Preeclampsia, a pregnancy-specific hypertensive disorder affecting 2-8% of pregnancies globally (1), is a major contributor to maternal mortality, accounting for approximately 18% of direct maternal deaths (2). Characterized by new-onset hypertension and potential damage to vital organs like the liver and kidneys (3), the prevalence and severity of preeclampsia underscore the urgent need to understand its underlying causes. Despite extensive research, the precise mechanisms driving its development remain elusive, hindering the development of effective early diagnostic tools and curative treatments beyond delivery (2).

Among the factors that may contribute to the pathogenesis of preeclampsia, cellular senescence has gained increasing attention. Characterized by permanent cell cycle arrest (4), cellular senescence has been implicated in various pregnancy-related complications, including preeclampsia (5). Abnormal expression of cellular senescence biomarkers is observed in the placentas and peripheral blood of preeclampsia patients. Moreover, elevated activity of cell cycle inhibitors not only induces cell cycle arrest but also promotes senescence (5), while reduced expression of senescence-suppressing factors further exacerbates the process (6). Furthermore, elevated circulating levels of senescence-associated secretory phenotype (SASP) factors (7), upregulated SASP gene expression in placentas from pregnancies complicated by preeclampsia and fetal growth restriction (FGR) (8), and links between placental lactate accumulation, histone lactylation, and premature trophoblast senescence (9) suggest that these interconnected biological processes may play a pathogenic role in the development of preeclampsia and FGR. However, much of the current evidence stems from observational or single-omics association studies, limiting the ability to establish causal relationships between specific senescence-related genes and preeclampsia development. Establishing causality is crucial for understanding disease mechanisms and identifying targeted interventions.

To address this challenge, our study employs a Mendelian Randomization (MR) framework, leveraging multi-omics data through the Summary-data-based Mendelian Randomization (SMR) approach. SMR integrates summary statistics from large-scale Genome-Wide Association Studies (GWAS) with molecular quantitative trait loci (QTL) data (e.g., expression QTLs - eQTLs, methylation QTLs - mQTLs, protein QTLs - pQTLs) to infer potential causal associations between molecular traits and disease risk (10). By using genetic variants robustly associated with molecular traits as instrumental variables (IVs), MR mimics a randomized controlled trial design, thereby minimizing biases from confounding and reverse causation that often affect observational studies. We incorporated the HEIDI (Heterogeneity in Dependent Instruments) test to detect potential horizontal pleiotropy (where a genetic variant affects the outcome via pathways independent of the exposure) (11) and colocalization analysis to assess whether identified associations between QTLs and preeclampsia GWAS signals are likely driven by shared causal variants, strengthening biological plausibility (12). This multidimensional analytical strategy overcomes the limitations of traditional single-omics research, offering a more comprehensive perspective for deciphering the molecular mechanisms of complex diseases.

Given the evidence linking cellular senescence to preeclampsia, we used a hypothesis-driven approach to deeply investigate this specific biological pathway. This method complements broader genome-wide screens by offering targeted insights into potential causal factors within the senescence network. Therefore, this study aims to systematically screen for and prioritize cellular senescence-associated genes with potential causal effects on odds of preeclampsia using the SMR methodology. By identifying potentially causal genes, our findings aim to provide a data-driven foundation of prioritized candidates for developing new biomarkers and therapeutic targets.

This study employed a multi-omics Mendelian Randomization approach to investigate the potential causal roles of cellular senescence-related genes in preeclampsia pathogenesis. Our findings implicate several genes operating through diverse pathways—including ATG16L1, MAP3K14, PMVK, CDC25A, and NSUN2—as potential mediators linking senescence processes to odds of preeclampsia. These genetic associations were further explored through tissue-specific MR analysis and RT-PCR in placental tissue.

Our MR analyses highlighted genes central to cellular stress and response pathways. Higher genetically predicted ATG16L1 expression, a key autophagy gene (37), was linked to increased odds of preeclampsia. While autophagy is often protective, its dysregulation (either insufficient or excessive) is implicated in placental pathology (38), potentially impairing trophoblast function or modulating senescence-associated inflammation (39, 40). This aligns with our preliminary RT-PCR results, which suggested ATG16L1 upregulation in preeclamptic placentas compared to controls. For MAP3K14, involved in NF-κB signaling and inflammation (41), lower genetically predicted expression was associated with increased risk. This finding, potentially supported by the observed trend of MAP3K14 downregulation in our placental samples, hints that impaired NF-κB signaling or altered inflammatory resolution might contribute to preeclampsia, possibly intersecting with SASP regulation (7, 8). The mevalonate pathway gene PMVK also emerged, with higher genetically predicted expression (linked to higher methylation at cg16318349) associated with lower odds of preeclampsia. Our RT-PCR analysis showed a trend towards PMVK downregulation in PE placentas, consistent with the MR prediction that lower expression confers higher risk. While some studies link PMVK downregulation to p53 activation and cell cycle arrest (42), its precise role in the placenta, potentially involving p53-independent metabolic or signaling functions or specific effects of cg16318349 methylation, warrants further investigation to resolve this apparent complexity. While the differential expression analysis does not confirm causality, it demonstrates that the gene dysregulation predicted by our genetic analysis is present at the tissue level in the disease state, thus providing important downstream support for its potential involvement.

It’s worth noting that the expressions of 2 genes (NEDD4 and ERBB2) showed strong colocalization evidence under a stringent threshold (PPH4 > 0.8). The NEDD4 gene encodes an E3 ubiquitin protein ligase. It plays a crucial role in various cellular processes, including protein degradation, cell signaling, cell differentiation, and cell cycle regulation (15021885). In the context of preeclampsia, NEDD4 is involved in the regulation of trophoblast necrosis by mediating the ubiquitination of TAK1, thereby affecting placental development and function (43). Additionally, the NEDD4L gene is associated with hypertension and epithelial sodium transport (44). Given that preeclampsia is a pregnancy-specific hypertensive disorder, the involvement of NEDD4L in hypertension suggests a potential link between NEDD4 and preeclampsia. On the other hand, the ERBB2 gene encodes a receptor tyrosine kinase and is a member of the epidermal growth factor receptor family. However, there is limited direct research on the relationship between the ERBB2 gene and preeclampsia. Given its role in cell growth and differentiation, ERBB2 may indirectly influence placental development and function, which are critical factors in the pathogenesis of preeclampsia.

Using uterine eQTL data as a proxy for maternal contributions, we found that higher genetically predicted expression of CDC25A and NSUN2 was associated with decreased odds of preeclampsia. CDC25A regulates cell cycle transitions, and its dysregulation is linked to senescence (45); a previous MR study also hinted at its involvement in preeclampsia (46). NSUN2, an RNA methyltransferase, adds to growing evidence implicating RNA modifications in placental development and function, such as decidualization (47, 48). The observed trends of CDC25A and NSUN2 downregulation in our preliminary RT-PCR analysis of preeclamptic placentas align with these protective MR associations. Collectively, these findings suggest that disruptions in placental cell cycle control (CDC25A) and RNA modification pathways (NSUN2) may contribute significantly to preeclampsia pathology.

Synthesizing these results, dysregulation across interconnected pathways—autophagy (ATG16L1), inflammation (MAP3K14), metabolism (PMVK), cell cycle (CDC25A), and RNA methylation (NSUN2)—appears to converge on promoting placental cellular senescence, contributing to preeclampsia pathophysiology. Senescence, particularly in trophoblasts, can impair placentation via multiple routes, including SASP-mediated inflammation disrupting spiral artery remodeling and causing endothelial dysfunction (49), and intrinsic impairment of trophoblast proliferation, migration, and differentiation. Connecting these senescence-associated genes and pathways to the broader pathophysiology of preeclampsia highlights potential cross-talk with these core mechanisms. For instance, the SASP can exacerbate systemic inflammation and endothelial dysfunction, key features of preeclampsia. Additionally, metabolic alterations (PMVK) may contribute to oxidative stress, further impairing placental function. Investigating functional links between these senescence-related genes and established preeclampsia susceptibility loci (e.g., near FLT1 (17)) represents an important future direction.

The identification of these candidate genes and pathways offers potential avenues for clinical translation, such as novel biomarkers (e.g., methylation patterns, circulating RNAs, SASP components) or therapeutic targets. However, considerable hurdles remain. Biomarker validation requires large, diverse prospective cohorts to establish sensitivity, specificity, and clinical utility. Therapeutic modulation of these pathways (e.g., using autophagy modulators, anti-inflammatories, senolytics) faces significant challenges regarding efficacy and, critically, safety during pregnancy, necessitating rigorous preclinical evaluation.

A key strength of our approach was the integration across multiple omics levels. The analysis identified CpG methylation sites apparently influencing gene expression, where both methylation and expression were associated with odds of preeclampsia, provides compelling candidates for epigenetic regulation in the disease. For ATG16L1, higher methylation at cg19193136 was unexpectedly associated with both higher gene expression and increased odds of preeclampsia. Conversely, for MAP3K14, higher methylation at cg08823240 was linked to lower expression but higher disease risk, aligning more closely with canonical gene silencing. The PMVK association (higher methylation linked to higher expression and lower risk) also deviated from simple promoter-silencing models. These complex methylation-expression relationships, particularly the positive correlations, underscore the need for further investigation (50). Potential mechanisms include methylation within gene bodies or enhancer regions affecting regulatory element binding, the influence of 5-hydroxymethylcytosine (5hmC) (51). Methylation in enhancer regions can either promote or stabilize the binding of transcriptional activators, thereby positively influencing gene expression (52). Additionally, 5hmC, which is often associated with active gene expression, may contribute to the regulation of ATG16L1 expression through distinct mechanisms compared to traditional 5-methylcytosine. Elucidating these precise epigenetic mechanisms requires targeted studies mapping CpG locations relative to functional elements.

Several limitations must be acknowledged. Firstly, our SMR findings showed limited replication in the independent GCST90301704 cohort. While this could be due to insufficient power in the replication sample or false positives in our discovery analysis, it may also reflect the significant clinical and etiological heterogeneity of preeclampsia. Our discovery cohort (FinnGen) used a broad, EHR-based definition of preeclampsia, whereas the replication cohort employed more stringent clinical criteria. This difference could lead to discordance if the identified genes are associated only with specific subtypes of the disease. This highlights the challenge of replicating genetic associations for complex, heterogeneous syndromes. Secondly, the RT-PCR experiments had a small sample size, particularly in the control group (n=5), which limits the statistical power and reliability of the findings. This increases the risk of both Type I (false positives) and Type II (false negatives) errors. Although the observed expression trends for our candidate genes are consistent with our MR predictions, these results are preliminary and must be interpreted cautiously. Thirdly, a major limitation is the use of blood and uterine QTL data as proxies for the placenta. This tissue-context mismatch could significantly impact causal inference, as we may have missed true causal genes regulated by placenta-specific QTLs that are inactive in blood and uterus. Conversely, an observed association could be misleading if a genetic variant affects gene expression differently in the placenta compared to the proxy tissues. Future research must prioritize placenta-specific multi-omics QTL datasets to validate and build upon our findings. Fourthly, our study found a discrepancy between significant SMR results and non-significant colocalization results for top candidate genes like ATG16L1 and CDC25A. This highlights the complementary nature of these methods. SMR tests for causal associations, while colocalization assesses shared causal variants. The lack of colocalization does not invalidate SMR findings but suggests a complex genetic architecture, such as causality driven by different variants in high LD. Thus, we consider SMR results as primary evidence for a causal link, acknowledging that further fine-mapping and functional studies using placenta-specific data are needed to elucidate the precise mechanisms. Fifthly, a major limitation is the lack of statistical significance after correction for multiple testing. This is common in under-powered MR omics studies. As a result, the aims of this study should be generating hypothesis, requiring substantial validation. Sixthly, sample overlap between exposure and outcome populations can potentially bias MR estimates. However, since our primary instruments were strong (F > 10), this bias is expected to be minimal. Finally, standard MR limitations apply, including the possibility of residual confounding or pleiotropy despite statistical tests (SMR/HEIDI, colocalization) and the reduced generalizability due to the primary reliance on European ancestry datasets.

Future research should prioritize validating these findings in larger, multi-ethnic and subtypically graded meta analyses and utilizing placenta-specific multi-omics QTL data when available. Prospective studies correlating biomarkers derived from these pathways with pregnancy outcomes are needed. Combining data from cohorts with detailed clinical information will be essential to explore the specific genetic architecture of different preconceptions. Ultimately, functional studies in relevant cell and animal models are essential to confirm the causal impact of ATG16L1, PMVK, MAP3K14, NSUN2, and CDC25A on senescence phenotypes and preeclampsia pathogenesis.

In conclusion, this multi-omics MR study, combined with preliminary experimental insights, pinpoints ATG16L1, PMVK, MAP3K14, NSUN2, and CDC25A as key candidate genes potentially mediating the link between cellular senescence pathways and odds of preeclampsia. While requiring further validation, these findings provide a valuable foundation for future research into the molecular mechanisms of preeclampsia and the development of targeted interventions.