Individuals in many mammalian species devote a fraction of their lives to reproduction. Females in these species experience a gradual midlife decline in reproductive performance with age, and eventually menopause, the abrupt termination of reproduction 1. In contrast, males are typically able to reproduce until old age, with only a slight decrease in reproductive ability. Although anthropologists have suggested that menopause is unique to primates 2 or even just humans 34, the phenomenon appears to be widespread in mammals – in a meta-analysis of 42 species, Cohen (2004) found support for post-reproductive lifespans (PRLS) in 83% of taxa. More recent studies have shown that there are a range of species with relatively little or no parental investment that also exhibit reproductive senescence and cessation, including species generally considered to be 'r-selected'. These species include guppies 5, parakeets 6, mice 7, and beetles 8.

Although reproductive senescence or menopause may be relatively common, only a handful of species have extremely long post-reproductive lifespans. Why do females of some species live so long after reproduction ceases? On the surface, this would appear to be a maladaptive life-history trait. Several hypotheses, adaptive and non-adaptive, have been proposed to explain the evolutionary significance of menopause.

Longitudinal sightings data were collected from two neighboring populations of fish-eating killer whales over the last 30 years, 1978–2007 233738: the Northern and Southern Resident killer whales which inhabit the inland and nearshore waters of Washington state (USA) and British Columbia (Canada). These populations are discrete; they do not interbreed, and neither immigration nor emigration have been observed 24. During annual photographic surveys, nearly every individual in the population has been recorded. Each individual has unique pigmentation, scars, and fin shapes, allowing us to track the survival and reproductive performance of each female over time. Although detailed age and birth data do exist for recent years, information on birth defects, still births, or mortality risk to pregnant females is not available. Due to low adult mortality 22, the majority of females in our study are expected to live beyond the onset of menopause.

The age structure of females from the northern and southern populations was calculated to illustrate that over the last 30 years, each population has fluctuated in size, but more importantly, the age structure of both populations have changed (Fig. 2). In the smaller Southern Resident population, recruitment of young females has generally declined, while the proportion of older animals has remained relatively constant. The larger Northern Resident population appears to have increased in most years since the 1970s, and has shown an increase in the youngest component of the population (Fig. 2). The proportion of post-reproductive females has also fluctuated through time (Fig. 2), and it is unclear what role these animals may have in maintaining population structure.

To evaluate support for the attentive mother hypothesis, we examined whether covariate attributes of mothers (age, dead/alive, birth order) had any impact on the survival of calves (ages 0 to 1) or on juvenile survival (< age 5) for individuals born since 1978. Binomial generalized linear models (GLMs) with a logit link were used to model survival (φ) as a function of covariates, e.g. log⁡(ϕ1−ϕ)=B0+B1⋅agemom MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGagiiBaWMaei4Ba8Maei4zaCwcfa4aaeWaaeaadaWcaaqaaiabew9aMbqaaiabigdaXiabgkHiTiabew9aMbaaaiaawIcacaGLPaaakiabg2da9iabdkeacnaaBaaaleaacqaIWaamaeqaaOGaey4kaSIaemOqai0aaSbaaSqaaiabigdaXaqabaGccqGHflY1cqWGHbqycqWGNbWzcqWGLbqzdaWgaaWcbaGaemyBa0Maem4Ba8MaemyBa0gabeaaaaa@4875@. The Schwarz criterion or BIC 39 was calculated for each competing model, and differences between BIC values were used as an approximation to the Bayes factor 40. If there are two competing models, the approximate Bayes factor supporting model 1 over model 2 is BF₁₂ = exp [-(BIC₁ - BIC₂)/2]. When two models are considered, this formula allows the posterior probability of each model given the data to be computed, Pr [M₁|Y] = BF₁₂/(1+BF₁₂), because the Bayes factor can also be viewed as the ratio of the posterior model probabilities.

To examine support for the grandmother hypothesis, we first looked at impacts of grandmothers on their daughters, and second evaluated potential impacts of grandmothers on grandoffspring survival. The reproductive lifespan of mature females was modeled using GLMs to evaluate support for including mothers' survival status as a covariate. Because the precise onset of reproductive termination cannot be determined, we defined the response variable (reproductive lifespan) as the number of years between the last and first birth. Only females whose mothers' survival status did not change over their reproductive histories were included in this calculation. An alternative effect grandmothers may have is that by providing additional care, daughters may be more productive (shortening the time between births). To examine support for this hypothesis, we constructed a model of interbirth intervals (IBI) as the response variable. The survival status of each females' mother (alive/dead) was included as a covariate to evaluate whether females with alive mothers had a shorter IBI compared to females with dead mothers. As the second component of evaluating the grandmother hypothesis, we expanded GLMs of survival to include the grandmothers' survival status as a covariate – in these models, we compared the age-specific survival rates of juveniles with living grandmothers to those with dead grandmothers. Two analyses were performed; in the first case we examined the impact of grandmothers on survival of calves (aged 0 to 1), in the second we examined this factor across juveniles (aged < 5).

One hypothesis for why survival of primiparous calves may be relatively low is that older mothers may be more experienced than younger mothers 4142. An alternative hypothesis is that there may be an evolutionary tradeoff between current survival and future reproduction – young females may prioritize their own survival above that of potential offspring, delaying reproduction 36. To evaluate evidence supporting this hypothesis, we first had to quantify lifetime reproductive success (LRS) for each female. Several different approaches have been used to measure LRS. Lahdenpera et al. (2004) used the number of births as a response variable, however using the raw number of births has been criticized because it doesn't account for survival to maturity 14. A second approach is to calculate the individual growth rate, λ 43 – this method has only been applied to short-lived species and cannot be applied to right-censored data, such as that in our study. A third approach involves estimating total recruits to the population, R₀ 44. This latter method may be more robust than calculating the individual λ because it is considered rate-insensitive 45.

The total number of recruits could not be calculated for every animal in our study because many females either have partially observed reproductive histories or have not yet reached menopause. Instead, we developed a proxy for recruits, defining a recruit to be an offspring that lives to maturity (age 10) 22. For each post-reproductive female whose entire reproductive history is known, we calculated the number of recruits, which was treated as a fixed constant (Fig. 3a). For females whose reproductive histories were incomplete, we performed Monte Carlo simulations to generate hypothetical distributions of future recruitment until the onset menopause. The reproductive performance of these individuals in the beginning of their reproductive lives may help inform potential tradeoffs between age at maturity and reproductive success; specifically, mothers that delay reproduction may produce more recruits. Three types of uncertainty were included in these simulations of future recruitment for these individuals (aged greater than 25): uncertainty in the future survival of the mother, uncertainty in the future reproductive performance of the mother, and uncertainty in the future survival of newborns to reproductive maturity. Estimates of natality from previous work were used as age-specific probabilities of producing calves 26. Survival estimates of juveniles from the best model in our analysis (constant survival) were used with previously published estimates of adult survival for projecting individuals forward from birth 41. Given the recruits per female at each iteration of the simulation, we used the known age at maturity for each female (age at first birth) to estimate the potential effect on the number of recruits. Recruits were modeled using Poisson GLMs, with age at maturity as a covariate: R₀~Poisson(u), log(u) = B₀ + B₁ · age_mat.

Hamilton (1966) first posited the idea of the "wall of death" – the idea that once females cease reproduction, mortality should increase rapidly as there is not selection pressure for continued survival 19. We know that in human populations, there is little support for the "wall of death" hypothesis, even in hunter-gather populations where one cannot make the argument that survival past reproduction is due to artificially extended lifespan 46. Until recently, it has been argued that menopause is specific to primates and that the "wall of death" still applies to non-primates. However, biologists have shown that menopause is a much more prevalent life-history trait in non-primates than was previously supposed 1229. As we show in this study, killer whales show even less support for the "wall of death" idea than human hunter-gatherer populations 46. In these hunter-gatherer populations, approximately 30% of human females lived past reproduction; in contrast, more than 50% of killer whale females are expected to live beyond the onset of menopause (Fig. 1) 2226.

Although there is some debate in the literature, we follow Cohen 29 and lump the attentive mother hypothesis and the helpful grandmother hypothesis into adaptive hypotheses for menopause. One explanation for the long post-reproductive lifespans that have been observed in killer whales is the presence (and potential helpfulness) of older matriarchal females within matrilineal units 33. Using multiple modeling approaches, we found limited support for either of these adaptive explanations of menopause.

If the attentive mother hypothesis alone was responsible for the cessation of reproduction in killer whales, we would expect females to live just long enough to raise their last offspring to independence (2–3 years, to age 42–43). On average, female killer whales appear to live an average of 7–8 years longer than this (> 50 years, 22) – roughly equivalent to the time needed for offspring to reach reproductive maturity. There is some uncertainty in the estimated life expectancy beyond menopause, however, due to the small sample size used and uncertainty in some ages of the oldest individuals 22. For the attentive mother hypothesis to be responsible for menopause in killer whales, several biological conditions must be met: the probability of older females giving birth must decline with age, offspring born prior to the cessation of reproduction must experience some benefit, and the rates of birth defects, stillbirths and maternal mortality may increase with maternal age 47. While the first two conditions appear to be met for this species (26 and this study, respectively), data on stillbirths, birth defects, and maternal mortality do not exist. While it is unlikely that this data will be collected on killer whales or other free-ranging marine mammals, this question may be best explored in future research on terrestrial mammals with relatively long post-reproductive lifespans.

Although not directly related to the attentive mother hypothesis, we found that the oldest mothers may also be the best mothers – calves born to females approaching menopause had higher estimated survival rates than calves of younger mothers. These older females may be more successful in raising young because of maternal experience, or they may allocate more effort to reproduction relative to younger females. This same result did not hold when older females were used to model juvenile survival, suggesting that older mothers may be more experienced in caring for calves during their most dependent period (ages 0–1). After weaning, juveniles may depend on additional benefits from related individuals – fathers provide little to no parental care in this species, but older siblings or related adult females may contribute in raising young.

If the grandmother hypothesis was responsible for the cessation of reproduction in killer whales, either daughters or grandoffspring would be expected to benefit from the presence of post-reproductive females. From the perspective of the grandmother's fitness, the difference between these responses amounts to a tradeoff between the quantity and quality of grandoffspring produced. Post-reproductive females did not appear to have an effect on the number of calves produced by their daughters (either a shorter interbirth interval, longer reproductive lifespan or increased calving probability). Post-reproductive females also did not appear to benefit newborn grandoffspring, but following other studies, one of the age groups that may receive the largest benefit are newly weaned animals. In human studies, grandmothers have been shown to be one of the most common forms of kin selection 48. An analysis of human menopause in 20^th century Gambia suggested that the benefit of grandmothers may occur during an extremely small temporal window, immediately after grandchildren are not completely dependent on their mothers 49. The most compelling evidence supporting a positive benefit of killer whale grandmothers is the survival of grandoffspring between their 2^nd and 3^rd birthdays (Fig. 4). While including the survival status of grandmothers is not supported across all age classes, the posterior probability of the grandmother effect increases when only 2 year olds are included, but is still approximately equal to 50:50 support (Pr [M|Y] = 0.48 versus 0.05).

In a review of current evidence for post-reproductive lifespans in multiple taxa, Cohen's 29 non-adaptive hypothesis for the evolution of menopause suggests that there may be two forms of selection acting separately on reproductive and somatic lifespans. Cohen's hypothesis may explain the phenomenon of long post-reproductive lifespans in killer whales, but cannot be evaluated with existing data. To test this hypothesis using the longitudinal data in our study, we would need to examine whether total lifespans are uncorrelated with the length of reproductive lifespans 29. Due to the extreme longevity of this species, females that were reaching reproductive maturity at the beginning of this study are currently reaching menopause, and are expected to live for another 15 years.

In summary, like other studies of long-lived mammals 1229, we found weak support for both of the adaptive hypotheses for menopause. Perhaps for anthropogenic reasons, the attentive mother and helpful grandmother hypotheses seem natural explanations for the presence of old females in social species – especially species that form multi-generation matriarchal groups. Aside from weak evidence in human populations, multiple studies have failed to find strong support for either of these adaptive hypotheses in other mammals. This does not mean that the hypothesis of "social transfer" 33 (the cultural transmission of knowledge and skills) or coalition formation 50 does not play a part in the evolution of menopause or the presence of old females in social groups. Rather, it suggests that if social transfers are playing a role, it is through more subtle effects than simply increasing the survival of daughters and granddaughters.

The authors declare that they have no competing interests.

EW performed the analyses and took the lead in writing the manuscript; KP and EH assisted in writing and revising the manuscript, as well as in developing the framework for the analysis; KB provided data and assisted with the initial editing and manuscript preparation; JF provided data and expertise in killer whale ecology, and detailed knowledge of problems in working with these datasets.