Association studies can take the form of hypothesis-driven candidate gene or hypothesis-free GWAS, with the latter described in a subsequent section in this review. Case-control association testing compares genotype frequencies between unaffected and affected individuals and takes considerable differences between these two groups as evidence for or against disease susceptibility. Association studies in complex disorders can yield useful information if findings are replicated, or alternatively, if associations are confirmed with linkage analyses studying large families. Examples of replicated association findings are the discovery of the genes ANXA11 and BTNL2 in sarcoidosis and DTNBP1 and NRG1 in schizophrenia (Riley and Kendler, 2006; Spagnolo and du Bois, 2007). However, such findings are rather rare and many studies run the risk of artefactual positive association due to case-control selection bias, population admixture, or alleles residing in linkage disequilibrium (LD) with an allele directly affecting phenotype expression. Furthermore, in candidate gene association studies, the actual gene(s) of interest must already be identified for sequencing or genotyping. Late-onset disorders make selecting control groups challenging and can present a problem in young asymptomatic or undiagnosed individuals with risk alleles (Silverman and Palmer, 2000).

Case-control association testing has been a relatively widely employed approach and has comprised the majority of genetics research conducted in tinnitus patients to date (Table 2, Figure 1). Tinnitus candidate gene selection has included genes enriched in cardiovascular function (ACE, ADD1), neurotrophic factors (BDNF, GDNF), ion recycling pathways (KCNE1, KCNE3, SLC12A2), GABA_B receptor subunit (KCTD12), and serotonin receptor/transporter (HTR1A, SLC6A4) function. Presently, nine case-control studies have examined a combined total of 18 genes that are summarized in Table 2 (Kleinjung et al., 2006; Deniz et al., 2010; Sand et al., 2010, 2011, 2012a,b; Pawełczyk et al., 2012; Orenay-Boyacioglu et al., 2016; Yüce et al., 2016). For brevity, we describe selected case-control association studies with potentially significant results.

ADD1 encodes ubiquitously expressed alpha-adducin. A well-studied polymorphism (p.G460W) has been linked to cardiovascular disease and hypertension (Staessen and Bianchi, 2005). Hypertension-associated auditory primary lesion sites are the organ of Corti and stria vascularis (Gates et al., 1993). An association study investigated the relationship between severe chronic tinnitus and the p.G460W polymorphism in 89 patients with severe chronic tinnitus and 104 age-matched Turkish-Caucasian controls (Yüce et al., 2016). Clinical tinnitus evaluation and severity assessment were performed by the Structured Tinnitus Interview and the Tinnitus Handicap Inventory, respectively. PCR-based restriction fragment length polymorphism (RFLP) analysis of the ADD1 GW genotype (p = 0.009, χ² = 9.4) and the W allele (p = 0.021, χ² = 5.3) revealed significantly increased allele frequencies in the patient group (Yüce et al., 2016). This study asserted the potential involvement of the p.G460W genotype and W allele in ADD1 in tinnitus pathophysiology.

Tinnitus is thought to stem from central nervous system hyperexcitability and auditory cortical neuronal plasticity. Accumulating evidence indicates that tinnitus adaptation is dependent on cortical tonotopic map remodeling (Eggermont, 2016). An understanding of neurotrophins as important drivers of neural circuit remodeling in the auditory pathway have rationalized their relevance as candidate genes for tinnitus (Tan et al., 2007; Sand et al., 2012b).

PCR-based RFLP analysis in 240 German patients with Tinnitus Questionnaire (TQ)-scored subjective chronic primary tinnitus (Goebel and Hiller, 1994) has been performed for the genes BDNF and GDNF, encoding brain and glial cell-derived neurotrophic factors, respectively (Sand et al., 2012b). Both genes are essential in early central auditory pathway development. Two and three markers were investigated in BDNF (rs2049046 and rs6265) and GDNF (rs1110149, rs884344, and rs3812047), respectively. Comparison with reference data did not show significance after multiple testing correction; however, the authors could not exclude a weak modulatory effect. Furthermore, questionnaire intensity scores did not correlate with genetic variants, although notably, age-corrected multiple regression models with joint BDNF and GDNF genotypes indicated tinnitus severity could be predicted in women (p = 0.04, uncorrected; Sand et al., 2012b).

These three GDNF markers were similarly screened in a replication study including 52 Turkish patients with chronic tinnitus and 42 controls aged between 18 and 55 years (Orenay-Boyacioglu et al., 2016). No statistically significant distribution was detected in allele frequencies for all three markers between tinnitus and control groups. The only parameter reaching significance was heterozygosity (C:G) in the SNP rs1110149 (p = 0.02, χ²), that was found to have a lower frequency in tinnitus patients compared to controls (Orenay-Boyacioglu et al., 2016).

Pharmacological research has highlighted ion regulation and transport as potential therapeutic targets (Sand et al., 2011). Voltage-gated ion channels that are involved in auditory neural transmission by regulating endocochlear potentials are intriguing for exploration of tinnitus pathophysiology (Sand et al., 2010). The genes KCNE1 and SLC12A2 each encode a homologous β-potassium channel subunit and an inner ear Na⁺/2Cl⁻/K⁺ co-transporter, respectively, which have been screened in case-control association studies (Sand et al., 2010; Pawełczyk et al., 2012).

KCNE1 screening in 201 Caucasian chronic TQ-scored tinnitus patients detected four coding and three non-coding variants, including one novel p.Val47Ile substitution and another novel 3′ UTR variant that were concluded as having a non-significant (p = 0.05, Fisher's exact test) dominant genotype or compound genotype effect without correction for multiple testing (Sand et al., 2010). These variants were not found to be causal by themselves or in compound heterozygosity for chronic TQ-scored tinnitus.

In a Polish genotyping study that included 10 potassium recycling genes, 128 noise-exposed subjects with tinnitus, and 498 noise-exposed controls responded to a questionnaire and underwent analysis. Case and control groups were divided into noise-resistant (normal audiograms) and noise-susceptible (abnormal audiograms) groups and individuals with a family history of hearing loss and other clinical indications or medication exposures were excluded. KCNE1 and SLC12A2 were associated with tinnitus based on significance in only one genetic marker per gene (rs915539 in KCNE1, p = 0.018; rs10089 in SLC12A2, p = 0.026). p-values were not subjected to multiple testing correction and, therefore, suggested as nominally significant results (Pawełczyk et al., 2012).

Abundant data support chronic tinnitus with neuronal hyperactivity at different levels of the central auditory pathway, making drugs that increase inhibitory neurotransmission or block excitatory neurotransmission candidates for the treatment of tinnitus, and genes encoding these respective receptor complexes of potential interest (Eggermont and Roberts, 2004; Wang et al., 2011; Smith et al., 2012; Sand et al., 2012a). The gene KCTD12 encodes a potassium channel tetramerization domain-containing protein that is tightly associated with the GABA_B2 receptor carboxy-terminus (Sand et al., 2012a) and was subsequently subjected to association testing. The genomic DNAs from 95 German chronic TQ-scored tinnitus patients were obtained and the KCTD12 open reading frame and adjacent 3′ untranslated regions were sequenced. Two rare synonymous and non-coding heterozygous variants were detected. Further, analysis disclosed one significant tinnitus-associated variant (rs34544607; p = 0.04, Fisher's exact test), but this significance weakened after screening 50 additional cases (p = 0.07, Fisher's exact test). No novel variants were detected and no variants were correlated with or predicted intensity of tinnitus; however, the authors acknowledge the study was underpowered.

There is considerable overlap in patients reporting disabling tinnitus in conjunction with other comorbidities and a particularly strong association among patients with comorbid depressive disorder that affects ~5–10% of the general population. There is an estimated 30% concordant overlap between comorbid depressive disorder and tinnitus that implies common molecular mechanisms and, therefore, overlapping genes attributing to both phenotypes (Tyler et al., 2006). As such, genes involved in serotonin regulation, a critical process associated with depressive psychiatric disorders, have been proposed as tinnitus candidate genes. Serotonin is present in hair cells, eighth nerve fibers, brainstem auditory nuclei and nuclei of the lateral lemniscus and superior olivary complex (Tyler et al., 2006).

The gene SLC6A4 regulates serotonin neurotransmission and has been evaluated for tinnitus-association (Tyler et al., 2006; Deniz et al., 2010). A functional 5′-HTTLPR 44 base pair insertion-deletion polymorphism in the promoter region has been implicated in major depressive disorder (Hoefgen et al., 2005). This polymorphism and a 17 base pair variable number tandem repeat region in intron 2 were screened in 54 patients with subjective tinnitus and 174 population-matched controls. Tinnitus severity and psychoacoustic characteristics were assessed using the Beck Depression Inventory and visual analog scale, respectively. A significant association was detected between the 5′-HTTLPR polymorphism and visual analog scores that measured tinnitus quality of life impact using χ² tests (severity, p = 0.004; tinnitus discomfort level, p = 0.002; attention deficit, p = 0.04; sleep disorder, p = 0.04). This study linked a polymorphism in the SLC6A4 promoter with neurophysiological symptoms in tinnitus patients (Deniz et al., 2010).

GWAS can be powerful for the association of common variants and genetic loci in complex disorders and are appropriate for dissection of the “common disease-common variant” hypothesis. This hypothesis assumes a significant proportion of phenotypic divergence arises from common variants, typically with a minor allele frequency >5%, and that these variants are important for disease susceptibility (Sharma et al., 2014).

GWAS utilizes up to several million SNP genotypes most commonly generated from genotyping arrays to tag haplotype blocks, occasionally spanning more than 100 kb, on which functional variants reside. These studies utilize non-random co-inheritance of variants in linkage disequilibrium (LD) to test for case-control trait association (Edwards et al., 2013). p-value thresholds for statistical significance are very rigorous, typically below 10⁻⁹, to reduce the likelihood of false positive results, accommodate multiple testing burden, and provide enough stringency in studies that include lower frequency variants (minor allele frequency >5%; LaFramboise, 2009; Fadista et al., 2016). The conclusion after a successful GWAS is that one or more tag SNPs co-reside on haplotype blocks with variants having a biological function related to the phenotype. Interestingly, over 90% of disease-associated variants from GWAS reside in non-coding regions associated with transcriptional regulatory mechanisms involving promoter and enhancer element modulation (Maurano et al., 2012; Edwards et al., 2013). Following detection of statistically significant association signals, replication and functional experiments are required. A deeper understanding of these variants in a biological context requires experiments analyzing pathogenicity mechanisms such as transcriptional regulation, non-coding RNA function, and epigenetic regulation (Edwards et al., 2013). Functional assessment includes expression quantitative trait loci testing, in vitro protein and chromatin-structure assay analysis, as well as model organism experiments (Lee et al., 2014).

The first cross-sectional pilot tinnitus GWAS in ethnically homogeneous individuals between 55 and 65 years old was performed using 167 individuals with tinnitus and 749 non-tinnitus controls from Belgium (Gilles et al., 2017). These patients were previously included in a GWAS for age-related hearing loss in which a polygenic architecture was detected (Fransen et al., 2015). The association between tinnitus phenotype and 4,000,000 SNPs was tested and a gene-set enrichment analysis followed. 3.2% of the phenotypic variance was due to additive genetic effects. Although none of the SNPs reached genome-wide significance, potentially attributed to the limited sample size, the most interesting associations were revealed in the gene-set enrichment analysis that showed significance in seven metabolic pathways. The three most prominent pathways detected were the nuclear factor erythroid 2 like 2 (NRF2)-mediated oxidative stress response, endoplasmic reticulum (ER) stress response and serotonin reception mediated signaling pathways with low FDR-corrected p-values ranging from 0.004 to 0.02. NRF2-mediated oxidative stress plays a role in noise-induced hearing loss and tinnitus. Interestingly, tinnitus patients have been identified with substantially increased oxidative index levels compared to controls (Delmaghani et al., 2015; Koç et al., 2016). Moreover, ER stress has been associated with hearing loss via apoptosis (Van Rossom et al., 2015; Xue et al., 2016). Delayed hearing loss progression has been shown in transgenic mice with ER stress inhibitor treatment (Hu et al., 2016). Evidence of serotonin receptor mediated pathway involvement has been proposed from an apparent beneficial outcome of antidepressant usage among tinnitus patients (Baldo et al., 2012). Other pathways reaching significance include RAS, vascular smooth muscle contraction, coenzyme A biosynthesis, and NDK dynamin pathways (Gilles et al., 2017). Several limitations described by the authors included limited power to detect significant individual tinnitus-associated SNPs, the sample set was not selectively enriched for tinnitus patients and comprehensive controlling for risk factors was not undertaken. However, insight into seven potentially implicated pathways in tinnitus was highlighted from gene-set enrichment analysis.

Twin-based epidemiological studies serve as a means to estimate heritability by comparing disease concordance in monozygotic (MZ) vs. dizygotic (DZ) twins. Increased concordance in genetically identical MZ vs. DZ twins, who share on average half of their alleles, suggests a role for genetic factors. It is assumed both MZ and DZ twins share the same family environment, thus yielding important information about the contribution of genetic factors to disease etiology. Two recent twin studies have been published. While there are differences in the experimental approaches that are detailed below, they independently concluded that genetic factors contribute to tinnitus.

A twin study by Bogo and colleagues evaluated the genetic effects of self-reported tinnitus in male twins aged 52–96 years who were included in a previous longitudinal study (Bogo et al., 2015) that analyzed genetic influence of age-related hearing loss. Male MZ and DZ twins who were born between 1914 and 1958 were included in baseline (n = 1084 individuals) and follow-up (n = 576 individuals) assessments 18 years apart that included audiometry and self-reported answers to questions about tinnitus status and severity. The hypothesis was that individuals with faster hearing deterioration had the greatest tinnitus risk and that genetic factors influenced tinnitus (Bogo et al., 2016).

No difference in tinnitus prevalence between MZ and DZ twins at either time point was detected and those who reported tinnitus disclosed a mild severity. Individuals (n = 576) were placed in one of four categories (never reported tinnitus, n = 361; tinnitus only at baseline, n = 24; tinnitus only at follow-up, n = 139; and tinnitus at both time points, n = 52). Those who reported tinnitus at baseline and both baseline and follow-up assessments showed remarkably poorer hearing at follow-up across all frequencies compared to the reference group (those never reporting tinnitus). Those with tinnitus only at follow-up did not have significantly different hearing thresholds compared to the reference group. MZ twin concordant rates were much higher than for DZ twins at both time points (baseline: MZ 0.46; DZ 0.07; follow-up: MZ 0.51; DZ 0.32), proposing that genetic factors were important. A genetic correlation measuring the extent of genetic influences correlated in tinnitus and hearing thresholds ranged from 0.33 to 0.49, and suggested a partial overlap of genes associated with tinnitus and hearing loss; however, the authors also concluded that most of the genetic variation in tinnitus was unique to tinnitus and not associated with co-occurring hearing loss. There was a greater hearing threshold difference between discordant DZ twin-pairs compared to MZ twin-pairs in cases and controls. Interestingly, the hearing thresholds among MZ twins discordant for tinnitus were more similar than for discordant DZ twins, which may be due to genetic background. Compared to controls, individuals with tinnitus have statistically significant hearing threshold shifts. An overall heritability of 0.4 was calculated, which demonstrates a moderate genetic influence on tinnitus. The authors disclosed that their study was underpowered, and that noise exposure and other risk factors were not assessed (Bogo et al., 2016).

Another twin study by Maas and colleagues took a slightly different approach and controlled for tinnitus laterality but did not assess hearing thresholds (Maas et al., 2017). Cross-sectional data from the Swedish Twin Registry, that includes participants from the “Screening Across the Lifespan Twin study” and the “Study of Twin Adults: Genes and Environment” (Magnusson et al., 2013) who were born between 1900 and 1985. Concordance rates between MZ and DZ twin-pairs (n = 10,464 twin pairs) were assessed. As opposed to the previously described study that enrolled only male twins, this study also included opposite-sexed DZ twins to assess differences due to sex and shared environments. A higher concordance of tinnitus was observed in MZ twins (0.32) vs. DZ twins (0.20). Concordance between DZ same-sex (0.20) and opposite-sex (0.19) twins, same-sex male DZ (0.11) vs. same-sex female DZ (0.13), as well as male MZ (0.25) vs. female MZ (0.23) were all similar. When comparing bilateral tinnitus concordance in both twin groups, a higher concordance rate was detected in MZ (0.49) vs. DZ (0.30) twins that was observed in both sexes. Younger MZ females (0.39) had a greater concordance than DZ females (0.20), but this was not observed in males. Heritability was greater in men (0.68) than in women (0.41), except when considering female twins younger than 40 years of age, where a heritability of 0.62 was determined, although it was noted by the authors that this was a highly variable group. Overall, this study concluded that while tinnitus may be environmentally driven, bilateral tinnitus may have a genetic etiology (Maas et al., 2017).

Identifying the most homogeneous tinnitus patients in terms of etiology, age, sex, severity, onset, and audiometric profile will increase study robustness by limiting genetic variance that would be expected if tinnitus is associated with these factors (Lopez-Escamez et al., 2016). Furthermore, consideration for co-occurring psychiatric disorders and quality of life evaluations are important to recognize as potentially contributing determinants (Langguth et al., 2013). The present collection of case-control association studies has grouped patients into unspecific “chronic tinnitus” cohorts from questionnaire data and only occasionally accounted for concomitant hearing loss, impeding the selection of stratified patient groups. The clinically heterogeneous nature of tinnitus makes assigning patients to one of many clinical sub-categories difficult. Current tinnitus assessments are comprised of self-report questionnaires and psychoacoustic measures (Langguth et al., 2013). There have been several classification systems proposed that consider the origin (i.e., auditory system or head and neck) or classify tinnitus according to auditory, somatosensory or psychopathology alterations (Levine and Oron, 2015). Homogeneous patient selection relies on accurate patient sub-typing; however, the present definition of tinnitus sub-types lacks consensus and is complicated by the many recognized etiologies and risk factors (Lopez-Escamez et al., 2016). The application of universal assessment protocols by clinicians would potentially benefit genetics studies in that once genetic datasets are obtained, the merging of multiple datasets by collaborating working groups can be streamlined to enhance recruitment of the several thousand patients and controls required for adequate statistical power. Furthermore, there are over 100 instruments available for clinical trial primary outcome measures, which further asserts there is a lack of tinnitus assessment consensus (Hall et al., 2016; Müller et al., 2016). Determining the standardized instruments used by practitioners would consolidate some of the assessment practices for accurate phenotyping.

The unraveling of the human genome has delivered basic knowledge of a reference genome, advanced fundamental knowledge of disease architecture, and catalyzed technological advancements that fuel complex trait research. Recent developments in genetics technology and increased affordability and availability of genotyping array and sequencing data will undoubtedly propel research in the field forward. The present body of research has begun to disclose a genetic architecture for tinnitus but there is still much work ahead for the identification of specific variants influencing critical gene expression and gene products in tinnitus pathophysiology. As previously discussed, study design and patient inclusion are particularly important for a clinically heterogeneous phenotype such as tinnitus. Equally important is the method selection to support appropriate scale and resolution of data for analysis. The next section will discuss the feasibility of case-control association testing, GWAS, and twin and familial approaches in the context of tinnitus research.

Population based association studies have long been a popular strategy to identify polymorphisms correlated with complex traits and have thus far been the most widely employed genetic study in tinnitus genetics research (Table 2). Historically, case-control candidate gene studies have not yielded abundant success and independent replications are often not possible. The same problem has also been encountered in tinnitus, which makes a reevaluation of the current and future study designs essential.

Future tinnitus association studies should overcome several problematic design setbacks. One key aspect is to exclude controls with unassessed or unrecognized tinnitus burden and select stratified patient cohorts with sufficient statistical power. Furthermore, control individuals should ideally be matched for age, sex, population background, stress/anxiety traits, and other recognized co-morbidities. Studying already characterized tinnitus patients with homogeneous tinnitus phenotypes, families with transgenerational tinnitus aggregation, or cohorts from previously performed epidemiological health studies where individuals experiencing chronic tinnitus can be re-contacted for in-depth tinnitus scoring and auditory assessment would be beneficial (Lopez-Escamez et al., 2016). Finally, drawing upon knowledge from current GWAS for complex traits such as body weight and neuropsychiatric disorders, tens to hundreds of thousands of cases and controls are required to pinpoint significant loci (Ripke et al., 2014; Locke et al., 2015). However, such studies also underscore the power of a GWAS approach for the association of common variants and genetic loci (LD blocks) with specific diseases or traits. Learning from the scale of studies that are required for other complex traits and considering the clinical complexity of tinnitus, it is likely that future studies in tinnitus need to be increased by several orders of magnitude.

GWAS approaches have successfully dissected the genomic architecture of complex diseases and remain a robust approach for future tinnitus research. Looking more specifically at the collection of studies presently published, it is reasonable to assume several technological advances will transform study designs. The past decade has endowed affordable sequencing technologies that have revolutionized novel gene discovery by providing an avenue from which to effectively approach complex and Mendelian disorders (Koboldt et al., 2013). Next generation sequencing (NGS), also termed high-throughput sequencing, allows parallel amplification and sequencing of the entire protein coding region of the genome (whole exome sequencing) or the entire sequence of an individual's genome (whole genome sequencing). As the majority of significant findings from GWAS are tag SNP haplotype blocks in non-coding regions and NGS provides nucleotide-resolution of scalable coding and non-coding sequence (i.e., whole exome or whole genome sequencing), NGS can be regarded as a complementary methodology to GWAS and both methods have the potential to contribute important findings to the field. This would be especially promising if a combination of causal rare coding and common non-coding regulatory variants underlie tinnitus pathology. Examples from Alzheimer and Parkinson disease research underscore this parallel approach from the implication of both common and rare variants in these diseases that were detected from GWAS and NGS approaches, respectively (Sharma et al., 2012, 2014; Guerreiro et al., 2013).

The present collection of case-control studies for tinnitus has emphasized the need for several design considerations. There has yet to be a tinnitus case-control study with adequate statistical power. Complex disorders are more challenging to detect common susceptibility alleles and require larger sample numbers to separate signal from noise and to detect frequent alleles of modest effect (Risch and Merikangas, 1996). Furthermore, population differences can have drastically differing allelic architecture that can complicate the mapping of putative risk factors when attempting association replication in patients of differing ethnicities (Sharma et al., 2014). Although underpowered, even if significance were achieved in the initial association study analyzing GDNF markers in German tinnitus patients, the replication study that followed studied the same three markers in a Turkish tinnitus group (Sand et al., 2012b; Orenay-Boyacioglu et al., 2016).

The only pilot tinnitus GWAS published to date in a Belgian tinnitus cohort has highlighted the worthwhile investment of this approach in tinnitus patients (Gilles et al., 2017). In light of these results, it would be worth repeating a GWAS with a stratified patient cohort of extremely severe tinnitus patients or patients who are young and therefore have a greater chance of having tinnitus due to a genetic etiology. The patients included would need to have the same tinnitus sub-type and controls should be specifically assessed for tinnitus. Replication using a tinnitus cohort would be important to highlight the same pathways and potentially achieve genome-wide significance.

Early studies utilizing questionnaire data from families with tinnitus returned low heritability estimates. Although both of the twin studies that have been performed each had their limitations and utilized different inclusion criteria, these recently published studies served as groundbreaking evidence suggesting tinnitus is multifactorial with both genetic and non-genetic factors contributing to its etiology (Bogo et al., 2016; Maas et al., 2017). Another central nervous system disorder, Parkinson disease, has witnessed a similar period of debate about genetic factors contributing to the disease. Early evidence disclosed low heritability estimates from twin studies (Wirdefeldt et al., 2004) and lack of familial aggregation (Levy et al., 2004), while concurrent evidence also uncovered familial aggregation (Maher et al., 2002; Payami et al., 2002; Marder et al., 2003) and heritability estimates were later calculated at 0.34 (Wirdefeldt et al., 2011). Continued research has indeed uncovered the genetic complexity of Parkinson disease. It remains to be seen whether continued research into the genetic underpinnings of tinnitus will uncover similar observations. With respect to twin studies, future studies would have to take into account the same clinical phenotyping and study design details that other genetic studies must also address.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.