly [*] all widespread anoles are restricted to adjacent continental regions. This is exactly the type of pattern you’d expect to see if geographical distance and oceanic barriers restricted anole movement. That is, widespread fragmented ranges should be difficult to maintain and should therefore be rare.
Biogeographic rates. Left: extinction (nodes) and dispersal (edges) rates. Right: within-region speciation (nodes) and between-region speciation (edges) rates.
Reassuringly, FIG inferred that dispersal was limited by distance, especially over water. Similarly, between-region speciation rates split ranges the fastest when the inhabited regions were far apart and/or separated by water. Region size and island-status did not have a predictable effect (or non-effect) on extinction and within-region speciation rate. The network diagrams above summarize different regional rates of evolutionary change.
Dispersal rates, between-region speciation rates, and distances.
We also were interested in a critical distance that we defined – the maximum range cohesion distance – beyond which widespread species tended to split into two rather than remain as one. The idea is that if dispersal rates are low, interpopulation migration rates should be low, and therefore rates of allopatric speciation should be high. If we assume that FIG’s dispersal rates approximate the rates of pulsed migration needed to maintain gene flow between regions, we predicted the corresponding distance at which organismal movement became too slow to maintain range cohesion. So how far is too far?
Our scrappy little estimator predicted that range cohesion is almost certainly degraded beyond ~470km over land and ~160km over water. As it turns out, six pairs of continental regions had average distances closer than 470km, where we find all widespread continental anoles today. We don’t expect that these distances perfectly describe the limits of range cohesion for all anole species in all contexts, but a cursory review of the anole population genetic literature convinced us that our estimates were at least reasonable.
Extinction rates, within-region speciation rates, and sizes.
We didn’t recover an analogous relationship for the ratio of extinction and within-region speciation rates with region size. But neither does FIG reject those relationships. Still, recalling the histogram of regional species richness above, it seems unlikely that size alone dictates richness in anoles across continental and insular regions – as they’re defined here – since Amazonia has so few species relative to much smaller regions, like the island of Hispaniola. More work on this issue is needed.
In its conclusion, our paper reflects on how phylogenetic models of biogeography treat allopatric speciation. With anoles under FIG, for example, the consequence of dispersal into a new region depends on the geographical context. If dispersal is between adjacent continental regions, dispersal tends to result in range expansion. But, if dispersal is between distant regions or involves insular regions, it tends to result in cladogenesis that’s consistent with allopatric speciation. In large part, this inference was made possible because regional features inform evolutionary rates in FIG, letting us predict where anoles tend to “split or spread.” We think this historical view of allopatric speciation in a phylogenetic context will be worth exploring further.
Not to pander, but Anolis has played and continues to play such an instrumental role in the development of biologically meaningful models of ecology and evolution. Many of the statistical phylogenetic models that I’ve looked to for inspiration in my own research over the years were introduced by anolologists. Maybe this is a second meaning for model clade? Rambling aside, our project was a true collaboration among some of the most creative organismal and mathematical biologists I’ve known. Our hope is that our small contribution lives up to the high standard of modeling anole diversity.
*–An innocent question from an outsider: is A. sagrei really a single species? Pretty wild.
Paper:
MJ Landis, I Quintero, MM Muñoz, F Zapata, MJ Donoghue. 2022. Phylogenetic inference of where species spread or split across barriers. Proceedings of the National Academy of Sciences 119: e2116948119. doi:10.1073/pnas.2116948119
Code:
https://github.com/mlandis/fig_model
Lab site:
http://landislab.org
- Inferring Where Anole Ranges Tend to Spread or Split - April 29, 2022
Leave a Reply