Group selection in Monkeyflowers?

I found a good very recent article that can be used to illustrate my point that we have a lot of data sets that show multilevel selection; however, because of the data sets were collected for a different purpose, they have never been used to look for multilevel selection.  The paper is on the effects of community composition on the pollination biology of Mimulus (Arceo-Gómez and Ashman 2014.  Am. Nat 183:E50 – E63).  First off, it is a good paper, and I have no concerns about their analysis at all.  Dr. Arceo-Gómez was kind enough to give me access to his data, for which I am quite grateful.  They collected data in several experiments, and only some of those data can be used for contextual analysis.  There is also the hidden joy of looking for group selection in one of Doug Schemske’s favorite plants – Doug was a starting assistant professor at the University of Chicago when I was a graduate student.  We never did see eye to eye on the matter of group selection . . .

In this study Arceo-Gómez and Ashman studied monkey flowers growing in isolated “seeps”.  Seeps are springs where the soil is moist and thus can support a variety of flowering plants.  They are surrounded by much drier soils that are primarily grassland.  In each seep they measured a number of traits, including the flower morphology of the Mimulus, and the number of co-flowering species in the seep.  The goal of this study was to discern whether the number of coflowering species affected the reproductive biology of the monkeyflowers, and as you might expect this is exactly what they found.

Mimulus guttatus

Mimulus Guttatus, the monkeyflower.  (from

For example, they observed a significantly lower pollinator visitation rate and a higher deposition of heterospecific pollen in sites with a high diversity of coflowering species.

pollinator visits by site diversity HP pollen by site diversity

Mean (+/- SE) for insect visitation rate (A), and proportion of heterospecific pollen (HP) on the stigmas (C) of Mimulus guttatus flowering at sites with high coflowering community diversity and low coflowering community diversity. Asterisks denote significance of preplanned contrasts between high and low diversity; one asterisk indicates P <.05 and two asterisks indicates P < .001.  (not shown:  (B) conspecific pollen (CP) receipt on stigma after 1 day of open pollination) (Arceo-Gómez and Ashman 2014.  Am. Nat 183:E50 – E63)


They also found that flower longevity was associated with the number of coflowering species.

flower longevity by sp richness

Correlation between Mimulus guttatus flower longevity assessed under common greenhouse conditions and the in situ number of coflowering species at the site (r = 0.52, P = .01, n = 23). (Arceo-Gómez and Ashman 2014.  Am. Nat 183:E50 – E63)

Some of their data cannot be used in a contextual analysis.  In particular, the pollinator visitation rate data is based on visitation to Mimulus in general, not to the specific individuals. Thus, we cannot do a regression of individual fitness on the trait.  If we look the effect of coflowering on flower longevity it appears that there is is “community selection”  occurring.  In this case it turns out that there is a strong relationship between the mean flower longevity, and species richness, I get the same values they get, but that there is enough variation at the individual level flowering times that the effect of species richness is not quite significant (p = 0.079).   Nevertheless, it argues that to the extent that longer flower life is a fitness component that there is “community selection” for increased coflowering.

Another rather iffy measure of “fitness” is amount of heterospecific pollen.  As shown above it is significantly affected by coflowering species richness.  Again we can do the regression and find that we come up with the same result.  It is actually interesting to throw in a community, a population and an individual trait.  For the community trait I chose species richness, for the population trait, mean flower size, and for the individual trait, flower size.  If we run that analysis with amount of heterospecific pollen as our fitness trait we get:

heterospecific pollen

Contextual analysis of heterospecific pollen load as a function of community, population and individual level traits.

You can decide whether the individual trait is significant, but the community and population traits clearly are significant.

My problem is that I don’t see why heterospecific pollen is a good “fitness trait”.  Thus, I think it makes more sense to use conspecific pollen load as our fitness trait.  If we do the same analysis using conspecific pollen load as the dependent variable we get:

conspecific pollen analysis

Contextual analysis of conspecific pollen load as a function of community, population and individual level traits.

In this case, it turns out that there is only one trait that is important, that is the mean size of flowers in the population.  The smaller the flowers, the more pollen gets deposited apparently.  This analysis seems to confirm the pattern we have seen before:  Group selection is very important in plants, and, in many cases, probably of considerably greater importance than individual selection.

I doubt that this is the best data set I could have used for this demonstration, and there are a number of things that would make me less than enthusiastic about publishing my “re-analysis”.  One is that I would want to understand the biology of monkeyflowers a lot better before I started calling flower longevity and heterospecific pollen loads as fitness traits.  I also picked flower size more or less randomly.  Again, understanding the biology would help a lot.

However, the big issue here is that (miss?) using contextual analysis it is possible make the claim of  “community selection”  any time a trait is correlated with a community level characteristic.  This is the problem of all phenotypic based selection analyses.  That is, there is no statement about heritability.  I am very doubtful that there is any sort of heritability for species richness in Mimulus, although there may be contextual heritability for flower size.  Thus, while we may indeed be able to show that plants in some types of communities have higher fitnesses than plants in other types of communities, as long as we are working strictly at the phenotypic level, it almost seems a matter of taste whether we call that community evolution or community ecology.  This raises a big question about the heritability of contextual traits.  Its not one that I have a nice clean answer for, but it is something I can at least discuss some of the issues.

Leave a Reply