In this cross-sectional study we demonstrate that couples attending for infertility treatment exhibit strong correlations for a range of physical, social and behaviour characteristics and modest to weak correlations for a range of lipids and some other metabolic measures. The similarity in correlations for height, education and ethnicity, with those found in other publications not restricted to couples seeing fertility treatment (2–4, 7, 10), suggest that conventional assortative mating is similar in infertile couples as in the general population’ couples. That diet is the principal source for several of the metabolites; docosahexaenoic acid, histidine, phenylalanine and omega-3-fatty acids, would suggest that convergence due to a shared environment and active co-participation in daily activities including food consumption facilitates convergence of some metabolites. However, we acknowledge that for the vast majority of the metabolic measures assessed here correlations were weak and in this study we have no information for how long the couples have co-habited, beyond the minimum of two years.
In line with previous research, the present analyses provide evidence of assortative mating for age (30–32), height (3) and educational levels (2, 7), with strong evidence of endogamy with respect to self-declared ethnicity (11). Age is well established as showing the greatest level of couple similarity among all personal characteristics, with spousal age correlations typically ranging from 0.70 to 0.90 (11, 30–32). The reasons for our slightly lower estimate (0.61 95% CI 0.53, 0.69) are unclear but may reflect recruitment of participants with known fertility issues, as both maternal and paternal age are known to be independently inversely associated with fecundity, however, the median age gap was similar to that observed in the general population (33). The observed modest estimate for height is similar to previous meta-analyses, which have suggested moderate assortative pairing for height across human populations (r = 0.23, 95%CI 0.21, 0.23), and that the strength of this assortment appears to be relatively constant over time (3). For ethnicity although endogamy remains the norm in Scotland, it has declined over recent years with similar declines observed in other Western countries (11, 34). In contrast to this decline in ethnic endogamy, most studies in line with our own study, have indicated a sustained increase in educational homogamy (32), with moderate partner similarities for potential drivers for education including socioeconomic status, abilities and intelligence all documented (11). Despite previous meta-analyses suggesting weak associations (r = 0.10 to 0.15) for BMI, weight and related indices including waist circumference and waist to hip ratios (8), we did not observe any correlation. This is likely to reflect our unique population, as even when we restricted the analyses to couples where the male BMI was also ≤ 30kg/m2 to account for the female BMI treatment eligibility criteria we still did not observe a correlation.
The observed convergence of additional lifestyle factors like alcohol consumption (r = 0.62, 95%CI 0.50, 0.74), with a dominance of consumption of a low number of units compliant with national guidelines (≤ 14 units per week), may in part reflect that the population were drawn from an infertility clinic where healthy preconceptual lifestyle behaviours may be anticipated. Meta-analyses have previously suggested an overall moderate similarity for alcohol use (r = 0.36)(35), though levels of similarity observed in different studies have ranged from negligible to high. For exercise, studies have generally reported correlations between 0.15 and 0.30 (36–38) albeit some higher than 0.40 (39, 40). That our observed correlation of smoking status (r = 0.47 95%CI 0.36, 0.57), is marginally higher than previous estimates reported in a meta-analyses of smoking habits (r = 0.23, 95% CI: 0.12, 0.36) (8) may reflect our eligibility criteria, as in Scotland placement on the waiting list for public funding of fertility treatment is dependent on confirmation of non-smoking by cotinine breath testing for both partners.
Limited evidence from twin and family studies of multiple metabolites suggest that the heritability (h2; proportion of phenotypic variance due to genetic factors) of lipids and lipid-like molecules have a mean h2 levels of 47% (range from h2 = 0.11 to h2 0.66), while for organic acids and derivatives the mean is 0.41 (0.14–0.72), essential amino acids 0.42 (0.23–06.4) and non-essential amino acids 0.39 (0.22–0.69)(19–22, 41). As direct genetic variation in metabolites profiles would not produce a correlation between couples due to the invisible nature of both genes and metabolites, our observed correlations are likely to be due to through indirect pathways including assortative mating for social and behavioural characteristics. In a systematic review for coronary risk factors, significant but low (upper limit of 95% confidence interval, maximal 0.10) spousal correlations were identified for total and LDL cholesterol and total triglycerides (8). These meta-analysis estimates are very similar to those reported here; total cholesterol 0.07 vs 0.11, LDL cholesterol 0.06 vs 0.10 and triglycerides 0.08 vs 0.12, with the detailed NMR breakdown of the lipid subclasses and lipoproteins providing further similar estimates of spousal correlation for lipid metabolism. To date given limited prospective longitudinal studies, inference on whether assortative mating and/or cohabitation and thereby a shared environment underlie these associations has been achieved by using marriage duration as a surrogate for a common environment and potential convergence (8). These support initial indirect assortative mating (i.e. on social or behaviour factors that influence metabolism), and that a shared environment may further influence lipid metabolism but to a lesser degree (8).
We observed weak to moderate spousal correlations for a range of essential amino acids and omega-3-fatty acids including the subtype docosahexaenoic acid all of which have diet as their principal source (42, 43). The sharing of a common household larder and most main meals has been proposed as a mechanism by which couples have similarities in types of food, and nutrient intakes (44). Although gender asymmetry in the spousal adoption of health-related dietary changes has been reported, with female partners more likely to adopt male partners changes than vice versa (45), this may not apply to preconceptual diets where females may have a dominant role in preparation for pregnancy. Consistent with the suggestion that shared diet may have a critical influence, heritability variance estimates for circulating serum levels of histidine, docosohexaenoic acid, phenylalanine have been all lower than those observed for lipids, with environmental factors such as diet having a much greater contribution (41).
Glucose, the downstream glycolysis product of pyruvate and then the citric acid cycle carbon flux product of citrate, plus lactate and the glyceroneogenesis pathways were all weakly correlated. A meta-analyses of six studies, estimated that history of spousal diabetes was a risk factor for diabetes in their partner (effect estimate 1.26 (95% CI 1.08, to 1.45)(46). A data mining study of 5,643 couples and 5643 non-couple pairs similarly found strong associations of having diabetes within couples (5.2% both of the couple had diabetes) than non-couples (0.1%)(47). Heritability and shared environmental factors are proposed to account for at least half of the variability in normalised fasting glucose (48), however, our study is unable to delineate their respective contributions to the weak association observed here.
Our studies adds to the small number (3) of studies we could identify that have previously explored spousal metabolite correlations (19–21). It has a similar sample size to one of those previous studies(19) and examines a similar number of metabolic traits to two of them (19, 21). It adds to those previous studies which were examining correlations in twin studies, where one or both of a spousal pair had been recruited based on being a twin. We do however acknowledge several limitations. Participants were couples awaiting IVF and this homogeneous relatively healthy population may have resulted in some selection bias and may mean that our results do not generalise to a general population of couples of reproductive age or same-sex populations. Previous population studies have suggested that regardless of sex composition of the partnership, all couples demonstrate substantial within couple similarity in demographics including for age, education, race/ethnicity, work hours, and earnings (49). Determination of metabolite concentrations were undertaken on non-fasting samples, however, collaborations of several studies using this same NMR analysis platform results have not differed notably between studies in which the analyses were undertaken in fasted and non-fasted participants (18). Our analyses are cross-sectional and included couples within a narrow age range. With repeat assessments of couple correlations over time, or with cross-sectional data including couples with a wide age range and number of years of being together, it would be possible to explore the relative contributions of assortative mating and convergence on the weak metabolite correlations we have observed. Previous studies that have tried to explore this using marriage / cohabitation duration as a surrogate have found little evidence of any convergence for physical measures such as BMI or blood pressure, while behaviours such as smoking and alcohol converged during the initial period of a relationship prior to marriage / cohabitation, whereas convergence in physical activity was sustained throughout life (50)(51). The NMR platform used misses a proportion of the currently quantifiable metabolites in human serum/plasma, including markers of microbiota metabolism, vitamins, co-factors, and xenobiotics, that may be influenced by diet and preconceptual supplements. We do not have detailed dietary questionnaires, which would allow us to confirm our suggestion that a shared environment and common food would contribute to the observed correlations of metabolites.
We have explored within couple correlations of multiple metabolomic traits and find weak to modest positive correlations for the vast majority that are not influenced by adjustment for traits know to be influenced by assortative mating or shared couple behaviours. This suggests assortative mating, for example via genes linked to assortative characteristics such as height and education, might have some potential weak to modest impact on couples having similar metabolic traits. Whilst we acknowledge replication in a general population would be valuable the broadly similar within couple correlations of physical, social, and behavioural traits in these couples provides some evidence that our findings might be generalisable. Longitudinal studies would be valuable to fully explore the relative roles of assortative mating and convergence.