Abstract

   The relationship between hearing loss (HL) and Parkinson’s disease (PD)
   remains unclear. Using individual-level and summary-level data from the
   UK Biobank and the largest genome-wide association studies, we examined
   this link through observational, Mendelian randomization and genetic
   pleiotropy analyses. Among 158,229 participants, PD risk rose with HL
   severity especially in elder and males, and hearing aids significantly
   reduced PD risk in males. Although our results did not support a causal
   association, genetic correlation analysis suggested a localized genetic
   overlap (17q21.31). We identified 1545 SNPs and 63 genes with
   pleiotropic effects on HL and PD, including 79 novel SNPs across 6
   loci, with 3 showing strong co-localization. These loci were enriched
   in key tissues like brain, heart, liver and pancreas, linked to the
   dihydrolipoyl dehydrogenase complex pathway, and targeted by drugs such
   as Warfarin and Phenprocoumon. Overall, this study reveals the risk
   association, genetic basis, and pleiotropic loci connecting HL and PD.

   Subject terms: Risk factors, Parkinson's disease

Introduction

   Parkinson’s disease (PD) is an age-related neurodegenerative disorder,
   which affects more than 6 million people worldwide, and its incidence
   rate continues to increase with the aging population^[44]1, imposing a
   substantial burden on patients and caregivers. In parallel, hearing
   loss (HL) affects more than 400 million people globally; and even mild
   HL leads to considerable difficulties in communication, cognition,
   social isolation, and depression^[45]2. Although PD and HL are
   seemingly distinct diseases, recent studies imply they likely share
   common pathological basis. For instance, the primary pathological
   features of PD include the degeneration of dopaminergic neurons^[46]3
   and the abnormal accumulation of α-synuclein^[47]4. In the cochlea,
   differential expression of α-synuclein may contribute to efferent nerve
   degeneration, potentially being a cause of early-onset
   presbycusis^[48]5. In advanced stages of PD or cases of PD with
   dementia, co-pathologies such as β-amyloid (Aβ)^[49]6 and
   microtubule-associated protein (tau)^[50]7 can also be observed. In
   mouse models where Aβ or Aβ-related molecules were overexpressed in
   cochlear hair cells, mice displayed auditory deficits, particularly in
   high-frequency sound perception, along with evidence of Aβ and tau
   synergistically contributing to hearing damage and hair cell
   loss^[51]8.

   Epidemiological studies have evaluated the link between HL and PD, but
   obtained inconsistent findings. For example, a nested case-control
   study (N[PD] = 1055) found that HL significantly increased the risk of
   developing PD by 66% (6–158%) within 2 years before diagnosis and 73%
   (16–157%) in the 2–5 years prior, but not more than 5 years before PD
   diagnosis^[52]9; an elevated risk of PD by 53% (17–99%) due to HL was
   also discovered by another study (N[PD] = 4976)^[53]10. Conversely,
   recent work suggested that self-reported HL was not a risk factor of PD
   over a decade-long follow-up^[54]11.

   These conflicting results suggest current epidemiological studies are
   still insufficient to explain the relation between HL and PD. Reverse
   correlation, information bias and unknown confounding in conventional
   observational studies make it difficult to interpret these results as
   causality, and residual confounding in prospective observational
   studies also challenge the validity of findings when interpreting them
   as causal. We aimed to revisit the association of HL and PD from a
   Mendelian randomization (MR) perspective. Rounded in its foundational
   trio of assumptions (Fig. [55]S1), MR is a robust approach that can
   mitigate confounding bias by leveraging single nucleotide polymorphisms
   (SNPs) as instrumental variables (IVs) to deduce causal
   relation^[56]12. A recent MR study found various types of HL, including
   self-reported HL, conductive HL, sudden idiopathic HL, mixed conductive
   and sensorineural HL, and sensorineural HL, did not causally influence
   PD^[57]13. Note that, in two-sample MR a linear exposure-outcome
   relation is assumed; however, such an assumption might not be satisfied
   for the link between HL and PD; thus non-linear MR methods should be
   conducted^[58]14. A well-designed MR analysis would be of unprecedented
   significance in addressing the limitations of traditional
   epidemiological studies when exploring the causal connection between HL
   and PD.

   The lack of causal evidence also implies the observed relation may
   involve biological mechanisms such as genetic pleiotropy^[59]15.
   Indeed, recent genome-wide association studies (GWAS) provide an
   in-depth understanding of genetic landscape underlying both HL^[60]16
   and PD^[61]17. There exists fragmented evidence supporting genetic
   overlap between the two traits; for instance, MAPT, an well-known PD
   susceptibility gene acting via tau protein aggregation pathways^[62]18
   by encoding the microtubule-associated protein tau primarily expressed
   in neurons^[63]19, was also identified to be significantly associated
   with speech-in-noise perception deficit, a primary complaint of
   individuals with audiometric HL^[64]20. Unfortunately, to our
   knowledge, no formal and large-scale cross-phenotype genetic analyses
   have been conducted to systematically investigate whether HL and PD
   share common genetic architecture and to examine the extent and nature
   of such sharing.

   Overall, we here attempted to construct a robust phenotypic and genetic
   connection between HL and PD under a relatively complete framework for
   the triangulation of evidence from the prospective cohort study,
   MR-based causal inference and pleiotropy analyses. To this objective,
   we first utilized the UK Biobank (UKB) cohort to explore the
   epidemiological association between HL and PD^[65]21. We also
   investigated whether the use of hearing aids could protect against the
   development of PD. Subsequently, we employed genome-wide
   cross-phenotype analysis to comprehensively characterize shared genetic
   architecture and causal relation. Our analyses included causal
   inference through bidirectional MR analysis, global and local genetic
   correlation analysis, determination of shared loci through
   cross-phenotype meta-analysis, identification of shared genes and
   tissues through transcriptome-wide association studies (TWAS), and a
   series of functional analyses to elucidate the biological mechanisms
   underlying HL and PD. The design overview of this study is outlined in
   Fig. [66]1.

Fig. 1. Study design overview for investigating the relationship between
hearing loss and Parkinson’s disease.

   [67]Fig. 1
   [68]Open in a new tab

   Insufficient HL, −5.5 dB ≤ SRT ≤ -3.5 dB; Poor HL, SRT > −3.5 dB; HL,
   hearing loss; PD, Parkinson’s disease; LDSC, linkage disequilibrium
   score regression; IVW, inverse-variance weighted; GPA, Genetic analysis
   incorporating pleiotropy and annotation; CPASSOC, cross-phenotype
   association analysis; TWAS, Transcriptome-wide association study;
   PLACO, Pleiotropic analysis under composite null hypothesis.

Results

Population characteristics

   The used UKB cohort included 158,229 individuals with available speech
   reception threshold (SRT) data to define hearing status. Among them,
   61,484 (38.9%) suffered from HL (SRT ≥ −5.5 dB), including 51,826
   (32.8%) insufficient HL (−5.5 dB ≤ SRT ≤ −3.5 dB) and 9658 (6.1%) poor
   HL (SRT > −3.5 dB). The mean age of participants at baseline was 56.6
   (±8.2) years, 45.5% were females, and 92.1% were of white ethnicity.
   During a median follow-up time of 13.4 years (interquartile range
   [IQR], 12.8–14.2 years), 900 (0.6%) incident PD cases were reported.

   The prevalence of PD was higher among participants with HL compared to
   those without HL, with a prevalent rate of 4.7‰, 6.8‰ or 10.0‰ for
   those without HL, insufficient HL, or poor HL, respectively. Among
   hearing aid users, 41.1% encountered insufficient HL and 26.5% suffered
   from poor HL. Further, HL and PD usually occurred in the elderly, or
   among participants with a history of drinking, diabetes, cardiovascular
   diseases, and family history of PD (Table [69]S1).

Relationships between HL and PD in observational analyses

   Cox proportional hazards (PH) models were used to assess the
   association between HL and the risk of incident PD, with hazard ratio
   (HR) and the 95% confidence intervals (CIs) reported. To enhance
   interpretability, our results were mainly expressed as the relative
   increase in risk. We observed that HL could lead to an increased risk
   of developing PD. Briefly, compared to individuals without HL, those
   with poor HL significantly encountered a 29.5% (2.9–62.8%,
   P = 2.73 × 10^−2) greater risk of occurring PD (Table [70]1). Those
   with insufficient HL suffered from an 8.2% greater risk of occurring PD
   although such an increase was not significant (P = 0.28). We did not
   identify a statistically significant protection against the occurrence
   of PD through the use of hearing aids although a 18.2% lower risk was
   seen (P = 0.22).

Table 1.

   Cox analysis revealed the associations between the severity of hearing
   loss and Parkinson’s disease
   Hearing loss Major analysis Sensitivity analyses
   Main model (N[case] = 899) Model 1 (N[case] = 872) Model 2
   (N[case] = 849)
   Insufficient HL 1.082 (0.938–1.247), 0.28 1.086 (0.941–1.254), 0.26
   1.139 (0.985–1.316), 0.08
   Poor HL 1.295 (1.029–1.628), 2.73 × 10^−2 1.303 (1.033–1.643),
   2.54 × 10^−2 1.360 (1.070–1.729), 1.19 × 10^−2
   Hearing-aid 0.818 (0.592–1.130), 0.22 0.805 (0.581–1.117), 0.19 0.742
   (0.528–1.043), 0.09
   [71]Open in a new tab

   Note: The severity of hearing loss was determined based on the worst
   ear, with the results presented via HR (95% CI) and P-value.
   Insufficient HL, −5.5 dB ≤ SRT ≤ −3.5 dB; Poor HL, SRT > −3.5 dB; Model
   1, only cases of Parkinson’s disease aged 50 years or older at baseline
   were included; Model 2, only white participants were included; HL,
   Hearing loss; HR, hazard ratio; CI, confidence interval.

   Compared to participants without HL, those with bilateral HL
   encountered a 26.1% (4.3–52.4%, P = 1.68 × 10^−2) greater risk of PD.
   Those with unilateral HL had a slightly elevated risk (5.9%) but this
   increase was nonsignificant (P = 0.45). Again, we did not detect
   significant evidence supporting the protection of hearing aids against
   the development of PD (P = 0.24) (Table [72]S2). Similar association
   patterns were observed when only analyzing PD cases aged 50 years or
   older at baseline or individuals of white ethnicity (Table [73]1).
   After adjusting for aging, both insufficient HL and poor HL were
   significantly related to an increased risk of PD, but wearing hearing
   aids remained not to affect PD (Table [74]S3).

   Among participants with age at onset exceeding 65 years, insufficient
   HL and poor HL significantly increased the risk of PD by 48.8%
   (27.5–73.6%, P = 4.20 × 10^−7) and 111.6% (65.4–170.7%,
   P = 2.48 × 10^−9), respectively. In males, insufficient HL and poor HL
   substantially increased the risk of PD by 23.7% (3.8–47.5%,
   P = 1.72 × 10^−2) and 41.5% (7.0–87.2%, P = 1.48 × 10^−2),
   respectively. Particularly, we discovered wearing hearing aids
   significantly reduced the risk of PD by 35.3% (2.7–57.0%,
   P = 3.64 × 10^−2) among males (Table [75]2).

Table 2.

   Stratified analyses revealed the associations between the severity of
   hearing loss and Parkinson’s disease
   Hearing loss Stratified by age at onset Stratified by sex
   Onset age≥65 (N[case] = 747) Onset age<65 (N[case] = 152) Male
   (N[case] = 587) Female (N[case] = 312)
   Insufficient HL 1.488 (1.275–1.736), 4.20 × 10^−7 1.110 (0.785–1.571),
   0.55 1.237 (1.038–1.475), 1.72 × 10^−2 0.840 (0.658–1.074), 0.16
   Poor HL 2.116 (1.654–2.707), 2.48 × 10^−9 1.425 (0.782–2.598), 0.24
   1.415 (1.070–1.872), 1.48 × 10^−2 1.110 (0.742–1.662), 0.60
   Hearing-aid 1.056 (0.283–2.128), 0.75 0.776 (0.749–2.598), 0.62 0.647
   (0.430–0.973), 3.64 × 10^−2 1.387 (0.817–2.355), 0.22
   [76]Open in a new tab

   Note: The stratified analyses were conducted based on the age at onset
   of Parkinson’s disease and sex. The results were presented via HR (95%
   CI) and P-value. Insufficient HL, −5.5 dB ≤ SRT ≤ -3.5 dB; Poor HL,
   SRT > −3.5 dB. HL hearing loss; HR hazard ratio; CI confidence
   interval.

Bidirectional Mendelian randomization analysis

   We here assessed whether the observed relation between HL and PD was
   driven by causality by conducting bidirectional MR analyses. All IVs of
   HL had the F statistics exceeding 10, indicating they were sufficiently
   strong (Table [77]S4). However, the inverse-variance weighted (IVW)
   analysis suggested a lack of compelling evidence for the causal
   relation between genetically determined HL and the risk of PD
   (OR = 0.98, 95%CI 0.80–1.20, P = 0.85). Similar null associations were
   identified by the maximum likelihood method (P = 0.84), the weighted
   median method (P = 0.80), or the MR-Egger regression (P = 0.86). The
   null result seemed not to be due to the nonlinear relation between HL
   and PD (P[non-linearity] = 0.27) and we also did not find heterogeneity
   among IVs (Cochran’s Q test, P > 0.05) (Fig. [78]S2). The leave-one-out
   analysis demonstrated the stability of the null association, and both
   the MR-Egger intercept test and MR-PRESSO indicated the absence of
   horizontal pleiotropy (P > 0.05). Further, the reverse MR analysis did
   not discover any evidence supporting the causal effect of PD on HL
   (Table [79]S5).

Global and local genetic correlation

   In the linkage disequilibrium (LD) score regression (LDSC) analysis, no
   significant global genetic correlation was detected between HL and PD (
   [MATH: <msub><mrow><mover
   accent="true"><mrow><mi>r</mi></mrow><mrow><mo>^</mo></mrow></mover></m
   row><mrow><mi>g</mi></mrow></msub> :MATH]
    = −0.019, P = 0.57). In the ρ-HESS analysis, a total of 1693 regions
   were retained for calculating local genetic correlation between HL and
   PD. Although none of them were significant after corrected via
   Bonferroni’s method (P > 0.05/1693), one region (Chr 17:
   43056905-45876022) located at 17q21.31 displayed a suggestively
   significant local genetic correlation (genetic correlation = −0.77,
   P = 2.18 × 10^−2) (Fig. [80]S3). Moreover, we found that 865 local
   regions showed positive genetic correlations with an average of 0.519
   and 828 displayed negative genetic correlations with an average of
   −0.518. Both positive and negative local genetic correlations nearly
   distributed equally across the whole genome and would cancel each other
   out when averaging them, which offered an explanation why the global
   genetic correlations were so small and non-significant.

SNP-level genetic pleiotropy analyses

   The genetic analysis incorporating pleiotropy and annotation (GPA)
   analysis suggested that the SNP-level pleiotropy proportion was 1.6%
   (P = 1.20 × 10^−13) between HL and PD. Using cross-phenotype
   association analysis (CPASSOC), we identified a total of 1,545
   pleiotropic SNPs for HL-PD (Table [81]S6), including 79 likely novel
   SNPs. These shared SNPs can be mapped to 10 independently pleiotropic
   loci (Fig. [82]2), including 6 novel loci (1q44, 1p13.3, 6q23.3,
   8p21.3, 9q21.13, and 16p11.2) (Table [83]3).

Fig. 2. Manhattan plot displays the negative logarithm (base 10) of
P[CPASSOC].

   [84]Fig. 2
   [85]Open in a new tab

   The black threshold line is set at 5 × 10^−8, with red dots indicating
   the SNPs which have the minimum P-value in the identified loci.

Table 3.

   Identifying independent loci from pleiotropic SNPs related to HL and PD
   Loci CPASSOC Fine-mapping Colocalization Novel
   Total Index SNP (P[CPASSOC]) N.set (Top SNP) Accuracy gain (%) PP4
   (Best casual)
   1q44 45 rs10927035 (9.40 × 10^−12) 16 (rs10927035) 64 0.91 (rs10927035)
   Yes
   1q32.1 4 rs823118 (2.77 × 10^−9) 4 (rs823118) 0 0.89 (rs823116) No
   1p13.3 2 rs12741221 (2.94 × 10^−9) 2 (rs12741221) 0 0.20 (rs12741221)
   Yes
   6q23.2 1 rs667955 (1.87 × 10^−14) 1 (rs667955) 0 0.99 (rs667955) No
   6q23.3 3 rs6934436 (8.44 × 10^−10) 3 (rs6934436) 0 0.79 (rs6934436) Yes
   8p21.3 1 rs73543600 (2.69 × 10^−8) 1 (rs73543600) 0 0.04 (rs73543600)
   Yes
   9q21.13 6 rs12001698 (2.21 × 10^−8) 6 (rs12001698) 0 0.21 (rs11143667)
   Yes
   14q24.3 39 rs2287407 (4.96 × 10^−9) 30 (rs2287407) 23 0.73 (rs2287407)
   No
   16p11.2 22 rs73524556 (3.37 × 10^−9) 12 (rs1870293) 45 0.86
   (rs12924903) Yes
   17q21.31 1409 rs7220839 (1.28 × 10^−13) 3 (rs7220839) 99.8 0.27
   (rs58879558) No
   [86]Open in a new tab

   Note: “Loci” denoted the identified independent loci; “Total”
   represented the number of pleiotropic SNPs in the each identified
   locus; “Index SNP” referred to the SNP with the minimum P-value in the
   identified loci; “N.set” was the number of SNPs in the 90% credible set
   identified by fine-mapping analysis; “Top SNP” designated the SNP with
   the highest posterior probability within the credible set; “Accuracy
   gain” was calculated by (Total-N.set)/Total×100%; “PP4” was the
   posterior probability of colocalization; “Best casual” designated the
   SNP with the highest posterior probability within the loci; “Novel”
   represented the newly discovered pleiotropic loci between hearing loss
   and Parkinson’s disease.

Fine-mapping and colocalization analyses

   By performing the fine-mapping analysis, we successfully narrowed the
   set of candidate causal SNPs down for four loci, including 1q44 with an
   accuracy gain of 64%, 14q24.3 with an accuracy gain of 23%, 16p11.2
   with an accuracy gain of 45%, and particularly 17q21.31 with an
   accuracy gain of 99.8%. The colocalization analysis indicated strong
   evidence of colocalization at 6 loci (posterior probability
   [PP4] > 0.7), including 1q44, 1q31.2, 6q23.2, 6q23.3, 14q24.3, and
   16p11.2 (Table [87]3). More information on fine-mapping and
   colocalization can be found in Table [88]S7 and Figure [89]S4.

Gene-level pleiotropy analyses

   The average gene-level pleiotropy proportion between HL and PD was
   12.5%, including a substantial pleiotropic tissue overlap in brain
   spinal cord cervical BA24 (21.9%) (Fig. [90]3). By performing TWAS and
   pleiotropic analysis under composite null hypothesis (PLACO), we
   identified 25 unique pleiotropic genes shared by HL and PD across
   various tissues (Table [91]S8). By integrating FUMA-mapped genes in
   terms of the pleiotropic SNPs identified by CPASSOC (Table [92]S9) and
   these pleiotropic genes identified by PLACO, we ultimately obtained 63
   unique pleiotropic genes (Table [93]S10).

Fig. 3. Heatmap about the proportion of pleiotropy calculated by the GPA
method.

   [94]Fig. 3
   [95]Open in a new tab

   The heatmap displaying the proportion of pleiotropy calculated by the
   GPA method, covering tissues associated with hearing loss and
   Parkinson’s disease. HL, hearing loss; PD, Parkinson’s disease. The
   size of the square and the color gradient from blue to red indicate the
   proportion of pleiotropy ranging from low to high.

Functional enrichment analyses

   We applied all identified pleiotropic genes to subsequent functional
   enrichment analyses. In the tissue-specific enrichment analysis, we
   observed significant functional enrichments in brain (e.g., putamen
   basal ganglia, cerebellum, hypothalamus, amygdala, cerebellar
   hemisphere, anterior cingulate cortex BA24), heart related tissue
   (heart atrial appendage), liver and pancreas (Fig. [96]4). Note that,
   some brain tissues also displayed a higher gene-level pleiotropy
   proportion as shown above. In the pathway analysis, these pleiotropic
   genes were associated with the dihydrolipoyl dehydrogenase complex
   (adjusted P = 0.013) of cellular component (Table [97]S11). In the drug
   targets analysis, noteworthy connections were established with Warfarin
   (adjusted P = 4.39 × 10^−4) and Phenprocoumon (adjusted
   P = 9.13 × 10^−3) (Table [98]S12).

Fig. 4. Tissue-specific enrichment of identified pleiotropic genes between
hearing loss and Parkinson’s disease.

   [99]Fig. 4
   [100]Open in a new tab

   The x-axis represents 54 tissues derived from GTEx v8, and the y-axis
   indicates the -log10 P-value of enrichment. Red bars denote tissues
   with significant gene enrichment, while blue bars represent tissues
   without significant enrichment.

Discussion

   In this study, we have profoundly investigated the connection between
   HL and PD through observational, causal inference and genetic
   pleiotropy analyses. The observational analysis suggested that,
   compared to no HL, poor HL significantly increased the risk of PD
   especially among individuals aged over 65 and males. Interestingly,
   wearing hearing aids only exerted a significantly protective effect
   against PD among males. But our MR analysis showed the relation between
   HL and PD might be not mediated by vertical causality. Further, despite
   the absence of global genetic correlation, a suggestively significant
   local genetic correlation was discovered in the genomic region located
   at 17q21.31. Importantly, the genetic pleiotropy analysis identified
   many SNPs, genes and loci with pleiotropic impacts, which set up the
   basis for the observed association between HL and PD.

   Based on the UKB cohort, we demonstrated that HL was an early non-motor
   risk factor of PD and that hearing aids conferred a protective effect
   against PD in males. One of our key findings was that, compared to no
   HL, poor HL significantly increased the risk of PD by 29.5%, whereas
   insufficient HL did not, which suggests that only when HL reaches a
   certain degree does it significantly elevate the risk of PD. Similar
   results were previously discovered for AD^[101]22,[102]23; but to our
   knowledge, this is the first report of a dose-response relation between
   HL and PD. We validated the stability of our results through a series
   of supplementary and sensitivity analyses; for example, bilateral HL
   enhanced the PD risk by 26.1%, whereas unilateral HL did not. Across
   different existing studies, we found considerable variation in the
   impact of HL on PD; for instance, a large representative case-control
   study from Germany (N[PD] = 138,345 and N[control] = 276,690) indicated
   that HL increased PD risk by 14%, with HL occurring more than 5 years
   before PD diagnosis^[103]24; another study from Taiwan demonstrated a
   higher prevalence of PD among individuals with HL (53% increased
   risk)^[104]10; both studies were consistent with our results. However,
   these aforementioned studies did not differentiate the severity of HL,
   which may overestimate the impact of mild HL on PD and weakened the
   effect of moderate to severe HL. Additionally, a recent study indicated
   the absence of association between HL and PD^[105]11, possibly due to
   smaller PD cases (N[PD] = 151) and inaccurately self-reported HL. In
   stratified analyses, HL significantly impacted late-onset PD (onset age
   ≥65), where insufficient HL and poor HL significantly increased the
   risk of PD by up to 48.8% and 111.6%, respectively, while not affecting
   early-onset PD (onset age <65), in line with a previous study which
   also found that PD patients with HL were more likely to be older (age
   ≥60) and males^[106]25. In HL patients, cognitive decline worsens over
   time, heightening the risk of PD and dementia^[107]26,[108]27. Aging
   contributes to neurodegeneration and the accumulation of α-synuclein
   aggregates^[109]28–[110]30, which may be related to the auditory
   dysfunction associated with PD^[111]31. This could explain the closer
   association between HL and late-onset PD. We also discovered that HL
   significantly increased the risk of PD occurrence among males, where
   insufficient HL and poor HL significantly raised the risk of PD by
   23.7% and 41.5%, respectively, but had no effect on females. A study
   also reported a higher proportion of males among PD patients with
   HL^[112]25. It is important to emphasize that both HL and PD are more
   prevalent in males, a gender difference that may be attributed to
   occupational exposures such as greater exposure to inorganic
   dust^[113]32, pesticides^[114]33, and engagement in agriculture and
   horticulture^[115]34. In addition, estrogen in females displayed a dual
   protective effect against HL and PD^[116]35–[117]37. Moreover,
   considering that the significant role of aging in the progression of HL
   and PD, we thus attempted to adjust for aging and discovered that HL
   could still cause an increased risk of PD, which indicated the impact
   of HL on PD risk was independent of aging, further validating the
   robustness of our results.

   Another novel finding of our work was that wearing hearing aids
   significantly reduced PD risk by 35.3% among males. A wealth of
   evidence suggested that early intervention for HL contributed to
   reducing the risk of dementia, possibly due to improvements in
   cognitive function^[118]23,[119]38. To our knowledge, our research is
   the first time to reveal that hearing aids could protect against PD
   occurrence, and this protection exhibited gender specificity. It has
   been reported that cognitive impairment and the improvement of
   cognitive function through hearing aids use only occurred in males
   after HL^[120]39, and males appeared to be also more accepting of
   hearing aids^[121]40,[122]41.

   However, MR analysis did not detect a significant causal connection
   between HL and PD, differing from the observational findings. This
   discrepancy may be due to MR examining HL’s lifelong impact on PD,
   unlike observational studies that focus on specific stages^[123]42.
   Further, current GWAS studies have not classified HL by severity or
   distinguished between early or late-onset PD or genders, possibly
   contributing to the inconsistency. Actually, a recent MR study also did
   not find a link between HL and PD, even with different HL
   subtypes^[124]13. Compared to that study, besides exploring the
   possible nonlinear relation, we further ruled out reverse causation
   through bidirectional two-sample MR and provided a detailed STROBE-MR.
   A similarly null relation was also observed between HL and
   AD^[125]43–[126]45, although observational studies suggested the
   presence of an association between HL and subsequent
   AD^[127]46–[128]49. Notably, recent research has found that pleiotropic
   genetic loci may establish a connection between HL and AD, such as
   rs2657879 in GLS2, which affected various traits such as glutamate
   levels, leading to HL and AD^[129]43. This highlights the importance of
   considering pleiotropic effects in understanding the complex relation
   between HL and neurodegenerative diseases including PD.

   Although we did not find significant global genetic correlations
   between HL and PD, we identified a potentially significant local
   genetic correlation in 17q21.31 (Chr 17: 43056905-45876022), which
   contains MAPT. The MAPT H1 haplotype was significantly associated with
   PD, and MAPT genotype had a significant effect on PD progression to
   dementia^[130]50. Furthermore, MAPT encodes tau protein, whose
   expression in cochlear hair cells could exacerbate hearing defects in
   mice^[131]8.

   This study identified 1545 pleiotropic SNPs and 63 genes shared between
   HL and PD, among which 79 SNPs were likely novel. These SNPs were
   ultimately localized to 6 loci, with 3 showing strong evidence of
   co-localization: 1q44 (PP4 = 0.91), 1p13.3, 6q23.3 (PP4 = 0.79),
   8p21.3, 9q21.13, and 16p11.2 (PP4 = 0.86). The 1q44 locus includes a
   significant pleiotropic gene, AKT3. AKT is a serine/threonine kinase
   and is one of the key downstream mediators of the phosphoinositide
   3-kinase (PI3K) signaling pathway. Evidence suggests that all AKT
   isoforms, including AKT3, are expressed in the cochlea of mice.
   Untreated AKT1 and AKT2/AKT3 double knockout mice exhibit significant
   HL, indicating these isoforms play a role in maintaining normal
   hearing^[132]51. The PI3K-AKT pathway crucially regulates cell
   survival, metabolism, and protein balance. A study has shown that AKT
   activation reduced in the substantia nigra of PD patients compared to
   non-PD samples, and constant AKT activation could protect dopaminergic
   neurons in preclinical PD models^[133]52. AMIGO1, located at the 1p13.3
   locus, is one of AMIGO family members. It is mainly expressed in the
   nervous system and has been demonstrated to enhance axon bundling and
   neurite outgrowth^[134]53. Axon bundling, referring to the process by
   which neuronal axons gather to form bundled structures, is vital for
   axon transport and the transmission of nerve signals. Disrupted axonal
   transport worsens neurotransmitter dysfunction and dopaminergic neuron
   loss, aggravating motor symptoms like tremors, rigidity, and slow
   movement in PD^[135]54. In the auditory system, axon bundling is
   crucial for linking spiral ganglion neurons with cochlear hair cells,
   facilitating efficient auditory signal transmission and guiding nerve
   fibers via repulsive signaling^[136]55. Given the key role of AMIGO1 in
   axon bundling, it may exhibit a pleiotropic impact on both HL and PD by
   maintaining the function of neural connections. The locus 6q23.3 has
   been discovered to be associated with various tumors, including
   Hodgkin’s lymphoma^[137]56, ocular adnexal marginal zone B cell
   lymphoma^[138]57, invasive epithelial ovarian cancer^[139]58.
   Additionally, the locus 6q23.3 interacted with schizophrenia
   loci^[140]59, and psychiatric disorders were common non-motor symptoms
   in PD^[141]60. Similarly, the locus 8p21.3 also contained potential
   tumor-suppressor genes associated with various cancers (B-cell
   lymphoma, hepatocellular carcinoma, head and neck squamous cell
   carcinoma, etc.)^[142]61–[143]63, and may be related to
   schizophrenia^[144]64. Microdeletions in 9q21.13 have been demonstrated
   to be associated with intellectual disability, delayed speech,
   epilepsy, autism-like behavior, and moderate facial
   dysmorphism^[145]65,[146]66. The 16p11.2 region has been linked to a
   range of neurodevelopmental disorders, including autism spectrum
   disorders, schizophrenia, intellectual disabilities, epilepsy, and
   cognitive impairment^[147]67–[148]69. Further, evidence in mice
   suggested that 16p11.2 copy number variations (CNVs) could lead to
   transcriptional dysregulation of various downstream pathways, resulting
   in changes in developmental trajectories and synaptic changes in
   distributed brain regions^[149]68. It is evident that these identified
   novel risk loci have been previously associated with neurodevelopmental
   disorders, cognition, language, movement, psychiatric disorders,
   cancer, and cardiovascular diseases, significantly overlapping with PD
   risk factors, complications and prodromal symptoms^[150]24. Although
   rarely significant correlation was found between these loci and HL, the
   relation between the auditory system and neurodevelopment is
   well-known.

   In functional enrichment analyses, pleiotropic genes were functionally
   enriched in brain (e.g., putamen basal ganglia, cerebellum, brain
   hypothalamus, amygdala, brain cerebellar hemisphere, anterior cingulate
   cortex BA24), heart atrial appendage, liver and pancreas. HL may cause
   challenges in sound perception and localization, possibly triggering
   increased neuronal activity in the putamen, a crucial basal ganglia
   component for motor control. Putamen dysfunction is closely associated
   with the hypokinesia and bradykinesia seen in PD
   patients^[151]70,[152]71. The cerebellum is essential for motor and
   cognitive functions, and its functional connections with the basal
   ganglia might be key in PD pathophysiology. Long-term HL can disrupt
   cerebellar functional circuits^[153]72,[154]73. The cerebellar
   hemispheres coordinate fine motor movements. Ischemia of the left
   internal auditory artery, from the anterior inferior cerebellar artery,
   can cause deafness and temporary vertigo^[155]74. The hypothalamus
   affects auditory function by regulating the
   hypothalamic-pituitary-adrenal (HPA) axis and promoting the release of
   glucocorticoids. The stress response regulated by the HPA axis can
   trigger or exacerbate motor symptoms in PD^[156]75–[157]77. Abnormal
   amygdala responses in PD patients may lead to emotional
   deficits^[158]78, while atrophy in the amygdala and temporal lobe
   regions, like the insula, is associated with HL and behavioral
   disorders^[159]79. The anterior cingulate cortex (ACC) is key for
   processing and regulating cognitive, sensorimotor, emotional, and
   autonomic functions^[160]80. It is linked to tinnitus distress, where
   increased activation and synchronized alpha activity correlate with
   emotional and auditory processing issues in severe cases^[161]81.
   Additionally, reduced gray matter in the ACC may lead to visual
   hallucinations in PD^[162]82. In individuals diagnosed with atrial
   fibrillation (AF), the atrial appendage emerges as a prevalent location
   for thrombus development, with AF serving as a distinct risk factor for
   both HL and PD^[163]83,[164]84. Additionally, genetic variants in
   allele genes specific to the atrial appendage have been linked to an
   increased susceptibility to PD^[165]85. A recent study has suggested
   that the liver X receptor (LXR), which is prominently expressed in
   organs such as the brain and liver, may play a key role in ameliorating
   PD neuropathology^[166]86. This potential effect of LXR may be
   attributed to its ability to decrease inflammatory signaling, mitigate
   neuroinflammation, reduce oxidative stress, address mitochondrial
   dysfunction, and enhance brain-derived neurotrophic factor
   signaling^[167]86. The association between pancreatic function and PD
   has been documented in the literature, particularly in relation to
   insulin and pancreatic amyloid peptide secretion^[168]87. A recent case
   study identified a 60-year-old PD patient with concurrent exocrine
   pancreatic insufficiency, potentially linked to the unfolded protein
   response^[169]88. An interesting illustration of the connection between
   HL and liver function is exemplified by Beethoven, who may have
   experienced hepatitis B-induced HL during the immune clearance
   phase^[170]89. Additionally, the link between diabetes and HL was
   close, with pancreatic dysfunction being a major culprit of
   diabetes^[171]90. The dihydrolipoyl dehydrogenase complex, identified
   as an enriched pleiotropic gene pathway, was a component of the
   pyruvate dehydrogenase complex, which was associated with pyruvate
   metabolism^[172]91. Disorders in pyruvate metabolism have been observed
   in Leigh syndrome, AD, and PD, potentially through mechanisms impacting
   cellular energy production^[173]92. The drug targets of Warfarin and
   Phenprocoumon are both anticoagulants widely used in the treatment of
   AF, and they may lead to various types of bleeding^[174]93,[175]94.
   Consequently, AF patients receiving these anticoagulant treatments may
   potentially be a high-risk population for PD, but further explorations
   are still needed.

   This study has several limitations. Firstly, our analyses were
   predominantly performed with Europeans, which minimizes population bias
   but also limits the generalizability of our findings to other
   populations^[176]95. Secondly, our prospective study, which utilized
   data from the UKB, was still constrained by limited PD cases especially
   early-onset patients, potentially impacting the robustness of our
   analysis. Thereby, we performed a range of sensitivity and stratified
   analyses to validate the reliability of our findings. Thirdly, since
   the definition of hearing aids use was based on self-report,
   respondents may misunderstand the question or provide inaccurate
   answers^[177]96. This information bias could lead to misjudgment of the
   actual use of hearing aids, thus affecting the reliability of our study
   results. In the MR analysis, the summary data for HL came from
   individual reports rather than SRT measurements as in our study, which
   possibly led to some degree of loss in comparability for association
   evidence available from various avenues. Finally, it is well-known that
   HL and PD are both age-related diseases. Although we made every effort
   to control for aging through rigorous statistical methods and study
   designs, the observed risk association may be still subject to bias due
   to the lack of solid functional evidence. Future laboratory research
   will help understand biological mechanisms underlying the observed
   relation, and further exploration is needed to investigate the relation
   between different types of HL and PD.

   In conclusion, early poor HL was significantly associated with an
   increased risk of developing PD especially late-onset cases and among
   males, and this relation was primarily driven by shared genetic factors
   rather than a direct causal link. The use of hearing aids protected
   against the onset of PD specifically among males. Overall, this work
   sheds light on novel perspectives and important references for future