Evolution · Things I've Learned

Natural Selection in the Fossil Record

table.figure {
background-color: #f0f0f0;
table.figure caption {
caption-side: bottom;
font-size: 90%;
padding: 1em;
background-color: #f0f0f0;

For quite some time now, I’ve had a question:

We can see evolution in the present.
And we can see natural selection in the present.
And we can see lots of evolution in the fossil record.
But can we see natural selection in the fossil record?

In case it’s not clear what this question means, recall that “evolution” means a change in traits in a population over the course of generations. For instance, in the well-known example of peppered moths in England, the fact that the population changed from mostly-light to mostly-dark during the Industrial Revolution constitutes evolution.

The next question, of course, is why did this change occur? Natural selection is one mechanism by which evolution can occur (and, indeed, Charles Darwin’s great contribution to science lay not in demonstrating that evolution occurs, which was well known at the time, but in proposing a mechanism to explain it).

But natural selection isn’t the only mechanism by which evolution can occur. In one of my earliest posts here, I looked at neutral drift, I found that a trait that makes absolutely no difference in survival rates can still take over a population, through the mechanism of neutral drift.

So the question remains: we can see evolution in the present, like finch populations changing the average length of their beaks over the course of a few decades. And we can note which factors influence reproductive success (natural selection). We can also set up experiments in which one trait confers a selective advantage, and make populations evolve as a result of this selection. And if you’ve ever visited a natural history museum, you know that there used to be plants and animals that don’t exist anymore (like dinosaurs), that there are plants and animals that didn’t used to exist (like dachshunds and grass), and that successive species often show a smooth transition, so evolution has occurred.

But can we tell whether some given change in the fossil record was caused by natural selection?

Last year, I was fortunate enough to run into paleontologists Nathan Jud and Thomas Holtz at UMD, and they were kind enough to point me at the paper Evolution in Fossil Lineages: Paleontology and The Origin of Species, by Gene Hunt of the Smithsonian’s museum of Natural History, which talks about this very thing.

To begin with, what does natural selection look like? And given that, what sorts of signs would we be able to see in the fossil record? Go back to my earlier post to see what neutral drift looks like.

For contrast, I whipped up a version of Richard Dawkins’s weasel program, except that it tries to evolve Hamlet’s soliloquy rather than the short sentence “Methinks it is like a weasel”.

The string it’s trying to evolve is 1529 characters long. My program starts out with a random string, figures out which one has the largest number of correct letters in the correct position, mutates that 100 times to make the next generation, and starts over. The following graph shows the average fitness (number of correct letters) over the first 6,000 generations:

As you can see, at first the fitness increases rapidly, when there are many ways to improve the string. But as more and more letters are correct, fewer and fewer of them are wrong, there are fewer opportunities for a random change to get something right, so later generations improve more slowly.

In fact, if we look at the first 15,000 generations, we see a second pattern:

After the first 6,000 generations or so, progress virtually stops. Recall that the original string has 1529 characters. And after the first 6,000 generations, the average fitness of the generated strings wobbles around 1500, or about 98% fit. After that, there’s so little room for improvement, and so little to be gained from it, that natural selection can’t overcome the forces of randomness.

In his paper, Hunt mentions a third type of tendency: stasis. I believe this is what we’re seeing on the right-hand side of the weasel graph: the fitness line wobbles a bit due to randomness, but not nearly as much as it would if it were random. Natural selection is acting to preserve the status quo, and any candidate that departs too far from that is prevented from reproducing.

So this gives us something to look for: in the real world, it’s not just abstract notions of “fitness” that change in populations over time, it’s actual traits like body size or furriness or leg length. Hunt illustrates three types of change:

Figure 4: Examples of evolutionary sequences from fossil lineages that represent the three canonical modes of evolution, as indicated by the small‐sample‐size Akaike Information Criterion. Left, directional change in test shape from the planktonic foraminifera Contusotruncana (Kucera and Malmgren 1998); center, a random walk in protoconch diameter from the benthic foraminifera Disclocyclina (Fermont 1982); right, stasis in the length‐to‐width ratio of the lower first molar of the mammal Cantius (Clyde and Gingerich 1994). Each point represents a sample mean; error bars indicate 1 SE. Time is in millions of years elapsed from the start of the sequence.

So if we had enough fossils from successive generations of a given population, we could look at some trait, like tibia length, or shell smoothness, or body length, or some other trait that fossilizes readily, and see whether it matches one of the modes of change: convergence on a target (natural selection or directional change), slight wobbling around a mean (stasis) or lots of wobbling without a goal.

Unfortunately, fossilization is such a rare event that a lot of evolution can happen between fossils. But sometimes, we get lucky:

A remarkably favorable case study for documenting the transformation of a fossil species via natural selection involves skeletal armor reduction in a lineage of stickleback fish from a 10‐million‐year‐old lake in Nevada (Bell et al. 2006). Here preservation is excellent, and fossil fish are numerous and articulated. Most unusually, sediments in this ancient lake were deposited in undisturbed yearly layers called varves. Thus, in principle, time in this environment can be resolved to individual years. In practice, fish are not so abundant in each varve as to allow meaningful analysis, and so specimens were lumped into 250‐year temporal bins. Even with this lumping, the temporal resolution of this system is outstanding compared with what is usually attainable in the paleontological record.

And when you look at a fossilized feature of this population over time (in this case, the average number of spines on the fish’s back), what do you get?:

Figure 3: Evolutionary trajectory in a lineage of stickleback fish from an ancient lake deposit. Each circle represents the average number of dorsal spines (one aspect of skeletal armor) over intervals of approximately 250 years. The dashed line shows the expected evolutionary trajectory for the best‐fit adaptive model, and the gray area is the 95% probability interval for the fit.

I’ll spare you the mathematical discussion, but the gist of it is that this pattern is best explained by a model where having one spine is better than having two or more, and natural selection has pushed the stickleback population that way. Elsewhere, Hunt shows an example of a trait in trilobites whose evolution is best explained by two periods of stasis interrupted by a burst of directed change.

So the short answer is yes, we can see the effects of natural selection in the fossil record. Not usually, but sometimes we can get lucky, and get enough data to draw a conclusion.

And that’s something that awes me about science, and humans: that we’re clever enough to figure out how to make mute rocks speak, and answer questions a pessimist might have given up on.

I also think it testifies to the fact that in science, no person or idea is sacred, that 150 years after the publication of Origin of Species, Hunt was willing and able to say, in effect, “So, this Darwin chap: was he right? Well, as it turns out, mostly yes, and here are the numbers to prove it.”