Placement Of Mexican Amber Fossil Responsible For Extremely Old Age Estimate For Anolis

Although we’ve been focusing a lot of attention on Nicholson et al.’s new classification for anoles, Daniel Scantlebury recently called attention to the fact that this monograph also contains “a bold hypothesis of the biogeographic history of” anoles.  I’m going to focus here on only one aspect of Nicholson et al.’s biogeographic analyses – namely, their use of two remarkable amber fossils to calibrate a Bayesian relaxed clock analysis supporting the hypothesis that anole diversification dates back to the Cretaceous.

Nicholson et al.’s hypothesis that anoles first appeared more than 90 million years ago and that most major clades of anoles originated prior to 70 mybp is likely to be one of the most controversial aspects their hypothesized biogeographic scenario.  These extremely old ages are significant because they make anole diversification compatible with a scenario that has long attracted the attention of vicariance biogeographers (Rosen 1975Savage 1982Crother and Guyer 1996).  Under this scenario, anoles occupied an ancient volcanic arc that originated in the Pacific ~120 mybp and formed a landbridge between North and South America in the Late Cretaceous (75-70 mybp) before moving on to form the present day West Indian islands.

I have characterized the ages for anole diversification in Nicholson et al.’s biogeographic reconstruction as “controversial” and “extremely old” because they are older than the age estimates obtained by most other studies.  Hedges et al. (1992) were among the first to use molecular methods to estimate ages for terrestrial vertebrate fauna of the West Indies, and reported ages for anoles and other taxa that were far too young to be compatible with Cretaceous vicariant events and the hypothesized Greater Antillean Landbridge between North and South America.  Hedges et al. (1992) suggested instead that anoles arrived in the West Indies via over-water dispersal.  Although Crother and Guyer (1996) criticized Hedges et al.’s use of immunological data and their resulting conclusions about over-water dispersal, more recent work has tended to support Hedges et al.’s conclusions by recovering ages for anoles and other terrestrial West Indian vertebrates that are too young to be compatible with the vicariant scenario hypothesized by Savage (1982), Crother and Guyer (1996) and Nicholson et al. (2012).

Daza et al.’s (2012) cladistic analysis of fossil data, for example, includes an update of the time calibrated tree generated by Conrad (2008) from available fossil material; this tree suggests that the Polychrotidae (the possibly non-monophyletic clade that includes anoles and other putative relatives like Polychrus) split from the Hoplocercidae sometime in the Eocene (~50 mybp).   Townsend et al.’s (2011) analysis of a multi-locus molecular phylogenetic dataset for iguanian lizards that used a BEAST analysis with 18 fossil calibrations suggests a split between Anolis carolinensis and the Corytophanidae at 50-70 mybp.  Most recently, Mulcahy et al.’s (In press) analysis of a multi-locus phylogenetic dataset for squamates in BEAST that relies on 14 fossil calibrations suggests that Anolis carolinensis split from Enyalioides laticeps 25-75 mybp (penalized likelihood analyses conducted by Mucahy et al. suggest a considerably older split between these two species that dates to around 80 mybp).

Recently published trees with estimates for the age of Anolis from Daza et al. 2012, Townsend et al. 2011, and Mulcahy et al. in press.

Why is there a discrepancy between the ages for anoles reported by Nicholson et al. and other studies?  There are many possible answers to this question (e.g., sampling differences, use of alternative methodologies, etc.), but I believe it is relatively easy to isolate the cause of the relatively old age estimates reported by Nicholson et al. to a single fossil calibration.  Before discussing whether this particular calibration was appropriately deployed by Nicholson et al., I’d like to first address the more salient (and somewhat easier to answer) question of whether this lone calibration is really driving the age estimates reported by Nicholson et al.

To investigate the impact of the two fossil calibrations used by Nicholson et al. on the inferred age of anoles, we need to take a quick look at a time calibrated tree.  Because Nicholson et al.’s monograph does not include any trees where branch lengths are scaled to time, I’ve crudely generated a time calibrated tree of my own using the same dataset and program used by Nicholson et al.  In my analyses, I’ve excluded fossil calibrations in favor of a simply assigning an arbitrary age to the root of anoles.  The reason for generating this tree is to illustrate the relative temporal positions of the two nodes that are calibrated with fossils by Nicholson et al.

Time calibrated tree generated using BEAST and a similar mtDNA dataset to the one analyzed by Nicholson et al. (they differ only in that this tree lacks a handful of recently sequenced species). This tree was generated without any fossil calibration data and was merely assigned an arbitrary root age mean of 100. The colored bars indicate the relative temporal positions of the two calibrated nodes in the Nicholson et al. analysis.

I’ve indicated the temporal position of Nicholson et al.’s two calibrations in the above figure, with the red dot and line indicating a node that is assigned a minimum age of 28 mybp and a blue dot and line indicating a node that is assigned a minimum age of 23 mybp.  Notice that the younger (i.e., closer to the tips) red node is assigned an older minimum age than the older (i.e., closer to the root) blue node.  As a result, assigning the minimum permissible age to the older blue node (23 mybp) will inevitably involve assigning the younger red node an age that is younger than its minimum age (28 mybp).  In other words, the blue node can’t actually have a date that is close to its minimum because this would violate assumptions of the red node’s calibration.  On the other hand, trees that assign the younger red node its minimum value (28 mybp) will remain compatible with the constraint on the older blue node (minimum age of 23 mybp).  Because of the fact that Nicholson et al. have placed the older calibration at the younger node, we expect that the minimum age of 28 mybp assigned to the younger red node will drive the overall tree calibration and root age while the 23 mybp calibration assigned to the blue node will have a relatively minor impact on the overall calibration.

Think about the age we’d likely recover for the root if we only used one fossil or the other.  If we used only the blue calibration, the scale of the figure above suggests that we’d likely arrive at a root age somewhere around 50 mybp.  If we used only the red calibration, however, the scale of the figure above suggests that we shouldn’t be surprised to recover root ages in excess of 100 mybp.

With this background we can return to a consideration of the two fossils used to calibrate this tree.  The first is the largely intact skeleton of a juvenile lizard from Dominican amber.  Although interpreting the phylogenetic position of this specimen was complicated by the fact that it is a juvenile from which some characters important to cladistic analyses of anole relationships could not be scored, de Queiroz et al.’s (1998) cladistic evaluation of the available data suggested that this fossil likely belonged to the clade that includes extant species of Hispaniolan trunk-crown anoles.  The Dominican amber is also well-dated and is thought to result from forests that existed 15-20 mybp (Iturralde-Vinent and MacPhee 1996).  Given this information, it seems reasonable to place this fossil calibration as a minimum age at the stem of the chlorocyanus species group, and this is exactly what Nicholson et al. have done (this fossil corresponds with the blue dot and line in the figure above).  Although we could quibble about the use of a 23 mybp estimate that seems slightly older than most of the current research on the Dominican amber would have us believe, I’m not particularly concerned about this point because, as I’ve argued above, I don’t think this fossil calibration has a big impact on the ages recovered by Nicholson et al.

The second fossil used to calibrate Nicholson et al.’s tree is from Mexican amber and includes fragmentary remains of a juvenile lizard.  Nicholson et al. rely on Lazell’s description of this fossil to place it in their phylogenetic tree.  Lazell’s description of the fossil includes informal pre-cladistic comparisons of features observable on the fossil to features of other extant anole species.  Based on these comparisons, Lazell regarded the amber fossil as most phenotypically similar to three different species, including representatives of both of the two deeply divergent anole clades that occur on the mainland (see also Jonathan Losos’s post on this fossil).  Because only one of these two clades occurs in Mexico today, and because Lazell considered only one of the Mexican species known in 1965 – limifrons – as worthy of comparison to A. electrum, Nicholson et al. assign this fossil calibration to the stem of limifrons/zeus.  I don’t want to argue that this is clearly the wrong place for this fossil.  It’s not difficult, however, to argue that its placement here is based on somewhat tenuous evidence and that a more conservative placement at the stem of other fuscoauratus-like anoles, or even at the stem of Norops, would generate a much younger estimate for the root age of anoles.

In addition to potentially placing this fossil closer to the tips of the tree than the available data can justify, Nicholson et al. also assign this node a minimum age of 28 mybp.  Lazell believed the Mexican amber fossil was “Oligocene or Miocene in age,” but most modern authorities now assign the Mexican amber to the Miocene with an age of 15-20 million years and believe that it results from the same type of trees that occurred contemporaneously on Hispaniola.  In response to a previous post on this topic, Nicholson commented “I seriously doubt there would be a significant difference between using some date [for the Mexican amber] between 15 – 20 mya, vs. 28 mya.”  It is my argument here that any such change would lead to near-commensurate changes in the ages of nodes across their tree.

Overall, it is my opinion that Nicholson et al.’s biogeographic scenario for anoles rests on two potentially flawed decisions about a single fossil calibration.  As reanalayses of Nicholson et al.’s dataset are implemented, I predict that reasonable calibration scenarios for Lazell’s Mexican fossil will produce much younger ages for the root of anoles and will challenge the vicariant scenario hypothesized in Nicholson et al.

Previous

Discussion Of Nicholson Et Al. Monograph Continues

Next

Explaining New Binomials And Species Epithets From The Nicholson Et Al. Classification

7 Comments

  1. 220mya

    Your post illustrates very nicely why fossil calibrations can’t simply be mined from the literature verbatim. Each worker needs to do the leg-work to justify the phylogenetic placement and geologic age of the fossil using modern methods.

    • Yes. If we’ve learned anything over the years that relaxed clock calibration methods have become popular it is that accurate fossil placement is essential, but often complicated by the absence of the characters or other information needed to reliably place fossils in a phylogenetic tree of extant taxa.

  2. It has to be possible to date that amber. We can date bone and wood from the C isotopes. Why not amber? Get us a precise date for A. electrum.

    • Although these fossils are too old to be dated with C isotopes it seems likely that other isotopes could be used to provide reasonable age estimates for the amber fossils. However, I’m not sure this work is really necessary; I think the Mexican amber has already been dated with a reasonable degree of precision using basic geologic principles.

  3. Bob Thomson

    Nice discussion Rich. I’m inclined to agree that the age estimates reported are likely too old, although I don’t completely follow your logic on the ‘red’ node inevitably inducing a minimum age on the ‘blue’ node. If we follow Nicholson et al. and apply lognormal distributions as age priors for the blue and red nodes, we could get an inferred age for ‘blue’ that is younger than the inferred age for ‘red’ if there is a relative speedup in the rate of substitution in the clade that contains ‘blue’. The relative probability of this (as given by the relaxed clock model) would depend on the particular shape of the lognormal priors used for the node ages and the particular shape of the lognormal prior on among-lineage variation in substitution rate (these details don’t seem to be given in Nicholson et al…or maybe I just can’t find them). Informally, the UCLN model is making a tradeoff between how much we’ve told it substitution rates can vary across the tree and what ages for those two particular nodes are credible. If we employ very diffuse priors on the node ages (such that ages for ‘blue’ that are slightly older than 23 million years are roughly equiprobable with ages double that) then the process you outline might certainly be going on. Conversely, if we specify tighter prior age distributions such that there is considerable prior mass on ages that are slightly older than 23 million years, but this attenuates somewhat strongly as ages double, then the model would favor explanations of the data that entail more among-lineage substitution rate variation and node ages that are closer to the minimal prior constraints.

    The fossil calibrations employed by Nicholson et al. could still be problematic for all of the reasons that you outline and I agree with you that the anole crown age looks to be suspiciously old here. One quick (but informal) way to gain an intuition for how much the fossil calibrations might be inducing a bias would be to take a look at a phylogram of the data used in Nicholson et al. and ask how clocklike the tree looks (i.e. do the tips roughly line up?) and then look at the marginal posterior density of the substitution rates in the ‘blue’ vs. ‘red’ clades from their BEAST analysis. If there is a big mismatch, it could be saying something about misspecification of the age priors.

    • Thanks for weighing in Bob. I agree that the scenarios you’ve described are possible. We’re trying to rerun some of these analyses now to see exactly what’s going on. I’ll be surprised if the younger calibration at the older node has much to do with the remarkably old age for anoles reported by Nicholson et al.

  4. cybokat

    What would happen with the age estimates if one were to use the dataset of Alföldi et al. (2011) adding the relevant taxa for ND2 from Nicholson et al. ?

Leave a Reply

Your email address will not be published.

Powered by WordPress & Theme by Anders Norén

%d bloggers like this: