Monday, February 26, 2007

Of Caterpillars and Compounds...

The theme du jour is the influence of plant defenses (particularly secondary compounds) on caterpillar growth and anti-predator defense in tropical forests (primarily at BCI and La Selva). Over the weekend, I read a proposal seeking to document the influences of mixtures of similar plant compounds in anti-herbivore (specifically, caterpillar) defense. This study focused primarily on the intraplant chemical level, looking at which compounds or compound mixtures best deterred caterpillars. I followed this up by reading an article for Journal Club today focusing on the role of plant quality (supposably including secondary compounds, though these weren't investigated directly - more on this later) in influencing caterpillar growth rates and anti-predator defenses. And I myself have focused more on the next trophic level up - namely, the role of diverse arthropod defenses (which are presumably due to some combination of bottom-up - i.e., plant-driven - and top-down - i.e., predator-driven) on limiting food availability to insectivorous birds. I found this to be interesting and thought-provoking reading - for all the time I've spent thinking about the relationship between caterpillars and birds in terms of top-down selection for anti-predator defenses and bottom-up food limitation acting to reduce metabolic and foraging rates and population densities of the birds, I've spent very little time thinking about other - namely, plant-driven - selective forces driving caterpillar defensive trait evolution. This adds a whole new level of questions to my research. Funny how ecological research feels a bit like jumping down the rabbit hole sometimes - the more you learn, the more new worlds of ideas and questions open up.

The other interesting aspect of reading these two papers back-to-back was the contrasts in methodological techniques and quality. The proposal I read was directly testing the roles of secondary compounds on herbivory, by synthesizing the compounds and adding different concentrations and mixtures to leaves which are then offered to caterpillars, allowing direct testing of causation. The second paper, however, primarily investigated the role of nutrient quality on caterpillar growth, discovered that nitrogen and water did not fully explain the regressions, and from there made a leap to saying that the residuals are likely therefore explained by secondary compounds. They did throw in a few other related experiments - e.g., exposing caterpillars fed on different leaves to ant predators - but it seemed they were confusing correlation for causation. Another problem I had with the methods of this paper is that they arbitrarily broke continuous characters (e.g., leaf-expansion rate, number of spines on a caterpillar) into dichotomous categories, and worse, did not explain the justification for the divisions. Dividing the data into discrete classes allowed them to use ANOVA and t-tests which produced significant differences between groups, but I can't help but wonder if that isn't partially just a construct, a result of their categorizations rather than reflecting biological reality. I wonder if keeping the data continuous and using single and/or multiple regressions wouldn't have better reflected reality - and if the significant results they found would still appear this way. As scientists we're taught to strive for that p<0.05 value, but I wonder if we don't sometimes rearrange and reclassify our data in ways not completely reflecting the complexity of nature in order to reach that holy grail...

1 comment:

Rebecca Hazen said...

Right on, sister! Great job explaining the Coley et al paper. I posted a quick blurb about the ANOVA analysis, but I really like what you say here about the possible misrepresentation due to dichotomous categorization.

As we discussed in Journal Club, as research scientists, it's easy to understand that at a certain point you just have to say enough is enough, or else you'll never get any research done. However, at the same time, our efforts may turn out to be counter-productive if we are arbitrarily "binning" our data just so we can publish.

Thanks for the well-spoken insight!