Genetic architecture distinguishes tinnitus from hearing loss

Abstract

Tinnitus is a heritable, highly prevalent auditory disorder treated by multiple medical specialties. Previous GWAS indicated high genetic correlations between tinnitus and hearing loss, with little indication of differentiating signals. We present a GWAS meta-analysis, triple previous sample sizes, and expand to non-European ancestries. GWAS in 596,905 Million Veteran Program subjects identified 39 tinnitus loci, and identified genes related to neuronal synapses and cochlear structural support. Applying state-of-the-art analytic tools, we confirm a large number of shared variants, but also a distinct genetic architecture of tinnitus, with higher polygenicity and large proportion of variants not shared with hearing difficulty. Tissue-expression analysis for tinnitus infers broad enrichment across most brain tissues, in contrast to hearing difficulty. Finally, tinnitus is not only correlated with hearing loss, but also with a spectrum of psychiatric disorders, providing potential new avenues for treatment. This study establishes tinnitus as a distinct disorder separate from hearing difficulties.

Fig. 1. Risk locus discovery and polygenic risk score analyses for tinnitus across different ancestries and data sources.

a Overlaying Manhattan plots from meta-analyses of tinnitus GWAS, showing 29 genome-wide significant (GWS) loci for the European ancestry (EA; full circles), and 30 GWS loci for the cross-ancestry (hollow circles) analyses. The y axis represents -log₁₀ p-values from two-sided z-test for meta-analyses effect estimates. The red line represents genome-wide significance at p < 5 × 10⁻⁸. b Genetic risk score (PRS) predictions for tinnitus comparing different training and target data. The y axis represents tinnitus odds ratios relative to the lowest quintile of PRS. Cross-dataset predictions from the EA MVP to UKB (red circles) and UKB to EA MVP (orange circles) indicate similar prediction accuracies. Using the EA GWAS meta-analysis as training data (full circles in a) to predict tinnitus in a non-overlapping sample from MVP release 4 (MVP R4, blue circles) indicates that individuals in the highest quintile have 65% higher odds to develop tinnitus than individuals in the lowest quintile. Lower prediction accuracies are found using the EA meta-analysis to predict tinnitus in subjects of non-European ancestry (Hispanics (LAT) (purple circles) and African (AA) (light blue circles)). Sample sizes are shown in Supplementary Data 8.

Fig. 2. Gene-based analyses of tinnitus identify 62 genes and gene expression enriched in brain tissue and specific ear cells.

a Manhattan plot of gene-based tinnitus meta-analysis across 481,874 subjects of European ancestry from MVP and UKB, showing 62 genome-wide significant genes. The y axis represents -log₁₀ p values from the two-sided F-test for gene-based analysis. The red dotted line indicates the gene-wide significance threshold at p < 2.66 × 10⁻⁶ (Bonferroni correction for 18,972 genes tested). b MAGMA tissue expression analysis for gene expression of GTEx v8 data sets, showing significant enrichment in the brain (inlet), with 11 of 13 brain regions being enriched. To test for positive relationships between gene expression in a specific category and genetic associations, SNPs were mapped to 17,196 protein-coding genes and gene-property tests were performed for average gene-expression per tissue type conditioning on average expression across all tissue types. (Note: cerebellar hemisphere and cerebellum are the same tissue, with different RNA preservation after death). Bars denote -log₁₀ p values from one-sided t-tests. The dotted line indicates significance at p < 9.26 × 10⁻⁴ (Bonferroni correction for 54 specific tissues tested) and at p < 1.67 × 10⁻³ (Bonferroni correction for 30 general tissues tested) in the inlet. c Diagram of the ear, with red arrows showing direction of sound waves through the external ear, vibrating through the tympanic membrane and ossicles to the oval window into the inner ear. The inner ear is composed of three fluid-filled chambers, and the Organ of Corti rests atop the basilar membrane (black) in the center chamber, the scala media (orange in the upper figure). Organ of Corti is shown in cross-section below. Stria vascularis, the locale for melanocytes, maintains the endocochlear potential required for outer hair cell (OHC) and inner hair cell (IHC) nerve firing. Deiter’s cells support OHCs, and IHCs are surrounded medially by border cells and laterally by inner pillar cells. OHCs lengthen and shorten, acting as a dampener/amplifier of the fluid wave. IHCs provide mechanotransduction, sending a neural signal through cochlear nerve fibers to the brain. Diagram of the ear from Encyclopædia Britannica, Inc., copyright 2009. All rights reserved. d Cochlear cell type enrichment analyses based on mouse data (Jean et al. 2023). Clustered bar chart depicting results of conditional MAGMA gene-property analyses for 34 different cell types. Bars (denoting -log₁₀ p-values from one-sided t-tests) are clustered and colored by the general cell type tested: circulating (red), glia (blue), hair (light blue), lateral wall (gray), neurons (royal blue), supporting (yellow), and surrounding structures (purple). Bars depict the cell type association conditioned on the average of all 34 cell types. Solid line denotes p < 0.05. Dotted line indicates significance after Bonferroni-adjustment for 34 comparisons (p < 1.47 × 10⁻³). e Organ of Corti cell type enrichment analyses based on mouse data (Hoa et al. 2020). Bar chart depicting results of conditional MAGMA gene-property analyses for five different cell types. Bars (denoting -log₁₀ p-values from one-sided t-tests) are colored by the cell type tested: hair cells (blue), supporting cells (yellow), and melanocytes (gray). For each tissue type, the bar depicts MAGMA analysis conditioned on the average of the five cell types. Solid line denotes p < 0.05. Dotted line indicates significance after Bonferroni-adjustment for five comparisons (p < 0.01).

Fig. 3. Genetic architecture of tinnitus and its relationship to hearing difficulty (HD).

a Quantification of the polygenic overlap between tinnitus in the MVP and UKB. Bivariate MiXeR modeling estimates 8,262 shared causal variants between the cohorts (gray), accounting for the majority of variants influencing tinnitus (88.7% in MVP, 91.8% in UKB, respectively), with very little unique polygenic components (blue shades). b Polygenic overlap between tinnitus and hearing difficulty (HD), indicating an estimated 3,850 causal variants are shared, thus accounting for 95.4% of the variants influencing HD, but only 40.7% of the variants influencing tinnitus. The numbers (in thousands, with standard errors in parenthesis) in the Venn diagrams indicate the estimated quantity of causal variants per component, explaining 90% of SNP heritability for each phenotype. The size of the circles reflects the degree of polygenicity. Genetic correlations (rg) estimated between the two phenotypes are shown below the Venn diagrams. c–e Manhattan plots of tinnitus GWAS and different models incorporating hearing difficulty indicate unique and shared risk loci between the two phenotypes. The y axis represents -log₁₀ p-values from two-sided z-tests. The red dotted line indicates the genome-wide significance threshold at p < 5 × 10−⁸. c GWAS meta-analysis of tinnitus, including participants from MVP (N = 308,879) and UKB (N = 172,995) identified 29 genome-wide significant (GWS) loci. The red line indicates the genome-wide significance threshold at p < 5 × 10−⁸. Diamonds indicate 8 tinnitus loci that will remain significant after adjustment for hearing difficulty (in d). d Tinnitus meta-analysis including a covariate for HD (based on ICD and self-report). Diamonds indicate 4 tinnitus loci that become significant after adjustment for HD. e Case-case GWAS of tinnitus and HD, showing 10 loci with significantly different effects between the two phenotypes. Blue diamonds indicate 2 loci with larger effects for tinnitus, green diamonds indicate 7 HD loci. The red diamond represents a locus with opposite effect (not GWS in either of the GWAS).

Fig. 4. Enrichment of specific brain regions in MAGMA tissue expression analyses, comparing GWAS on tinnitus with a GWAS on tinnitus adjusted for hearing difficulty, with a hearing difficulty GWAS, and a case-case GWAS on tinnitus versus hearing difficulty.

MAGMA tissue expression analysis for gene expression of GTEx v8 data sets, showing significant enrichment in specific brain regions for GWAS meta-analyses across the same set of MVP and UKB European ancestry participants (N = 481,874). To test for positive relationships between gene expression in a specific category and genetic associations, SNPs were mapped to 17,196 protein-coding genes and gene-property tests were performed for average gene-expression per tissue type conditioning on average expression across all tissue types. Bars denote -log₁₀ p-values from one-sided t-tests. The dotted line represents Bonferroni-adjusted significance at p < 9.26 × 10⁻⁴ for 54 specific tissues (only brain tissues are shown here, none of the other tissues were significant).

Fig. 5. Genetic correlations of tinnitus with psychiatric disorders and health related traits.

a Genetic correlations of tinnitus with 789 health-related traits and disorders across 13 domains. GWAS summary data was derived from the Psychiatric Genomics Consortium and the Complex Trait Genetics Virtual Lab (see Methods and Supplementary Data 18). The y axis represents -log₁₀ p-values from the two-sided z-tests for genetic correlations. Orientation of the triangle indicates positive (up) or negative (down) correlations. Black arrow indicates that 3 hearing-related traits (from UKB) with p-values extending beyond the range of the X-axis (p = 1.26 × 10⁻⁷²). Dotted line indicates Bonferroni-corrected significance at (p < 6.34 × 10⁻⁵). Selected psychiatric (blue, b) and health-related (red, c) traits were included in gSEM analyses. b Path diagram and standardized estimates from the best fitting confirmatory-factor model (CFA) of tinnitus, hearing difficulty, and 10 psychiatric disorders. Exploratory factor analyses of the genetic correlation matrix produced from multivariable LD-score regression of odd chromosomes were used to inform CFAs, fit to the covariance matrix from the even chromosomes. In this best-fitting CFA, four correlated latent genetic factors (F1_g, F2_g, F3_g, F4_g,) represent shared genetic liability for the conditions. Single-headed arrows represent partial regression coefficients and reflect the degree of relationship between the latent factor and each variable. Variation explained by latent factors can be computed by squaring the factor loadings. Curved, double-headed arrows represent correlations between factors. Unique variance not explained by the model in each condition is represented by the ‘u’ oval estimates. c Path diagram and standardized estimates from the best fitting CFA of tinnitus and 13 health-related traits. TIN tinnitus, HD hearing difficulty, PTSD post-traumatic stress disorder, ADHD attention-deficit/hyperactivity disorder, ALCH problematic alcohol use, MDD major depressive disorder, TS Tourette’s syndrome, AN anorexia nervosa, OCD obsessive compulsive disorder, SCZ schizophrenia, BP bipolar disorder, ASD autism spectrum disorder, SWB subjective well being, NEUR neuroticism, SLEEP sleep duration, STRESS No Illness/injury/bereavement stress in last 2 years, PAIN1 Neck or shoulder pain experienced in last month, PAIN2 no pain experienced in last month, TIRED frequency of tiredness/lethargy in last 2 weeks, LILL long-standing illness/disability/infirmity, ILLN Number of self-reported non-cancer illnesses, TXN Number of treatments/medications taken, MED1 no medication for pain relief/constipation/heartburn used, MED2 Paracetamol used, MIG migraine.