2.1 The Continuous Darkness-Exposed Rat Model Evolved PCOS-Like Phenotypes

Our previously studies revealed that 8-week darkness exposure to rats resulted in PCOS-like phenotypes, including hyperandrogenism, disturbances in the estrous cycle, and polycystic ovarian [11, 12], which prompted us to investigate when individual phenotypes emerged and how they dynamically changed.

To elucidate the longitudinal alterations in the continuous darkness model, 6-week-old female SD rats were randomly divided into two groups after 1 week of acclimatization: control group, exposed to normal 12:12 h light–dark cycle, and dark group, exposed to continuous darkness. At weeks 0, 2, 4, 6, and 8, the rats were sacrificed for observation and sampling (Figure 1A). No changes were observed in their ovarian mass or body weight during the 8-week period (Figure 1B,C). Next, the appearance of reproductive endocrine alterations was evaluated. As expected, the dark group exhibited major phenotypes in the PCOS diagnostic criteria, including irregular estrous cycles (Figure 1D), polycystic ovary (Figure 1E), and hyperandrogenism (Figure 1F). Notably, the testosterone levels of the dark group increased in a time-dependent manner, with a significant difference at week 6 and a further increase at week 8 (Figure 1F). Although there was no significant change in sex hormone-binding globulin (SHBG) levels at any point within 8 weeks (Figure 1G), it was postulated that the free androgen index (FAI) was higher at weeks 2 and 8 in the dark than in the control group (Figure 1H). Furthermore, luteinizing hormone (LH) levels significantly increased at weeks 6 and 8 whereas the follicle-stimulating hormone (FSH) levels remained unchanged and the LH/FSH ratio significantly increased (Figure 1I–K). The dark group also exhibited impaired inflammation and lipid metabolism at week 8 (Figure S1). These findings indicate that phenotypes in rats with prolonged darkness exposure emerge at different time points, implying the existence of various epigenetic regulatory mechanisms governing acute response and chronic adaptation.

2.2 Continuous Darkness Exposure Leads to Dynamic Methylomic and Transcriptomic Profiles Changes in Ovarian GCs

To identify the epigenetic regulatory mechanism responsible for the dynamic phenotypes at different time points of continuous darkness, SureSelectXT rat DNA methylation target-captured bisulfite sequencing (MC-seq; Agilent Technologies, CA, USA) was employed, and the DNA methylation of rat ovarian GCs collected at five time points spanning 8 weeks was examined (Figure 1A). Genome-wide DNA methylation profiles indicated that 15–85% of tested CpG sites were methylated (binomial test, q-value < 0.05), consistent with previous findings [21]. Global DNA methylation showed only mild overall changes over time (Figure 2A). Subsequently, each set of standardized DNA methylation data was aligned to a gene model that includes promoting/transcribing starting points (TSSs) and transcribing ending points. The methylation patterns of the 10 groups were homologous. And there was a sharp dip in methylation levels 1 kb upstream of the TSS, followed by a nadir upstream of the TSS and high methylation levels throughout the gene body region (Figures 2B and S2A,B). This indicates that DNA methylation profiles were generally longitudinally stable and that 8-week continuous darkness exposure would not cause significant disturbance to DNA methylation.

To identify the dynamic patterns of longitudinal methylation profiles, we focused on scrutinizing the methylation levels in gene promoter regions via short time-series expression miner (STEM) analysis. The results indicated four clustering modules in the dark group (p < 0.05) (Figure 2C). The first module was progressively hypomethylated over time, which exhibited a similar methylation profile to patients with PCOS [13] and people working in night shift [22]. Subsequently, we performed Gene Ontology (GO) enrichment, Kyoto Encyclopedia of Genes and Genomes (KEGG), and Reactome Pathway analyses to predict the function and pathway profile of this module with hypomethylation trends. The genes were mainly enriched in immune inflammation and the AKT and metabolic pathways (Figure 2D). The STRING protein interaction network (http://string-db.org/) demonstrated that persistently hypomethylated genes were mostly enriched in the immune system, encompassing antigen presentation, ubiquitination modifications, and inflammatory factors (Figure S2C–G). To investigate the association between differentially methylated sites (DMSs) and each phenotype, weighted correlation network analysis was conducted to obtain three modules (MEdarkorange, MEgrey60, and MEblack) exhibiting significant and negative associations with testosterone, FAI, LH, leptin, and interleukin 1 beta (IL-1b). Furthermore, KEGG analysis revealed differentially expressed genes (DEGs) within the modules enriched in metabolism-related pathways (Figure S2H–J). These data suggest that changes caused by darkness are enriched in neural, metabolic, and inflammatory pathways.

Although the average differences of methylation sites across all time points were small, 386–511 differentially methylated regions (DMRs) and 80,926–86,927 DMSs were identified between each dark and control group pair (Figure S3A,B). These may hold the key to driving disease phenotypes. Dynamic volcano plots revealed a significant increase in DMSs at week 6 and a significant elevation in hypomethylated sites in the dark group (Figure 2E). Notably, the highest number of DMRs and DMSs in promoter TSS regions was still observed in week 6 (Figure S3C,D). We herein demonstrate the changes in the CpG sites in the TSS region of the genes Igf2r and timeless over the time course. Although the methylation changes varied from gene to gene, their differences expanded at week 6 (Figure S3E,F). Principal component analysis revealed similar phenomenon that rats at week 6 were better clustered according to the CpG site methylation (Figure 2F). As the most notable methylation variations occur during week 6, it will be of great interest to investigate the progressive alteration-induced expression changes at this point.

We remodeled rats in continuous darkness for 6 weeks to extract RNA from ovarian GCs for RNA sequencing. Compared with the control group, the dark group had similar body weights (Figure S4A), notable abnormalities in testosterone levels (Figure S4B), ovarian histological sections (Figure S4C), and glucose metabolism (Figure S4D,E). A total of 378 DEGs were identified between the dark and control groups, including 90 and 288 up- and downregulated genes, respectively (Figure 2G). Functional enrichment and clustering analyses of the DEGs revealed enrichment of terms related to lipid metabolism and neural pathways (Figure 2H). This finding is consistent with the results of the enrichment analysis of DNA methylation changes induced by 6-week darkness exposure (Figure 2I–K).

2.3 Darkness-Induced Serpine1 CpGs Hypomethylation Near TSS is Highly Correlated with Serum Testosterone Levels

To further explore the association between this epigenetic modification and expression changes induced by 6-week darkness exposure, we conducted an integrative analysis between 2818 DEGs and 4272 promoter DMSs. Notably, of the four quadrants, quadrant II showed 108 genes with reduced promoter CpG methylation and higher expression (Figure 3A). Venn analysis was conducted between genes in quadrant II and those with persistent hypomethylation in STEM-D1, identifying 16 overlapping genes. These genes maintained sustained hypomethylation levels while being overexpressed during the 6-week darkness exposure (Figure 3B). Notably, four candidate genes, Serpine1 [17], Lep [23], IL-2 [24], and Agt [25], encoding secretory proteins and having evident associations with inflammation and lipid metabolism were clustered in the STRING network (Figure 3C). The heat map showed a gradual trend of decreasing methylation in the promoter region of these four genes over time (Figure 3D). Interestingly, significantly negative correlations were observed between the promoter region methylation of Serpine1 and both serum testosterone and FAI (Figure 3E), suggesting a potential association between Serpine1 methylation and androgens. These findings suggest that hypomethylation contributes to Serpine1 overexpression and, hence, hyperandrogenism.

Several studies have reported inverse associations between DNA methylation and gene transcription levels in SERPINE1 [26]. However, research on specific Serpine1 CpGs that can induce transcriptional changes is lacking. We validated Serpine1 overexpression over 8 weeks in the GCs via qPCR (Figure 3F). Furthermore, we confirmed hypomethylation of the CpG (position 2) observed in MC-seq and nearby CpGs from the Serpine1 promoter region via pyrosequencing (Figure 3G,O), followed by a gradual increase in serum PAI-1 levels via ELISA (Figure 3H). This finding was also confirmed in the liver (Figure 3I,J), the main organ involved in Serpine1 synthesis [27], as well as in adipose tissues (Figure S5A) and brown adipose tissues (Figure S5B). These data indicate that hypomethylation and high Serpine1 expression are universal darkness-induced phenomena, suggesting that DNA methylation plays a pivotal role in the regulation of Serpine1 expression.

To further validate this methylation modification switch-based gene activation, we treated human KGN cell line cells with 5-Aza-2′-deoxycytidine (5-Aza), a DNA methyltransferase inhibitor that binds to DNA and inhibits DNMT activity and expression [27]. qPCR and Western blotting revealed that 5-Aza promoted SERPINE1 expression in a dose- and time-dependent manner (Figure 3,3). Similar results were observed in primary rat GCs (Figures 3M,N and S5C).

Next, we used a targeted DNA methylation editing tool, dCas9–Dnmt3aCD, for epigenetic alterations at specific CpG sites to infer a cause–effect relationship. Specific guide RNAs (sgRNA) were designed to bring the dCas9–DNMT3aCD system to the sequence located near the TSS of the first exon of Serpine1, including positions 1, 2, and 3 in the rat normal liver cell line BRL-3A (Figures 3O and S5D). Pyrosequencing revealed that dCas9–Dnmt3aCD/gRNA transduction significantly increased DNA methylation at positions 1, 2, and 3 and at adjacent position 4 but not at distant positions 7 and 8 (Figure 3P). DNA methylation of long interspersed nuclear element 1, which reflects global DNA methylation in cells, did not significantly differ among the sgRNA treatments and controls (Figure 3Q). Chromatin-immunoprecipitation results also validated the specific binding of the DNMT3A protein to the target promoter region (Figure S5E). These results indicated successful on-target methylation. Subsequently, qPCR results showed a significant reduction in Serpine1 transcripts in targeted methylated cells (Figure 3R). These results suggest that Serpine1 expression was efficiently decreased by methylation near the TSS.

2.4 SERPINE1 Inhibits Aromatase Expression and Promotes Androgen Accumulation Through the PI3K/AKT Pathway

Since multiomics sequencing results indicated a potential correlation between Serpine1 and testosterone, we conducted Pearson's analysis on serum SERPINE1 (PAI-1) and testosterone and found a significantly positive correlation (Figure 4A).

Therefore, investigating the function of SERPINE1 in the androgen metabolic pathway in GCs is valuable. First, we conducted an indirect assay for aromatase activity by detecting estrogen levels in cell supernatants after adding androgen, and found a significant decrease in aromatase activity after SERPINE1 overexpression and vice versa after knockdown in either KGN or primary rat GCs (Figure 4B,C). Second, we examined the expression of several key enzymes involved in the androgen metabolic pathway. CYP19A1 expression significantly decreased after SERPINE1 overexpression and vice versa after knockdown (Figures 4D–K and S6). Third, we used an activity assay kit to directly detect aromatase activity and the results were consistent with those with the indirect method (Figure 4L,M). Using immunofluorescence, we verified that SERPINE1 inhibited CYP19A1 expression, and identified that SERPINE1 was mainly localized in the cytoplasm and extracellular matrix; instead, aromatase was distributed in the nucleus and cytoplasm, with more in the cytoplasm (Figure 4N,O). Accordingly, we hypothesized that SERPINE1 in the cytoplasm and extracellular matrix might transmit repressive signals through specific signaling pathways, thereby inhibiting CYP19A1 transcription and expression.

To further validate the hypothesis, we interfered KGN cells with two SERPINE1 inhibitors, Loureirin B (LrB) and tiplaxtinin (TPX). qPCR results indicated a 10-fold sharp increase in CYP19A1 expression after 72 h of LrB treatment (Figure 5A) or a threefold increase with TPX (Figure 5B). Western blotting results supported this phenomenon, with a decreasing level of PAI-1 and an abrupt surge of aromatase expression at 72 h (Figure 5C,D). Cell Count Kit-8 experiments showed that the inhibitors had no effects on cell proliferation during this process (Figure S7A). Then, we used Phospho Explorer Antibody microarray (PEX100) to analyze signaling pathways protein changes at 0 and 15 min of TPX treatment. Differentially phosphorylated proteins were divided into three clusters according to the following pathways: AKT, Ras, and MAPK (Figure 5E). We used protein interaction network (PPI) networks and Metascape to identify protein interactions and Cytoscape to visualize them (Figure 5F). Functional enrichment analysis from high-degree proteins and the top five pathways revealed that AKT1 had a high degree of protein interactions and the PI3K/AKT pathway had the highest number of interacting proteins (Figure 5G). Meanwhile, KEGG and pathway mapping of differentially phosphorylated proteins revealed significant PI3K/AKT pathway enrichment (Figures 5H,I and S7B). For further confirmation, we conducted western blotting and found that AKT were significantly phosphorylated after 15 min of TPX treatment (Figure S7C). When two different inhibitors were added for 72 h, phosphorylated AKT showed a similar pattern, increasing in the first 48 h and then decreasing (Figures 5J,K and S8). These findings suggest that PI3K/AKT plays a pivotal role in the downregulation of aromatase caused by elevated SERPINE1 level.

2.5 TPX Effectively Improves Disorders Caused by Dehydroepiandrosterone and Continuous Darkness in Rats

As we previously observed hypoandrogenic effects after SERPINE1 inhibition, we investigated the therapeutic potential of SERPINE1 inhibitors in a preclinical study. We first tested the effect of simultaneous gavage of TPX or LrB at different concentrations for 21 days during dehydroepiandrosterone (DHEA) modeling (Figures 6A and S9A). Compared with controls, the DHEA-treated rats exhibited impaired ovulation, as evidenced by follicles mostly blocked in the preantral follicular phase and stalled and disturbed estrous cycles. Notably, TPX 2 mg/kg, rather than TPX 20 mg/kg or LrB, effectively reduced the number of preantral follicles, with increasing the number of corpus luteum and restoring normal cycles (Figures 6B,C and S9B). TPX treatment effectively suppressed PAI-1 levels (Figure 6D) and restored abnormal testosterone and FAI levels in DHEA-treated rats in the absence of significant SHBG and gonadotropin fluctuations (Figure 6E–J). Notably, the PAI-1 level did not significantly decrease in the DHEA-treated group, leading to the speculation that SERPINE1 might serve as an upper stimulus, instead of a response, for elevated androgen level in vivo (Figure 6D). In line with the high androgen levels, the aromatase-promoting effect of TPX induced a slight increase in E2 levels (Figure 6K). Furthermore, the body weight of the rats decreased with TPX 2 mg/kg (Figures 6L and S9C). The IPGTT results indicated that the glucose metabolic effect of TPX 2 and 20 mg/kg was significantly more potent than that of LrB at any dose (Figures 6M and S9D,E). These findings indicate that PAI-1 plays a therapeutic role in hyperandrogenism-induced PCOS.

Next, we administered continuous TPX treatment via gavage to the rats during the 8-week continuous darkness exposure (Figure 6N). Consistent with previous findings, the dark group exhibited manifestations of ovarian morphology, estrous cycles, and elevated PAI-1 levels, as previously observed (Figure 6O–Q). TPX significantly decreased the PAI-1 levels and the androgen levels (Figure 6Q–T), further validating the downregulatory effect of SERPINE1 on androgens in vivo. As expected, the LH level elevation in the dark group was significantly reversed by TPX, but the FSH levels decreased after TPX treatment (Figure 6U–W), suggesting that TPX exerts a potential suppressive effect on gonadotropin secretion. Interestingly, contrary to the DHEA models, we did not observe a similar accumulation of E2 levels (Figure 6X). Although the body weight of rats did not change (Figure 6Y), the TPX treatment partially rescued darkness-induced abnormalities in glucose metabolism (Figure 6Z). These results indicate the therapeutic effect of PAI-1 in circadian disruption-induced hyperandrogenism.

2.6 SERPINE1 is Hypomethylated and Overexpressed in Women with PCOS

To determine whether our PCOS findings in rats might be associated with those in humans, we first confirmed the serum PAI-1 level in a cohort of 57 women with PCOS and 57 controls. As a result, the PAI-1 levels were significantly elevated in those with PCOS (Figure 7 and Table S1). In addition, we characterized a significant positive correlation between serum PAI-1 and testosterone levels in this cohort (Figure 7B). PAI-1 also had a strong correlation with BMI and HOMA-IR (Figure S10A,B), as reported in previous literature [28, 29]. But we found no difference in PAI-1 in follicular fluid from patients with PCOS and controls (Figure S10C and Table S2). Consistent with data from rat GCs, the methylation level of CpG in the SERPINE1 promoter decreased in the peripheral leukocytes of patients with PCOS (Figure 7C,D and Table S3). Notably, the two CpG sites (positions 10 and 11) that underwent significant changes were located in the promoter region close to the TSS, which was very similar to the changes observed in the rats.

4.1 Animal Use and Care

The animal study protocols were approved by the Animal Ethics Committee of Shanghai Model Organisms Center (2021-0011-01) and Shanghai University of Traditional Chinese Medicine (PZSHUTCM211101047). All animal experiments were carried out in accordance with the guidelines of the local animal ethics committee.

4.2 Continuous Darkness-Exposed Rat Model

The continuous-darkness-exposed SD rat model has been generated as previously described [11, 12]. A total of 80 female SD rats (5–6 weeks) were used after 1 week of a 12:12 h light–dark cycle acclimatization. The rats were randomly divided into two groups: (1) the dark group, caged with impermeable black cage cloth, resulting in a continuous dark environment and (2) the control group, kept under a 12:12 h light–dark cycle. Eight rats from each group were sacrificed every 2 weeks for total 8 weeks.

4.3 DHEA Treatment

Immature (21 d old) female SD rats were injected daily with DHEA (60 mg kg⁻¹/d, dissolved in 0.2 mL of sesame oil, D4000; Sigma Aldrich, St. Louis, USA) subcutaneously, as the DHEA group. The control group received injections of the same volume of sesame oil daily. The above animals were all treated for 3 weeks.

4.4 TPX Treatment

TPX was administered by gavage as a rescue treatment. For the darkness-treated model, there were three groups: (1) dark + TPX, kept under a continuous dark environment and gavaged with TPX (2 mg kg⁻¹/d, diluted in 1% DMSO in corn oil, HY-15253; MedChemExpress, New Jersey, USA), (2) dark + oil, kept under a continuous dark environment and gavaged with corn oil (HY-Y1888; MedChemExpress), (3) control + oil, kept under a 12:12 h light–dark cycle and gavaged with corn oil. The above animals were all treated for 8 weeks. For the DHEA-treated model, there were three groups: (1) DHEA + TPX, injected with DHEA (60 mg kg⁻¹/d) and gavaged with TPX (2 mg kg⁻¹/d), (2) DHEA + oil, injected with DHEA (60 mg kg⁻¹/d) and gavaged with corn oil, (3) oil + oil, injected with sesame oil and gavaged with corn oil. The above animals were all treated for 3 weeks.

4.5 MC-seq DNA Methylation Data Analysis

For MC-seq, raw image files obtained from sequencing were identified by base identification and error filtering to obtain raw sequencing fragments that could be used for analysis, and the results were stored in FASTQ file format. Sequence data quality was examined using FastQC (ver. 0.11.5). The bases with low quality scores and the adapters in all the sequenced reads were removed by Trim_galore (ver. 0.4.1). The trimmed reads were mapped to Rattus norregicus genome (RAT6.0, ensemble) by Bismark pipelines (ver. v0.15.0; bowtie v2.2.9). Duplicated reads were removed by deduplicate_bismark. All CpG sites with coverage ≥10× depth were retained for analysis to ensure high MC-Seq data quality. Each sample provides approximately 20 G of raw data after sequencing, with a targeted capture area of approximately 90 Mb. The target read ratio varies from 71.27 to 84.26%, and the threshold value was set as: p < 0.05, methylation difference degree >10%. The DMRs were obtained by the R package DSS call DMR function with the DMR length (less than 1000 bp) and CpG number (at least 3 CpGs in a DMR), and the threshold value was set as: pairwise Chi-squared tests with subsequent multiple testing correction; adjusted p < 0.05, a minimal difference cutoff of 10%. The promoter region is defined as the regions from −2.5 to +0.5 kb of the tTSSs. The selected DMS-associated genes were mapped to each term in the GO database, the KEGG database and the Reactome database using KOBAS-i (KOBAS 3.0), and the number of genes in each term was calculated. Fisher's exact test was used to measure the gene-enrichment in annotation terms. Principal component analysis was performed using R (v.3.3.1).

4.6 RNA-seq Data Analysis

The raw pair-end RNA-seq FASTQ data were trimmed to remove adaptor sequences and low-quality bases by Seqtk (GitHub—lh3/seqtk: Toolkit for processing sequences in FASTA/Q formats) with default settings. The trimmed reads were mapped to Rattus norregicus genome (RAT6.0, ensemble) using Hisat2 (version:2.0.4). The expression level of each gene was calculated as fragments per kilobase of exon model per million mapped reads (FPKM) by Stringtie (version:1.3.0) and TMM (trimmed mean of M values). DEGs analysis was performed by using edgeR. The DEG were as follows: at least 1.5-fold change and adjusted p value (attained by the Wald test and corrected for multiple testing using the Benjamini–Hochberg method by default) less than 0.05. The DEGs were submitted to the DAVID 6.8 website to perform KEGG analysis, and Fisher's Exact test was used to measure the gene enrichment.

4.7 Correlation Analysis of Methylation and Transcription

To obtain correlation between differential methylation and gene expression, we considered only DMSs in promoter regions. We plotted log2-fold expression changes versus methylation differences, and the distribution of genes among the four resulting quadrants was tested for directionality using Fisher's exact test.

4.8 CRISPR–dCas9-Based Targeted Methylation in BRL-3A Cells

The sgRNA–ZsGreen1–T2A–Puro (PGMLV-SpCas9) vectors (GM-8199) and GPLVX–CMV–3×Flag–NLS–dCas9–Dnmt3aCD–T2A–Blasticidin vectors (10948) were purchased from Genomeditech (Shanghai, China). To generate stable cell lines with integrated dCas9–Dnmt3a transgenes, the BRL-3A cells were seeded into a six-well plate and transfected the next day at 30–50% confluence. We added 5 µg/mL polybrene (Sigma–Aldrich) to 10% FBS DMEM. BRL-3A cells were transduced with lentiviral particles of GPLVX–CMV–3×Flag–NLS–dCas9–Dnmt3aCD–T2A–Blasticidin according to the provider's protocol. Stably integrated cells were selected with blasticidin S (25 µg/mL, GM-040404; Genomeditech) for 48 h. Then stable dCas9–Dnmt3aCD–BRL–3A cells were transduced with lentiviral particles of different sgRNA–ZsGreen1–T2A–Puro vectors separately (Table S4). Stably integrated cells were selected with puromycin (1.5 µg/mL, GM-040401; Genomeditech). Cells were harvested 3 days postinfection in this study.

4.9 Phospho-Antibody Array Analysis

Identification of PAI-1 targeting proteins was performed by phospho-antibody array (Full Moon, Sunnyvale, USA) analyzed with phospho-Exploerer (PEX100) by Wayen Biotechnologies (Shanghai, China) using KGN cells. KGN cells were treated with TPX (10 µM) for 15 min, then the cells were collected for the array. The phosphorylation signal ratio was calculated as follows: phospho-ratio = phospho/unphospho. The ratio of the phospho-ratio between 15 and 0 min was calculated as follows: ratio of phospho-ratio = phospho-ratio (15 min)/phospho-ratio (0 min). The Fold Change threshold was set to 1.5 to obtain the differentially phosphorylated proteins.

4.10 Clinical Samples

This study enrolled patients who underwent in vitro fertilization-embryo transfer treatment at Department of Reproductive Medicine in Ren Ji Hospital Affiliated to Shanghai Jiao Tong University School of Medicine from January 2015 to December 2017. Discarded serum, plasma and follicular fluid from patients were collected for the study. The Medical Ethics Committee of Ren Ji Hospital Affiliated to Shanghai Jiao Tong University School of Medicine approved this study process (no. 2017041411), and all participants provided written informed consent. The main reference for the diagnostic criteria of PCOS was the Rotterdam Consensus for PCOS revised in 2003. Basic and PCOS-related anthropometric variables are presented in Tables S1–S3.

4.11 Statistical Analysis

All data are presented as mean ± SD. Each experiment/group used 3–5 samples/independent replicates. Statistical analyses were conducted using IBM SPSS Statistics and GraphPad Prism 8 (GraphPad Software). Statistical significance was analyzed using Levene's test for equality of variances. Differences between two groups were determined by two-tailed Student's t test where data was normally distributed data. Differences between more than two groups were determined by one-way ANOVA followed by Dunnett's post hoc test or the Kruskal–Wallis test followed by Dunn's post hoc test. The correlation between the two data sets was analyzed using the Pearson correlation coefficient when equal variance was assumed. p < 0.05 was considered as statistically significant.