PPP1R1C: a specific placental mRNA biomarker for distinguishing preeclampsia from healthy pregnancies

Abstract

Background

Preeclampsia (PE) poses a serious threat to maternal and fetal health, and its early, reliable diagnosis remains challenging.

Methods

We conducted a two-phase study using a discovery and validation design. Initially, multiple RNA-sequencing datasets from NCBI GEO were aggregated, and differentially expressed genes (DEGs) were identified via meta-analysis. The DEGs were then analyzed with weighted gene co-expression network analysis (WGCNA) to pinpoint the module most associated with PE. A signature biomarker model was developed using binary logistic regression and subsequently validated by real-time PCR (RT-PCR) on placental tissues from 30 PE patients and 30 matched controls.

Results

Meta-analysis yielded over 4000 DEGs, from which WGCNA identified a module of 100 genes most correlated with PE. Within this module, 24 genes exhibited |logFC| > 1, and four candidates (FSTL3, PNCK, PPP1R1C, TBC1D26) demonstrated a high discriminative ability as a combined signature biomarker (AUC = 0.90). Subsequent RT-PCR analysis confirmed the overexpression of FSTL3 and downregulation of PPP1R1C, with PNCK and TBC1D26 undetectable. However, FSTL3 did not significantly differentiate PE patients (AUC = 0.6, p > 0.05), whereas PPP1R1C achieved high accuracy (AUC = 0.94, p < 0.0001). Additionally, the alterations of FSTL3 and PPP1R1C were not correlated and combining them did not improve the diagnostic ability compared to individual usage of PPP1R1C.

Conclusion

Placental downregulation of PPP1R1C mRNA effectively distinguishes PE patients from healthy controls, indicating its promise as a diagnostic biomarker.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-026-08779-x.

Introduction

Preeclampsia (PE) is a growing concern that affects 2–8% of all pregnancies [1]. It is characterized by the advent of hypertension (systolic blood-pressure > 140 mmHg, diastolic blood pressure > 90 mmHg) and often organ damage after 20 weeks of gestation [2]. Proteinuria is the other most common symptom of PE. Severe PE (sPE) is associated with maternal renal, pulmonary and neurologic deficits and growth restriction of the fetus; however, most severe cases could lead to stillbirth and maternal death [3, 4].

A pressing problem in PE studies is the identification of highly efficient and yet specific biomarkers for early diagnosis [5]. Due to the fact that the main underlying mechanisms of PE still remain unclear, proposing molecular biomarkers associated with PE development is challenging [6]. Recent developments in high-throughput data analysis, including RNA sequencing and microarrays, have revealed novel findings about the molecular basis of PE, particularly concerning genetic and epigenetic alterations [7]. Interestingly, these alterations could be utilized for developing highly specific biomarkers, as the disease is likely caused by these aberrations; therefore, disease-related changes in the transcriptomic profile of a tissue or just in blood flow have been proposed as diagnostic biomarkers not only in PE but also in a wide range of diseases [8]. Changes in certain indicators in the placenta, including but not limited to RNA and proteins, may be the cause of changes in the levels of circulating markers, providing a reference for the discovery of future circulating markers [9].​

Recently, the identification of coordinated expression aberrations among multiple transcripts has provided new insights into disease etiology, particularly for idiopathic diseases such as PE, because interrelated genes that form a module are considered functionally related or part of the same pathway [10]. Co-expression analysis alongside expression analysis of high-throughput data makes a powerful approach which could reveal not only the differentially expressed genes (DEGs) but also the correlation network between these DEGs that eventually that could be used to propose particular biomarkers for each disease [11].

Taken together, this study aims to identify PE-associated DEGs through a meta-analysis approach based on pre-existing RNA-seq data in the NCBI GEO database. Moreover, weighted gene co-expression network analysis (WGCNA) is utilized to discover the most correlated topological modules with the onset of PE. Subsequently, these results are used to propose a unique and novel diagnostic biomarker for early detection of PE. To the best of our knowledge, this work is the first to perform an integrative analysis of multiple datasets combined with WGCNA to suggest potential biomarkers, which are subsequently validated via real-time polymerase chain reaction (RT-PCR).

Methods and materials

RNA-seq data collection and Preparation

We have searched NCBI GEO for RNA-seq datasets that evaluate expression patterns in PE. Importantly, each dataset was rigorously checked to ensure adherence to internationally recognized PE diagnostic standards, minimizing potential bias from variations in diagnostic criteria across different studies.​ Five datasets (Table 1) were qualified and included in this work as they met our selection criteria, which comprised RNA Integrity Number (RIN) ≥ 7, adequate sequencing depth, high base quality scores, successful adapter/contaminant removal, and high read mapping rates to the reference genome. Furthermore, the patients must not have been subjected to any form of treatment. These datasets were then downloaded using SRAtoolkit [12], and their quality was assessed using FastQC [13]. Removal of adapter content and quality-enhancement trimming was carried out by Trimmomatic [14]. Next, HISAT2 was used to align the remaining reads to the reference genome (GhRC38) [15], and aligned reads were counted using featureCounts software [16]. Later, the percentage of assigned reads was visualized by MultiQC, and samples with abnormal traits were excluded.

Table 1.

Dataset characteristics

Dataset	Platform	Sample Size	Mean gestational age	Sequencing type	Ref
GSE234729	Illumina NovaSeq 6000	Case: 73 Control: 50	NA	Total RNA	[17]
GSE186257	Illumina NovaSeq 6000	Case: 18 Control: 26	NA	Total RNA	[18]
GSE177049	Illumina HiSeq 2000	Case: 5 Control: 5	NA	Total RNA	[19]
GSE143953	Illumina HiSeq 4000	Case: 4 Control: 4	Case:37.9 Control:38.17	Total RNA	[20]
GSE114691	Illumina HiSeq 2000	Case: 21 Control: 20	NA	Total RNA	[21]

Defining metadata and its quality control

The expression tables of each dataset were introduced as a batch to the ComBat-seq function from the SVA package under R 4.2.0 software [22]. After the batch effect was corrected by ComBat-seq, the quality of the metadata was assessed using batch median correlation and density plots.

Metadata characteristics

The suitability of data for statistical analysis was later confirmed through dispersion assessment via boxplot, principal component analysis (PCA), and hierarchical clustering. Dispersions were corrected by normalization using the LIMMA package. Next, samples with improper behaviors were identified as outliers and eventually excluded.

Differential expression analysis

DESeq2, an advantageous method for RNA-seq data analysis, was used to compare the mean of each transcript between the PE and control groups [23]. In addition to mean difference, base expression, log fold change (LogFC), P value (P.Val), and adjusted P value (adj.P.val) were calculated and reported for each gene. The significance level was set to 0.05 to determine DEGs.

WGCNA analysis

Module membership analysis was carried out on DEGs based on their expression patterns using the WGCNA package. Subsequently, we constructed an adjacency matrix to describe the correlation strength between the nodes. The correlation coefficients between genes were calculated using the following formula: aij=|Sij|β (aij: adjacency between gene i and gene j, Sij: Pearson’s correlation between gene i and gene j, β: soft threshold). Then, the adjacency matrix was converted into a topological overlap matrix (TOM). In WGCNA, the linkage hierarchical clustering was carried out for the genes dendrogram based on dissimilarity measure (1-TOM), and the minimum size (gene group) was set as 30 in order to classify the genes with similar expression profiles into the same gene module.

To further determine the key modules in the co-expression network, the relevance of module eigengene (ME) to PE was calculated, and the association of each gene with clinical significance was measured by gene significance (GS). Module membership (MM) represents the average gene significance (GS) of all genes within a module. Finally, the GS-MM correlation was calculated for each group to ensure the selection of the most appropriate module.

Network visualization

The output of DEGs’ interactions within the selected module was visualized using CytoScape software. Genes are presented as nodes, and the links between these nodes are representative of the interaction. Next the degree filter was applied on the network to identify the most interacted genes as the hub-genes.

Filtration and regression analysis

After selection of the most correlated module, DEGs within this module were filtered using |LogFC| > 1. Next, binary logistic regression analysis was performed on the filtered genes to develop a model biomarker for discrimination of patients from healthy controls. Binary logistic regression was selected as the primary modeling approach due to its interpretability and suitability for binary outcome prediction in the context of our study. This method allowed us to directly assess the relationship between predictor variables and the outcome of interest using odds ratios, which are intuitive and clinically meaningful. To minimize risks of overfitting and ensure reproducibility, data preprocessing steps—including normalization and feature selection—were applied consistently across analyses. While alternative machine learning models may offer enhanced flexibility, the transparency and established reliability of logistic regression in biomedical research supported its use here. We believe this approach appropriately balances robustness with interpretability for our study’s aims.

Finally, ROC curve analysis was conducted to evaluate the diagnostic ability of the suggested model according to the expression of DEGs in meta-data. Moreover, the diagnostic efficacy model biomarker was assured using its associated sensitivity and specificity values.

Functional enrichment analysis

EnrichR was utilized to investigate the biological function and cellular processes mediated by DEGs. Significant results were identified with p < 0.05 and subsequently presented as a dot plot.

Acquiring human placenta samples

In order to validate the accuracy of the proposed biomarker from the discovery phase, the expression patterns of candidate genes were investigated in human placental tissue obtained from the Vali-E-Asr Reproductive Health Research Center from October 2024 to January 2025. A total of 30 placenta samples were obtained from PE patients as cases, and 30 age-matched samples were enrolled in this study as healthy controls. Placental biopsies were collected within 30 min after delivery. Samples were taken from the villous parenchyma of central cotyledons, approximately 5–8 mm below the chorionic plate, to include trophoblast-rich villous tissue. Maternal decidua, chorionic membranes, umbilical cord, and large vessels were carefully excluded. Areas with infarction, calcification, or hemorrhage were avoided. Four biopsies per placenta were pooled, rinsed briefly in PBS to remove blood, and preserved in RNAlater before storage at − 80 °C.

The diagnostic criteria for PE used in this study align with the guidelines set forth by the American College of Obstetricians and Gynecologists (ACOG).​ Specifically, PE was defined by the presence of new-onset hypertension (systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg on two occasions at least four hours apart after 20 weeks of gestation) or other features of severe PE, such as thrombocytopenia, impaired liver function, renal insufficiency, pulmonary edema, or new-onset cerebral or visual disturbances [24]. Healthy controls had no history of hypertension, proteinuria, or other complications during pregnancy. While samples were primarily age-matched as indicated in Table 2, the study acknowledged the importance of other potential confounders like gestational age and familial history. The demographic characteristics provided in Table 2 present the distribution of these variables across groups.

Table 2.

Demographic characteristics of participants

		Case	Control	Statistical Result
Count		30	30
Age (mean)		31.1	28.5	p = 0.12
Gestational age at delivery (weeks, mean)		35.2 ± 2.1 (95% CI: 34.4–36.0)	38.6 ± 1.3 (95% CI: 38.1–39.1)	p < 0.001
Blood Pressure		155/100 mmHg	120/80 mmHg	p < 0.01
Proteinuria		2+	Absent	p < 0.001
Recurrence		27%	-	-
Previous pregnancy	No	18 (60%)	10 (33%)	-
	1st	8 (27%)	12 (40%)	-
	2nd	3 (10%)	6 (20%)	-
	3rd	1 (3%)	2 (7%)	-
Familial History	Yes	12 (40%)	5 (17%)	-
	No	18 (60%)	25 (83%)	-
Abortion History	Yes	9 (30%)	3 (10%)	-
	No	21 (70%)	27 (90%)	-

This study was approved by the ethics committee of Baqiyatallah University of Medical Sciences, and written informed consent was acquired from all participants. Samples were collected during the labor, and the health state has been confirmed by the specialist gynecologist. The samples were stored at -80℃, and the RNA content of the obtained tissues has been preserved with RNA-later solution (YTA, Tehran, Iran).

RNA isolation from placenta tissue

The tissues were initially homogenized using liquid nitrogen, and next their nucleic acid content was isolated using the SinaPure One extraction kit (SinaClon BioScience, Cat. No. EX6051). Afterwards, samples were treated with DNase I to eliminate DNA contamination and increase sample stability. Subsequently, the concentration and quality of extracted RNA were measured by a Nano-Drop spectrophotometer.

Synthesis of cDNA and RT-PCR

Complementary DNA (cDNA) was synthesized from 500 ng of total RNA using the ExcelRT™ Reverse Transcription Kit II (SMOBio Technology, Inc., Cat. No. RP1400), following the manufacturer’s protocol.

Quantification of candidate genes was performed via real-time PCR (RT-PCR) using the ExcelTaq™ 2X Fast Q-PCR Master Mix (SMOBio Technology, Inc., Cat. No. TQ1100) mix on synthesized cDNA. Each sample was analyzed in technical triplicates to ensure reproducibility and accuracy of expression measurements. Primers targeting each gene of interest were either designed using Primer3Plus and OligoAnalyzer based on reference sequences (listed in Supplementary file 1, Table 1). Primer specificity and amplification efficiency were confirmed via melting curve analysis and standard curve evaluation, ensuring single product amplification and minimizing primer-dimer formation. GAPDH was used as the internal control gene for normalization. The RT-PCR conditions were initial denaturation at 95 °C for 10 min, followed by 40 amplification cycles (95 °C, 58 °C, and 72 °C). (See supplementary file 1, Figs. 1 and 2)

Statistical analysis

RT-PCR results were analyzed by the Livak method (2^−ΔΔct) to compare the mean difference between candidate gene expression in PE and control groups. Initially, normality tests were utilized to evaluate the distribution patterns of the CT values. Next, the significance of the difference between the means of two groups was tested by Welch’s t-test or Mann-Whitney based on their distribution patterns. Later, ROC curve analysis was performed for each gene, accompanied by the development of a combined model. The entire analysis was carried out by GraphPad Prism 9.0.

Results

RNA-seq data characteristics

Overall, the baseline expression of over 26,000 transcripts was assessed through sequencing in each dataset, of which 10,800 were included for further analysis after filtering the genes without any expression or with low expression. Next, the quality of the batch effect correction was ensured through batch effect variation analysis (Fig. 1A), and the correlation of samples was evaluated via median pairwise correlation (Fig. 1B). Later the proper distribution patterns of remaining genes were confirmed through visualization of density (Fig. 1C) and Q-Q plots (Fig. 1D); moreover, the similarity of expression profile of all samples within the meta-data was checked by PCA, confirming the samples are qualified for meta-analysis after correcting the batch effect (Fig. 1E). (For PCA plot prior to batch effect correction see supplementary file 1, Fig. 3)

Fig. 1.

Comprehensive assessment of batch effects using multiple complementary approaches. A Heatmap of row wise Z scores for four summary statistics calculated per batch — mean expression, variance, skewness, and kurtosis — each standardized across batches (Z score computed within each metric). These standardized values highlight significant differences between batches (all p < 0.0001 at both sample wise and gene wise levels). B Median pairwise correlation plot showing inter sample similarity and separation by batch. C Density plot of expression distributions across batches (N = 10 851 features; bandwidth = 0.2113). D Normal Q Q plot illustrating deviation from theoretical normality. E Principal component analysis (PCA) plot showing sample distribution along the first two principal components (PC1, PC2), colored by batch

Identifying the DEGs in PE

According to the DeSeq2 expression analysis, the concentration levels of 4141 genes out of 10,851 genes were significantly different in placenta tissue of PE patients compared to the healthy controls. The alterations of these DEGs were highly diverse and included both up- and down-regulation of specific genes (Fig. 2A-B). SERPINA3 was the most overexpressed gene, as its levels were amplified more than 16 times in patients (LogFC = 4.57), and KDM5A was the gene with the least significant overexpression, which was only increased by 3% (LogFC = 0.05). On the other hand, the downregulations were less intense, and their calculated LogFC varied from − 1.63 to -0.05, where SLAMF1 was the most decreased gene and ZNF426 was on the other side of the range. Outstandingly, over 90 genes had LogFC > 1 (Adj.P.Val < 0.05), meaning their expression was at least doubled in PE patients, indicating the excessive amount of disruption in gene expression patterns leading to an increased chance of PE. On the contrary, only 20 genes were downregulated more than two times (LogFC < -1, Adj.P.Val < 0.05), proposing the fact that even though 46% of DEGs are allocated to the downregulated group, the overexpression of particular genes may be the main contributor in PE development. Figure 2C depicts the schematic correlation between significant results using a heatmap.

Fig. 2.

Differential gene expression and state based clustering in preeclampsia (PE) versus control samples are shown as follows: A MA plot mapping log₁₀ transformed mean expression against log₂ fold change, with genes colored to indicate significant upregulation (green), downregulation (red), or nonsignificant change (black); B volcano plot presenting log₂ fold change versus –log₁₀ transformed p value to highlight statistically significant genes; and (C) heatmap with hierarchical clustering dendrograms grouping genes and samples by similar expression profiles, with sample clustering performed according to sample state (case vs. control), providing insights into potential biomarkers and the underlying molecular mechanisms of PE

WGCNA analysis and module detection

Initially, hierarchical clustering analysis was applied, leading to the identification of one outlier sample (Fig. 3A). Afterwards, the required TOM matrix was built using a soft threshold, which was set to be 7 according to the scale-free topology model fit and mean connectivity analysis (Fig. 3B). Next, samples were re-clustered considering the TOM-based dissimilarity; subsequently, a distinct color was assigned to each cluster (Fig. 3C). Initially, 14 unique modules were distinguished, and each included at least 30 genes. (see supplementary file 1, Table 2) Next, the eigengene — the first principal component of the module’s expression data — was calculated for each module. Based on the eigengene similarity, modules with more than 80% similarity (Pearson correlation > 0.80; height < 0.20) were merged to minimize redundancy, following standard WGCNA practice and recommendations from Langfelder & Horvath (2008), thereby ensuring that retained modules represented distinct expression patterns. This merging step resulted in eight final modules (Fig. 3C). Finally, correlation analysis identified the relationship between each module and the trait (PE).

Fig. 3.

This figure presents the workflow for clustering and module analysis across five panels: A a hierarchical clustering dendrogram that identifies outlier samples using a red threshold line; B two plots that determine the optimal soft threshold power, with the left showing the Scale Free Topology Model Fit (signed R²) over powers 1–10 and the right depicting the corresponding decline in mean connectivity; C a detailed gene clustering dendrogram annotated with color bars representing both the initial dynamic tree cut and the merged clusters; D a bar plot of module eigengenes that summarizes the predominant expression patterns across identified modules; and (E) a scatter plot correlating module membership with gene significance for the green/yellow module, featuring a regression line that highlights a strong positive relationship (cor = 0.73, p = 2.12e-19)

Interestingly, all eight modules were significantly related to the onset of PE; however, half of these modules were positively correlated (ρ > 0), and the remaining half were negatively correlated (Fig. 3D). The green-yellow module was the most correlated with the PE and, thereby, was selected for further analysis. Likewise, this module presented the highest correlation analysis in the GS-MM correlation analysis (Fig. 3E). (See supplementary file 2)

Green-yellow module visualization

As illustrated in Fig. 4, a total number of 202 genes make up the green-yellow module. According to the applied degree filter, some genes, including SASH1, PDZD7, GPR146, P4HA1, PNCK and FSTL3, had the most connections and were detected as network hub genes. Within this network, nodes with bigger size and distinct color are representative of the genes with the most interactions and are located in the center, whereas genes with fewer interactions are positioned in the margins and have smaller size. Considering the fact that these genes are both differentially expressed and were allocated to the most related module while also constituting the hub genes of the network gives rise to the idea that they are the main drivers of the PE.

Fig. 4.

Network visualization of the green-yellow module, displays the interconnectedness of genes within this specific module. The nodes in the network represent individual genes, and the links between them signify their interactions. Genes with a larger node size and more distinct color indicate higher interaction counts and are centrally located within the network, while genes with fewer interactions are smaller and positioned at the margins

Efficiency of the combined model biomarker

In order to develop a proficient diagnostic biomarker, genes with an absolute value of LogFC over 1 were extracted from the green-yellow module, leaving 24 unique genes. To refine this set, backward binary logistic regression analysis was applied, which iteratively removed non-contributory variables until the most parsimonious model was achieved. This procedure retained four genes (FSTL3, PNCK, PPP1R1C, and TBC1D26) as the final predictors, each assigned a regression coefficient (β) reflecting its contribution to the outcome (PE vs. control). The probability value was derived directly from the logistic regression model, calculated as the weighted sum of these gene expression values plus the intercept term, and transformed using the logistic function:

Performing binary logistic regression analysis on the count table of these 24 DEGs resulted in the formula below:

ROC curve analysis was then performed on the calculated probability values, with individual expression values provided in Supplementary file 3. The AUC for the suggested model biomarker was 90.3%, and its associated sensitivity and specificity were 84% and 88%, respectively (Fig. 5A). The relative expression of the four genes within the formula is presented in Fig. 5B. (For relative expression of candidate genes and their associated AUC values see Figs. 4 and 5 in supplementary file 1)

Fig. 5.

A ROC curve analysis of model biomarker, B Relative logFC of genes in model biomarker in discovery phase. C Dot plot of functional enrichment analysis

Functional enrichment analysis

Functional enrichment analysis of the gene set (green-yellow module) revealed its association with a range of biological functions, suggesting a multifaceted role in PE progression (Fig. 5C). The findings indicate that the gene set is involved in hormonal signaling and peptide hormone biosynthesis, processes that are essential for cellular communication and endocrine regulation. In addition, the analysis highlighted significant alterations in metabolism, specifically in the pathways linked to amino acid and creatine metabolism, suggesting a metabolic shift that may support cellular adaptation and energy homeostasis. Furthermore, the enrichment results pointed to the involvement of extracellular matrix remodeling and cell adhesion processes, implying that modifications in tissue architecture and intercellular interactions may be critical aspects of the biological impact of these genes. Moreover, the regulatory network appears to extend to pathways involved in nutrient transport and immune modulation, as evidenced by the enrichment in transport mechanisms and oncostatin M signaling. Collectively, these insights underscore the complex interplay between endocrine, metabolic, structural, and immunological factors, and they provide a foundation for further investigation through network analysis, experimental validation, and comparative studies across different datasets.

Expression differentiation of selected genes in placenta tissue

Initially we tried to assess the expression levels of these genes; however, the detectable levels were only observed for FSTL3 and PPP1R1C and not for PNCK and TBC1D26. The undetectable levels of PNCK and TBC1D26, despite initial attempts to assess their expression, can be attributed to several factors.​ It is possible that the expression of these genes is genuinely very low or absent in the bulk placental tissue samples analyzed, falling below the detection limits of the RT-PCR assay. This could be because their expression is confined to specific, specialized cell types within the placenta that were either not present or were underrepresented in the collected samples, as bulk RNA-seq provides an average expression profile that might obscure signals from rare cell populations. Additionally, variations in sample quality, preparation methods, or the intrinsic sensitivity and efficiency of the RT-PCR assay itself could also contribute to these genes remaining undetectable.

Normality tests have identified that FSTL3 is normally dispersed, whereas PPP1R1C does not follow a normal distribution pattern. According to the Welch’s t-test, the FSTL3 is slightly (FC = 1.71, p = 0.049) but significantly upregulated in the PE group compared to healthy controls (Fig. 6A). On the contrary, the overexpression of PPP1R1C was initially detected by RNA-seq analysis; however, the RT-PCR analysis pointed out a strong decrease in PE patients (FC = 0.53, p < 0.0001) (Fig. 6B). Additionally, correlation analysis between these genes did not yield any significant results (r = 0.07, p = 0.6), meaning the expression of FSTL3 and PPP1R1C are not accordant despite what was suggested in WGCNA analysis (Fig. 6C). (See supplementary file 3)

Fig. 6.

A, B Relative expression of FSTL3, PPP1R1C in placental tissue of PE patients compared to healthy mothers. C correlation of FSTL3 and PPP1R1C in placenta tissue. ROC curve analysis of FSTL3 (D), PPP1R1C (E), and combined biomarker (F)

Practical value of proposed biomarker

Although the initial suggested model was comprised of expression levels of 4 mRNAs, we were able to detect the expression of two of them. Primarily, the diagnostic ability of these genes was individually assessed by performing the ROC curve analysis on the – Δct values. As shown in Fig. 6D, FSTL3 was not significantly able to distinguish patients from control (AUC = 0.6, p = 0.19, Sensitivity = 0.57, Specificity = 0.50); however, on the other hand, the calculated AUC for PPP1R1C was 0.94 (95% = 0.88 − 1.00, p < 0.0001, Sensitivity = 0.82, Specificity = 0.92), which is already higher than the proposed model, proposing the ability of placental PPP1R1C mRNA to distinguish the PE patients from healthy mothers successfully without combination with other transcripts (Fig. 6E).

Regarding the combination of diagnostic power of FSTL3 with PPP1R1C, despite the significance and potency (AUC = 0.93, p < 0.0001, sensitivity = 0.81, specificity = 0.90), the combined model not only has no superiority over PPP1R1C, but it also has relatively decreased ability. Therefore, individual application of PPP1R1C could be ideal, as it presents better results with minimum cost and effort (Fig. 6F).

Discussion

In recent years, a growing body of data has highlighted the limitations of conventional diagnostic tools for PE and underscored the urgent need for biomarkers with improved sensitivity and specificity [25]. Evidence from extensive proteomic, genomic, and imaging studies suggests that novel candidate biomarkers may better reflect early disease processes and offer critical insights into underlying pathophysiological mechanisms [26]. This general landscape of existing research not only reveals promising correlations between certain markers and disease states but also points to their potential utility in enhancing early diagnosis and guiding personalized treatment strategies [27]. Our investigation into the molecular underpinnings of PE has illuminated the potential diagnostic utility of Follistatin-Like 3, or FSTL3 and Protein Phosphatase 1 Regulatory Inhibitor Subunit 1 C, also known as PPP1R1C. While initial RNA-seq analysis suggested a combined model biomarker involving FSTL3, PNCK, PPP1R1C, and TBC1D26, subsequent RT-PCR validation on placental tissue samples revealed a more nuanced picture. Although detectable levels were only observed for FSTL3 and PPP1R1C, and not for PNCK and TBC1D26, individual assessment highlighted PPP1R1C as a particularly potent diagnostic marker. Indeed, while FSTL3 exhibited a slight but significant upregulation in PE, PPP1R1C demonstrated a strong decrease in PE patients, boasting an AUC of 0.94 in ROC curve analysis.

FSTL3 exhibited slight overexpression in our study, yet its diagnostic value was not significant, a finding that conflicts with prior research [28] where more pronounced overexpression and significant diagnostic utility were reported.​ This divergence suggests a potential difference in the magnitude of FSTL3 upregulation observed across studies. Several factors might contribute to this discrepancy, including variations in sample size, which can affect the statistical power to detect significant changes, and ethnic differences among study populations, which could influence FSTL3 expression levels. Furthermore, variations in the specific PE subtypes investigated in different studies might lead to varying degrees of FSTL3 overexpression, as the molecular mechanisms underlying different PE presentations can differ. In stark contrast, PPP1R1C demonstrated a substantial decrease in PE patients within our study, showing a robust diagnostic capability with an AUC of 0.94 in ROC curve analysis. This strong performance positions PPP1R1C as a promising biomarker, warranting further investigation to understand its role in PE pathology and to clarify the reasons for the observed inconsistencies in FSTL3 findings.

The observed discrepancy between RNA-seq data showing PPP1R1C overexpression and RT-PCR results indicating its downregulation necessitates a more detailed analysis.​ This difference is likely due to factors such as variations in sample preparation protocols, differences in assay sensitivity, or the influence of post-transcriptional regulatory mechanisms. Further discussion should explore potential post-transcriptional regulation. This could include miRNA-mediated repression, which can lead to mRNA degradation or translational inhibition, or methylation, which might affect gene expression without altering the mRNA sequence detected by RNA-seq. Such mechanisms could explain how a gene might appear overexpressed at the RNA level but downregulated at the protein or functional level. Moving forward, it is important to acknowledge the inherent complexity of PE, where various factors and pathways interact. Additional studies should not only aim to confirm PPP1R1C’s diagnostic accuracy and predictive value but also incorporate detailed investigations into the proposed post-transcriptional regulatory mechanisms. This comprehensive approach will provide a clearer understanding of PPP1R1C’s true role in PE.

FSTL3 is a secreted glycoprotein that belongs to the follistatin family, known for their roles in regulating the transforming growth factor-beta (TGF-β) superfamily [29]. FSTL3 is expressed in various tissues, including the heart, placenta, gonads, and pancreas [30]. The molecular functions of FSTL3 are diverse and context-dependent. It is capable of binding the activin-A protein and is involved in many inflammatory processes in the body [30]. Normally, FSTL3 performs a protective role, preventing airway remodeling. FSTL3 also gets involved in the regulation of glucose and lipid homeostasis [29]. However, FSTL3 has been discovered to promote lipid buildup and the release of inflammatory cytokines, as well as trigger the expression of CD36 and lipoxygenase 1 in foam cells [29]. FSTL3 has also been demonstrated as conducive to the development of lung, kidney and gastric cancer and applicable as a novel marker of poor survival. It has been reported that hypoxia enhances the expression of FSTL3 in term human trophoblast [31]. Aberrant expression of FSTL3 in PE led to the dysfunction of trophoblast, indicating its involvement in the pathogenesis of PE [28]. A recent study has identified that elevated circulating FSTL3 levels are tightly correlated with postpartum cardiovascular dysfunction and severe maternal morbidity [32]. The findings underscore its complex and multifaceted role in various physiological and pathological processes.

PPP1R1C, also known as IPP5, is a key regulator of protein phosphatase 1 (PP1), a major serine/threonine phosphatase involved in a wide array of cellular functions (PPP1R1C Gene Set, 2008). PE is characterized by inadequate trophoblast invasion within a hypoxic microenvironment, disrupting the balance between cellular proliferation and apoptosis critical for placental development. Notably, trophoblasts in PE pregnancies often exit the cell cycle prematurely at the G1 phase, favoring apoptotic pathways over cell division [33]. Oxidative stress induced by hypoxia, marked by elevated reactive oxygen species, exacerbates this apoptotic tendency [34]. The protein PPP1R1C has been implicated in cell cycle inhibition and apoptosis, suggesting its dysregulation may contribute to impaired trophoblast function in PE [35]. Besides, aberrations in the WNT/β-catenin signaling axis are also involved in the pathogenesis of PE. PPP1R1C appears to function within a molecular network interconnected with WNT signaling, which regulates trophoblast differentiation and placental development [36]. In PE placental tissue, Wnt2 expression is notably downregulated while sFRP4—an antagonist of WNT signaling—is upregulated [37]. Moreover, increased WNT-1 and β-catenin expression suggest a role for this pathway in placental calcification, particularly in LOPE [37]. Lastly, complications, including eclamptic seizures, constitute a significant risk in PE. The condition is associated with decreased plasma osmolality and sodium levels, which can alter cerebral levels of glutamate—a key excitatory neurotransmitter [38]. Reduced glutamate concentrations have been observed in women with PE. PPP1R1C modulates glutamatergic signaling through its interaction with group I metabotropic glutamate receptors, impacting synaptic transmission [39]. Disruption of this signaling may underlie cerebral symptoms of PE, which also features heightened sympathetic activity and impaired baroreceptor reflexes.

Based on the existing literature, integrating the molecular and biological functions of FSTL3 and PPP1R1C with the enrichment analysis results of our study, we can hypothesize potential dysregulated pathways in PE. However, the protein encoded by PPP1R1C cannot be secreted, making its direct use as a non-invasive diagnostic indicator impossible. It has been shown that FSTL3 is significantly associated with the prognosis and progression of lung adenocarcinoma and the infiltration of immune cells. These findings suggest a potential link between FSTL3, immune dysregulation, and PE. Immune dysregulation is a proposed factor in PE, highlighting the need to explore molecular markers and their relationship with immune modulation. Additionally, one of the suggested mechanisms involved in PE is hypoxia. Studies have reported that hypoxia enhances the expression of FSTL3 in term human trophoblast. Moreover, FSTL3 inhibits the action of TGFβ ligands such as activin. It has been reported that FSTL3 is also an inhibitor of myostatin, which is a muscle growth inhibitor. It is important to note that the GSE dataset used in this study is not single-cell, which may lead to confounding factors in the analysis.​ Nevertheless, this study can provide a valuable reference for future exploration of the diagnostic value of secretable proteins in single cells for PE. For example, similar studies have investigated potential secretable markers [40], though it should be noted that these studies have not definitively confirmed or excluded that these secretory proteins are derived from the placenta and a specific cell population thereof.

In summary, FSTL3 could promote tumor immune evasion and attenuate response to anti-PD1 therapy in CRC. The current study suggests that FSTL3 promotes the infiltration and polarization of TAMs in CRC. As for PPP1R1C, PP1 is a major serine/threonine phosphatase that regulates a variety of cellular functions. Perturbations in this complex interplay could potentially disrupt trophoblast function and contribute to the pathogenesis of PE.

Limitations

It is important to acknowledge the limitations of this study to provide a comprehensive and balanced perspective. One key limitation is the discrepancy observed between the RNA-seq and RT-PCR results for PPP1R1C. While RNA-seq initially indicated overexpression, RT-PCR showed significant downregulation. This inconsistency could stem from various factors, including differences in sample preparation protocols, varying sensitivities of the assays, or the influence of post-transcriptional regulatory mechanisms that are not fully captured by RNA-seq. Another limitation pertains to the nature of the RNA-seq datasets used. The study primarily utilized bulk RNA-seq data, which provides an average expression profile across a heterogeneous population of cells. This approach may obscure cell-specific expression patterns or subtle changes in gene expression within particular cell types that are critical to PE pathogenesis. Future research employing single-cell RNA sequencing (scRNA-seq) could address this by offering a more granular understanding of gene expression dynamics and identifying potential secretable proteins from specific placental cell populations.

Finally, while the study identified PPP1R1C as a promising diagnostic biomarker, it is crucial to note that the protein encoded by PPP1R1C is not secreted. This non-secreted nature precludes its direct use as a non-invasive diagnostic indicator in clinical settings, such as through blood tests. Further investigation is needed to explore potential secreted surrogate markers or other non-invasive methods that reflect PPP1R1C’s expression and diagnostic value.

The study’s generalizability is limited by the sample size and patient demographics. The RT-PCR validation was performed on a relatively small cohort. Additionally, the manuscript does not specify the ethnic or racial diversity of the participants, which is a crucial factor given that PE incidence and severity can vary across different ethnic groups. A larger, more diverse patient cohort would enhance the robustness and applicability of these findings to a broader population.

Conclusion and future directions

While initial investigations suggested a multi-gene panel as a potential biomarker for PE, subsequent validation studies revealed PPP1R1C as a standout diagnostic marker. The marked downregulation of PPP1R1C in placental tissue of PE patients, underscored by its impressive AUC value, positions it as a promising candidate for diagnostic strategies. The individual application of PPP1R1C could be ideal as it presents better results with minimum cost and effort, and its implications extend beyond mere diagnostics, hinting at a pivotal role for PPP1R1C-mediated pathways in the pathophysiology of PE.

Further investigation is needed to explore potential secreted surrogate markers or other non-invasive methods that reflect PPP1R1C’s expression and diagnostic value, especially given that the protein encoded by PPP1R1C is not secreted, precluding its direct use in non-invasive clinical tests. Comprehensive investigations are essential to fully elucidate the functional consequences of PPP1R1C dysregulation in the context of PE, prioritizing understanding how altered PPP1R1C expression impacts trophoblast function, placental development, and maternal-fetal interactions. Moreover, future investigations could benefit from further detailing explicit matching strategies or multivariate adjustments for these key confounders to enhance the generalizability and robustness of the findings. In this regard, single-cell RNA sequencing represents a promising future direction, as it would allow precise determination of PPP1R1C expression within specific trophoblast subtypes and help clarify its cell-type specificity, whereas in the current study bulk RNA-seq was used to develop a diagnostic biomarker from whole placental tissue without separating individual cell populations.

In conclusion, our exploration into the molecular mechanisms of PE has shed light on the distinctive diagnostic capabilities of PPP1R1C, showing that placental downregulation of PPP1R1C mRNA effectively distinguishes PE patients from healthy controls. This indicates its promise as a diagnostic biomarker, though further research is required to translate these findings into widespread clinical application.

Authors’ contributions

Study workflow was designed by A.E and M.Z-D. In-silico analysis was carried out by A.E and Z.H was responsible for sample collection. A.E performed the molecular experiments and A.M supervised the process. Statistical analysis was accomplished by A.E. A.E and M.Z-D drafted the manuscript and M.Z-D revised the final version. All authors have approved the final version of manuscript.

Funding

This study was approved and financially supported by Baqyatallah university of Medical Sciences (grant number: 402000518).

Data availability

The analyzed datasets are publicly available in NCBI.GEO database; however, other analyzed and generated data will be available upon reasonable demand. R codes are available in supplementary file 4.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Baqiyatallah University of Medical Sciences (Ethics approval code: IR.BMSU.BLC.1403.064) and was conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants prior to their inclusion in the study.

Consent for publication

All participants consented to publishing of results through written consent form.

Competing interests

The authors declare no competing interests.