An obvious solution is to use only especially good calibrations, based on parts of the fossil record that paleontologists consider reliable. What qualifies as “especially good” and “reliable” is subjective, but there are some cases that do seem convincing. Consider the branching point between the bird lineage and the crocodile lineage, for example, which has often been used to calibrate bird molecular clocks (see Figure 6.1). A key fossil in this instance is from a creature called Arizonasaurus (discovered in Arizona, of course, although it has also been found in Texas) dating from the early part of the mid-Triassic, about 240 million years ago. Arizonasaurus was an impressive-looking ten-foot-long carnivore with a sailfin on its back made up of long, upward extensions of its vertebrae. The sailfin made Arizonasaurus look superficially like some much older sailfin reptiles related to mammals,26 but details of its anatomy indicate that it belongs on the crocodile branch. Assuming that’s the case, it is the oldest fossil known to come after the bird-croc split, and as such, it sets the youngest possible age for that branching point.27 Just a little further back in time, in the period from about 245 to 250 million years ago, there are many fossils of close relatives of the bird/croc group from all over the world, but nothing from within that group itself. In other words, there are indications that the bird/croc group did not yet exist at that time. We might therefore conclude, conservatively, that the bird-croc branching point could not have occurred earlier than 250 million years ago. Splitting the difference between our minimum and maximum ages for the bird-croc split—240 and 250 million years ago, respectively—we might then come up with a calibration age of 245 million years. There are quite a few other branching points, for vertebrates and other groups, for which similarly reasonable ages can be given.
This bird/croc example brings up another aspect of dealing with the calibration problem, namely, incorporating uncertainty in the age of a branching point into the analysis. In the bird/croc case, the calibration could be entered as the whole range of ages between 240 and 250 million years ago, rather than as exactly 245 million years ago. Using a range of ages instead of just a single value will make the molecular clock calibration less precise but more defensible. By analogy, if you were estimating how long a drive would take and didn’t know the driver, it would be prudent to calculate the time based on a range of possible speeds rather than just one. Many recent studies do incorporate such uncertainty for all calibration points.28 Also, some investigators have used several to many calibration points, then figured out which ones were out of synch—making the clock run especially fast or slow in comparison to the other points—and removed these outliers in the final analysis. Do these various improvements solve the calibration problem? Not completely. Bad calibrations can creep in, and it’s even conceivable that a set of calibration points could be consistent with each other but consistently inaccurate; by analogy, it is certainly better to have multiple witnesses to a crime than one witness, but it’s also possible that all of those witnesses are lying. Nonetheless, in studies that use “especially good” calibrations and remove outliers, calibration points that are far out of line with reality are probably rare.
6.1 Fossil calibration of the branching point between birds and crocodylians. Arizonasaurus, from 240 million years ago, is placed on the croc branch, indicating that the bird-croc split had already occurred by then. Mya = millions of years ago. Between 245 and 250 million years ago there are many fossils of close relatives of birds and crocs, but none that seem to be within those two lineages, suggesting they did not exist at that time. The bird-croc branching point is therefore assumed to have occurred sometime between 240 and 250 million years ago.
The second big issue—the problem with molecules—is the well-documented unclock-like nature of the clock, that is, the fact that genetic changes occur at different rates in different lineages.29 Basically, the more variable, and therefore unpredictable, the clock is, the more difficult it becomes to estimate the ages of branching points, and it turns out that the clock is pretty variable. I mentioned in Chapter Five that the clock tends to run rapidly in mammals and extremely slowly in sharks, and that, within mammals, it runs especially fast in rodents (about ten times faster in mice than in humans). This is just the tip of the iceberg; in studies that include many groups, it’s fairly rare to find that the clock is running at close to the same rate in all lineages.
The reasons for the inconstancy of the clock are complex and only partially understood. Lineages with short generation times often have fast clocks, probably because mutations mostly occur when the DNA is replicating, and short generations mean more replications over a given period of time. That might explain why, for instance, rodent clocks run faster than those of people or rhinos or elephants. Ultraviolet radiation causes mutations, and that could be why at least some plants that live in exposed, sunny places have fast-running clocks. The effect of natural selection differs among lineages and undoubtedly has an effect on overall rates of genetic change in some cases. For instance, selection can be relaxed in parasites, because some of their physiological functions are performed by their hosts, and this means that many mutations that would be weeded out in nonparasites can spread within such parasite populations; that effect can translate to a faster clock. (I worried about that for the tapeworms, and got around it by only using changes in the DNA that should not have been influenced by selection.) Similarly, in small populations, the influence of natural selection, which mostly acts to prevent genetic change, is easily overwhelmed by random substitutions in the DNA sequence (genetic drift), again leading to faster rates of change.
The bottom line, however, is that none of these factors, or any of the many others that might speed up or slow down molecular clocks, have such clear and consistent effects that they can be relied upon to predict the rate for a particular group. So, when it comes to molecular dating, none of these insights into what might change clock speed have much practical impact; we’re still faced with a fickle, inconstant, and unpredictable clock. What this inconstancy means is that, even if you have accurate calibration points, you cannot know for sure how fast the clock is running in other parts of the tree, including the parts for which you’re trying to calculate ages. To put the problem in extreme terms, it’s like trying to predict how long a journey will take without knowing whether you’ll be riding in a horse carriage, a car, or an airplane.
The inconstant clock, even more than the problem with fossil calibrations, has been considered the Achilles’ heel of molecular dating. To make use of the flood of DNA sequences—that big pile of bottles—people realized that this problem had to be addressed. As a result, over the past ten years or so, there has been a big push to develop so-called relaxed clock methods, which do not assume the clock is constant and instead estimate when and how much it shifts. These methods don’t use generation times, exposure to UV radiation, and the like, but instead are more like mathematical puzzle-solving devices that search for the set of rates of genetic change that best fits the data at hand, that is, the DNA sequences from the different lineages and the collection of calibration points. In essence, these methods are designed to figure out which lineages are horse carriages and which are airplanes (or anything in between) and calculate ages accordingly.
When these relaxed clock methods have been tested using computer programs that make simulated DNA sequences evolve along branches of a simulated evolutionary tree, they usually give pretty accurate ages. Of course, simulations are not the real thing, and one could always argue that such computer games have left out some important aspect of the actual evolutionary process. Nonetheless, these methods seem to be improvements over the older methods, which assumed a constant clock.30 When several different relaxed clock methods are applied to a very large amount of data (meaning many genes and many sampled groups), in conjunction with a large number of calibration points scattered throughout the evolutionary tree in question, the results can be convincing.
Consider, for
instance, a 2011 study done by a team of evolutionary biologists, led by a mammalogist named Robert Meredith, to construct a “timetree” for mammals as a whole. This research team analyzed parts of 26 different genes (a total of about 35,000 base pairs—the bits of information—for each sampled species) and included representatives of all mammalian orders and more than 97 percent of the families. My friend and colleague John Gatesy, who was part of this team, told me that, by using calibrations from different groups—for instance, only whales versus only rodents—one could get vastly different branching-point ages for the whole tree. That kind of problem is expected for a large group like mammals, and it reflects the inconstancy of the clock. However, this group of scientists didn’t use just one or a few calibration points; they included 82 calibrations spread throughout the evolutionary tree of mammals, with each one entered as a distribution of possible ages rather than as a single age. Then they analyzed their dataset using several kinds of relaxed clock methods that made different assumptions about the evolutionary process. Basically, they did exactly what you’re supposed to do in a molecular dating study: they used many species, many genes, and many fossil calibration points, and they incorporated uncertainty about the ages of those calibrations and about how genetic change happens. No doubt someone will improve on this study, probably in the near future, by using more species, more genes, more calibrations, and yet-to-be-invented methods that take into account more of the complexity of evolution. Nonetheless, it would be surprising if the mammal timetree from this study was severely distorted. It would be surprising if it was very far from the truth.
For our purposes, an important finding from this state-of-the-art mammal timetree is that it validates the biogeographic conclusions of all the mammal cases mentioned in this book, despite the fact that those conclusions were based on earlier studies using less sophisticated methods and far fewer data. Other groups lack such a standard for comparison. However, a massive survey of molecular dating results, called the Timetree of Life project, suggests general agreement among recent studies for most groups. For the taxon I know best, the snakes, the two most convincing studies gave strikingly similar branching ages despite using different sets of genes. This general agreement among molecular dating results is one reason to believe the forest—that is, the overall picture—even if some published studies, individually, remain “pretty suspect.”
THE MOLECULAR CLOCK AND THE PITFALLS OF EXTREMISM
The incompleteness of the fossil record has led some biologists to turn to an alternate source for calibrating molecular clocks, either for particular fossil-poor groups or in general. Specifically, these researchers have used fragmentation events, such as the separation of landmasses due to continental drift or sea-level rise, to set the tempo of the clock. The notion here is that if an evolutionary branching point came about because of such a geologic or climatic event—that is, because of vicariance—then the ages of the branching point and the physical fragmentation would have to be the same. This kind of calibration would introduce circular reasoning if the goal of the study was to evaluate the reality of that particular vicariance event, but the method can be useful for other purposes. For instance, a team of Japanese researchers used a fossil-calibrated clock of cichlid fishes to indicate that deep evolutionary separations in this group were caused by Gondwanan breakup. They went on to argue that those deep splits could be used as calibration points for future studies in place of or in addition to fossil calibrations. That all seems perfectly reasonable. However, in other cases, researchers have relied entirely and uncritically on calibrations that assume vicariance, and this approach has produced some very bizarre age estimates for groups. In fact, the ages are so anomalous that they serve as strong warnings against accepting vicariance too readily.
Consider, for example, a 2004 study on amphisbaenians, limbless or two-legged, burrowing reptiles that are often called “worm lizards” because they look like oversized earthworms. To calibrate a molecular dating analysis, the authors of this study chose a branching point separating a group of South American worm lizards from their African relatives, setting the age of the split by the opening of the Atlantic Ocean 80 million years ago. (This age for that geologic event is actually too young, but a more accurate age would only make the example more extreme.) Thus, they were accepting that this South America–Africa branching point in the worm-lizard tree had been caused by Gondwanan breakup. Using that calibration, they estimated the ages of some branching points within a genus called Bipes (the single group of two-legged amphisbaenians). These ages within Bipes seem surprisingly old for the separation of very similar-looking species—one point was estimated at 69 million years ago, which was before the extinction of the dinosaurs—but there aren’t any fossils within Bipes that contradict those estimated ages. The analysis also implied, however, that the earliest branching points within amphisbaenians as a whole occurred more than 200 million years ago, and that result is decidedly strange. Amphisbaenians are not an especially early evolutionary branch within the lizard group, yet that age of 200-plus million years is considerably older than any lizard fossil. The fossil record of lizards is substantial, so this is a conundrum; finding that worm lizards are older than the oldest known lizards of any kind is sort of like discovering that you’re older than your grandmother. The problem here almost certainly lies not with the lizard fossil record, but with that calibration point, based on the opening of the Atlantic. In other words, the problem is with the assumption of ancient vicariance.
An even more egregious example comes from Michael Heads, who has stated explicitly that tectonic and other fragmentation events should be used as calibration points in place of what he sees as horribly unreliable fossil calibrations. In criticizing a molecular dating study of hedgehog-like mammals called tenrecs, Heads recommended calibrating the analysis with the split of Malagasy and African tenrecs, setting the age of that point as the time of the separation of Madagascar from Africa 120 to 165 million years ago. As with the amphisbaenian case, this calibration assumes that vicariance was the cause of the piecemeal distribution, and it leads to a similar kind of paradox: tenrecs are not an early branch within placental mammals, yet this calibration age within tenrecs predates the first known placental mammal fossils by at least 55 million years! Another way to look at it is that, if tenrecs are as ancient as Heads suggests, then the placental mammal group as a whole must be about 200 million years old, more than three times older than the earliest known placental fossil. It is probably safe to say that no one with a deep knowledge of the mammalian fossil record believes this is even remotely possible. The problem here is the uncritical acceptance of vicariance and the related notion that tectonic events must be used to calibrate molecular clocks.
WHAT FOSSILS SAY ABOUT THE CLOCK
Making comparisons among different molecular studies for the same group isn’t the only way to evaluate molecular dating results. Another argument focuses on comparing age estimates from the molecular approach with those based on the fossil record. The key here is to limit comparisons to branching points, such as the bird-crocodile split, for which the fossil record should give accurate ages. These fossil-based ages can then serve as yardsticks to judge how well molecular dating analyses are working.
When such comparisons are made, they show that the molecular age estimates match the fossil-based ones well up to branching ages of about 400 million years. Up to that age, the observed data—the points relating molecular estimates to the corresponding fossil ages—are generally not far from the line that represents a perfect match between the two kinds of estimates (see Figure 6.2). The simplest explanation for this result is that, although there is a fair amount of slop in the data, molecules and fossils are converging on the truth, that is, on the real ages of these branching points. The alternative is that both fossil and molecular estimates are wrong, but, for some unknown reason, they just happen to be off in the same direction and to the same degree. That wo
uld be a strange coincidence and, therefore, seems unlikely. Also, and importantly, there is little suggestion from the graph in Figure 6.2 that the molecular estimates are biased to be too young (which would show up as points tending to fall above rather than below the line of perfect match). This lack of bias is critical because it means that, even if molecular estimates are off, which they must certainly be in some cases, there is no tendency for them to generally support oceanic dispersal hypotheses by giving spuriously young age estimates. Errors are likely going both ways, sometimes erroneously strengthening the case for recent dispersal but, just as often, mistakenly weakening the case. In short, if it turns out that molecular dating suggests that ocean crossings have been very important in general, that’s probably because ocean crossings really have been very important.
The Monkey's Voyage Page 17
