When did modern placental mammals diversify?

February 3, 2014 • 1:36 pm

by Greg Mayer

Almost exactly a year ago, I reported in two posts here at WEIT on a paper in Science by Maureen O’Leary and colleagues on the radiation of placental mammals. Placentals are one of three major groups of living mammals, the others being the marsupials (dominant in Australia, plus a fair number in South and Central America, and a few in North America) and the monotremes (the egg-laying playtpus and spiny anteaters: a handful of Australasian species). Placentals are by far the most species rich and abundant of the mammals, including the cats, dogs, cattle, deer, and us that are the dominant land animals of our world today.

What O’Leary et al. argued for was a view of placental evolution called the “Explosive Model” (see figure below). In fact, they argued for an ‘extra-explosivey’ model, since they thought the common ancestor of the modern placental orders of mammals arose after the extinction of the dinosaurs (i.e. in the Paleogene, not the Cretaceous); in the figure below by Ken Rose, the ‘explosive’ evolution begins at the very end of the Cretaceous).

Models of placental mammal radiation (Rose, 2006).
Models of placental mammal radiation (Rose, 2006); the thicker lines represent extant orders of placental mammals..

The O’Leary et al. study was widely misinterpreted by the press, which said they had discovered the common ancestor of mammals, a beast called Protoungulatum. This interpretation is completely wrong, and not what O’Leary et al. claimed. My earlier posts emphasized correcting this misinterpretation.

I also noted that O’Leary et al. used the fossil record in a quite literal way to infer dates of lineage splitting. But fossils only provide a minimum date of separation of lineages, and there may be a considerable unrecorded history predating the earliest known fossil. A new paper in Biology Letters (open access), a Royal Society publication, takes O’Leary and colleagues to task on precisely this issue.

Mario dos Reis and colleagues use various approaches to calibrate the molecular clock of placental divergence and accounting for the imperfections of the fossil record. Under all three methods they use (a, b, and c in the figure below), the divergence of the modern orders begins in the Cretaceous, their estimates ranging from 72 to 107 mya (the former, though, not very far from O’Leary et al’s 65 mya). O’Leary’s view is shown in panel d of the figure.

dos Reis et al. 2014, Figure 1.
dos Reis et al. 2014, Figure 1. a, b, and c are the estimates they contemplate; d is the view of O’Leary et al.

So, who’s right here? The first thing I would note is that although the Cretaceous-Paleogene boundary has great psychological weight (and is when a lot of things went extinct, including the dinosaurs), Rose’s depiction of the “Explosive” model had divergence beginning in the Late Cretaceous, and 72 to 107 mya is still Late Cretaceous (or very close to it). So while dos Reis et al. strongly object to O’Leary’s methodology, the results of the two papers are not very different: O’Leary has the most recent explosion, dos Reis has a somewhat earlier explosion, and both bracket the time of divergence depicted by Rose. dos Reis’s timing does have biogeographic implications differing from O’Leary’s, since an extra 25 million years allows for greater influence of plate tectonic events on mammalian distribution.

The second thing to note is that a defender of O’Leary’s might make the empirical retort that we have a fair number of Late Cretaceous mammal fossils, and none of them are clearly progenitors of the modern placental orders. Absence of evidence is not evidence of absence, unless, of course, you’ve looked where the evidence should be, and that’s what an O’Leary defender could argue.

And a third issue is that dos Reis et al. rely heavily on Bayesian statistics to make their inferences. It would be a very long and dry argument to explore this here, but suffice it to say that I find Bayesian statistics, in most cases, to be logically unjustified, and thus I’m not entirely sanguine about dos Reis’s inferences. It’s a fairly arcane issue in the logic of scientific inference, so I’ll just point in the references below to two sources (Royall and Sober) that I have found helpful.

In one of my original posts, I asked

[I]sn’t using the literal fossil record a pretty crude way of determining ages of taxon splits, since such ages are always minimum ages? And shouldn’t the richer information available in molecular sequence data that is time-calibrated by securely known fossil dates be used? Well, the critics will answer “yes” to both questions, and will also point out that the fossil record is imperfect, so to say we don’t have any fossils dated to the Cretaceous is different from saying no such animals existed then. O’Leary et al. might reply that all molecular dating requires geological calibration, so that the fossil data is primary, not the molecular extrapolation; and that we have lots of Cretaceous mammal fossils, and none of them are obviously the varied precursors of the Cenozoic placental radiation.

And concluded by asking

Who’s right? I don’t know. But that’s what the upcoming arguments will be about.

Indeed, dos Reis et al. have made a sharp statement in this ongoing argument.


dos Reis, M.,  P.C.J. Donoghue, and Z. Yang. 2014. Neither phylogenomic nor palaeontological data support a Palaeogene origin of placental mammals. Biology Letters 10. pdf

O’Leary, M.A., et al. 2013. The placental mammal ancestor and the post-K-Pg radiation of  placentals. Science 339:662-667. (abstract)

Rose, K.D. 2006. The Beginning of the Age of Mammals. Johns Hopkins University Press, Baltimore. (Google Books)

Royall, R. 1997. Statistical Evidence: A Likelihood Paradigm. Chapman & Hall, London. (Google Books)

Sober, E. 2002. Bayesianism — its scope and limits. in R. Swinburne, ed., Bayes’ Theorem, Proceedings of the British Academy 113:21-38. pdf

12 thoughts on “When did modern placental mammals diversify?

  1. I suppose it is important to read the whole article. I read the title and the first five words and assumed it was creationist rant!!

    1. I think it is a stretch to try to settle choices of scientific methods on purely philosophical grounds (logical justification). Parsimony, maximum likelihood, Bayesian methods… we can talk till the sun burns out about which is the better method but we are unlikely, especially after so many decades of discussion already, to find the deadly logical flaw that rules out a method. Rather, methods will survive or fade away based on their actual success in practice.

  2. Oh no, the Bayesian gauntlet has been thrown down. All statistical methods have implicit priors; none are the magic objectivity generators many people would like them to be. It is just that some methods are a bit more honest about this than others which, to many people, is like an admission of guilt. And if you have, you know, actual evidence, priors quickly cease to matter anyway, so worrying about priors too much is essentially to say you don’t have much evidence so… yeah.

    That said, any method can also be abused. If you think your priors are as good as evidence you are deluded. This is a fault with you, not the method. I have no idea if the authors here made their inferences appropriately but if they did not it probably is not the fault of their use of Bayesian statistics, per se (nor their use of Macs vs Windows), but due to their own over-interpretation of their data.

  3. As I’m a statistician, I haven’t read the biology paper yet, but I’ve just read Sober’s article instead.

    Rather nicely, he says pretty much what I’ve been saying for the last 25 years.

Leave a Reply