The results support a general evolutionary theory of late life
Most evolutionary theories suggest a rapid rise in age-specific fecundity at early ages followed by a long decline after some peak value. Our interpretation of the evolutionary theory of late life, based on the decline in the force of natural selection, was that population fecundity will plateau at very late ages, like age-specific mortality rates (Rauser et al., 2003). We made this prediction because the force of natural selection acting on age-specific fecundity asymptotically falls to such a low level that it can no longer distinguish fitness differences in fecundity at different ages. Our experimental work supports this interpretation. We found that the rate of the decline in fecundity greatly slows at late ages, or plateaus, in 10 independent populations at some number of eggs laid per day greater than zero. Although some populations show a more defined plateau in fecundity than others, the two-stage model, with a second-stage plateau, fit the data in all 10 populations. The data do not have to fit the model, but they do. Note that if there was not a slowing in the decline in fecundity at late ages, the model would not have fit, or only the first stage of the model would have fit the data (the way the model-fitting algorithm was written, a failure to fit was allowed). This, however, was never the case. It is also important to note that each of the five populations from a selection regime are replicates and that we expect to see subtle differences between these replicate populations because of differential mutation and drift effects and because each pair-wise comparison was performed at different times. Thus, not all replicates will seem, on visual inspection, to fit a plateau model; this is sampling variation. Furthermore, the law of large numbers guarantees that the right end of the plateau will show ‘disintegration’, with a tendency to values much above or below the inferred plateau. Lastly, we do not necessarily expect to observe an exact plateau for each of these 10 populations because they are not that old (approximately 150 and 360 generations old for the CO and ACO populations, respectively) and thus are not fully converged. This line of reasoning suggests that the CO and the newly derived NRCO populations should have a less defined late-life plateau compared to the ACO populations, which have been maintained for many more generations. Upon visual inspection, this is exactly what we observe (Figs 3 and 6). Rose et al. (2002) observed a similar result for late-life mortality-rate plateaus. That is, the B populations, which had been maintained for 450 generations, showed the most well defined late-life mortality-rate plateau compared to the O, CO and ACO populations.
In addition to our experimental work, we have preliminary computer simulations of populations evolving with recurrent mutations that explore the evolution of late-life age-specific fecundity. These simulations also generate plateaus in fecundity at ages after the point when the force of natural selection falls to zero.
Previously, many studies fit mathematical models to fecundity data post hoc, using statistical goodness-of-fit criteria to determine the best function (e.g. Hoem et al., 1981; Gage, 2001; Müller et al., 2001; Novoseltsev et al., 2002, 2003; Carey, 2003), with no deep theoretical motivation for any of the functions tested (note, however, that these studies are fitting models to individual fecundity data, rather than population fecundity data). Furthermore, these studies did not observe late-life fecundity plateaus because they fitted their models to data sets that started with far fewer females than our own (n = 3200 per cohort), greatly reducing the likelihood of females surviving to ages late enough to observe the occurrence of late-life plateaus in population fecundity.
In this study, we not only demonstrate that population fecundity plateaus at late ages, but we show that these plateaus evolve according to the age of last survival in these populations’ evolutionary history. Together, these results corroborate the basic evolutionary theory of late life and its prediction that fecundity, just like mortality-rates, should plateau some time after the force of natural selection acting on fecundity plateaus. Our experiments could have refuted the basic evolutionary theory of late life if there had been no difference between populations in the age at which their late-life fecundity plateaus commence, after long maintenance with very different last ages of reproduction. The theory could have also been refuted if the difference between these break days had been in the opposite direction from the difference in the day of last reproduction. However, neither of these outcomes occurred.
As for the viability of the eggs laid throughout life, including late life, we found that viability declines over all parental ages, regardless of selection regime. This result is different from what Kern et al. (2001) found in similarly selected populations. They found that offspring viability did not decline in some of the later reproducing populations. This discrepancy may have arisen because they did not include very late age viability in their analysis. However, like the results of Kern et al. (2001), we found that viability declines more rapidly in populations selected for earlier reproduction relative to those selected for later reproduction. Although they concluded that offspring viability is a general feature of senescence, we found that it does not follow the same pattern of senescence as mortality and fecundity. That is, viability does not deteriorate so rapidly, nor plateau at late ages.
We also separated mid-life from late-life parental ages and found that viability did not decline during mid-life in the populations selected for earlier reproduction, but that it did decline in the populations selected for later reproduction. The opposite result was true for parental ages that occurred after the onset of late-life fecundity plateaus. Our results indicate that aging is reflected in offspring quality, regardless of selection regime, when very late ages are also considered. The contrast we observed between mid- and late-life parental ages for viability (changing in opposite directions for the two selection regimes) suggests that there may be a trade-off in age-specific fitness characters. Offspring viability declines in mid-life, but stops declining late in life in those populations selected for later reproduction. Note, however, that when performing the analysis for viability we pooled the data from the replicated populations within each selection regime. Therefore, any random effects associated with each replicated population were not taken into account.
Lastly, we tested antagonistic pleiotropy as a genetic mechanism affecting late-life fecundity. With antagonistic pleiotropy between early and late ages, some of the alleles that enhance early reproduction will depress later survival or fecundity (Williams, 1957,1966; Rose, 1985; Charlesworth, 1994). Natural selection on early reproduction will therefore tend to increase mortality rates and decrease fecundity later in life. We found that late-life fecundity was significantly responsive to selection for early reproduction imposed for a small number of generations. That is, antagonistic pleiotropy between early and late ages resulted in an earlier decline in the force of natural selection acting on fecundity in the NRCO populations, which caused an earlier decrease in fecundity before the start of the plateau. These results are also consistent with our general prediction that the last age of survival and the start of late-life fecundity plateaus should be positively connected.
Random genetic drift and mutation accumulation could not have a significant effect in the 24 generations of selection for earlier reproduction. Drift fixes mutations at a rate of 4Ne generations, or 4000 generations in this case, as these populations have been maintained at an effective population size of at least 1000 individuals. Similarly, the impact of mutation accumulation on the differentiation between the CO and NRCO populations, with 24 generations of accumulation of deleterious mutations, will not exceed about 0.1% of the break day differentiation of the ACO and CO stocks, adapting the mutation-accumulation calculations of Passananti et al. (2004). This small an effect would be undetectable in experiments of our size. Although our experimental results implicate antagonistic pleiotropy in the evolution of late life, it is important to note that the two genetic mechanisms of antagonistic pleiotropy and mutation accumulation are not mutually exclusive and that a positive result for antagonistic pleiotropy does not necessarily mean that mutation accumulation is not involved in the evolution of late life.
In our antagonistic pleiotropy experiment, we measured fecundity again in four of the five CO populations and determined the break day to be 56.16 days (Table 4), which is later than the break day determined in the ACO-CO pair-wise comparison (49.86 days, Table 2). This discrepancy is most likely due to environmental effects arising because these comparisons were performed at different times. Note that pair-wise comparisons were performed for both sets of experiments because evaluating relative differences in ages of plateau onset is the only way to properly control for possible environmental effects that may arise from performing replicate assays at different times.
Further implications for late life
Late-life fecundity plateaus are yet another surprising feature of late life. Although these fecundity plateaus are at a low number of eggs per female, they are significantly greater than zero. Thus they are analogous to late-life mortality-rates that plateau below 100%. This finding potentially has profound implications for our understanding of pleiotropy and selection in evolution. Perhaps there are some alleles that generally foster survival and fecundity, at both early ages, when the force of natural selection is great and at later ages, when the force of natural selection is weak. That is, some alleles may not have an age-specific effect, but may instead have an effect at all ages (Charlesworth, 2001).
Although the evolutionary theory of late life readily explains both the occurrence and evolution of late-life fecundity plateaus, this study does not attempt to test the influence of individual female heterogeneity in fecundity on the occurrence of these plateaus. It is conceivable that individual female fecundity does not plateau, but that the plateaus we observe are aggregate population characteristics. For example, it is possible that high egg-layers die early, leaving only the lifetime low egg-laying females alive at late enough ages to contribute to the population fecundity plateau we observe. This heterogeneity in fecundity idea is analogous to the demographic heterogeneity theories that have been proposed to explain late-life mortality-rate plateaus (e.g. Vaupel et al., 1998). However, unlike the case of mortality-rates, the amount of heterogeneity in individual fecundity patterns can be measured directly. We have data on individual female fecundity trajectories from large out-bred cohorts, not presented here, which shows that heterogeneity in fecundity is not sufficient to cause plateaus in fecundity at late ages in our populations. The present study was designed to test the predictions made by the evolutionary theory, and so does not test this alternative hypothesis.
Other non-evolutionary theories that have been proposed to explain late-life mortality-rate plateaus do not naturally lend themselves as explanations of late-life fecundity plateaus. One such theory is the reliability theory of Gavrilov & Gavrilova (2001). Under this theory, death occurs when the first of several essential physiological blocks fails. Each of these blocks contains redundant systems that can be characterized by the failure rate of their components. The structure of these systems can lead to both exponential rates of increase of mortality rates at young ages and mortality-rate plateaus at advanced ages. The age of onset of these mortality rates is a function of the failure rate of the component systems. Reliability theory explanations do not negate the influence of natural selection on the age of onset of a plateau through its influence on the rate that physiological systems fail. These failure rates could certainly be set by evolutionary forces, like those described by Mueller & Rose (1996). However, it is also clear that the physiological processes that determine the numerical age-specific decline of eggs produced by females would not naturally follow processes described by reliability theory without resorting to a number of ad hoc assumptions.
The ability of evolutionary theories based on the force of natural selection to explain the existence and experimental manipulation of both mortality-rate (Rose et al., 2002) and fecundity plateaus adds to their plausibility as causal explanations of these phenomena.