Sex, male origin microchimerism, and mortality in a Danish cohort

Our original study of the association between MOM and mortality in women was carried out as a cohort study using peripheral blood samples and questionnaire data (including lifestyle, height, weight, and parity) from 272 women randomly sampled from the Danish Diet, Cancer and Health (DCH) cohort [12], linked with nationwide registry data (including date and cause of death). Women were followed up for all-cause mortality until the end of 2009. To study associations between MOM and female malignancies [13‐15] and cardiovascular [16] outcomes, we subsequently obtained peripheral blood samples and questionnaire data for an additional 494 randomly sampled women from the DCH cohort. A total of 774 women had available MOM exposure data from laboratory analysis of blood samples, confounding variable data from DCH questionnaires, and all-cause mortality data from the Danish Civil Registration System [17] until September 19, 2017. Only women with complete information on the baseline variables used for constructing the prognostic scores were included in the study (n = 766). In addition, we included a random sample of 1,000 men participating in the DCH cohort. For men, we had no measure of MOM, so we only included variables from the DCH baseline questionnaire for descriptive comparison with women and register-based mortality until September 19, 2017.

We obtained 1.2 μg DNA purified from buffy coat equivalent to 180,000 maternal cells from each woman. To identify presence of male DNA, we applied a validated qPCR assay targeting Y-chromosome gene DYS14 sequences [18], which should never be present in women. Samples were run in 6-plicate with DNA corresponding to 30,000 cells per well. qPCR was performed as previously described by our group [10]. We set the threshold cycle to 40, and if the intensity of the fluorescent light from a minimum of one of six wells crossed the threshold, the test was considered positive for the presence of MOM. Six wells with no template controls (NTC) were included for each run, and for a run to be accepted, all six NTC should be negative. We dichotomized the presence of MOM as positive versus negative by denoting a sample positive if at least one well tested positive, and negative if all wells proved negative. The estimated concentration of MOM was categorized as 0, > 0 to < 2, ≥2 to < 4, and ≥ 4 MOM cells per 10⁶ cells tested. To avoid male contamination, laboratory analyses were performed by female technicians who were blinded to the case/control status of the woman from whom the sample was taken.

In epidemiological investigations, confounding variables are often identified theoretically either a priori or guided by directed acyclic graphs, or empirically by e.g. change-in-estimate or forward/backward selection methods [19]. We averted the theoretical approaches because they require knowledge of the associations between the possible confounding variables and the exposure variable, which is not currently available for our exposure, namely MOM [20]. Likewise, empirical methods require that the decisions made about the associations are correct and do not result from sampling variability to avoid introduction of bias. Given these issues, we addressed confounding by incorporating available knowledge about factors influencing mortality, rather than MOM, as illustrated in a directed acyclic graph depicting the anticipated relationship between sex, MOM, and mortality (Fig. 1). We estimated the prognostic score [21, 22] as a single summary measure used to weigh the data and secure prognostic balance across exposure levels. Specifically, we fitted a Cox Regression model, setting death as the dependent variable, baseline predictors of death (smoking, alcohol intake, body mass index (kg/m²), educational attainment (a fundamental indicator of socioeconomic position shaping health related behavior across life [23]), parity, physical activity (hours/week), and dietary habits) as independent variables, and age as the underlying time scale. This model was derived using only data of women who tested negative for MOM (the unexposed). To correct for potential bias due to small sample cell sizes, we used Firth’s Penalized Likelihood (coxphf package for the R). The model coefficients for all seven predictors were retrieved, thereby allowing the prognostic score to be calculated for each woman regardless of her exposure status but for complete cases only. This prognostic score was assigned to each woman as a summary variable, informing on her risk of all-cause mortality that would have been expected, had she been MOM negative. Subsequently, we performed optimal full matching on the prognostic score between women who tested negative and positive for MOM to assign a weight to balance the two exposure groups [21, 24]. Finally, the weights were included in the Cox regression models using the survival package to obtain adjusted estimates of the association between MOM exposure and mortality. We made the R code openly available at the Open Science Framework [https://osf.io/q4ypr/?view_only=cbb0b51b548141cfaf49e6e32227c582], DOI https://doi.org/10.17605/OSF.IO/Q4YPR.

Characteristics of the study population are presented for women who tested positive and negative for MOM, and for men. Based on Cox regression models, we generated prognostic score weighted survival curves for women according to MOM status. Time-at-risk was counted from the time of the interview, i.e. enrolment in the cohort until death, migration, or end of follow-up, whichever occurred first. MOM was analyzed as a binary variable (positive versus negative) and estimates of association with mortality are presented as hazard ratios (HRs) with 95% confidence intervals (95% CIs). Next, we fitted Cox regression models to estimate the total effect of sex on survival according to attained age for women who tested positive and negative for MOM, and for men. These results are presented graphically as unweighted survival curves. The absolute measures of years lost in survival were calculated using the restricted mean survival time (RMST) approach [25] for women who were tested positive versus negative for MOM. RMST represents the area under the survival curve for each group and reflects the expected survival time. While including all ages would equate RMST to life expectancy at birth, in this analysis, it is limited to the years lost within the age ranges under study.

In sub-analyses, we first re-ran the association between MOM and mortality using traditional confounder adjustment rather than weighing for the prognostic score. We adjusted for the same variables used to calculate the prognostic score for comparability. Second, we calculated the E-value [26] as the minimum strength of association a group of uncontrolled confounders would need to have to explain away the associations observed between MOM and mortality. Third, mortality was evaluated according to the concentration of MOM in groups of 0, > 0 to < 2, ≥2 to < 4, or ≥ 4 MOM cells per 10⁶ cells tested. Fourth, we assessed the joint exposure to MOM and hormonal exposure at baseline on survival as previously described by our group [16] by constructing a variable with the following four levels: (1) women tested MOM positive who reported hormone exposure (nulliparity or ever use of hormonal replacement therapy or never use of oral contraceptives), (2) women tested MOM positive who reported no hormone exposure (parous and never use of hormone replacement therapy and ever use of oral contraceptives), (3) women tested MOM negative who reported hormone exposure, and (4) women tested MOM negative who reported no hormone exposure. This joint effect analysis was performed unadjusted and adjusted for smoking, alcohol intake, body mass index, educational attainment, physical activity, and dietary habits (the variables used to calculate the prognostic score, but without parity, which was included in the estimation of hormonal exposure). We assessed statistical interaction by incorporating a product term between MOM and hormonal exposure alongside their respective main effects. From the HR estimates, we calculated the relative excess risk due to interaction (RERI) with 95% CI [27]. Fifth and finally, we conducted unweighted cause-specific mortality analyses, treating death from other causes as a censoring event to appropriately account for competing risks. Results are reported as HRs, representing the relative event rates for the following comparisons: women tested MOM positive versus MOM negative, women tested MOM positive versus men, and women tested MOM negative versus men. Causes of death were categorized according to major chapters in the International Classification of Disease (version 10) system as follows: cancer (C), cardiovascular disease (I), respiratory diseases (J), neurological diseases (F, G), and other observed causes (A, B, D, E, K, M, N, R, V, W, X).

We evaluated the proportional hazards assumption using the cox.zph function in the survival package [28] correlating scaled Schoenfeld residuals for each covariate with time. Data management and statistical analysis were conducted using the R version 4.2.2 for Windows [29].

The research protocol was approved by the Danish Data Protection Agency (journal number 2011-41-6911) and by the Danish National Committee on Health Research Ethics (journal number H16021411) following national legislation.