Single-Cell Immune Cis-eQTL Mendelian Randomization Prioritizes Cell-Type-Resolved Candidate Causal Genes for Endometriosis

Endometriosis has a substantial immune component, but the cellular contexts of genetically regulated disease susceptibility remain incompletely resolved.

We tested single-cell cis-eQTL instruments from 20 OneK1K peripheral-blood cell types against FinnGen R12 endometriosis (20,190 cases; 130,160 controls) using single-variant Mendelian randomization (MR), colocalization and sensitivity analyses. Benjamini–Yekutieli (BY) correction distinguished conservative findings from an exploratory Benjamini–Hochberg (BH) screen. Directional consistency, bulk-blood MR and lesion expression provided contextual comparisons.

Across 9157 unique gene–cell-type tests, 135 associations in 61 genes reached BH-FDR < 0.05. Colocalization at the default prior prioritized 10 genes (6 protein-coding) across 28 pairs. LOC102723878 and RUVBL1 were also BY-robust; IGF1R in natural killer (NK) cells was BH-supported but not BY-robust (PP4 = 0.89; OR 1.11 per SD higher expression). Its NK attribution reflects instrument availability and does not establish exclusivity. Colocalization was prior-sensitive. Directional concordance in a partly overlapping sample (8/8) was descriptive. Bulk-blood MR nominated 3/6 testable protein-coding genes. IGF1R lesion expression showed a nonsignificant upward trend (P = 0.101).

The analysis prioritizes immune-context candidate genes for functional study. IGF1R remains an exploratory, mechanistically unresolved nomination. Genetic support and protein tractability do not establish therapeutic benefit or reproductive safety; experimental validation is required.

endometriosis, Mendelian randomization analysis, quantitative trait loci, single-cell gene expression, killer cells, natural, receptor, IGF type 1

Endometriosis, the growth of endometrium-like tissue outside the uterine cavity, affects approximately 10% of women of reproductive age and is a leading cause of chronic pelvic pain, dysmenorrhea, and infertility.1–3 Despite its prevalence and morbidity, diagnosis is frequently delayed by years, and medical management remains dominated by hormonal suppression, which is contraceptive, incompletely effective, and poorly tolerated for long-term use.1 There is a pressing need for non-hormonal, mechanism-based therapeutic hypotheses, ideally ones that can be connected to pharmacology that already exists. Although retrograde menstruation seeds ectopic endometrial fragments in most women, only a minority develop endometriosis, implicating a failure of immune clearance in disease establishment.1 The immune dimension is well documented:4 cytotoxicity of peripheral and peritoneal natural killer (NK) cells is reduced in affected women, impairing elimination of ectopic cells;5,6 peritoneal macrophages are polarized toward a lesion-supportive phenotype;7 and altered T-cell subsets and B-cell-associated responses have also been described, although findings vary by anatomical compartment and menstrual-cycle phase and the role of B cells remains uncertain.8 Local cytokines, including TGF-β, IL-6 and IL-10, can suppress NK-cell function.6 These observations frame endometriosis as, in part, a disorder of immune surveillance, but they do not identify which genes, in which immune cell types, causally drive risk. Human genetics offers a route to causal genes. Genome-wide association studies (GWAS) have identified dozens of endometriosis risk loci and implicated hormonal and inflammatory pathways,9–11 yet, as for most complex traits, the great majority of associated variants are non-coding and are presumed to act by altering gene expression.12 Translating a non-coding association into a specific effector gene, and into the cell type in which that gene acts, remains the central bottleneck. Mendelian randomization (MR) addresses the first half of this problem. By using cis-expression quantitative trait loci (cis-eQTLs) as instrumental variables for gene expression, cis-eQTL (or “drug-target”) MR estimates the effect of a gene’s expression on disease and has become a workhorse for genetic target prioritization.13–16 Because a single locus can harbor both an eQTL and a GWAS signal through linkage disequilibrium (LD) alone, MR is paired with statistical colocalization, which tests whether the two signals share one causal variant and thereby guards against LD-driven false positives.17–19 Applied to bulk tissue, this framework averages expression over constituent cell types and can obscure the cellular context of a regulatory association. Large single-cell and cell-type-resolved eQTL resources now make cell-type-resolved causal inference feasible.20 The OneK1K cohort mapped cis-eQTLs across peripheral-blood immune subsets in 982 donors;21 the eQTL Catalogue reprocesses these data uniformly, provides them as 20 cell-type sub-populations, and fine-maps them with SuSiE, yielding credible sets of candidate causal variants per gene and cell type.22,23 Coupling such instruments to MR and colocalization has recently been applied to other immune-mediated diseases.24 In endometriosis, a recent study integrated bulk peripheral-blood eQTL Mendelian randomization with single-cell transcriptomic localization to nominate candidate genes;25 a complementary question is how single-cell eQTL instruments, paired with colocalization, can refine genetic prioritization across immune-cell contexts in this disease. The insulin-like growth factor (IGF) axis provides a concrete motivation: IGFs stimulate proliferation of endometrial stromal cells,26 and the IGF1R protein has been investigated as a therapeutic target in oncology,27 raising the prospect that cell-resolved genetics could connect this axis to a specific immune context. Here we integrate single-cell immune cis-eQTLs for 20 cell types with a large endometriosis GWAS to (i) prioritize candidate causal genes, (ii) describe the immune-cell contexts in which genetic evidence is detectable, and (iii) annotate their tractability and the evidence that supports or qualifies each nomination. We show that endometriosis risk is associated with genetically predicted expression in several immune-cell contexts, and we nominate NK-cell IGF1R as a prioritized, hypothesis-generating candidate whose NK-cell attribution derives from the cell-type-resolved exposure data. Throughout, we distinguish statistical prioritization from therapeutic readiness: none of the genes reported here is a therapy-ready target.

Study Design, Ethics and Reporting This is a two-sample, cell-type-resolved cis-eQTL Mendelian randomization study with colocalization and secondary analyses (Figure 1). Reporting follows the STROBE-MR guideline28 (Supplementary Table 1). This study analyzed publicly available GWAS and eQTL summary statistics from previously published human genetic studies, together with de-identified sample-level gene-expression measurements from GSE7305. No participants were recruited, no new biological samples were collected, and no directly identifying participant information was accessed. The original studies were responsible for their ethical-review and participant-consent procedures. In accordance with items 1 and 2 of Article 32 of the Measures for Ethical Review of Life Science and Medical Research Involving Human Subjects, issued jointly by the National Health Commission, the Ministry of Education, the Ministry of Science and Technology and the National Administration of Traditional Chinese Medicine of the People’s Republic of China on 18 February 2023 (Guo Wei Ke Jiao Fa [2023] No. 4), research using lawfully obtained public data or anonymized information may be exempt from ethical review when it causes no harm to humans and involves no sensitive personal information or commercial interests. This secondary analysis was conducted under that public-data and anonymized-information exemption provision. No new participant consent was sought for this secondary use of public data. The study was conducted in accordance with the Declaration of Helsinki. Data Sources, Provenance and Ancestry The characteristics, ancestry, sample sizes, data-collection context and release dates of every dataset are summarized in Table 1. Exposure data were the cis-eQTL summary statistics of the OneK1K cohort,21 982 donors of European ancestry recruited in Hobart, Tasmania, Australia, whose peripheral blood mononuclear cells were profiled by single-cell RNA sequencing (1,267,758 cells; study published April 2022); we used the eQTL Catalogue reprocessing of these data (study QTS000038, release files dated 20 May 2024), which provides 20 immune cell-type sub-populations uniformly processed, mapped to GRCh38 and fine-mapped with the sum of single effects (SuSiE) model.22,23 Per-cell-type sample sizes were read from the allele-number field of each summary-statistics file and ranged from 534 to 981 donors. The discovery outcome was the FinnGen data freeze 12 (R12; public release 4 November 2024; 500,348 participants of Finnish ancestry recruited through hospital and population biobanks with linkage to national health registers)29 endpoint N14_ENDOMETRIOSIS (20,190 cases; 130,160 female controls), defined from hospital-discharge and cause-of-death registers (ICD-10 N80, ICD-9 617, ICD-8 625.3), on GRCh38. Both the exposure and the outcome datasets are therefore of European ancestry; FinnGen is a Finnish founder population with partly distinct allele frequencies and linkage disequilibrium, so instruments were matched by genomic position and alleles rather than by rsID, and harmonization retained only verified forward-reference allele matches (below). The OneK1K donors are not part of FinnGen, so the two samples do not overlap. FinnGen’s active sample-collection phase was 2017–2023, with legacy biobank specimens and historical register follow-up extending beyond that collection window (FinnGen Data Handbook). Recruitment dates were not reported in the accessed OneK1K summary-statistic metadata. The replication outcome, the whole-blood eQTL meta-analysis, the leiomyoma endpoint and the lesion transcriptome dataset are described in their own sections and in Table 1. | Table 1 Characteristics and Provenance of the Data Sources | Instrument Selection, Linkage Disequilibrium and Credible Sets For each gene in each cell type, one cis-eQTL instrument was selected as the variant with the maximum posterior inclusion probability (PIP) across all available SuSiE 95% credible sets, without first selecting a set by marginal P value. Ties retained the first record in the source-file order. SuSiE accounts for local LD in fine-mapping; retaining only one instrument per gene–cell-type exposure obviated within-exposure LD clumping. It does not imply independence across genes or cell types. Instruments were restricted to autosomes, exposure P 10.13,14,30 Alleles were coordinated using GRCh38 chromosome, position and REF/ALT identities. We retained unambiguous forward-reference matches, including A/T and C/G variants when the reference and alternate alleles agreed across sources; allele complementation based on rsID alone was not used. Opposite REF/ALT indel representations were not treated as recordings of the same variant; unmatched identities were excluded, and one outcome record per retained gene–cell-type instrument was required. The harmonization audit and exclusions are included in the accompanying code and data archive; 9157 unique gene–cell-type tests remained. Mendelian Randomization and Non-Applicability of Multi-Instrument Sensitivity Analyses The association of a standard-deviation increase in genetically predicted expression with endometriosis was estimated by the single-variant Wald ratio, with first-order delta-method standard errors.31 MR-Egger regression,32 the weighted median33 and MR-PRESSO34 require multiple instruments per exposure and cannot be meaningfully estimated from one ratio estimate. We therefore used colocalization, an approximate directionality screen, a phenome-wide association lookup and cross-disease comparisons as complementary checks. These checks cannot rule out horizontal pleiotropy or prove the exclusion restriction, and this remains a limitation. Instrument Strength and Statistical Power Instrument strength was summarized as F = (β/SE)2 and the approximate variance explained R2 = F/(F + n − 2), using the cell-type-specific donor count.35 Conditional F statistics address conditional strength in multivariable settings; I2GX assesses dilution in MR-Egger with multiple instruments. Neither supplies an additional diagnostic for this single-exposure, single-variant ratio analysis. Binary-outcome power was approximated using:36,37 with δ = |log(OR)|√[N k(1−k)R2], power = Φ(δ−z[1−α/2]) + Φ(−δ−z[1−α/2]), N = 150,350 and k = 20,190/150,350. We report 80% power minimum detectable odds ratios and prespecified OR scenarios at nominal two-sided α = 0.05 and at α = 0.05/m, where m is the number of tested gene–cell pairs. The latter is a conservative multiplicity sensitivity analysis, not the power of the data-dependent BH procedure. Estimates rely on approximate R2 and may be optimistic after eQTL selection. Summaries are in Supplementary Table 2; complete instrument-level values accompany the code and data archive. Multiple Testing Under Dependence We used Benjamini–Hochberg (BH) FDR < 0.05 across all 9157 gene–cell-type tests as an exploratory screen.38 Because positive regression dependence on the subset of true null hypotheses (PRDS) is not established for this screen, Benjamini–Yekutieli (BY) correction under arbitrary dependence was used to distinguish more conservative findings.39 Counts of shared variants and within-gene signed-Z correlations describe observed dependence; they are not a proof of PRDS or BH calibration. Two supplementary analyses used BH across distinct chromosome-position-REF-ALT variant identities and within-gene Bonferroni correction over distinct instruments followed by BH across genes. Missing rsIDs were not treated as missing variant identities. Nearby genes can share regulatory loci, so the gene-level analysis does not establish independent replication or independent loci. Among genes passing the MR and colocalization screens, tier 1 denotes at least one BY-significant association; tier 2 denotes support from BH and gene-level criteria without BY robustness (Supplementary Table 3). Colocalization on Complete Regional Windows For every BH-significant gene–cell-type pair, and additionally for every other cell type in which a colocalizing gene was instrumentable, colocalization between the cis-eQTL and endometriosis association signals was tested with coloc.abf17 over a complete ±500 kb window centred on the instrument. Regional summary statistics were sliced from the tabix-indexed eQTL Catalogue release files; we note for transparency that an earlier version of this analysis had retrieved these windows through the now-retired eQTL Catalogue API, whose paging truncated them, and that every colocalization result reported here was recomputed on complete windows. Distinct variants were matched by GRCh38 chromosome, position, reference and alternate alleles to FinnGen, retaining different alternate alleles at the same position rather than deduplicating by position alone. The principal regional slices contained SNPs. For KLF12, whose MR instrument is an indel, we additionally retrieved the full regional slice including indels, checked reference alleles and left alignment/minimal representation, and report the inclusive analysis as primary with the SNP-only result as sensitivity (Supplementary Table 4E); the eQTL sample size was taken from each file and the outcome was modelled as case–control with case fraction 0.134. The coloc software default priors were used (p1 = p2 = 1 × 10−4, p12 = 1 × 10−5); colocalization was defined as a posterior probability of a shared causal variant PP4 > 0.8, with the posterior probability of two distinct causal variants (PP3) and the conditional posterior quantity PP4/(PP3 + PP4) reported alongside.19 Sensitivity to p12 (1 × 10−6 and 5 × 10−5) and, as an exploratory check, to the window width (±250 kb and ±100 kb) was recorded for every pair (Supplementary Table 4). coloc.abf assumes a single causal variant per trait in the window; loci at which either dataset carries several independent signals can therefore return a low PP4 even when one of the signals is shared, and we flag such loci where the published SuSiE fine-mapping of either dataset indicates multiple credible sets. Gene–cell-type pairs in the major histocompatibility complex (chr6:28.5–33.5 Mb) were excluded a priori from colocalization and target reporting. Directionality, Pleiotropy Screen and Tractability An approximate directionality screen compared the exposure R2 estimate with the outcome signal-strength proxy z2/N. This is a simplified Steiger-style comparison, not a liability-scale conversion or proof of causal direction.40 Lead instruments of the colocalized genes were looked up in the archived full GWAS Catalog associations snapshot41 (archive member dated 3 October 2026) as a coarse horizontal-pleiotropy scan, distinguishing reproductive and immune associations from potential off-target liabilities. Protein-coding colocalized genes were annotated for small-molecule and antibody tractability, including approved-drug and clinical-precedence categories, in the Open Targets Platform42 (accessed October 2026). We use “tractability” in its Open Targets sense—whether the encoded protein can in principle be modulated pharmacologically—and not as evidence that modulating it would treat endometriosis. Directional Consistency and Instrument Cross-Validation Colocalized genes were assessed for directional consistency in the endometriosis GWAS meta-analysis of Rahmioglu et al excluding 23andMe9 (GWAS Catalog GCST90205183, released 2023; 21,779 European-ancestry cases and 449,087 controls from Greece, the United States, Belgium, Finland, Poland, Denmark, the United Kingdom, Australia, Iceland and Estonia, plus 1713 Japanese cases and 1581 controls; 99.3% European). This sample partly overlaps the discovery cohort because it incorporates an earlier FinnGen data freeze; its estimates are therefore partly correlated with discovery and are interpreted as a consistency check rather than as independent replication.43 Lead instruments were matched by rsID, harmonized to the exposure effect allele and re-estimated by the Wald ratio; directional concordance with discovery was summarized descriptively, without treating correlated loci and overlapping samples as independent replications. Independently, the single-cell lead variants were cross-validated in the eQTLGen whole-blood cis-eQTL meta-analysis (31,684 individuals, predominantly of European ancestry; data release December 2019)44 by comparing the direction of the lead-variant effect on expression. Bulk Benchmark, Cross-Disease Comparison and Disease-Tissue Expression To ask whether a conventional bulk design would have nominated the same protein-coding genes, each was instrumented by its strongest eQTLGen whole-blood cis-eQTL and tested against FinnGen R12 endometriosis by the same Wald-ratio pipeline. Cross-disease associations were compared by re-estimating each colocalized instrument against uterine leiomyoma (FinnGen R12 endpoint CD2_BENIGN_LEIOMYOMA_UTERI; 42,107 cases and 239,957 controls), reporting whether genes reached BH-FDR < 0.05 for leiomyoma without interpreting a nonsignificant result as proof of disease specificity. Expression of the protein-coding genes at the disease site was examined in GSE730545 (Affymetrix HG-U133 Plus 2.0; 10 ovarian lesions and matched eutopic endometrium from the same 10 women; public release 9 April 2007). Patient identifiers in GEO source labels defined the 10 pairs. RMA intensities were transformed as log2(intensity + 1); the probe with the highest mean expression across all 20 samples represented each gene. Lesion-minus-eutopic differences were assessed by two-sided paired t-tests with BH correction across the six genes. The earlier unpaired Welch tests are retained as a sensitivity comparison. These three analyses are small and descriptive and are interpreted as such. Software and Data Availability Analyses used R 4.6.0 (data.table, coloc 5.2.3) and Python 3. All instrument, MR, power, colocalization, replication, cross-disease, tissue-expression and annotation tables are provided as supplementary material and, together with the analysis code, in the accompanying code and data archive described in the Data Sharing Statement.

Instrument Strength and Power at Explicit Significance Thresholds The harmonized screen contained 9157 unique gene–cell-type tests in 3821 genes. Instruments were selected by the global maximum PIP across available SuSiE credible sets. F ranged from 30.2 upward (median 64.9; interquartile range 42.3–125.2), with approximate median expression variance explained of 6.2%. At nominal two-sided α = 0.05, the 80% power minimum detectable OR had a median of 1.09 and maximum 1.13. These values do not quantify discovery power after multiple testing. At α = 0.05/9157, the corresponding median and maximum detectable ORs were 1.18 and 1.27; 35.3% of instruments reached 80% power for OR = 1.15 under this conservative threshold (Supplementary Table 2). IGF1R in NK cells had F = 137.8 and approximate R2 = 0.12. Strong selected eQTLs do not eliminate winner’s curse or limited power for smaller true effects. Cell-Type-Resolved MR Screening Across the Immune Compartment In total, 135 associations in 61 genes reached a Benjamini–Hochberg FDR < 0.05 (Figure 2 and Supplementary Table 5 lists them with their strength, power and alternative FDR values). Significant signals were distributed across immune lineages rather than confined to one subset, with the largest numbers in naive CD4+ T cells (31); central-memory CD4+ T cells (27); NK cells (18); effector-memory CD8+ T cells (16), and smaller contributions from the remaining B-, T-, NK and monocyte populations. This breadth motivated a cell-type-resolved treatment of each signal rather than a single pooled analysis. Dependence Among Tests and Dependence-Aware Significance Tiers The 9157 tests used 7110 distinct lead-variant identities, with 2047 additional tests reusing variants across genes and/or cell types. Among within-gene pairs using different variants, signed-Z correlation was 0.80; the random interchromosomal comparison was −0.002. These are descriptive dependence summaries and do not establish PRDS. Variant-deduplicated BH supported 136 gene–cell associations in 61 genes, and within-gene Bonferroni followed by BH supported 46 genes. BY retained 15 associations in 7 genes. We distinguish BY-robust tier 1 from exploratory tier 2 rather than treating agreement between related BH analyses as independent confirmation. IGF1R/NK was tier 2 (BH FDR = 0.0085; gene-level FDR = 0.012; BY FDR = 0.082; Supplementary Table 3). Colocalization on Complete Windows Prioritizes 10 Genes Colocalization was tested for every Benjamini–Hochberg-significant gene–cell-type pair and for every other cell type in which a colocalizing gene was instrumentable (123 pairs on complete ±500 kb windows; Supplementary Table 4); gene–cell-type pairs in the major histocompatibility complex were excluded a priori because its extensive LD and pleiotropy make colocalization unreliable. Ten genes colocalized at PP4 > 0.8, together spanning 28 gene–cell-type pairs, and all colocalizing pairs passed the approximate directionality screen (Figure 3, Table 2 and Supplementary Table 6). Six were protein-coding—KLF12, ATG7, RUVBL1, IGF1R, USP53, SZT2—and four were non-coding transcripts, which are annotated below. Two genes were tier 1 (LOC102723878 and RUVBL1) and eight tier 2. The conditional posterior probability of H4 given either H3 or H4, PP4/(PP3 + PP4), was ≥ 0.83 for every colocalizing pair, and PP4 remained above 0.8 on the narrower ±250 kb and ±100 kb windows for 28 of 28 pairs, indicating stability across the evaluated window widths. They do depend on the prior: only 1 of 28 pairs retained PP4 > 0.8 at p12 = 1 × 10−6 (all did at 5 × 10−5), so the colocalizations reported here are supported at the default prior and are interpreted with corresponding caution (Supplementary Table 4). ESR1 produced the single strongest marginal MR association (FDR = 6.6 × 10−17) yet did not colocalize (PP4 = 0.00): this analysis did not provide evidence of a shared signal at ESR1 under the single-causal-variant model, so ESR1 was not carried forward. This exclusion illustrates why colocalization is essential and why marginal MR significance alone is insufficient. | Table 2 The 10 Colocalized Genes with Their Statistical Evidence and Significance Tier | Distribution of MR and Colocalization Evidence Across Cell Types Figure 4 distinguishes gene–cell pairs that passed both MR BH-FDR 0.8 from tested pairs that did not pass both criteria, and from subsets lacking a retained instrument. For IGF1R, a qualifying instrument was available only in NK cells; this identifies where the association was testable, not biological exclusivity. Across other genes, screening evidence varied among subsets, but threshold differences can reflect instrument choice, power or regulation and do not establish effect heterogeneity. In particular, USP53 in naive CD4 T cells had PP4 > 0.8 but did not pass the MR threshold. The matrix therefore should not be read as a direct test of cell-specific causality. NK-Cell IGF1R as an Exploratory Candidate Under our instrument-selection criteria, a qualifying, harmonized IGF1R instrument was retained only in NK cells among the 20 subsets. Its NK-cell cis-eQTL colocalized with endometriosis on the complete window (PP4 = 0.89; PP4/(PP3 + PP4) = 0.94; PP4 = 0.46 and 0.98 at p12 = 1 × 10−6 and 5 × 10−5); the MR instrument rs2311766 is the maximum-PIP variant across the available SuSiE credible sets and lies 3 kb downstream of the IGF1R transcription start site, near the local eQTL and GWAS association peaks, which are approximately 4.1 kb apart (Figure 5). Higher genetically predicted NK-cell IGF1R expression was associated with higher endometriosis risk (odds ratio 1.11 per standard deviation). Four qualifications bound this result. First, it is tier 2: significant by the Benjamini–Hochberg and gene-level criteria but not under the worst-case Benjamini–Yekutieli procedure. Second, the NK-cell attribution describes where the signal could be detected, not where it must act. Third, the encoded protein is tractable—the anti-IGF1R antibody teprotumumab demonstrated efficacy in a Phase 3 trial of active thyroid eye disease46 and the small-molecule inhibitor linsitinib reached Phase 3 trials47—but tractability is a property of the protein, not evidence that modulating it would benefit endometriosis; the direction, cellular mechanism and safety of any such intervention are unresolved and are examined in the Discussion. Fourth, the local GWAS signal is sub-genome-wide (minimum P ≈ 3.8 × 10−6 in the window), so the colocalization rests on a modest outcome association. We therefore describe NK-cell IGF1R as a prioritized, hypothesis-generating candidate gene. Annotation of the Non-Coding Colocalized Transcripts The four non-coding transcripts were annotated by biotype, genomic position and proximity to protein-coding candidates (Supplementary Table 7). They comprise LOC102723878 (lncRNA; ends 6 bp 5’ of CCDC88B); LINC-PINT (an intergenic p53-induced lncRNA on chromosome 7q32.3); ENSG00000278022 (lncRNA; sense intronic to IGF1R); MRPL42P1 (transcribed processed pseudogene; inside the USP53 cis region). Two share a lead variant or lie within 250 kb of a colocalized protein-coding candidate. These annotations are compatible with shared regulation but do not distinguish co-regulation from an independent transcript effect. The four nominations remain part of the prespecified screen; no RNA-targeted functional validation was performed. Tractability annotation focuses on encoded proteins and should not be taken to exclude a functional role for the non-coding transcripts. Tractability and Pleiotropy of the Protein-Coding Genes In the Open Targets Platform, the protein encoded by IGF1R carried an approved-drug tractability category, and ATG7 and RUVBL1 carried small-molecule precedence (ligand-bound structures or high-quality chemical matter), USP53 had a localization-based antibody-tractability annotation, whereas KLF12 and SZT2 had no small-molecule or antibody category in the archived lookup (Supplementary Table 8 and Table 3). A phenome-wide look-up of the lead instruments of the protein-coding genes in the GWAS Catalog found few reported associations (ATG7: none; IGF1R: Risk-taking behavior (multivariate analysis); KLF12: Eosinophill percentage (UKB data field 30,210); RUVBL1: Pelvic organ prolapse, Uterine prolapse; SZT2: none; USP53: Circulating MSTN levels, Immature fraction of reticulocytes, Low density lipoprotein cholesterol levels, Sex hormone-binding globulin levels, Sex hormone-binding globulin levels adjusted for BMI, Total cholesterol levels, Waist-hip ratio, Waist-to-hip ratio adjusted for BMI) (Supplementary Table 9); the traits found are anthropometric, hematological, metabolic, reproductive, hormonal or behavioral and are not obviously indicative of confounding, though they are equally consistent with genuine pleiotropy at multi-function loci. By contrast, two of the lead instruments of the non-coding transcript LOC102723878 at 11q13.1 carry reported associations with several immune-mediated diseases (including hypothyroidism, sarcoidosis, primary sclerosing cholangitis and rheumatoid arthritis), marking that locus as broadly immune-pleiotropic rather than endometriosis-specific. A single-variant catalog look-up has limited sensitivity and does not constitute a safety assessment; it is reported as a coarse pleiotropy screen only. | Table 3 Triangulation and Tractability of the 6 Protein-Coding Colocalized Genes | Directional Consistency and Instrument Cross-Validation Of the colocalized genes, eight had instruments available in the Rahmioglu et al meta-analysis excluding 23andMe; eight of eight showed effect directions concordant with discovery (3 at nominal P < 0.05), including the NK-cell IGF1R effect (odds ratio 1.05, P = 0.087) (Supplementary Table 10). Because this sample incorporates an earlier FinnGen data freeze and the targets are correlated, the directional concordance is descriptive and is not independent replication. In eQTLGen whole blood, 5 of the 6 protein-coding lead variants were present as cis-eQTLs, 5 with concordant direction of effect on expression, providing a comparison with cis-regulatory effects detectable in bulk blood (Supplementary Table 11); the IGF1R lead variant was concordant. Bulk Benchmark, Cross-Disease Comparison and Lesion Expression When each protein-coding gene was instead instrumented by its strongest eQTLGen whole-blood cis-eQTL, all six had a bulk instrument in the outcome and three of these were nominally associated (P < 0.05), whereas RUVBL1, IGF1R and SZT2 were not (Supplementary Table 12). For IGF1R the reason is instructive: the endometriosis-colocalizing variant is only a weak eQTL in bulk blood, whereas the strongest bulk IGF1R eQTL lies at a distinct variant without endometriosis association. This comparison is small (6 genes) and reflects lead-variant selection rather than an intrinsic inability of bulk MR, so we draw from it only that, for these genes, cell-type resolution changed which variant instruments the gene and thereby which genes were nominated. Against uterine leiomyoma, seven of the ten colocalized genes did not reach BH-FDR < 0.05 for leiomyoma, including IGF1R (leiomyoma FDR = 0.96); LOC102723878, ATG7 and LINC-PINT were shared with leiomyoma (Supplementary Table 13). In ovarian lesions vs matched eutopic endometrium from the same 10 women (GSE7305; paired t-tests), four of six protein-coding genes were differentially expressed at BH-FDR < 0.05 (RUVBL1, USP53 and ATG7 lower in lesions; KLF12 higher); IGF1R showed a non-significant upward trend (log2 fold-change +0.30, P = 0.101) (Figure 6 and Supplementary Table 14). Lesion-tissue expression is confounded by cell composition and by the ovarian-lesion-versus-eutopic-endometrium comparison, so these results indicate that the genes are expressed and variably dysregulated at the disease site; they do not test the causal direction. Instrument-level values for all 9157 tests—exposure and outcome statistics, MR estimates, strength, power and every FDR variant—are provided as machine-readable tables in the accompanying code and data archive (Data Sharing Statement).

This screen prioritizes candidate genes in immune-cell contexts, with an explicit distinction between 2 BY-robust colocalized genes (LOC102723878, RUVBL1) and exploratory BH-supported candidates, including IGF1R. The empirical dependence summaries do not prove BH calibration; the BY comparison is therefore important to interpretation. Cell-type attribution identifies where an instrument and supporting evidence were available and cannot by itself establish exclusivity or differences between cell types. Tractability describes a property of a protein and does not establish that its modulation would be effective or safe in endometriosis. Interpreting the IGF1R signal requires particular care. The MR estimate indicates that higher NK-cell IGF1R expression is associated with higher endometriosis risk, which by the logic of drug-target MR would point to IGF1R inhibition as the risk-lowering direction. This does not map cleanly onto NK biology: IGF-1/IGF1R signaling canonically promotes, rather than suppresses, NK-cell development and cytotoxicity,48 so the intuitive narrative that higher IGF1R impairs NK clearance of ectopic cells is not supported and would run against the literature on reduced NK cytotoxicity in endometriosis.5,6 The positive association is therefore best regarded as a genetically supported but mechanistically unresolved observation: the NK IGF1R eQTL may act through non-cytotoxic NK programs, may mark rather than drive an NK state, or may exert its disease-relevant effect outside the effector function measurable in blood; the disease-relevant NK deficit is, moreover, documented chiefly for peritoneal NK cells and shaped by the local suppressive milieu, which a peripheral-blood eQTL may not capture. The lesion-expression trend (P = 0.101) and a nonsignificant leiomyoma association establish neither disease specificity, mechanism nor direction. Any therapeutic reading must also consider reproductive safety: granulosa-cell-specific deletion of Igf1r in female mice disrupts steroidogenesis and follicular development and causes infertility.49 This genetic experiment raises concern about reproductive effects of IGF1R targeting but does not directly establish the effects of systemic pharmacological inhibition. Hyperglycemia and hearing impairment reported in the teprotumumab trial are additional safety considerations.46 For all these reasons we present NK-cell IGF1R as a prioritized candidate gene whose therapeutic direction, cellular mechanism and safety remain to be resolved, and explicitly not as a repurposing target. The remaining protein-coding genes nominate hypotheses for follow-up. ATG7, an enzyme essential for autophagy, implicates cell-survival programs; KLF12, a Krüppel-like transcriptional repressor that colocalized in memory B cells with a protective direction and was higher in lesion tissue, has prior experimental links to endometriosis through miR-141-3p regulation of stromal-cell behavior; the present memory-B-cell genetic association adds a different cellular context and does not establish the same mechanism; RUVBL1, a chromatin-associated AAA+ ATPase linked to proliferation, implicates growth programs; SZT2, a component of the KICSTOR complex that restrains mTORC1 signaling under nutrient stress, colocalized in naive and central-memory CD4+ T cells with a protective direction, suggesting a T-cell metabolic-checkpoint hypothesis; USP53, a deubiquitinase-like protein colocalizing across B- and T-cell subsets, has no established immune function and is a candidate for mechanistic study. Prior KLF12 experimental evidence in endometriotic stromal cells is described by Zhang et al.50 For KLF12, the matched regional input now includes its indel instrument; the default PP4 was 0.944 with all matched variants vs 0.948 in the original SNP-only sensitivity. Support fell to PP4 = 0.630 at p12 = 10−6, so this candidate remains prior-sensitive. These candidates require targeted functional follow-up before therapeutic inference. Methodologically, the study illustrates both the value and the limits of pairing MR with colocalization at cell-type resolution. ESR1 produced the strongest marginal MR signal yet failed colocalization and would not have met the combined evidence criteria if reported on MR significance alone. Corrections made during revision are important to the interpretation. The colocalization windows of the submitted version had been truncated by the paging of a data-access interface, so that in many cases they did not contain the instrument; all colocalization results here were recomputed on complete ±500 kb windows, which changed the colocalized set: under comparable MR-BH and non-MHC reporting criteria, the earlier results contained 46 eligible pairs, of which 25 are retained and 21 no longer pass; three additional eligible pairs (KLF12/memory B, SZT2/central-memory CD4 and SZT2/naive CD4) yield 28 in the corrected regional analysis. The submitted count of 47 was a raw PP4 count that also included CCDC88B/NK despite MR FDR above 0.05. Among previously tested pairs there were 22 downward and two upward PP4 threshold crossings; two newly tested PP4-positive pairs were additionally identified, of which USP53/naive CD4 did not pass MR-BH. These are distinct counting stages. The submitted version also described the consistency sample as European; it is 99.3% European and includes a small Japanese component, as now stated. The bulk whole-blood benchmark, which we previously overstated as showing that cell-type attribution is inaccessible to bulk analysis, is better described as showing that cell-type resolution changed which variant instruments a gene and thereby which genes were nominated: 3 of 6 testable genes were also nominated by bulk MR. Several limitations temper interpretation. Each exposure has one instrument, so multi-instrument robust estimators and MR-Egger/MR-PRESSO analyses are unavailable, and pleiotropy cannot be excluded. coloc.abf assumes one causal variant per trait per window; low PP4 at a locus with several signals can be inconclusive, rather than refuting all shared effects (Supplementary Table 4). Default-prior support is sensitive to a smaller p12. The directional-consistency sample overlaps discovery, and nearby genes can share loci. Peripheral-blood European-ancestry eQTLs may not represent lesion-resident cells or other ancestries. Nominal power estimates overstate detection ability at stricter screening thresholds; smaller effects may be missed, and selected eQTL effect sizes may be inflated. De-identified bulk lesion expression is confounded by tissue and cell composition and cannot validate the direction of a cell-specific genetic effect. All nominations remain computational hypotheses requiring functional validation and reproductive-safety assessment. Future work should extend this design to tissue-resident and lesion-derived immune eQTLs as they become available, to non-European cohorts, to multi-signal colocalization frameworks that relax the single-causal-variant assumption, and to functional models that test whether modulating IGF1R in NK cells alters the clearance of ectopic endometrial cells.

Single-cell immune cis-eQTL Mendelian randomization with complete-window colocalization prioritizes 10 candidate causal genes for endometriosis, 6 of them protein-coding, in a tiered framework that separates Benjamini–Yekutieli-robust nominations (LOC102723878, RUVBL1) from those supported by Benjamini–Hochberg and gene-level criteria, and that distinguishes instrument availability from variation in screening evidence, neither of which alone establishes biological exclusivity. NK-cell IGF1R heads the second tier as a prioritized, hypothesis-generating candidate whose therapeutic direction, mechanism and reproductive safety are unresolved. None of these genes is a therapy-ready target; their value is to focus experimental work on specific genes in specific immune contexts in a disease still managed largely by hormonal suppression. Abbreviations BH, Benjamini–Hochberg; BY, Benjamini–Yekutieli; CI, confidence interval; eQTL, expression quantitative trait locus; FDR, false discovery rate; GWAS, genome-wide association study; IGF, insulin-like growth factor; LD, linkage disequilibrium; MHC, major histocompatibility complex; MR, Mendelian randomization; NK, natural killer; OR, odds ratio; PIP, posterior inclusion probability; PP3/PP4, posterior probability of two distinct/one shared causal variant(s); SD, standard deviation; SuSiE, sum of single effects. Data Sharing Statement The public source accessions are OneK1K/eQTL Catalogue QTS000038 (QTD000606–QTD000625), FinnGen R12 N14_ENDOMETRIOSIS and CD2_BENIGN_LEIOMYOMA_UTERI, GWAS Catalog GCST90205183, the eQTLGen cis-eQTL resource and GEO GSE7305. Open Targets and the GWAS Catalog supplied annotations. The revised analysis code, complete result tables, regional input data and an input/output manifest are publicly available at https://doi.org/10.5281/zenodo.23273877. This corrected R1 archive supersedes the earlier analysis deposit at https://doi.org/10.5281/zenodo.21234009 for the revised manuscript. The archive identifies the exact revision inputs and outputs. Ethics Approval and Informed Consent This study analyzed publicly available GWAS and eQTL summary statistics from previously published human genetic studies, together with de-identified sample-level gene-expression measurements from GSE7305. No participants were recruited, no new biological samples were collected, and no directly identifying participant information was accessed. The original studies were responsible for their ethical-review and participant-consent procedures. In accordance with items 1 and 2 of Article 32 of the Measures for Ethical Review of Life Science and Medical Research Involving Human Subjects, issued jointly by the National Health Commission, the Ministry of Education, the Ministry of Science and Technology and the National Administration of Traditional Chinese Medicine of the People’s Republic of China on 18 February 2023 (Guo Wei Ke Jiao Fa [2023] No. 4), research using lawfully obtained public data or anonymized information may be exempt from ethical review when it causes no harm to humans and involves no sensitive personal information or commercial interests. This secondary analysis was conducted under that public-data and anonymized-information exemption provision. No new participant consent was sought for this secondary use of public data. The study was conducted in accordance with the Declaration of Helsinki. Consent for Publication Not applicable; no identifiable participant information or identifiable images are reported. Acknowledgments The authors thank the participants and investigators of OneK1K, the eQTL Catalogue, FinnGen, the endometriosis GWAS consortia, eQTLGen, GEO, the GWAS Catalog and Open Targets for making their data available. Claude Code (Anthropic; Fable 5.1) and Codex (OpenAI; GPT-6 Astra) assisted with code development and execution, computational quality checks, interpretation support and language revision. The authors have checked both tools’ terms of use and confirm their suitability for this publication. The authors have verified the originality and accuracy of all content, including references, and take full responsibility for the integrity of the entire manuscript and its scientific conclusions. Author Contributions All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. Xin Wang and Guanglei Yao contributed equally to this work and are joint first authors. Funding No external funding was received for this work. Disclosure The authors report no conflicts of interest in this work.

1. Zondervan KT, Becker CM, Missmer SA. Endometriosis. N Engl J Med. 2020;382(13):1244–18. doi:10.1056/NEJMra1810764 2. Giudice LC, Kao LC. Endometriosis. Lancet. 2004;364(9447):1789–1799. doi:10.1016/S0140-6736(04)17403-5 3. Zondervan KT, Becker CM, Koga K, Missmer SA, Taylor RN, Viganò P. Endometriosis. Nat Rev Dis Primers. 2018;4(1):9. doi:10.1038/s41572-018-0008-5 4. Symons LK, Miller JE, Kay VR, et al. The immunopathophysiology of endometriosis. Trends Mol Med. 2018;24(9):748–762. doi:10.1016/j.molmed.2018.07.004 5. Jeung I, Cheon K, Kim MR. Decreased cytotoxicity of peripheral and peritoneal natural killer cell in endometriosis. Biomed Res Int. 2016;2016:2916070. doi:10.1155/2016/2916070 6. Jiang W, Xu W, Chen F. Dysfunction of natural killer cells promotes immune escape and disease progression in endometriosis. Front Immunol. 2025;16:1657605. doi:10.3389/fimmu.2025.1657605 7. Kobayashi H, Imanaka S. Understanding the molecular mechanisms of macrophage polarization and metabolic reprogramming in endometriosis: a narrative review. Reprod Med Biol. 2022;21(1):e12488. doi:10.1002/rmb2.12488 8. Vallvé-Juanico J, Houshdaran S, Giudice LC. The endometrial immune environment of women with endometriosis. Hum Reprod Update. 2019;25(5):565–592. doi:10.1093/humupd/dmz018 9. Rahmioglu N, Mortlock S, Ghiasi M, et al; DBDS Genomic Consortium; FinnGen Study; FinnGen Endometriosis Taskforce; Celmatix Research Team; 23andMe Research Team. The genetic basis of endometriosis and comorbidity with other pain and inflammatory conditions. Nat Genet. 2023;55(3):423–436. doi:10.1038/s41588-023-01323-z 10. Sapkota Y, Steinthorsdottir V, Morris AP, et al; iPSYCH-SSI-Broad Group. Meta-analysis identifies five novel loci associated with endometriosis highlighting key genes involved in hormone metabolism. Nat Commun. 2017;8:15539. doi:10.1038/ncomms15539 11. Nyholt DR, Low SK, Anderson CA, et al. Genome-wide association meta-analysis identifies new endometriosis risk loci. Nat Genet. 2012;44(12):1355–1359. doi:10.1038/ng.2445 12. Schaid DJ, Chen W, Larson NB. From genome-wide associations to candidate causal variants by statistical fine-mapping. Nat Rev Genet. 2018;19(8):491–504. doi:10.1038/s41576-018-0016-z 13. Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Genet. 2014;23(R1):R89–98. doi:10.1093/hmg/ddu328 14. Schmidt AF, Finan C, Gordillo-Marañón M, et al. Genetic drug target validation using Mendelian randomisation. Nat Commun. 2020;11(1):3255. doi:10.1038/s41467-020-16969-0 15. Sekula P, Del Greco M F, Pattaro C, Köttgen A. Mendelian randomization as an approach to assess causality using observational data. J Am Soc Nephrol. 2016;27(11):3253–3265. doi:10.1681/ASN.2016010098 16. Zhu Z, Zhang F, Hu H, et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet. 2016;48(5):481–487. doi:10.1038/ng.3538 17. Giambartolomei C, Vukcevic D, Schadt EE, et al. Bayesian test for colocalisation between pairs of genetic association studies using summary statistics. PLoS Genet. 2014;10(5):e1004383. doi:10.1371/journal.pgen.1004383 18. Hormozdiari F, van de Bunt M, Segrè AV, et al. Colocalization of GWAS and eQTL signals detects target genes. Am J Hum Genet. 2016;99(6):1245–1260. doi:10.1016/j.ajhg.2016.10.003 19. Wallace C. Eliciting priors and relaxing the single causal variant assumption in colocalisation analyses. PLoS Genet. 2020;16(4):e1008720. Correction in: PLoS Genet. 2026;22(5):e1012155. doi:10.1371/journal.pgen.1012155. doi:10.1371/journal.pgen.1008720. 20. Schmiedel BJ, Singh D, Madrigal A, et al. Impact of genetic polymorphisms on human immune cell gene expression. Cell. 2018;175(6):1701–1715.e16. doi:10.1016/j.cell.2018.10.022 21. Yazar S, Alquicira-Hernandez J, Wing K, et al. Single-cell eQTL mapping identifies cell type-specific genetic control of autoimmune disease. Science. 2022;376(6589):eabf3041. doi:10.1126/science.abf3041 22. Kerimov N, Hayhurst JD, Peikova K, et al. A compendium of uniformly processed human gene expression and splicing quantitative trait loci. Nat Genet. 2021;53(9):1290–1299. doi:10.1038/s41588-021-00924-w 23. Wang G, Sarkar A, Carbonetto P, Stephens M. A simple new approach to variable selection in regression, with application to genetic fine mapping. J R Stat Soc Ser B Stat Methodol. 2020;82(5):1273–1300. doi:10.1111/rssb.12388 24. Antonatos C, Vasilopoulos Y. Single-cell cis-Mendelian randomization reveals cell-specific genetic mechanisms underlying atopic dermatitis. Int J Mol Sci. 2026;27(5):2226. doi:10.3390/ijms27052226 25. Dou S, Wei Y, Lin Z, et al. A new perspective on endometriosis: integrating eQTL Mendelian randomization with transcriptomics and single-cell data analyses. Funct Integr Genomics. 2025;25(1):75. doi:10.1007/s10142-025-01543-y 26. Ghazal S, McKinnon B, Zhou J, et al. H19 lncRNA alters stromal cell growth via IGF signaling in the endometrium of women with endometriosis. EMBO Mol Med. 2015;7(8):996–1003. doi:10.15252/emmm.201505245 27. Pollak M. The insulin and insulin-like growth factor receptor family in neoplasia: an update. Nat Rev Cancer. 2012;12(3):159–169. doi:10.1038/nrc3215 28. Skrivankova VW, Richmond RC, Woolf BAR, et al. Strengthening the reporting of observational studies in epidemiology using Mendelian randomization: the STROBE-MR statement. JAMA. 2021;326(16):1614–1621. doi:10.1001/jama.2021.18236 29. Kurki MI, Karjalainen J, Palta P, et al; FinnGen. FinnGen provides genetic insights from a well-phenotyped isolated population. Nature. 2023;613(7944):508–518. Correction in: Nature. 2023;615(7952):E19. doi:10.1038/s41586-023-05837-8. doi:10.1038/s41586-022-05473-8 30. Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol. 2013;37(7):658–665. doi:10.1002/gepi.21758 31. Hemani G, Zheng J, Elsworth B, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife. 2018;7:e34408. doi:10.7554/eLife.34408 32. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44(2):512–525. doi:10.1093/ije/dyv080 33. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol. 2016;40(4):304–314. doi:10.1002/gepi.21965 34. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50(5):693–698. Correction in: Nat Genet. 2018;50(8):1196. doi:10.1038/s41588-018-0164-2. doi:10.1038/s41588-018-0099-7 35. Burgess S, Thompson SG; CRP CHD Genetics Collaboration. Avoiding bias from weak instruments in Mendelian randomization studies. Int J Epidemiol. 2011;40(3):755–764. doi:10.1093/ije/dyr036 36. Burgess S. Sample size and power calculations in Mendelian randomization with a single instrumental variable and a binary outcome. Int J Epidemiol. 2014;43(3):922–929. doi:10.1093/ije/dyu005 37. Brion MJ, Shakhbazov K, Visscher PM. Calculating statistical power in Mendelian randomization studies. Int J Epidemiol. 2013;42(5):1497–1501. doi:10.1093/ije/dyt179 38. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B. 1995;57(1):289–300. doi:10.1111/j.2517-6161.1995.tb02031.x 39. Benjamini Y, Yekutieli D. The control of the false discovery rate in multiple testing under dependency. Ann Statist. 2001;29(4):1165–1188. doi:10.1214/aos/1013699998 40. Hemani G, Tilling K, Davey Smith G. Orienting the causal relationship between imprecisely measured traits using GWAS summary data. PLoS Genet. 2017;13(11):e1007081. Correction in: PLoS Genet. 2017;13(12):e1007149. doi:10.1371/journal.pgen.1007149. doi:10.1371/journal.pgen.1007081 41. Buniello A, MacArthur JAL, Cerezo M, et al. The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res. 2019;47(D1):D1005–D1012. doi:10.1093/nar/gky1120 42. Ochoa D, Hercules A, Carmona M, et al. The next-generation Open Targets Platform: reimagined, redesigned, rebuilt. Nucleic Acids Res. 2023;51(D1):D1353–D1359. doi:10.1093/nar/gkac1046 43. Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol. 2016;40(7):597–608. doi:10.1002/gepi.21998 44. Võsa U, Claringbould A, Westra HJ, et al; BIOS Consortium; i2QTL Consortium. Large-scale cis- and trans-eQTL analyses identify thousands of genetic loci and polygenic scores that regulate blood gene expression. Nat Genet. 2021;53(9):1300–1310. doi:10.1038/s41588-021-00913-z 45. Hever A, Roth RB, Hevezi P, et al. Human endometriosis is associated with plasma cells and overexpression of B lymphocyte stimulator. Proc Natl Acad Sci U S A. 2007;104(30):12451–12456. doi:10.1073/pnas.0703451104 46. Douglas RS, Kahaly GJ, Patel A, et al. Teprotumumab for the treatment of active thyroid eye disease. N Engl J Med. 2020;382(4):341–352. doi:10.1056/NEJMoa1910434 47. Fassnacht M, Berruti A, Baudin E, et al. Linsitinib (OSI-906) versus placebo for patients with locally advanced or metastatic adrenocortical carcinoma: a double-blind, randomised, phase 3 study. Lancet Oncol. 2015;16(4):426–435. doi:10.1016/S1470-2045(15)70081-1 48. Ni F, Sun R, Fu B, et al. IGF-1 promotes the development and cytotoxic activity of human NK cells. Nat Commun. 2013;4:1479. doi:10.1038/ncomms2484 49. Baumgarten SC, Armouti M, Ko C, Stocco C. IGF1R expression in ovarian granulosa cells is essential for steroidogenesis, follicle survival, and fertility in female mice. Endocrinology. 2017;158(7):2309–2318. doi:10.1210/en.2017-00146 50. Zhang Y, Yan J, Pan X. miR-141-3p affects apoptosis and migration of endometrial stromal cells by targeting KLF-12. Pflugers Arch. 2019;471(8):1055–1063. doi:10.1007/s00424-019-02283-2 © 2026 The Author(s). This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms and incorporate the Creative Commons Attribution - Non Commercial (unported, 4.0) License. By accessing the work you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms.