Thursday, April 16, 2009

Coffee Shop Phlogenetics #2: What's More Prehistoric the Robin or the Blue Jay?

I just showed up for a latte at my favorite coffee shop and my buddy Ian was there waiting with a question: "What's more prehistoric, the Robin or the Blue Jay"? He was thinking it was the Blue Jay due to overall physical appearance and their prehistoric squawking calls. It's impossible to answer this question in a manner that's going to satisfy the serious phylogeneticist, but because I like to think of myself as a phylogeneticist of the people I'm going take stab at this one. To avoid troublesome inference about which species is more primitive or more advanced we should focus simply on which extant species has been around for longer. If we look at Hackett et al.'s recent phylogenomic analysis of birds, we find that the Robin's genus (Turdus) is included, but the closest thing to a blue jay is the con-familial crow (Corvus). Let's approach this question from the family level by contrasting the phylogenetic position of the crows and jays (Corvidae) with that of the thrushes (Turdidae). It's clear that the Corvidae branched off from a clade including the Turdidae and a range of other families relatively deep in the Oscine radiation. This pattern certainly fails to reject Ian's hypothesis, but its unclear that anything shy of a comprehensive species-level, time-calibrated phylogeny would be able to do more. Any ornithologists or paleontologists care to weigh in on this important topic?


Brian O'Meara said...

"It's impossible to answer this question in a manner that's going to satisfy the serious phylogeneticist."

Yup, pretty much, especially if you're just using topology.

Though if you interpret "prehistoric" as "has undergone less evolution", you could do root-tip length for molecular substitutions (correcting for node-density effect) or maybe morphological branch lengths -- the shorter one is more "prehistoric."

Glor said...

I was taking a simpler perspective by trying to ask whether the species we know today as Turdus migratorius has been around for longer than the species we know today as Cyanocitta cristata. This is definitely a question better suited to fossils than a phylogeny, but assuming no extinction and comprehensive family-level sampling the topology is telling us that the family Corvidae has been around for longer than the family Turdidae.

Liam Revell said...

I agree with Brian. If I was asked this question, I would surely be most inclined to provide the answer that since both species have exactly equal times of separate history since they shared a common ancestor (both of them being extant), then neither species is more or less prehistoric.

If that answer proved unsatisfactory, then I might say that the species with more apomorphies is more derived, and thus the species with fewer is more ancestral (i.e., more prehistoric). This may or may not correspond to the most basally located species in the topology. (If it tends to on real phylogenies, this could very well be an artifactual consequence of the natural tendency of taxonomists to measure only characters that vary among species, thus minimizing the number of inferred autapomorphies along basally divergent branches.)

The "apomorphy" approach thus presumes that we have measured all relevant morphological (or genetical) characters, and if not we should probably restrict our conclusions to "species X exhibits a more ancestral condition for trait Y," etc.

The whole question seems somewhat analagous (although for most lay-people, less inflammatory) to the old "do people really come from monkeys" line. . . .

Glor said...

Good point. I realize that Ian's original question was inspired by his intuition of primitive versus derived, or specialized versus unspecialized character states, but I was trying to avoid this in favor of asking more simply which extant species has existed for longer (i.e., did the speciation event resulting the extant Blue Jay occur prior to the speciation event resulting in the extant Robin). Perhaps I cast my original non-answer from the tree a bit too optimistically given that sampling is far too incomplete and that time calibration is lacking.

Brian O'Meara said...

I'm not sure that even relative age of the most recent cladogenic events gets at the question, though. First, there's the "chronospecies" issue: if the robin-sister split was 10 MYA, and the jay-sister split 5 MYA, but the robin's morphology and other traits radically changed 2 MYA but not the jay's, the age of the robin species as paleontologist with a great time series would see it is 2 MY, not 10 MY. Second, the form of the last speciation event matters. If it's a dumbbell allopatric event where neither descendant species is morphologically identical to the ancestor, maybe that's of interest to the lay audience, but what if the main robin lineage has remained unchanged for millions of years but occasional birds get blown to islands and form new species [the giant flightless Aleutian robin, perhaps]? Presumably, Ian wouldn't care that the last time this happened was 200,000 years ago for robins but 1 MYA for jays -- it's not the kind of speciation event that "counts" in terms of age of the focal species to most people. So I think the question, to make any sense at all, has to relate to character change. Liam makes a good point about which morphological traits are examined biasing the results. Perhaps there's some pheneticist out there with a data set?

John Harshman said...

I think you're conflating at least two separate questions:

1. Which species is more primitive?

2. Which species has been around longer?

Neither is, I'm afraid to say, a sensible question in my opinion.

The first requires that we adopt some kind of objective index of primitiveness, with particular character contributions weighted in some fashion. Any given species has some primitive features and some derived features. This question should go the way of arguments among ornithologists about whether crows or finches are more advanced, and should thus be listed last in a linear classfication. (I'm not kidding about that -- there were many trees killed over that question in the last century.)

The second question demands that we articulate a species concept that can be extended over evolutionary time, when the prevalent species concepts apply only to restricted time slices. We can learn nothing about the answer from any phylogenetic tree, even one that contains all modern species. All we can learn from that is when a given lineage diverged from its closest living relative. We can't learn when that lineage became the species it is now, which, depending on species concept, might be either long before or long after that divergence. Even fossils don't help much unless you decide to use a morphological species concept, i.e. one in which a species is a bundle of sufficiently similar morphotypes. In birds, this is especially suspect, as many modern avian species pairs are difficult, perhaps impossible, to distinguish on the basis of skeletal comparisons. And of course the bird fossil record isn't complete enough to track morphotypes over time anyway.

Glor said...

All good points gentlemen. I should have followed my own advice and avoided even trying to come up with a serious to answer this question in the first place!

John Harshman said...

You're welcome. I only dropped by because I was reading the Joe Felsenstein interview, and I happened to see that you had shown a part of my tree. And such a nice tree is is.

Remember: there are no stupid questions. Well, creationists manage it. So there are no stupid questions, but there are full tilt bozo questioners.

Glor said...

This was by far the dumbest of four posts we've had about your paper: 1, 2, 3. Comments on the others would be greatly appreciated if you have a moment. Your results set off an interesting discussion about the application of Bayesian methods to large phylogenetic datasets

John Harshman said...

Thanks. As those threads seem long dead, perhaps I can make a comment or two here.

I do think MrBayes has problems with large data sets, though I obviously haven't investigated that rigorously. And it may be just with large numbers of taxa, not bases, as I've experienced similar difficulties (high posteriors for contradictory nodes in runs upwards of 20 million generations) with a couple hundred taxa but only a few thousand bases. My theory, which one of the posters mentioned, is that with lots of taxa, not enough topological changes are being proposed or accepted to result in a reasonable sampling of tree space.

I admit I have not really investigated the effects of alterations in the various MCMC parameters, and would appreciate it anyone has citations for such studies. I haven't seen much. As a matter of fact, if anyone would like to perform such a study using the Early Bird data set, I'm sure everyone involved would be thrilled. But I bet it would take a very long time, unless the parameters we picked just happened to result in unusually long search times and unusually poor mixing.

Then again, I like Garli and RAxML pretty well. They both seemed to converge in reasonable time on the big data set. If you will note, we did Garli bootstraps, so any nodes with high bootstrap must have converged consistently. And multiple runs did give us similar likelihoods. To me, Garli's major disadvantage was its lack of partitioned analysis. But I'm not sure that's a serious problem with a big sequence sample of this type. We've done comparisons of multi-partition and single-partition analyses with smaller taxon samples of the same data, and there was no difference in topology. (See Harshman et al. 2008. PNAS 105:13462.)

Interesting discussions, all.