Genome‐wide association study and polygenic risk score analysis for hearing measures in children

An efficient auditory system contributes to cognitive and psychosocial development. A right ear advantage in hearing thresholds (HTs) has been described in adults and atypical patterns of left/right hearing threshold asymmetry (HTA) have been described for psychiatric and neurodevelopmental conditions. Previous genome‐wide association studies (GWASs) on HT have mainly been conducted in elderly participants whose hearing is more likely to be affected by external environmental factors. Here, we investigated HT and HTA in a children population cohort (ALSPAC, n = 6,743). Better hearing was associated with better cognitive performance and higher socioeconomic status. At the group level, HTA suggested a left ear advantage (mean = −0.28 dB) that was mainly driven by females. SNP heritability for HT and HTA was 0.13 and 0.02, respectively (n = 4,989). We found a modest negative genetic correlation between HT and reading ability. GWAS for HT (n = 5,344) did not yield significant hits but polygenic risk scores for higher educational attainment (EA, ß = −1,564.72, p = .008) and schizophrenia (ß = −241.14, p = .004) were associated with lower HT, that is, better hearing. In summary, we report new data supporting associations between hearing measures and cognitive abilities at the behavioral level. Genetic analysis suggests shared biological pathways between cognitive and sensory systems and provides evidence for a positive outcome of genetic risk for schizophrenia.

sensory hair cells of the inner ear, whose depolarization is initiated by deflection of mechano-sensitive hair bundles (Schwander, Kachar, & Müller, 2010). The auditory nerve transmits these signals to the cochlear nucleus in the brainstem. The majority of the input is transmitted to the contralateral superior olivary complex, while a minor part of the input is transmitted ipsilaterally (Felix, Gourévitch, & Portfors, 2018;Warren & Liberman, 1989). Greater contralateral medial olivonuclear suppression in the right compared to the left ear (Khalfa & Collet, 1996) has been suggested as the underlying correlate of a fundamental functional asymmetry between the left and right ear.
This hearing threshold asymmetry (HTA) has typically been reported as HT left-HT right so that positive values indicate an advantage of the right and negative values indicate an advantage of the left ear. In a study of more than 50,000 adults, Chung, Mason, Gannon, and Willson (1983) reported a right ear advantage (HTA between 1 and 4 dB) with more pronounced HTA in males than in females. In 1,191 children, a right ear advantage has been reported, albeit to a smaller extent than in adults (Eagles, 1973). Other authors found a general right ear advantage in males (n =~400, HTA between 0.1 and 0.5 dB) and a left ear advantage in females for specific frequencies (n =~400, HTA between À0.1 and À0.4 dB) (Kannan & Lipscomb, 1974). Smaller studies reported a general left ear advantage in children (Rahko & Karma, 1989;Roche, Siervogel, Himes, & Johnson, 1978).
An absence of HTA has been reported in schizophrenia (Gruzelier & Hammond, 1979;Mathew, Gruzelier, & Liddle, 1993) and ADHD (Combs, 2002). Moreover, symmetrical contralateral suppression in the olivary complex in the left and the right ear in schizophrenia  is in contrast with the right ear advantage typically found in controls. In children and adolescents, a right ear advantage has been reported in a sample of n = 22 with autism spectrum disorder (ASD) while no asymmetry was found in the control group (Khalfa et al., 2001). A developmental effect toward stronger HTA has been reported in controls that was absent in ASD children (n = 24) (James & Barry, 1983). Reduced laterality in processing auditory stimuli was reported in ASD and bipolar disorder (BIP) suggesting that HTA is linked to neurodevelopmental disorders (Reite et al., 2009;Schmidt, Rey, Oram Cardy, & Roberts, 2009).
However, older subjects have had more exposure to environmental factors, which might affect hearing, such as extensive noise (Stanbury, Rafferty, & Rosenman, 2008), medication (Ruhl, Cable, & Martell, 2014), chemicals (Morata, 2003), and medical conditions (Kakarlapudi, Sawyer, & Staecker, 2003). An investigation of HT in 250 monozygotic (MZ) and 307 dizygotic (DZ) twin pairs from 36 to 80 years of age suggests that environmental effects become more significant with age (Karlsson, Harris, & Svartengren, 1997). Despite this age effect, no study has ever investigated genetic factors involved in hearing function in children.
Here, we analyzed hearing measures in children from the Avon Longitudinal Study of Parents and Children (ALSPAC) (n = 6,743).
Consistent with previous studies, we found that better hearing is associated with enhanced cognitive skills and higher socioeconomic status (SES). We report the first GWAS for HT in children (n = 5,344).
In addition to single marker trait associations, we conducted genebased and gene set analysis and tested the effects of polygenic risk scores (PRS) for a range of neurodevelopmental disorders, IQ, and educational attainment (EA). Our results suggest that PRS for higher EA and schizophrenia are associated with better hearing.

| Cohort
ALSPAC is a longitudinal cohort representing the general population living in the Bristol area. Pregnant women resident in the county of Avon, United Kingdom, with expected dates of delivery from April 1, 1991, to December 31, 1992, were invited to take part in the study, resulting in 14,062 live births and 13,988 children who were alive at 1 year of age Fraser et al., 2013). From age 7, all children were invited annually for assessments on a wide range of physical, behavioral and neuropsychological traits. Informed written consent was obtained from the parents after receiving a complete description of the study at the time of enrolment into ALSPAC, with the option to withdraw at any time. Ethical approval for the present study was obtained from the ALSPAC Law and Ethics Committee and the Local Research Ethics Committees. The ALSPAC study website contains details of all the data that is available through a fully searchable data dictionary (http://www.bris.ac.uk/alspac/researchers/dataaccess/data-dictionary/).

| Phenotypes
Audiometry was performed according to British Society of Audiologists standards. Hearing tests were carried out in a room with minimal external noise. Testing was stopped if the background noise level exceeded 35 dBA. The air conduction threshold, that is, the lowest intensity in decibels at which a tone is perceived 50% of the time (dBHL, decibel hearing level), was tested using either a GSI 61 clinical audiometer or a Kamplex AD12 audiometer. Lower dBHL values indicate better hearing. For each ear, the air conduction threshold level was tested at 500 Hz, 1 kHz, 2 kHz, and 4 kHz. For each frequency, stimuli were first presented on the right and then on the left ear. The average threshold across different frequencies was derived for each ear.
HT was defined as the average air conduction threshold on the better ear. HTA was defined as the absolute difference in air conduc- Cognitive skills were assessed using tests for reading ability (Rust, Golombok, & Trickey, 1993), communication skills (Bishop, 1998), listening comprehension (Rust, 1996), short-term memory (Gathercole, Willis, Baddeley, & Emslie, 1994), total IQ, verbal IQ, performance IQ (Wechsler, Golombok, & Rust, 1991), and EA measured as capped General Certificate of Secondary Education (GCSE) scores (for detailed descriptions, see Appendix S1). Maternal highest educational qualification during pregnancy was used as a proxy for SES (Rashid et al., 2018).
Educational qualification was grouped into "CSE and no education," "Vocational," "O level," "A level," and "Degree." Children were assigned to neurodevelopmental and control subgroups as defined for the ALSPAC sample previously (Scerri et al., 2011).

| Genotype quality control (QC) and imputation
Genotypes were generated on the Illumina HumanHap550-quad array at the Wellcome Trust Sanger Institute, Cambridge, United Kingdom, and the Laboratory Corporation of America, Burlington, NC, United States. Standard QC was performed as described elsewhere (Brandler et al., 2013). Population stratification was assessed by multidimensional scaling analysis and compared with Hapmap II (release 22) European descent, Han Chinese, Japanese, and Yoruba reference populations. All individuals with non-European ancestry were removed. In total, 9,115 subjects and 500,527 SNPs passed QC filtering. Haplotypes were estimated using ShapeIT (v2.r644). QC-filtered autosomal SNPs were imputed using Impute v3 using the HRC 1.1 reference data panel. Poorly imputed SNPs (Info score < 0.8) and SNPs with low minor allele frequency (MAF < 0.05) were excluded from further analysis.

| SNP heritability
Genome-wide genotype data were available for 5,344 children with phenotypes (2,691 males, 2,653 females, mean age = 7.58 years, SD = 0.31 years). HT and HTA were inverse rank-transformed to achieve a normal distribution. We estimated SNP heritability (SNP h 2 ) using genome-based restricted maximum-likelihood (GREML) analysis implemented in GCTA (Genome-wide complex trait analysis) (Yang, Lee, Goddard, & Visscher, 2011), which compares phenotypic similarity and genotypic similarity based on a genetic-relationship matrix (GRM) (Yang et al., 2010). We created a GRM based on directly genotyped SNPs and including unrelated individuals (n = 4,989; identity by descent [IBD] < 0.05). SNP h 2 was estimated on the basis of this GRM using sex and the first two principal components as covariates.

| Genetic correlation analysis
We used bivariate GREML to estimate genetic correlations (r g ) between HT and the eight cognitive skills used for behavioral analysis based on the GRM described above (n = 4,989 unrelated children) and using sex and the first two principal components as covariates.

| GWAS
Association testing was performed using BOLT-LMM v2.3.4 (Loh et al., 2015) under the standard infinitesimal linear mixed model (LMM) framework specifying sex and the first two principal components as covariates. Since BOLT-LMM allows for the inclusion of related individuals (Loh, Kichaev, Gazal, Schoech, & Price, 2018), GWAS was run on the whole available sample with phenotypes and genotypes (n = 5,344). Overall, 5,305,352 SNPs that were either directly genotyped or imputed and passed QC were tested for association. The genomic inflation factor (λ) was calculated for all SNPs and revealed no evidence of population structure (HT: λ = 1.02, HTA: λ = 1.00).

| Gene-based and gene set analyses
SNPs were clumped based on LD (r 2 ≥ 0.1) within a 250 kb window. PRS were derived as the weighted sum of risk alleles based on odds ratios or beta values from the training GWAS summary statistics.
Sex and first two principal components were included as covariates.
Results are presented for the optimal training GWAS p-value threshold (explaining the highest proportion of phenotypic variance in HT) as well as GWAS p-value thresholds of .001, .05, .1, .2, .3, .4, .5, and 1 (all SNPs included). For optimal training GWAS p-value thresholds and number of SNPs included in the PRS, see Table S1. For six training GWAS, the Bonferroni-corrected significance level was set to .05/6 = .0083.
Data preparation and visualization was performed using R v.4.0.0.
While females showed a significant left ear advantage (t (3343) = À7.30, p = 3.50 Â 10 À13 ), there was no ear advantage in males Bivariate Pearson correlations revealed significant positive correlations among the different cognitive traits as previously reported (Scerri et al., 2011). There were significant negative correlations after conservative Bonferroni correction (45 comparisons, p < .0011) for all cognitive traits but listening comprehension with HT (Figure 2), indicating that lower HT (better hearing) is associated with better cognitive performance (correlation plots are shown in Figure S2). There was no association between HTA and cognitive traits.
Two-sample t-tests revealed no difference between children affected by neurodevelopmental disorders and sex-matched controls in HT (Table S2) or HTA (Table S3). However, there was a consistent pattern across neurodevelopmental subgroups with more negative HTA compared to the control group, indicating more leftward asymmetry.

| SNP heritability
In n = 4,989 unrelated children, SNP h 2 was 0.13 (SE = 0.07) for HT and 0.02 (SE = 0.06) for HTA. Because of the extremely low SNP h 2 for HTA, subsequent genetic analyses were only performed for HT.

| Genetic correlation
HT showed a trend toward negative genetic correlation with reading ability (r g = À.34, SE = 0.23, Figure 3), indicating a small degree of
Since there was no significant SNP h 2 for HTA, we did not perform genetic correlation or PRS analyses for HTA. However, since it might benefit future studies interested in HTA, we conducted an exploratory GWAS (Table S6; Figures S8-S11).

| PRS
PRS were tested for IQ, EA, and four neurodevelopmental conditions based on the behavioral correlations between HT and cognitive traits ( Figure 2). PRS for ADHD showed an association with HT (Table 1, Figure S12), indicating that higher genetic risk for ADHD is associated with higher HT, that is, worse hearing. In contrast, schizophrenia PRS showed a negative association with HT, suggesting that higher genetic risk for schizophrenia is associated with lower HT, that is, better hearing (Table 1, Figure S13). PRS for EA also reached significance for HT (  Figure S14). The negative association suggests that a genetic liability toward higher EA is associated with better hearing. We next tested whether the associations between ADHD and schizophrenia PRS with HT were mediated by cognitive skills and SES.
We thus reran the PRS analysis on HT with ADHD and schizophrenia as training GWAS and using sex, the first two principal components, total IQ, GCSE, and SES as covariates. The effect of ADHD PRS on HT disappeared after adjusting for total IQ, GCSE, and SES (Table 1, Figure S15), suggesting that the association is mediated by these variables. In contrast, the effect of schizophrenia PRS on HT remained similar after adjusting for total IQ, GCSE, and SES (Table 1, Figure S16), suggesting that the association is not mediated by cognitive skills and SES.

| DISCUSSION
We dissect the relationship between hearing measures and cognitive abilities and neurodevelopmental disorders both at the phenotypic and genetic level. We confirm that better hearing is associated with better performance on a range of cognitive tasks (Figure 2). We also report the results of the first GWAS on HT in children. Single marker, gene-based, and gene set enrichment analyses did not lead to any statistically significant results. However, PRS for EA and schizophrenia were significantly associated with HT (Table 1), with genetic liability for both traits associated with better hearing.
Phenotypic analysis for hearing measures showed sex-specific developmental effects for both HT and HTA. Our data (n = 6,743) show lower HT in boys compared to girls (Figure 1). The sex effect was very small and might only be detectable in reasonably sized samples, explaining why no sex effect was reported in a previous study of n = 1,869 children and adolescents (Park et al., 2016). In adults, the reverse effect has been reported with better hearing in females compared to males (McFadden, 1993). In our data, HTA revealed an overall left ear advantage with a lower air conduction threshold of 0.28 dB on average (Figure 1), replicating the results from smaller studies in children (Rahko & Karma, 1989;Roche et al., 1978). The effect was driven by females, suggesting that the sex effect on HTA reported in adults (i.e., stronger right ear advantage in males) seems to be established already in children. It is possible that a developmental shift toward the right ear in both sexes is driven by environmental factors, as continuous noise exposure (e.g., industrial noise) has been shown to affect the left ear more than the right (Nageris, Raveh, Zilberberg, & Attias, 2007). A possible limitation of the current study is that stimulus presentation was not randomized, but the right ear was always tested first. Testing the right ear first has been more common in previous studies than vice versa, which could either result in a learning effect (favoring the left ear) or in a fatigue effect (favoring the right ear) (Pirilä, Jounio-Ervasti, & Sorri, 1992). However, in adults, a right ear advantage is more common even in studies in which the order of stimulus presentation has been randomized (Pirilä et al., 1992), so the effect of stimulus presentation should be minimal.
F I G U R E 3 Results of genetic correlation (r g ) analysis between HT and cognitive traits Consistent with previous studies (Moore et al., 2020), we found that in the normal range of variation, HT is negatively associated with several cognitive skills on the behavioral level ( Figure 2). Although not significant, we found more negative HTA (i.e., more leftward asymmetry) in neurodevelopmental conditions including ASD compared to controls. Different types of asymmetry such as structural brain asymmetry (Postema et al., 2019), frontal alpha asymmetry (Gabard-Durnam, Tierney, Vogel-Farley, Tager-Flusberg, & Nelson, 2015), language processing (Herringshaw, Ammons, DeRamus, & Kana, 2016), and handedness (Markou, Ahtam, & Papadatou-Pastou, 2017) have been implicated in ASD. Therefore, it might be worth collecting HTA measures more systematically to further investigate the link between asymmetries and disorders.
Because there was no significant SNP h 2 for HTA, downstream genomic analyses were only performed for HT. Although no single marker trait associations reached significance in the GWAS ( Figure S4), the top marker on chromosome 15 (rs1039444) is located in an intron of RAB8B, which encodes for a GTPase that is expressed in inner and outer hair cells and is involved in autosomal recessive deafness (Heidrych et al., 2008). Targeted analysis for markers reported in previous GWAS for HT in adults replicated association with one marker, rs12955474, which is located in an intron of the CCBE1 gene (Fransen et al., 2015). Other markers in this gene have been associated with depression (Power et al., 2013) and left entorhinal cortex volume (Zhao et al., 2019). The sample size of our study is rather small for general GWAS standard, but of comparable size to GWAS conducted for hearing measures in adults previously. In fact, air conduction thresholds are not routinely collected in large-scale population studies. For example, the UK Biobank includes phenotypic information on hearing ability as self-reported hearing difficulty or use of hearing aids (n > 300,000) (Kalra et al., 2020;Wells et al., 2019).
Similarly, a GWAS on age-related hearing loss in the Genetic Epidemiology Research on Adult Health and Aging (GERA) cohort (n > 50,000) used a case control design, identifying cases based on health records (ICD-9 diagnosis) (Hoffmann et al., 2016). GWASs on quantitative measures of hearing ability were limited to smaller sample sizes below 6,000 subjects for individual samples (Nolan et al., 2013;Vuckovic et al., 2015). Systematic collection of air conduction thresholds in both ears in children would enable larger genetic studies to dissect the links between hearing, cognition and neurodevelopment.
Genetic correlation analysis ( Figure 3) revealed a tentative modest negative correlation between HT and reading ability. As lower HT indicate better hearing, this suggests an overlap of markers involved in better hearing and improved reading skills. In this analysis, standard errors were quite high, showcasing the need for follow up in larger datasets. We found no genetic correlation between verbal, performance or total IQ and HT. Similarly, PRS derived from large datasets did not identify an association between PRS for IQ (capturing mainly verbal and total IQ, Genç et al., 2021) and HT, suggesting that correlations we found on the behavioral level ( Figure 2) are not mediated by shared biological pathways. Instead, the analysis of PRS for ADHD and schizophrenia support a role of the genetic risk for these neurodevelopmental disorders contributing to HT. Hearing deficits in ADHD have been reported in terms of speech perception (Fuermaier et al., 2018), but not in air conduction thresholds. In our sample, there was no association between HT and ADHD on the behavioral level; however, this analysis was based on a too small sample of children meeting the criteria for ADHD (n = 21) to make conclusive results (Table S2). Moreover, the effect of ADHD PRS on HT disappeared after adjusting for cognitive skills and SES, suggesting that the effect was mediated by cognitive factors.
Higher PRS for schizophrenia were associated with better hearing after adjustment for cognitive skills and SES. Most PRS studies report shared genetic risk for different outcomes, for example, risk for one disorder tends to increase risk for other disorders (Andlauer et al., 2021) or poor cognitive performance (Gialluisi et al., 2019).
Instead, we found that PRS for schizophrenia were associated with better hearing. Similarly, PRS for schizophrenia have recently been associated with better language skills, but not overall school performance (Rajagopal et al., 2020). Therefore, our study expands the range of positive outcomes associated with risk of schizophrenia.
The correlations between better hearing and cognitive skills (including language measures) found on the behavioral level ( Figure 2) could be based on shared biological pathways which also increase the risk for schizophrenia. On the phenotypic level, previous research did not detect differences in HT between individuals affected by schizophrenia and controls in small samples (n = 87) (Prager & Jeste, 1993). In the future, this association might be worth being investigated more systematically in larger cohorts. In summary, our results highlight behavioral and genetic overlap between cognitive and sensory domains. We find that PRS for EA and schizophrenia are associated with HT, extending previous research reporting positive outcomes for schizophrenia PRS. Future studies should explore in more detail associations between sensory function and cognitive traits.

ACKNOWLEDGMENTS
The authors are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and labo-

CONFLICT OF INTEREST
The authors declare no competing financial interests in relation to the work described.

DATA AVAILABILITY STATEMENT
Data used for this submission will be made available on request to the ALSPAC Executive (alspac-exec@bristol.ac.uk). The ALSPAC data management plan (http://www.bristol.ac.uk/alspac/researchers/dataaccess/documents/alspac-data-management-plan.pdf) describes the data sharing policy, which is through a system of managed open access.