CPT1A, CYGB, and MRPL3 in pre-eclampsia: biomarkers identified from oxidative stress-related gene analysis Authors: Ling Chen, Meiting Wu Published: 2026 ABSTRACT BackgroundThe involvement of oxidative stress (OS) in pre-eclampsia (PE) has been reported, and the present study probed into the OS-related feature genes in PE.MethodsThe dataset GSE60438 was used to identify the OS-related features in PE, and corresponding feature genes were identified using machine learning algorithms including weighted gene co-expression network analysis (WGCNA), Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis and support vector machine-recursive feature elimination (SVM-RFE). The diagnostic value of the feature genes and their correlation with immune infiltration were explored. Gene set enrichment analysis (GSEA) was performed to identify enriched pathways of these genes. Furthermore, regulatory networks involving the feature genes were constructed. The role of these genes in hypoxia/reoxygenation (H/R)-induced trophoblasts was investigated.ResultsTwo WGCNA-identified modules (MEpurple and MEturquoise) were intersected with OS-related genes, yielding 359 common genes. Through machine learning algorithms and expression validation, three genes (CPT1A, CYGB and MRPL3) were identified and used to establish a diagnostic model, which reached an AUC of 0.930. The transcription factor (TF) SPI1 and multiple miRNAs formed a potential regulatory network for CPT1A, CYGB and MRPL3, and the expression of these three genes was closely associated with the infiltration of various immune cells (such as monocytes and macrophages M1) in PE. Cellular experiments confirmed that CPT1A silencing alleviated trophoblast dysfunction, reduced reactive oxygen species (ROS) levels, and decreased apoptosis under H/R conditions.ConclusionThe present study examined the OS features in PE, hoping to contribute to the management of PE. Introduction As a major cause of maternal and neonatal morbidity and mortality, pre-eclampsia (PE) is characterized by elevated blood pressure, proteinuria and edema (Dimitriadis et al., 2023; Montgomery et al., 2024; Guo et al., 2023). Advances in the prediction and prophylaxis of PE has contributed to the gradual evolution of this field worldwide (Kornacki et al., 2023). Current management of PE focuses on the alleviation of maternal symptoms and prolonging pregnancy fetal growth as long as possible, thereby reducing complications associated with iatrogenic premature delivery (Zhang et al., 2022). However, in low- and middle-income countries, where PE is prevalent and represents a major contributor to maternal mortality, using biomarkers for predicting PE are difficult to implement due to the scarcity of screening tools (Amelia et al., 2025). Furthermore, despite progress in the development of screening options, further work is needed before a screening test can be clinically implemented to predict the onset of PE (MacDonald et al., 2022). During the progression of PE, increased oxidative stress (OS) triggers the production of lipid peroxides and reactive oxygen species (ROS), leading to placental vascular endothelial dysfunction and ultimately contributes to PE development (Ahmad, Zimmerman & Moore, 2019). In addition to placental vascular endothelial dysfunction, OS can alter placental remodeling and cause ischemia/reperfusion (I/R) injury with an elevated xanthine oxidase activity (San Juan-Reyes et al., 2020). Therefore, targeting OS may represent a potential therapeutic method to improve both short-term and long-term maternal and fetal outcomes in patients with PE (Afrose, Alfonso-Sánchez & McClements, 2025). At present, a better and more detailed understanding of the mechanisms of action of proposed anti-OS strategies is critical for identifying companion biomarkers and personalized medicine approaches to developing effective treatments of PE. Driven by interest in identifying gene expression profiles, largely based on microarray data for diagnosis and risk prediction, one study suggested that PE may be a complicated condition influenced by maternal genes (Vennou et al., 2020). Another study employed sequencing data to identify key genes involved in the onset of PE (Li et al., 2024b). It is well recognized that public databases provide ample data for re-analysis and re-evaluation, thereby enhancing the utilization of available resources for more robust analyses (Clough et al., 2024; Egorova et al., 2025). Additionally, machine learning algorithms, particularly feature selection technique, are highly effective in revealing crucial information from high-throughput genomic data, thereby facilitating the identification of crucial genes involved in disease progression (Auslander, Gussow & Koonin, 2021; Cho & Kang, 2020; Shang et al., 2024; Zhang et al., 2024; Zhang et al., 2026). The present study aimed to analyze and screen feature genes related to OS in PE using relevant data from Gene Expression Omnibus (GEO) database (He et al., 2024), applying algorithms including weighted gene co-expression network analysis (WGCNA) and machine learning approaches. Subsequently, functional enrichment analysis and immune infiltration analyses were performed to elucidate the significance of these feature genes in PE. Further experimental assays were conducted to investigate their implication in PE. Analytical methods Data source The dataset GSE60438 (Transcriptome profiling of deciduas from pre-eclamptic and normotensive pregnancies) was downloaded from GEO database (Yong et al., 2015). Two datasets WG6v3 (the training set) and HT12v4 (the validation set) were obtained. Specifically, the WG6v3 dataset comprised 23 cases of normal basal decidua samples and 25 cases of basal decidua samples from PE patients, and the HT12v4 dataset comprised 42 normal basal decidua samples and 35 basal decidua samples from PE patients. Additionally, a total of 1,405 OS-related genes were retrieved from GeneCards database (https://www.genecards.org/) based on the correlation score, with the top 10% genes selected for further analysis. Identifying feature gene modules by WGCNA The R package “WGCNA” was applied to identify feature genes modules associated with OS in PE using the training set data (Langfelder & Horvath, 2008). Prior to network construction, the expression matrix was preprocessed to remove outlier samples and genes with low expression in order to improve the reliability of the analysis. The pickSoftThreshold function was subsequently applied to evaluate the scale-free topology fit across a range of candidate soft-thresholding powers. The minimum power achieving a scale-free topology fit index of 0.85 was selected as the soft threshold β for network construction. An adjacency matrix was computed using the chosen β and subsequently transformed into a topological overlap matrix (TOM). Genes were clustered based on the TOM dissimilarity, and modules were identified using the dynamic tree-cut method, with the minimum module size set to 200. Modules whose eigengenes showed high similarity were merged using a mergeCutHeight = 0.2. Finally, the correlation between each module eigengene and the clinical phenotype was calculated, and modules satisfying —cor— ≥ 0.3 with P < 0.05 were retained as candidate key modules for downstream analysis (Li et al., 2024a). Machine learning algorithms to reveal feature genes Genes from the WGCNA-identified modules of interest were intersected with OS-related genes from GeneCards database. The overlapping genes were then subjected to Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis using the R package “glmnet” (Engebretsen & Bohlin, 2019). Additionally, these genes were analyzed by the support vector machine-recursive feature elimination (SVM-RFE) algorithm using the R package “e1071” (Kebede, Le Cornet & Fortner, 2023). The results were selected based on the minimum error criterion, the overlapping genes identified by both algorithms were designated as key genes for further analyses. Evaluation of expression and the diagnostic potential of the key genes in PE Box plots were plotted to display the expression of the key genes in the samples of both training set and validation set, and the diagnostic potential of these genes was explored by developing a nomogram using the R package “rms” (Wang et al., 2024). The corresponding ROC curve, AUC value, calibration curve and decision curves were generated or calculated to evaluate the predictive potential and reliability of the nomogram. Immune infiltration analysis Immune infiltration of 22 types of immune cells was analyzed using the R package “CIBERSORT” based on data from the website of CIBERSORT (https://cibersortx.stanford.edu/) (Craven, Gökmen-Polar & Badve, 2021). The immune infiltration profile was compared between sample groups, and the correlation between key genes and immune infiltration was further assessed and visualized using a heatmap. Gene set enrichment analysis (GSEA) PE samples were divided into low- and high-expression groups by the median of the key genes. GSEA was performed using the GSEA software (version 4.4.0) (Wu et al., 2020) based on the background gene sets c2.cp.kegg_medicus and c5.go to elucidate the functional roles of the key genes in PE. Plotting the regulatory networks incorporating the key genes The potential transcription factors (TFs) modulating the key genes in the placenta tissues were predicted using the database hTFtarget (https://guolab.wchscu.cn/hTFtarget/#!/). Additionally, miRNAs targeting the key genes were obtained from the ENCORI database (https://rnasysu.com/encori/). The TFs and miRNAs validated in at least three studies were taken and the corresponding regulatory networks were constructed (He et al., 2025). Experimental validation Cell culture and hypoxia/reoxygenation (H/R) modeling: Roswell Park Memorial Institute-1640 medium (C3250, Solarbio, Beijing, China) added with 5% fetal bovine serum (S9020, Solarbio, China) was used to culture human trophoblasts cell line HTR8/SVneo (CRL-3271, ATCC, Manassas, VA) in the incubator with 5% CO2 at 37 °C. STR identification was performed on the cells, and the mycoplasma detection results were negative. For the modeling process, HTR8/SVneo cells were exposed to hypoxic conditions (2% O2) at 37 °C for 8 h to induce hypoxia, followed by reoxygenation for 16 h under normoxic conditions (5% CO2 in air) (Wang et al., 2025). Cell liposome transfection: To knockdown CPT1A in HTR8/SVneo cells, small interfering RNA (siRNA) against CPT1A (si-CPT1A #1, sequence: 5′-GCCATGAAGCT CTTAGACAAA-3′; si-CPT1A #2, sequence: 5′-CGTAGCCTTTGGTAAAGGAAT-3′) were synthesized and ordered from Sangon Biotech (Shanghai, China). Cells transfected with the scramble control siRNA were considered as the controls. Transfections were performed using lipofectamine 2000 transfection reagent (11668-027; Invitrogen, Carlsbad, CA, USA) according to the instructions. Cells were harvested 48 h later, and transfection efficiency was tested. Quantitative real-time PCR assay: TriZol reagent (15596-026; Invitrogen) was utilized to extract total cellular RNA from HTR8/SVneo cells following the manual, and RNA concentration was quantified in NanoDrop 2000 spectrophotometer (ND-2000; Thermo Fisher Scientific, Waltham, MA, USA). Following the synthesis of cDNA using the first-strand cDNA synthesis kit (K16325; Solarbio), quantitative PCR was performed in CFX384 touch real-time PCR system (Bio-Rad Laboratories, Inc., Hercules, CA, USA) using the SYBR Greem qPCR mix (SR1110; Solarbio). With GAPDH as a normalization control, the relative mRNA level was finally quantified using the 2−ΔΔCt method (Livak & Schmittgen, 2001). Primers were listed in Table 1. 10.7717/peerj.21572/table-1Table 1Sequences of primers for the PCR assay.GenePrimers (5′–3′)ForwardReverseCPT1AGATCCTGGACAATACCTCGGAGCTCCACAGCATCAAGAGACTGCGAPDHGTCTCCTCTGACTTCAACAGCGACCACCCTGTTGCTGTAGCCAA Sequences of primers for the PCR assay. Cell viability assay: Following the liposome transfection and H/R modeling, HTR-8/SVneo cells at 1 × 104 cells/well were planted into 96-well plates. After cell incubation for 24, 48 and 72 h at 37 °C, CCK-8 working solution (CA1210; Solarbio) was supplemented for another 1-h incubation at 37 °C. The optical density was recorded at 450 nm with a microplate reader (Bio-Rad Laboratories, Inc.). Cell migration assay: In detail, HTR-8/SVneo cells were seeded into 6-well plates at 1 × 105 cells/well until 90% confluence. Thereafter, a straight wound on monolayer was made with a sterile 100 µL pipette tip. Next, the cells were incubated in serum-free medium for 48 h. The gap area was imaged using an inverted optical microscope (Olympus, Tokyo, Japan), and the wound closure was analyzed by ImageJ software (version 1.48v; National Institutes of Health, Bethesda, MD, USA). Cell invasion assay: A total of 5 × 104 HTR-8/SVneo cells were suspended in 200 µL serum-free medium and plated into the upper chambers of Transwell inserts (pore size: 8 µm; #3422, Corning, Inc., Corning, NY, USA) coated with Matrigel (356234, BD Biosciences, Franklin Lakes, NJ, USA) at 37 °C for 1 h. The lower chambers contained normal culture medium with 10% serum. Following 48-h cell culture at 37 °C, the invaded cells were fixed with 4% paraformaldehyde (P1110; Solarbio) for 20 min at 37 °C and colored with 0.1% crystal violet (C8470; Solarbio) for 10 min at ambient temperature. The images were obtained under an inverted optical microscope (Olympus, Japan), and the invading cells were quantified using ImageJ software. Quantification on the ROS level: Cellular ROS level was measured as previously described (Chen, Wu & Zhou, 2024). Briefly, cells were stained with 10 µM 2,7-dichlorodihydrofluorescein diacetate (DCFH-DA) (D6470; Solarbio) for 20 min at 37 °C in the dark. ROS level was then analyzed with the CytoFLEX LX flow cytometer (Beckman Coulter, Indianapolis, IN, USA), and the mean fluorescence intensity (MFI) was recorded. Cell apoptosis assay: using a commercial kit (CA1020; Solarbio), apoptosis of HTR8/SVneo cells was determined via flow cytometry. After centrifugation, the supernatant was removed and cells were subsequently added with the working solution of Annexin V-FITC and propidium iodide for 20-min incubation at room temperature in the dark. CytoFLEX LX flow cytometer and FlowJo software (v. 11; FlowJo, LLC., Ashland, OR) were employed for final analysis. Statistical analyses Computational analyses were performed in R software (version 3.6.0; R Core Team, 2019) unless specified, and the laboratory data analyses were performed using GraphPad Prism software (version 8.0.2). Unpaired t-test was used for two-group comparisons. One-way ANOVA followed by Tukey’s multiple comparisons test and two-way analysis of variance (ANOVA) followed by Šídák’s multiple comparisons test were applied where appropriate. All experiments were repeated at least three times, and data were expressed as mean ± standard deviation (SD). A p < 0.05 stood for statistical significance. Results Identification of PE-relevant gene modules via WGCNA Data from the training set (WG6v3 data of dataset GSE60438) were analyzed by WGCNA to identify gene modules. First, by evaluating the scale-free topological fit index across a range of soft thresholds, the optimal soft threshold (β) for network construction was 10, at which the network structure conformed to a scale-free topology (Fig. 1A). A gene clustering dendrogram based on the topological overlap matrix was generated, and 10 co-expressed gene modules were identified using the dynamic tree-cut method, each assigned a distinct color (Fig. 1B). Additionally, Fig. 1C presents a heatmap of module-phenotype correlations. The MEpurple module (correlation coefficient: 0.37, P = 0.01) was significantly positively correlated with PE, while the MEturquoise module was negatively linked to PE (correlation coefficient: −0.37, P = 0.09). Therefore, these modules were selected as feature modules for subsequent analysis. Figure 1D shows the intersection of genes from these two modules with OS-correlated genes, yielding 359 overlapping genes. This result provided the basis for subsequent screening of key genes using machine learning algorithms. Identification of pre-eclampsia-relevant gene modules via WGCNA. (A) Soft-threshold power determination; (B) closely connected genes in the key modules in cluster dendrogram. (C) Relationships between the identified gene modules and PE based on the coefficients and p-values. (D) Identification of key genes from the gene modules MEpurple and MEturquoise and OS-correlated genes from GeneCards database. Machine learning algorithms on revealing the key genes Thereafter, two machine learning algorithms were applied to further identify key genes from the 359 common genes. Using LASSO regression analysis, 18 feature genes were selected when the error was minimized (lambda.min = 0.0759, Figs. 2A–2B). Based on SVM-RFE analysis, the minimum error was achieved when the number of gene was 16; therefore, 16 feature genes were selected (Fig. 2C). The intersection of the genes identified by the two methods yielded eight key genes: ATF4, SMN1, CPT1A, MRPL3, CYGB, HSPD1, CLN6 and HMGB1 (Fig. 2D). Machine learning algorithms on revealing the key genes related to pre-eclampsia. (A–B) LASSO regression analysis was applied as a machine learning algorithm to reveal the key genes related to pre-eclampsia. (C) SVM-RFE analysis was another machine learning algorithm to reveal the key genes related to pre-eclampsia. (D) Identification of the common genes related to both LASSO regression analysis and SVM-RFE analysis. Validation of the diagnostic potential of the key genes Subsequently, the expression level of these eight genes in the placenta samples of PE patients and healthy controls of both the training set and validation set were quantified and compared (Figs. 3A–3B). Only three genes (CPT1A, MRPL3, and CYGB) showed consistent expression trends across the two datasets and were therefore retained for further analysis. Specifically, in placental samples from PE patients, CPT1A and MRPL3 were significantly upregulated, whereas CYGB was significantly downregulated. Quantification on the levels of key genes in pre-eclampsia. (A–B) Expression trends of key genes CPT1A, CYGB, and MRPL3 in the placenta samples of pre-eclampsia patients and healthy controls of both the training set WG6v3 and validation set HT12v4. A nomogram incorporating the expression level of these three key genes was plotted to explore the diagnostic potential (Fig. 4A), and the corresponding ROC curve demonstrated a strong reliability of the nomogram (Fig. 4B, AUC = 0.93). Additionally, calibration curves and decision curves showed that the calibration curve closely aligned with the actual curve (Fig. 4C) and that the net benefit of the nomogram was evidently higher than any of the single gene (Fig. 4D). Validation on the diagnosis performance of the three key genes in pre-eclampsia. (A) Nomogram incorporating the level of the three key genes (CPT1A, CYGB, and MRPL3) plotted to explore the diagnostic potential. (B–D) The corresponding ROC curve (B), calibration curve (C) and the decision curve (D) for the nomogram. Correlation of the key genes with immune infiltration The immune infiltration in PE was analyzed using the training set. Significant differences were observed in the infiltration of four types of immune cells, including macrophage M1, T cells CD4 memory resting, T cells CD4 memory activated, and monocytes (Fig. 5A). Further correlation analysis revealed distinct correlation patterns. CPT1A was positively linked to the infiltration of T cells CD4 memory resting and macrophages M2, but negatively related to that of monocytes and macrophages M0. MRPL3 was positively correlated with the infiltration of T cells CD4 memory resting and macrophages M1, but negatively linked to that of monocytes and macrophages M0. CYGB was positively correlated with the infiltration of monocytes (Fig. 5A). Correlation of the key genes with immune infiltration in pre-eclampsia. (A) Relative abundance of immune cells in the placenta tissue based on the data of CIBERSORT. (B) Correlation of the key genes with immune infiltration in pre-eclampsia. *** p < 0.001, ** p < 0.01, and * p < 0.05. GSEA to reveal enriched pathways GSEA revealed that the three key genes were enriched to mitochondria-related pathways including mitochondrial translation, mitochondrial protein containing complex, and mitochondrial gene expression, along with VGCC CA2 apoptotic pathway (Figs. 6A–6C). GSEA on revealing the enriched pathways. (A–C) GSEA was performed to reveal the enriched pathways in pre-eclampsia patients with high/low CPT1A, CYGB and MRPL3 expression levels. Plotting the regulatory network based on the key genes Additionally, the hTFtarget database was applied to identify the common TF(s) regulating the three genes, and SPI1 was considered as the common TF. According to the ENCORI database, 14 and 13 miRNAs that potentially targeted CPT1A and MRPL3, respectively, with each targeting relationship validated in at least three studies, with each targeting relationship validated in at least three studies. In contrast, no miRNA targeting CYGB met the selection criteria. These regulatory relationship were uploaded into the cytoscape software to construct the regulatory network (Fig. 7). Plotting of the regulatory network based on the relevant transcription factors and miRNAs. The regulatory network was plotted based on the transcription factors which may regulate the key genes in the placenta from hTFtarget prediction and the miRNAs from the ENCORI database. Experimental validation on the involvement of CPT1A in H/R-modeled trophoblasts A series of laboratory assays were performed to validate the effects of CPT1A in H/R-modeled trophoblasts. An elevated level of CPT1A was observed in the modeled trophoblasts (Fig. 8A), and the corresponding knockdown assays were applied lower CPT1A expression (Fig. 8B). Furthermore, CCK-8, scratch and Transwell assays demonstrated that CPT1A silencing could promote the survival, migration and invasion of H/R-modeled trophoblasts (Figs. 8C–8E). Flow cytometry also showed that silencing CPT1A visibly reduced the ROS content and apoptosis of H/R-modeled trophoblasts (Figs. 8F–8G). Validation on the effects of CPT1A on hypoxia/reoxygenation-induced trophoblasts HTR8/SVneo. (A) The quantified CPT1A mRNA expression level in HTR8/SVneo cells with/out hypoxia/reoxygenation modeling. (B) Verification on the knockdown efficiency of CPT1A-specific small interfering RNAs. (C) The effects of CPT1A silencing on the viability of HTR8/SVneo cells following hypoxia/reoxygenation modeling detected by CCK-8 assay. (D) The migration of HTR8/SVneo cells following hypoxia/reoxygenation modeling detected by scratch assay. (E) The invasion of HTR8/SVneo cells following hypoxia/reoxygenation modeling detected by Transwell assay. (F) The ROS content of HTR8/SVneo cells following hypoxia/reoxygenation modeling. (G) The results of flow cytometry assay on the apoptosis of HTR8/SVneo cells following hypoxia/reoxygenation modeling. All procedures were performed in triplicate (n = 3). * vs. Control; ˆvs. si-NC. ˆp < 0.05, ˆˆp < 0.01, ˆˆˆp < 0.001, and **** p or ˆˆˆˆp < 0.0001. Discussion Although increasing attention has been paid to PE, no effective treatment for PE is currently available. Symptoms of PE can generally be alleviated only after placental delivery, which has therefore highlighted the importance of prediction and prevention (Roberts, 2024). In our in-silico analysis using WGCNA and the two machine learning algorithms identified three PE-related genes (CPT1A, MRPL3 and CYGB). Subsequently, a nomogram incorporating these three genes was created, showing a strong diagnostic potential for PE. Further cellular validation assays demonstrated an elevated level of CPT1A in the H/R-modeled trophoblasts, and that silencing CPT1A promoted the survival, migration and invasion yet reduced the ROS content and apoptosis of modeled trophoblasts. These data therefore underscored the therapeutic potential of the three genes in PE. Several previous reviews have revealed the mechanisms of physiopathogenesis of PE, including OS (San Juan-Reyes et al., 2020). In the placenta, OS is believed to arise from disruption of the balance between ROS production and detoxification (Wu et al., 2025). Subsequently, the by-products trigger inflammatory responses and are detrimental to cells, thereby leading to premature placental aging. Such placental prematurity may result in reduced functional capacity and abnormal pregnancy outcomes such as PE (Joo et al., 2021). Studies have used the data of GEO database to investigate genes implicated in PE progression and detection, with special attention paid to the relationship between OS and PE (Cao et al., 2024; Guo, Li & Xiong, 2025; Hamdan, 2025; Yu et al., 2025). Integration of data from GEO can increase sample size in a cost-effective manner; however, meticulous attention should be paid to methodological details to avoid compromising results. In the present study, the dataset GSE60438 was used to identify OS-related genes in PE, leading to the identification of three key genes (CPT1A, MRPL3, and CYGB) for subsequent analysis. The carnitine palmitoyltransferases family member CPT1A is a rate-limiting enzyme for fatty acid β-oxidation, and its inhibition has been demonstrated to reduce OS-induced chondrocyte senescence (Schlaepfer & Joshi, 2020; Jiang et al., 2022). Notably, our computational analysis revealed that CPT1A expression was elevated in PE placental tissues, a finding that aligns with recent experimental evidence demonstrating CPT1A upregulation in PE placentas and its role in promoting PE progression via the PI3K/AKT/mTOR pathway (Chen et al., 2023). MRPL3 is a mitochondrial ribosomal protein encoded by the nuclear genome and transported into mitochondria for ribosome biogenesis (Ahmad et al., 2025). CYGB is predominantly expressed in hepatic stellate cells and plays a critical role in attenuating mitochondrial OS (Okina et al., 2023). Concerning their association to PE, previous studies observed increased levels of CYGB in the serum of PE patients (Tayyar et al., 2019), while we detected reduced levels of CYGB in PE tissue. This discrepancy may reflect tissue-specific expression patterns or distinct regulatory mechanisms. The diagnostic model incorporating these three genes demonstrated strong diagnostic potential for PE. Furthermore, our cellular validation assays validated that CPT1A was upregulated in H/R-modeled trophoblasts, and that silencing CPT1A promoted cell survival, migration, and invasion while reducing ROS content and apoptosis. These functional observations were consistent with our computational findings of CPT1A overexpression in PE tissues and corroborate recent reports that CPT1A knockdown enhances trophoblast invasion and migration (Chen et al., 2023). Moreover, CPT1A-mediated fatty acid oxidation has been implicated in endothelial cell function and angiogenesis in PE (Wu et al., 2024), further supporting its pathogenic role. All these results suggest a direct connection between CPT1A overexpression in PE tissues and its functional influence on trophoblast behavior. Future studies are warranted to clarify the mechanism of the other two genes related to PE. Immune cells function crucially in supporting embryo implantation and developing maternal-fetal interface. In PE development, immune dysfunction and inflammation have been recognized as the key risk factors (Deer et al., 2023). Disruption of the complex immune microenvironment during pregnancy can lead to pro-inflammatory reactions, OS, and endothelial dysfunction, thereby potentially exciting PE and adverse outcomes (Aneman et al., 2020; Opichka et al., 2021). Based on the data of CIBERSORT, our current study unveiled significant difference in the infiltration of monocytes, T cells CD4 memory, macrophage M1. Additionally, correlation analysis between the key genes and immune infiltration revealed distinct patterns. Specifically, CPT1A was positively linked to the infiltration of T cells CD4 memory resting and macrophages M2, but yet negatively linked to that of monocytes and macrophages M0. MRPL3 was positively linked to the infiltration of T cells CD4 memory resting and macrophages M1, but negatively linked to that of monocytes and macrophages M0. CYGB was positively correlated with the infiltration of monocytes. Previous studies have unveiled the lower activation of memory CD4+ T cell subsets in and after PE, whereas altered functional activity of monocyte-macrophage system, key component of innate immunity, suggests a dysfunctional maternal immune response (Kieffer et al., 2019; Vishnyakova et al., 2019). These findings suggest a crucial role for these identified genes in immune dysfunction associated with PE. Our GSEA revealed that the three genes were enriched to several mitochondria-related pathways including mitochondrial gene expression, mitochondrial translation, mitochondrial protein containing complex, and VGCC CA2 apoptotic pathway. Notably, mitochondrial dysfunction has been observed in the pathogenesis of PE (Smith et al., 2021; Hu & Zhang, 2022). In particular, in PE placentas, mitochondrial fission mediated by hypoxia may activate mitophagy machinery, thereby promoting mitochondrial fragmentation and damages to placental tissues (Kobayashi et al., 2024). The mitochondrial calcium overload, additionally, has been considered as a pro-apoptotic pathway that induces the swelling of mitochondria (Giorgi et al., 2012). However, calcium deficiency in women is related to a higher PE risk (Lu et al., 2024). These findings suggest potential biological processes underlying PE and provided a reference for the development of management strategies. Nonetheless, there are some limitations that should be addressed. First, the screening of key genes is primarily based on a single public database and specific machine learning algorithms, which may introduce dataset-specific and algorithmic biases. We plan to integrate multiple independent datasets and employ diverse computational methods for cross-validation to enhance the robustness and generalizability of the screening results. Second, the diagnostic model was evaluated only using the training set and has not been validated using an independent external dataset. We plan to collect multicenter clinical samples and utilize additional public data resources to for external validation to objective assess the model’s generalizability. Third, although transcription factors and miRNAs regulating CPT1A, MRPL3, and CYGB were predicted using computational analyses, their regulatory relationships have not yet been experimentally validated. In the future, we plan to use techniques such as dual-luciferase reporter assays and chromatin immunoprecipitation to clarify direct interactions among these molecules. Finally, the application of the three identified genes to early pregnancy screening still faces challenges regarding test sensitivity and stability. Future study could integrate these genes with higher-precision techniques, such as single-cell sequencing, and evaluate their clinical value as early predictive biomarkers in prospective cohorts. These improvements are expected to further solidify the scientific value and clinical translation potential of the findings from this study. Conclusion Collectively, the present study revealed three OS-related key genes (CPT1A, MRPL3 and CYGB) and unraveled their diagnostic potential and association with the immune microenvironment in PE. Further explanation using H/R-modeled trophoblasts has demonstrated the potential involvement of CPT1A in PE. These findings further provided data for the development of relevant prevention and management strategies for PE. Supplemental Information MIQE checklist