Science Drinks – Mar 25, 2014
Attendees for the latest Science Drinks session at the William Wallace included:
Lilly started the session by introducing a paper on allometry in cervids by Lemaître and colleagues recently published in Biology Letters. Allometry is the study of scaling relationships, and is typically quantified by measuring the slope of the line fit between a log-transformed “body size” index and log-transformed trait values. But a review by MacLeod (2010) pointed out several problems with standard assumptions by scientists studying allometry, including that the best-fit line is straight. Lemaître and colleagues show that the allometric relationship between (log) body mass and (log) antler length in cervids is curved, with the increase in antlers leveling off once mass reaches about 110 kg. they argue that this curvilinearity may be explained by energetic constraints on investment in antlers among the largest species., and caution that researchers should be mindful of possible nonlinearities when assessing scaling relationships.
Adam shared a different comparative study, this one focusing on the lengths of telomeres in chimps and humans and recently published in the American Journal of Human Biology. Telomeres are a section of repetitive DNA that caps the end of each chromosome, and that tend to shorten progressively as an animal ages because with every cell replication, a tiny bit of the telomere at the very end of the chromosome cannot be copied. The enzyme telomerase can restore lost telomere, but because this kind of repair mechanism is costly, the length of telomeres may reveal both individual-level information about biological age and susceptibility to degenerative diseases, as well as species-level investment in somatic repair that has evolved in conjunction with maximum longevity (see Monaghan 2010 for a great review of telomere dynamics in a life history context). Tackney and colleagues (2014) predicted that because the maximum longevity of humans exceeds that of chimps, chimps might have shorter telomeres or faster rates of telomeric loss than humans. The computed the relative signals of telomeric and single copy gene signals (T/S ratios) in white blood cells, and found that while their estimates of telomeric loss for chimps (approx. 73 nucleotides per year lost) exceeded that of humans (approximately 40 nucleotides per year), these rates were not statistically distinguishable. Moreover, the chimp telomeres were actually longer, having a T/S ratio around twice that of humans. It’s unclear what the implications of this finding are, but it raises many intriguing questions about the complex relationship between longevity and investment in repair that concerns all students of life history.
Toby showed some new figures, hot off the printer, illustrating some of his Hons research findings about the relationships between eye facet size divergence (which involves adaptations for photosensitivity) and the angles between facets (which may limit acuity) in dance flies. Claudia showed off some new plots and tables of coefficients of her own, involving the study of courtship and mating as a function of diet and social environment in crickets. I don’t want to step on imminent punch-lines: both Toby and Claudia will submit their dissertations in a few short weeks, and we’ll have a chance to see summaries of their work then. Stay tuned!
I (Luc) wrapped up the session by calling attention to a paper just published as an “early-view” manuscript on the journal Evolution‘s homepage by Walker (2014). This paper deals with a problem that is well-known but still underappreciated: when dealing with observation data (unmanipulated experimentally), coefficients of regression analyses will be biased upwards if any of the predictors are correlated with other causal factors that have been omitted from the model through ignorance or negligence. Walker uses simulations to show that this upward bias can be substantial, and argues that the problem is especially acute if one seeks to interpret coefficients as evidence for causal relationships. He further argues that the problem is not necessarily avoided by having prior knowledge of the “right traits” to include in one’s model, and carefully measuring candidate rival traits that one can include as covariates. He writes,
“forty percent of the time an additional covariate is added, even using omniscient prior knowledge of effect size, the error in the estimate of the effect is worse than if the confounder had been left unmeasured”.
This is deeply unsettling research for someone like me who spends a lot of time trying to find patterns of covariance in observational data (much of it by necessity since some of our work is on organisms that do little in the lab except die). Walker does point to directed acyclical graphs (structural equation models) as promising some clarity concerning causal relationships in observational data, which is satisfying news given my recent obsession with them, but overall the MS is more cautionary than prescriptive:
“Unless a functional model generates very specific predictions of effect magnitude, observational data is better used for testing model assumptions and not the mere presence of an effect. For studies where experimental manipulation or functional modeling is not an option, we may have to be content with simply not knowing the magnitude of effects very precisely.”