An fascinating new paper published this week in the journal PLoS One demonstrates how selection acting at least three different levels produces distinct selective pressures that shape the song behavior of male Dupont’s lark (Chersophilus duponti) in the Ebro Valley of northwestern Spain. Authored by Paola Laiolo and José Ramón Obeso and entitled “Multilevel Selection and Neighbourhood Effects from Individual to Metapopulation in a Wild Passerine“, the paper presents compelling evidence that male birds seeking to attract females and repel competitor males face conflicting selective pressures.
Male song repertoire has been shown to be under sexual selection pressure in a variety of birds, but no study has explored how the social interplay of individuals producing songs affects selection. If females tend to prefer males with larger repertoires, why do we see broad variation in the repertoires displayed by males, and what sets the limit on repertoire size? To tackle this question, Laiolo and Obeso designed an ingenious observational study that allowed them to consider song production and population dynamics at three different levels: 1) the individual bird; 2) a population of individual birds; and 3) a metapopulation of populations. Taking advantage of the fact that sound recordings could be used as both a means of censusing populations (because each bird has its own unique ‘audio fingerprint’) and assessing the song repertoire of individuals, Laiolo was able to gather data relevant to all three scales of social organization. This allowed two different comparisons of selection operating at different levels:
As this diagram from the paper shows, the success of both individuals and their neighbors (their local population) could be assessed by considering the individual life span versus individual song repertoire. This allows for what the authors call a “Level 1″ comparison of the effect of having a particular repertoire within a population versus the effect of the average repertoire in a particular neighborhood. Similarly, a comparison can be made at “Level 2″ between different local populations and the overall metapopulation within which they are nested; for this comparison, overall intrinsic growth rate rather than life span could be used as a more direct measure of fitness.
One of the challenges of detecting selection acting on multiple levels is that you need a lot more data, and one of the ways that this study produces a lot of data is to use the audio fingerprints of each bird as a means of both determining individual lifespan and estimating population size. In essence this is a variant of the classic mark-recapture method, but in this case only the songs of individual birds need to be ‘captured’. Data was collected from 2004-2008, so that also added to the sample size. Because the study was designed to consider variation at three different scales of social organization, it had to remove any significant sources of covariation before comparing each level. Analysis of their data suggested that patch size was a covarying factor, and multiple regression models were used to correct for this covariation.
The results at Level 1 (individual versus local group) are shown in Figure 2:
This is a fascinating and unusual result. The classic case of multilevel selection would involve a trait that was advantageous at the individual level but disadvantageous at the group level. With each level of selection pushing in the opposite direction, the expected result would be an intermediate average trait value: a compromise between each level of selection pushing in opposite directions. But in these larks, what we see is a little bit different. There is still a compromise between levels of selection, but in this case the nature of selection at each level is more complex. At the individual level we see what is called ‘disruptive selection’, with the lowest-fitness individuals possessing song repertoires of intermediate size. The authors suggest that this may be due to the stability of both “honest signaling” by males with large repertoires and a ‘freeloading strategy’ by floaters who try to gain matings not through their song repertoire by fooling females; the study did not, however, collect data that could support or refute this suggestion. At the local neighborhood level, we see a very different pattern. The groups with the highest average life-span are those with an intermediate repertoire size: an example of ‘stabilizing selection’, a selection pressure that is directly opposite to the disruptive selection found at the individual level. This is strong evidence that multilevel selection is important, because selection at each level is ‘pushing’ traits in different evolutionary directions.
Why might there be this pattern? What is causing what is good at the individual level to be disfavored at the level of the group? Well, those with large song repertoires do best within their groups because their large song repertoire is an honest signal of their fitness and leads them to have the greatest lifespan in their group. But when many other males with large repertoires are in that same neighborhood group, this causes a decrease in fitness, presumably because strong interference competition has an adverse effect on life-span. Groups with low average repertoire size also display lower life-span, presumably because the low overall repertoire size is an ‘honest’ indication of low overall fitness. The best groups are those with intermediate average repertoire sizes. This makes sense because competition may be less intense in these groups, but exactly why is not clear: it could either be because most individuals have moderate-sized repertoires and therefore engage in less intense competition, or because these intermediate-average groups are heterogeneous, composed of both individuals with either very small or very large repertoires. Where along the spectrum of these possibilities the actual observed groups fell is not entirely clear in the paper.
This is unfortunate, because this is a key point. In their introduction, the authors ask how we can distinguish between frequency-dependent selection and group selection, both of which make the environment of the group important for fitness. Without having an understanding of the variance displayed by groups of intermediate average repertoire size, it is a bit difficult to know what kind of selection this represents. If the groups are homogenous, that points towards unequivocal group selection, as the trait that is disadvantageous within the group would be best for the group as a whole. If these groups are heterogeneous, that suggests that frequency-dependent selection is the explanation: it is good to be the lone male with a large repertoire in a group, but add too many other large-repertoire males and that disadvantage disappears. But if you really think about it, there is some question whether ‘group selection’ and ‘frequency dependent selection’ are really all that different in this context. Is it not the case that frequency dependent selection emerges from properties of the group and not the individual?
The biggest problem with this Level 1 data is that it represents a fairly indirect measure of fitness, the survival rate of males as expressed by life-span. This might not be so problematic if this were not a sexually-selected trait. But with a sexually-selected trait, one might find that life-span has very little to do with fitness. And if you argue that life-span is a good proxy for fitness, to some degree you remove the importance of female choice of mates. This conflict produces the most confusing elements of this paper, because in their explanation of the Level 1 results, the authors use explanations that are about attracting females, not living longer. It is too bad that a more direct measure of male fitness could not have been incorporated in this study.
This shortcoming disappears in the Level 2 comparison, because the fitness of local populations and metapopulations can be measured via the intrinsic growth rate, a much more direct proxy for reproductive fitness. When the authors compare the intrinsic growth rate of local populations to the intrinsic growth rate of their neighbor populations, they find some fascinating results:
As before, at the two levels considered selection is acting in opposite manners. But unlike before, this is the more simple ‘directional selection pushing in opposite directions’ scenario. From the perspective of the local population, it is better to have a larger average repertoire, presumably because females are attracted to patches with lots of males that can produce a variety of songs. But at the metapopulation level, it is bad to be around other patches occupied by males with a large repertoire, presumably because the females will be at least partially attracted towards these competing patches. While the paper does not gather data to test these presumptions, the pattern of selection is interesting even if its true explanation has not been uncovered.
I think that most readers will assess the results of this paper as entirely intuitive. If you can outcompete less fit males through the display of a vast song repertoire, it is good to be in possession of such virtuosity. But this virtuosity becomes a liability when you are paired with other similarly-fit males, as the competition is likely to sap both you and your competitors of vigor. But on the level of your entire patch, it is good to be with all these other virtuosic singers, because you will be able to attract more females. This is not, however, the end of the story, because your fitness also depends on the other patches around you: if they too are virtuosic, the fitness of your patch will be lowered due to competition. What is remarkable about this paper is that it has quantified selection at each of these levels. Perhaps our intuitive understanding of evolution in social groups already incorporates multilevel selection thinking, we just do not usually conceptualize what is going on in these systems in this manner; and more importantly, we almost never characterize the nature of selection at each level.
I am still a little confused about whether this study looks at three or four levels. On the one hand it is easy to see that it is a paper about individuals, populations, and metapopulations, but it is at the ‘middle level’ of population that I find myself confused. While both Level 1 and Level 2 comparisons involve populations of individuals, it is not entirely clear if these are the same, as in Level 1 the populations refer to local neighbor groups and in Level 2 they refer to groups each occupying a patch. These might be two different levels of social organization, which would help explain why the Level 1 populations and the Level 2 populations seemed to be under different selective regimes. All of this is made more confusing by the two different fitness measures used in Level 1 and Level 2: this difference may make comparing these two comparisons meaningless. This is a remarkable paper, but it is not perfect.
Imperfections acknowledged, I do not think that it is an exaggeration to say that this is a landmark study. It is the first study that I am aware of to explicitly partition the effects of selection on three levels in a natural system, showing that the evolutionary process is a compromise between mechanistic processes occurring at (at least) three different levels. Interestingly, this application of multilevel selection theory is not really about cooperation per se, but simply about the influence of individual versus group dynamics. I guess you could think of individuals with moderate-sized repertoires as ‘altruistically avoiding the adverse effects of competition’, but this seems like a bit of a stretch to me. What we really are seeing are the limits of competition: some competition is obviously going to be present in any system because of the individual-level benefits of being highly competitive (in this case highly robust as expressed by life span), but too much competition creates in this case a chaos that harms all in the most competitive groups. What higher-level selection is doing is effectively suppressing competition by opposing individual-level selection for larger repertoires.
It is interesting to me that this dynamic emerges from a system in which the focal trait is sexually selected. One can easily see where a sexually-selected trait might display a very large range of feasible values, because while the trait is going to be costly (which makes not expressing it a possible advantage) there is also the possibility that female choice will push the trait to extreme values (through Fisherian “run away” process). This variation of possible trait values makes this sexually-selected trait ripe for multilevel selection, because just as processes at one level can push it to one extreme, processes at the other can push it back towards the other extreme. Is this sort of multilevel selection more likely for sexually-selected traits? Maybe, but I do not think that this finding is idiosyncratic to sexually-selected traits. In fact, a paper considering only two levels of predator fitness (individual and group) suggests that a similar dynamic may be present in felid carnivores (Fryxell et al. 2007). What makes this study of the Dupont Lark so amazing is how well it unravels the role of selection at each level via direct empirical observations. I would love to see this same approach applied to social carnivores, who might also show the same observable patterns of opposite selection regimes operating at different levels of organization.
Another wrinkle here has to do with the limitations on sexual selection. If females in a particular group prefer males with larger song repertoires, one would question what prevents males from evolving to produce larger and larger repertoires. Another way of asking this question is to ask ‘what are the limits to run-away selection’? Of course biophysical limits can always be invoked to answer this question, as can tradeoffs with other selective pressures: there are only so many songs that a bird brain (even one under strong selective pressure) can learn, and dedicating so much effort towards singing might be to the detriment of other essential traits such as finding food. This study introduces a new sort of limit, one created by the social impact of individual behaviors. This is a kind of tradeoff, but different than the trait tradeoffs we usually associate with putting the brakes on selection for sexually-exaggerated traits. By singing too many different songs and eliciting similar responses from his competitors, a virtuosic bird lowers his own fitness.
You can bet that this paper has a huge target on its back, as critics will be excited to take shots at the assumptions, methodologies, and interpretations of this paper and maintain the evolutionary orthodoxy that ‘group selection is possible in theory but practically impossible’. I predict that many will say that this is just frequency-dependent selection operating on multiple levels, which might be a productive criticism if it spurs people to consider whether what we have called frequency-dependent selection is really just equivalent to selection at a higher level. There are some shortcomings here, even in the context of the overall-ingenious nature of the paper. Chief among these is that reproductive fitness is not directly inferred at the individual level. I suspect that some will quibble with the fact that the reproductive success of individual males was not tallied in parallel with the population and metapopulation intrinsic rates of increase. Still, it is hard to imagine that a more detailed study at the individual level would not find that males with more (and to some degree less) diverse repertoires sired the most offspring. After all, this is what many studies of sexual selection performed solely at the individual level have found; perhaps we should fault these studies for failure to consider all possible levels of selection.
I hope that more studies take the lead from this important paper and investigate selection above the level of the individual. It is pretty ridiculous to say that we never observe selection above the level of the individual if in fact we rarely look for selection above the level of the individual. This was an impressively-large-scale study, and the fact that data collected from 2004-2008 is only now being published tells you something about the analytical challenges these authors must have faced. That, or perhaps it is really hard to get a paper published if it reports data sympathetic to multilevel selection. It is really nice to see PLoS One publishing work of this quality and value. Expect more papers like this as empirical biologists are progressively inspired by the growing body of theoretical work that shows that selection is not just about individual fitness.