The cells of the sensory epithelium, and not the stria vascularis, are the main cochlear cells related to the genetic pathogenesis of age-related hearing loss

Summary

Age-related hearing loss (ARHL) is a major health concern among the elderly population. It is hoped that increasing our understanding of its underlying pathophysiological processes will lead to the development of novel therapies. Recent genome-wide association studies (GWASs) discovered a few dozen genetic variants in association with elevated risk for ARHL. Integrated analysis of GWAS results and transcriptomics data is a powerful approach for elucidating specific cell types that are involved in disease pathogenesis. Intriguingly, recent studies that applied such bioinformatics approaches to ARHL resulted in disagreeing findings as for the key cell types that are most strongly linked to the genetic pathogenesis of ARHL. These conflicting studies pointed either to cochlear sensory epithelial or to stria vascularis cells as the cell types most prominently involved in the genetic basis of ARHL. Seeking to resolve this discrepancy, we integrated the analysis of four ARHL GWAS datasets with four independent inner-ear single-cell RNA-sequencing datasets. Our analysis clearly points to the cochlear sensory epithelial cells as the key cells for the genetic predisposition to ARHL. We also explain the limitation of the bioinformatics analysis performed by previous studies that led to missing the enrichment for ARHL GWAS signal in sensory epithelial cells. Collectively, we show that cochlear epithelial cells, not stria vascularis cells, are the main inner-ear cells related to the genetic pathogenesis of ARHL.

Two recent studies that integrated genome-wide association study and single-cell transcriptomics analyses resulted in conflicting results as for the key cell types linked to the genetic pathogenesis of age-related hearing loss (ARHL). We show that cochlear epithelial cells, not stria vascularis cells, are the main cells for the pathogenesis of ARHL.

Main text

Age-related hearing loss (ARHL) is a major health concern, affecting ∼25% of people 65–74 years old and ∼50% of those older than 75.^1,2,3 ARHL is associated with social withdrawal, depression, and cognitive decline.⁴ At present, there is no preventive treatment for ARHL. It is expected that increasing our understanding of pathophysiological processes that underlie ARHL will pave the way for more rational development of novel therapies. Recent genome-wide association studies (GWASs) discovered a few dozen genetic variants in association with elevated risk for ARHL.^5,6,7 Integrated analysis of GWAS results and transcriptomics data have proved to be very powerful in elucidating specific cell types and molecular pathways that are involved in disease pathogenesis.^8,9,10 These bioinformatics approaches are based on the assumption that the cells most relevant for the disease pathophysiological processes are the cells in which the “disease risk genes” (that is, the genes affected by the disease risk variants) are most robustly expressed.

Intriguingly, two recent applications of such bioinformatics approaches to ARHL resulted in conflicting results as for the key cell types in the cochlea that are most strongly linked to this late-onset hearing impairment (the main cell types in the cochlea are shown in Figure S1). Kalra et al.⁵ conducted a meta-analysis of multiple hearing-related traits in the UK Biobank (n = up to 330,759) using GWAS summary statistics from Wells et al.⁷ and identified 31 genome-wide significant risk loci for self-reported hearing difficulty. Integrated analysis with transcriptomics data performed by this study pointed to cochlear epithelial cells as the key ones for the genetic pathogenesis of ARHL. More recently, Trpchevska et al.⁶ performed a genome-wide association meta-analysis of clinically diagnosed and self-reported hearing impairment on 723,266 individuals and identified 48 significant loci. In contrast to Kalra et al.,⁵ and rather unexpectedly, the integrated analysis with transcriptomic data carried out by Trpchevska et al.⁶ found no enrichment for ARHL risk signal in cochlear epithelial cells (cells from the organ of Corti—Deiters’ cells, inner hair cells [IHCs] and outer hair cells [OHCs]) (see Figure 2C in Trpchevska et al.⁶). However, their analysis revealed enrichment of ARHL risk signal in lateral wall spindle and root cells, and basal cells of the stria. The authors concluded that their findings strongly support a prominent role of the stria vascularis in the genetic basis of ARHL.⁶

We were puzzled by the contrasting conclusions made by these two recent studies and, in this study, sought to resolve this discrepancy. Specifically, we wished to determine which cochlear cells are the key mediators of ARHL genetic risk: the sensory epithelial cells of the organ of Corti or cells of the lateral wall and stria vascularis. To this goal, we integrated the analysis of the two ARHL GWAS datasets with three independent, previously published, and new mouse inner-ear single-nucleus RNA-sequencing (snRNA-seq) datasets (collectively spanning a broad range of ages from postnatal day 15 [P15] to 15 months)^11,12,13 and using two complementary computational approaches (MAGMA¹⁴ and sLDSC¹⁵—see supplemental methods). Our analysis clearly points to the cochlear epithelial cells as the key cells involved in the genetic predisposition to ARHL. Importantly, in all datasets, the cochlear epithelial cells were the most enriched for ARHL risk signal, and their enrichment was at least an order of magnitude stronger than in any of the lateral wall and stria vascularis cells (Figures 1 and S2A). To further substantiate our results, we analyzed two additional ARHL GWAS datasets that were recently published^16,17 (Figures S2B and S2C) and used a third computational approach, single-cell disease relevance score (scDRS),¹⁸ which calculates disease-association scores at the single-cell level (Figure S2D). Collectively, our observation was consistent over four GWAS datasets, three computational approaches, and various scRNA-seq datasets and ages.

Figure 1.

Enrichment of different cochlear cell types for ARHL genetic risk signal

Enrichment for ARHL signals obtained using the GWASs from Wells et al.⁷ and the MAGMA tool. Dashed bar represents the 5% Bonferroni correction threshold. B, B cell; BC, basal cell; CC, chondrocyte; OB, osteoblast; CEC, capillary endothelial cell; DC/PC, Deiters cell and pillar cell; EC, epithelial cell; FB, fibroblast; FC1, fibrocyte 1; FC2, fibrocyte 2; FC3, fibrocyte 3; FC4, fibrocyte 4; HC, hair cell; HeC, Hensen’s cell; IC, intermediate cell; IHC, inner hair cell; IPhC/IBC, inner phalangeal cell and inner border cell; ISC, inner sulcus cell; M/MP, macrophage; MC, marginal cell; Neu, granulocyte/neutrophil; Nudt4+, Nudt4+ pillar cell; OHC, outer hair cell; OSC, outer sulcus cell; PVM_M, perivascular resident macrophage-like melanocyte; RBC, red blood cells; RMC, cells in Reissner’s membrane; SC, Schwann cell; SGC, satellite glial cell; SGN, spiral ganglion neuron; SL, spiral ligament; SMC, smooth muscle cell; SpC/RC, spindle cell/root cell; SV, stria vascularis; T, T cell; TBC, tympanic border cell.

We next turned to examine why the analysis performed by Trpchevska et al.⁶ failed to detect enrichment of ARHL GWAS signal in the cells of the organ of Corti. This study analyzed three inner-ear transcriptomic datasets: two datasets of mouse cochlear scRNA-seq from Milon et al.,¹⁹ which included the spiral ganglion region and the lateral wall and stria vascularis (neither set included IHCs, OHCs, or Deiters' cells), and a third dataset from Ranum et al.,²⁰ which contained only IHCs, OHCs, and Deiters' cells. Since the study performed by Ranum et al.²⁰ used a different transcriptomic profiling method than the study from Milon et al.,¹⁹ Trpchevska et al.⁶ analyzed these datasets separately. Their analysis was based on the calculation of gene expression cell-type specificity scores within each dataset. This index measures, for each gene, how specific its expression is in each of the profiled cell types relative to all other cell types included in the dataset. After calculating these scores, the top 10% most cell-type specific genes in each cell type were tested for enrichment for ARHL risk signal. Using this approach, no enrichment was detected in the epithelial sensory cells. However, importantly, note that cell-type specificity scores, on which the Trpchevska et al. analysis was based,⁶ strongly depend on the cell-type composition of the analyzed dataset. In particular, Ranum’s dataset contained only three cell types (Deiters’ cells, IHCs, and OHCs)²⁰ while Kalra et al.’s analysis included cells from the entire organ of Corti.⁵ Consequently, if ARHL risk is mainly mediated by genes whose expression is shared by IHCs, OHCs, and the Deiters' cells (or by any two of these cell types), an analysis that seeks enrichment for ARHL GWAS signals among the genes that are highly specific to only one particular cell type out of these three types will miss the effect. To test this hypothesis, we focused on the scRNA-seq from Xu et al.¹² (P28 dataset in Figure 1), as this dataset extensively covered a large spectrum of inner-ear cell types, including the main cell types of the organ of Corti and of the stria vascularis. We applied the same method used by Trpchevska et al.⁶ for defining the top 10% cell-type specific genes in each cell type. Notably, when the calculation of these specificity scores included all the cell types in the dataset, Deiters’ cells and OHCs were the two cochlear cell types most highly enriched for ARHL (Figure 2). However, when we confined this analysis only to the three cell types that were included in Ranum et al. dataset,²⁰ the enrichment of Dieters’ cells and OHCs for ARHL signal was completely lost (Figure 2).

Figure 2.

Effect of the dataset’s cellular composition on the results of GWAS signal enrichment tests

Enrichment for ARHL obtained when either all the cell types in the Xu. et al. dataset (P28)¹² or only Deiters’ cells, OHCs, and IHCs (inset) were considered in the analysis. Confining the analysis to only the three epithelial cell types significantly attenuates the enrichment of these cells for ARHL signal. Dashed bar represents the 5% Bonferroni correction threshold.

In conclusion, first, our results indicate that the epithelial cells of the organ of Corti, rather than cells of the stria vascularis, are the prominent ones for mediating the genetic pathogenesis of ARHL. Second, from the methodological perspective, our analysis emphasizes the importance of including a wide spectrum of cell types in statistical enrichment analyses that are based on cell-type specificity scores. Last, our analysis is focused on the genetic predisposition to ARHL. It does not rule out an important role for the lateral wall and stria vascularis in mediating environmental (e.g., noise exposure) effects on ARHL. One limitation of our study and previous related studies is the reliance on mouse inner-ear samples. At present, no scRNA-seq dataset is available for adult human cochlea. Given the inaccessibility of human inner-ear samples, emerging developments in generation of human organoid models hold promise as a viable alternative.²¹

Data and code availability

Raw and processed data for the snRNA-seq cochlear dataset generated in this study (month 3 [P90] mice) have been submitted to GEO. The accession number for this data is GEO: GSE234926. It has also been submitted to the gEAR portal (https://umgear.org/p?s=653896d7).

Published: February 7, 2024