Predicting immune-related thyroiditis using polygenic risk scores in patients with advanced melanoma Authors: Ahmad A Tarhini, Mohammad Ali Khaksar, Zhihua Chen, Sandra J Lee, F Stephen Hodi, Tingyi Li, Sean J Yoder, Howard Streicher, Vernon K Sondak, John M Kirkwood, Xuefeng Wang, Peter A Kanetsky Published: 2026 ABSTRACT AbstractBackgroundInherited genetic variation may impact patient response and risk of toxicity following immune checkpoint inhibitor (ICI) therapy.MethodsWe conducted an agnostic genome-wide analysis study for inherited genetic variants that may predict thyroiditis in patients with melanoma treated with ipilimumab and investigated Polygenic Risk Scores (PRSs) previously reported to be associated with thyroid disease. Germline DNA from 744 participants in a phase 3 adjuvant trial was analyzed following genome-wide genotyping.ResultsAn agnostic genome-wide cohort-level analysis identified top associated single nucleotide polymorphisms (SNPs), and a custom 10-SNP PRS was significantly associated with thyroiditis risk and severity. Among the SNPs identified, five mapped to intronic, four intergenic and one exonic regions. The latter was the lead SNP and mapped to CNOT6L, located near CXCL13. Among the other variants, most were located within regions with potential relevance to immune regulation and autoimmunity. Separately, multiple thyroid disease-related PRSs derived from Polygenic Score Catalog weights were tested, and several were significantly associated with thyroiditis, but the custom PRS had stronger discrimination for thyroiditis risk (area under the curve 0.82).ConclusionsThese findings indicate that inherited genetic background contributes to thyroid immune toxicity risk following ICI and support PRS-based approaches for risk-adapted monitoring during immunotherapy. WHAT IS ALREADY KNOWN ON THIS TOPIC Immune-related thyroiditis is a common toxicity following immune checkpoint inhibitor therapy, and inherited genetic susceptibility may contribute to its development. Immune-related thyroiditis is a common toxicity following immune checkpoint inhibitor therapy, and inherited genetic susceptibility may contribute to its development. WHAT THIS STUDY ADDS This study identified germline variants associated with ipilimumab-related thyroiditis in melanoma and demonstrated that a cohort-derived 10-single nucleotide polymorphism Polygenic Risk Score (PRS) strongly predicted thyroiditis risk and severity. Several previously published thyroid disease-related PRSs were also significantly associated with thyroiditis. This study identified germline variants associated with ipilimumab-related thyroiditis in melanoma and demonstrated that a cohort-derived 10-single nucleotide polymorphism Polygenic Risk Score (PRS) strongly predicted thyroiditis risk and severity. Several previously published thyroid disease-related PRSs were also significantly associated with thyroiditis. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings support the potential utility of germline PRS-based approaches for risk stratification and personalized monitoring of immune-related toxicities during immunotherapy. These findings support the potential utility of germline PRS-based approaches for risk stratification and personalized monitoring of immune-related toxicities during immunotherapy. Background The incidence of melanoma continues to show an upward trend, and it is estimated that 112,000 new cases of invasive melanoma will be diagnosed in the USA alone in 2026.1 Melanoma that is locoregionally advanced or metastatic carries a poor prognosis for patients diagnosed with this disease. However, advances in cancer immunotherapy, particularly the introduction of immune checkpoint inhibitors (ICIs), have undeniably transformed the prognosis of patients with advanced melanoma and improved their clinical outcomes in the adjuvant and neoadjuvant settings.24 Nevertheless, ICI may also expose patients to significant risks, including the development of a spectrum of immune-related adverse events (irAEs) that may necessitate treatment delay or discontinuation, and in severe cases lead to hospitalization, life-threatening complications and even death. There is a great need to identify biomarkers of irAE susceptibility to inform risk stratification and clinical decision making. The frequency and severity of irAEs differ by ICI type and dosage,3 5 and thyroid dysfunction is among the most frequently reported irAEs following ICI therapy. Prior studies have suggested an association between genetic variation within the CTLA-4 gene region and risk of thyroiditis following cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) blockade,6 7 and reported rates of thyroid dysfunction range from 10% to 20% in patients with high-risk melanoma treated with adjuvant ipilimumab, a CTLA4 blocking antibody. Thus, patients treated with ipilimumab monotherapy are of special interest in investigating biomarkers of thyroiditis.3 8 The observed overlap in the pathophysiology of autoimmune thyroid disease and ICI-induced thyroid dysfunction, together with evidence supporting a role for genetic variation in thyroid autoimmunity,9 suggest that inherited genetic susceptibility may contribute to endocrine irAEs in patients receiving ICIs. Therefore, germline profiling, including Polygenic Risk Score (PRS) analysis, may help identify patients at higher risk of treatment-related irAEs leading to mitigation of these irAEs and increased patient safety. In this study, we conducted an agnostic genome-wide investigation of inherited genetic variation in relation to the risk of thyroiditis following ipilimumab adjuvant therapy as tested in the E1609 Eastern Cooperative Oncology Group-American College of Radiology Imaging Network Cancer Research Group (ECOG-ACRIN) phase 3 clinical trial in melanoma. We pursued inherited genetic variants that may predict susceptibility risk for ipilimumab-induced thyroiditis. In addition, we interrogated PRSs previously reported to be associated with thyroid dysfunction derived from non-melanoma patient populations, the majority of whom were not exposed to immunotherapy. We assessed the predictive value of our custom PRS, and the previously published thyroid-dysfunction associated PRSs in relation to the risk of thyroiditis following ipilimumab as tested in E1609 in melanoma. Methods Patient selection The present study used banked biospecimens from 744 consenting patients who met eligibility criteria for the E1609 trial (NCT01274338), including those with stage IIIB, IIIC, M1a or M1b melanoma based on American Joint Committee on Cancer seventh edition.3 Patients had melanoma, of cutaneous or unknown primary origin, that was histologically confirmed. The study required an ECOG performance status of 0 or 1 and screening safety laboratory test criteria. A history of autoimmune disorders or conditions of immunosuppression that necessitated the use of systemic corticosteroids or other immunosuppressants were not allowed. Within 12 weeks of surgical resection, patients underwent trial registration and randomization to receive adjuvant therapy with ipilimumab 3 mg/kg or 10 mg/kg versus interferon-α2b. Details of the randomization process, treatment regimens tested, clinical trial endpoint points evaluated, and trial oversight procedures were previously published.3 Thyroiditis definition Trial patients were screened for prior history of thyroid disorders, including hypothyroidism and hyperthyroidism, based on documentation available before study enrollment. Thyroid function tests were performed at baseline, at scheduled clinic visits, during acute hospital admission, and whenever any thyroid-related irAE was suspected by investigators. Thyroid-related adverse events were defined and graded according to Common Terminology Criteria for Adverse Events (CTCAE) V.4.0.10 After treatment discontinuation, patients continued to be followed and monitored for treatment-related thyroid disorders. Genotyping, GWAS and PRS calculation for thyroiditis risk Details related to biospecimen collection, processing and banking were previously published.11 Germline DNA (n=711) and peripheral blood mononuclear cells (PBMC; n=33) biospecimens banked at the ECOG-ACRIN Central Biorepository and Pathology Facility at MD Anderson Cancer Center were shipped to Moffitt Cancer Center CCSG Shared Resources (Tissue Core and Molecular Genomics Cores). Germline DNA was extracted from PBMC, and all DNA samples underwent Qubit quantification. Genome-wide genotyping was performed using the Illumina Infinium Global Screening Array V.3.0 + Multi-Disease BeadChip, containing 730,059 designed loci. Genotypes were called using Illumina Genome Studio V.2.0 and IAAP CLI V.1.1.0, then subjected to quality control, and harmonized with the 1,000 Genomes Phase 3 (1KGP3) reference dataset.12 13 Genotype imputation was conducted against the Haplotype Reference Consortium V.r1.1 reference panel using the Michigan Imputation Server 2 pipeline V.2.0.6.14 Next, we conducted a genome-wide association study (GWAS) using genome-wide genotyping data to identify germline single nucleotide polymorphisms (SNPs) associated with thyroiditis following ipilimumab therapy. Each variant was then evaluated individually to characterize locus-specific genetic associations. As a next step, eighteen PRS for thyroiditis available in the Polygenic Score (PGS)-Catalog v20240318 (PGS000759, PGS000761, PGS000820, PGS000965, PGS001181, PGS001816, PGS002024, PGS002336, PGS002408, PGS002457, PGS002506, PGS002555, PGS002604, PGS002653, PGS002702, PGS002766, PGS004446, PGS004516),91520 were calculated21 using the following equation, where N represents the total number of SNPs in each PGS model, βi denotes SNP effect size for each PGS model, and G{ij} is genotype dosage for SNP i in individual j (0–2 or imputed). Statistical analysis Logistic regression implemented in PLINK (V.2.00a4LM glm with Firth fallback) was used to estimate associations between inherited genetic markers and the occurrence of thyroiditis, with adjustment for genetic ancestry. Two complementary non-parametric statistical approaches were used to assess associations between PRS and thyroid-related irAE. First, the Kruskal-Wallis test was applied to compare PRS distributions across all reported irAE grades, treating irAE grade as a multilevel categorical variable to detect global differences according to toxicity severity. Second, to enhance statistical power and facilitate clinically meaningful interpretation, the Wilcoxon rank-sum test was performed after dichotomizing irAE grades into grade 0 (no toxicity) versus grade ≥1 (any thyroiditis). The prognostic impact of thyroiditis on recurrence-free survival (RFS) and overall survival (OS) was assessed using Kaplan-Meier curves and multivariable Cox proportional hazards regression models adjusted for sex and primary tumor site. In addition, receiver operating characteristic (ROC) curve analyses were conducted to evaluate the discriminatory performance of PRSs with model performance quantified using the area under the curve (AUC). All statistical analyses were conducted in R (V.4.1.1) and Python (V.3.12.2), and two-sided p values ≤0.05 were considered statistically significant. Results Study population A total of 1,670 patients were enrolled in the E1609 phase 3 adjuvant trial following complete surgical resection of the locoregionally advanced melanoma (stages IIIB, IIIC, M1a, and M1b). Patients were randomized to receive adjuvant therapy with ipilimumab (3 mg/kg or 10 mg/kg) versus high-dose interferon-α2b given for up to 1 year.3 Only 1,027 patients treated with ipilimumab were potentially eligible to be included in our analysis given our interest in irAEs induced by anti-CTLA4 ICI therapy. Of the ipilimumab-treated patients, 744 patients who provided a written informed consent and with available germline DNA for research were included in this biomarker project. Subsequently, five patients were excluded following quality control measures, and one was excluded due to missing clinical information. Thyroiditis events were defined using CTCAE V.4.0. Baseline demographics and clinical characteristics are summarized in online supplemental table S.1. Patient disposition from E1609 is described in the consort diagram (online supplemental figure S1). PRS/SNP analyses based on E1609 cohort-derived germline variants We evaluated cohort-derived germline SNPs in order to identify locus-specific genetic variants associated with ipilimumab-related thyroiditis. Given genome-wide association analyses, we prioritized 10 SNPs that demonstrated the strongest associations with thyroiditis incidence in this cohort (table 1). Genomic annotation of the top ten SNPs revealed that most mapped to non-coding regions, including intronic (n=5) and intergenic (n=4) loci, with one variant located within an exonic region. Several intronic variants were identified within genes with potential relevance to immune regulation or cellular signaling, including CNOT6L, CDH20, LINC02135, LARGE-AS1, and translocase of inner mitochondrial membrane 23 homolog B (TIMM23B). Whereas intergenic variants were located near genes including FOXG1-AS1, CHEK2P2, BZW2, SLC10A2. In addition, one variant was mapped to an exonic region and was located near the LINC02388 gene. Furthermore, among these variants, a lead SNP located at chromosome 4, position 78679319 (A/G) demonstrated the strongest association with thyroiditis (online supplemental figure S2). This variant mapped to an intronic region of CNOT6L and was located near CXCL13, a chemokine gene implicated in immune cell trafficking and immune activation. Table 1Top cohort-derived germline SNPs associated with ipilimumab-related thyroiditisSNPChr (Position)Effect/referenceOR (95% CI)P valueFeatureGene IDGene symbolrs285806074 (78679319)A/G2.36 (1.66 to 3.35)1.65474e-06Intron246175 CNOT6L rs195638814 (29132877)A/G0.50 (0.38 to 0.67)1.92255e-06Intergenic*59764†103695363 FOXG1-AS1 rs215959418 (59048700)A/C2.06 (1.52 to 2.78)2.56023e-06Intron28316 CDH20 rs718512316 (86322333)A/C2.54 (1.67 to 3.88)1.55753e-05Intron146513 LINC02135 GSA-rs11776486315 (20044342)A/G3.19 (1.88 to 5.43)1.81114e-05Intergenic*443398†646096 CHEK2P2 rs7060457 (16754957)G/A1.78 (1.37 to 2.33)2.06511e-05Intergenic*8808†28969 BZW2 rs797010412 (58962959)G/A0.49 (0.35 to 0.68)2.56738e-05Exon101927653 LINC02388 rs951892713 (103745231)G/A0.42 (0.28 to 0.63)2.95855e-05Intergenic*26463†6555 SLC10A2 rs23933022 (34135007)G/A1.79 (1.36 to 2.35)2.99303e-05Intron100506195 LARGE-AS1 rs7492294110 (51520713)G/A5.26 (2.41 to 11.47)3.01339e-05Intron100652748 TIMM23B *Intergenic SNPs were annotated using a linear genomic distance approach to the nearest upstream and downstream genes.†Distance to nearest gene.Chr, chromosome; SNP, single-nucleotide polymorphism. Top cohort-derived germline SNPs associated with ipilimumab-related thyroiditis Intergenic SNPs were annotated using a linear genomic distance approach to the nearest upstream and downstream genes. Distance to nearest gene. Chr, chromosome; SNP, single-nucleotide polymorphism. From our genome-wide association analysis, when combining the 10 independent SNPs that were significantly associated with thyroiditis irAE using log (OR) as weights, these SNPs were significantly associated with the severity of irAE. The discriminatory ability for distinguishing patients who developed thyroiditis from those who did not (grade 0 vs grade ≥1), achieved an AUC of 0.82 (figure 1). Moreover, using Kruskal-Wallis analysis indicated our custom PRS were significantly associated with increasing toxicity severity (p<2.3 × 10−29) (figure 2). In addition, Wilcoxon rank-sum testing demonstrated that patients who developed thyroiditis of any grade had higher median scores compared with those without thyroiditis (p<3.9 × 10−31) (figure 3). ROC curves and corresponding AUC values for the cohort-derived 10-SNP PRS and selected significant PGS Catalog-based PRSs in predicting ICI-related thyroiditis. ROC curve shows the discriminatory performance of a custom 10-SNP PRS for ipilimumab-related thyroiditis (AUC=0.82). For the selected statistically significant PGS Catalog-based PRSs, AUC values ranged from 0.57 to 0.60. AUC, area under the curve; ICI, immune checkpoint inhibitor; PGS, Polygenic Score; PRS, Polygenic Risk Score; ROC, receiver operating characteristic; SNP, single nucleotide polymorphism. Distribution of thyroid disorder-related PRSs across irAE grades. Boxplots depict PRS values by thyroiditis severity category with Kruskal-Wallis p values shown for PGS. Each panel corresponds to a distinct PRS derived from the PGS Catalog and our custom PGS. Individual points represent patient-level PRS values, with boxes indicating median and IQR. irAE, immune-related adverse event; PGS, Polygenic Score; PRS, Polygenic Risk Score; SNP, single nucleotide polymorphism. Custom PRS across cross-group comparisons of thyroiditis toxicity grades. Box plots illustrate the distribution of the custom thyroiditis PRS across multiple cross-comparisons of thyroiditis toxicity grade groupings. Comparisons include grade 0 versus grade 1–2, 2–3, and grade 3 alone; grade 0 versus combined all grades (1–2–3); grades 0 and 1 versus grades 2 and 3; and grade 0–1–2 versus grade 3. Group differences were evaluated using the Wilcoxon rank-sum test, with corresponding p values shown in each panel. PGS, Polygenic Score; PRS, Polygenic Risk Score. Analysis of PRS derived from non-melanoma cohorts To complement the analyses based on our GWAS study, we examined 18 distinct PRSs previously reported to be associated with thyroid disease and generated from non-melanoma patient populations. The examined PRS as derived from the PGS Catalog included: PGS000759, PGS000761, PGS000820, PGS000965, PGS001181, PGS001816, PGS002024, PGS002336, PGS002408, PGS002457, PGS002506, PGS002555, PGS002604, PGS002653, PGS002702, PGS002766, PGS004446, PGS004516. Of the 18 thyroid-related PRSs available in the PGS Catalog, nine demonstrated a statistically significant association comparing the distribution of PRS across grades of thyroid toxicity (figure 2): PGS000759 (p=0.012), PGS000761 (p=0.0028), PGS000965 (p=0.032), PGS001181 (p=0.035), PGS001816 (p=0.0075), PGS002024 (p=0.002), PGS002336 (p=0.03), PGS002653 (p=0.0036), and PGS002702 (p=0.0076). For 10 PRSs (online supplemental figure S3), patients who experienced thyroiditis (grade ≥1) had statistically significantly higher median PRS values compared with patients who did not develop thyroiditis (grade 0): PGS000759 (p=0.0079), PGS000761 (p=0.00031), PGS000820 (p=0.011), PGS000965 (p=0.0042), PGS001181 (p=0.0058), PGS001816 (p=0.0032), PGS002024 (p=0.0013), PGS002336 (p=0.011), PGS002702 (p=0.0016), and PGS002653 (p=0.002). Additional ROC curve analyses (figure 1) demonstrated the following AUC values for these evaluated PGS Catalog-derived PRSs: PGS000759 (AUC=0.57), PGS000761 (AUC=0.60), PGS000820 (AUC=0.57), PGS000965 (AUC=0.58), PGS001181 (AUC=0.57), PGS001816 (AUC=0.58), PGS002024 (AUC=0.59), PGS002336 (AUC=0.57), PGS002702 (AUC=0.59), and PGS002653 (AUC=0.58). Therefore, our results support a subset of thyroid-related PRS, namely PGS000759, PGS000761, PGS002653, PGS000965, PGS001181, PGS001816, PGS002024, PGS002336, PGS002702, and PGS002653 as the informative germline markers associated with the incidence of immune-related thyroiditis in patients with melanoma treated with ipilimumab. Association with survival outcomes Given that several thyroid-related PGSs were associated with the development of thyroiditis, we next evaluated whether the most informative scores were also associated with survival outcomes, namely RFS and OS. For each PGS, Patients’ PRSs were stratified according to percentile thresholds (10th, 50th, and 90th percentiles), and RFS and OS were analyzed using Kaplan-Meier curves and Cox proportional hazards models. Results from Kaplan-Meier analysis did not demonstrate a statistically significant difference in RFS or OS across PRS percentile categories for published thyroiditis PRS or the PRS derived from the top 10 SNPs identified in this cohort. Consistent with these findings, multivariable Cox proportional hazards models adjusted for sex and primary tumor status that were previously shown to be significantly prognostic in the E1609 cohort,22 23 showed no statistically significant associations between the custom 10-SNP PRS and either RFS (HR=0.74, 95% CI 0.18 to 3.11; p=0.69) or OS (HR=0.76, 95% CI 0.12 to 4.89; p=0.77) (figure 4). Association of relapse-free and overall survival with our custom PRS. Kaplan-Meier survival curves and Cox proportional hazards models depict survival outcomes stratified by quartiles of a custom PRS constructed from the top ten germline SNPs associated with ipilimumab-related thyroiditis. (A) and (B) show RFS, demonstrating no significant differences in RFS across PGS quartiles. (C) and (D) show OS, similarly revealing no significant association between the custom PRS and OS. Forest plots display multivariable HRs adjusted for sex and primary melanoma status (known vs unknown). OS, overall survival; PGS, Polygenic Score; PRS, Polygenic Risk Score; RFS, relapse-free survival; SNP, single nucleotide polymorphism. As illustrated in figure 5, Kaplan-Meier analysis demonstrated longer RFS among patients who developed thyroiditis following ipilimumab compared with those without thyroiditis, both when defined as any grade (1–3) and when restricted to low-moderate grades (1–2). In multivariable Cox regression adjusted for sex and primary tumor site, thyroiditis of any grade (grades 0 vs 1–3) was associated with a lower hazard of relapse (HR=0.84, 95% CI 0.63 to 1.11, p=0.21). A similar association was observed for low-moderate thyroiditis (grades 0 vs 1–2; HR=0.79, 95% CI 0.59 to 1.05, p=0.11). However, these two did not reach statistical significance. In contrast, grade 3 thyroiditis was associated with an increased risk of relapse, but the estimate had a wide confidence interval (HR=2.42, 95% CI 0.99 to 5.89, p=0.05), probably consistent with the limited sample size. RFS according to thyroiditis grade after ipilimumab therapy. Kaplan-Meier survival curves and multivariable Cox proportional hazards models evaluating RFS by thyroiditis grade following ipilimumab. (A) and (B): analyses comparing patients without thyroiditis (grade 0) versus those who developed thyroiditis of any grade (grades 1–3). (C) and (D): analyses comparing patients without thyroiditis (grade 0) versus those with low-to-moderate thyroiditis (grades 1–2). (E) and (F): analyses isolating high-grade thyroiditis (grade 3) compared with grade 0. RFS, relapse-free survival. Individuals with low-moderate grade (1–2) thyroid toxicity consistently also exhibited higher OS probabilities on survival follow-up compared with those without thyroiditis (figure 6). Nevertheless, in multivariable Cox regression adjusted for sex and primary tumor site, low-to-moderate thyroid toxicity was not significantly associated with improved OS (grades 0 vs 1–2: HR=0.85, 95% CI 0.58 to 1.24, p=0.41). A similar result was observed when thyroiditis of any grade was considered (grades 0 vs 1–3: HR=0.96, 95% CI 0.67 to 1.37, p=0.83). Interestingly, high-grade thyroid irAEs (seven patients, all grade 3) were significantly associated with worse OS, but here again with a wide CI (grade 0 vs 3: HR=5.53, 95% CI 2.25 to 13.62, p<0.001). We investigated these cases and found that six of seven patients had other serious irAEs, including colitis and nephritis (1), hypophysitis (2), adrenal insufficiency (2), and hypophysitis and adrenal insufficiency (1). Therefore, it is likely that these patients had more serious toxicities besides thyroid dysfunction that may explain the negative impact on survival. OS according to thyroiditis grade after ipilimumab therapy. Kaplan-Meier survival curves and multivariable Cox proportional hazards models evaluating OS by thyroiditis grade following ipilimumab. (A) and (B): analyses comparing patients without thyroiditis (grade 0) versus those who developed thyroiditis of any grade (grades 1–3). (C) and (D): analyses comparing patients without thyroiditis (grade 0) versus those with low-to-moderate thyroiditis (grades 1–2). (E) and (F): analyses isolating high-grade thyroiditis (grade 3) compared with grade 0. Discussion Although endocrine irAEs are among the most common and earliest toxicities following ICI therapy, they have been associated with improved treatment outcomes. Notably, endocrine irAEs are frequently permanent and can result in additional long-term clinical burden.24 Here, we conducted an agnostic GWAS of inherited genetic variation in relation to the risk of thyroiditis following ipilimumab adjuvant therapy as tested in the E1609 trial. We identified ten independent loci that were significantly associated with thyroiditis incidence in our patient cohort. Among these 10 SNPs, 9 were located in non-coding regions (intronic or intergenic), while only one SNP was located within a coding region. Although intergenic SNPs are located in non-coding regions between genes, many fall within regulatory elements such as promoters, enhancers, and long non-coding RNA (lncRNA) loci, where they can modulate transcriptional control of nearby genes and influence disease susceptibility. SNP Rs28580607, located on chromosome 4 at position 78679319 (A/G), demonstrated the most statistically significant association with thyroiditis among patients who received ipilimumab. This variant mapped to an intronic region of CNOT6L, which is located near CXCL13, a chemokine gene implicated in immune cell trafficking and immune activation. Interestingly, Zhang et al25 demonstrated that CNOT6L expression is upregulated in individuals with type 2 diabetes. They further reported that higher expression levels were associated with diabetes complications, inflammation and poor clinical outcomes, suggesting a role for this gene in inflammatory dysregulation and adverse metabolic outcomes. In addition, Sato et al reported that CNOT6L knockdown mice subjected to transverse aortic constriction-induced pressure overload demonstrated significantly reduced left ventricular contractility and increased myocardial fibrosis and inflammation. Moreover they observed that CNOT6L deadenylase attenuates heart failure progression via downregulation of tenascin-C expression in cardiac fibroblasts.26 Together, these findings suggest that CNOT6L may exert context- and tissue-specific effects, potentially reflecting a compensatory response in chronic inflammatory states. In terms of the CXCL13 gene, it encodes a key chemokine involved in inflammation and immune reaction and its dysregulation is associated with various autoimmune diseases.27 28 Rs1956388 was another SNP identified as being associated with thyroiditis following CTLA-4 blockade and mapped to FOXG1-AS1. FOXG1-AS1 is an immune-related lncRNA associated with the FOXG1 locus and has been implicated in autoimmune diseases, in particular thymoma and myasthenia gravis.29 This observation suggests a potential role for FOXG1-AS1 in modulating immune response, possibly through interactions with immune cells and regulatory signaling pathways. Moreover, gene ontology (GO) analysis revealed that target genes of FOXG1-AS1 were predominantly enriched in small GTPase binding. GTPases play an important role in inflammatory cell migration, and dysregulation of GTPase signaling has been implicated in a variety of inflammatory diseases.30 Genetic variation at the Cadherin 20 (CDH20) locus was also associated with susceptibility to thyroiditis. To date, there is limited evidence linking CDH20 directly to autoimmune disease. However, Li et al31 reported that CDH20, in cooperation with β-Catenin, inhibits cervical cancer cell migration and invasion, suggesting a role for CDH20 in cell–cell adhesion and tissue integrity that may be relevant to immune-mediated tissue injury. The next SNP identified in our analysis mapped to the CHEK2P2 gene was GSA-rs117764863. According to existing literature, CHEK2P2 itself has not been directly implicated in autoimmune disease, as it is a non-coding pseudogene. CHEK2P2 is annotated as a pseudogene related to CHEK2, and like other pseudogenes, may have potential regulatory roles; however, direct evidence that CHEK2P2 regulates CHEK2 expression has not yet been demonstrated in published studies. CHEK2 is a key mediator of genomic integrity that promotes cell cycle arrest or apoptosis in response to DNA double-strand breaks.32 Notably, Xu et al observed that tumors harboring CHEK2 deficiency mutations exhibited improved responses to programmed cell death protein 1 (PD-1) immunotherapy. There are two potential explanations for this observation. First, the canonical role of CHEK2 in response to DNA damage response pathways, leading to increased genomic instability and accumulation of somatic mutations. This, in turn, may result in elevated tumor mutational burden, a feature that has been associated with enhanced responsiveness to ICIs and improved clinical outcomes. Second, beyond that canonical role, CHEK2 appears to play a novel immunomodulatory role within the tumor microenvironment. CHEK2 dysfunction may directly potentiate anti-tumor immunity through activation of innate immune signaling pathways, including cyclic GMP-AMP synthase-stimulator of interferon genes(cGAS-STING), thereby promoting CD8+ T-cell infiltration and enhancing responsiveness to ICI.33 34 Therefore, a better clinical response to ICIs is associated with a higher probability of developing irAEs. Of note, another SNP identified in our analysis maps to the SLC10A2 gene, which encodes the ileal sodium-dependent bile acid transporter, a key mediator of bile acid reabsorption in the terminal ileum. Dysfunction of SLC10A2 has been reported in Crohn’s disease and is associated with impaired enterohepatic circulation of bile acids, resulting in bile acid malabsorption and chronic diarrhea.35 One of the intronic SNPs we identified maps to the TIMM23B gene, which is involved in protein import into the mitochondrial matrix, thereby it plays a pivotal role in mitochondrial function and hemostasis. Mitochondrial dysfunction promotes chronic inflammation through a constant vicious cycle in which mitochondrial damage triggers inflammatory responses that, in turn, further exacerbate mitochondrial injury. This process results in reduced ATP production, increased generation of reactive oxygen species, and the release of proinflammatory mitochondrial damage-associated molecular patterns, such as mitochondrial DNA, which activate innate immune pathways, including the NLRP3 inflammasome, thereby sustaining chronic inflammation.36 Additionally, mitochondrial dysfunction in myeloid cells—particularly dendritic cells, macrophages, and neutrophils—leads to dysregulation of these cells and represents an early initiating step in the development of autoimmunity.37 38 As a next step, a PRS was constructed using these top 10 SNPs, and its predictive performance in relation to thyroiditis risk following ipilimumab therapy was assessed. Cross-comparisons of toxicity grade groupings demonstrated a consistent association between higher PRS values and more severe thyroiditis phenotypes. These findings suggest that germline genetic susceptibility contributes to interindividual variability in the incidence and severity of immune-related thyroiditis following ICI therapy. Importantly, the use of cross-grade comparisons, rather than reliance on a single maximum-grade definition, strengthens the robustness of this association by demonstrating consistency across multiple clinically relevant toxicity groupings. Additionally, the observed AUC of 0.82 in ROC curve analysis indicates good discriminatory performance for distinguishing individuals with thyroiditis. In addition to our custom PRS, we evaluated thyroid disorder-related PRSs from the PGS Catalog generated from non-melanoma patient populations. Several of these PRSs demonstrated significant differences in score distributions. Notably, PGS000759 and PGS000761 were derived from the study by Khan et al,15 which analyzed seven large clinical trials testing atezolizumab, a programmed death-ligand 1 inhibitor, in combination with chemotherapy and bevacizumab, in patients with urothelial carcinoma, renal cell carcinoma, non-small cell lung cancer, small cell lung cancer and triple-negative breast cancer. PGS000820 was another PRS that showed a significant association with the development of thyroiditis in our patients with melanoma treated with ipilimumab. This PRS has also been evaluated in a study of patients with non-small cell lung cancer receiving PD-1 blockade.9 All other PGS evaluated were generated from studies of subjects who were not previously exposed to immunotherapy. PGS000965 and PGS001181 were among the PRSs that showed significant associations in our analysis. These PRSs were originally derived by Tanigawa et al using UK Biobank data, a large general population cohort.16 We also evaluated PGS001816, PGS002024, PGS002336, PGS002653, and PGS002702 in our cohort; these PRSs were likewise developed using UK Biobank data in separate studies.17 18 Although previous studies derived PRS models using UK Biobank data, these investigations differed in purpose, including PRS construction, cross-ancestry portability assessment, and methodological algorithm development, and were conducted in general population cohorts rather than treatment-specific melanoma patient populations. Of note, four SNPs—rs1956388, rs2159594, rs9518927, and rs239330—were shared between the PGS Catalog–derived PRSs and our custom PRS. These do not include the lead SNP (rs28580607) from our agnostic GWAS analysis, discussed earlier. To evaluate predictive strength, we assessed discrimination using ROC curve analysis and AUC metrics. Our cohort-derived PRS achieved a higher AUC than other PRS models; nevertheless, external validation is required. We observed that patients who developed grades 1–2 thyroiditis following ipilimumab had improved median OS and RFS compared with those who did not experience thyroiditis; however, these differences were not retained after adjustment for sex and primary tumor status, likely related to the limited sample size in this cohort. Interestingly, an older study reported a statistically significant association between the development of autoimmunity—specifically thyroid dysfunction—and improved clinical outcomes (RFS and OS) in patients with melanoma treated with adjuvant interferon-alfa-2b, suggesting that autoimmune thyroid dysfunction is an indicator of a more robust immune response against melanoma, correlating with better prognosis.39 In our study, the occurrence of grade 3 thyroiditis in seven patients was significantly associated with worse survival, but we also found that the majority of these patients had other serious irAEs besides thyroid dysfunction that may explain the negative impact on survival. Conclusion In conclusion, inherited genetic susceptibility significantly contributes to the development of immune-related thyroiditis in patients with melanoma treated with ipilimumab. Integration of established thyroid autoimmunity PRS with cohort-derived germline variants demonstrated strong predictive performance for thyroid irAEs, including a custom 10-SNP score with high discriminatory ability. Genetic risk for thyroiditis was independent of relapse-free and OS, highlighting a distinction between toxicity predisposition and oncologic outcomes. Clinically, low-grade thyroiditis showed a favorable survival trend, whereas high-grade toxicity was associated with worse outcomes. These findings support the potential utility of germline genetic profiling to inform risk-adapted monitoring and personalized immunotherapy strategies. Supplementary material