In this study, we examined a putative causal relationship between three hormonal reproductive traits (AAM, ANM and AFB) and an autoimmune inflammatory disease RA which affects mainly women. We capitalized on the summary statistics of the largest GWAS(s) conducted for these traits in European ancestry populations and constructed strong instruments using hundreds of SNPs associated with the exposures (F-statistic for AAM 11.89, for ANM 65.18, for AFB 11.68). We did not find convincing evidence in support for a causal effect of reproductive factors on RA using univariable MR analyses. Consistent null associations were identified by sensitivity analysis and multivariable MR analysis, demonstrating the robustness of our findings.
Current results from conventional epidemiological studies on this topic remain controversial, yet many studies point towards a positive association. For example, a study enrolling 121,700 female nurses found that age at menarche ≤ 10 years was associated with an increased risk of seropositive RA (RR 1.60, 95%CI 1.10-2.40).  A community-based health survey including 30,447 subjects (18,326 women) between 1991 and 1996 found an association between early age at menopause (≤ 45 years) and subsequent development of RA (OR 2.42, 95%CI 1.32-4.45), which remained significant after adjusting for smoking, level of education and length of breastfeeding (OR 1.92, 95%CI 1.02-3.64).  A prospective cohort study of 31,336 North America women reported similar findings (RRmenopause >51 vs. menopause <45 0.64, 95%CI 0.41-1.00). 
Our large-scale MR, however, did not identify a putative causal link between the three well-defined hormonal exposures and risk of RA. Several reasons underlie such a discrepancy. First of all, reproductive factors are highly complicated and heterogenous traits shaped by both genetic and environmental factors and genetics alone does not fully capture the phenotypic variance of these traits. For example, age at first birth is a human behavioral trait influenced largely by psychosocial, cultural and financial factors rather than the genetics. Secondly, results from previous epidemiological studies are likely to be impaired by confounding factors. For example, obesity is an important confounder affecting both the exposure and the outcome. An MR study demonstrated that a one-year delay in age at menarche reduced adult BMI by 0.38kg/m2 (95%CI 0.25-0.51kg/m2).  Global adiposity is a robust causal risk factor for RA as demonstrated by our recently published MR.  It is likely that traditional epidemiological investigations did not adequately control for the confounding effects from obesity. The protective effect of education on RA has been reported by observational studies. [27, 28] An MR study identified that a one-year later in age at menarche increased 0.14 years (53 days) of time spent in education.  We performed a MVMR to control for the effect of adiposity and education, and the negative results corroborating our main findings on a null association. Finally, it is also likely that the true causal effect of reproductive factors on RA is modest, which our study is underpowered to identify.
The biological mechanisms underlying hormonal factors and the development of RA remain unclear. The effect of sex hormones on the immune system and their interaction with environmental and genetic factors may partly explain the higher prevalence of RA observed among women. Estrogen is a complex modulator to the immune system exerting both a stimulatory and an inhibitory effect.  For example, estrogens at periovulatory to pregnancy levels stimulate B cells and the Th2 response and support the survival of auto-reactive T and B cell clones. On other hand, estrogens could inhibit cell-mediated responses such as the differentiation to Th17 cells. [30-32] A reduced risk of RA onset during pregnancy compared to an increased risk postpartum, suggests a role the hormonal changes or the exposure to fetus paternal HLA in RA onset. 
Our study has several strengths. To the best of our knowledge, no MR has been performed to assess the relationship between reproductive factors and RA. We incorporated three different reproductive traits (age at menarche, age at natural menopause and age at first birth) reflecting the length of reproductive period and complementing each other well. Moreover, we conducted important sensitivity analyses to verify MR model assumptions. We selected the most significant independent SNPs identified by the largest GWAS, so all were robustly associated with exposure of interest, guaranteeing “relevance” assumption. We excluded SNPs associated with potential confounders on the exposure-outcome relationship to satisfy “exclusion restriction” assumption. The consistent results observed across different approaches, further lend support to our findings.
We have to acknowledge several limitations. Firstly, our analysis was performed using the European populations which restricted its generalizability. Secondly, the genetic instruments of three exposures (AAM, ANM and AFB) we used as proxies for hormonal reproductive characteristics captured only a modest proportion of phenotypic variance. Reproductive factors are complex traits influenced by different components such as genetic, environmental and socioeconomic factors as well as their complex interactions. The design of our study disables us to take into account environmental impacts. Thirdly, the association between genetically predicted age at each of the reproductive events and risk of RA was evaluated fitting the exposure as a continuous variable – we can still not exclude a non-linear effect which was not captured by our study with the current availability of data. Future work on such topics may be focused on categorized age of reproductive events. Fourthly, our study was conducted using overall RA (a majority of which are seropositive RA, >85%) without specifying disease subsets characterized by the presence/absence of antibodies to citrullinated peptides or rheumatic factors. It is possible that hormonal factors influence different RA subsets via a distinct way. It is also likely that other factors such as hormone use and health conditions confound our results, in addition to the only two confounders (obesity and education) considered in the current study. However, it is difficult to control for the effect of hormone therapy due to limited availability of genetic data underlying this trait. Finally, power calculations showed that potential weak effects were difficult to be detected in our analysis.