Results

A total of 72,297 (5229 from LifeGene and 67,068 from MoBa) women were included in the final GWAS meta-analysis. Of these, 17,511 (24.2%) met the criteria for PSs (1962 [37.5%]); cases were oversampled for genotyping, while the prevalence in the whole cohort was 10.8% from LifeGene and 23.1% (n = 15,549) from MoBa. In LifeGene, compared with the control women, women with PSs were more likely to have lower educational attainment, not to have a partner, and to have a prior history of depression or anxiety disorder (Table 1). Similar characteristics were observed among MoBa participants.

 | LifeGene | MoBa
Without PSs, n = 3267 | With PSs, n = 1962 | p | Without PSs, n = 51,519 | With PSs, n = 15,549 | p
Age, Years | 34 ± 8.43 | 33 ± 7.79 | – | 29.47 ± 4.56 | 30.1 ± 4.68 | –
Educational Levela
Presecondary education | 112 (3.42%) | 48 (2.5%) | .060 | 1137 (2.2%) | 427 (2.7%) | <.001
Secondary education | 777 (23.7%) | 513 (26.1%) | 15,815 (30.7%) | 5393 (34.7%)
Postsecondary education | 2378 (72.4%) | 1401 (71.1%) | 31,936 (62.0%) | 8850 (56.9%)
Unknown | 12 (0.4%) | 5 (0.3%) | 2631 (5.1%) | 879 (5.7%)
Civil Statusa
Single/separated/widower | 1922 (58.8%) | 1228 (62.5%) | .024 | 915 (1.8%) | 301 (1.9%) | .003
Married/cohabitant | 1342 (41.0%) | 733 (37.3%) | 46,648 (90.5%) | 13,937 (89.6%)
Unknown | 3 (0.09%) | 1 (0.05%) | 3956 (7.7%) | 1311 (8.4%)
Depressionb
No | 2795 (85.6%) | 1390 (70.9%) | <.001 | 47,111 (91.4%) | 13,208 (84.9%) | <.001
Yes | 472 (14.4%) | 572 (29.1%) | 4408 (8.6%) | 2341 (15.1%)
Anxiety Disorderb
No | 2738 (83.8%) | 1316 (67.1%) | <.001 | 47,895 (93.0%) | 13,882 (89.3%) | <.001
Yes | 529 (16.2%) | 646 (32.9%) | 3624 (7%) | 1667 (10.7%)
Age at Menarche | 13.39 ± 3.64 | 13.20 ± 3.50 | – | 13.05 ± 1.37 | 12.93 ± 1.39 | –
≤10 years | 82 (2.5%) | 81 (4.1%) | .005 | 970 (1.9%) | 387 (2.5%) | <.001
11–14 years | 2577 (78.9%) | 1552 (79.1%) | 43,036 (83.5%) | 13,172 (84.7%)
≥15 years | 419 (12.8%) | 232 (11.9%) | 6860 (13.3%) | 1807 (11.6%)
Unknown | 189 (5.8%) | 97 (4.9%) | 653 (1.3%) | 183 (1.2%)
Paritya
0 | 1941 (59.5%) | 1196 (60.9%) | .134 | 29,505 (57.3%) | 7611 (48.9%) | <.001
1 | 398 (12.1%) | 247 (12.6%) | 14,740 (28.6%) | 4912 (31.6%)
2 | 697 (21.3%) | 367 (18.7%) | 6031 (11.7%) | 2458 (15.8%)
≥3 | 231 (7.1%) | 152 (7.8%) | 1243 (2.4%) | 568 (3.7%)

GWAS Analysis

QQ analysis of the GWAS meta-analysis indicated moderate inflation (λ = 1.075) (Figure S1). One locus at 12p13.3 was genome-wide significant (rs758170, p = 1.53 × 10−8, OR = 0.93, 95% CI [0.90 to 0.95], risk allele C) (Figure 1 and Table 2). While the effect sizes were comparable in the 2 cohorts (LifeGene OR = 0.95 vs. MoBa OR = 0.92, p for heterogeneity [_p_het] = 0.59), the association was not significant in LifeGene (p = .242). Moreover, 6 other loci showed a suggestive level of significance (p < 1.0 × 10−6) (Table 2), 3 of which were nominally significant in both cohorts (lead SNPs: rs76665457, rs147346386, and rs4773561).

Figure 1: Manhattan plot from the meta-analyzed genome-wide association study of premenstrual symptoms. This analysis included 17,511 cases and 54,786 controls. The dashed red line indicates a genome-wide significant threshold of p = 5 × 10−8. The pale dashed red line indicates a suggestive significance threshold of p = 1.0 × 10−6. Green dots show SNPs in linkage disequilibrium (r2 > 0.8) with the lead SNP of each genome-wide significant or suggestive locus. The point estimates of lead SNPs are provided in Table 2. SNP, single nucleotide polymorphism.

Chr | Position | rsID | Gene | A1 | A2 | Sample | OR (95% CI)a | p Value | p for Heterogeneity
12 | 2361460 | rs758170 | CACNA1C | C | T | LifeGene | 0.95 (0.86–1.04) | .242 | 
MoBa | 0.92 (0.90–0.95) | 2.68 × 10−8 | 
Meta-analysis | 0.93 (0.90–0.95) | 1.53 × 10−8 | .5949
5 | 144181270 | rs76665457 | CTB-85P21.1 | G | C | LifeGene | 0.80 (0.63–0.99) | .032 | 
MoBa | 0.86 (0.81–0.91) | 5.30 × 10−7 | 
Meta-analysis | 0.86 (0.80–0.86) | 5.96 × 10−8 | .5668
10 | 56883744 | rs12770903 | PCDH15 | G | T | LifeGene | 0.81 (0.62–1.02) | .066 | 
MoBa | 0.85 (0.80–0.91) | 7.31 × 10−7 | 
Meta-analysis | 0.84 (0.78–0.90) | 1.34 × 10−7 | .7665
4 | 15764240 | rs147346386 | CD38 | C | T | LifeGene | 0.60 (0.39–0.85) | .001 | 
MoBa | 0.84 (0.78–0.90) | 5.89 × 10−6 | 
Meta-analysis | 0.82 (0.77–0.89) | 1.95 × 10−7 | .0813
13 | 90887997 | rs4773561 | KRT18P27 | G | A | LifeGene | 0.89 (0.80–0.98) | .013 | 
MoBa | 0.93 (0.91–0.96) | 2.96 × 10−6 | 
Meta-analysis | 0.93 (0.90–0.95) | 2.06 × 10−7 | .3235
1 | 151581008 | rs77519409 | RP11-404E16.1 | G | A | LifeGene | – | – | 
MoBa | 1.17 (1.10–1.25) | 4.61 × 10−7 | 
Meta- analysis | 1.17 (1.10–1.25) | 4.61 × 10−7 | –
18 | 77629986 | rs112526506 | KCNG2 | G | A | LifeGene | 1.08 (0.95–1.23) | .183 | 
MoBa | 1.10 (1.06–1.15) | 1.50 × 10−6 | 
Meta- analysis | 1.10 (1.06–1.14) | 6.12 × 10−7 | .8539

Functional Annotation

The top SNP, rs758170 (intron), was positionally mapped to the CACNA1C gene (Figure 2). FUMA indicates that the C allele associated with PS risk leads to increased expression of CACNA1C in the cerebellum (p = 2.53 × 10−6). Locus 10q21.1 is mapped to RP11-478B11.2, which is a noncoding RNA gene. The allele G of rs12770903 was associated with decreased expression of RP11-478B11.2 in blood samples (p = 2.1 × 10−4). Other loci were not associated with differential gene expression in blood or brain tissues.

Figure 2: Regional plot for the lead SNP rs758170. SNPs that are not in LD of the lead SNP in the selected region are colored gray. Only SNPs that are in LD of the lead SNP are displayed in the plot. LD, linkage disequlibrium; SNP, single nucleotide polymorphism.

Next, we conducted several sensitivity analyses using LifeGene samples to test the robustness of our findings. To reduce the influence of comorbid psychiatric disorders, we also adjusted for depression and anxiety, yielding comparable associations for top loci (Table S5). To minimize the misclassification of probable PMD cases, we restricted analysis to cases confirmed by both questionnaire assessment and clinical diagnosis (n = 762). Statistically comparable point estimates were observed across cohorts, although statistical significance was not consistently observed in both datasets (Table S6).

SNP Heritability and Genetic Correlation

The SNP-based heritability was estimated as 0.072 (SE = 0.01, p = 2.46 × 10−12) and on the liability scale as 0.065 (SE = 0.009, p = 2.6 × 10−12). We observed significant positive genetic correlations between PSs and all considered psychiatric disorders (_r_g = 0.32−0.62), with the strongest correlation being with major depression (_r_g = 0.62, 95% CI [0.49 to 0.74], empirical p = 3.04 × 10−22) (Figure 3). There was a significant positive genetic correlation between PMDs and endometriosis (_r_g = 0.17, 95% CI [0.01 to 0.32], empirical p = .029). In addition, a positive genetic correlation was found for body mass index (_r_g = 0.1, 95% CI = [0.03 to 0.17], empirical p = .003), while significant negative genetic correlations were observed with age at first childbirth (_r_g = −0.35, 95% CI = [−0.47 to −0.22], empirical p = 7.01 × 10−8), subjective well-being (_r_g = −0.34, 95% CI = [−0.50 to −0.19], empirical p = 1.82 × 10−5), and educational attainment (_r_g = −0.15, 95% CI = [−0.22 to −0.07], empirical p = 8.39 × 10−5) (Table S4). Finally, we did not observe any significant correlation with blood steroid hormone levels.

Figure 3: Genetic correlations (rg) between premenstrual symptoms and psychosocial characteristics, gynecological traits, and steroid hormones. Dots indicate the point estimate, while caps indicate the 95% CI. Green dots denote empirical p value < .05. The actual estimates are provided in Table S4. ADHD, attention-deficit/hyperactivity disorder; BMI, body mass index.

Additional Analyses

To test the consistency between the LifeGene and MoBa cohorts, we performed several additional analyses. In the cross-cohort polygenic prediction analysis, we found a significantly positive association between the polygenic risk score based on the MoBa summary statistics and the probable case in the LifeGene cohort for SNPs with p value < .05 (OR = 1.18, p = 1.27 × 10−9). We also tested for SNPs in different p-value thresholds, p value < .005 and p value < .0005, and obtained similar results (OR = 1.08, p = .002 and OR = 1.09, p = .001, respectively). Moreover, the genetic correlation between the LifeGene and MoBa cohorts was estimated as 0.66 (SE = 0.38, p = .08). Last, the heterogeneity testing for SNPs with p < 1 × 10−4 from the MoBa cohort showed that 118 of 147 (80.3%) SNPs were statistically comparable with the corresponding point estimate from LifeGene (_p_het > .05).