A discovery cohort of three women with PPD and three control participants, along with a validation cohort (n = 50 with PPD, n = 50 controls), was recruited from women who attended a 42-day postpartum follow-up visit at Zhongda Hospital and Nanjing Women and Children’s Healthcare Hospital between June 2024 and December 2024. All participants completed the psychological questionnaire, including the Edinburgh Postnatal Depression Scale (EPDS), Social Support Rating Scale (SSRS), Pittsburgh Sleep Quality Index (PSQI), and sociodemographic questionnaires under the guidance and supervision of a trained physician, following predefined inclusion and exclusion criteria. The control group was defined by EPDS scores ($\le$4), while the PPD group was defined by EPDS scores ($\ge$13) and the clinical structured interview based on the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), respectively. All participants provided written informed consent prior to the commencement of the study.

Participant selection criteria were as follows: (1) age from 18 years to 45 years, who received routine prenatal care, delivery, and postpartum follow-up at the participating hospital; (2) no serious cardiovascular, neurological, pulmonary, or renal diseases; (3) currently breastfeeding and not taking any medication; and (4) submission of a signed informed consent form. Participant exclusion criteria were: (1) twin pregnancy; (2) pre-existing diabetes before conception; (3) pre-pregnancy hypertensive disorders and preeclampsia; (4) hyperthyroidism; (5) history of miscarriage or stillbirth; (6) clinical diagnosis of schizophrenia and bipolar disorder; (7) family history of mental illness; (8) lack of clinical information; and (9) missing questionnaire information and incomplete clinical information.

Venous blood samples (2 mL) were collected from the antecubital vein of each participant in the early morning following an overnight fast, using K2EDTA-coated vacuum blood collection tubes (ST750EK, Streck, NE, USA). After gentle inversion, samples underwent centrifugation (1000 $\times$g, 15 min, 4 °C). The resulting plasma supernatant was then aliquoted into nuclease-free vials for cryopreservation at –80 °C until analysis.

For circRNA microarray analysis, sample preparation and microarray hybridization were performed following the Arraystar microarray protocols (AS-S-CR-M-1.0 and ASLNC-MV3.0, Arraystar, TX, USA). To enrich circRNAs, total RNA samples were digested with RNase R (RNR07250, Epicentre, WI, USA) to remove linear RNAs. The enriched circRNAs were then amplified and transcribed into fluorescent circRNA utilizing a random priming method (Arraystar Super RNA Labeling Kit; AS-MS-0250, Arraystar, TX, USA). Labeled circRNAs were hybridized onto the Arraystar Human circRNA Array V2 (8x15K, Arraystar, TX, USA). After washing, the slides were scanned using the Agilent Scanner G2505C (G2505C, Agilent Technologies, CA, USA), and array images were analyzed with Agilent Feature Extraction software (version 11.0.1.1, Agilent Technologies, CA, USA).

Validation of differentially expressed circRNAs: 22 circRNAs identified as differentially expressed in the microarray analysis were selected for validation. Among them, 5 circRNAs (circDYM, circHIPK2, circZBTB25, circSTAG1, circATF7IP) that had previously been reported to be closely associated with MDD are included for validation using reverse transcription-quantitative polymerase chain reaction (RT-qPCR). In the validation cohort, total RNA was extracted from plasma using the miRNeasy® Serum/Plasma Kit (cat. No. 217184, Qiagen, MD, USA) according to the manufacturer’s instructions. RNA concentrations were detected using the One Drop spectrophotometer (Thermo Fisher Scientific, MA, USA). Reverse transcription was performed using the HiScript III All-in-one RT SuperMix Perfect for qPCR Kit (R333-C1, Vazyme, Nanjing, Jiangsu, China). The reaction system was configured, and amplification was carried out using SYBR Green Premix Pro Taq HS qPCR Tracking Kit (AG11733, Accurate Biology, Changsha, Hunan, China) following the manufacturer’s protocol, with cDNA as the template. Relative gene expression levels were calculated using the 2^{-Δ⁢Δ⁢CT} method. Controls were normalized for each experiment. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as an internal control. We employed the “reverse primer approach” by designing upstream and downstream primers on either side of the circRNA-specific antisense splicing site, allowing the primers to extend outward across the junction. Simultaneously, we verified the single peak in the melting curve to ensure primer specificity. Primers were synthesized by Geneary (Shanghai, China), and the primer sequences are shown in Supplementary Table 1.

SPSS version 26.0 (International Business Machines Corporation, Armonk, NY, USA) and GraphPad Prism version 8.0.2 (GraphPad Software, Boston, MA, USA) were used for data analysis. The normality of continuous variables was assessed using the Shapiro-Wilk test, while the categorical variables were analyzed using the chi-square test. For continuous variables, an independent samples t-test was employed when the data were normally distributed; otherwise, the Mann-Whitney U-test was used. Spearman’s rank correlation was utilized to examine linear relationships between levels of circRNAs and severity of PPD, degree of social support, and sleep condition. To adjust for potential confounders, both univariate and multivariate logistic regression models were employed, reporting odds ratios (ORs) with corresponding 95% confidence intervals (CIs). The diagnostic efficacy of candidate circRNAs for PPD was assessed by the area under the curve (AUC) of the receiver operating characteristic (ROC), with the Youden index determining the optimal sensitivity and specificity cutoff. Statistical significance was defined as a two-tailed p-value 0.05.

In the discovery cohort, the mean ages of the PPD and control groups were 30.33 $\pm$ 5.86 and 27.67 $\pm$ 1.16 years, respectively. The average gestational age at delivery was 38.76 $\pm$ 1.00 weeks for the PPD group and 39.62 $\pm$ 0.36 weeks for the control group. The postpartum body mass index (BMI) was 23.77 $\pm$ 2.24 and 21.83 $\pm$ 0.50, respectively. All participants were Han Chinese, married, and had no history of anxiety or depression. Additional sociodemographic data, clinical data, and laboratory findings are summarized in Supplementary Table 2. The validation cohort consisted of 50 individuals with PPD and 50 controls. Univariate analysis identified statistically significant differences between the two groups in several factors, including natural conception, spontaneous labor, history of anxiety and depression, exercise during pregnancy, white blood cell (WBC) levels, thyroid-stimulating hormone (TSH) levels, SSRS scores, PSQI scores, and self-injurious ideation (p 0.05) (Table 1). Univariate logistic analysis identified the following risk factors for PPD: a history of pre-pregnancy anxiety or depression (OR = 14.75, 95% CI: 1.67–1940.94, p = 0.01), assisted reproduction (OR = 3.92, 95% CI: 1.01–15.22, p 0.05), postpartum WBC levels (OR = 1.44, 95% CI: 1.05–1.98, p = 0.03), gestational TSH levels (OR = 1.51, 95% CI: 1.00–2.27, p 0.05), and PSQI scores suggesting sleep disturbance (OR = 2.21, 95% CI: 1.62–3.03, p 0.001). Protective factors included cesarean section (OR = 0.44, 95% CI: 0.20–0.99, p 0.05), exercise during pregnancy (OR = 0.22, 95% CI: 0.07–0.74, p = 0.01), and SSRS score suggestive of good social support (OR = 0.74, 95% CI: 0.66–0.83, p 0.001) (Table 2).

CircRNA microarrays were used to compare plasma circRNA levels between PPD patients and matched controls in the discovery cohort. Among the 7683 circRNAs detected by microarray profiling (Fig. 1A–C), 22 were found to be differentially expressed between the PPD group and the control group ($|$Fold Change$|$ $>$1.20, p 0.05), including 12 upregulated and 10 downregulated circRNAs. Since primers could not be designed for two circRNAs, only 20 differentially expressed circRNAs were validated in the validation cohort ($|$Fold Change$|$ $>$1.2 and p 0.05). The complete microarray screening results and the corresponding RT-qPCR findings from the validation cohort are presented in Supplementary Table 3. Among the differentially expressed circRNAs, circIFNGR2 (p 0.05) and circABCC5 (p = 0.01) exhibited significantly increased expression levels in the PPD group, consistent with the microarray results. CircDYM (p 0.01) had decreased expression levels in the PPD group, and circATF7IP (p = 0.01) had elevated expression levels, consistent with previously reported expression trends in MDD-related studies (Fig. 1D–G).

Fig. 1.

Screening and validation of differentially expressed circular RNAs. (A) Experimental flowchart. (B,C) Volcano plot and heatmap of microarray analysis results. (D–G) Differentially expressed circRNAs verified by RT-qPCR in the validation cohort. (D) Relative expression levels of circIFNGR2. (E) Relative expression levels of circABCC5. (F) Relative expression levels of circDYM. (G) Relative expression levels of circATF7IP. Note: * p 0.05, ** p 0.01. Abbreviations: Con, control group; FC, fold change; RT-qPCR, reverse transcription-quantitative polymerase chain reaction; MDD, major depressive disorder; circRNA, circular RNA; circIFNGR2, circRNA derived from the Interferon Gamma Receptor 2 gene; circABCC5, circRNA derived from the ATP-Binding Cassette Subfamily C Member 5 gene; circDYM, circRNA derived from the Dystrophia Myotonica Protein Kinase gene; circATF7IP, circRNA derived from the Activating Transcription Factor 7 Interacting Protein gene.

Spearman correlation analysis demonstrated that circIFNGR2 was not correlated with EPDS scores (r = 0.123, p = 0.224), SSRS scores (r = –0.158, p = 0.117), or PSQI scores (r = 0.152, p = 0.132). CircABCC5 was not correlated with EPDS scores (r = 0.192, p = 0.056) or SSRS scores (r = –0.095, p = 0.350) but was positively associated with PSQI scores (r = 0.199, p = 0.047). CircDYM was not significantly correlated with EPDS scores (r = –0.155, p = 0.123) but was positively associated with SSRS scores (r = 0.219, p = 0.029) and negatively associated with PSQI scores (r = –0.248, p = 0.013). CircATF7IP was positively correlated with EPDS scores (r = 0.236, p = 0.018) and PSQI scores (r = 0.337, p = 0.0006) and negatively associated with SSRS scores (r = –0.306, p = 0.002) (Fig. 2). Further studies indicate that only circABCC5 level shows a positive correlation with WBC count (r = 0.205, p = 0.041) (Supplementary Fig. 1).

Fig. 2.

Correlation between plasma circIFNGR2, circABCC5, circDYM and circATF7IP levels and questionnaire assessments in participants. (A,D,G,J) Correlation between four differentially expressed circular RNAs and EPDS scores. (B,E,H,K) Correlation between four differentially expressed circular RNAs and SSRS scores. (C,F,I,L) Correlation between four differentially expressed circular RNAs and PSQI scores. Abbreviations: EPDS, Edinburgh Postnatal Depression Scale.

CircIFNGR2 showed an AUC of 0.62 (95% CI: 0.51–0.73, p 0.05), with a sensitivity of 28% and a specificity of 96%. CircABCC5 exhibited an AUC of 0.64 (95% CI: 0.53–0.75, p = 0.01), with a sensitivity of 36% and a specificity of 96%. CircDYM showed an AUC of 0.69 (95% CI: 0.58–0.79, p = 0.01), with a sensitivity of 54% and specificity of 84%. Lastly, CircATF7IP showed an AUC of 0.65 (95% CI: 0.54–0.76, p = 0.01), with a sensitivity of 62% sensitivity and 72% specificity. The combined model incorporating circIFNGR2, circABCC5, circDYM, and circATF7IP yielded an AUC of 0.80 (95% CI: 0.71–0.88, p 0.0001), with a sensitivity of 78% and a specificity of 70% (Fig. 3A,B, Supplementary Table 4). Multivariate logistic regression analysis adjusting for assisted reproduction, cesarean section delivery, history of anxiety and depression, exercise during pregnancy, WBC levels, and TSH levels identified that circIFNGR2 (adjusted OR = 2.26, 95% CI: 1.21–4.23, p = 0.01), circABCC5 (adjusted OR = 2.09, 95% CI: 1.35–3.23, p 0.01), and circATF7IP (adjusted OR = 1.74, 95% CI: 1.01–3.02, p 0.05) as independent predictive factors for PPD (Table 3). The combined model incorporating circIFNGR2, circABCC5, and circATF7IP yielded an AUC of 0.75 (95% CI: 0.65–0.85, p 0.0001), with a sensitivity of 66% and a specificity of 80%. The AUC for the combination of circIFNGR2 and circABCC5 was 0.67 (95% CI: 0.58–0.79, p 0.01), with a sensitivity of 54% and a specificity of 82%. The AUC for the combination of circIFNGR2 and circATF7IP was 0.69 (95% CI: 0.59–0.79, p 0.01), with a sensitivity of 52% and a specificity of 80%. The AUC for the combination of circABCC5 and circATF7IP was 0.72 (95% CI: 0.62–0.82, p 0.0001), with a sensitivity of 60% and a specificity of 76% (Fig. 3C–F, Supplementary Table 4).

Fig. 3.

AUC for ROC analysis of circRNAs that distinguish individuals with PPD symptoms from a normal control population. (A) AUC for the independent prediction of PPD by 4 circRNAs. (B) AUC for the combined prediction of PPD by 4 circRNAs. (C) AUC for the combined prediction of PPD by 3 circRNAs identified as relatively independent predictive factors. (D–F) AUC for the combined prediction of PPD by 2 circRNAs in each combination. Abbreviations: AUC, area under the curve; ROC, receiver operating characteristic.

Several limitations should be considered when interpreting our findings. First, our study included a small sample size. Due to the low incidence of PPD and challenges in sample collection, a 3-to-3 sample ratio was adopted for microarray analysis in the discovery cohort. Although the included samples were sufficiently representative, including individuals with severe depressive symptoms and self-harm tendencies who subsequently required hospitalization, the small sample size still poses a risk of false positives. Given the exploratory nature of this study, we validated as many relevant circRNAs as possible in the validation cohort, including those associated with MDD. The generalizability and reliability of these findings still require validation in larger, multicenter, multiethnic cohorts. Second, the dynamic changes of circRNA levels during the course of the disease cannot be fully captured by the single-timepoint sampling approach used in this study. Future research should adopt longitudinal sampling strategies to monitor circRNA expression profiles at multiple stages of pregnancy and the postpartum period. Lastly, the specific pathophysiologic mechanisms through which different circRNAs contribute to PPD require additional exploration.