Introduction
Premenstrual disorders (PMDs) are a heterogeneous group of conditions characterized by the occurrence of affective, cognitive, and somatic symptoms during the luteal phase of the menstrual cycle [Reference Takeda1, Reference Yonkers and Simoni2]. The severity of PMD symptoms ranges from mild to debilitating [Reference Takeda1–Reference Meth, Noga, Dubol, Xue, Sundstrom-Poromaa and Benedict3]. PMDs reduce the quality of life [Reference Wang, Keijser, Chen, Yu, Hysaj and Hagg4], increase the risk of suicidal behavior [Reference Yang, Sjolander, Li, Viktorin, Bertone-Johnson and Ye5], and consequently burden both individuals and healthcare systems [Reference Islas-Preciado, Ramos-Lira and Estrada-Camarena6–Reference Prasad, Wollenhaupt-Aguiar, Kidd, de Azevedo Cardoso and Frey9]. While the pathogenesis of PMDs remains unknown, their heritability is estimated to range from 30 to 60% [Reference Epperson, Steiner, Hartlage, Eriksson, Schmidt and Jones10, Reference Abdellaoui, Yengo, Verweij and Visscher11]. The pathophysiology includes interactions between fluctuating hormone concentrations – including progesterone, allopregnanolone, and estradiol – during the menstrual cycle, and neurotransmitter systems, such as the serotoninergic and GABAergic systems [Reference Takeda1, Reference Yonkers and Simoni2]. In the absence of etiological therapies, women with PMDs receive symptomatic treatment, which has limited efficacy and is associated with adverse effects [Reference Takeda1, Reference Yonkers and Simoni2]. However, thanks to recent initiatives, such as the introduction of a model diagnostic category into the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) [12] and the International Classification of Diseases, 11th Revision (ICD-11) [13], the challenges of PMDs are receiving due attention [Reference Marwick, Reilly, Allan and Golightly14].
The greater prevalence of PMDs in women with major depression (MD), the more severe clinical characteristics of bipolar disorder (BIP) in women with premenstrual dysphoric disorder, and the effectiveness of antidepressants in the treatment of PMDs indicate shared pathophysiology between PMDs and major psychiatric disorders (MPDs) [Reference Takeda1–Reference Meth, Noga, Dubol, Xue, Sundstrom-Poromaa and Benedict3, Reference Forrester-Knauss, Zemp Stutz, Weiss and Tschudin15, Reference Slyepchenko, Frey, Lafer, Nierenberg, Sachs and Dias16]. This is further supported by the association between the PMD symptoms and genetic liability for MPDs, including schizophrenia (SCZ), BIP, and MD [Reference Jaholkowski, Shadrin, Jangmo, Frei, Tesfaye and Hindley17]. Our recent genome-wide association study (GWAS) revealed that the symptoms of PMD are more genetically correlated with MPDs than with other disorders of the reproductive system [Reference Hysaj, Jaholkowski, Shadrin, Bergstedt, Lu and Bertone-Johnson18]. Moreover, the genes mapped to the putative genetic loci [Reference Hysaj, Jaholkowski, Shadrin, Bergstedt, Lu and Bertone-Johnson18] suggest overlapping mechanisms with SCZ, BIP, and MD [Reference Zhang, Xiao, Li and Li19, Reference He, An, Wang, Li, Li and Chen20].
Here we aimed to identify genetic variants shared between the symptoms of PMD and SCZ, BIP, and MD. We employed the conjunctional false discovery rate (conjFDR) method, leveraging GWAS associations of two traits to enhance the statistical power of identifying genetic variants shared between these traits [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21, Reference Andreassen, Thompson and Dale22]. The conjFDR method aids in identifying shared genetic variants, regardless of their effect directions [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. This is important because the genetic architecture shared among human traits typically comprises a mixture of variants with both concordant and discordant effect directions [Reference Tesfaye, Jaholkowski, Shadrin, van der Meer, Hindley and Holen23–Reference Fominykh, Shadrin, Jaholkowski, Bahrami, Athanasiu and Wightman27]. Subsequently, to gain insight into the biological pathways underlying the shared loci between PMDs and MPDs, we conducted a functional analysis of the genes mapped to the identified genetic variants, including drug enrichment methods to investigate potential drug repurposing opportunities.
Methods
Samples
Owing to the lack of uniform PMD diagnostic criteria until recently, and to methodological challenges in prospective PMD studies – for example, very high attrition rates of up to 50%, particularly among participants with the most severe symptoms [Reference Cohen, Soares, Otto, Sweeney, Liberman and Harlow28, Reference Henz, Ferreira, Oderich, Gallon, Castro and Conzatti29] – a large-scale GWAS of PMDs based on recommended prospective case ascertainment remains lacking. In light of evidence supporting retrospective, questionnaire-based PMD assessments for population-based research [Reference Bertone-Johnson, Hankinson, Johnson and Manson30], we accordingly used GWAS results derived from self-reported PMD-specific symptoms (S-PMD). We prepared summary statistics from the GWAS of S-PMD [Reference Hysaj, Jaholkowski, Shadrin, Bergstedt, Lu and Bertone-Johnson18] by restricting the analysis to the sample from the Norwegian Mother, Father and Child (MoBa) [Reference Brandlistuen, Kristjansson, Alsaker, Valen, Birkeland and Royrvik31, Reference Paltiel, Anita, Skjerden, Harbak, Bækken and Nina Kristin32] cohort (15,549 cases, 51,519 controls) (Supplementary Materials) to avoid sample overlap with GWAS for MPDs. Women were classified as S-PMD cases if they reported that they usually felt depressed or irritable before menstruation that subsided after the onset of their period, or if they had received a clinical PMD diagnosis recorded in the primary care registry (Supplementary Materials). We acquired summary statistics for GWASs of MPDs – SCZ (53,386 cases, 77.258 controls) [Reference Trubetskoy, Pardinas, Qi, Panagiotaropoulou, Awasthi and Bigdeli33], BIP (58,112 cases, 769,223 controls) [Reference O’Connell, Koromina, van der Veen, Boltz, David and Yang34], and MD (398,906 cases, 1,519,526 controls) [Reference Adams, Streit, Meng, Awasthi, Adey and Choi35] – from the Psychiatric Genomics Consortium. In the case of MD and BIP, to avoid sample overlap with the GWAS for S-PMD, we obtained the summary statistics from the GWAS after excluding the Norwegian cohorts. For all GWAS samples used, summary statistics were generated only from individuals of European ancestry.
GWASs investigated in the study were approved by the ethics committees, and informed consent was obtained from all participants.
Conjunctional false discovery rate (conjFDR)
We utilized the conjFDR approach to identify genetic variants shared between the S-PMD and MPDs [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. The conjFDR approach leverages pleiotropy between the analyzed phenotypes to increase the power to detect genetic variants shared between them [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. This method is robust to substantial differences in sample size between the analyzed GWASs, as demonstrated across various complex human trait pairs, including MPDs [Reference Jaholkowski, Tesfaye, Fominykh, Parekh, Wiström and Parker36–Reference Holen, Shadrin, Icick, Filiz, Hindley and Rodevand39].
The initial step involved demonstrating the presence of bidirectional cross-trait genetic enrichment between the S-PMD and MPDs, a criterion necessary for conducting conjFDR analysis [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. For this purpose, we generated QQ plots that depict the distributions of p-values for the primary trait conditioned on the p-values in the GWAS of the secondary trait, and vice versa. To counteract spurious enrichment, we constructed QQ plots by implementing random pruning and averaging over 500 iterations [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. For every linkage disequilibrium (LD)-independent region (r 2 > 0.1), we included only one randomly selected SNP per analysis iteration [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. To avoid potential bias in the estimated null distribution, we removed SNPs located within long-range LD regions that show strong associations with brain traits (major histocompatibility complex region chr6:25119106–33854733, chr8:7200000–12500000, and chr19:44909039–45912650; build GRCh37) [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. The presence of cross-trait enrichment is indicated on a QQ plot by successive leftward deflections of a primary trait, which correlate with increasing levels of association with a secondary trait [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21].
For each MPD, we run two conditional (condFDR) analyses [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21] that recalculate the associations between variants and a given MPD based on their associations with the S-PMD and vice versa. We defined the higher of the two obtained condFDR values for a specific variant as the conjFDR value, which serves as a conservative estimate of the association between that variant and the two traits of interest [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21]. For the cond/conjFDR analyses, we removed the same genomic regions as during the construction of the QQ plots. Following previous studies [Reference Smeland, Frei, Shadrin, O’Connell, Fan and Bahrami21], we used a conjFDR threshold of 0.05 to ensure sufficient power to detect true associations while restricting the expected proportion of false discoveries among significant findings to a low level.
We defined independent genomic loci according to the FUMA protocol [Reference Watanabe, Taskesen, van Bochoven and Posthuma40]. SNPs with conjFDR < 0.05 and LD r 2 < 0.6 with each other were labeled as independently significant. For each independent significant variant, all candidate variants were identified as those with LD r 2 ≥ 0.6 with the independent significant variant, regardless of their significance. For a given independent significant variant, the borders of the genomic locus were defined as the minimum and maximum positional coordinates across all corresponding candidate variants. Then, loci separated by < 250 kb were merged, and independent significant variants with LD r 2 < 0.1 were selected as lead variants for the locus. The most significant lead variant within a locus was assigned as the locus lead variant. The LD r 2 values were computed based on the 1000 Genomes Project European ancestry haplotype reference panel [Reference Auton, Abecasis, Altshuler, Durbin, Abecasis and Bentley41].
Additionally, for each locus identified in the conjFDR analysis, we performed a colocalization analysis using coloc (v5.2.3) R package [Reference Wallace42] to assess whether the same causal variant underlies the observed associations for both traits.
Functional analysis
To map genes for the lead SNPs identified in the conjFDR analyses, we applied the SNP2GENE function of FUMA (version v1.5.2) [Reference Watanabe, Taskesen, van Bochoven and Posthuma40]. Next, we performed functional and drug enrichment analyses for the combined set of genes mapped to loci shared between S-PMD and the three MPDs, using the ToppFun function of ToppGene [Reference Chen, Bardes, Aronow and Jegga43] We identified gene set enrichment within the Gene Ontology (GO) [Reference Ashburner, Ball, Blake, Botstein, Butler and Cherry44], the Human Molecular Signatures Database (MSigDB) [Reference Subramanian, Tamayo, Mootha, Mukherjee, Ebert and Gillette45, Reference Liberzon, Birger, Thorvaldsdottir, Ghandi, Mesirov and Tamayo46], and Kyoto Encyclopedia of Genes and Genomes (KEGG) [Reference Kanehisa, Furumichi, Sato, Matsuura and Ishiguro-Watanabe47] gene sets. All p-values were adjusted for multiple testing using the Benjamini–Hochberg procedure.
We merged the list of genes associated with S-PMD and the three MPDs, and we utilized the tissue expression data from the Adult Genotype Tissue Expression Project (GTEx version 10; data obtained 19/03/25) [Reference Consortium, Aguet, Anand, Ardlie, Gabriel and Getz48] to assess the expression of mapped genes in various tissues. These include the CNS, liver, nerve tibial, cervix, fallopian tube, adrenal gland, thyroid, ovary, vagina, adipose tissue, breast mammary tissue, skeletal muscle, and whole blood. According to the GTEx data presentation approach, isoforms were collapsed into a single gene, and no normalization or threshold procedures were applied. We performed the clustering and visualization procedures using the pheatmap R package [Reference Kolde49]. Similarly, we merged the list of genes associated with S-PMD and the three MPDs, and we utilized the whole human brain 10× RNA-seq gene expression data from the Allen Brain Cell Atlas to visualize anatomy- and cell type-specific expression [Reference Siletti, Hodge, Mossi Albiach, Lee, Ding and Hu50, 51]. We followed the data preparation and clustering procedures from the Allen Institute [51] and visualized the heatmaps of gene expression for specific cell clusters using the pheatmap R package [Reference Kolde49] for each MPD.
We merged the list of genes associated with S-PMD and the three MPDs, and we utilized the BrainSpan RNA sequencing data [Reference Miller, Ding, Sunkin, Smith, Ng and Szafer52] for females and the cerebroViz R package [Reference Bahl, Koomar and Michaelson53] to plot heatmaps visualizing the spatiotemporal expression of these genes. Additionally, using nonlinear LOESS modeling with the ggplot2 R package [Reference Wickham54], we visualized mean expression across lifespan periods in cortical structures, subcortical structures, and for both groups combined.
Results
Conjunctional FDR
Using conditioned QQ plots, we show bidirectional cross-trait genetic enrichment between S-PMD and all SCZ, BIP, and MD (Figure 1). This is visible as a consistent leftward deflection from the dashed line (indicating no enrichment) for subsets of variants with higher significance in the conditional trait.
Conditional QQ plots show cross-trait enrichment in symptoms of premenstrual disorders (S-PMD) and major psychiatric disorders (MPDs) at different threshold values. (A) S-PMD conditioned on schizophrenia (SCZ) (S-PMD |SCZ), (B) SCZ conditioned on S-PMD (SCZ | S-PMD), (C) S-PMD conditioned on bipolar disorder (BIP) (S-PMD | BIP), (D) BIP conditioned on S-PMD (BIP | S-PMD), (E) S-PMD conditioned on major depression (MD) (S-PMD | MD), and (F) MD conditioned on S-PMD (MD | S-PMD).
Figure 1. Long description
A six-panel grid of Q Q plots labeled A through F. Each plot features an x-axis representing Expected minus log sub 10 (q) and a y-axis representing Observed minus log sub 10 (p). A dashed diagonal line from the origin (0,0) to (7,7) represents the null distribution.
* Panel A (S-P M D conditioned on S C Z): Shows four curves. The blue line for All S N P s follows the diagonal. As the conditioning threshold for S C Z becomes more stringent (p sub S C Z less than 10 super minus 1, 10 super minus 2, and 10 super minus 3), the curves shift progressively upward and to the left, with the purple line (10 super minus 3) showing the greatest enrichment.
* Panel B (S C Z conditioned on S-P M D): Similar upward shift as conditioning on S-P M D becomes more stringent, with curves for p sub S-P M D thresholds showing a steep vertical departure from the diagonal.
* Panel C (S-P M D conditioned on B I P): Displays a similar pattern to Panel A, where more stringent B I P thresholds result in higher observed minus log sub 10 (p) values for S-P M D.
* Panel D (B I P conditioned on S-P M D): Shows a sharp upward divergence of B I P association signals when conditioned on lower p-value thresholds of S-P M D.
* Panel E (S-P M D conditioned on M D): Follows the same trend of leftward and upward shifting curves as the M D p-value threshold decreases.
* Panel F (M D conditioned on S-P M D): Shows the enrichment of M D signals at various S-P M D significance levels, with the purple curve (p sub S-P M D less than 10 super minus 3) reaching the top of the y-axis at an expected x-value of approximately 3.
ConjFDR analysis at FDR < 0.05 identified 9 loci shared between S-PMD and SCZ, 3 loci shared between S-PMD and BIP, as well as 14 loci shared between S-PMD and MD (Figure 2, Supplementary Tables S1–S3). Loci chr12:2345295/2361460 and chr18:77603283/77640207/77616819 were significant in conjFDR analyses of all three MPDs. Locus chr1:50458443/50528591 was significant between S-PMD and both SCZ and MD, and locus chr13:66954754/67095885 was significant in the conjFDR analyses between S-PMD and both BIP and MD.
The conjFDR Manhattan plots show common genetic variants that are jointly associated with (A) schizophrenia (SCZ) and premenstrual disorder symptoms (S-PMD), (B) bipolar disorder (BIP) and S-PMD, as well as (C) major depression (MD) and S-PMD at the conjunctional false discovery rate (conjFDR) < 0.05. The y-axis shows the -log10-transformed conjFDR. Chromosomal position is presented along the x-axis. The threshold for significant shared associations (conjFDR <0.05) is represented by the horizontal dotted line. Lead SNPS are indicated by a black perimeter. Lead SNPs with concordant effect direction for both given traits are depicted in red, while the ones with discordant effect direction are depicted in blue.
Figure 2. Long description
A three-panel multiplot of conj F D R Manhattan plots. Each panel shares a common x-axis representing Chromosome numbers 1 through 22 and a y-axis representing minus log 10 transformed conj F D R. A horizontal dotted line at 1.3 on the y-axis indicates the significance threshold of conj F D R less than 0.05.
Panel A, titled conj F D R S C Z and S-P M D, shows significant peaks at chromosomes 1, 2, 3, 6, 12, and 18. Notable lead S N P s include E L A V L 4, K L H L 20, L I N C 0 1 1 2 2, C N T N 4, I T I H 3, G R M 1, C A C N A 1 C, N O S 1, and K C N G 2. C A C N A 1 C is the highest peak, exceeding 4.0 on the y-axis.
Panel B, titled conj F D R B I P and S-P M D, shows significant peaks primarily at chromosomes 12, 13, and 18. Lead S N P s include C A C N A 1 C, P C D H 9, and K C N G 2.
Panel C, titled conj F D R M D and S-P M D, shows a higher density of significant associations across chromosomes 1, 7, 8, 11, 12, 13, 14, and 18. Lead S N P s include A G B L 4, R P L 3 1 P 1 2, H S 2 S T 1, R P 5 minus 1 0 5 2 I 5 dot 2, C 1 orf 5 3, R A B G A P 1 L, A C 0 0 7 5 6 8 dot 1, R P 1 1 minus 2 3 D 5 dot 1, P A X 6, Z 8 3 3 0 7 dot 3, C A C N A 1 C, R P 1 1 minus 1 1 4 G 2 2 dot 1, G P R 1 3 3, P C D H 9, R P 1 1 minus 6 6 1 D 1 9 dot 3, and K C N G 2.
In all panels, red circles represent concordant effect directions and blue circles represent discordant effect directions.
Seven of the nine (78%) lead SNPs shared between S-PMD and SCZ have concordant effect directions. For S-PMD and BIP, two of the three shared lead SNPs have concordant effect directions. Additionally, 13 out of 14 (93%) lead SNPs shared between S-PMD and MD have concordant effect directions (Figure 2).
The colocalization analysis of loci shared between S-PMD and SCZ (Supplementary Table S5) suggested that, at the locus chr12:2285731–2408194, both traits share a single common causal variant (86% probability), whereas, at the locus chr18:77588847–77659561, the traits have distinct independent causal variants (88% probability). For the locus chr12:2274051–2420526 shared between S-PMDs and BIP, the colocalization analysis (Supplementary Table S6) suggested a single common causal variant (95% probability) for both traits. The colocalization analysis suggested that, for three loci shared between S-PMD and MD, both traits share a single common causal variant at each locus (chr1:49463925–50591851, 87% probability; chr1:72628347–72959039, 81% probability; and chr13:66913378–67013856, 92% probability), whereas at the locus chr18:77553520–77986130 the traits have distinct independent causal variants (93% probability) (Supplementary Table S7).
Functional analysis
Functional annotation of shared lead SNPs identified in conjFDR analyses showed that intronic SNPs accounted for 77.8%, 100%, and 50% of SNPs shared between S-PMD and SCZ, BIP, and MD, respectively; 11:31808280 is located within a noncoding exonic RNA region of the PAX6 gene.
Gene-set enrichment analysis of positionally mapped genes for the Kyoto Encyclopedia of Genes and Genomes identified the “calcium signaling pathway.” Gene Ontology gene-set analysis identified three cellular components, namely “neuron projection,” “dendritic tree,” and “axon.” Meanwhile gene-set analysis for cell types in the Human Molecular Signature Database identified “Manno; midbrain neurotypes; human GABAergic neurons,” “Manno; midbrain neurotypes; human oculomotor and trochlear nucleus,” “Durante; adult olfactory neuroepithelium immature neurons,” and “Lake; adult kidney C15 connecting tubule” (Figure 3A). The gene list enrichment analysis of the mapped genes identified 36 drugs, including those registered for the treatment of central nervous system disorders (Figure 3B; Supplementary Table S4).
(A) The functional enrichment analyses for genes mapped to loci identified in the conjunctional false discovery rate (conjFDR) analysis for symptoms of premenstrual disorder (S-PMD) and major psychiatric disorders (MPDs). The dark violet bar depicts the findings from the Kyoto Encyclopedia of Genes and Genomes. The light green bars illustrate the results from the Gene Ontology, specifically the cellular component category. The light violet bars represent the findings for the cell type signature gene sets in the Human Molecular Signatures Database. The left panel shows the proportion of overlapping genes in each gene set. The middle panel illustrates the -log10-adjusted enrichment p-value (all groups remained significant after correction for multiple comparisons). The right panel displays the genes identified for each gene set, using the corresponding color coding. (B) The drug enrichment analyses for genes mapped to loci identified in the conjFDR analyses for symptoms of premenstrual disorder and major psychiatric disorder. The top panel (green bars) illustrates the adjusted p-value. The middle panel (light blue bars) shows the proportion of identified genes in the total gene set. The bottom panel depicts genes linked with a given drug, labeled by a navy blue filled rectangle.
Figure 3. Long description
Panel A contains three sub-plots organized by gene set categories: Kyoto Encyclopedia of Genes and Genomes (dark violet), Gene Ontology cellular component (green), and Human Molecular Signatures Database cell type signatures (light violet). The left sub-plot shows the proportion of overlapping genes in gene sets on an x-axis from 0.05 to 0.00. The middle sub-plot shows the negative log 10 adjusted enrichment p-value on an x-axis from 0.00 to 0.05. The right sub-plot is a grid matrix mapping specific genes (A G B L 4, C A C N A 1 C, C N T N 4, E L A V L 4, G P R 1 3 3, G R M 1, K L H L 2 0, N O S 1, P C D H 9, R A B G A P 1 L) to pathways like Calcium signaling, Neuron projection, and various midbrain neurotypes.
Panel B displays drug enrichment results. The top sub-plot uses green bars to show adjusted p-values ranging from 0.000 to 0.050. The middle sub-plot uses light blue bars to show the proportion of identified genes in the total gene set, ranging from 0.00 to 0.06. The bottom sub-plot is a large matrix where navy blue rectangles indicate links between 11 genes (listed vertically) and 38 drugs or compounds (listed horizontally), including Riluzole, Resveratrol, and Potassium.
Analysis of tissue-specific expression of the mapped genes showed expression within the CNS and identified two groups (clusters; Supplementary Figure S4) with distinct expression patterns. The first group includes genes such as PAX6, ELAVL4, and GRM1, which are predominantly expressed in the CNS. The second group includes genes such as HS2ST1 and CACNA1C, which are expressed across most of the analyzed tissues. An exception was the ITIH3 gene, which is predominantly expressed in non-CNS tissues. Furthermore, co-expression analysis of the mapped genes within the CNS identified two principal clusters: one encompassing the cerebellum and the other encompassing the cerebrum, diencephalon, mesencephalon, and spinal cord.
The analysis of spatial expression revealed that genes mapped to loci shared between S-PMD and MPDs are more highly expressed in neuronal cells compared to other cell types of the CNS (Figure 4A). Nevertheless, we did not observe any specificity towards particular nerve cell clusters (Figure 4A). We observed marked expression of these genes in most of the examined CNS structures, with notably lower expression in the spinal cord, myelencephalon, midbrain, and pons (Supplementary Figure S1). We found that these genes are expressed in nerve cells that release various neurotransmitters, with expression being higher in cells that release glutamate, serotonin, and dopamine (Figure 2).
(A) Expression of genes mapped to loci identified in the conjunctional false discovery rate (conjFDR) analysis for symptoms of premenstrual disorder (S-PMD) and major psychiatric disorders (MPDs) within brain cell clusters. The hierarchically clustered heatmap shows gene expression levels for each cluster. The color scale represents expression from low (pale blue) to high (dark blue). (B) Mean expression across lifespan periods of genes mapped to loci identified in conjFDR analysis for symptoms of S-PMD and MPDs in cortical structures (purple line), subcortical structures (green line), and all structures combined (black line, with gray shading indicating the 95% confidence interval). Lines were fitted using nonlinear LOESS modeling.
Figure 4. Long description
Panel A is a hierarchically clustered heatmap. The vertical y-axis lists 31 brain cell clusters, including Cerebellar inhibitory, Thalamic excitatory, and various hippocampal and glial cells. The horizontal x-axis lists 20 genes, including P C D H 9, A G B L 4, and C A C N A 1 C. Dendrograms on the top and left indicate clustering relationships. A color scale bar to the right indicates expression levels from pale blue (low) to dark blue (high). High expression is concentrated in the leftmost columns for most neuronal clusters.
Panel B is a line graph titled Mean gene expression on the y-axis, ranging from 0.10 to 0.25. The x-axis represents lifespan periods: First trimester, Second trimester, Third trimester, Infancy, Early childhood, Late childhood, Adolescence, Young adulthood, and Adulthood. Three lines represent different structures: a purple line for Cortical, a green line for Subcortical, and a thick black line for Combined with a gray 95 percent confidence interval band. All three lines show a bimodal trend: an initial peak during Infancy, a sharp decline in Early childhood, and a second, higher peak during Young adulthood before a slight decline in Adulthood. The Cortical expression is generally higher than Subcortical expression after the Third trimester.
The analysis of the spatiotemporal expression of genes mapped to loci common to S-PMD and MPDs showed widespread expression across the examined CNS structures, with notably higher levels after birth, particularly during infancy and young adulthood (Figure 4B and Supplementary Figure S3).
Discussion
In the current study, we identified a total of 20 loci shared between S-PMD and three MPDs, including 19 novel loci for S-PMD. Functional analysis of genes mapped to the identified loci revealed associations with neuronal cells and their projections, and with the calcium signaling pathway. Moreover, the drug enrichment analysis of the mapped genes identified 36 drugs, including riluzole and mianserin, indicating the possibility of repurposing them for treatment of PMDs.
The finding of genetic enrichments between S-PMD and MPDs, along with the concordant effect direction of most identified SNPs shared between these traits, suggests that common genetic factors underlie the high comorbidity observed between PMDs and MPDs [Reference Takeda1, Reference Yonkers and Simoni2, Reference Forrester-Knauss, Zemp Stutz, Weiss and Tschudin15, Reference Slyepchenko, Frey, Lafer, Nierenberg, Sachs and Dias16]. These observations are consistent with previous studies that found an association between S-PMD and the polygenic risk score for MPDs [Reference Jaholkowski, Shadrin, Jangmo, Frei, Tesfaye and Hindley17], as well as a genetic correlation between S-PMD and MPDs [Reference Hysaj, Jaholkowski, Shadrin, Bergstedt, Lu and Bertone-Johnson18]. Moreover, they are typical of the results observed for closely related complex human traits and disorders [Reference Tesfaye, Jaholkowski, Hindley, Shadrin, Rahman and Bahrami24, Reference Hindley, Frei, Shadrin, Cheng, O’Connell and Icick55].
The finding of gene-set enrichment for neuronal projections and neural cell types, coupled with the increased expression of mapped genes in neurons compared to other CNS cells and the elevated expression within brain structures rich in synaptic connections, suggest that the co-occurrence of PMDs and MPDs may be related to neuronal network functioning. Notably, genetic variants associated with SCZ, BIP, and MD, along with the pharmacological treatments used for these conditions, are largely connected to neuronal functioning [Reference Yao, Harder, Darki, Chang, Li and Nikouei56–Reference Nierenberg, Agustini, Kohler-Forsberg, Cusin, Katz and Sylvia60]. Similarly, previous studies on PMDs have highlighted the roles of serotonergic and GABAergic systems, as well as the therapeutic efficacy of drugs that modulate synaptic transmission [Reference Takeda1, Reference Yonkers and Simoni2]. Furthermore, cyclic sex hormone fluctuations modulate GABA signaling and affect dendritic spine plasticity [Reference Gilfarb and Leuner61, Reference Beeson and Meitzen62]. Gene products mapped in our study to loci shared between the S-PMD and MPDs play diverse roles in the functioning of the neural network, including neurotransmitter receptor (GRM1 [Reference Ayoub, Angelicheva, Vile, Chandler, Morar and Cavanaugh63]), ion channel (CACNA1C [Reference Harrison, Husain, Lee, Los Angeles, Colbourne and Mould64]), adhesion molecule (PCDH9 [Reference Uemura, Furuse, Yamada, Kushida, Abe and Imai65]), and transcription regulator (PAX6 [Reference Ochi, Manabe, Kikkawa and Osumi66]). This provides new insights into the biological pathways underlying these disorders and their co-occurrence.
Our finding of the temporal expression pattern of genes shared between S-PMD and MPD may mirror the hypothalamic–pituitary–gonadal axis (HPG) activity. High expression levels of these mapped genes during infancy correspond to the transient activation of the HPG, known as mini-puberty, which is observed during this developmental phase [Reference Andersson, Toppari, Haavisto, Petersen, Simell and Simell67]. This is followed by low expression levels in early childhood when HPG activity, and particularly the level of luteinizing hormone, is lower [Reference Andersson, Toppari, Haavisto, Petersen, Simell and Simell67]. From adolescence, when HPG activity increases again, we observed a renewed rise in the expression of these mapped genes [Reference Rohayem, Alexander, Heger, Nordenstrom and Howard68]. This observation may be associated with the developmental trajectories of PMDs.
Our enrichment analysis identified the calcium signaling pathway as one of the gene sets shared between S-PMD and MPD. This finding is consistent with previous observations linking calcium signaling to the pathogenesis of SCZ, BIP, and MD [Reference Heyes, Pratt, Rees, Dahimene, Ferron and Owen69, Reference Harrison, Hall, Mould, Al-Juffali and Tunbridge70], studies indicating that higher calcium intake may reduce the risk of premenstrual syndrome [Reference Bertone-Johnson, Hankinson, Bendich, Johnson, Willett and Manson71, Reference Li, Goff, Rudzinskas, Jung, Dubey and Hoffman72], and with randomized clinical trials suggesting an effect of calcium supplementation in the treatment of PMD [Reference Robinson, Ferreira, Iacovou and Kellow73, Reference Thys-Jacobs, Ceccarelli, Bierman, Weisman, Cohen and Alvir74]. Although prior studies have implicated changes in serum calcium concentration across the menstrual cycle, noted similarities between S-PMD and hypocalcemia, and indicated differences in calcium-regulatory hormone levels among women with premenstrual syndrome, the biological underpinnings of these phenomena remain unclear [Reference Thys-Jacobs and Alvir75–Reference Thys-Jacobs77]. Our study suggests that genetic variants within the identified calcium-related genes may underlie the association between PMDs and calcium physiology. At the same time, our results suggest that alterations in calcium homeostasis and signaling may constitute pathophysiological mechanisms shared by PMDs and MPDs and may underlie their co-occurrence. These processes may help explain the pathophysiology of PMDs and MPDs, elucidate the therapeutic mechanism of calcium supplementation, and support the development of new therapeutic strategies.
In addition to the three genes highlighted in our analysis (CACNA1C, GRM1, and NOS1), the functions of two other genes (PCDH9, PAX6) shared between S-PMD and MPDs are linked to calcium [Reference Kim, Yasuda, Tanaka, Yamagata and Kim78, Reference Leiper, Walczysko, Kucerova, Ou, Shanley and Lawson79]. PCDH9 encodes protocadherin 9, a member of the cadherin family, which are calcium-dependent cell adhesion molecules [Reference Xiao, Zheng, Chang, Ma, Yao and Luo80]. It is predominantly expressed in the brain, with its product frequently localized at synapses, and has been linked to susceptibility to psychiatric disorders [Reference Uemura, Furuse, Yamada, Kushida, Abe and Imai65, Reference Xiao, Zheng, Chang, Ma, Yao and Luo80, Reference Miozzo, Murru, Maiellano, di Iasio, Zippo and Zambrano Avendano81]. PAX6 encodes a transcriptional factor required for the development and functioning of the eyes and central nervous system [Reference Pratt, Vitalis, Warren, Edgar, Mason and Price82, Reference Grant, Bobilev, Rasys, Branson Byers, Schriever and Hekmatyar83]. It directly controls the expression of matrix metalloproteinase-9 (MMP-9), a protease involved in the cleavage of the extracellular matrix and in the plasticity of excitatory synapses, which has been linked to the pathogenesis of MPDs [Reference Lepeta and Kaczmarek84, Reference Kaczmarek, Protokowicz and Kaczmarek85]. Interestingly, subnormal propagation of injury-induced calcium waves was observed in the corneal epithelium of the Pax6 +/− mouse model [Reference Leiper, Walczysko, Kucerova, Ou, Shanley and Lawson79]. While calcium plays a role in a wide variety of CNS processes [Reference Burgoyne, Helassa, McCue and Haynes86, Reference Berridge87], our identified genes and associated calcium-related pathways implicate more specific mechanisms.
Current strategies for the treatment of PMDs are based on the use of selective serotonin reuptake inhibitors (SSRIs), oral contraceptives, nonsteroidal anti-inflammatory agents, and gonadotropin-releasing agonists [Reference Thys-Jacobs77]. Due to their associated side effects and limited efficacy, alternative, safer, and more effective treatments are needed. In this context, the results of our drug enrichment analysis are of interest. The results highlighted riluzole and mianserin, two well-tolerated drugs used in the treatment of CNS diseases. Riluzole inhibits glutamate release, blocks voltage-activated calcium and sodium channels, and activates potassium channels [Reference Vingerhoets, Hirdes, McAlonan, Amelsvoort and Zinkstok88, Reference Doble89]. It is approved for the treatment of amyotrophic lateral sclerosis and is being explored for treatment of mental disorders [Reference Vingerhoets, Hirdes, McAlonan, Amelsvoort and Zinkstok88–Reference Pajer, Bellak, Grosz, Nogradi, Patai and Sinko90]. Mianserin is an antidepressant that blocks serotonin 5-HT2, histamine H1, and α2-adrenoceptors [Reference Olianas, Dedoni and Onali91, Reference Peroutka and Snyder92]. Due to its distinct pharmacodynamic properties compared to SSRIs, mianserin is a viable alternative in cases of SSRI intolerance or ineffectiveness. This makes it an intriguing candidate for repurposing trials in PMD treatment, including cases with co-occurring affective disorders. Additionally, it remains an issue whether the absence of SSRIs among the drugs identified in our enrichment analysis reflects methodological limitations or a distinct mechanism of action of SSRIs in MPD and PMDs that has been suggested [Reference Modzelewski, Oracz, Żukow, Iłendo, Śledzikowka and Waszkiewicz93].
There are certain limitations to our study. First, the GWAS used for S-PMD is based on a retrospective questionnaire assessment, lacking the recommended prospective assessment [Reference Marwick, Reilly, Allan and Golightly14, 94]. Another limitation is the use of GWASs for MPD based on both sexes, which stems from the lack of sufficiently large female-specific GWASs. Additionally, the generalizability of our study results is limited due to the constraints of the tools used and the focus on data for individuals with European ancestry.
Conclusions
In the present study, we identified genetic variants shared between S-PMD and SCZ, BIP, and MD, with predominantly concordant directions of effect, consistent with clinically observed comorbidity. Additionally, the associated genes were linked to neuronal projections and calcium signaling, highlighting the biological basis and potential role of calcium-related interventions. The temporal expression pattern suggests that PMDs may have a developmental aspect. Furthermore, gene-set enrichment analysis suggested several drugs with potential for repurposing in PMD treatment. Ultimately, these findings provide a genetic and functional basis for the relationship between PMDs and MPD, with implications for future drug development.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1192/j.eurpsy.2026.12260.
Data availability statement
The GWAS summary statistics for major depression, bipolar disorder, and schizophrenia are publicly available through the Psychiatric Genomics Consortium (https://pgc.unc.edu). The code for the conjunctional False Discovery Rate (conjFDR) analysis is available on GitHub (https://github.com/precimed/pleiofdr). The databases and software used in this study are accessible at the following locations: FUMA (https://fuma.ctglab.nl/), the Adult Genotype Tissue Expression Project (https://gtexportal.org/), ToppGene (https://toppgene.cchmc.org/), the Allen Brain Cell Atlas (https://brain-map.org/), the Kyoto Encyclopedia of Genes and Genomes (https://www.genome.jp/kegg/).
Acknowledgments
The authors would like to thank the study participants, staff, and researchers of the Norwegian Mother, Father and Child Cohort Study (MoBa) and the Psychiatric Genomics Consortium (PGC) for making this research possible. This work was partly performed on the TSD (Services for Sensitive Data) facilities, owned by the University of Oslo, operated and developed by the TSD service group at the University of Oslo, IT Department. Computations were also performed on resources provided by UNINETT Sigma2 – the National Infrastructure for High Performance Computing and Data Storage in Norway.
Financial support
Research Council of Norway (grants: 324252, 344121, 326813, 296030, 334920, 223273, and 248778), European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie Actions Grant (grant 801133), European Union’s Horizon 2020 Research and Innovation Programme (grants: 847776 CoMorMent and 964874 RealMent), EEA and Norway grant (grant EEA-RO-NO-2018-0573), The South-East Norway Regional Health Authority (grant 2022-087), National Institutes of Health (grants: 5R01MH124839-02, U24DA041123, and U24DA055330).
Competing interests
O.A.A. has received speaker fees from BMS, Lundbeck, Janssen, Otsuka, Lilly, and Sunovion, and is a consultant to Cortechs.ai. and Precision Health. MH has served as a speaker for Lundbeck. All other authors report no potential conflicts of interest.
Open access
Comments
No Comments have been published for this article.