Postpartum depression (PPD) is one of the most prominent mood disorders affecting 15% of parturient women [1]. Its clinical symptoms typically occur within two weeks after delivery, with patients reporting persistent and severe moodiness, including negative emotions (e.g., anxiety, sadness, hopelessness and/or worthlessness), low energy, and social withdrawal [2]. PPD may result in both short-term and long-term negative effects on neonatal and infant development, cognition, emotion, and behavior. Previous studies have shown that maternal depression is a risk factor for higher rates of premature and low-birth-weight babies, infant malnutrition and stunting, and infant diarrhoeal, which can lead to mothers’ reluctance to care for their children and impair normal mother-child relationships. And it’s becoming a major public health problem [3].

PPD is the result of the combined influence of multiple factors. Previous studies have reported hypotheses regarding the physiopathologic mechanisms of PPD, such as the abnormal changes of hormones, neurotransmitters, inflammatory factors [4], among others. The molecular pathways implicated in the PPD was involved in genetic inheritance [5], HPA axis [6], serotonin transporter [7], and GABA receptor [8]. However, there is no exact evidence to support any of the above conclusions, and the physiopathologic mechanisms of PPD remain largely obscure.

Despite the severity of PPD on adverse infant and maternal outcomes, early diagnosis would help establish mechanisms for the prevention and early cure of PPD. Therefore, it is necessary to identify reliable diagnostic biomarkers of PPD. In practical work, it is difficult to acquire brain tissue for clinical testing. Peripheral blood mononuclear cells (PBMCs) share mRNA expression patterns in brain tissues [9]. In both PBMCs and the prefrontal cortex, the expression levels of many genes and biological processes are similar [10]. On the one hand, mounting evidence has indicated that gene expression changes in PBMCs may intervene molecular alterations in the region of brain [11]. On the other hand, gene expression of peripheral lymphocytes may be affected by the central nervous system (CNS) via neurotransmitters, cytokines, and hormones [12]. Therefore, PBMCs can serve as an effective tissue to search the gene expression in PPD.

With the development of high-throughput expression microarray technology, gene expression profiling has become a powerful tool for in-depth study of the pathogenesis and diagnostic biomarkers of mental disorders at the genomic level [13]. For example, microarrays have been used to detect differences in gene expression characteristics between healthy controls and PPD patients [14]. Many microarray-based studies have focused only on screening for differentially expressed genes (DEGs) without identifying their links. Genes with similar expression patterns may be functionally related. Weighted gene co-expression Network analysis (WGCNA) algorithm can divide genes into different modules according to the similarity of gene co-expression in multiple samples. This produces a set of genes that function similarly, allowing selection of disease-related biomarkers and pathways. In the present study, we, for the first time, used WGCNA to construct a co-expression network of genes and determined key modules and hub genes associated with PPD.

PPD is a major public health concern, affecting 10 to 15 percent of mothers worldwide [25]. Investigations are being performed to develop a non-invasive and quantitative clinical test to support PPD diagnosis, nevertheless, no specific or sensitive biomarkers in whole blood are currently available for the diagnosis of PPD [26, 27]. Thus, we performed several bioinformatic methods to identify the exact pathological mechanisms and effective diagnostic biomarkers of PPD. In the first part of our study, we determined the hub genes and pathways associated with PPD within whole blood samples. A total of strictly screened 673 DEGs ($|$LogFC$|$$>$ 0.5 was used as a cutoff, with the p value 0.05 regarded as statistically significant, 163 upregulated genes and 510 downregulated genes) and 30 hub genes were explored from the GSE45603 dataset. Next, we applied a new approach, WGCNA, to validate gene co-expression modules related to the molecular mechanisms underlying PPD. GEO2R identified 8979 genes which were analyzed to construct a co-expression network and generate a highly correlated co-expression gene group. There were three highly correlated modules (turquoise, yellow, and green modules) obtained from the WGCNA analysis. We regarded the top three hub genes (HnRNPA2B1, FBL, and SPI1), both in the PPI network and co-expression network, as “real” hub genes, which may function as important regulators in the pathogenesis of PPD.

Recently, it has been widely accepted that endocrine [28], immune response [29], and genetic and epigenetic factors [30], are all involved in the pathogenesis of PPD. Our study found that genes in the three modules identified by WGCNA (turquoise, green, and yellow modules) were mainly enriched in innate immune response, signal transduction, intracellular signal transduction, inflammatory response, and viral processes. The green module was enriched in the regulation of transcription (DNA-templated), gene expression, response to ionizing radiation, and RNA processing.

According to the cross-validation of data from the three modules and DEGs, eight genes acted as high degree genes. Among these eight genes, we were interested in the HnRNPA2B1, Fibrillarin (FBL), and Spi-1 proto-oncogene (SPI1). HnRNPA2B1, an RNA-binding protein and one member of the heterogeneous nuclear ribonucleoprotein family, initiates and amplifies the innate immune response. HnRNPA2B1 dimerization is required for nucleocytoplasmic translocation and initiation of IFN-$\alpha$/$\beta$ expression [31]. Furthermore, hnRNPA2B1 is also involved in the pathogenesis of several devastating neurodegenerative diseases. Mutations in prion-like domains of hnRNPA2B1 play a causal role in multisystem proteinopathy (MSP) and may be involved in the pathogenesis of amyotrophic lateral sclerosis (ALS) [32]. Fibrillarin (FBL), a molecular marker of transcriptionally active RNA polymerase I, catalyzes the 2’-O-methylation of ribosomal RNAs. FBL plays an important role in ribosome biogenesis, particularly in the methylation of ribosomal RNAs and rDNA histones [33, 34]. Recent work has indicated that FBL may be an important component of ribosome size, life duration in multicellular organisms, and ribosomal RNA production, all of which are correlated with age in healthy humans [35, 36]. Spi-1 proto-oncogene (SPI1), the E26 transformation-specific (ETS) family transcription factor PU.1, serves as a master regulator of myeloid and lymphoid development and is primarily expressed in monocytes/macrophages, neutrophils, mast cells, B cells, and early erythroblasts [37, 38]. The function of SPI1 is closely related to microglial viability and phagocytic capability. The targets of SPI1 demonstrate a significant relationship with the pathways defined as “Endocytosis”, “Fc$\gamma$ receptor-mediated phagocytosis”, “Chemokine signaling pathway”, and “MAPK signaling pathway” [38, 39]. UBA52, a ubiquitin-ribosomal fusion gene and located in chromosome 19, is a major source of ubiquitin protein for covalent modification of proteinaceous substrates recycled by ubiquitin-proteasome system (UPS) [40]. UBA 52 encodes a fusion protein comprising ubiquitin at the N-terminus and RPL40 at the C-terminus [41]. Upon translation, ubiquitin and RPL40 are immediately cleaved from the translated product [42]. RAD51 has the function of discovering and invading homologous DNA sequences, and can be repaired accurately and timely [43]. NHP2 gene is a member of the H/ACA snoRNPs (small nucleolar ribonucleoproteins) gene family. snoRNPs are involved in various aspects of rRNA processing and modification [44]. RPL13A encodes a member of the ribosomal protein L13P family and is a component of the 60S subunit. The protein encoded as a component of the IFN-$\gamma$ activated translational inhibitor (gait) complex also plays a role in inhibiting inflammatory genes [45]. However, the functions of RAD51, NHP2, and RPL13A in PPD did not study in detail. Interestingly, the pathway enrichment analysis indicated that genes in the three modules were enriched in “neurotrophin signaling pathway” and “chemokine signaling pathway”, which partly coincided with SPI1 targets. The neurotrophin signaling pathway includes several neurotrophic factors, largely a family of secreted proteins that benefit in the growth, survival, and differentiation of neurons. Due to its beneficial and therapeutic effects on the neuronal physiology, researchers have intensively studied neurotrophic factors for decades [43]. Recently, neurotrophic factors have been considered as having significant links with the pathophysiology of neurological and neuropsychiatric disorders. Emerging evidence has demonstrated that brain-derived neurotrophic factor (BDNF), a well-known neurotrophic factor, is involved in PPD and its treatments. Clinical studies have observed a link between PPD symptoms and single nucleotide polymorphisms (SNPs) in BDNF [44]. Postpartum female mice induced by chronic unpredictable stress exhibited low expression of BDNF mRNA in the mPFC region [45]. In the serum of patients with PPD at admission and during development, BDNF levels were lower than healthy individuals. Fluoxetine, a classic treatment for women with PPD, may mediate its antidepressant effect in PPD by upregulating BDNF expression [46]. Second, to protect the fetus against pathogens and receive the semi-allogenic markers as the fetus, the immune system of the mother would go through a tremendous adaptation during pregnancy [47]. Inflammatory cytokine levels may obviously increase in the periphery during cervical ripening [48]. Furthermore, since the mother’s immune system recovers to the previous non-pregnant state after delivery, subsequently, the change in inflammatory cytokine levels may affect mood [49]. Brewster et al. [50] reported that there was a close relationship between chemokines and PPD, which is manifested by increased levels of pro-inflammatory cytokines, such as IL-18 and C-X-C motif chemokine 1 (CXCL1). PPD was positively correlated with multiple genes in the immune response of an immobilization stress-induced mouse PPD model. The neurotrophin signaling pathway and chemokine signaling pathway may be deeply involved in the mechanisms of PPD; however, it remains a major challenge that required continued exploration.

There were two previous studies to find biomarkers of PPD based on GSE45603 dataset. Lauren et al. [51] reported that the objective of this study was to independently replicate their previously published prediction model of postpartum depression and women without a history of psychiatric disorders, and to further investigate the DNA methylation status of postpartum depression biomarkers associated with changes in pregnancy hormone levels and the timing of major hormonal changes. Maria et al. [52] focused on whether abnormal patterns of circulating levels of cytokines and chemokines may offer a suitable biomarker for disease development and/or therapeutic response. The objective of the present study and the previous two study were both searching for biomarkers of PPD. However, there were several differences among these studies. Our present study only observed the postpartum mental state and healthy controls. Lauren et al. collected different gestation process samples and used their own model to identify biomarker, their key point was changes in hormone levels. The emphasis of Maria’ study was chemokines levels. Our present study did not classify genes, we put all DEGs into WGCNA and obtained hub genes, and the functions of genes were involved in various aspects of cell.

There were several limitations to our present study. First, in order to comprehensively determine dysfunctions related to PPD, both brain tissue and blood samples would need to be integrated. However, our present study did not perform an analysis of brain tissue. Second, only one microarray was included in the computerized analysis, and likely leading to one-sided results and a high false-positive rate. Third, our present study only performed data mining and data analysis and did not conduct any experiments to validate the results. Finally, there were only 5 healthy control samples and 16 PPD samples selected for analysis; thus, it is necessary to increase the sample size to validate the accuracy of the results in a follow-up study.

Based on our knowledge, we for the first time used the system biology-based WGCNA method to predict several potential biological pathways and diagnostic biomarkers involved in PPD. WGCNA and co-expression network analysis identified key biological processes and signaling pathways, especially the neurotrophin signaling pathway, chemokine signaling pathway, Fc$\gamma$ receptor-mediated phagocytosis, and the MAPK signaling pathway, which may contribute to the elucidation of the pathogenesis and progression of PPD. The potential diagnostic biomarkers included HNRNPA2B1, IL10, RAD51, UBA52, NHP2, RPL13A, and FBL. Finally, in-depth molecular biological experiments are required to determine the exact functions of the biological pathways and diagnostic biomarkers of PPD.

GW designed this study. GW and XH provided administrative support. DZ and CW collected the data, writed the manuscript. LJ, DA, YY, TJ, and YC analyzed the data. All the authors approved the final version of manuscript.

Not applicable.

Not applicable.

This study was supported by grants from the Wuhan Clinical Medical Research Project (WX20C15).

The authors declare no conflict of interest.