Treating fossils as terminal taxa in divergence time estimation reveals ancient vicariance patterns in the palpimanoid spiders

Wood et al. Systematic Biology 62(2): 264-284. DOI:10.1093/sysbio/sys092. Treating fossils as terminal taxa in divergence time estimation reveals ancient vicariance patterns in the palpimanoid spiders.

This is a guest post with April Wright. Below, we give our first impressions of this article. Please comment below, or tweet AprilWill or Lynsey (maybe use #pegejc). Think of this as a journal club discussion group!


April Wright

When I started graduate school, I envisioned doing some fusion of paleontology and molecular phylogenetics. What I didn’t envision is other researchers constantly asking me “Why?”. Why use morphology when you can have a bajillion base pairs of sequence data? Why use morphology when we have such nice, explicit models for sequence evolution?

But in the past year and a half, there’s been a series of really lovely papers forging a kind of truce between the morphological and molecular worlds, and I think this paper highlights why this is important: Fossil taxa are the only record we have of extinct organisms, and we can learn a lot about the world of yore from them. Seems like an obvious point, but from the sheer volume of people asking me “Why?!”, it apparently is not.

For a little bit of background, in 2011, Alexander Pyron authored a paper treating fossil taxa as tips, rather than calibration points on nodes, in a chronogram. And people, in both the molecular and morphological spheres were pretty excited about this. It’s an intuitively appealing idea. Often, fossils are placed on a tree as calibration points, but we don’t really know the fossil belongs where we’ve placed it. Treating the fossil as a tip in the tree allows the fossil to be placed with confidence. It’s a nice concept.

In discussions with coworkers, people at meetings and randos off the street, it became clear to me that not everyone was sold on the utility of idea. While many people liked the idea of treating fossils as tips conceptually, there are still questions about if this practice will actually result in any noticeable effect on tree estimation or the inferences drawn from those trees. The paper for this week is quite nice in that it makes use of real data from fossil and extant spiders that the authors want to use to make an inference about historical biogeography to test the effect of treating fossils as tips.

A challenge in integrating morphological and molecular data is the degree of asymmetry between data types. In combined analyses, often there are many species for whom molecular data is available, some species for whom morphological data is available, and only a handful of species with both. The net result of this is basically a molecular tree and a morphological tree held together by a couple of taxa. This isn’t the case with this paper, and I was impressed by the care taken with the sampling of morphological characters in extant and fossil spiders, though the fossils are not well-intercalated with the extant taxa (more on this in a moment).

One of the interesting results in this paper is that treating fossils as tips on the tree resulted in node ages that were older than when fossils were used as calibrations. This hasn’t been found in other studies (but, this is one of the first studies of its kind, so this pattern may be quite common, and we don’t know it yet). As I mentioned before, the fossils are not intercalated in with the extant taxa, instead branching from a single point. This odd result highlights that we need to do more research to understand the effects of using fossils as tips.

This study takes it one step further and uses the dated phylogeny they obtained to make a biogeographical inference about spiders. Using the software LaGrange, the authors looked at the historical ranges of the spider clades for which they had data. The authors support the conclusion that the breakup Pangaea into Laurasia and Gondwana lead to a vicariance event within spiders. This is a very cool result, though likely not very different from the one they would have received treating fossils as calibrations only.

I’m going on a bit long, so I’ll wrap up by saying that I really enjoyed this paper. I think this is a great example of going about a new type of analysis in a very thoughtful way. I think the primary result of a vicariance event in spiders at the Pangaea split is a pretty neat, punchy result, but there’s plenty in this paper for any methods dork to have fun with.

Will Pearse

Will Pearse

In theory, I’m a phylogeneticist, so I should probably have an opinion about how we best use fossil calibration points. As such, I’m not going to talk about spiders, or biogeography, and am essentially just going to talk about what I remember of Joe Felsenstein‘s talk at Evolution 2012 on this issue. I got very excited coming out of that talk, so I hope I’ve remembered the details correctly!

There is a very big problem in phylogenetics that I don’t think enough people talk about: how do we date a phylogeny? We can now build massive phylogenies with RAxML, but the output doesn’t tell us when things evolved, just what’s closely related to what. Programs like BEAST let us simultaneously estimate phylogenetic structure and timing of evolutionary divergence, but we need to calibrate our results with fossil data. Otherwise we’re just inferring dates based on molecular data, and ignoring when we know certain groups must have evolved given what we find in the fossil record.

Wood et al. argue that dating clades by using fossils to set prior distributions on how old they’re likely to be may not be the best approach. I agree with them. Instead of using fossils to date clades, they’re putting the fossils in as extinct taxa, and they building phylogenies around them. This is kind of neat; it means the species fossils represent become part of the tree, and the extant species get dated in the process of making a phylogeny with those dated fossils in it.  They argue that, when they use this method, their results are less driven by their prior distributions, and as a rather naïve Bayesian statistician (that’s a pun, stats fans), I agree with that. I want the signal in my data to drive my answer, not the constraints and assumptions I made at the beginning.

Felsenstein outlined what I view as almost an extension of this method. In it, you use morphometric measurements of the actual fossils, along with measurements of extant species, to figure out where fossils go within a clade. Essentially, this means you can figure what a fossil’s closest relatives within a clade were, what branch of a phylogeny they’re more like, and get the dating of the phylogeny for free because we know roughly when the fossil was put down. I view this as an improvement on the present method, because the fossil taxa are not left orphaned in their own sister group to the extant species (see figure 1 in the paper) – they’re nestled in there with them, which of course reflects how the clade actually evolved. The disadvantage is that (as far as I’m aware) it’s not implemented yet, and it is probably vastly more data-hungry.

Lynsey McInnes

Lynsey McInnes

First, thanks April for providing our first guest post and for picking a whopper of a paper!

Man, this paper was dense and I commend the authors whole-heartedly for steering a relatively clear path through the huge number of analyses performed and for extracting the relevant conclusions thoughtfully. These guys certainly know their methods, and their spiders!

I was convinced by their argument to include fossils as terminal taxa and liked their inclusion of uncertainty around the fossil ages. It would have been a shame to take one step forward (including fossils as terminal taxa) and one step back (pretending their age is known with certainty).  Their conclusion – use all the data – was hardly surprising but very nicely demonstrated.

I do wonder however whether the fossils would be as crucial if extinct and extant taxa had overlapping/the same ranges? While this paper is a cool example of distributions shaped by the break up of Pangaea, making the fossil information central to any valid conclusions, how important would fossil info be if it didn’t provide additional/new information about ancestral distributions? Presumably, in these cases, the terminal fossils would function more like extant taxa in the same area, lending more weight to any conclusions on distributions possible using extant taxa alone. I guess that would be a bit boring, and so Wood et al’s situation makes for a much more gripping tale.

Although any of the following would have made the paper excessively long and I imagine may have been covered in the Pyron or Ronquist papers the authors refer to, I wonder, totally naively, if one could partition out how strongly the fossils drive the results found, how much fossil data is needed to change the story, what happens when one fossil is misplaced (in time or space), what happens if preservation biases mean that more fossils are found in one region than another so the signal is somehow skewed…hm…I am sure there already exists a whole literature on dealing with these issues when using fossils as calibration points and this could be preyed upon to find out what would happen in this ramped up fossil use case.

That was a harsh paragraph to end on when discussing an admirably thorough, thoughtful AND neat paper and came out mostly because I’m at a bit of a loss as to how to critique such a piece of work. So I’ll just stop rambling right here!


About will.pearse
Ecology / evolutionary biologist

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: