Main

Fear and anxiety are critical survival responses; thus, ANX may result from dysregulation of the brain’s threat–response circuits. Although perturbations in various neurotransmitter systems, such as serotonin or gamma-aminobutyric acid (GABA), have been proposed as a basis of their etiology, no reliable biomarkers have yet been identified1. The major ANX, including generalized anxiety disorder (GAD), panic disorder and phobias (specific phobia, social phobia and agoraphobia), represent different clinical presentations of that underlying common diathesis2,3,4. Up to 25% of the population will develop an ANX at some point during their lifetime5,6,7. These disorders tend to onset early in life, are persistent and are highly comorbid with other psychiatric conditions for which they often present as a predisposing risk factor; for example, major depressive disorder (MDD) and substance-use disorders6,8,9,10. ANX are also associated with other medical conditions, such as neurological, cardiovascular and gastrointestinal disorders as well as cancers11,12,13,14. These features make ANX a leading source of worldwide disability15,16.

Each ANX aggregates in families (odds ratio, 4–6) primarily owing to genetic risk factors17. Estimates from twin studies indicate that ANX are moderately heritable (h2 = 30–50%)2,17, similar to other common psychiatric disorders like MDD but lower than less prevalent disorders like schizophrenia and bipolar disorder. Different ANX exhibit overlapping clinical features and strong comorbidity, which may be a result of shared genetic susceptibility17,18,19 and environmental risk factors20,21,22. Research implicates mechanisms that affect the structure and functional capacity of brain networks involved in emotion and cognition23,24,25. Twin studies report substantial genetic correlations between ANX and other psychiatric conditions, particularly MDD26, helping to explain their high comorbidity. In addition, ANX and depression both share genetic risk with heritable personality traits such as neuroticism27,28. Anxiety symptoms often precede suicidal behaviors29, with possible causal implications30. Therefore, examining the genetic relationship between ANX and related phenotypes on the internalizing spectrum is essential.

The combination of high prevalence, extensive comorbidity and high polygenicity makes it particularly difficult to identify genetic variants underlying risk for ANX. Prior genome-wide association studies (GWAS) have identified a handful of genetic loci with inconsistent results31,32,33,34,35,36. A recent meta-analysis using five publicly available datasets reported ten additional novel associations37. Genome-wide single nucleotide polymorphism (SNP)-based heritability estimates range from 10–28%, supporting that ANX have a polygenic basis. Consistent with twin studies, previous psychiatric GWAS have demonstrated that ANX polygenic risk is highly correlated with that of MDD and neuroticism38,39,40,41,42. Similar to other complex genetic phenotypes, sufficiently large samples are required to achieve the necessary power to detect the small effects of common variants.

Here, we present a GWAS meta-analysis from the Anxiety Disorders Working Group of the Psychiatric Genomics Consortium (PGC-ANX), consisting of 122,341 individuals diagnosed with any ANX and 729,881 controls, all of European (EUR) ancestry. We analyzed the data at the level of variant, gene, pathway/gene set and tissue by using both functionally informed and functionally agnostic methods. Subsequently, these results were compared with those of other phenotypes and investigated for possible molecular mechanisms and avenues for drug repurposing.

Results

GWAS meta-analysis

We performed a GWAS meta-analysis of 36 case–control cohorts (122,341 ANX cases and 729,881 controls; Supplementary Table 1). Details about phenotype, quality control and GWAS analysis for each individual cohort are provided in Supplementary Note 2. Among the 7.2 million autosomal SNPs analyzed, we identified 58 independent, genome-wide significant (GWS) SNPs associated with ANX (Fig. 1 and Table 1; further information is provided in Supplementary Table 2, Supplementary Fig. 1 (quantile–quantile plot) and Supplementary Figs. 2–56 (regional association plots of each significant SNP and forest plots indicating each cohort’s effect size)). Estimates of the genomic inflation factor ((\lambda) = 1.41, ({\lambda }_{1000}=1.00)), linkage disequilibrium (LD) score regression (LDSC) intercept (1.05, standard error (s.e.) = 0.01), and attenuation ratio (0.082, s.e. = 0.014) suggest that inflation was probably caused by polygenicity and not by cryptic population structure. LDSC estimates a SNP-based heritability of 10.1% (s.e. = 0.004), assuming a 20% population prevalence.

Fig. 1: Manhattan plot of the main ANX GWAS showing 58 GWS loci.

The x axis shows the position in the genome (chromosomes 1 to 22), and the y axis represents −log10(P values) (two-sided, not adjusted for multiple testing) for the association of variants with ANX using an inverse-variance weighted fixed effects model (122,341 ANX cases and 729,881 unaffected controls). The horizontal red line shows the threshold for GWS (P = 5 × 10−8). Dots represent each SNP that was tested in the GWAS, with a green diamond indicating the lead SNP of a GWS locus and green dots below representing SNPs within the locus with high levels of LD with the lead SNP.

A series of sensitivity analyses, including GWAS Cochran’s Q (Supplementary Fig. 57) and I² statistics (forest plots in Supplementary Figs. 2–56), revealed no substantial genome-wide heterogeneity across the 36 cohorts. Furthermore, we performed subgroup-specific meta-analyses, subdividing our study cohorts based on (1) their ascertainment strategy (five subgroups: clinical, comorbidity, community, biobanks and self-reported professional diagnosis (SRPD); Manhattan and quantile–quantile plots in Supplementary Figs. 58–62) and (2) their assessment strategy (three subgroups: interview, ICD-10 codes and SRPD; Manhattan and quantile–quantile plots in Supplementary Figs. 63–66). We then used confirmatory factor analysis in GenomicSEM43 to test whether these subgroups fit a one-factor model. In both cases, a single latent factor best explained the genetic covariance between the subgroups (ascertainment fit statistics: CFI = 1, SRMR = 0.04; assessment fit statistics: CFI = 1, SRMR = 3.67 × 10−9). The factor loadings across both subgroup models were high (0.75–1), with the factor explaining 81.8% and 95.6% of the total genomic variance in the ascertainment and assessment models, respectively (see Supplementary Note 3 for details on the subgrouping and Supplementary Table 5 and Supplementary Fig. 67a,b for GenomicSEM results). Using parallel analysis based on multivariate LDSC (paLDSC44), we identified one non-spurious dimension in exploratory genomic factor analysis, including 14 cohorts with more than 10,000 individuals and at least 1,000 cases. This finding supports our hypothesis that the genetic association signals were generally consistent across samples and study designs and tapped into a common underlying ANX genetic vulnerability.

Replication and validation of GWAS SNPs

We conducted two replication analyses of the 58 significant loci: one in a large independent EUR ANX GWAS from 23andMe, and the other in an African-American (AFR) ancestry ANX GWAS from the Veterans Affairs Million Veteran Program (MVP). The 23andMe sample consisted of 1,175,012 ANX self-report cases and 1,956,379 controls (see Methods for details). Among the 58 SNPs identified in the discovery GWAS, all but one (rs7121169) were available for replication testing in the 23andMe genotype platform. Two additional variants failed quality control procedures (rs72704544 and rs11599236). Considering the remaining 55 loci tested, all showed the same direction of effect as the primary GWAS, and 51 were significant at a Bonferroni-corrected P value of P = 0.0009 (0.05 / 55) (Supplementary Table 6). At the time of this analysis, only the MVP had published an ANX GWAS in a reasonably sized non-EUR sample (MVP-AFR: military ascertainment, AFR ancestry; 5,664 cases and 26,410 controls)34. Analyzing those data, we compared the direction of effect and P values of association for our 58 lead SNPs to examine consistency with our EUR results (Supplementary Table 7). Among the 53 SNPs available in MVP-AFR, only 27 (50.9%) showed the same sign. Given differences in LD and allele frequency between EUR and AFR genomes, we also searched for the most significant SNP in a 50-kb window around each lead SNP in the MVP-AFR cohort. A total of 36 of these SNPs were nominally associated, but only two were significantly associated after adjustment for multiple testing.

We further compared our associations with those reported in previous ANX case–control GWAS31,32,33,34,37 (Supplementary Table 8). A recent GWAS using broader anxiety-related case–control and symptom-based phenotypes reported 40 EUR-ancestry significant SNPs45; all but one showed the same direction of effect, while ten were also GWS in our analysis. Importantly, most of the associations in our GWAS are novel discoveries, with only 15 reported in prior ANX GWAS. We note that some of the previously identified SNPs are in LD with each other, and all previously published ANX GWAS partially overlap with our samples. Therefore, these are not independent replications but demonstrate the consistency of results when additional samples are incorporated.

To study the generalizability of our results across different ancestral groups, we tested the extent to which polygenic risk scores (PRS) derived from our GWAS (excluding UK datasets) predicted ANX in the UK Biobank for participants of EUR, AFR and South Asian ancestry (see Supplementary Table 9). The PRS predicted 2.27% of the variance (P < 2.0 × 10−16) in ANX liability for those of EUR ancestry, assuming a prevalence of 20%. The variance explained for those of South Asian and AFR ancestries was 1.94% (P = 6.37 × 10−5) and 0.54% (P = 0.051), respectively, revealing significant polygenic overlap across EUR and South Asian ancestries.

Characterization and functional annotation of GWAS SNPs

To identify potential causal variants, we conducted statistical fine mapping of our GWS loci using FINEMAP (v.1.3.1) with stringent inclusion thresholds46. This process identified six credible SNP sets defined as having a posterior probability of >0.95 and five or fewer SNPs per credible set to avoid excessive false positive rates (Supplementary Table 10). The lead SNPs of these credible sets were located at the following chromosomal positions: 3:67,895,104 (within SUCLG2-GT), 10:104,654,873 (within SORCS3), 17:8,187,590 (near TRI-AAT-5) and 20:20,876,379 (near KIZ); and two within the major histocompatibility complex (MHC) region: 6:28,329,086 (within ZSCAN31) and 6:30,170,699 (within TRIM15).

To examine the biological relevance of our GWS SNPs, we performed functional annotation in FUMA (v.1.6.1) to link our GWS SNPs with expression quantitative trait loci (eQTL) and brain chromatin interaction (Hi-C) data. The results suggest that most of the identified loci were associated with established gene regulatory mechanisms (circos plots in Supplementary Figs. 68–87). Although these results on their own do not provide enough evidence for involvement of respective genes in the etiology of ANX, they add to a broader picture that includes our summary-data-based Mendelian randomization (SMR) and other analyses (Supplementary Table 20).

We conducted stratified LDSC to partition the heritability into different functional genetic annotations and cell types. As noted in Supplementary Table 11, the association signal is highly conserved across species and significantly enriched for introns, monomethylated and polyacetylated histone marks (H3K4me1 and H3K4ac) and DNase I hypersensitivity sites in both adult and fetal tissues. Similar to other psychiatric GWAS, our findings are enriched for certain non-coding features rather than coding regions. Cell-type-specific enrichment was observed for central nervous system structures, including multiple cortical and subcortical areas, as well as cervical spine.

We also examined whether genetic associations with ANX were enriched among transcriptomic profiles of human tissues and/or individual cell types, using FUMA (v1.6.1)47. Tissue-enrichment analyses for general tissue types using data from the GTEx (v.8) consortium suggested that the expression patterns related to brain and pituitary tissues were significantly associated with the genetic risk of ANX (P = 1.18 × 10−13 and P = 6.50 × 10−5, respectively; Supplementary Table 12a and Supplementary Fig. 88). All individual brain tissues showed significant enrichment (Supplementary Table 12b and Supplementary Fig. 89), with cortex overall (P = 2.62 × 10−12) as well as frontal and anterior cingulate cortices and nucleus accumbens as most significant. At the level of individual cell types, we found a consistent association of GABAergic neurons with genetic variation associated with ANX (Supplementary Fig. 90). Our strongest association (P = 3.24 × 10−8) was found with GABAergic neuroblasts (via GSE76381)48.

Gene-based association and enrichment

Using MAGMA (v.1.08)49, we identified 91 significantly associated genes (adjusted P < 0.05 / 18,490 = 2.7 × 10−6; Supplementary Table 13). Historically interesting candidates include CLOCK, GABBR1, PCLO, NCAM1 and DRD2.

To test whether our loci significantly co-localize with known functional QTLs, we used SMR50 to conduct transcriptome-wide, proteome-wide and methylome-wide analyses (T-SMR, P-SMR and M-SMR, respectively). We used the largest available eQTL, protein QTL and methylation QTL reference datasets, respectively, for both brain and blood tissues (Supplementary Table 14). By using the conservative P values adjusted for the HEIDI test (see Methods), we detected 27 Bonferroni-corrected significant genes or isoforms in the brain associated with changes in the methylome, 16 in the transcriptome and seven in the proteome (Supplementary Tables 15–17). To improve signal detection in brain transcriptome and methylome data, we used Primo51 to jointly analyze blood and brain statistics (see ref. 52). We did not jointly analyze proteome data because of the low number of brain probes. These between-tissue concordance analyses yielded 22 significant ANX signals (posterior probability of >0.95) for the transcriptome and 133 for the methylome (Supplementary Tables 18 and 19). BTN3A2 remains a leading signal in both analyses, and interesting sub-threshold genes from single-tissue analyses become strong findings in the joint T-SMR (ZDHHC5, FURIN and NEGR1).

To highlight genes for which there was the strongest support, we summarized the findings across multiple (equally weighted) analyses in Supplementary Table 20, which includes an expanded set of 151 genes associated with ANX susceptibility. Starting with the 91 significant associations from MAGMA, we added genes supported by joint T-SMR or joint M-SMR with a posterior probability of >0.95. We annotated these using additional support from P-SMR, eQTL and Hi-C data. Figure 2 lists the 66 genes with three or more sources of support (score of ≥3). Most of these have prior reported associations with one or more psychiatric phenotypes, possibly suggesting gene-based pleiotropy, while a small proportion appear specific to ANX risk (reviewed in the Discussion).

Fig. 2: List of 66 most highly supported ANX genes.

Genes that were implicated in at least three of the six SNP-based (eQTL, Hi-C) or gene-based (MAGMA, M-SMR, P-SMR, T-SMR) tests. The left side indicates the position of the gene in the genome. Significance is indicated by a colored dot. eQTL (blue dots) compares results from brain-related eQTL studies for overlap in significance between our GWAS and the eQTL studies. Hi-C (green dots) uses brain-related Hi-C information available through FUMA to functionally annotate our results. MAGMA (gray dots) tests genetic associations at the gene level for the combined effect of SNPs in or near protein-coding genes. M-SMR, P-SMR and T-SMR (yellow, red and pink dots, respectively) refer to transcriptome-wide, proteome-wide and methylome-wide analyses that assessed likely causal associations between traits and genes, proteins and genomic regions by inferring the association between the trait and gene expression, protein concentration and methylation, as predicted from genomic data.

To test whether pre-existing gene sets are enriched for our ANX risk loci, we examined 10,894 gene sets obtained from MsigDB (v.5.2) (curated gene sets, 4,728; Gene Ontology terms, 6,166). Specifically, we used MAGMA to test for enrichment of our ANX signals (see Supplementary Table 21). Overall, one gene set was significant after correction for multiple testing: dawson_methylated_in_lymphoma_tcl1 (P = 1.71 × 10−6), including 57 genes that are hypermethylated in at least one of the lymphoma tumors in transgenic mice overexpressing TCL1 in germinal center B lymphocytes; the top three genes were also supported by T-SMR or M-SMR (NCAM1, HMGN1 and ZDHHC5). On the surface, it is difficult to appreciate the relevance of this cancer gene pathway for anxiety etiology. We also note that the overlap between this gene set and MAGMA gene signals is small (three out of 54; namely, NCAM1, HMGN1 and ZDHHC5). Among the next highly associated sets were genes related to commissural neuron axon guidance (P = 5.24 × 10−5) and GABAergic synapse (P = 9.67 × 10−5), the latter with 66 genes, including GABBR1, DRD2, CDH13 and LRFN5.

Gene–drug associations

To reveal possible drug repurposing opportunities for ANX, we used DrugTargetor53 (v.1.3) with our main ANX summary statistics. Among the 161 drug classes analyzed, several that are already successfully being used for ANX treatment demonstrated significant associations (q valueBF < 0.05; Supplementary Table 22): psycholeptics (drugs with a calming effect) and psychoanaleptics (mostly antidepressants), as well as other sedating drugs like antihistamines, antipsychotics, general anesthetics and opioids. However, none of the more than 1,500 individual compounds cataloged in ChEMBL54 and DgiDB55 yielded a significant signal (Supplementary Table 23), possibly because of the moderate power of this GWAS.

Genetic overlap between ANX and other phenotypes

To examine the overlap between our ANX association signals and other phenotypes, we conducted a phenome-wide association study (PheWAS). Of the 58 SNPs significantly associated with ANX, 15 were deemed ANX-specific (red diamonds in Fig. 3); that is, variants not reported as GWS in other extant GWAS. A total of 43 variants were associated with at least one other phenotype. We note that the higher number of overlapping associations with cardiometabolic, hematological and immunological outcomes reflects both the robust genetic architectures of these phenotypes and the number of GWAS that have been published in these domains. Overlap of ANX-related SNPs with cardiometabolic and hematological traits was heavily skewed towards a subset of variants (rs2710323, rs58825580 and rs174560). Figure 4 depicts a dendrogram-based heatmap showing the association with psychiatric or personality traits among 24 possibly pleiotropic SNPs (other heatmaps for cognitive and behavioral domains are found in Supplementary Figs. 91 and 92). Not surprisingly, more ANX SNPs overlap with internalizing phenotypes (neuroticism, depression) than with psychotic disorders (schizophrenia, bipolar disorder).

Fig. 3: Overview of SNP associations with other phenotypes.

The (rotated) Manhattan plot of the −log10(P values) of the ANX meta-analysis (left; as in Fig. 1) and PheWAS alluvial plot of potentially pleiotropic variants (right). The colored ribbons depict variants that are associated with at least one other published GWAS finding and correspond with the color of the ribbon in the alluvial plot. The red diamonds in the Manhattan plot depict the most significant variant in the region corresponding with potentially ANX-specific SNPs; that is, a variant that reached the GWS threshold for ANX but not in any other published GWAS.

Fig. 4: Heatmap of SNP associations with other psychiatric and personality traits.

Dendrogram-based heatmap indicating the number of unique GWS associations with psychiatric or personality traits among 24 SNPs that reach significance for multiple such phenotypes. Shading indicates the number of GWAS reporting associations between a specific SNP and the outcomes. Symptom dimensions (mood disturbance, mania, psychosis) and self-reported professional diagnoses (depression, anxiety, distress) are from the UK Biobank.

We used bivariate LDSC to estimate the genetic correlations between ANX and a wide variety of other traits. We included 112 previously published GWAS on various traits, including psychiatric, substance use, cognition or socioeconomic status, personality, psychological, neurological, autoimmune, cardiovascular, anthropomorphic, dietary and fertility phenotypes. After false discovery rate correction, we found that 82 traits showed significant genetic correlation with ANX (Fig. 5 and Supplementary Table 24). Among the psychiatric disorders and traits, ANX showed the strongest correlations with MDD (({r}_{g}=0.91)), followed by childhood internalizing symptoms ({(r}_{g}=0.76)), mood disturbance ({(r}_{g}=0.76)), symptoms of depression ({(r}_{g}=0.71)), post-traumatic stress disorder (PTSD) ({(r}_{g}=0.71)), psychosis ({(r}_{g}=0.68)), mania ({(r}_{g}=0.66)), suicide attempt ({(r}_{g}=0.58)) and obsessive–compulsive disorder ({(r}_{g}=0.41)). Genetic correlations were also high with total neuroticism score ({(r}_{g}=0.70)) and its various clusters and items. We found somewhat lower correlations with other psychiatric and substance-use disorders. ANX genetic risk was also modestly correlated with that of several neurological disorders, as well as adult-onset asthma and heart disease (positive) and inflammatory bowel diseases (negative). As shown in Supplementary Figs. 93 and 94 and Supplementary Table 24, the different ANX data subgroups show a variable but overall similar pattern of correlations.

Fig. 5: Genetic correlations (*rG*) between the main ANX GWAS and 112 phenotypes.**

Genetic correlations (r**g) between ANX and psychiatric, substance use, cognition/socioeconomic status (SES), personality, psychological, neurological, autoimmune, cardiovascular, anthropomorphic/diet, fertility and other phenotypes. References and sample sizes of the corresponding summary statistics of the GWAS studies can be found in Supplementary Table 24. The ANX summary statistics are of the main meta-analysis (ncases = 122,341; ncontrols = 729,881). Red circles indicate significant associations with a P value adjusted for multiple testing with the Benjamini–Hochberg procedure to control the false discovery rate (FDR < 0.05). Black circles indicate associations that are not significant. Error bars represent 95% confidence intervals for the genetic correlation estimates. ADHD, attention-deficit hyperactivity disorder; ALS, amyotrophic lateral sclerosis; BMI, body mass index; embarras., embarrassment; freq., frequency; fr., from; HDL, high-density lipoprotein; LDL, low-density lipoprotein; neurot., neuroticism; nr., number; OCD, obsessive–compulsive disorder; sat., satisfaction.

These results highlight the complex interrelations between the three internalizing phenotypes that also have the highest genetic correlations with ANX: MDD56, PTSD57 and neuroticism39. To examine potential directional effects underlying these correlations, we applied bi-directional generalized SMR (GSMR)58 with the latest available GWAS summary statistics. These results (Supplementary Table 25) indicate a highly significant bi-directional effect between ANX and each of these phenotypes. Based on beta-values, the strength of reverse (MDD → ANX = 0.657) and forward (ANX → MDD = 0.545) effects are similar between ANX and MDD. However, both PTSD (PTSD → ANX = 0.891 vs ANX → PTSD = 0.239) and neuroticism (neuroticism → ANX = 1.25 vs ANX → neuroticism = 0.17) effects on ANX are stronger than the reverse.

Discussion

In this GWAS meta-analysis, we identified 58 independent genome-wide loci associated with anxiety risk by including data from a composite phenotype created from five lifetime anxiety disorders (36 cohorts including 122,341 ANX cases and 729,881 controls; neffective = 390,560). Three-quarters of the identified variants are novel, with only 15 reported in prior anxiety GWAS. A total of 51 of these SNPs were replicated in an independent EUR-ancestry sample from 23andMe, strengthening their relevance. These results represent a major advance in identifying validated susceptibility loci for anxiety disorders.

The SNP-based heritability estimated at 10.1% captures approximately one-quarter of the broad-sense heritability from twin studies of adult ANX17, similar to other complex traits like MDD40. We divided the cohorts into subgroups based on ascertainment and assessment strategies and conducted separate GWAS as a sensitivity test. We observed moderate to high genetic correlations between these subgroups, supporting our decision to combine all samples into a single meta-analysis. SNP-based heritability varied from 23.7% in the clinical subgroup to 6.9% in the community subgroup (ascertainment) and from 7.7% in the interview subgroup to 13.2% in the ICD-10 subgroup (assessment), consistent with the hypothesis that more severe syndromes have higher heritability59,60,61. The overall meta-analytic SNP heritability is probably diminished by the effects of heterogeneity across these subgroups.

Along with replication in an independent EUR cohort from 23andMe (51 loci replicated at a Bonferroni-corrected P value), we tested the transferability of our results. First, we examined replication in the MVP-AFR ancestry sample, in which nominally significant proxy loci were identified for 36 lead SNPs, but only two showed significant association after Bonferroni adjustment. This is not surprising given both ancestry and ascertainment differences. Second, we applied PRS to estimate the variance explained in ANX liability. The PRS explained 2.27% of the variance in EUR individuals, which is comparable to PRS reports of MDD40. We then tested whether our findings would generalize to non-EUR samples. The EUR-ANX PRS explained 1.94% of the variance in the South Asian subsample of UK Biobank (significant) but only 0.54% for the AFR subsample (non-significant), in line with the low replication in the MVP-AFR ancestry cohort. This shows that for anxiety, as for other phenotypes, genetic liability estimated from EUR samples more closely reflects that of South Asian than AFR ancestry31. These findings stress the need for more diverse ancestry inclusion in future ANX GWAS.

Using LDSC, we found that, consistent with prior twin studies and extant GWAS, ANX shares the largest genetic overlap with MDD (r**g = 0.91), with which it has the highest lifetime comorbidity. This is followed by PTSD (r**g = 0.71), which is expected given their high comorbidity and the prior classification of PTSD among anxiety disorders[62](https://www.nature.com/articles/s41588-025-02485-8#ref-CR62 “American Psych