In the last bit of tidying up after our trip, Holly asked us to comment on a recent paper in Nature (An epistatic ratchet constrains the direction of glucocorticoid receptor evolution, Bridgham et al., 461, 515-519, 24 September 2009), related to the question of whether evolution can, or does, ever go backwards. This is a long-standing question among evolutionary theorists; indeed, in 1890, Belgian paleontologist Louis Dollo addressed it by proposing what is now known as Dollo's Law: "An organism is unable to return, even partially, to a previous stage already realized in the ranks of its ancestors."
But, what does it mean to ask whether evolution can reverse itself? As Bridgham et al. themselves point out, there are many ways to make a given trait, not all genetically equivalent. So, is evolutionary reversal the replication of an older form, which wouldn't necessarily have to involve the same genes as the original iteration? Or, does it mean replication of the same form, produced by the same genes? The first, reproduction of an ancestral form in a novel way, is much more plausible than the second, given the apparent strength of contingency and cooperation in development; what gets made next depends on what's here now, and that depends on what was just here. Thus, to reproduce an ancestral form would require that each step in the evolution of a protein be reversed, along with each step in the evolution of the proteins with which it works. And, because, say, a receptor cooperates with a ligand, although imprecise receptor/ligand binding often works, neither can change too drastically too quickly, lest the signal telling the cell what to do next won't get sent. So, reversal of form and DNA sequences would require reversal of each of the co-evolutionary steps that led to the modern form, which seems highly implausible.
In addition, reversal to a previous form, via the same genes or not, would require the return of the environment in which that form was once successful, and that makes it that much more implausible, since environments are always changing. As Bridgham et al. say, "The past is difficult to recover because it was built on the foundation of its own history, one irrevocably different from that of the present and its many possible futures." So, because of so much that is known about biology, we would a priori agree with Dollo's Law.
All that said, Bridgham et al. actually put the question to a test. They engineered a hormone receptor protein to mimic its ancestral state and the steps it took to become the modern receptor it is, and evaluated its function along the way. In its ancestral state, the receptor was 'promiscuous', able to bind with different classes of hormones, but over 40 million years, and with an estimated 37 different amino acid changes, it lost that ability and became specific to a single class of hormone. The authors found that only 2 amino acid changes were necessary for the receptor's evolved specificity but that subsequent amino acid changes that "optimized the new specificity of the glucocorticoid receptor, also destabilized elements of the protein structure that were required to support the ancestral conformation." Thus, building in just the 2 original changes wasn't sufficient to allow them to resurrect the receptor's ancestral function. They conclude that too many changes are required for successful reversal of the receptor to its ancestral function. Does this prove that evolutionary reversal is impossible? No, but it does suggest that contingency and cooperation are indeed foundational principles in development and evolution, and are important reasons why reversal is unlikely.
Perhaps the points can be more clearly and immediately seen if we ask how probable rather than plausible they are. A mutation replacing an A with a T in DNA can be reversed, the T being replaced with an A--changes that do occur. But because in general mutation is very improbable at any given spot in DNA, the same specific mutation is even less probable. And even if that chance were, say, 1/1000 (much much greater than is actually the case), if 100 of these reversals were needed, the chance would be 1/1000 to the hundredth power, infinitessimally small. And this doesn't take account of the order, viability of intermediate stages in the reversal process, etc. So while most things like this might be possible, they are too unlikely to take seriously if we're thinking about anything at all complicated.
And we can seal the 'no' deal in two other ways. First, we already know that things can reverse. People who have a certain height can have shorter children but their children could again be taller. Much of the time this will be due to different genotypes.
And many mutations involve deletions of DNA, sometimes of chunks many nucleotides long. It would take a fairy godmother to wave a wand to reverse this and somehow conjure up the exact chunk to be re-inserted some time later.
Darwinism as natural law
Here's another way to think of it. Evolutionary theory, since Darwin, has attempted to be a natural science based on natural law. Darwin was very Newtonian in this, suggesting that natural selection was a kind of 'force', rhetoric often used today even though we have a strong sense of probabilism (including both mendelian sampling of genes from parent to offspring, and genetic drift in which the frequency of genetic variants changes in a population strictly by chance).
In a perfectly Newtonian world, nature is predictable and retrodictable. If you know the state today you can predict tomorrow, or tomorrow you can predict today (e.g., by changing the sign of the equation from plus to minus, so to speak). This was an ideal until the 20th century, roughly speaking, when disturbances such as quantum theory showed that things were not so uniformly homogeneous in time.
Largely stimulated by Darwinian thinking, even physicists began to realize that time had a direction. If change is probabilistic, we can go from today to some state tomorrow, but tomorrow we can't tell what today's state had been.
In principle, if we compare DNA sequences between species, we see that evolution diverges forward in time, as mutational changes occur in different descendant copies of a gene from generation to generation, producing a branching or tree-like structure of sequence relationships. These are presumably related to traits, like fingers and leaves, and natural selection (and drift) produce differentiated, adaptive organisms over evolutionary time.
In a sense, this would not seem to be reversible. Once you can make a limb (on a tree or on your body), it is so complex a process that you can destroy it but you can't go back to the state before it was a limb.
However, if you just look at the nucleotides, as we said above, traits of organisms involve many different genes, and the probability of everything being exactly reversed is trivially small, even if mathematically possible.
If you want to be a stickler for exactness then you have the answer: reversal is technically possible but in practice impossible. But if you look at traits or function, then evolution certainly reverses itself. That's what happened in the evolution of flightless birds. Different lineages of insects have repeatedly evolved, or de-evolved, similar states related to the number of ocelli (small central eyes), and different lineages of amphibians have gained or lost tadpole stages in their development. Some of these may involve reverse mutations in the same genes, but there are undoubtedly different genetic pathways 'forward' as well as 'backward' in this phenotypic sense.
As so often is the case, the answer depends on the question. Dollo's principle seems reasonable (if not a 'law' in the cosmic sense) in regard to complex adaptive traits. Different versions of what seem to be the 'same' trait usually have at least some genetic differences. In that purist sense, you really can't go home again.
-Ken and Anne