The centerpiece of the work is a causal inference strategy built on Mendelian randomization, a statistical technique that exploits naturally occurring genetic variants as instruments to test whether a molecule genuinely influences disease risk, rather than merely correlating with it. The team drew on plasma proteomic data from two of the largest protein-genetic resources in existence: the deCODE cohort from Iceland, encompassing 35,559 individuals, and the UK Biobank pharmaceutical proteomics resource, covering 54,219 participants. In these datasets, genetic variants that alter circulating levels of thousands of proteins have been mapped, allowing the researchers to ask whether genetically predicted protein abundance is associated with genetic risk of premature ovarian insufficiency, using summary statistics drawn from the MRC IEU OpenGWAS database.

Because Mendelian randomization can be confounded by the phenomenon of linkage disequilibrium, in which a genetic instrument for one protein also tags a nearby gene, the authors added Bayesian colocalization analysis. This method tests whether the same causal variant drives both the protein level and the disease association, filtering out spurious signals. Triangulating across these two independent proteomic cohorts, the analysis converged on seven high-confidence candidate targets: EPHA4, FSTL3, NUCB2, OXT, SERPINA12, TNFRSF6B and FABP1. Each of these proteins passed the dual-cohort causal test, making them far stronger candidates than targets nominated by a single dataset alone.

Genetic evidence alone, however, says nothing about where in the ovary a protein acts or in which disease context it becomes relevant. To address this, the team turned to single-cell RNA sequencing, examining expression of the seven genes across ovarian cell populations. This cellular-level validation matters because the ovary is a complex tissue containing oocytes, granulosa cells, theca cells, stromal cells and immune cells, and a circulating protein flagged by blood proteomics may or may not reflect activity in the follicles that govern reproductive function. The integration of population genetics with cell-type-resolved transcriptomics is part of a broader movement in the aging field, where recent single-cell atlases of the aging mouse and human ovary have begun to reveal which cell populations deteriorate first.

The most striking experimental validation came from disease models. Three of the seven targets, EPHA4, FSTL3 and NUCB2, were significantly dysregulated in both cisplatin-induced mouse models of ovarian insufficiency and in human granulosa cell models exposed to the chemotherapeutic agent, with statistical significance ranging from P less than 0.05 to P less than 0.001. Cisplatin, a widely used chemotherapy drug, is a well-established trigger of ovarian damage, making these models clinically relevant to the many women who experience treatment-induced ovarian failure. Beyond their dysregulation, the three genes showed remarkable diagnostic performance, with area under the receiver operating characteristic curve values of 0.92 to 0.96, a range that approaches the accuracy expected of clinically deployable biomarkers.

The biology of these three lead targets is intriguing. EPHA4 belongs to the Eph receptor family of tyrosine kinases, signaling molecules previously implicated in tissue repair and fibrosis through their interactions with ephrin ligands. FSTL3, follistatin-like 3, is a secreted antagonist of activin and other TGF-beta family ligands, and earlier transgenic mouse work showed that overexpression of FSTL3 in gonads causes defects in gonadal development and function, a finding that dovetails strikingly with the new causal genetic evidence. NUCB2, the precursor of the anorexigenic peptide nesfatin-1, has been linked to the metabolic control of reproduction and is regulated by gonadotropins in the mouse ovary and uterus. Together, the targets point toward fibrosis, inflammation and metabolic dysregulation as central processes in premature ovarian insufficiency, a picture consistent with the functional enrichment analyses reported in the study.

Having nominated causal targets, the researchers then asked a question with immediate translational potential: are there existing drugs that could modulate these proteins? Through computational pharmacology, they performed molecular docking, predicting how small molecules fit into the three-dimensional structures of the target proteins, using structural models from the Protein Data Bank and AlphaFold alongside chemical libraries such as PubChem. The strongest predicted interaction was between cycloheximide and EPHA4, with a binding energy of minus 7.8 kilocalories per mole. To ensure the docked poses were not computational artifacts, molecular dynamics simulations were run on the top complexes, and the protein-ligand systems remained stable, with root mean square deviation values below 2.0 angstroms, indicating that the predicted binding modes are physically plausible.

Drug repurposing is an attractive strategy for a neglected condition like premature ovarian insufficiency because established drugs come with known safety profiles, manufacturing routes and, in some cases, off-label clinical experience, dramatically shortening the path to first-in-patient studies compared with de novo drug discovery. The docking and dynamics work described here provides a structural starting point for medicinal chemistry, whether that means repurposing an existing compound or optimizing its scaffold into a lead with better affinity and selectivity for EPHA4. The authors are careful to frame these computational results as a foundation rather than a proof: docking scores and simulation stability are hypotheses that must be tested in granulosa cell assays and animal models before any clinical consideration.

The multiomics-to-therapy pipeline itself may prove as influential as any individual target it identified. By chaining together proteome-wide Mendelian randomization, colocalization, single-cell expression mapping, disease-model validation, diagnostic performance assessment and structure-based drug prediction, the study offers a template that could be applied to other forms of ovarian aging and to age-related conditions more broadly. GeroScience, the journal publishing the work, focuses on the biology of aging, and the framing here is significant: premature ovarian insufficiency is increasingly understood as an accelerated form of ovarian aging, so targets and drugs identified for the disease may illuminate the normal aging process of the ovary as well.

Considerable work remains before these findings reach the clinic. The proteomic datasets used have restricted access and reflect circulating proteins, which may not perfectly mirror ovarian tissue biology; the diagnostic accuracy figures require validation in independent patient cohorts; and the candidate drugs must demonstrate efficacy and safety in preclinical systems. Nevertheless, the study delivers something the field has lacked: a shortlist of genetically supported, experimentally validated, structurally actionable targets for a condition that has long been managed without mechanism-based therapies. For women facing early loss of ovarian function, and for researchers probing the fundamental biology of reproductive aging, the seven proteins now on the table represent the most concrete starting points in years.

Subject of Research: Identification of druggable targets and repurposable drugs for premature ovarian insufficiency using multiomics and causal inference

Article Title: Identifying novel druggable targets and repurposable drugs for premature ovarian insufficiency by integrated multiomics and causal inference analysis

Article References: Liu, C., Wang, R., Liu, X., Li, Y., Ren, L., Zhao, Z., Fan, Z., & Xiao, J. (2026). Identifying novel druggable targets and repurposable drugs for premature ovarian insufficiency by integrated multiomics and causal inference analysis. GeroScience. https://doi.org/10.1007/s11357-026-02491-6

Image Credits: AI Generated

DOI: 10.1007/s11357-026-02491-6

Keywords: premature ovarian insufficiency, ovarian aging, Mendelian randomization, proteomics, single-cell RNA sequencing, molecular docking, drug repurposing, EPHA4, FSTL3, NUCB2, biomarkers, GeroScience