Just as ecology is beginning to refocus on integrating evolutionary dynamics and community ecology, a paper from Yoel Stuart and Jonathan Losos (2013) suggests that perhaps the best-known eco-evolutionary hypothesis - Ecological Character Displacement (ECD) – needs to be demoted in popularity. They review the existing evidence for ECD and in the process illustrate the rather typical path that research into pattern-based hypotheses seems to be taking.
ECD developed during that period of ecology when competition was at the forefront of ecological thought. During the 1950s-1960s, Connell, Hutchinson and McArthur produced their influential ideas about competitive coexistence. At the same time, Brown and Wilson (1956) first described ecological character displacement. ECD is defined as involving first, competition for limited resources; second, in response, selection for resource partitioning which drives populations to diverge in resource use. Ecological competition drives adaptive evolution in resource usage – either resulting in exaggerated divergence in sympatry or trait overdispersion. ECD fell in line with a competition-biased worldview, integrated ecology and evolution, and so quickly became entrenched: the ubiquity of trait differences between sympatric species seemed to support its predictions. Pfennig and Pfennig (2012) go so far as to say ‘Character displacement...plays a key, and often decisive, role in generating and maintaining biodiversity.’
One problem was that tests of ECD tended to make it a self-fulfilling prophecy. Differences in resource usage are expected when coexisting species compete; therefore if differences in resource usage are observed, competition is assumed to be the cause. In the ideal test, divergent sympatric species would be found experimentally to compete, and ECD could be used to explain the proximal cause of divergence. But the argument was also made that when divergent sympatric species were not found to compete, this was also evidence of ECD, since “ghosts of competition past” could have lead to complete divergence such that competition no longer occurred. This made it rather difficult to disprove ECD.
There was pushback in the 1970s against these problems, but interestingly, ECD didn’t fall out of favour. A familiar pattern took form: initial ecstatic support, followed by critical papers, which were in turn rebutted by new experimental studies. Theoretical models both supported or rebutted the hypothesis depending on the assumptions involved. In response the large literature, several influential reviews were written (Schluter (2000), Dayan and Simberloff (2005)) that appeared to suggest at least partial support for the ECD from existing data. Rather than dimming interest in ECD, debate kept it relevant for 40+ years. And continued relevance translated to the image of ECD as a longstanding (hence important) idea. Stuart and Losos carry out a new evaluation of the existing evidence for ECD using Schluter and McPhail’s (1992) ‘6 criteria’, using both the papers from the two previous reviews and more recent studies. Their results suggest that strong evidence for ECD is nearly non-existent, with only 5% of all 144 studies meeting all 6 criteria. (Note: this isn't equivalent to suggesting that ECD is nearly non-existent, just that currently support is limited. There's a good discussion as to some of the possible reasons that ECD has been rarely observed, in the paper).
The authors note that there are many explanations for this finding of weak support: the study of evolution in nature is difficult, particularly given the dearth of long term studies. The 6 criteria are very difficult to fulfill. But they also make an important, much more general point: character displacement patterns can result from multiple processes that are not competition, so patterns on their own are not indicative. Patterns that result from legitimate ecological character displacement may not show the predicted trait overdispersion. The story of the rise and fall of ECD is a story with applications to many pattern-driven ecological hypotheses. There are many axiomatic relationships you learn about in introductory courses: productivity-diversity hump shaped relationships, the intermediate disturbance hypothesis, ECD, etc, etc. These have guided hypothesis formation and testing for 40 years and have become entrenched in the literature despite criticism. And similarly, there are recent papers suggesting that long-standing pattern-based hypotheses are actually wrong or at least misguided (e.g. 1, 2, 3, etc). Why? Because pattern-driven hypotheses lack mechanism, usually relying on some sort of common-sense description of a relationship. The truth is that the same pattern may result from multiple processes. Further, a single process can produce multiple patterns. So a pattern means very little without the appropriate context.
So have we wasted 40 years of time, energy and resources jousting at windmills? Probably not, data and knowledge are arrived at in many ways. And observing patterns is important - it is the source of information from natural systems we use to develop hypotheses. But it is hopeful that this is a period where ecology is recognizing that pattern-based hypotheses (and particularly the focus on patterns as proof for these hypotheses) ask the right questions but focus on the wrong answers.
ECD developed during that period of ecology when competition was at the forefront of ecological thought. During the 1950s-1960s, Connell, Hutchinson and McArthur produced their influential ideas about competitive coexistence. At the same time, Brown and Wilson (1956) first described ecological character displacement. ECD is defined as involving first, competition for limited resources; second, in response, selection for resource partitioning which drives populations to diverge in resource use. Ecological competition drives adaptive evolution in resource usage – either resulting in exaggerated divergence in sympatry or trait overdispersion. ECD fell in line with a competition-biased worldview, integrated ecology and evolution, and so quickly became entrenched: the ubiquity of trait differences between sympatric species seemed to support its predictions. Pfennig and Pfennig (2012) go so far as to say ‘Character displacement...plays a key, and often decisive, role in generating and maintaining biodiversity.’
One problem was that tests of ECD tended to make it a self-fulfilling prophecy. Differences in resource usage are expected when coexisting species compete; therefore if differences in resource usage are observed, competition is assumed to be the cause. In the ideal test, divergent sympatric species would be found experimentally to compete, and ECD could be used to explain the proximal cause of divergence. But the argument was also made that when divergent sympatric species were not found to compete, this was also evidence of ECD, since “ghosts of competition past” could have lead to complete divergence such that competition no longer occurred. This made it rather difficult to disprove ECD.
There was pushback in the 1970s against these problems, but interestingly, ECD didn’t fall out of favour. A familiar pattern took form: initial ecstatic support, followed by critical papers, which were in turn rebutted by new experimental studies. Theoretical models both supported or rebutted the hypothesis depending on the assumptions involved. In response the large literature, several influential reviews were written (Schluter (2000), Dayan and Simberloff (2005)) that appeared to suggest at least partial support for the ECD from existing data. Rather than dimming interest in ECD, debate kept it relevant for 40+ years. And continued relevance translated to the image of ECD as a longstanding (hence important) idea. Stuart and Losos carry out a new evaluation of the existing evidence for ECD using Schluter and McPhail’s (1992) ‘6 criteria’, using both the papers from the two previous reviews and more recent studies. Their results suggest that strong evidence for ECD is nearly non-existent, with only 5% of all 144 studies meeting all 6 criteria. (Note: this isn't equivalent to suggesting that ECD is nearly non-existent, just that currently support is limited. There's a good discussion as to some of the possible reasons that ECD has been rarely observed, in the paper).
 |
| From Stuart and Losos (2013). Fraction of cases from Schluter 2000, Dayan and Simberloff 2005, and this study that meet either 4 or all 6 of the criteria for ECD. |
The authors note that there are many explanations for this finding of weak support: the study of evolution in nature is difficult, particularly given the dearth of long term studies. The 6 criteria are very difficult to fulfill. But they also make an important, much more general point: character displacement patterns can result from multiple processes that are not competition, so patterns on their own are not indicative. Patterns that result from legitimate ecological character displacement may not show the predicted trait overdispersion. The story of the rise and fall of ECD is a story with applications to many pattern-driven ecological hypotheses. There are many axiomatic relationships you learn about in introductory courses: productivity-diversity hump shaped relationships, the intermediate disturbance hypothesis, ECD, etc, etc. These have guided hypothesis formation and testing for 40 years and have become entrenched in the literature despite criticism. And similarly, there are recent papers suggesting that long-standing pattern-based hypotheses are actually wrong or at least misguided (e.g. 1, 2, 3, etc). Why? Because pattern-driven hypotheses lack mechanism, usually relying on some sort of common-sense description of a relationship. The truth is that the same pattern may result from multiple processes. Further, a single process can produce multiple patterns. So a pattern means very little without the appropriate context.
So have we wasted 40 years of time, energy and resources jousting at windmills? Probably not, data and knowledge are arrived at in many ways. And observing patterns is important - it is the source of information from natural systems we use to develop hypotheses. But it is hopeful that this is a period where ecology is recognizing that pattern-based hypotheses (and particularly the focus on patterns as proof for these hypotheses) ask the right questions but focus on the wrong answers.
 |
| Long-term studies of Darwin's finches have provided strong evidence for ECD. |
8 comments:
A study shows that among 30 ecperiments, only 3 have shown that a feather falls with the same speed of a ball of iron (the 3 experiments are in void, whereas the other 27 are carried out in the atmosphere, as can be read in the supplementary material). This meta-analysis clearly shows that the claim of Galileo that speed of fall is independent of mass is practically not-existent on earth. Possibly there is some truth in the ideas of Galileo, but they are context-dependent (the presence of air or not) and therefore they prevent the progress of science.
Hi Hans - is this paper proof that ECD isn't important? No, not at all. I didn't go into detail, but they discuss the different possible reasons we aren't doing a good job of experimentally/observationally capturing ECD. Definitely worth reading. (i'll adjust the text to make this more clear). Likely ECD is important but more limited than its most optimistic supporters have suggested.
Although I yield your point that a meta-analysis is inappropriate in such a scenario as the one you describe, the point of the authors' meta-analysis was more to point out that the amount of support is less than often suggested. No one has considered the role of context, in the current or previous meta-analysis. This is obviously a fair critique of meta-analyses in general though.
I think that's separate from the overall point. If this was physics, we would have a strong mechanistic basis and we could predict what we expect given the context of the experiment (air/no air).
"The truth is that the same pattern may result from multiple processes. Further, a single process can produce multiple patterns."
Absolutely! But you pose this as a criticism of pattern based inference when in fact it would be just as damning to mechanism driven work in the absence of patterns. If a single process can produce multiple patterns, then process has no more power to predict than pattern. So while "we could predict what we expect give the context of the experiment", we would still have to test it empirically (using patterns).
"So a pattern means very little without the appropriate context."
No, the pattern IS the context! Mechanism also means very little without the pattern! We can insert a different pattern (than ECD) and think about this. Lets use the latitudinal diversity gradient. There are MANY proposed processes for why there are more species in the tropics. If you found that net diversification (speciation minus extinction) was higher in the tropics, but had no idea that there were more species in the tropics, what would that information tell you? You would be left with a process that explains nothing. Or, you might discover net diversification is higher in temperate regions that in the tropics, and infer that species diversity is higher in temperate regions that tropical regions. See the issue that you really raise, is that induction often fails, not that pattern based inference is somehow less robust than process based inference.
The only reason that we are able to sit around and talk about ECD (or any other ecological phenomenon) is that we have documented general patterns (e.g. divergent species in sympatry). Sometimes when we use pattern to infer process we discover we are wrong, the same can be said for process without pattern.
*quick footnote
My main beef is the death to pattern based hypothesis testing angle, but just a quick note about ECD. ECD may not be explain divergent species in sympatry, but that does not mean that competition is not the mechanism driving this pattern. Divergence may have occurred in allopatry and those individuals move back into sympatry after the matter, but it's likely competition that prevents co-occurence until after divergence.
Hey Devin - too bad you aren't around to debate with in person these days :)
I think we will ultimately disagree on some of these issues, but I think one issue is that I haven't defined my terms clearly enough that so we're speaking about the same thing. I'll try to clarify. I think a good starting point is that observational patterns have played a big role in ecology, so that's what I'm particularly referring to. *Also patterns are clearly a source of information, and we have needed them to develop hypotheses and inform us. What I object to is their use as "proof" on their own, without manipulation or understanding of the mechanisms at play, of the activity of a particular process.* So the latitudinal gradient exists, has informed many hypotheses, but on its own, its existence alone does not "prove" that speciation rates (for example) are higher in the tropics. You need to start with that hypothesis, understand its predictions (including context-dependent aspects), etc. Why? because multiple hypotheses could be supported by the existence of the latitudinal diversity gradient pattern.
So I'm hardly saying, ignore the patterns in nature that inform how we understand nature. I'm saying good science doesn't become caught up in looking for patterns, and suggesting they are evidence for a particular popular hypothesis (whether a latitudinal gradient, a productivity-diversity curve, etc). I think that community ecology in particular has been guilty of this.
Let me see if I can respond to your points. (sorry I couldn't figure out a tidier way of quoting)
"If a single process can produce multiple patterns, then process has no more power to predict than pattern. So while "we could predict what we expect give the context of the experiment", we would still have to test it empirically (using patterns). "
->Of course process has power, it makes predictions about expected (including context dependent) results. It's not unexpected that multiple possible outcomes could result from a given process. It's totally fine to experimentally test for the outcome (what you're calling patterns). If you are moving from mechanistic-hypothesis to predicted evidence and then you find that evidence, that's great. It's not "proof" in a strict framework, but it's support for your hypothesis.
A simple example from physics: you drop a ball from a tower. The speed at which it falls is dependent on multiple contexts (wind, ball materials, etc), so predicting the landing point depends on understanding the underlying process (gravity), and the contextual information. Varying context means there are different landing points or speed (multiple patterns), but this hardly makes the process "powerless".
""So a pattern means very little without the appropriate context."
No, the pattern IS the context! Mechanism also means very little without the pattern! We can insert a different pattern (than ECD) and think about this. Lets use the latitudinal diversity gradient. There are MANY proposed processes for why there are more species in the tropics. If you found that net diversification (speciation minus extinction) was higher in the tropics, but had no idea that there were more species in the tropics, what would that information tell you? You would be left with a process that explains nothing. "
->I disagree. Latitudinal gradients obviously exist, and they allow us to generate hypotheses about process. BUT they can't really be used to inductively ascertain the processes at play.
(cont'd below)
"The only reason that we are able to sit around and talk about ECD (or any other ecological phenomenon) is that we have documented general patterns (e.g. divergent species in sympatry). Sometimes when we use pattern to infer process we discover we are wrong, the same can be said for process without pattern. "
->Probably! We are often wrong. Information is not wasted. Understanding systems is important. But we are never content with just describing and identifying patterns - we want to understand them, which requires process and then testing these processes deductively.
"*quick footnote
My main beef is the death to pattern based hypothesis testing angle, but just a quick note about ECD. ECD may not be explain divergent species in sympatry, but that does not mean that competition is not the mechanism driving this pattern. Divergence may have occurred in allopatry and those individuals move back into sympatry after the matter, but it's likely competition that prevents co-occurence until after divergence."
->Yes! but this is the point of the paper, no? Just checking the boxes for look, I observed a pattern of divergence in sympatry presence isn't tell us that much about the dynamics in that system. It's possible, if you want, to understand more detailed molecular relationships between species, to experimentally test for competition or the absence thereof, and if we are content to stop at pattern we miss this much more useful approach to understand.
Great article, I appreciate your point that multiple processes can produce the same pattern but I agree with Devin that this is equally a problem for process-based and pattern-based studies.
Post a Comment