Polygenic Risk Score-Based Association Analysis of Speech-in-Noise and Hearing Threshold Measures in Healthy Young Adults with Self-reported Normal Hearing

Abstract

Purpose

Speech-in-noise (SIN) traits exhibit high inter-subject variability, even for healthy young adults reporting normal hearing. Emerging evidence suggests that genetic variability could influence inter-subject variability in SIN traits. Genome-wide association studies (GWAS) have uncovered the polygenic architecture of various adult-onset complex human conditions. Polygenic risk scores (PRS) summarize complex genetic susceptibility to quantify the degree of genetic risk for health conditions. The present study conducted PRS-based association analyses to identify PRS risk factors for SIN and hearing threshold measures in 255 healthy young adults (18–40 years) with self-reported normal hearing.

Methods

Self-reported SIN perception abilities were assessed by the Speech, Spatial, and Qualities of Hearing Scale (SSQ12). QuickSIN and audiometry (0.25–16 kHz) were performed on 218 participants. Saliva-derived DNA was used for low-pass whole genome sequencing, and 2620 PRS variables for various traits were calculated using the models derived from the polygenic risk score (PGS) catalog. The regression analysis was conducted to identify predictors for SSQ12, QuickSIN, and better ear puretone averages at conventional (PTA_0.5–2), high (PTA_4-8), and extended-high (PTA_12.5–16) frequency ranges.

Results

Participants with a higher genetic predisposition to HDL cholesterol reported better SSQ12. Participants with high PRS to dementia revealed significantly elevated PTA_4-8, and those with high PRS to atrial fibrillation and flutter revealed significantly elevated PTA_12.5–16.

Conclusion

These results indicate that healthy individuals with polygenic risk of certain health conditions could exhibit a subclinical decline in hearing health measures at young ages, decades before clinically meaningful SIN deficits and hearing loss could be observed. PRS could be used to identify high-risk individuals to prevent hearing health conditions by promoting a healthy lifestyle.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10162-023-00911-4.

Introduction

Speech-in-noise (SIN) perception is necessary for performing many routine tasks. SIN processing is known to be challenging for individuals suffering from sensorineural hearing loss e.g., [1]. SIN performance shows large inter-subject variability, even for individuals with normal hearing thresholds [2]. Behavioral hearing thresholds are not highly predictive of SIN performance [3]. About 26 million US adults with normal hearing experience SIN difficulties in daily life [4], and around 5% of clinical patients seeking professional help for SIN difficulties in daily living exhibit normal conventional audiograms [5]. SIN abilities depend on a complex network of interdependent auditory-cognitive processes involving acoustic signal transduction, suprathreshold neural coding, auditory grouping (i.e., grouping parts of the acoustic signals), and unmasking speech signals at the cortical levels e.g., [6–9]. Cognitive processes involved in selective attention and working memory could be critical for providing lexical access to speech sounds e.g., [10, 11]. Emerging literature suggests that SIN measures are sensitive for detecting early-stage dysfunction in the auditory-cognitive system [12–14].

Neurophysiological processes underlying SIN deficits, such as temporal and spectral auditory processing, exhibit high heritability (h² > 0.7), suggesting that genetic variability could influence inter-subject differences in SIN performance [15]. Auditory neuropathy spectrum disorder, characterized by a higher degree of SIN deficit than predicted from hearing thresholds, has been associated with mutations in cochlear genes, such as OTOF, VGLUT3, and CABP2 (for review, [16]). Genetic variants in PMP22 causing Charcot-Marie-Tooth disease type 1, a common peripheral neuropathy, is linked to SIN deficit [17]. Cognitive processes that are necessary for SIN perception are influenced by genetic variability [18, 19]. The above evidence suggests that inter-subject variability in SIN perception could be influenced by genetic variability. Genetic interrogation of SIN perception could provide insight into the biological processes and comorbidities influencing inter-subject variability in SIN measures.

In the past decade, genome-wide association studies (GWAS) have uncovered the genetic basis of adult-onset and heritable human traits and health conditions. GWAS identified a highly polygenic architecture underlying complex human conditions, where genetic loading of multiple risk variants (typically in 100 s to 1000 s), each modifying the risk of a complex trait with a small effect, impact an individual’s probabilistic susceptibility to health conditions [20]. GWAS summary statistics, estimating the effect sizes across the risk alleles, could be used to construct a model for calculating polygenic risk scores (PRS). PRS is typically calculated as the standardized-effect size-weighted sum of the risk allele (i.e., 0, 1, or 2), summarizing a multifactorial genetic susceptibility of a complex trait into a single score for each person and thereby quantifying the degree of genetic risk e.g., [21, 22]. Polygenic score (PGS) catalog provides access to PRS models for various human traits and diseases [23]. PRS models could be used to calculate PRS for each individual, independently of the cohort included in the study, to obtain the polygenic susceptibility of reported traits. Large-scale studies involving population-based cohorts and electronic medical databases have documented the predictive utility of PRS models e.g., [24–26], despite methodological challenges in their design and verification [27].

The present study used the PGS catalog to calculate PRS variables for various human traits and health conditions (Supplementary File S1). The study investigated the genetic epidemiology of SIN perception using PRS in healthy young adults with self-reported normal hearing (N = 255). Low-pass whole genome sequencing data with imputation were used to calculate 2620 PRS variables based on the PRS models derived from the PGS catalog. SIN perception was evaluated with a self-reported questionnaire (i.e., Speech, Spatial, and Quality of Hearing Questionnaire—12 questions) and a standardized clinical assessment (i.e., QuickSIN). Hearing sensitivity was evaluated with puretone audiometry from 250 to 16,000 Hz.

Materials and Methods

The institutional review board of the University of Iowa approved the study protocol. The participants were recruited from the University of Iowa campus. Informed written consent was obtained from all participants before the data collection process.

Participants

An initial screening questionnaire was distributed via a campus-wide mass email. The initial screening questionnaire inquired about demographic details (age, sex, ethnicity), medical history, and hearing health (e.g., hearing loss, reoccurring ear infections). A total of 2248 individuals filled out the initial screening questionnaire. Individuals reporting no medical history and good health were invited for the present study. Individuals with active ear infections and middle ear conditions were excluded. A sample of 255 individuals (84 males and 171 females) aged 18–40 years (mean age: 21.7 years, SD = 3.4 years, range: 18–40 years) was recruited for the present study. All participants were English speakers (native or non-native) reporting normal hearing.

Audiometric Measures

Participants completed the Speech, Spatial, and Quality of Hearing Scale-12-item version (SSQ12) [28]. SSQ12 evaluates self-reported speech perception abilities on a 0–10 scale (10, perfect perceptual experience; 0, no perceptual experience). SSQ12 includes speech perception in the following pragmatic subscale: speech-in-noise, multiple speech streams, speech-in-speech, localization, distance and movement, segregation, identification of sound, quality of naturalness, and listening effort. The scores were averaged across 12 responses to derive the average SSQ12 score.

The participants underwent otoscopic examination and immittance audiometry (Titan IMP440, Interacoustics, Middelfart, Denmark). All audiometric procedures were conducted in a double-walled sound-treated booth meeting ANSI standards. Participants with normal otoscopic and tympanometric findings (e.g., Jerger type A tympanograms–static compliance: 0.3–1.75, middle ear pressure: + 50 to − 100 daPa) were tested with puretone audiometry and QuickSIN. Audiometric thresholds were tested with MedRx AVANT clinical audiometer (MedRx, Largo, FL) using the modified Hughson-Westlake procedure. Conventional hearing thresholds at 250, 500, 1000, 2000, 3000, 4000, 6000, and 8000 Hz were tested with insert receivers IP30 (RadioEar, Middelfart, Denmark). Extended high-frequency audiometry was performed at 9000, 10,000, 11,200, 12,500, 14,000, and 16,000 Hz with circumaural headphones DD450 (RadioEar, Middelfart, Denmark). Puretone average was calculated with HTs at 500, 1000, and 2000 Hz (PTA_0.5–2), with HTs at 4000, 6000, and 8000 Hz (PTA_4-8), and with HTs at 12,500, 14,000, and 16,000 Hz (PTA_12.5–16). All participants had symmetrical hearing (i.e., the difference between PTAs between ears < 20 dB). Better ear PTAs were used for the statistical analysis.

Speech-in-Noise Performance

Speech-in-noise performance was tested with QuickSIN [29]. It is a widely used clinical test with a superior ability to separate performance between individuals with and without hearing loss [30]. QuickSIN was performed binaurally at 70 dB HL fixed speech level with 3 test sets, each containing six sentences with a signal-to-noise ratio (SNR) ranging from 0 to 25 dB in a 5-dB step. The SNR loss is calculated for each test set by subtracting the correct words from 25.5. The average SNR loss across three trials was calculated to obtain suprathreshold speech-in-noise performance. The average QuickSIN SNR loss was used for the statistical analysis.

Saliva Sample Collection and Low-Pass Whole Genome Sequencing

Saliva samples were collected using Oragene OGR-600 DISCOVER kit (DNA Genotek, Ottawa, Canada). The saliva samples were stored at room temperature before DNA extraction and sequencing. DNA extraction and sequencing were performed by BGI Americas Corporation (Cambridge, MA). The salivary DNA sample was extracted using prepITL2P reagent kit (DNA Genotek, Ottawa, Canada). One microgram of genomic DNA sample was subjected to low-pass genomic sequencing. DNA was randomly fragmented, and fragmented DNA was selected by Agencourt AMPure XP-Medium kit (average size: 200–400 base pair). The selected fragments were through end-repair, 3′ adenylated, adapters-ligation, and polymerase chain reaction (PCR) amplifying processes. AxyPrep Mag PCR clean-up kit was used to recover the PCR products. The double-stranded PCR products were heat denatured and circularized using the splint oligo sequence. The single-strand circle DNAs were formatted as the final library and were subjected to quality control quantification. DNA sequencing was performed with the DNBSEQ-G400 platform on the qualified libraries achieving quality control standards. DNA sequencing data were subjected to an imputation pipeline generating genotype data for > 10 million markers (for methodological details, review [31]).

PRS Calculation

Custom software was used to calculate PRS. The methodological details of the PRS calculator are described elsewhere [21]. PRS calculation was performed with the weighted-allele count approach that assumes an independent effect across the genetic risk loci derived from the genome-wide association analysis. PRS calculation was performed using the following equation:

$$PRSi={\sum }_{j=1}^n\left(x_{\mathit{ij}},{\beta }_j\right)$$

Here, PRS for an individual i is calculated by adding β (i.e., effect size, usually defined as log odds ratio per dosage of each effect allele (0 ≤ $x_{\mathit{ij}}$≤2), derived from GWAS summary statistics) for genetic variant j. The PGS catalog was used to obtain weights for 2620 PRS models, including phenotypes such as sensory disorders, cardiovascular diseases, metabolic diseases, body measurement, cancer, biological processes, lipid or lipoprotein measurement, and neurological and inflammatory markers [23] (Supplementary file S1). The PGS catalog follows a standardized framework for reporting ancestry data [32], endorsed by the PRS reporting standards [33]. Most models were derived from GWAS conducted on individuals with European ancestry. The performance of some models was evaluated in other ancestral groups. PRS model-specific details are available on the PGS catalog website (https://www.pgscatalog.org/).

Statistical Analysis

Plink2 (version v2.00aLM, https://www.cog-genomics.org/plink2) was used to perform the principal component analysis (PCA) on the imputed genomic data, and genomic PCAs were calculated [34]. We conducted a multiple linear regression for the following outcome measures: SSQ12, QuickSIN, better ear PTA, and better ear EHF PTA (MATLAB, MathWorks, Inc., Natick, MA). Age, sex, and ethnicity were included as covariates. The top five genomic PCAs were included in the model to control for possible population stratification. The analysis was conducted using the standard methods employed by the genome-wide association studies, where the multiple linear regression model was fit to each PRS variable using the following equation:

$$Outcome=\mathrm{intercept}+\mathrm{B}1\ast \mathrm{age}+\mathrm{B}2\ast \mathrm{sex}+\mathrm{B}3\ast \mathrm{ethnicity}+\mathrm{B}4\ast \mathrm{PCA}1+\mathrm{B}5\ast \mathrm{PCA}2+\mathrm{B}6\ast \mathrm{PCA}3+\mathrm{B}7\ast \mathrm{PCA}4+\mathrm{B}8\ast \mathrm{PCA}5+\mathrm{B}9i\ast \mathrm{PRS}i$$

Here, i is a vector of PRS with a length of 2620 (i.e., a total number of traits), and beta (B1–B8) are regression coefficients for non-PRS variables. B9i is a regression coefficient for the ith PRS variable. B9i, adjusted R², p-value, and false-discovery rate (FDR)-adjusted p-value (FDR rate = 0.05) were calculated for all PRS variables. The Benjamini–Hochberg FDR-adjusted p-value < 0.05 was used as a statistical significance threshold to identify the PRS association with the outcome measures [35]. Two baseline models were created before including PRS variables in the analysis. The first model included age, sex, and ethnicity as predictors. The second model included age, sex, ethnicity, PCA1, PCA2, PCA3, PCA4, and PCA5. The baseline models estimated the effects of demographic factors and genomic PCAs on outcome measures.

Results

The study sample included 255 participants (84 males and 171 females) who completed the questionnaire. The study included 180 participants reporting European ethnicity. Two hundred eighteen participants (73 males and 139 females) agreed to the time commitment necessary for audiometric testing. Among the participants completing audiometric tests, 153 reported European ethnicity. Table 1 presents a correlation matrix for demographic and audiological measures. SSQ12 showed a significant negative correlation coefficient with PTA measures. QuickSIN and SSQ12 showed a trend of association (r =  − 0.011, p = 0.08), but the p-value could not achieve statistical significance. QuickSIN was associated with ethnicity, with participants reporting non-European ethnicity showing significantly higher SNR loss than their counterparts (mean difference (MD) = 1.24 dB, p < 10⁻⁶). Ethnicity showed no association with SSQ12 (p > 0.05). About 70% of participants reporting non-European ethnicity were non-native English speakers. The association analysis for SSQ12 was conducted on a sample of 255 participants. The association analyses for PTA_0.5–2, PTA_4-8, PTA_0.5–2, and QuickSIN were conducted on a sample of 218 participants with audiometric data.

Table 1.

A correlation matrix for the audiometric and demographic variables included in the analysis

	SSQ12	QuickSIN	PTA0.5–2	PTA4-8	PTA12.5–16	Age	Sex	Ethnicity
SSQ12	1
QuickSIN	− 0.119	1
PTA0.5–2	− 0.242**	0.121	1
PTA4-8	− 0.202**	0.069	0.441***	1
PTA12.5–16	− 0.152*	0.082	0.268**	0.543**	1
Age	0.173**	− 0.048	− 0.048	0.003	0.119	1
Sex	0.001	− 0.120	− 0.131	− 0.051	− 0.096	− 0.013	1
Ethnicity	− 0.008	− 0.339**	0.022	0.080	0.044	− 0.087	− 0.075	1

*p < 0.05, **p < 0.01, ***p < 0.001

PRS Associated with SSQ12 Scores

The results of the association analysis are presented in Table 2. High-density lipoprotein (HDL) cholesterol PRS significantly associated with SSQ12 scores. Participants with a higher genetic predisposition to HDL showed significantly higher (better) SSQ12 scores (Fig. 1). Four HDL PRS models included in the present study (PGS002597, PGS002548, PGS002695, PGS000845) showed significant association with SSQ12. HDL PRS variables were highly correlated with one another (p < 10⁻⁸) (Supplementary File S1). One hundred twenty-eight PRS variables revealed a suggestive association with SSQ12 (unadjusted p < 0.05), but they could not achieve the FDR-adjusted p-value threshold of 0.05. Apolipoprotein-A PRS (PGS0002101) revealed a suggestive association with SSQ12 (adjusted p = 0.063) (Table 2). Other top results included PRS related to dementia in Parkinson’s disease, sodium in the urine, and recent feelings of depression. SSQ12 showed suggestive association with neuroimaging phenotype-related PRS, such as volume of gray matter in parahippocampal gyrus (PGS001586), weighted-mean mean-diffusivity in tract inferior longitudinal fasciculus (right) (PGS001753), weighted-mean intra-cellular volume fraction in left tract superior longitudinal fasciculus (PGS001682), and mean intra-cellular volume fraction in left anterior corona radiata on fractional anisotropy skeleton (PGS001450). Hearing aid use PRS (PGS000763) showed a suggestive association with SSQ12 (r =  − 0.20, p < 0.05), indicating that individuals with a high genetic predisposition to hearing aid use (e.g., aidable hearing loss) showed lower SSQ12 scores. No other PRS related to hearing health showed a significant association with SSQ12.

Table 2.

Top 15 PRS models associated with SSQ12 (N = 255)

PRS ID	Reported trait	Beta	p	Adj p	Adj R2
PGS002597	HDL cholesterol	0.42	1.82E-05	0.044*	0.08
PGS002548	HDL cholesterol	0.40	5.49E-05	0.044*	0.08
PGS002695	HDL cholesterol	0.39	6.88E-05	0.044*	0.07
PGS000845	HDL cholesterol	0.40	8.53E-05	0.044*	0.07
PGS000064	HDL cholesterol	0.39	0.0001	0.050	0.07
PGS002101	Apolipoprotein A	0.37	0.0001	0.063	0.07
PGS002329	HDL cholesterol	0.36	0.0003	0.11	0.06
PGS002172	HDL cholesterol	0.35	0.0003	0.11	0.06
PGS001888	Apolipoprotein A	0.34	0.0004	0.12	0.06
PGS000777	Parkinson’s disease dementia	− 0.33	0.0005	0.13	0.06
PGS000309	High-density lipoprotein	0.34	0.0006	0.14	0.06
PGS000695	Sodium in urine (mmol/L)	0.36	0.0008	0.17	0.06
PGS002401	HDL cholesterol	0.34	0.0012	0.22	0.05
PGS001586	Volume of grey matter in parahippocampal gyrus, posterior division (R)	0.33	0.0013	0.24	0.05
PGS002112	Recent feelings of depression	− 0.31	0.0020	0.33	0.05

Baseline model 1 (predictors: age, sex, ethnicity): adjusted R² = 0.018, p = 0.054 (age was a significant predictor)

Baseline model 2 (predictors: age, sex, ethnicity, PCA1-5): adjusted R² = 0.02, p = 0.09 (Age and PCA5 were significantly associated with SSQ12)

Fig. 1.

A scatter plot between SSQ12 and PGS002597 (trait: HDL level), showing a significant positive correlation with SSQ12 scores. A bar chart presenting average SSQ12 scores between low, mid, and high PRS groups. Participants with a high genetic predisposition to HDL levels had better SSQ12 scores

PRS Associated with QuickSIN

QuickSIN SNR loss was associated with sex and ethnicity. Males revealed significantly higher SNR loss than females (MD = 0.46, p < 0.05), and non-Europeans showed significantly higher SNR loss (MD = 1.29, p < 10⁻⁶). Genomic PCAs were significantly associated with SNR loss in the baseline model (p < 0.001). After controlling for the effects of these confounders, 119 PRS variables showed suggestive association with QuickSIN (unadjusted p < 0.05, Supplementary File S1). However, no PRS predictors achieved statistical significance after applying the FDR correction. The top results included PRS for lung cancer (PGS156), cylindrical power of eye movement (PGS000156), neutrophil count (PGS001358), QRS duration in cardiogram (PGS000182), and diastolic blood pressure (PGS002443) (Supplement file S1).

PRS Predictors of PTA_0.5–2

PTA_0.5–2 showed no association with age, sex, ethnicity, and genomic PCAs. 191 PRS variables achieved suggestive significance (unadjusted p < 0.05). However, no PRS predictors were significantly associated with PTA_0.5–2 after applying the FDR correction for the multiple comparisons. The top results included PRS related to diabetic retinopathy (PGS000862), weighted-mean L1 in the track left inferior longitudinal fasciculus (PGS001708), and myocardial infarction (PGS000710).

PRS Predictors of PTA_4-8

Table 3 presents the results of the association analysis for PTA_4-8. Dementia PRS (PGS002035 and PGS000929) revealed a significant association with PTA_4-8. Participants with high PRS (i.e., increased polygenic risk of dementia) revealed significantly elevated PTA_4-8 than their counterparts (Fig. 2). PRS of a family history of Alzheimer’s disease (AD)-related dementia (PGS001347) showed a significant association with PTA_4-8. 138 PRS revealed a significant association with PTA_4-8. The top results included ten PRS models of AD and dementia (Table 3). Atrial flutter PRS showed suggestive significance with PTA_4-8. Neuroimaging-related PRS, such as weighted-average L1 in tract superior longitudinal fasciculus (PGS001710), showed a suggestive association with PTA_4-8.

Table 3.

Top 15 PRS models associated with PTA_4-8 (N = 218)

PRS ID	Reported trait	Beta	p	Adj p	Adj R2
PGS001347	Family history of Alzheimer’s/dementia	2.15	3.55E-05	0.046*	0.05
PGS000929	All cause dementia (algorithmically-defined)	2.07	7.03E-05	0.046*	0.04
PGS002035	Dementias	2.08	7.07E-05	0.046*	0.04
PGS001349	Alzheimer’s disease (time-to-event)	1.98	0.0001	0.071	0.04
PGS001348	Alzheimer’s disease (algorithmically-defined)	1.94	0.0001	0.07	0.03
PGS001179	Vascular dementia (time-to-event)	1.93	0.0002	0.07	0.03
PGS001827	Dementias	1.90	0.0002	0.07	0.03
PGS000946	Unspecified dementia (time-to-event)	1.89	0.0002	0.07	0.03
PGS001828	Alzheimer’s disease	1.89	0.0002	0.07	0.03
PGS002289	Late-onset Alzheimer’s disease	1.93	0.0003	0.07	0.03
PGS000945	Dementia in Alzheimer's disease (time-to-event)	1.79	0.0004	0.10	0.02
PGS000779	Alzheimer’s disease	1.83	0.0009	0.18	0.02
PGS001710	WA L1 in tract superior longitudinal fasciculus (R)	− 1.86	0.001	0.21	0.02
PGS000334	Late-onset Alzheimer’s disease	1.71	0.001	0.21	0.02
PGS001263	Atrial flutter	1.87	0.002	0.37	0.01

Baseline model 1 (predictors: age, sex, ethnicity): adjusted R² =  − 0.005, p > 0.05 (no predictors achieved statistical significance)

Baseline model 2 (predictors: age, sex, ethnicity, PCA1-5): adjusted R² =  − 0.02, p > 0.05 (no predictors achieved statistical significance)

Fig. 2.

Average behavioral hearing thresholds as a function of test frequencies for low (< 33.3 percentile), mid (33.3–66.6 percentile), and high (> 66.6 percentile) PRS groups of family history of Alzheimer’s disease-related dementia. The error bar indicates ± 1 standard error

PRS Predictors of PTA_12-16

PTA_12-16 revealed a significant association with 7 PRS variables related to atrial fibrillation and flutter (Table 4). These results showed that participants with high PRS for atrial fibrillation and flutter showed higher PTA_12-16 (Fig. 3). Atrial fibrillation and flutter PRS variables showed high collinearity (Supplement file S2). One hundred thirty-four PRS variables achieved suggestive significance. PRS related to breast cancer (PGS002294 and PGS000050), cystatin levels (PGS000228 and PGS002165), and cathepsin L1 serum levels showed suggestive association with PTA_12-16 (Table 4).

Table 4.

Top 15 PRS models associated with PTA_12-16 (N = 218)

PRS name	Reported trait	Beta	p	Adj p	Adj R2
PGS001339	Atrial fibrillation and flutter (time-to-event)	5.01	1.8E-07	0.0003**	0.12
PGS001340	Atrial fibrillation	5.10	2.5E-07	0.0003**	0.12
PGS001263	Atrial flutter	4.96	3.9E-07	0.0003**	0.11
PGS002050	Atrial fibrillation and flutter	4.14	2.4E-05	0.015*	0.08
PGS000331	Atrial fibrillation	4.03	3.4E-05	0.017*	0.07
PGS001841	Atrial fibrillation and flutter	4.00	4.9E-05	0.021*	0.07
PGS000016	Atrial fibrillation	4.14	9.4E-05	0.035*	0.07
PGS002294	Breast cancer	3.20	0.001	0.32	0.05
PGS001947	Cystatin C	− 2.94	0.001	0.36	0.04
PGS000035	Atrial fibrillation	2.95	0.001	0.42	0.04
PGS000228	Cathepsin L1 (CTSL1) serum levels	2.82	0.001	0.42	0.04
PGS000338	Atrial fibrillation	2.85	0.002	0.44	0.04
PGS000050	Breast cancer	2.57	0.002	0.48	0.04
PGS001356	Atrial fibrillation	− 3.01	0.002	0.48	0.04
PGS002165	Cystatin C	− 2.84	0.002	0.48	0.04

Baseline model 1 (predictors: age, sex, ethnicity): adjusted R² = 0.01, p > 0.05 (no predictors achieved statistical significance)

Baseline model 2 (predictors: age, sex, ethnicity, PCA1-5): adjusted R² = 0.002, p > 0.05 (no predictors achieved statistical significance)

*p < 0.05, **p < 0.001

Fig. 3.

Average behavioral hearing thresholds as a function of test frequencies for low (< 33.3 percentile), mid (33.3–66.6 percentile), and high (> 66.6 percentile) PRS groups of atrial fibrillation and flutter. The error bar indicates ± 1 standard error

Association Analyses in a Subsample of Europeans

The results achieving statistical significance in the multi-ethnic sample of the present study were subjected to association analyses (with linear regression model described in the statistical analysis) in a subsample of Europeans. The results showed that the top results presented in Tables 2, 3, and 4 showed significant associations with their respective phenotypes in the subsample of Europeans (for details, Supplementary File S2), suggesting that the top results obtained are not likely inflated by potential population stratification due to including a multi-ethnic sample.

Association Analyses Between Hearing Trait-PRSs and Audiological Measures

We conduct the association analyses for hearing trait-PRSs and audiological traits (individual SSQ questions, SSQ12, QuickSIN, better PTA_0.5–2, PTA_4-8, and PTA_12-16, and poorer ear PTA_0.5–2, PTA_4-8, and PTA_12-16). PGS000763 (hearing aid use), PGS001252 (hearing difficulty and deafness), PGS001253 (hearing difficulty), and PGS001533 (tinnitus severity) revealed modest associations with SSQ traits. No hearing-trait PRSs revealed associations with hearing thresholds (PTAs).

Comparison Between SSQ12 and QuickSIN

The present study investigated self-reported SIN abilities (SSQ12) and SIN performance (QuickSIN) in healthy young adults. Figure 4 presents a scatter plot between SSQ12 and QuickSIN for participants reporting English as a native versus non-native language. QuickSIN and SSQ12 revealed a modest correlation (p < 0.05) in the subsample of participants reporting European ethnicity; all were native English speakers. However, the relationship was insignificant in the subsample of non-native English speakers. About 70% of non-native English speakers reported non-European ethnic backgrounds. QuickSIN SNR loss varied drastically between native and non-native English speakers (MD = 1.2, p < 10⁻⁶), but SSQ12 scores remained comparable between the groups (p > 0.05).

Fig. 4.

A scatter plot between SSQ12 and QuickSIN for native and non-native English speakers. SSQ12 showed a modest correlation with QuickSIN in the subsample of the native English speaker. Although the trend was similar, no significant correlation between these measures was found for non-native English speakers, possibly due to high inter-subject variability. Box whisker plots present group differences in SSQ12 and QuickSIN between native and non-native English speakers. Non-native English speakers showed significantly poorer QuickSIN, yet their SSQ12 scores were comparable to native English speakers

Discussion

The present study investigated the genetic epidemiology of SIN measures and hearing sensitivity in healthy young adults with self-reported normal hearing. The study used the PGS catalog to calculate PRS for 2620 variables covering a range of human traits and health conditions. The major findings of the study were as follows: [1] HDL PRS revealed a significant association with SSQ12, suggesting that young adults with a lower genetic predisposition to HDL experience poorer SIN perception; [2] atrial fibrillation and flutter PRS showed a significant association with the extended high-frequency hearing thresholds (PTA_12-16), showing that individuals with a higher genetic predisposition to atrial fibrillation and flutter exhibit elevated extended high-frequency thresholds; and [3] dementia PRS revealed a significant association with high-frequency hearing thresholds (PTA_4-8), indicating that healthy young adults with a genetic predisposition to dementia (with and without AD) could exhibit elevated high-frequency hearing thresholds. These results suggest that SIN perception and peripheral hearing measures in healthy young adults with self-reported normal hearing could be influenced by a genetic predisposition to reduced HDL levels, atrial fibrillation/flutter, and dementia.

Genetic Risk to Atrial Fibrillation and Extended High-Frequency Hearing Thresholds

Atrial fibrillation, a common form of cardiac arrhythmia, is a major risk factor for developing stroke, embolism, and cardiac arrest [36]. Older adults in high-income countries are at substantially higher risk of atrial fibrillation [37]. Atrial fibrillation in older adults is associated with sensory and cognitive comorbidities, including hearing impairment, cognitive decline, social isolation, anxiety, and depression [38]. The present study suggests that a reduction in hearing sensitivity at the extended high-frequency range in individuals with a high genetic predisposition to atrial fibrillation could start decades before a clinically meaningful hearing loss in the conventional frequency range (250–8000 Hz) is observed.

Emerging evidence suggests that older adults with atrial fibrillation and hearing loss exhibit accelerated cognitive decline [39]. Early audiological intervention could prevent or delay cognitive decline [40]. Knowledge about genetic predisposition to atrial fibrillation/flutter could help clinicians identify individuals at risk of hearing impairment decades before they acquire clinically significant hearing loss. Atrial fibrillation PRS could be used to initiate an early audiological intervention for preventing or delaying hearing loss-related cognitive decline in individuals carrying a genetic risk of atrial fibrillation.

Atrial fibrillation PRS (PGS001263, PGS001339, PGS001340) showed a suggestive association with PTA_4-8. The observed effect size (i.e., beta value) was substantially smaller at 4000–8000 Hz (e.g., PGS001263 beta = 1.87, p = 0.002, adj p > 0.05) than at 9000–16,000 Hz (e.g., PGS001263 beta = 4.96, p < 10⁻⁶, adj p = 0.0003). These results suggest that the effect of atrial fibrillation PRS was higher (i.e., more severe) at the extended high-frequency hearing thresholds. The finding is consistent with emerging literature indicating that extended high-frequency hearing thresholds are more sensitive for detecting early-stage auditory dysfunction than conventional audiometry [41]. Extended high-frequency assessment is not typically carried out in clinics e.g., [42], potentially due to limited clinical applicability, additional time commitment, and instrumentation necessary for the evaluation. Knowledge about genetic susceptibility could help clinicians identify candidate patients, such as those with high PRS for atrial fibrillation, for whom extended high-frequency assessment could be valuable.

Genetic Risk to Dementia and Hearing Loss

Healthy young adults with a genetic risk of dementia revealed significantly elevated PTA_4-8. The observed effect sizes for dementia PRSs (PGS001347, PGS000929, and PGS002035) were about 2 dB for PTA_4-8, which remained consistent throughout the extended high-frequency range (Supplementary File S1—see results for PTA_12-16). The p-values reached statistical significance (p < 0.05), but they could not remain significant after applying the FDR correction. The results were not significant at the lower frequencies (PTA_0.5–2). These results indicate that the genetic risk of dementia impacts hearing thresholds consistently at high and extended-high frequencies, but the effects are more clearly observed at high frequencies (4000–8000 Hz) than at the extended-high frequencies (12,000–16,000 Hz), potentially due to high inter-subject variability in hearing thresholds at the extended high frequencies (Fig. 2). A growing body of research suggests a link between dementia and hearing loss in older adults e.g., [43–46]. Individuals with AD-related dementia exhibit significantly higher thresholds than their counterparts, with the magnitude of the difference in hearing thresholds for adults with dementia increasing at higher audiometric frequencies [47]. A genetic study suggested that dementia and hearing loss do not share common genetic underpinnings, but they have common genetic vulnerabilities sharing molecular pathways [48]. Taken together, the present study suggests that a decline in auditory sensitivity starts at young ages for individuals with a high genetic predisposition to dementia, potentially due to shared molecular pathways between dementia and hearing loss.

Genetic Risk to HDL Levels and Self-reported Speech Perception Abilities

HDL PRS was inversely related to SSQ12, suggesting that young adults with a genetic predisposition to low levels of HDL struggle with SIN difficulties. The study obtained a suggestive association between Apolipoprotein-A PRS and SSQ12 (Table 2). Metabolic syndrome, which includes clinical symptoms such as hypertension, dyslipidemia, central obesity, and glucose intolerance, is known to increase the risk of adult-onset health conditions e.g., [49]. Past studies suggest that low HDL-C levels (< 40 mg/dL) exhibit the strongest association with hearing thresholds among other components of metabolic syndrome [50, 51]. While the exact molecular mechanisms underlying the relationship between HDL and hearing loss remain elusive, emerging evidence suggests that HDL could be important for preventing apoptotic cell death following oxidative stress and inflammatory events contributing to better hearing outcomes [52]. Early staged auditory dysfunction caused by a genetic predisposition to metabolic dysfunction in healthy young adults might not be reflected in puretone audiometry but could compromise suprathreshold auditory processing abilities. These observations are consistent with mounting evidence suggesting that speech perception measures are more sensitive for detecting early staged auditory dysfunctions than puretone audiometry e.g., [12, 53].

Association Between Hearing Trait-PRSs and Audiological Measures

We hypothesized that hearing trait-PRSs would reveal strong associations with audiological measures. Contrary to our hypothesis, we found that PGS000763 (hearing aid use) showed a modest relationship with SSQ12 (Table 5), and no other hearing trait-PRSs revealed significant associations with SSQ12, QuickSIN, and PTAs. A follow-up analysis investigating the effects of hearing trait-PRSs on single SSQ12 items and better and poorer ear PTAs showed modest associations between hearing trait-PRSs and SSQ12 traits (Supplementary File S3) but no significant influence on PTAs. Hearing-trait PRSs were derived from GWAS using older participants with self-reported hearing loss and hearing aid use e.g., [54]. Age-related decline in hearing thresholds could be observed around the 3rd decade of life e.g., [55]. Recent evidence suggests that suprathreshold auditory measures, such as auditory evoked potentials elicited with suprathreshold stimuli, could be more sensitive for detecting early-stage decline than audiometric hearing thresholds e.g., [12, 53, 56]. Future research is necessary to evaluate the influence of polygenic risk on suprathreshold auditory measures.

Table 5.

Results of the hearing trait PRSs on audiological measures

PRS name	Reported trait	SSQ12		QuickSIN		PTA0.5–2		PTA4-8		PTA12.5–16
		Beta	p-value	Beta	p-value	Beta	p-value	Beta	p-value	Beta	p-value
PGS000763	Hearing aid use	− 0.20	0.04*	0.03	0.73	− 0.57	0.15	0.03	0.95	0.67	0.43
PGS001252	Hearing difficulty and deafness	− 0.13	0.11	− 0.13	0.23	0.36	0.40	0.16	0.78	0.07	0.93
PGS001253	Hearing difficulty	− 0.13	0.11	− 0.15	0.20	0.35	0.42	0.13	0.83	− 0.14	0.88
PGS002104	Hearing difficulty	− 0.04	0.65	0.09	0.39	0.42	0.29	− 0.28	0.61	0.02	0.97
PGS000762	Hearing difficulty	− 0.003	0.96	− 0.06	0.58	0.28	0.46	0.96	0.08	− 1.30	0.15

*Unadjusted p < 0.05

Phenotyping SIN for Future Genetic Studies: Comparison Between SSQ12 and QuickSIN

The present study investigated self-reported SIN abilities (SSQ12) and SIN performance (QuickSIN) in healthy young adults. SSQ12 revealed significant correlation coefficients with PTAs, but QuickSIN showed no significant associations with PTAs (Table 1). Figure 4 presents a scatter plot between SSQ12 and QuickSIN for participants reporting English as a native versus non-native language. QuickSIN and SSQ12 revealed a modest correlation (p < 0.05) in the subsample of participants reporting European ethnicity; all were native English speakers. However, the relationship was not significant in the subsample of non-native English speakers. About 70% of non-native English speakers reported non-European ethnic backgrounds. QuickSIN SNR loss varied drastically between the groups (MD = 1.2, p < 10⁻⁶), consistent with a past study comparing QuickSIN performance in native and non-native English speakers [57]. In contrast, SSQ12 scores showed no significant group difference. SIN processing requires complex biological processes underlying audition (e.g., acoustic transduction, neural coding) and cognition (e.g., selective attention and working memory) e.g., [58, 59]. Linguistic, cognitive, and sociocultural factors influencing the design of SIN tests could influence inter-subject variability in test performance e.g., [60, 61]. A lack of a group difference in SSQ12 with a large difference in QuickSIN, significant relationships between SSQ12 and PTAs, and no significant relationships between QuickSIN and PTAs collectively suggest that speech perception abilities, evaluated with a self-reported rating of auditory experiences in a person’s environment, could be less influenced by cognitive, linguistic, and sociocultural confounders influencing standardized SIN tests. Taken together, the present study suggests that self-reported speech perception could help evaluate the genetic epidemiology of SIN deficits.

Summary

The present study evaluated the relationship between SIN perception and hearing thresholds with 2620 PRS variables in a sample of healthy young adults. A high polygenic risk of reduced HDL levels was associated with lower self-reported ratings of speech perception abilities. Individuals with a high polygenic risk of atrial fibrillation and flutter showed significantly elevated extended high-frequency hearing thresholds (9000–16,000 Hz). Individuals with a high polygenic risk of dementia showed elevated high-frequency hearing thresholds (4000–8000 Hz). These results indicate that individuals with polygenic risk of certain health conditions could exhibit a subclinical decline in hearing health measures at young ages, decades before a clinically meaningful hearing loss could be observed. Clinical knowledge about polygenic risk factors influencing hearing health could help clinicians identify susceptible individuals well before they acquire irreversible hearing loss and speech perception deficits, which could be used for promoting a healthy lifestyle and initiating prophylactic interventions for preventing (or reducing) the impact of hearing loss on cognition. Large-scale studies are necessary to investigate the genetic epidemiology of SIN perception.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

PRS

Polygenic risk score

GWAS

Genome-wide association study

SNP

Single nucleotide polymorphism

Funding

The study was funded by the National Institute on Deafness and Other Communication Disorders Grant R21DC016704-01A1.

Data Availability

The database will be available on NIH dbGaP after the completion of the project R21DC016704-01A1.

Declarations

Conflict of Interest

The authors declare no competing interests.