Insight into the association between ethnic diversity and the characteristics of a woman’s menopause transition may provide evidence for the timing of the appearance of menopause in human evolution. In this study, we evaluated perimenopause and menopause age, perimenopause duration, and the associated reproductive hormonal changes in a longitudinal, multi-ethnic sample from the cross-sectional study SWAN . While the results of this study revealed lack of a consistent ethnicity-specific difference in the timing of the menopause transition, there were significant changes in the associated hormones which, together with variations within populations in the menopause “window” and population-specific variation in menopausal symptoms, are suggestive of menopause as a recently evolved and still evolving trait in human evolution.
Before discussing the significance of these results, it is important to state the rationale for this study. From the women’s health point of view, it is important to know if menopause is still evolving and, if so, can it de-evolve, i.e. can it be delayed or even eliminated. To be able answer this question, we need to know the nature of variation in menopause. The SWAN study (and many others reviewed by Gold et al. ) looked into the effect of lifestyle factors on menopause. Controlling for sociodemographic, lifestyle, and health factors affecting the final menstrual period (FMP), Gold et al.  found that “racial/ethnic groups did not differ in age at the FMP. Higher educational level, prior oral contraceptive use, and higher weight at baseline, as well as being employed, not smoking, consuming alcohol, having less physical activity, and having better self-rated health over follow-up, were significantly associated with later age at the FMP”. These results suggest that menopause is a complex trait, but it is interesting to note that, in spite of the effect of socioeconomic factors, the onset-age “window” of menopause remains more or less stable from population to population (Fig 2). Our reduced data set was designed to increase the chances of discovering ethnic differences in the onset age of menopause; we found it to vary significantly in Hispanic population only. In the following, we review our findings and use them to propose a model of “shifting mate choice-shifting menopause” and predict that there is population variation in the onset age of perimenopause and that the shifting patterns of mate choice and marriage will lead to a shift, a delay in the onset age of menopause.
Ethnic variation in perimenopause age
At present, this is the first assessment of the age of perimenopause onset in a multi-ethnic sample. This study has identified a significantly earlier perimenopause onset in women of Hispanic descent compared to women of Caucasian, African-American, Chinese, and Japanese descent. Hispanics experience perimenopause approximately two years earlier, which suggests that changes in the reproductive systems are still taking place and that menopause has not stopped evolving separately in each population. While this result may be suspect because of small sample size and heterogeneity among samples originating from different Central American countries, as we show below, other studies have reported similar results.
Previous perimenopause literature is highly limited. Mentions of perimenopause age in other literature are either not quantified or they refer to the mean age of women in the perimenopause cohort, rather than onset age. Of the feasible literature, McKinlay et al.  has reported the inception of perimenopause to be 47.5 years, earlier than determined in our study (51.8 years). It should be noted that McKinlay et al.’s  perimenopause age was not specific to any ethnicity and was limited to women in Massachusetts. Sammel et al.  also examined the menopause transition across ethnicities, albeit limited to Caucasians and African-Americans. The specific age was not listed, but it was found that African-American women entered perimenopause earlier than Caucasians, a finding that was not observed in this study . Differences in the reported age in McKinlay et al.’s  study and the variation in African-American and Caucasian women perimenopause age of onset in Sammel et al.’s  study can be attributed to definition bias introduced in our study. Other studies define perimenopause onset using early perimenopause (menstrual irregularity). In contrast, we used late perimenopause (3 to 11 months of amenorrhea) due to a change in the questionnaire following the 1997-1999 follow-up (SWAN), which led to overestimation of perimenopause onset age. It is likely that a change in perimenopause definition will yield results similar to other studies, as our results are consistent with Sammel et al.’s  finding that entry into late perimenopause did not differ significantly between African-American and Caucasian women. Although it is probable that the definition of perimenopause influenced observed trends, it is likely that an ethnic variation in perimenopausal age exists. Perimenopause age in Indian women was reported to be 44.69 years, which is prominently earlier than any reported perimenopause age for US women as a whole and thus suggestive of differential biological timing of perimenopause present among populations . More studies with diverse population samples are needed.
Ethnic variation in menopause age
In agreement with wide variation in menopause age across women worldwide, heterogeneity in menopausal age among ethnicities was found only in Hispanics in our study. Hispanic women experienced menopause two years earlier than women of other ethnicities, which mimics the results found for perimenopause age. Again, this result may due to small sample size, but several studies have reported a two year difference in menopausal age between Hispanic and Caucasian women30-32. Age of menopause in Hispanic populations has been reported at 48.5 years for 15 countries across Latin America and 47.9 years in Mexican women, whereas Caucasian populations experience menopause at 50 years [33, 34]. Despite the higher menopausal age recorded in this study (51 vs. 53 years; Fig 2), which was attributed to differences in methodology, our findings support previous literature. Furthermore, we suggest this difference is unlikely to be the result of lifestyle factors such as smoking prevalence, which accelerates menopause by 1-2 years [27, 32, 35]. Previous analysis of SWAN participants have shown that smoking frequency was lower among Hispanics than Caucasians and African Americans, which was consistent with other studies [36, 37]. Hence, the earlier onset in Hispanic women may be due to a different biological clock for the timing of menopause onset compared to other ethnicities.
Studies of candidate gene polymorphisms affecting menopausal age have found that the rs16991615 SNP in MCM8 locus, a gene involved in the repair of double-stranded DNA breaks reduces age at menopause by 2.3 years for Hispanic women . The samers16991615 SNP exerts the opposite result in Caucasian women, increasing menopausal age by 0.82 years . The effect of an SNP may vary according to ethnicity, which suggests that menopause may have co-opted different genes in ancestral Hispanic populations.
Whether a difference in menopausal age between African-Americans and Caucasians exists is unclear. Bromberger et al.  proposed that African-American women experience menopause six months to a year earlier than Caucasian women. This seems plausible given that studies of African women, specifically from Ghana and Nigeria, determined that menopause occurred at 48-49 years, which is earlier than Caucasian women [30, 32, 34]. In contrast, several studies did not report differences in menopausal age between Caucasians and African-Americans [28, 30, 31]. Our findings support the latter, as both the mean and median menopausal age in both ethnicities was similar. It is possible that the small African American sample size (n=19) and high percentage of smokers could have contributed to Bromberger et al.’s  observations.
The similarity in menopausal age observed in both African-American and Caucasian women led us to expect similar evolution of menopause in their ancestral populations. Interestingly, however, many of the Caucasian associated SNPs studied in candidate gene polymorphisms associated with menopause age were poorly replicated or not replicable in African-Americans in GWAS studies, which suggest that African-Americans are not as similar to Caucasians as initially believed [39, 41]. Given the high linkage disequilibrium, African-Americans were suggested to have a variant of the SNP at the associated loci, although it was not identified [39, 41]. If variants exist in African-Americans, it is possible that the effect of the variant would be similar to the Caucasian SNP to result in the similar phenotype (i.e. menopause age) observed between these two ethnicities. This seems plausible since the SNP findings were able to be replicated in African Americans, such as the rs365132 SNP at the UIMC1 locus, with effects similar to Caucasians (i.e. increased menopausal age by ~0.4 years) . Further research is required to determine the variation between Caucasians and African-Americans, as SNPs associated with menopause onset in African-Americans have not been well-studied.
Menopausal age for Chinese women from China and Singapore and Japanese women from Japan was roughly 50 years in both populations, which resembles the Caucasian menopausal age [34, 41, 44]. Concurrent with our results, similarity in menopausal age between Caucasian, Chinese, and Japanese women have been reported in the literature [42, 45-47]. Several SNPs associated with menopausal age in Caucasians were replicated to the same effect and direction in Chinese women from Shanghai, suggesting that there are some shared menopausal evolutionary traits between these two ethnicities .
Recent studies have suggested that Japanese women entered menopause at later ages than Caucasian women [28, 30, 31, 37]. While the mean menopausal age did not differ significantly between Caucasian and Japanese women in our study, we did observe a higher percentage of Japanese women entering perimenopause and menopause later than other ethnicities, which supports past findings [28, 30, 31, 37]. If differences in menopause age exist between the two ethnicities, this may indicate differences in menopause affecting loci. One study reported several SNPs associated with menopause age in Caucasians could not be replicated in Japanese women in GWAS studies; however this was a very small study . Further research is needed to determine whether or not Caucasian SNPs can actually be replicated in Japanese women or whether other variants exist.
Ethnic variation in the duration length of perimenopause
This is also the first study reporting the duration of perimenopause among different ethnic groups. In spite of the heterogeneity in the timing of the menopause transition, our findings show that the length of perimenopause was consistent across all ethnicities (Fig 4). The duration of perimenopause appears negatively correlated with the age of menopause. This result presents the possibility that the intrinsic mechanisms underlying the progressive loss in ovarian function from the perimenopause to menopause stage is shared among individuals. Combined with the data for timing of the menopause transition, it is hypothesized that the timing of onset is what separates the menopause phenotype between ethnicities, as opposed to the duration. Future research can be directed to understanding why and how the timing of menopause has been shifted in different populations over the course of human evolution to result in the current ethnic variation in the onset of the menopause transition. One evolutionary theory explores the idea that women of ancestral populations with lower female biased dispersal would experience earlier menopause, although the required information about ancestral ecologies in these populations is lacking .
McKinlay et al.  determined the length of perimenopause to be 3.7 years, which was markedly higher than the 1.3 years reported in this study. Again, this can be partly attributed to the definition bias of perimenopause, which would have led to an underestimation of the perimenopause duration in this study (Fig 4). Alternatively, the methodological differences used to determine the duration was also a factor. For example, McKinlay et al.  determined duration by subtracting the mean age of menopause from perimenopause onset. This could lead to under/overestimation of the duration since it was not representative of the duration in individuals. There is individual variation in duration of perimenopause; hence, the duration should be determined by averaging the duration in each individual, which was done in this study. Future studies in perimenopause length are encouraged to average individual duration for greater accuracy of the duration. Additionally, further research is needed to determine if the similarity in duration in ethnic groups still exists when early perimenopause is used to define perimenopause onset, as opposed to late perimenopause.
Ethnic variation in hormone concentrations
Across all ethnicities, E2 concentrations declined over the menopause transition, as expected. As the menopause transition progresses, there is progressive loss of ovarian follicles, which produces E2, leading to overall reduced E2 levels . Consistent with studies conducted by Randolph et al.  and Kim et al. , mean E2 levels did not differ significantly across ethnicities. Similarities between E2 levels across ethnicities may be linked to earlier observations of the perimenopause duration. Klaiber et al.  noted that women with lower E2 levels had shorter menopause transition duration. Since all ethnicities were similar in E2 levels, the duration was not expected to differ between ethnicities, which are consistent with our results.
Since changes in E2 are correlated to changes in FSH, we did not expect differences in serum FSH across ethnicities either. However, we found a significantly lower FSH increase in Hispanics and African-Americans compared to other ethnicities during the menopause transition. These ethnic differences in FSH levels were thought to be linked to ethnic variation in pituitary-ovarian feedback during the menopause transition, such as variation in estrogen sensitivity [53-55]. Greater sensitivity to estrogen decline is associated with higher FSH levels, as estrogen is a negative regulator of FSH. Examination of polymorphisms in ESR1, which codes for an estrogen receptor, has shown that the rs2234693 polymorphism was significantly associated with Caucasian women menopause onset, although further research is needed for other ethnicities . In contrast, other studies have reported higher FSH levels in Hispanics and African-Americans, regardless of menopausal status [50, 53]. This was attributed to the lack of adjusted BMI hormonal values in our study and different hormonal collection times. Fundamentally, both our results and Randolph et al.’s [50, 53] have consistently noted an ethnic difference in FSH levels.
Similar to FSH levels, there was an ethnic difference in the increase of T levels across the menopause transition. Premenopausal T concentrations did not vary among ethnicities, but in menopausal women, T was significantly higher in Hispanic and African-American women compared to other ethnicities, suggestive of ethnic differences in T metabolism over the menopause transition. Literature to support this observation is scarce because T assays have difficulties reporting the low T concentrations in women, making it challenging to study in menopausal women . A study by Luckey et al.’s  provides a possible indicator of ethnic differences in T metabolism by examining rates of bone loss during the menopause transition, as lower T levels are suggested to contribute to greater bone loss. In Luckey et al.’s  study, African-American women were found to have lower rates of bone loss than white women, suggestive of higher T levels in African-Americans, which is consistent with our findings in this study. Other studies have found lower T levels for African-Americans and Hispanics compared to other ethnicities, which was attributed to BMI adjusted values [50, 51]. Overall, we can affirm that ethnic variation for T levels exists.
Several studies have correlated changes in SHBG to changes in T, which may suggest that ethnic differences in T could be attributed to ethnic differences in SHBG during the menopause transition [59, 60]. This was unlikely given that ethnic differences in SHBG in menopausal women and SHBG rate of increase over the menopause transition were not observed, which was consistent with findings from Burger et al. . Alternatively, studies have found associations between DHAS and T, which may also suggest ethnic differences in DHAS as a factor for ethnic differences in T [60, 61]. While ethnic differences were noted in DHAS levels in both premenopausal women and menopausal women, no ethnic differences were found in the rate of DHAS increase. Similar findings have also been reported in other studies . It thus seems that ethnic differences in DHAS changes are not specifically associated with the menopause transition, and is unlikely to be the cause of ethnic differences in T during the menopause transition, so other factors are likely responsible for the ethnic differences in T changes. One possibility involves androstenediol, which is secreted similarly to DHAS during the menopause transition . Its concentrations are thought to affect the balance of androgens over the menopause transition, hence research determining if ethnic differences exist for androstenediol, along with other possibilities for the ethnic differences in T, would be beneficial .
As mentioned before, there are several limitations to this study. The first is regarding the participants. The age of participants was limited to 42-55 years, which may have biased the perimenopause and menopause age to older ages. This may not be a significant limitation, given that, in a study done by Luborsky et al. , they reported ethnic variation in premature menopause (menopause age <40 years) consistent with our study. The exclusion of women with inconsistent bleeding patterns in this study could also have biased the results. Although not shown, we retested our findings with these exceptions added, but overall the findings remained the same. Secondly, late perimenopause was used to define perimenopause onset, leading to an overestimation of perimenopause age, underestimation of menopause duration, and irregularity in premenopausal hormone values. Additionally, use of BMI adjusted hormonal values could have altered our findings, as BMI has been shown to affect hormonal values significantly, positively for T and negatively for all other tested hormonal values . It is also possible that hormone therapy could have interfered with our findings; however, as explored by Kim et al. , hormone therapy does not deviate from the results comparing ethnicities; rather, it only reduces the differences between groups.
A “shifting mate choice-shifting menopause” model of female fertility
The main results regarding variation in age of menopause, from this study as well as others, can be summarized as follows. Firstly, all human populations, regardless of socioeconomic status, show a roughly 15-year window of menopause [45-60], which means that there is plenty of variation within populations. Secondly, many lifestyle factors are known to affect the age of menopause, but with the exception of premature menopause, the menopause window persists in all populations. Thirdly, there are no significant ethnic differences in the onset age of perimenopause, presumably because there has not been sufficient time since the origin of menopause to produce such differences. Furthermore, as we show below, the absence of variation among populations may indicate the role of similar mate choice mechanisms (i.e. male preference for younger mates) operating in all human populations. Fourth, there are significant changes in the levels of associated hormones within as well as between populations. Finally, different populations show ethnicity/race-specific menopause symptoms; for example, vasomotor symptoms were more common in Afro-American and Hispanic women than in other ethnicities . This suggests a role of both biological and cultural factors. We take these results to mean that menopause is a relatively recent trait, and most variation in the trait is present within populations with relatively little variation between populations. This is not surprising in view of what is known about human genetic variation in general .
We use these results to extend the mate choice theory of menopause , which holds that menopause is the result of men’s preference for younger mates, leading to the accumulation of deleterious mutations affecting fertility in women deprived of reproduction. We propose (1) that the present menopause window (roughly 45-60 years) is directly related to the “mate choice window” in the present-day human population (roughly 15-30 years); (2) that the lack of between-variation in perimenopause is the result of a similar mate choice system operating in all human populations; and (3) that changing patterns of mate choice/marriage will ultimately lead to a shift in the onset age of menopause. This shift will not occur immediately as the pressure from the shifting mate choice will only have its effect if women of premenopausal age will choose to engage in reproduction. Under the mate choice theory, menopause is a transient phase of male-imposed female fertility and it will be delayed with changing patterns of delayed mate choice/marriage and reproduction.
To allude to the clinical significance of this study I introduce Mayr’s concept of proximal and ultimate causes . Clinical practices are based on proximal causes of the disease and proximal or immediate cure or relief to the patient. Evolutionary studies, as the present one, relate to the ultimate causes and while they may not help the clinicians or the patients in the short term, they provide an evolutionary framework for understanding the disease and the variable symptoms and treatment response between patients in the long term. Menopause is not a disease; it is an evolutionary consequence of reproductive behavior imposed by mate choice practices in the long past. Our study shows that menopause is not an invariable trait and the results from this study will provide basis to help design personalized treatment. Integrated studies involving longitudinal menopausal studies linked with genomics and hormonal studies with diverse ethnic populations can provide valuable information bearing on women’s health and personalized medicine.