How does external mortality affect aging rate?

A long-standing prediction for the effect of external mortality factors on aging rate was (I think) first made by Williams, in the context of his 'antagonistic pleiotropy' theory of aging. The reasoning is alluringly simple: organisms which tend to die young (for example, from high predation rates) also reproduce while they're young. Therefore, the protection against aging afforded by natural selection disappears more quickly, and so these organisms age faster. The prediction from this idea is simply that external mortality pressure and aging rate should be positively correlated.

It turns out not to be that simple, however. (Somewhat) recent data on aging rates in fish have found that, in at least some cases, the opposite occurs: fish which have been exposed to higher predation rates actually age more slowly. I just read a paper from Mitteldorf and Pepper (Theory Biosci. 2007, 126:3-8) which attempts to square this circle by arguing for a role for group selection in evolution. Their argument is that population-level events can contribute to individual survival chances, for example in epidemics, where a high population density can negatively impact an individual's chance of survival. Another example they give is starvation, which could be triggered by an excessively large population; aging may be an adaptation to keep populations in check.

As the authors note, however, group selection is not a generally-accepted mechanism of evolution. The physical mechanisms underlying evolution certainly act on the individual level. However, viewed in game-theory terms, I can imagine a group selection mechanism: suppose two genotypes exist, one of which confers maximal individual reproductive success, and the second which is reduced on an individual level but decreases the probability that a population-wide calamity will occur. On short time-scales, individual selection would certainly predominate, and the maximal individual reproductive success would be the driving factor. However, certainly some subpopulations will end up with the second option, just by random chance -- especially if it is a reasonably good fitness, even if not absolutely optimal. On a longer time-scale, the re-occurence of the aforementioned calamity would tend to indirectly select for the second (apparently less fit) genotype. So it seems like the important variables would be (1) the difference in fitness between the two genotypes, and (2) the rate at which population-wide catastrophes occur. I'll have to tinker with this a bit...might be possible to construct a neat little model for this.

Although, stepping back for a moment, I wonder if this argument is necessary -- could the observed effects in fish (and several other organisms) be the result of genetic drift affecting isolated subpopulations? Recent experimental evidence indicates that neutral evolution could be a much more important driving force than previously thought.

Another caveat arises from a short review by Bronikowski and Promislow (TRENDS Ecol. Evol., 2005, 20). Evidently the conclusions for the empirical fish data (from Reznick) were obtained by fitting an exponential curve to the mortality rate versus age. One issue is that there is a substantial amount of scatter on the plot. I should look up the confidence intervals from the original paper. Also I need to check how they did their fits; in B&P it looks suspiciously like a linear regression to a semi-log plot, although hopefully that isn't actually the case! But, given the amount of scatter in the data, I wonder if the conclusions from Reznick are solid. Another interesting point raised by B&P is that not all forms of mortality are equivalent, from an evolutionary standpoint -- it matters whether the extrinsic factor being examined is independent from previous occurrences. The example they give is that while the probability of getting eaten (or not) by a predator today is likely the same as it will be tomorrow, due to acquired immunity, the situation is quite different with regard to death by infection. So this adds another (potential) dimension to the analysis of mortality rates -- you may need separate curves, one for causes of death which are (approximately) condition-independent, and one for condition-dependent causes.

Downloaded a pile o' papers on this topic, and clearly, I'm going to have to plow into these a bit before I try and get my hands (too) dirty with trying to construct a model from scratch...