Fossils, molecules, divergence times, and the origin of Salamandroidea

We paleontologists can sometimes suffer from envy when we look at the sheer amount of data that a molecular systematist can bring to bear on a question of evolutionary relationship between species. Our fossils are often fragmentary, and the fossil record can be stingy, especially failing to produce small, fragile taxa such as amphibians over periods of tens of millions of years. Nevertheless, paleontologists have exclusive access to some information from deep time, primarily the combination of primitive and derived states that can tell the story of character transformation within a lineage, assuming the taxon is sufficiently well represented in the fossil record. We also used to claim exclusive access to information on the timing of events of evolution, but because of the development of molecular clock techniques we now have to share this stage. Molecular divergence estimates are highly sensitive to the number, location, and topological distribution of calibration points (1–4), and fossils, being constraints on the younger end of cladogenesis, necessarily underestimate divergence timing. The result of these sets of potential biases is a frequent mismatch in the ages estimated by using either set of data. This discrepancy has certainly been evident with respect to the evolution of extant amphibians (frogs, salamanders, and caecilians; collectively Lissamphibia), with molecular estimates being much older (as much as 100 Myr) than the fossil record suggests (4–11). However, a study in PNAS (12) shows that rapprochement within at least salamander phylogeny may be within grasp.

Salamanders are the morphologically most generalized of the three extant groups, not having the locomotor specializations of their sister taxa frogs (i.e., jumping) or caecilians (i.e., burrowing). There is an important divergence near the base of the salamander phylogenetic tree (Fig. 1) between the cryptobranchoids (Cryptobranchidae and Hynobiidae), and salamandroids (all other salamanders, excepting Sirenidae, whose placement remains controversial). Until the PNAS study (12), the oldest known salamandroid was from the Cretaceous of Spain (13). Gao and Shubin's new species extends this range by 40 Myr, with important implications for our understanding of salamander evolution.

The specimen is not only the oldest salamandroid, but it is placed lowest on the stem, preserving an interesting mosaic of derived and primitive features. It retains gills and lateral line canal grooves, hallmarks of aquatic, larval life, but it is clearly adult from several osteological indicators, demonstrating it was neotenic like many modern salamanders. A laterally compressed and fin-bearing tail reinforces this interpretation. It retains several bones that, in extant salamanders, are either highly reduced or lost entirely, which is of great interest to specialists. Most significantly, it bears monocuspid, nonpedicellate teeth. In pedicellate teeth, a classic synapomorphy of Lissamphibia (14), the tooth cusp (usually bicuspid) is separated from the base (or pedicel) by a zone of poor mineralization (15). Nonpedicellate teeth are also seen in stem group salamanders such as Karaurus and Kokartus (16), and Gao and Shubin (12) argue that pedicely might be a condition derived within salamanders, thus questioning its status as a synapomorphy for the group as a whole. Although this inference is certainly possible, pedicely and bicuspidality are ontogenetically variable conditions within an individual salamander's life history (17). Furthermore, when one considers the most probable lissamphibian outgroup, the amphibamid temnospondyls (18), pedicely and bicuspidality have been reported, in some cases with a morphology nearly indistinguishable from what is seen in modern amphibians. Regardless, this is an intriguing specimen that has much to teach us about salamander evolution.

It is the 40-Myr age range extension that has the most exciting implications. A number of estimates for the divergence of cryptobranchoids and salamandroids have been made in the past several years (5–9). These estimates vary, but in general they have become progressively younger as the use of multiple calibrations from the fossil record (3, 4, 19) has become more commonplace. However, these estimates are still much older than suggested by a direct reading of the fossil record. One recent attempt to infer divergence by using just fossils (10) estimated this split to be at approximately 143 Mya, which is more than 30 Myr more recent than the youngest of the molecular estimates, and 80 Myr younger than the average of the molecular estimates.

This discrepancy results from a few issues with the fossil data that skew this estimate toward younger ages, and biases in the molecular data that skew estimates in the older direction. The first, and probably greatest, effect with respect to this particular fossil-based estimate is the incompleteness of the fossil record for lissamphibians. There are enormous periods of nonrepresentation during which we know there must have been fossils. Approximately 80 Myr span between Triadobatrachus (the first frog) and the stem salamanders Karaurus and Kokartus. Similarly, there are approximately 55 Myr between Triadobatrachus and the first caecilian Eocaecilia, and the next caecilian fossil does not occur for another 45 Myr (20). The fossil record improves as it gets progressively younger, but not evenly or even predictably. This incompleteness makes what is already an underestimate biased toward even younger ages. A second issue is a methodological decision made by Marjanović and Laurin (10) to use “hard” maximal constraints, i.e., making the statement that a clade can definitely not be older than a particular lower bound (in stratigraphic terms, older is toward the bottom and younger to the top of a stacked time series). However, we already know that the fossil record is an underestimate of true age of divergence, so those bounds themselves are just as prone to these same systematic biases, and most of the molecular studies have not used maximal constraints as a result. Finally, there is a disagreement over the age of the earliest cryptobranchoid Chunerpeton (21), which underscores the need for explicitness when using stratigraphic and fossil data to constrain divergence estimates (4).

This discrepancy is where the new fossil salamandroid fits into the picture. Gao and Shubin's new salamander extends the known range of salamandroids by 40 Myr, or 17 Myr older than the time that had been estimated by the fossils alone. This starkly underscores the caution we must take when making statements about evolutionary trends from a direct reading of the fossil record. However, the new fossil salamander is still much younger than most estimates from molecular clocks.

This may be a result of a fragmentary fossil record; fossils await discovery that will create more range extensions into deep time, eventually to approach the dates suggested by the molecular clocks. Alternatively, it might point toward systematic biases within the chosen calibration method (1) or the molecular data themselves (22).

One recent molecular study estimated a time of divergence that fits very well with this new fossil (5). The authors used fossil data in a number of innovative ways. First, they used a number of external and internal calibration points from the fossil record as hard minimal constraints (i.e., the clade must have diverged by this point). They then took the suggestion of using maxima (10), but instead of using these as absolute or hard constraints, used them as “soft” constraints by creating a distribution of possible ages for these events. The result of their method, depending on the exact parameters used, places the basal salamander divergence slightly (3 Myr) to reasonably (13 Myr) older than this first salamandroid (Fig. 1). Not using any maximal constraint increased the age of estimation for the divergence by 20 to 30 Myr, and using hard maxima made the estimate 10 Myr younger than the soft maxima. The preference for using the soft maxima withstood this test from the fossil record well, and may point the direction for future molecular clock work (1), but further analytical work will be necessary to ensure these results are repeatable and consistent before they become standard practice. However, notwithstanding the caution with which fossil data should be used, they appear to increase the accuracy of estimates of divergence made by molecular clock models, and should be incorporated (with appropriate provisos) whenever possible.