Discussion

To our knowledge, this is the first GWAS focused on PSs, including 17,511 individuals with PSs and 54,786 control individuals of European genetic ancestry. We identified a genome-wide significant risk locus on 12p13.3, with comparable point estimates in LifeGene and MoBa, although the association was not statistically significant in LifeGene. Moreover, we found 6 loci with borderline significance, 3 of which were nominally significant in both cohorts. A moderate SNP-based heritability of 7.2% was observed. Genetic correlations were found between PSs and a range of psychiatric disorders, with the strongest genetic overlap noted for major depression.

The CACNA1C gene, implicated by linkage to rs758170, encodes the calcium voltage-gated channel subunit alpha1C, crucial for calcium channel functioning essential to neurodevelopment (47,48). While we observed genome-wide significance for this locus in the meta-analysis, the association was not significant in the better-characterized cohort, LifeGene. Because the effect sizes were highly comparable in the 2 cohorts (LifeGene OR = 0.95 vs. MoBa OR = 0.92), we speculate that the less pronounced association in LifeGene is likely due to the relatively small sample size instead of difference by phenotyping (49). Moreover, rs2370419 (chr12:2423857), an intronic variant within the CACNA1C gene adjacent to rs758170 (_R_2 = 0.10 according to LDlink), was nominally significant (p = .03) in LifeGene, further supporting the implication of CACNA1C locus. However, replication in external populations with larger sample sizes and better assessment of PMD is warranted in the future. Previous in vivo research has found that the gene NUCB1, encoding a calcium-binding protein, is directly involved in regulating intracellular Ca2+ within the endoplasmic reticulum–Golgi compartment, contributing to the abnormal response to steroid hormones in patients with PMDD (50). Moreover, studies on calcium signaling in neural cells suggest potential interactions between KCNMA1 (calcium-dependent gene) and NUCB1 and CACNA1C as part of calcium regulation pathways (51,52). Because Ca2+ activity in developing neural cells is modulated by various membrane receptors, including GABA (gamma-aminobutyric acid), these findings highlight promising links between calcium regulation and GABA receptor function, which has been studied widely in relation to PMDs (53,54).

CACNA1C has been linked to a range of psychiatric disorders (55,56). In a study investigating the effects of CACNA1C haploinsufficiency on mouse behavior in tests with relevance to human mood disorders, researchers found that an intronic region of CACNA1C was involved in mood disorder pathophysiology (57). Moreover, several association studies have linked polymorphisms in CACNA1C to bipolar disorder and schizophrenia (55,56), potentially through variations in mean gray matter volume and mediotemporal emotional processing (58,59). rs758170 is highly correlated (LD _r_2 > 0.8) with variants that have been associated with bipolar disorder (60,61), schizophrenia (62), autism, and attention-deficit/hyperactivity disorder (ADHD) (63), suggesting potentially shared disease mechanisms between PMDs and these psychiatric disorders. However, the association of rs758170 remains similar after adjustment for depression and anxiety, indicating that our finding cannot be completely explained by psychiatric comorbidities. In addition, rs758170 has been linked to lipid metabolism (64), for which several epidemiological studies have illustrated a salient relationship between adiposity and PMDs (65,66).

The CD38 gene, linking to the borderline-significant variant rs147346386 (LifeGene OR = 0.60, p = .001 vs. MoBa OR = 0.84, p = 5.89 × 10−6), is involved in immune responses and has been linked to depression and anxiety (67, 68, 69). Considering the well-documented role of inflammatory processes in psychiatric disorders, the involvement of this gene is of particular significance in the context of our findings. Moreover, the KRT18P27 gene linked to borderline-significant variant rs4773561 (LifeGene OR = 0.89, p = .013 vs. MoBa OR = 0.93, p = 2.96 × 10−6) has been implicated through positional gene-set enrichment to alcohol and drug dependence (70). There are no prior reports on the other borderline-significant variant, rs76665457, or the CTB-85P21.1 gene. Future studies are needed to confirm the findings and to understand the potential mechanisms of these genetic variants and genes in relation to PMD.

While previous twin research indicates a sizable genetic contribution to PSs (16), our study represents the first attempt to estimate the genetic liability of probable PMDs. As reported in previous genetic studies of other complex traits (71,72), the SNP-based heritability estimate in our study (_h_2SNP = 0.07) is lower than that reported in twin studies. However, this is comparable to the SNP-based heritability estimate for major depression (73). Future research should focus on larger GWASs as well as rare alleles, other alleles not well captured by GWAS, and gene-environment interplay to better capture the genetic liability of PMDs.

Extensive clinical and questionnaire-based research has consistently demonstrated a high prevalence of psychiatric comorbidities, particularly depressive and anxiety disorders, among individuals diagnosed with PMDs (1,74). In the current analyses, we found the largest and most significant genetic correlation of PMDs with major depression, consistent with the confirmed phenotypic correlation in the literature (75). Together with the observed genetic correlation with anxiety and neuroticism, it suggests that PMDs may share substantial genetic liability with internalizing disorders. That said, the abnormal sensitivity to hormone fluctuations may represent a sex-specific internalizing risk that emerges during reproductive life. It is plausible that PMDs are rather hormonally triggered manifestations of a shared underlying vulnerability with depression or internalizing disorders. In addition, other psychiatric conditions such as bipolar disorder and ADHD have been found to co-occur with PMDs (45,76). Here, we observed genetic correlations with ADHD and bipolar disorder (including bipolar I, bipolar II, and schizoaffective type), echoing the previous report on the polygenetic association with these disorders observed in MoBa (31). Future studies are needed to understand the shared phenotypic and genetic link between these disorders, particularly ADHD given the emerging data (77,78), and PMDs. Moreover, we report a significant genetic correlation between endometriosis and PSs. This is consistent with previous research reporting a phenotypic link between endometriosis and premenstrual tension (79). These data support previous research conducted on endometriosis and the phenotypic relationship with major depression and other psychiatric disorders, potentially explained by shared dysregulated immunologic functions (80). Finally, no genetic correlation was observed with steroid hormone levels, lending support to the notion that PMDs are thought to reflect differential sensitivity to hormonal fluctuations rather than absolute hormone levels as shown in previous experimental studies (10).

This study is strengthened by its large sample size, the inclusion of 2 Nordic cohorts with banked biosamples, and the use of rich register and questionnaire data for case identification. This study also has some limitations. First, a portion of the individuals with PSs were identified through screening tools. Although validity has varied across studies, ranging from chance level (81) to fairly good (27,82), large epidemiological cohorts (83) have shown that retrospective questionnaires can achieve meaningful positive predictive value. This enables the identification of potentially important cases while preserving the large sample sizes needed for research discovery. Similarly, prospective symptom charting, as is required to establish the diagnosis, is not feasible in GWASs, where a large sample size is needed, and may result in a high attrition rate particularly for those with severe symptoms (84). Moreover, to complement our questionnaire assessment, we used clinical diagnoses recorded from national and regional health care registers. While we lacked information on clinical diagnostic process, prospective symptom charting has been outlined in clinical guidelines in Sweden (30). Although clinical guidelines are generally well followed in the Nordic countries due to the tax-funded health care system, studies have shown that in the United States, at least few health care providers use daily symptom monitoring for the diagnosis of PMD (85). With both questionnaire assessment and register-recorded diagnoses, we still might have captured moderate/severe PMS, PMDD, and false positive cases. However, such misclassification should be nondifferential in terms of the exposure (genetic variants) and would have attenuated the associations toward the null (86). Importantly, in a sensitivity analysis, we restricted cases to those confirmed by both questionnaire assessment and clinical diagnosis, which likely reflects higher diagnostic validity. While the analysis was limited in power and the point estimates differed somewhat, the effect directions for the top SNPs were broadly consistent. Second, depressive or anxiety symptoms may bias the assessment of PSs, leading to potential misclassifications. However, comparable results for the top loci were observed after adjustment for history of depression and anxiety. Nonetheless, given the high comorbidity and shared vulnerability (75,87, 88, 89), the presence of PMD in the current GWAS of depression and anxiety disorders can be substantial, which could spuriously boost the genetic correlation observed between these conditions. That said, the similarity between the current PMD GWAS and MDD GWAS is expected, and so would be for the genetic correlations with other psychiatric traits. Strong genetic correlations with anxiety and neuroticism have been observed for both PMD and MDD (31). Third, while it is unclear whether depression/anxiety symptoms were present at the assessment of PSs, our questionnaires are not designed to capture individuals with chronic mood disorders with perimenstrual worsening, i.e., premenstrual exacerbation (PME). In the absence of an established clinical diagnosis for PME, it is also difficult to identify this diagnosis from registers. However, PME represents a group potentially sharing underlying mechanisms with PMS/PMDD, and future research is needed to understand the shared or nonshared genetic architectures between these disorders. As the current study was not designed to formally assess pleiotropy with depression and related psychiatric traits, the specificity of the identified associations to PSs remains uncertain. Future studies incorporating formal cross-trait analyses will be needed to distinguish shared from trait-specific genetic effects. Fourth, due to the small numbers of participants, we removed participants of non-European ancestry. Future studies should include diverse ancestral backgrounds to aim for generalizable results across different groups, thereby making the findings relevant and potentially beneficial to a wider population, as the lack of replication in the current study is a limitation. Moreover, the difference in phenotyping impacts the prevalence rate between cohorts, enclosing a wider phenotypic heterogeneity. Finally, the phenotyping difference (different questionnaires to assess PSs) between the 2 cohorts and the relatively small sample size in LifeGene make it difficult to replicate the MoBa findings with the same level of statistical significance. However, the additional analyses of cross-cohort polygenic prediction, positive but nonsignificant genetic correlation between LifeGene and MoBa, as well as 80% to top SNPs in MoBa showing statistical comparable results in LifeGene lend further support to the cross-cohort phenotype compatibility and common genetic signals. Given differences in phenotype definition across cohorts, including both self-reported symptom measures and diagnostic phenotypes, our findings should be interpreted cautiously. The identified signals may reflect genetic influences on PSs captured in population-based samples and may also partially overlap with broader affective or internalizing liability. That said, our study underscores the need for larger studies with improved and harmonized phenotyping in the future.

Conclusions

In this first genome-wide association meta-analysis of PSs based on retrospective questionnaire assessment and register-based diagnoses, we report a significant locus and genetic correlations with a range of psychosocial and gynecological phenotypes. The identified genetic markers may help advance our understanding of the underlying mechanisms, which may further inform the development of early detection and clinical management for PMDs. Future studies with larger, more diverse samples and cases confirmed with prospective symptom charting are needed to further understand the genetic influence on PMDs and potentially the mechanisms of mood-regulating effects of sex hormones.