Saturday, February 9, 2013

The rise and fall of parallel evolution: dispatches from Belgium and Texas.

Evolutionary biology has traditionally sought support for the power of natural selection through evidence that similar phenotypes evolve in similar environments: flying animals all have wings, cave organisms predictably lose their eyes and pigments, and birds on small islands often lose their wings! If similar traits evolve in similar environments despite independent origins, then surely natural selection is overwhelmingly powerful in shaping the deterministic evolution of organisms. This pattern is generally called “parallel” when it occurs from similar ancestors and “convergent” when it occurs from different ancestors. (Although some interpretations focus more on whether similar [parallel] or different [convergent] genetic changes are involved.) In this vein, countless studies have reported parallel or convergent evolution and thereby provided overwhelming evidence for the power of natural selection.

I will here argue that parallel evolution is – or should be – on the wane. I don’t mean this in the simple semantic sense that evolution is never parallel but is instead convergent; as has been recently argued by a number of authors. I instead mean it in the more subtle, but probably more important, sense that parallel (and convergent) evolution is much less parallel (and convergent) than is normally promulgated – and that evolutionary biologists are increasingly realizing this fact. My inspiration to revisit this topic (see the earlier post) stems from meetings I recently attended in Leuven, Belgium, and Austin, Texas.

The typical approach in studies of parallel or convergent evolution is to test whether independent populations that have evolved in similar environments (or habitats or food types) are more similar than independent populations that have evolved in different environments. If a statistically significant effect of “environment type” is found across such populations, then parallel evolution is invoked. The trouble with using this result to invoke parallel evolution is that independent populations in similar environments, although often superficially similar, do show many subtle (or even obvious) differences. Wings look extremely different for pterosaurs, bats, birds, and insects! Not all cave organisms lose their eyes!  Not all birds on small islands lose their wings! Has our need to invoke the power of selection blinded us to the fact that evolution is not very predictable? Or, in short, how parallel (or convergent) is parallel (or convergent) evolution, really?

Several talks at the Leuven meeting invoked parallelism (or convergence): spiders in Hawaii, spiders in Galapagos, beetles in salt marshes, stickleback in freshwater, and many others. In most cases, however, examination of the data made clear that although all evolution generally occurred in roughly similar directions, the independent populations colonizing similar environments were never the same either morphologically or genetically. That is, they might share some phenotypic or genetic similarity but the differences among different populations in similar habitats were often, to my eye, as striking as the similarities. A couple of talks near the end of the meeting brought this to the idea to the front of my brain and thus precipitated this post. First, Joop Ouborg showed that the genetic basis for inbreeding depression was almost entirely non-overlapping between different plant species. Second, Freddy Chain showed that lake-stream stickleback divergence in five independent watersheds was almost entirely non-parallel at the genetic level. In short, the fact that evolution is generally non-parallel (or non-convergent) is remarkably parallel (or convergent).

Of course, these are just my interpretations and I have to confess that I have been wrong before – even quite recently. Take grappa, for instance; that crazy Italian drink made by fermenting the leftovers from the pressing of grapes to make wine. Distinctively musty, this drink is quintessentially Italian and is wonderful when drunk for the first time late at night in downtown Napoli surrounded by locals cheering their football team against Barcelona. Or is grappa really that good? Maybe I just liked it because of the context and novelty. Indeed, subsequent consumptions of grappa have further separated the drink from the context and somewhat decreased my enthusiasm. Well, I told the above sad tale at dinner to Nelson Hairson, Luc De Meester, and Isabelle Olivieri right after Nelson ordered grappa. Bah, says Nelson, grappa is good even outside of the context. Ok, then, I challenged him, name three mainstream drinks worse than grappa. And then Isabelle had a brilliant idea – Nelson should pick one drink on the menu that was worse that grappa. That will get him, I figured. So Nelson takes a careful academic look at the menu and pronounces that “Spiriteux de poivre” would be worse. So we ordered this drink and passed it around the other end of the table, where they had not heard our argument and didn’t know what the drinks were: a single-blind experiment if you will. And the verdict: grappa was not very good – but Spiriteux de poivre was infinitely worse.

The Texas gunslinger re-envisioned
Back to parallel evolution. The idea that (even) parallel evolution is often non-parallel was the focus of the meeting that I had in Austin with Dan Bolnick, Yoel Stuart, Dieta Hanson, Rowan Barrett, and Katie Peichel. We had recently received money from the NSF to quantify the non-parallel and parallel contributions to lake-stream stickleback divergence. We will use 16 independent lake-stream pairs to quantify parallelism at the ecological level, the morphological level, and the genetic level. We will then ask to what extent parallelism and non-parallelism at each level can be explained by parallelism and non-parallelism at the other levels. This meeting was set up to plan our first real field season. But, in reality, that was really just the excuse because my main memories involve eating outstanding tacos and barbeque, drinking copious amounts of beverages from Texas, California, and Scotland, and climbing, both in Dan’s backyard “cave” and at Reimer’s Ranch. (If you are at the NSF, I am just kidding.) To push a metaphor perhaps too far, all climbing routes go in an overall parallel direction (up!) but each one follows its own idiosyncratic route and often ends up in a different place.

Multitasking par excellence- or "why McGill hired Dr. Barrett"


  1. Do we need parallel and convergent evolution to prove evolution?
    The examples of the independent acquirement of flight throughout evolution are striking. Whether or not they represent convergence is irrelevant, or at least merely a semantic issue. When looking at form and/or function, they do indeed represent a convergence: the wings that evolved each time allow the owner to fly, representing convergence in function.
    Along the same line, we should then consider that blue whales, herrings, shovelers (ducks) and tiny water fleas display functional convergent evolution: all have evolved a filter apparatus to feed, and descend from non-filter feeding ancestors. The fact that kangaroos have large hind legs and grashoppers too, allowing them to jump far, is convergent evolution. The hard carapace of crabs and turtles, can be considered convergence if we consider their form and function. Sea urchins and hedgehogs. The list is incredibly long. And rather irrelevant.
    Not convergence in itself is interesting, but rather the pathways used by evolution to find solutions to similar selection pressures and ecological opportunities. Basically, it comes down to one of the fundamental questions of life: when is evolution deterministic, and to what degree? The raw material on which selection can act originates, to a large extent, through stochastic mutation mechanisms. (Let’s not dwell on the potential determinism of endogenous mutational mechanisms that organism have to steer mutations. See for this Shapiro 2012: Evolution, a 21st century view). The large convergences among disparate branches of the tree of life indicate that, at least in function and sometimes also in form, evolution can be deterministic. But can it also be in the genetic background and the pathways used to get there?

    Suppose we could rewind the tape of life to different steps back in time and start over from these points in time, when would evolution take the same course as it took in this world, for how long, and when would it bifurcate? And why?
    A superficial look at artificial breeding and selection indicates that for single traits, evolution through breeder selection can take similar pathways, mostly by selecting for mutations shutting down or at least disrupting expression of a particular gene. Mighty mice, the Belgian blue “double-muscled” cattle, Piedmontese cattle, texel sheep, some racing dogs, … all have a mutation in the myostatin gene, knocking it out and resulting in extraordinary musculature. In form and function, these are parallel evolutions. But each time, a different site in the myostatin gene was targeted, suggesting that evolution, already at this level, is unpredictable. Any mutation knocking out the gene gets the animals where the breeder wants them to go. Form and function are all what matters to evolution. If it walks like a duck, quacks like and duck and looks like a duck, evolution doesn’t care if it isn’t really a duck.

    1. Ah, yes, but the diversity of ducks is huge, which is precisely my point. Just saying that something is a duck isn't that informative since ducks vary hugely in their performance: some can fly, some can dive and others can't, some eat fish and some plants, some are found in the Arctic and some in the tropics, etc. In short, if something quacks like a duck, the important question is "OK, what kind of duck"? This is the same as asking just how parallel (or convergent) is parallel (or convergent) evolution rather than just saying something is or isn't. Moreover, not all ducks quack.

  2. Andrew,
    Your post hits on several concerns I have had on parallelism in recent years so I figured that I would stick my neck out and share my ideas. I may be completely wrong but lets discuss.

    First, have you read Gould's historical account of parallelism and convergence starting somewhere within chapter 10 of his grand treatise? This summary has significantly influenced the way I think about this topic and many of my ideas emerge from taking a historical perspective. Going back to Lankester, Scott, Osborn, and the turn of the 20th century what I took away from Gould's account was that parallelisms, at the time of their conception, represented a special case of convergence, one where similarity in form arose using similar genetic mechanisms BECAUSE of common ancestry. In other words, parallelisms were due to a combination of selection and commonalities in the underlying developmental-genetic architecture among lineages descended from a recent common ancestor. In contrast, convergence among more distantly related lineages clearly represented the primacy of natural selection without the additional affect of development. I think that parallelism and convergence were originally intended to be treated as a continuum, not as discrete entities as they are treated today. The likelihood of parallelism would decrease with the time separating the lineages.

    If we assume that Gould's historical account is accurate and look forward a century we can then ask what happened to these ideas. My feeling is that distinguishing between parallelism and convergence was difficult to operationalize in most systems. For example, developmental analyses and phylogenetic comparative methods have both only recently become mainstream. Therefore, mid 20th century biologists from different specialties latched on to particular pieces of the original parallelism definition and subsequently broadened the use of this term. Rather than needing to fulfill the complex requirements of the original parallelism definition, authors used the term based on only a single criterion (e.g., evolutionary proximity of species, same embryological origins [now same gene], same “starting point”). I believe that the theoretical breadth of parallelism significantly widened during the 20th century to the point where the idea is now entrenched throughout all of comparative biology. Unfortunately, as we now strive for more a synthetic understanding of biological evolution we are struggling to meaningfully reconcile the differences in opinion that have arisen among ecologists, evolutionary biologists, geneticists, and developmental biologists.

    In my opinion, the conceptual identity of parallelism has been battered and bruised to the point where it is no longer recognizable. There are too many bastardized forms of the idea floating around the literature and too many authors fail to be explicit with their definition. But I am not one to get hung up on terms. Who is to say that my definition is anymore valid than anyone else’s? But while the study of parallelism has lost its conceptual rigor, I do not believe that the original intent should be forgotten. I think that we need to move beyond biologically unrealistic expectations of searching for THE gene and accept that variation in most phenotypic traits likely have complex developmental-genetic origins. I also think that we need to pay more explicit attention to the underlying evolutionary mechanisms (e.g., latent developmental potential, inherited patterns of variation, gene flow introducing novel genetic variation, mutational target size) and stop diagnosing parallelisms based solely on molecular or phylogenetic patterns. I think that a quantitative framework analyzing the similarities among different levels of organization is the most promising way to dissect the effects of history and selection. Regardless of whether you decide to call the patterns you find in sticklebacks parallelism or convergence, I look forward to the conclusions of your upcoming study.

    Thom Sanger

  3. I didn't get beyond chapter 9 of Gould's treatise, so I didn't see his historical account of parallel evolution. Nonetheless, I think you did a good job of summarizing it. Can you also please summarize the rest of the chapters in the book? Maybe a Sanger's Notes (like Cliff's Note) for Gould.

    I agree entirely with you that the terms have become quite vague and diverse in their use - but they are still surprisingly useful. At least within a given context. In stickleback work, for example, everyone knows precisely what is meant by parallel evolution. However, my key argument is that we are still thinking about it in the wrong way. I think that we need to QUANTIFY just how parallel (or convergent) parallel (or convergent) evolution really. And we need to find out how parallelism and nonparallelism (and convergence and nonconvergence) at a given level of organization, such as phenotypes or genes, is influenced by variation at other levels of organization.

    And so I find myself in perfect agreement with your statement that: "I think that a quantitative framework analyzing the similarities among different levels of organization is the most promising way to dissect the effects of history and selection."

    ps. (I actually didn't make it beyond chapter 1 of Gould's book - but it does look good on my shelf.)


A 25-year quest for the Holy Grail of evolutionary biology

When I started my postdoc in 1998, I think it is safe to say that the Holy Grail (or maybe Rosetta Stone) for many evolutionary biologists w...