Integrated proteomic, transcriptomic, and epigenomic profiling identifies SRA1 as a novel therapeutic target for postpartum depression

Abstract

Postpartum depression (PPD) is among the most common complications of childbirth, and identifying novel treatments is vital. We aimed to identify potential drug targets for PPD by integrating the plasma proteome, transcriptome and epigenome. We designed a comprehensive analysis pipeline involving two-sample Mendelian randomisation (MR) (for proteins), colocalisation (for coding genes), and summary-based MR (SMR) (for mRNA and DNA methylation) to identify potential therapeutic targets for PPD. Genetic data on the plasma proteome were obtained from 4907 aptamers in 35,559 Icelanders and 7596 proteins in 828 FinnGen participants. The PPD genome-wide association study data were sourced from the Psychiatric Genomics Consortium (PGC) (Ncase = 17,339, Ncontrol = 53,426). A two-step MR approach was used to assess whether brain imaging-derived phenotypes (IDPs) and metabolites from blood, brain and cerebrospinal fluid mediated the observed effects. Across the two proteome datasets, the genetically predicted levels of 18 plasma proteins were nominally significantly associated with PPD, and the expression of steroid receptor RNA activator 1 (SRA1), a regulator of steroid hormone signalling, was significantly associated with PPD. SRA1, angiotensinogen (AGT, a key mediator of the renin-angiotensin and stress-response system), and glycerol-3-phosphate phosphatase (PGP, involved in lipid metabolism and cellular stress) showed increased colocalisation. The methylation of SRA1 at cg02434007 in the brain was associated with increased expression of SRA1 and a high risk of PPD, which aligns with the positive effect of SRA1 gene expression on PPD risk. Isoleucine (mediation proportion: 5.8%, p = 0.042) from blood metabolites and the IDP ICA100 edge 442 (mediation proportion: 7.6%, p = 0.044) may play mediating roles. This study reveals that SRA1 is a novel therapeutic target for PPD, which enhances the understanding of its molecular aetiology and the development of therapeutic strategies.

Introduction

Postpartum depression (PPD) is a severe psychiatric disorder affecting approximately 20% of women within the first year postpartum¹. Beyond its debilitating effects on mothers, PPD can significantly impact child development and family well-being. Despite its prevalence, the underlying biological mechanisms of PPD remain poorly understood, and current therapeutic options often fail to address the disorder effectively^2,3.

The aetiology of major depressive disorder (MDD) is heterogeneous, and PPD represents a more heritable subset of MDD^4–6. The current genome-wide association study (GWAS) data on the largest cohort of PPD in a European population provides an opportunity to examine potential targets on the basis of genetics⁶. Mendelian randomisation (MR) employs genetic variations as instrumental variables (IVs) to establish causal relationships between exposures and outcomes^7,8. Recent advances in multiomics approaches offer powerful tools for determining the molecular aetiology of complex diseases. Proteomic and transcriptomic MR analyses have demonstrated potential for identifying novel biomarkers and therapeutic targets^9,10. Integrating these approaches with genetic data allows for robust identification of causal relationships and potential drug targets. However, few studies have applied these methodologies to PPD.

In this study, we employed a thorough analysis pipeline that included proteome-wide MR, colocalisation, transcriptome-wide, and epigenomic MR analyses to identify potential causal proteins. Among the key proteins implicated in this process, steroid receptor RNA activator 1 (SRA1) is a bifunctional regulator of steroid hormone receptor signalling, which encompasses the oestrogen and progesterone pathways and has previously been linked to mood regulation and reproductive biology^11,12. To further explore the underlying mechanisms, we performed a two-step MR analysis.

Methods

Study design

The study design is illustrated in the graphic abstract. We defined a comprehensive analysis with the following steps: (1) selection of potential protein targets associated with PPD from the largest GWAS datasets using proteome-wide association studies conducted through deCODE and FinnGen on the SomaScan platform; (2) verification of shared coding gene loci of the identified proteins with PPD through colocalisation analysis; (3) assessment of the expression of candidate protein-coding genes and identification of consistent associations via a transcriptome-wide association study (for mRNA and DNA methylation) employing summary-based MR (SMR); (4) exploration of the associations between candidate protein biomarkers, MDD symptoms and other neuropsychiatric disorders and exploration of broader health implications using MR-phenome-wide association studies (MR-PheWAS); and (5) investigation of potential intermediate factors through blood, cerebrospinal fluid (CSF), brain metabolites and brain imaging-derived phenotypes (IDPs).

Data sources of plasma proteome, transcriptome, epigenomic and PPD

We acquired protein quantitative trait loci (pQTLs) associated with plasma proteins from two large independent GWAS datasets. The first dataset included 4907 aptamers measured in 35,559 Icelanders from the SomaScan platform¹³. The second dataset included 7596 proteins measured in 828 participants from the FinnGen study, also using the SomaScan platform¹⁴. According to these signals from the two GWAS datasets, cis-pQTLs were defined as single-nucleotide polymorphisms (SNPs) within 1 Mb of the gene encoding the protein. Gene expression data were obtained from the eQTLGen consortium and 49 tissues sourced from the GTEx (v8) project, with a focus on SNPs linked to changes in the expression of plasma protein-targeted genes. Our analysis concentrated on cis-expression quantitative trait loci (eQTLs) to confirm that genetic variants were relevant to alterations in gene expression, which enabled us to explore tissue-specific associations and evaluate potential off-target effects of drugs targeting these genes¹⁵. DNA methylation quantitative trait locus (mQTLs) association data from 27,750 individuals¹⁶ and brain mQTLs data (estimated effective n = 1160)¹⁷ were utilised in this analysis. The GWAS data for PPD were sourced from the largest meta-analysis to date of the Psychiatric Genomics Consortium (PGC), comprising 18 cohorts of European ancestry, totalling 17,339 PPD cases and 53,426 controls. All public databases utilised in this research have obtained ethical approval as stipulated in their respective original articles.

Potential mediators of blood, brain and CSF metabolic and brain IDPs

In accordance with Jia et al.^18,19, our study used summary statistics for the blood metabolome derived from three large-scale GWAS datasets with a total of 147,827 European participants. Specifically, Shin et al. analysed 452 metabolic traits in 7824 participants using 3 million SNPs²⁰; Kettunen et al. studied 123 traits in 24,925 participants with 12 million SNPs²¹; and Borges et al. examined 249 traits in 115,078 participants from the UK Biobank (https://gwas.mrcieu.ac.uk/). For overlapping metabolites, the statistics from the study with the larger sample size were retained. After excluding overlapping and unknown metabolites, a total of 431 metabolites were included for further analysis. Metabolites were selected on the basis of SNPs significantly associated with blood metabolites in these GWAS datasets (p < 5E-08), excluding SNPs in linkage disequilibrium. Among the 431 unique and annotated metabolites, 335 were excluded because of insufficient instrumental variables: 3 metabolites had 0 SNPs, 253 had only 1 SNP, and 79 had only 2 SNPs. The remaining 96 metabolites with at least 3 SNPs were retained; however, 3 were further removed because of weak genetic instruments (variance explained < 0.5%), resulting in a final set of 93 metabolites for MR analysis. Wang et al. conducted a genome-wide association study of 440 CSF metabolite levels from participants in five cohorts and 962 brain metabolite levels from the following three different sources. Comprehensive details can be found in the associated article²². Summary statistics for the brain IDPs originated from a recently published GWAS conducted by Smith et al., which examined 3953 brain IDPs in a sample size of up to 33,224 individuals drawn from UK Biobank cohorts²³. All findings are accessible via the Oxford Brain Imaging Genetics web browser (http://big.stats.ox.ac.uk/).

Statistical analysis

This study adhered to the STROBE-MR guidelines²⁴. We selected potential protein targets for PPD using MR analysis. In the discovery phase, using deCODE, a two-sample MR analysis was conducted where cis-pQTLs served as IVs for each protein, with the outcome derived from multivariate genomic analyses of PPD. For proteins with only one cis-pQTL, the Wald ratio method was employed to estimate the odds ratio (OR) and corresponding confidence interval (CI)²⁵. For proteins with multiple cis-pQTLs, estimates were obtained using the inverse variance weighted (IVW) method²⁶.

This approach necessitated adherence to three key assumptions: (1) the IVs must be associated with the exposure, (2) the IVs influence the outcome solely through the exposure, and (3) the IVs are not correlated with confounders. To evaluate these assumptions, we calculated F statistics to assess the strength of the IVs, performed MR‒Egger regression to check for horizontal pleiotropy, and tested for heterogeneity among the pQTLs using Q statistics. The MR estimators indicated the risk of PPD per standard deviation increase in genetically predicted levels of circulating proteins. In the replication phase, MR analysis was conducted using the plasma proteome from FinnGen as the exposure. To control for multiple testing, a false discovery rate (FDR)-corrected p-value (q < 0.05) was considered to indicate statistical significance, whereas a p-value < 0.05 was regarded as nominally significant. Two-sample MR analysis was conducted using the “TwoSampleMR (version 0.5.6)” package of R software (version 4.3.2).

We conducted a colocalisation analysis to estimate posterior probabilities for five potential hypotheses about the association of causal variants within a genomic region^27,28, which employs a Bayesian framework using the coloc package. For each shared locus, we extracted summary statistics for variants located within 1 Mb of the target gene and computed the posterior probability for H4 (PPH4) (association with both traits through a shared SNP) using both the coloc.abf and coloc.susie algorithms^29,30. For the coloc.abf algorithm, we interpreted a result as colocalisation if the PPH4 > 0.50. The coloc.susie (sum of single effects) method extends the standard coloc approach by first performing fine mapping with SuSiE to identify credible sets of causal variants for each trait and then performing pairwise colocalisation analysis across all combinations of credible sets. A result was considered colocalised if the PPH4 > 0.50 for any pair of credible sets. Evidence from either method was considered sufficient to support colocalisation. Strong colocalisation between the identified protein and PPD was defined as the PPH4 > 0.80.

Transcriptome-wide MR analysis was conducted using summary data to evaluate the associations between target gene methylation, expression, and protein abundance and the risk of PPD³¹. We developed a hypothetical mediation model in which SNPs influence a trait by altering DNA methylation (DNAm) levels, which subsequently regulate the expression of a functional gene. Consequently, we utilised SMR to analyse the target gene DNAm and gene expression in a multiomics context (mQTLs–eQTLs). We selected the top associated cis-QTLs by considering a window centred on each relevant gene within 1000 kb and applying a p-value threshold of 5E-08. The heterogeneity in the dependent instrument (HEIDI) test was used to distinguish between pleiotropy and linkage, with a p_HEIDI < 0.01 indicating likely pleiotropy, leading to exclusion from the analysis. Both the SMR analysis and the HEIDI test were conducted using version 1.3.1 of the SMR software.

We conducted a comprehensive analysis to investigate the causal relationships between candidate proteins and MDD³² and its symptoms³³, as well as other neuropsychiatric disorders, including attention-deficit/hyperactivity disorder (ADHD)³⁴, autism spectrum disorder (ASD)³⁵, bipolar disorder (BIP)³⁶, epilepsy³⁷, obsessive‒compulsive disorder (OCD)³⁸, posttraumatic stress disorder (PTSD), and schizophrenia (SCZ)³⁹. Additionally, we performed two-sample MR-PheWAS analysis. The Lee Lab’s SAIGE method analysed approximately 1400 ICD-based UK Biobank Phecode binary phenotypes (408,910 White British samples) across 28 million imputed variants, encompassing 17 systemic disease classifications^40,41. The SAIGE method ensures accurate p-value estimates even in unbalanced case–control settings. GWAS summary data from 783 phenotypes were filtered to exclude cases below 500 in the MR-PheWAS analysis⁴⁰.

We further explored candidate mediators, including brain IDPs and metabolites from blood, brain and CSF. To select candidate SNPs associated with metabolite traits as IVs, we conducted a rigorous filtering process at different genome-wide significance levels: a p-value threshold of less than 1E-05 for 3935 IDPs and 5E-08 for metabolites. To mitigate the effects of linkage disequilibrium, we set an r² < 0.001 threshold with a 10,000 kb clumping window and used the European population as a reference. This process ensured the selection of independent SNPs as IVs. We performed two-step MR to assess whether an intermediate factor had a mediating effect between the candidate proteins and PPD. The first step was to estimate the causal effect of the candidate proteins on the mediator (β1). The second step was to estimate the causal effect of each mediator on PPD (β2). The mediation proportion of each mediator in the association between candidate proteins and PPD was calculated as the product of β1 and β2 divided by the total effect of candidate proteins on PPD. The 95% CIs of the mediation proportions were calculated using the delta method.

Results

Identification of candidate proteins associated with PPD

Through proteome-wide MR analysis, we determined that SRA1 (q < 0.05) was associated with PPD in the discovery-stage two-sample MR analyses, along with 89 nominally significant proteins (p < 0.05) identified through the use of proteomics from deCODE (Fig. 1A). Notably, 19 proteins were consistently identified across both analyses (Fig. 1B), 17 of which showed directionally consistent effects on PPD risk. Among these 17 proteins, 11 were negatively associated with PPD, and 6 were positively associated with an increased risk of PPD (Fig. 1C). To identify robust causal candidates, colocalisation analysis was performed on all 17 directionally consistent proteins. SRA1, AGT, and PGP were colocalised with PPD (coloc.abf or coloc-SuSiE: PPH4 > 0.50) (Supplementary Figs. 1–3), confirming their direct genetic linkage. Among these proteins, SRA1 demonstrated the strongest colocalisation (coloc.abf: PPH4 = 0.91; coloc-SuSiE: PPH4 = 0.89). Specifically, SRA1, AGT, and PGP were positively associated with PPD, with combined OR estimates of 1.67 (1.32 to 2.12), 1.25 (1.05 to 1.49), and 1.24 (1.10 to 1.39), respectively. The detailed MR results for both datasets, including pleiotropy and heterogeneity and the colocalisation tests, are available in Supplementary Data 1–3.

Fig. 1. Proteome-wide MR results of the three identified proteins.

A Volcano plot of individual proteins across two data sources associated with the primary PPD outcome; the significant proteins (q < 0.05) and nominally significant proteins (p < 0.05) are shown. B Cross proteins in the two datasets are reported. C Forest plot of the identified proteins for the risk of PPD and colocalisation analyses. The ORs and 95% CIs of the cross proteins in the two datasets are reported. AGT angiotensinogen, AKR1A1 alcohol dehydrogenase [NADP(+)], C5 C5a anaphylatoxin, C8G complement component C8 gamma chain, CDON cell adhesion molecule-related/downregulated by oncogenes, CRNN cornulin, IL1R2 interleukin-1 receptor type 2, MANSC4 MANSC domain-containing protein 4, PGP glycerol-3-phosphate phosphatase, PIGR polymeric immunoglobulin receptor, PPD postpartum depression, PPH4 probability of hypothesis, ROR1 inactive tyrosine-protein kinase transmembrane receptor ROR1, SERPINA10 protein Z-dependent protease inhibitor, SRA1 steroid receptor RNA activator 1, STX7 syntaxin-7, TLR3 toll-like receptor 3, TIRAP toll/interleukin-1 receptor domain-containing adapter protein, TMEM106B transmembrane protein 106B.

Validation of protein associations with PPD using transcriptome-wide MR

The transcriptome-wide MR analysis, employing the SMR method, validated the identified associations, which was supported by significant evidence of gene expression from the eQTLGen datasets (p < 0.01, HEIDI test). Figure 2 presents the results of the SMR analysis for six genes, three of which were consistently aligned with proteins associated with PPD. SRA1 was linked to an increased risk of PPD in multiple blood or tissue-specific samples, including the brain (brain anterior cingulate cortex BA24, brain caudate basal ganglia, brain frontal cortex BA9, brain hippocampus, brain hypothalamus, brain putamen basal ganglia, and brain substantia nigra) and subcutaneous adipose tissue. The OR and 95% CI for these associations in the eQTLGen and GTEx datasets are detailed in Supplementary Data 4, reinforcing the evidence of a causal role for SRA1 in PPD.

Fig. 2. Transcriptome-wide MR results of the three identified proteins.

A Heatmap of identified protein-coding genes associated with PPD risk for the identified proteins. The colour represents the β estimators of the SMR analysis, where green represents a decreased PPD risk and red represents an increased PPD risk for per-SD increased gene expression. The missing values marked with “-” represent the genes without effective cis-eQTLs in the SMR analysis; “/” p > 0.05 in the SMR analysis; “×” represents failure in the HEIDI test; *p < 0.05; **multiple tests, p < 0.05/(3*15) (genes*tissues). B SMR analysis of the mQTLs, eQTLs and GWAS data. AGT angiotensinogen, CI confidence interval, OR odds ratio, PGP glycerol-3-phosphate phosphatase, PPD postpartum depression, SNP single-nucleotide polymorphism, SRA1 steroid receptor RNA activator 1.

Causal effect of SRA1-related DNA methylation on PPD

Our methylation MR analysis revealed that two CpG sites of SRA1 (cg01226231 and cg02434007) from blood and the brain were causally associated with the risk of PPD (p_SMR < 0.05, p_HEIDI > 0.01). For cg01226231 in blood for SRA1 expression, significant SMR and HEIDI test results were obtained (OR: 1.24, 95% CI: 1.20–1.28; p_SMR = 5.66E-34; p_HEIDI < 0.01). Using brain mQTLs available in the brain-mMeta mQTL dataset, we found that the association of cg02434007 with SRA1 expression was supported by the heterogeneity test results (OR: 1.11; 95% CI: 1.06–1.16; p_SMR = 7.13E-06; p_HEIDI > 0.01). In previous transcriptome MR analyses, we reported that SRA1 expression in multiple brain tissues is associated with increased PPD risk. In summary, methylation of cg02434007 of SRA1 in the brain is associated with increased expression of SRA1 and a high risk of PPD, which aligns with the positive effect of SRA1 gene expression on PPD risk across different tissues, including brain tissues (Fig. 2, Supplementary Data 5 and Supplementary Fig. 4).

Causal results of MDD symptoms and other neuropsychiatric disorders and MR-PheWAS

We found that increased SRA1 levels also increased the risk of MDD (OR: 1.12; 95% CI: 1.04–1.20; p = 0.002). With respect to the symptoms of MDD, increased levels of SRA1 were associated primarily with MDD (OR: 1.20; 95% CI: 1.08–1.35; p = 0.001), anhedonia (OR: 1.21; 95% CI: 1.09–1.34; p = 4.75E-04), and weight gain (OR: 1.24; 95% CI: 1.01–1.52; p = 0.040). For other neuropsychiatric disorders, elevated levels of SRA1 were also correlated with an increased risk of PTSD (OR: 1.01, 95% CI: 1.01–1.02; p = 0.001) (Fig. 3).

Fig. 3. Causal results of MDD symptoms and other neuropsychiatric disorders and MR-PheWAS.

A Causal results of MDD symptoms. B Causal results for other neuropsychiatric disorders. C MR-PheWAS analysis results for SRA1. ADHD attention-deficit/hyperactivity disorder, ASD autism spectrum disorder, BIP bipolar disorder, CI confidence interval, HEIDI heterogeneity in the dependent instrument test, MDD major depressive disorder, OCD obsessive-compulsive disorder, OR odds ratio, PTSD posttraumatic stress disorder, SCZ schizophrenia, SRA1 steroid receptor RNA activator 1.

The MR-PheWAS analysis revealed 38 phenotypes associated with genetically predicted SRA1 at nominal significance (IVW p < 0.05) (see Supplementary Data 6). Among these, a higher SRA1 index was linked to 29 phenotypes, including conditions such as cardiomyopathy, breast conditions, congenital or hormone-related disorders, and epilepsy, recurrent seizures, and convulsions. Although not statistically significant, a lower SRA1 index was found to be potentially associated with 9 phenotypes: other open wounds of the head and face; open wounds of the head, neck, and trunk; cancer of the oesophagus; viral enteritis; urinary incontinence; cholelithiasis with other cholecystitis; malignant neoplasms of the rectum, rectosigmoid junction, and anus; symptoms involving the nervous and musculoskeletal systems; and disturbances in skin sensation. These findings highlight the need for future investigations into potential adverse effects related to these conditions.

Associations between SRA1 and candidate mediating factors

To further explore the mediators of PPD due to increased levels of SRA1, we analysed the effects of 93 blood metabolites. We characterised 3 potential blood metabolites modified by SRA1 and 7 blood metabolites that affect PPD (Fig. 4). All the above detailed results are based on sensitivity analyses, which do not indicate heterogeneity or horizontal pleiotropy (Supplementary Data 7–9). Importantly, we found that SRA1 decreased isoleucine levels (β = −0.098 [95% CI: −0.136, −0.034], p = 0.003; p-FDR < 0.05), and higher isoleucine levels were associated with a decreased risk of PPD (β = −0.302 [95% CI: −0.516, −0.089], p = 0.005). In other words, SRA1 may increase the risk of PPD by reducing isoleucine levels, comprising 5.8% (p = 0.042) of the total effect. While alterations in SRA1 levels were associated with changes in 39 brain and CSF metabolites, only glutarylcarnitine exhibited marginal mediating effects between SRA1 and PPD (19.2%, p = 0.089) (Supplementary Data 10, 11 and Supplementary Fig. 5).

Fig. 4. Results of the MR-mediated analysis of SRA1, blood metabolites and PPD.

A Classification of 93 metabolites. B MR analysis of the effects of significant blood metabolites. C Mediation proportions of key blood metabolites. PPD postpartum depression, SRA1 steroid receptor RNA activator 1.

In addition to exploring the indirect effects of SRA1 on PPD through blood metabolites, we similarly investigated the possibility that SRA1 acts directly on the structure and function of the brain, leading to PPD. We ultimately confirmed 149 positive IDPs due to higher levels of SRA1, including 69 rfMRI connectivity IDPs, 30 regional and tissue volume IDPs, 9 cortical area IDPs, 7 WM tract ISOVF IDPs, 6 WM tract FA IDPs, 5 cortical thickness IDPs, 5 rfMRI node amplitude IDPs, 4 WM tract MO IDPs, 3 regional and tissue intensity IDPs, 3 WM tract diffusivity IDPs, 3 WM tract OD IDPs, 2 QC IDPs, 2 regional T2 IDPs, and 1 cortical grey‒white contrast IDP (Fig. 5 and Supplementary Data 12). Eleven of these IDPs are also found in PPD (Supplementary Data 13, 14). rfMRI connectivity (ICA100 edge 442) had a mediating effect (7.6%, p = 0.044) (Supplementary Fig. 6).

Fig. 5. Mediation MR analysis results for SRA1, IDPs and PPD.

A Classification of 3935 IDPs. B MR analysis of the effect of the significant IDPs. C Mediation proportion of the cross-key significant IDPs. D MR analysis of the effect of the cross-significant IDPs. IDPs imaging-derived phenotypes, PPD postpartum depression, SRA1 steroid receptor RNA activator 1.

ICA100 edge 442 is described in the UK Biobank as the resting-state functional magnetic resonance imaging (rfMRI) connectivity analysis results of rfMRI amplitudes for both ICA100 node 7 and ICA100 node 31. ICA100 node 7 primarily involves the medial prefrontal cortex (mPFC), posterior cingulate cortex (PCC), retrosplenial cortex (RSC), anterior temporal cortex (ATC), middle temporal gyrus (MTG) in the lateral temporal cortex, medial temporal lobe (MTL), and angular gyrus (AG) in the lateral parietal cortex of the default mode network (DMN)⁴². On the other hand, ICA100 node 31 primarily encompasses the right angular gyrus, right dorsolateral prefrontal cortex, and right frontal eye field from the right frontal parietal network (rFPN)⁴³.

Discussion

By integrating proteomic, transcriptomic, and epigenomic data, we identified SRA1 as a novel therapeutic target. The positive association between SRA1 expression and PPD risk highlights its potential as a biomarker and intervention point. SRA1 has emerged as a promising target in this context. In addition, through the two-step MR process, we further explored several potential mediators, which increased our understanding of the biological mechanism of SRA1 action on PPD.

It is important to note that SRA1 has dual molecular outputs, encoding both a conventional mRNA that is translated into a steroid receptor RNA activator protein (SRAP) and a long noncoding RNA (lncRNA) isoform arising from the same genomic locus^11,12. The present study primarily investigated the potential role of SRA1-encoded mRNA and its protein product in PPD susceptibility; however, the lncRNA isoform may also exert independent effects on PPD risk and warrants further dedicated investigation. Previous studies have shown that SRA1 is associated with metabolic dysregulation, including obesity and insulin resistance, as well as systemic inflammatory states^44,45, and these metabolic and inflammatory pathways have been proposed as potential contributors to the aetiology of postpartum depression across multiple studies⁴⁶. However, evidence from MR analyses suggests a more nuanced picture: while genetically predicted BMI has been causally linked to PPD risk, genetically predicted metabolic traits, including blood pressure, lipid profiles, glycaemic indices, and circulating CRP levels, showed no significant causal association with PPD, suggesting that these pathways may not function as primary aetiological drivers at the genetic level⁴⁷. Furthermore, SRA1 encodes a long noncoding RNA (lncRNA) that modulates nuclear receptor function, with potential roles in neuropsychiatric disorders. The influence of SRA1 on nuclear receptor pathways indicates a mechanism by which postpartum hormonal fluctuations could affect mood regulation. This is supported by extensive research linking SRA1 to both breast and prostate cancer, suggesting its role in regulating androgen and oestrogen pathways^48–50. These findings are consistent with earlier findings associating hormonal imbalances with PPD. Although direct investigations on the impact of SRA1 specifically on PPD are limited, emerging studies suggest that SRA1 can influence mood by regulating genes associated with depression and anxiety^51,52. For instance, research by Hofstra et al. revealed that SRA1 is a brain-specific eQTLs gene that is shared between Alzheimer’s disease and depression, highlighting its multifaceted role across neurological and psychiatric disorders⁵². By modulating steroid hormone pathways and downstream signalling, SRA1 may significantly affect mood regulation and neural plasticity.

DNA methylation, a crucial epigenetic modification, significantly affects gene expression, particularly in the context of neuropsychiatric conditions, where it is influenced by environmental factors such as traumatic stress hormones^53–55. Our MR methylation analysis revealed a significant association between specific CpG methylation sites within SRA1, including probes cg01226231 and cg02434007, and the risk of PPD, emphasising the importance of epigenetic regulation in mental health disorders. The cg01226231 probe is located in the N-Shore region of SRA1, a regulatory zone upstream of the CpG island known to influence promoter activity. In blood, cg01226231 methylation was positively associated with SRA1 expression; however, this signal did not pass the HEIDI heterogeneity test, indicating that the association between cg01226231 and SRA1 expression in blood is likely driven by pleiotropy or linkage disequilibrium with neighbouring variants rather than a shared causal mechanism. This site may therefore serve as an epigenetic marker of SRA1-related PPD risk rather than a direct functional effector. In contrast, cg02434007 is located in the S-Shore region of an SRA1-associated CpG island and maps to the 5’UTR and first exon of SRA1. In brain tissue, increased methylation at cg02434007 is strongly associated with elevated SRA1 expression. While methylation of CpG sites within the 5’UTR and first exon is canonically associated with transcriptional repression, methylation in S-Shore regions flanking CpG islands can paradoxically facilitate gene expression by disrupting the binding of transcriptional repressors or insulator elements, thereby relieving local silencing and permitting enhanced transcription^56,57. Consequently, hypermethylation at cg02434007 may derepress SRA1 transcription in the brain, leading to elevated SRA1 protein levels. This positions the methylation status of cg02434007 as a biologically plausible epigenetic biomarker for PPD risk and highlights the SRA1 regulatory axis as a candidate target for future epigenetic intervention strategies (Supplementary Fig. 4).

The two-step MR analysis conducted here highlights a significant association between isoleucine levels and PPD risk. Isoleucine, an essential amino acid, participates in various metabolic pathways vital for neurotransmission and metabolism⁵⁸. Previous studies have shown that isoleucine may be affected by gestational diabetes⁵⁹. Although direct studies linking isoleucine and PPD are scarce, our findings suggest that SRA1 might impact PPD risk by modulating isoleucine metabolism⁶⁰. SRA1 possibly influences the synthesis or breakdown of isoleucine, altering its concentration in neuronal tissues and thereby affecting mood regulation and neural network utility⁶¹. In our study, rfMRI connectivity (ICA100 edge 442) had a mediating effect. ICA100 node 7 (DMN) and ICA100 node 31 (rFPN) have been shown to be relevant to PPD or MDD in previous studies^62,63. The DMN encompasses internally focused, self-referential processing⁶⁴. Previous research has demonstrated abnormal neural activity in mothers with PPD compared with healthy mothers. This includes a significant reduction in cortico-cortical and cortico-limbic connectivity⁶⁵. In addition, the markedly disrupted connectivity between the posterior cingulate cortex (PCC, a key node of the DMN) and the amygdala raises the possibility that PPD may involve disturbances in outward, preventive aspects of self-referential thought, as well as in theory-of-mind and empathy processes⁶⁶.

The cognitive control functions of the brain areas located within the frontoparietal cortex have been reported⁶⁷. The most anterior part of the prefrontal cortex has been suggested as the apex of the executive network underlying decision-making⁶⁷. One study offered evidence indicating that individuals with PTSD and comorbid MDD exhibit heightened connectivity within the right frontoparietal network than individuals in mono-diagnostic groups and control groups do⁴³. The rFPN regulates and coordinates task-dependent switching⁶⁸ and has the greatest temporal variability in cognitive control networks⁶⁹; these findings suggest that PPD may be associated with abnormal executive function and increased symptom severity.

A study on the influence of sex hormones on the functional architecture of the human brain revealed that cortical network dynamics (particularly in the default mode, frontoparietal, and dorsal attention networks, whose hubs are densely populated with oestrogen receptors) are preceded and perhaps driven by hormonal fluctuations⁷⁰. A previous study in breast cancer patients revealed that different expression levels of hormone receptors, such as ERs or PRs, may be associated with cognitive impairment and provided evidence that the functional connectivity strength of the dorsolateral prefrontal cortex (a key structure in the FPN) may be selectively involved in cognitive impairment in the hormone receptor-negative group of breast cancer patients, especially with respect to subjective prospective memory⁷¹. It may be possible to explain the mechanism through which SRA1 influences PPD through the DMN and rFPN, but further research is still needed in the future.

The current FDA-approved pharmacological treatments for PPD, brexanolone and zuranolone, are synthetic analogues of allopregnanolone, the primary neuroactive metabolite of progesterone, which exerts its effects through positive allosteric modulation of GABA-A receptors, thereby enhancing GABAergic neurotransmission^3,72,73. This mechanistic framework raises the possibility that SRA1, as a coactivator of oestrogen receptors α and β and the progesterone receptor, may participate in the same neuroendocrine axis⁷⁴. Consistent with this hypothesis, the upregulation of SRA1 in murine ovarian granulosa cells has been shown to stimulate the secretion of oestradiol and progesterone in vitro⁷⁵. Whether SRA1 can exert effects analogous to those of progesterone through coactivation of the progesterone receptor and thereby modulate the neuroactive steroid signalling pathways implicated in PPD remains to be determined and represents a compelling direction for future functional genomics investigations.

Although there are currently no clinical trials specifically targeting SRA1, investigations into long noncoding RNAs as therapeutic targets are underway. Potential drug development could focus on modulating SRA1 expression or disrupting its interaction with downstream pathways. These insights not only enhance our understanding of the biological mechanisms underlying PPD but also lay the groundwork for subsequent studies aimed at therapeutic development.

While our study presents robust findings, several limitations warrant discussion. First, although the complexity of protein function is difficult to fully capture through genetically predicted protein levels, we have thoroughly integrated transcriptomic and DNA methylation data to strengthen our findings. Second, replication cohorts with diverse ethnic backgrounds are needed to validate these results. Future studies with individual-level data should examine whether SRA1-related pathways are differentially activated across pregnancies, particularly among women at elevated risk of PPD recurrence. Third, functional studies are needed to elucidate the precise mechanisms underlying the role of SRA1 in PPD. Although the results of the present study provide causal proteomic evidence through MR, the findings are derived entirely from summary-level genetic data and cannot capture the dynamic hormonal transitions of the peripartum period. Future work should therefore incorporate two complementary murine models: an oestrogen-exposure model, in which ovariectomised mice receive controlled 17β-oestradiol implantation to assess SRA1 expression during sustained hormonal elevation, and a hormone withdrawal (HW) model, which simulates abrupt postpartum endocrine decline. Single-cell RNA sequencing applied to the HW model would enable cell type-specific dissection of the transcriptional pathways through which SRA1 modulates the neuroactive steroid signalling implicated in PPD. Furthermore, the protein GWAS database and the transcriptomic and DNA methylation data utilised in this study were not stratified by sex-specific characteristics. While IDPs provide sex-specific GWAS data, the availability of valid instrumental variables is limited. Although studies have revealed that the signs of the effect sizes for genetic females and genetic males are consistently aligned, further validation is necessary. This limitation underscores the need for future GWAS and multiomics studies to incorporate sex-stratified analyses, given the critical influence of sex on the molecular mechanisms related to postpartum conditions. Additionally, it is important to note that the majority of the CSF and brain metabolite data in our study were derived from Alzheimer’s disease cohorts, as such data are exceedingly difficult to obtain from healthy individuals. It is also important to acknowledge the statistical constraints inherent to the current PPD GWAS landscape. Although we employed the largest available PPD GWAS to date, the phenotypic ascertainment of postpartum depression remains challenging, and the eligible study populations are relatively limited in size. Consequently, no single SNP has reached genome-wide significance in European or cross-ancestry meta-analyses, and the estimated SNP-based heritability of PPD is 0.14, indicating that common genetic variants collectively explain approximately 14% of the liability-scale variance. Against this backdrop, our proteome-wide screen revealed approximately 5000–7000 proteins, but only a single protein (SRA1) passed multiple-testing correction in both the discovery and replication datasets. These results are consistent with the modest statistical power available under current GWAS conditions and suggest that a larger number of biologically meaningful protein associations are likely present but remain below the threshold of detectability. Future studies leveraging substantially larger PPD GWAS datasets, as well as more comprehensive pQTLs resources with improved tissue specificity, will be essential for uncovering the broader proteomic architecture of PPD risk. Finally, not all the findings from the exploratory MR analyses passed multiple-testing correction. However, these exploratory results provide important insights that could guide future mechanistic studies. Despite these limitations, given the limited availability of therapeutic targets for PPD, any exploratory analyses remain highly valuable for advancing our understanding and treatment of this condition.

In conclusion, this study highlights SRA1 as a novel therapeutic target for PPD, offering insights into the molecular mechanisms underlying the disorder. The integration of proteomic, transcriptomic, and epigenomic data underscores the utility of multiomics approaches in psychiatric research. Future studies should focus on validating these findings and developing targeted interventions to mitigate PPD risk.

Author contributions

All authors contributed substantially to the study’s conceptualisation and design. M.C. and H.H. prepared the data. H.H. wrote the analytic code and carried out the analyses, with additional input from M.C., H.L. and X.X. The image of brain MR results was interpreted by J.M. and Y.L. H.H., H.R. and L.F. wrote the first draft of the manuscript. C.H. and Q.W. contributed to the critical revision of the manuscript.

Data availability

Publicly accessible summary statistics from GWAS were acquired from the websites (https://gwas.mrcieu.ac.uk/, https://www.ebi.ac.uk/gwas/summary-statistics, https://pgc.unc.edu/for-researchers/download-results/, https://www.decode.com/summarydata/, and https://www.finngen.fi/en/access_results).

Code availability

For further details or questions regarding the code, please contact the corresponding author.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44184-026-00232-3.