The genetic vulnerability to cisplatin ototoxicity: a systematic review

Ototoxicity is one of the major side-effects of platinum-based chemotherapy, in particular cisplatin (cis-diammine dichloroplatinum II). To our knowledge, no systematic review has previously provided a quantitative summary estimate of the impact of genetics upon the risk of developing hearing loss. We searched Embase, Medline, ASSIA, Pubmed, Scopus, and Web of Science, for studies documenting the genetic risk of ototoxicity in patients with cancer treated with cisplatin. Titles/abstracts and full texts were reviewed for inclusion. Meta-analytic estimates of risk (Odds Ratio) from the pooled data were calculated for studies that have been repeated twice or more. The search identified 3891 papers, of which 30 were included. The majority were retrospective (44%), ranging from n = 39 to n = 317, some including only patients younger than 25 years of age (33%), and some on both genders (80%). The most common cancers involved were osteosarcoma (53%), neuroblastoma (37%), prostate (17%) and reproductive (10%). Most studies performed genotyping, though only 5 studies performed genome-wide association studies. Nineteen single-nucleotide polymorphisms (SNPs) from 15 genes were repeated more than twice. Meta-analysis of group data indicated that rs1872328 on ACYP2, which plays a role in calcium homeostasis, increases the risk of ototoxicity by 4.61 (95% CI: 3.04–7.02; N = 696, p < 0.0001) as well as LRP2 rs4668123 shows a cumulated Odds Ratio of 3.53 (95% CI: 1.48–8.45; N = 118, p = 0.0059), which could not be evidenced in individual studies. Despite the evidence of heterogeneity across studies, these meta-analytic results from 30 studies are consistent with a view of a genetic predisposition to platinum-based chemotherapy mediated ototoxicity. These new findings are informative and encourage the genetic screening of cancer patients in order to identify patients with greater vulnerability of developing hearing loss, a condition having a potentially large impact on quality of life. More studies are needed, with larger sample size, in order to identify additional markers of ototoxic risk associated with platinum-based chemotherapy and investigate polygenic risks, where multiple markers may exacerbate the side-effects.

Criteria for considering studies for this review. All studies written in English were considered eligible for this review. There was no restriction on participant age since studies with both children and adults were included. All different study designs were taken into account. Studies that were not available in English were excluded as we did not have the resources to translate them. Both adults and children were included in the review as many of the studies have been in children and since it is known that cisplatin causes more severe ototoxicity in children than in elders 22 . In vitro and in vivo studies were excluded because cell lines and animals are not fully representative of the ototoxic effects that platinum-based chemotherapy could have on humans.
Data extraction and Management. Data extracted included study design, demographic characteristics, intervention and genetic association. Data extraction tables were developed and piloted for this purpose. Where data were missing or unclearly reported, an attempt was made to contact the relevant corresponding author of the study. Three articles were excluded after reading the full text. One paper was excluded because platinum-based chemotherapy was only studied by in vitro methods 23 . Another was excluded because there was no association between cisplatin ototoxicity and the mitochondrial mutations, which they analysed and there is no report of ototoxicity grade 27 . A third paper was excluded because the statistical results are based on comparison with craniospinal radiation 28 . A study by Upadhya et al. reported that 31.4% of patients had sensorineural hearing loss 6 months after radiation of the ear 29 . Radiation causes ototoxic effects, therefore is a confounding factor when investigating the ototoxic effect of chemotherapy. Studies by Brown et al. 30 , Drögemöller et al. 31 , Olgun et al. 32 and Wheeler et al. 33 are included in the socio-demographic and the cisplatin intervention tables but not in the forest www.nature.com/scientificreports www.nature.com/scientificreports/ plots since not all values about patients with or without ototoxicity in relationship with the genetic profile were available.
The data from each article was extracted and summarized in an extraction form (Table 1 & Sup. Table S1). The extraction form includes socio-demographic data of the study participants, details of the treatment intervention and audiological assessment and the results of the statistical analysis of the genes examined. The Critical Appraisal Skills Program checklist was used to assess the validity and results of each article included in the systematic review (CASP Critical Appraisal Skills Program Oxford UK, 2017).

Risk of bias (Quality assessment).
Risk of bias assessment was conducted by four authors (E.T., T.N., N.E. & C. R. C) on those study records included in the meta-analysis. Risk of bias criteria that were taken into consideration in this review were the study population (age, gender, ethnicity), type of cancer (any type of malignancies), type of intervention (other ototoxic drugs, irradiation, prior hearing loss) and measurement of hearing outcome. All these criteria were taken into account in the interpretation of the results.

Meta-analysis.
Forest plots were created using the Forest Plot add-in (version 8.0) for JMP 13.2.1 data analysis software to visually summarize the results from each study included in this review. The forest plots display the odds ratios (OR) and 95% confidence intervals that demonstrate the association between ototoxicity and the various genes and single-nucleotide polymorphisms (SNPs) reported in the literature. One forest plot was created to demonstrate the results for the genes tested in a single study. A second forest plot was created to compare the findings of different studies examining the same genes and SNPs and provide an estimate of the combined result of these studies. This is a meta-analysis of the data. The combined odds ratios and 95% confidence intervals were calculated using the values for the number of variant SNPs and controls in the patient groups with normal hearing and with ototoxicity after chemotherapy. statistical analysis. The   www.nature.com/scientificreports www.nature.com/scientificreports/ testing. Although Fisher's exact test is preferable whenever the computational power allows to carry it out, we also report results from the commonly used chi-squared (χ 2 ) test to ensure comparability with the literature.
Furthermore, we report odds ratios (OR) for quantifying the risk of ototoxicity. The OR for being affected by ototoxicity if also having the SNP variant was then calculated as:
Radiotherapy or other ototoxic drugs (such as aminoglycosides and vincristine) were used as part of the treatment of patients in 24 of the included studies. Six papers did not specify whether other ototoxic drugs or radiotherapy were used 33,[37][38][39][40]52 . Radiation to the head and neck causes ototoxic effects and is a confounding factor when investigating the ototoxic effect of chemotherapy 29 . Only 8 studies adjusted the statistical analysis to relevant clinical variables, such as age at diagnosis, gender, ethnic group, cumulative cisplatin dose, vincristine treatment and craniospinal irradiation doses 43,[47][48][49][50]55,59,60 .
Twenty SNPs of 9 genes were investigated once without having been repeated (Fig. 2). Three of the SNPs were shown to be otoprotective 36, 38,60 . Epoxides are among the many targets of GSTs. Converging this pathway, the Epoxide Hydrolase 1 (EPXH1) rs2234922 was related to otoprotection (OR: 0.05; 95% CI: 0.00-0.94; n = 84; p = 0.004) 36  Nineteen SNPs of 15 genes were investigated at least twice and the meta-analysis is shown in Figs 3 and 4. Seven of these SNPs showed no overall effect, namely the copper transport protein 1 CTR1 rs10981694, GSTM1 and T1 deletions, GSTP1 rs1695, SLC16A5 rs4788863, XPC rs2228001 [a component of nucleotide excision repair (NER)] and XPD rs1799793. Albeit, XPD rs1799793 did not present an overall effect, was significantly ototoxic in one study (OR: 2.621; 95% CI: 1.13-6.10; n = 106; p = 0.034) 47 . The low-density lipoprotein-related protein 2 (LRP2) encoding the protein megalin was shown positively associated with ototoxicity on 2 SNPs (rs2075252 and rs4668123, Fig. 4). Interestingly, while the latter did not appear significant in hypothesis testing in 2 studies 40,56 , the accumulated data supports a positive association (OR: 3.532; 95% CI: 1.48-8.45; n = 118; p = 0.0059), likely due to an increase in the statistical power. Three SNPs (rs12201199, rs1142345, rs1800460) on the thiopurine S-methyltransferase gene (TPMT) were found significant in 2 studies 42,43 and not in 3 others 44,46,57 . However, the overall pattern of the 5 studies merged together displayed significant associations with increased ototoxic risk from OR 2.47 to 2.82, with a total sample size of 786 (p < 0.0001). Another variant in COMT rs9332377  . Forest plot describing ABCC3 rs1051640, ACYP2 rs1872328, COMT rs4646316, COMT rs9332377, CTR1 rs10981694, GSTM1 del, GSTM3 rs1799735, GSTP1 rs1695, GSTT1 del tested in multiple studies. Black indicates a non-significant association with ototoxicity, blue indicates a significant association with ototoxicity and red a significant association with otoprotection. The square is centred on the odds ratio and the horizontal line represents the 95% confidence interval of each study. The diamond summarises each SNP average OR and the horizontal shows the 95% confidence interval. n = overall sample size. The asterisk (*) identifies studies in which the p value to reach significance differed between Fishers and χ 2 tests. . Forest plot describing LRP2 rs2075252, LRP2 rs4668123, SLC16A5 rs4788863, SLC22A2 rs316019, SOD2 rs4880, TPMT rs1142345, TPMT rs12201199, TPMT rs1800460, XPC rs2228001, XPD/ERCC2 rs1799793 tested in multiple studies. Black indicates a non-significant association with ototoxicity, blue indicates a significant association with ototoxicity and red a significant association with otoprotection. The square is centred on the odds ratio and the horizontal line represents the 95% confidence interval of each study. The diamond summarises each SNP average OR and the horizontal shows the 95% confidence interval. n = overall sample size. The asterisk (*) identifies studies in which the p value to reach significance differed between Fishers and χ 2 tests. (2019) 9:3455 | https://doi.org/10.1038/s41598-019-40138-z www.nature.com/scientificreports www.nature.com/scientificreports/ 1.18-2.05; n = 847; p = 0.002). The acylphosphatase-2 ACYP2 variant rs1872328, showed in Vos et al. 58 and in Xu et al. 54 reports a significant association (p = 0.0274 and p < 0.0001, respectively, by Fisher's test), in spite of an extremely large confidence interval (Fig. 3). High OR of this SNP was also presented in another study showing the strong relation with cisplatin ototoxicity 50 . The combined data reveals a strong positive association with an overall risk of 4.618 (95% CI: 3.04-7.02; n = 696; p < 0.0001). Another gene playing an important role in oxidative stress, superoxide dismutase 2, mitochondrial (SOD2), with the rs4880 showing a positive association with cisplatin ototoxicity in one of the two studies 45 , but also in the overall meta-analysis with an OR of 1.917 (significant with χ 2 , but not with Fisher's test, Fig. 4).

Discussion
One of the most prevalent adverse effects of cisplatin treatment is ototoxicity, in which the consequent hearing loss -although is not lethal -has a non-negligible impact on life quality. Hearing deficits are now 4 th in the leading causes of years lived with disability 8 . It is thus an important factor to consider when performing cisplatin interventions in patients, not only to inform patients of the risks, but also to determine whether a given individual has a greater risk and for whom the regimen of administration may be adjusted or alternatives to cisplatin being given. The confirmation that there is a non-negligible risk for ototoxicity when treating patients with cisplatin, and that this risk is influenced by genetics, is in support for greater cautiousness in considering auditory impairments that cancer patients may develop.
A striking finding from this systematic review is that studies with non-significant findings in isolation reached sufficient power when combined to show increased risk of developing cisplatin-induced ototoxicity. This is the case for LRP2 rs4668123, which emphasizes the need of considering larger sample sizes when performing such studies in order to provide more statistically solid evidence. Although our meta-analysis did not use individual data nor included adjustments (for instance for age, sex, the ethnic group, and the cumulative cisplatin dose), the summarized analysis emphasizes the need of large sample sizes to reveal biologically relevant associations that would otherwise been underestimated or missed.
We identified 8 different SNPs from 5 different genes (including rs4668123 from LRP2) from repeated studies showing significant associations with cisplatin ototoxicity ( Table 2). These genes are mainly related to anti-oxidant regulation, neurotransmission or to auditory function. ACYP2 encodes the acylphosphatase-2 expressed in the cochlea that hydrolyses phosphoenzyme intermediates of membrane pumps that affect Ca 2+ ion homeostasis 54 . While ATP-dependent Ca 2+ signalling has been shown to be involved in hair cell development, the exact role of ACYP2 on the cochlea remains unknown 54 . Interestingly this ACYP2 polymorphism showed the highest average risk (OR: 4.618), which suggests its major involvement in cisplatin ototoxicity and opens the possibility for more investigations addressing the contribution of this polymorphism in cisplatin-mediated ototoxicity.
With an OR ranging from 2.8 to 3.53, the LRP2 rs2075252 and rs4668123 polymorphisms also appear as important risk factors for developing cisplatin-mediated ototoxicity. LRP2 or megalin is currently the only gene associated with Donnai-Barrow syndrome, a condition characterized by craniofacial anomalies, ocular abnormalities, sensorineural deafness and developmental delay 62 . LRP2 is also connected with diabetic nephropathy, Lowe syndrome, Dent disease, Alzheimer's disease (AD) and gallstone disease 63 .
TPMT is a methyltransferase, which enzymatic activity varies depending on polymorphisms of TPMT gene in chromosome 6 64 . A decreased enzymatic activity leads to myelosuppression, gastrointestinal intolerance, pancreatitis and hypersensitivity 64 . Here, 3 polymorphisms were found with significant OR (Table 2), ranging from 2.47 to 2.82 suggesting a strong involvement of TPMT in cisplatin-mediated ototoxicity. Again, 2 out of 5 studies found positive associations, and the overall outcome was a clear risk in spite of the 3 negative associations. Interestingly, rs12201199 has been linked to the 3A haplotype, leading to a reduced activity of the TPMT enzyme and greater toxicity of the anticancer drugs thiopurine and mercaptopurine 65 . It was recently demonstrated that HEI-OC1 and UB/OC-1 cells derived from the cochlea are more sensitive to cisplatin when expressing the TPMT*3A variant instead of the wild-type Tpmt 65 . In contrast, Tpmt knock-out mice do not display an increased sensitivity to www.nature.com/scientificreports www.nature.com/scientificreports/ cisplatin when administered at comparable levels as found in humans 66 , however this result might not be surprising given the known resistance of mice to cisplatin ototoxicity when compared to rats or guinea pigs 67 .
Of the two polymorphisms tested for COMT, only rs9332377 appeared as an important risk factor, although displaying the smallest OR of all validated studies (OR: 1.553). Mutations in COMT genes are implicated in sensorineural deafness. Hearing loss is less severe in subjects with COMT Met allele, possible due to the protective effect of dopamine on the hearing system 68,69 . While COMT has not been described in the cochlea, a homolog sharing 30% sequence identity, Comt2 was found expressed in hair cells and mice homozygous for a missense mutation in Comt2 showed sensorineural deafness due to degeneration of hair cells 70 . Overall, there are strong indications that catecholamines play a potential role in the auditory function. Thus, a greater vulnerability to cisplatin ototoxicity may arise when the function of the auditory system is already weakened.
Another important polymorphism related with increased risk of cisplatin ototoxicity is SOD2 rs4880 presented an overall OR of 1.917. SOD2 catalyses the metabolism of the highly toxic superoxide anion to less but still toxic hydrogen peroxide. The SNP rs4880, which results in an exchange of valine against alanine, increases the catalytic activity of SOD2, leading to the accumulation of hydrogen peroxide and secondary ROS generation 45 . It is thus possible that altered mitochondrial function in the cochlea may increase the vulnerability to cisplatin ototoxicity. Notably, SOD2 polymorphisms (IVS3-23T/G; IVS3-60T/G; and V16A) have also been implicated in noise induced hearing loss (NIHL) 71,72 .
Three polymorphisms, which have been evaluated twice, were found with a significant oto-protective effect, namely ABCC3 rs1051640, GSTM3 rs1799735 and SLC22A2 rs316019. ABCC3 is an ATP-binding cassette member of the MRP subfamily which is involved in multi-drug resistance. This transporter regulates the efflux of organic anions, glutathione S-conjugates and xenobiotics 73,74 . MRP expression in cancer cells correlates with resistance to cisplatin 75 . The mechanisms by which ABCC3 regulates cisplatin-induced hearing loss are unclear, but some studies suggest it may act upstream of GST 73 . Indeed, consistent with the otoprotective effects of the ABCC3 variant, GSTM3 rs1799735 shown to be otoprotective by Peters et al. 40 but also in the overall analysis. GSTM3 variations are indeed thought to alter the susceptibility to potential carcinogens and toxins 76 . SLC22A2 is a solute carrier that encodes CTR1, which is a plasma-membrane transport-protein that has an essential role in cisplatin uptake into cochlea hair cells. Since cisplatin accumulates in the stria vascularis from the cochlea 9 , polymorphism that positively affects monocarboxylate transporter function may improve the strial function affected by cisplatin.
Only two studies have evaluated polygenic effects on the vulnerability to cisplatin ototoxicity. Oldenburg et al. evaluated the cumulated risks of combinations in variants of GSTT1, GSTM1 and GSTP1 48 . Such an approach makes sense when considering that the overall results for GSTM1 and T1 appeared inconclusive (Fig. 3). However, the combination of GSTM1 null, T1 null and P1 Ile105/Ile105 alleles had a major impact on the risk for severe hearing impairment 48 . These findings are consistent with the known association of GSTM1 null, T1 null and P1 Ile105/Ile105 genotypes with greater vulnerability of developing NIHL 77,78 and an 8.88-fold increase in the risk of developing presbyacusis (sensorineural hearing loss caused by natural ageing) 79 . The incapacity of these individuals to conjugate certain metabolites may ultimately cause oxidative stress and damage to the cochlea, which would be exacerbated in presence of cisplatin. Pussegoda et al. also performed plurigenic analyses but included more complex models incorporating clinical and genetic variables 43 . Here the combination of TPMT rs12201199, ABCC3 rs1051640, and COMT rs4646316 in a high risk group could reach an OR of 11 (95% CI: 3.2-37.6). Such studies highlight the need of considering multigenic screens when assessing the risk for ototoxicity which may be underestimated when considering a single marker.
There are a number of limitations to be noted in the present study. First, our meta-analysis was performed using group data and not individual data, which pre-empted the possibility of adjusting for e.g. age at diagnosis, gender, ethnic group, cumulative cisplatin dose. Second, in all studies reviewed, hearing loss associated with cisplatin chemotherapy was assessed immediately or soon after treatment. Given that cisplatin persists indefinitely in the human cochlea after such treatment 10 , the possibility of longer-term cochlear vulnerability (and hence progressive hearing loss) cannot be discounted. Third, ototoxic effects do not only lead to hearing loss, but also tinnitus, and vestibular toxicity 80 , which were not assessed in the present review, may help determining additional impacts on the auditory system that cannot be revealed with traditional audiometry. There are numerous ototoxicity grading scales used across the different studies. In the clinical trial setting, standardization is vital and the variability between different studies makes analysis more challenging 81 . Currently, there are 2 main categories of ototoxicity assessment criteria: (1) those measuring a change of hearing from baseline, such as the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE), and (2) those measuring absolute hearing levels, such as the Brock or Chang classifications 81 . Interestingly, Spracklen et al. used the CTCAE and Chang grading scales, but also the American Speech-Language-Hearing Association criteria (ASHA), and the resulting associations differed depending on which scale is used 60 . As a matter of fact NFE2L2 polymorphisms presented as significantly ototoxic in ASHA and CTCAE scales (the latter of which was included in Fig. 2) but not when Chang scale is used 60 . As a consequence, the selection of the grading scale can have a dramatic impact on the outcome of the study. These findings highlight the needs of determining the most sensitive measures in order to standardise the methodologies into the context of genetic testing in ototoxic vulnerability.
Finally, whilst a genetic predisposition to cisplatin mediated sensorineural hearing loss has been identified and may help identifying cancer patients with greater ototoxic risk, the specific mechanisms remain elusive. This would be an essential precursive step to the development of oto-protective therapy together with cisplatin interventions. It has recently been demonstrated that specifically targeting the p53 pathway protects from cisplatin ototoxicity while still maintaining cancer treatment efficacy 20 . Knowing the genetic predisposition to cisplatin is an important advancement for improving clinical treatment but now new therapies that target specific pathways are being developed to protect against cisplatin-induced ototoxicity.