Skip to main content
Download PDF

Leveraging large-scale datasets and single cell omics data to develop a polygenic score for cisplatin-induced ototoxicity

Human Genomics volume 18, Article number: 112 (2024) Cite this article

Abstract

Background

Cisplatin-induced ototoxicity (CIO), characterized by irreversible and progressive bilateral hearing loss, is a prevalent adverse effect of cisplatin chemotherapy. Alongside clinical risk factors, genetic variants contribute to CIO and genome-wide association studies (GWAS) have highlighted the polygenicity of this adverse drug reaction. Polygenic scores (PGS), which integrate information from multiple genetic variants across the genome, offer a promising tool for the identification of individuals who are at higher risk for CIO. Integrating large-scale hearing loss GWAS data with single cell omics data holds potential to overcome limitations related to small sample sizes associated with CIO studies, enabling the creation of PGSs to predict CIO risk.

Results

We utilized a large-scale hearing loss GWAS and murine inner ear single nuclei RNA-sequencing (snRNA-seq) data to develop two polygenic scores: a hearing loss PGS (PGSHL) and a biologically informed PGS for CIO (PGSCIO). The PGSCIO included only variants which mapped to genes that were differentially expressed within cochlear cells that showed differential abundance in the murine snRNA-seq data post-cisplatin treatment. Evaluation of the association of these PGSs with CIO in our target CIO cohort revealed that PGSCIO demonstrated superior performance (P = 5.54 × 10− 5) relative to PGSHL (P = 2.93 × 10− 3). PGSCIO was also associated with CIO in our test cohort (P = 0.04), while the PGSHL did not show a significant association with CIO (P = 0.52).

Conclusion

This study developed the first PGS for CIO using a large-scale hearing loss dataset and a biologically informed filter generated from cisplatin-treated murine inner ear snRNA-seq data. This innovative approach offers new avenues for developing PGSs for pharmacogenomic traits, which could contribute to the implementation of tailored therapeutic interventions. Further, our approach facilitated the identification of specific cochlear cells that may play critical roles in CIO. These novel insights will guide future research aimed at developing targeted therapeutic strategies to prevent CIO.

Background

Cisplatin is an effective platinum-based chemotherapeutic that is used to treat a spectrum of malignancies, including leukemias, lymphomas, sarcomas, as well as breast, testicular, ovarian, head and neck, and cervical cancers [1]. The effectiveness of this treatment is highlighted by the fact that the introduction of cisplatin into testicular cancer treatment protocols in 1978 led to a 9-fold increase in survival rates for these patients [2, 3]. Unfortunately, cisplatin treatments are also accompanied by the occurrence of adverse drug reactions (ADRs), which can significantly diminish patients’ quality of life [4, 5]. One such example is cisplatin-induced ototoxicity (CIO), a common ADR characterized by permanent, progressive, and bilateral hearing loss [6]. The exact mechanisms underlying the development of CIO are still not completely understood, but several pathways have been implicated [5]. This includes damage to cochlear cells through the cytotoxic effects of cisplatin, damage to nuclear and mitochondrial DNA [7, 8], and the activation of apoptosis [9] through direct and indirect means [10,11,12].

The incidence of ototoxicity after cisplatin treatment, which can be up to 80% for some cancers [13], adds to the challenges and burdens already faced by cancer patients and survivors. Further, the development of hearing loss is associated with adverse consequences throughout these individuals’ lives, including challenges in speech and language development for young children, as well as the potential for increased social isolation, or depression [14]. As sodium thiosulfate has recently been approved by the Food and Drug Administration (FDA) to reduce the risk of CIO in pediatric patients with non-malignant cancers [15], CIO risk prediction models could help to prioritize which individuals would benefit most from these otoprotectants. Several risk factors for CIO have been identified, including age, with pediatric and older adults at higher risk [5], high cisplatin administration schedules [16], and cranial irradiation, which is considered the most significant clinical risk factor for CIO [17]. As a result, current predictive models for CIO have focused on using clinical risk factors to predict risk of CIO [18, 19].

While clinical factors play an important role in risk of CIO, they do not account for all the variability observed for this ADR. In line with this, heritability studies suggest that genetic variation also plays an important role in CIO, with up to 47% of the variability in the occurrence of CIO attributed to genetics [20]. Notably, genome-wide association studies (GWAS) have identified an increasing number of genetic variants that contribute to CIO risk [17, 21, 22]. Unfortunately, due to the polygenic nature of CIO, individually, these genetic variants offer only partial insights into overall genetic risk for CIO [23]. This highlights the need for polygenic scores (PGS), which integrate risk information from multiple genetic variants across the genome to create a single risk score for each individual. However, accurately generating these scores for CIO remains challenging due to relatively small cohorts, resulting in inaccurate effect size estimates for risk variants [24]. Therefore, innovative approaches are required to overcome these challenges and enhance the accuracy of PGSs.

To overcome these limitations in this study, we integrated data from a large-scale multi-trait analysis of GWAS (MTAG) of hearing loss [25] with cisplatin-treated murine inner ear single nuclei RNA sequencing (snRNA-seq) data. Due to the genetic overlap between hearing loss phenotypes and CIO [26], we used MTAG data to increase our power to uncover variants that play a role in several diverse hearing traits. Further, given the potential tissue-specific effects of cisplatin on gene expression, the generation of inner ear snRNA-seq data obtained from cisplatin-treated mice allowed for the incorporation of a biologically informed refinement filter to enhance the relevance of our initial PGS (i.e., PGSHL) to CIO. By harnessing these unique data, we developed a refined PGS (i.e., PGSCIO) that was associated with CIO in two independent cohorts, providing further insights into the genetic factors that influence susceptibility to CIO.

Methods

Study cohorts

Base cohort

Given the observed overlap in the genetic architecture between hearing loss and CIO [26] we obtained summary statistics from a previously published MTAG of four genetically correlated, heritable and polygenic hearing loss traits from the UK Biobank (UKB) [25]. These data were used as a base cohort to develop a hearing loss PGS that would be of relevance to diverse hearing loss traits. The four self-reported hearing traits included in this MTAG were hearing difficulty with background noise (cases: n = 134,141; controls: n = 219,842), hearing difficulty (cases: n = 90,710; controls: n = 255,925), hearing aid user (cases: n = 10,942; controls: n = 208,416), and tinnitus (cases: n = 7,739; controls: n = 110,142). The use of this large-scale MTAG dataset allowed us to circumvent sample size limitations associated with current CIO GWAS, providing more accurate effect size estimates for variants that are associated across diverse hearing loss phenotypes [24].

Target cohort

To investigate the association between the PGSs and CIO, GWAS summary statistics from the PanCareLIFE (PCL) cohort were obtained. This cohort is described in detail elsewhere [27], but consists of 390 pediatric cancer patients of European (92%), and non-European ancestry (8%) (African-American/African-Caribbean and Latin American) who underwent cisplatin treatment, but were not subjected to cranial radiation therapy. Cases and controls were defined by the Muenster grading criteria, with 168 individuals exhibiting CIO (Muenster grade > 2b), while 222 individuals showed no or minimal ototoxicity (Muenster grade 0-2a). Genotyping was performed using the Illumina Infinium Global Screening Array. Quality control (QC) and imputation, using the Michigan Imputation Server and the Haplotype Reference Consortium reference (HRC r1.1), were performed as previously described [27].

Test cohort

To validate the association between the PGSs and CIO, individual-level genotype and clinical data from two St. Jude’s Children Research Hospital Medulloblastoma (SJMB) cohorts, the SJMB96 cohort consisting of 31 children enrolled between 1996 and 2003 and the SJMB03 cohort consisting of 207 children enrolled between 2003 and 2012, were used. All individuals enrolled in both the SJMB96 and SJMB03 cohorts underwent craniospinal irradiation (CSI) and cisplatin treatment. Further, cases and controls were defined by the Chang grading criteria, with 61% of individuals exhibiting CIO (Chang grade\(\:>\) 0), while 39% of individuals showed no ototoxicity (Chang grade 0). Genotyping of these samples was performed using the Illumina HumanOmni2.5+HumanExome BeadChip, as previously described [17].

Generation of mouse inner ear single-nuclei RNA-sequencing data

Intraperitoneal injections of 3.0 mg/kg cisplatin or saline were administered to postnatal day-6 (P6) CBA/CaJ mice in the control (n = 6) and treatment groups (n = 6), respectively. This dosing schedule has been shown to produce a clinically relevant mouse model for CIO [28]. P6 mice were chosen for the absence of ossification in the cochlear cartilaginous membrane, facilitating optimal dissection of intact cochlea samples and eliminating the need to include decalcification processes which may alter gene expression profiles, thereby enhancing compatibility with snRNA-seq. snRNA-seq was used as these data form part of a larger multiome dataset, which requires nuclei for ATAC sequencing. Four hours post-treatment, corresponding to the timepoint when cisplatin-induced gene expression changes were observed in other organs in mice [29], mice were euthanized via decapitation and their cochleae were micro-dissected as previously described [30]. After dissection, the cochleae were flash-frozen in liquid nitrogen for 45 s and stored in a liquid nitrogen tank. Single nuclei were isolated from pooled treatment and control cochlea samples using the 10X Genomics Chromium Nuclei Isolation Kit, according to the manufacturer’s protocol, modified to include only one round of washing to allow for maximum nuclei yield. Before single-cell library preparation, the concentration and viability of the nuclei were assessed using AO/PI staining and the CellDrop automated cell counter. Libraries were prepared using the Single Cell GEX & Fixed RNA Profiling Kit and a local Chromium Controller, according to the manufacturer’s protocol. These libraries were frozen and sent for sequencing with a target depth of 50 million reads/sample on the NovaSeq X Plus Sequencing System at Princess Margaret Genomics Sequencing Centre. Technical replicates were performed for each sample by preparing two separate libraries from the same nuclei suspension across two channels on the Chromium microfluidic device.

After sequencing, data processing was carried out, as detailed in Supplementary Fig. 1. Alignment of reads was performed using CellRanger, followed by QC checks using FastQC. Further QC was performed to correct for ambient RNA using SoupX, filtering low-quality cells using Seurat, and doublet removal using scDblFinder. Low-quality cells were identified based on their mitochondrial percentages deviating more than 3 median absolute deviations (MADs), as well as cells with log10-transformed gene counts and UMI counts exceeding ±3 MADs. The data was normalized and scaled using Seurat’s NormalizeData and ScaleData functions. Feature selection identified informative genes, followed by principal component analyses (PCA) and cell clustering. Uniform Manifold and Approximation Projection (UMAP) was then used to visualize the data, and clusters were annotated using Seurat’s data transfer method based on previously published murine cochlear data [31].

Construction of a hearing loss PGS (PGSHL)

MTAG summary statistics were harmonized using the S-PrediXcan harmonization pipeline [32] to ensure consistent formatting and compatibility across different datasets. These harmonized summary statistics, along with the GCTB sparse shrunk linkage disequilibrium (LD) reference panel consisting of 2.8 million common variants from the UKB, were used as inputs for SBayesR [33] to construct a PGS for hearing loss using Bayesian-based approaches. We used SBayesR default parameters with the “—robust” option to address potential differences in SNP effects across traits in the MTAG dataset [33]. Ambiguous variants (i.e., C/G, G/C, A/T and T/A variants) were excluded from downstream analyses.

Refinement of PGSHL to increase the relevance of the score to CIO (PGSCIO)

To enhance the relevance of PGSHL to CIO [24], a biologically informed pharmacogenomics filtering strategy was applied. To do this, analyses with MILO-R [34] were performed to identify cochlear cell types that exhibited differential abundance post-cisplatin treatment, when compared to saline-treated samples (LogFC<=-3, SpatialFDR < 0.1). Genes that were differentially expressed in these cells were identified using DESeq2 [35] (Padj<0.05). Next, Ensembl’s Variant Effect Predictor v.108.1 was used, with default settings, to map genetic variants included in the PGSHL to human Ensembl gene IDs. Corresponding mouse orthologues were identified using the R package, babelgene. These data were then used to filter the PGSHL to include only variants mapping to murine genes that were differentially expressed within cochlear cells that showed differential abundance in the snRNA-seq data post-cisplatin treatment (Fig. 1). This filtering strategy resulted in the generation of a biologically relevant PGS (PGSCIO) for downstream analyses, with effect size weights of variants included in this score derived from the base cohort.

Fig. 1
figure 1

Beeswarm plot illustrating cochlear cells showing differential abundance four hours post-cisplatin treatment vs. saline-treated samples. Each dot represents a neighborhood of cells. Red indicates decreased abundance and blue indicates increased abundance of cells treated with cisplatin relative to saline-treated controls

Full size image

Testing the association between PGSHL/PGSCIO and CIO in the target cohort

The association between PGSHL/PGSCIO and CIO was evaluated using the CIO GWAS summary statistics from the PCL cohort. To enable comparison between datasets, the summary statistics were harmonized using the S-PrediXcan pipeline. These harmonized summary statistics were used to assess the associations between the PGSHL/PGSCIO and CIO using ReACt [36]. The use of this tool circumvented the need to obtain individual-level data, thereby streamlining these analyses [37]. By reconstructing allelic frequencies [36] from the summary statistics obtained from the target (PCL) cohort, ReACt was used to calculate the mean PGS in cases and controls and test for significant differences between CIO cases and controls using a two-sample t-test [36]. P < 0.05 was considered statistically significant.

Validating the association between PGSHL/PGSCIO and CIO in the test cohort

Genomic QC was performed using PLINK v1.09 (Supplementary Fig. 2). Variants with missing data exceeding 5% and deviating from Hardy Weinberg Equilibrium (P < 1.0 × 10− 4) were removed, along with individuals exhibiting missing data exceeding 5%, related individuals (PI hat > 0.2), and those with mismatched sex. Imputation was performed using the TOPMed Imputation Server and the TOPMed r3 reference panel. Following imputation, all variants with R2 > 0.95 were included in the data to ensure the inclusion of high-quality imputed genotypes. After pruning for LD, and removing variants with minor allele frequencies < 0.01, PCA was performed using EIGENSOFT v5. Principal component clustering revealed that 69% of patients were of European descent.

Using PLINK v1.09, PGSs were generated for each individual in the test cohort for both the PGSHL and the PGSCIO. After scoring individuals, the distribution of PGSs were evaluated for normality using the Shapiro test. Deviation from normality was indicated by P < 0.05. To assess the association between PGSHL/PGSCIO and CIO, logistic regression was performed using the geneticriskR package [38] in R, including age at diagnosis, protocol (SJMB96 or SJMB03) [17], and the first ten ancestral principal components as covariates. These covariates were selected based on their previously reported significant associations with CIO [17]. Further, given the established role of CSI in ototoxicity [39], we incorporated an interaction term between CSI dose (< 25 Gy or ≥25 Gy) and PGS in our model. The ability of significant PGSs to distinguish cases (individuals with CIO) and controls (individuals without CIO) was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC) values. Lastly, the positive predictive value (PPV) and negative predictive value (NPV) were calculated to quantify the predictive accuracy of the PGS model. P < 0.05 was considered significant in all analyses.

Results

Generation of snRNA-seq data

To allow for biologically informed filtering of the PGSHL, we generated snRNA-seq from the inner ears of cisplatin-treated mice. After QC, we observed fewer cells in the cisplatin treatment group (replicate 1: n = 10,407 cells, replicate 2: n = 9,822) compared to the saline treatment control group (replicate 1: n = 13,212 cells, replicate 2: n = 12,724 cells). Closer investigation of the specific cells that are impacted by cisplatin treatment revealed that there was a decrease in type I neurons, osteoblasts, outer hair cells (OHC), spindle cells, macrophages, tympanic border cells, supporting cells, supporting structures 1 and 2, and osteocytes post-cisplatin treatment (MILO-R: LogFC<=-3, SpatialFDR < 0.1). Notably, macrophages were the only population of cells that demonstrated a significant increase in response to cisplatin treatment. Across the cochlear cell types that showed differential abundance, 159 differentially expressed genes (DEGs) were observed (DESeq2: Padj<0.05) (Supplementary Fig. 3).

Evaluation of the association between PGSHL/PGSCIO and CIO in the target cohort

A total of n = 2,370,365 non-ambiguous variants were included in PGSHL. Subsequent application of our refinement filter to PGSHL yielded PGSCIO, which contained n = 138,670 variants mapping to DEGs within cochlear cells that showed differential abundance. ReACt was then used to determine whether there was a significant association between the PGSs and CIO in the target PCL cohort. While the PGSHL was significantly associated with CIO in this cohort (P = 2.93 × 10− 3, R2 = 0.023), the PGSCIO was more significantly associated with CIO (P = 5.54 × 10− 5, R2 = 0.041).

Evaluation of the association between PGSHL/PGSCIO and CIO in the test cohort

Investigation of the distribution of PGSHL and PGSCIO confirmed that these scores were normally distributed in the SJMB cohort (PGSHL: P = 0.72; PGSCIO: P = 0.96) (Supplementary Fig. 4). Examination of the association between PGSHL/PGSCIO and CIO revealed that PGSHL was not significantly associated with CIO in the test cohort (P = 0.52, Nagelkerke R2 = 0.006), while PGSCIO demonstrated a significant association (P = 0.043, Nagelkerke R2 = 0.024). To establish baseline comparisons, we evaluated the independent predictive performance of a model including (i) only the PGSCIO, (ii) only clinical covariates (i.e., age at diagnosis, treatment protocol, CSI, and ancestral principal components), and (iii) the combined model, which incorporated both the PGSCIO and the relevant clinical covariates, including the interaction between CSI and PGSCIO. The model which included only PGSCIO yielded an AUC of 0.576 (95% CI: 0.503–0.650). Addition of PGSCIO to the model that included the relevant clinical covariates did not substantially increase the predictive performance of the clinical model (AUC of 0.732, 95% CI: 0.668–0.797 vs. AUC of 0.714, 95% CI: 0.648–0.78) (Supplementary Fig. 5). Examination of the PPV (81%) and NPV (48%) revealed that the predictive model, which includes both clinical and genetic variables, correctly identifies 81% of true positive CIO cases and 48% of true negative CIO cases.

Discussion

PGSs have shown considerable promise in various clinical contexts. For example, coronary artery disease (CAD) PGSs have identified a significantly larger number of at-risk individuals compared to methods relying on rare mutations [24, 40]. In addition, CAD PGSs have guided treatment decisions, such as the use of statins for patients classified as high-risk [24, 41]. Building on these successes, this is the first study to develop a hearing loss PGS that is specifically aimed at predicting CIO. By employing innovative approaches that integrate large-scale hearing loss GWAS data with cutting-edge snRNA-seq data, this study created a PGS that was significantly associated with CIO in two independent cohorts. This work has shown that the incorporation of omics datasets into PGS development pipelines can open new avenues for considering biological pathways underlying pharmacogenomic traits. Collectively, these novel methodologies offer a new approach for the development of pharmacogenomic PGSs.

The use of large-scale databases and summary statistics to overcome sample size and resource limitations associated with the development of PGS for CIO

As reviewed in detail by Johnson et al., previous studies have demonstrated the value of using large-scale GWAS of related phenotypes (e.g. schizophrenia GWAS) to create PGSs that can be used to predict pharmacogenomic traits (antipsychotic response) [42]. Given the observed overlap in the genetic architecture of hearing loss and CIO [26], we decided to build upon these approaches by using MTAG summary statistics from the UKB cohort. These summary statistics offer unique benefits as they provide information relating to associations across four genetically correlated hearing loss traits [43], thereby allowing for the identification of genetic variants of relevance across diverse hearing phenotypes. In addition, MTAG leverages the shared genetic architecture among these traits to refine effect size estimates, thereby enhancing our ability to detect associations, with these approaches being reported to increase the variance explained by the PGSs by as much as 25% [43]. Leveraging this approach, our study identified a significant association between the PGSHL and CIO in the PCL cohort. Further the use of summary statistics for both the base and target cohorts represents a cost- and resource-effective method for the preliminary evaluation of the performance of PGSs. Given the challenges related to data-sharing practices in pharmacogenomics research [37], this method provides researchers with a unique opportunity to assess the feasibility and robustness of PGSs for future investigation, even when individual-level data is not readily available.

The use of snRNA-seq data to increase the relevance of a hearing loss PGS to CIO

The use of large-scale GWAS data from related phenotypes has been shown to be a useful approach for the development of pharmacogenomic PGS [42]. However, not all variants uncovered from these GWASs are likely to be relevant to their drug-induced counterpart phenotypes. Therefore, to enhance the relevance of the PGSHL to CIO, we generated murine inner ear snRNA-seq data obtained from cisplatin-treated mice. By using snRNA-seq data to apply a pharmacologically relevant filter to our PGSHL, we were able to selectively include variants from the PGSHL that map to DEGs within cochlear cells that showed differential abundance. By enhancing the biological relevance of PGSHL to CIO, we developed a PGS (PGSCIO) with improved predictive capacity. Notably, evaluation of the performance of PGSHL and PGSCIO using summary statistics from the PCL cohort revealed that PGSCIO (P = 5.54 × 10− 5) outperforms PGSHL (P = 2.93 × 10− 3) in predicting CIO risk. This observation was replicated in the SJMB cohort, where PGSCIO was significantly associated with CIO (P = 0.043), while PGSHL was not (P = 0.52).

Uncovering cells and pathways of relevance to CIO

While initial evaluations of the PGSs using summary statistics from the PCL cohort provided valuable insights, further validation of PGSCIO using individual-level genotype and clinical data was essential to confirm the reliability of these findings. Subsequent analyses revealed that while the PGSCIO was significantly associated with CIO in the SJMB cohort (P = 0.043), the inclusion of this score in a logistic regression model resulted in only a marginal improvement in predictive performance (AUC = 0.732) when compared to a model built with existing clinical/demographic predictors (age at diagnosis, CSI, protocol, and ancestral principal components) (AUC = 0.714). Similarly, while PGSCIO accounted for more of the variance observed for CIO (Nagelkerke R2 = 0.024), compared to PGSHL (Nagelkerke R2 = 0.006), highlighting the value of our filtering approach, the variance explained was still relatively low. Consequently, the current clinical utility of this score remains limited. Nonetheless, by integrating human CIO GWAS data with murine snRNA-seq data, we have gained valuable insights into potential biological mechanisms underlying CIO.

Analysis of the snRNAseq data generated by this study revealed that cisplatin treatment led to a decrease in the abundance of several cell types within the inner ear, as well as an increase in the abundance of certain clusters of macrophages. The identification of cochlear cells that exhibit differential abundance in the snRNA-seq data, including specialized auditory cells (spindle cells, OHCs and type I neurons), supporting cells (supporting cells, supporting structures 1, supporting structures 2 and tympanic border cells), immune cells (macrophages), and bone cells (osteoblasts and osteocytes), allowed us to pinpoint specific cell types that are impacted by cisplatin and may therefore be important in CIO. The stria vascularis, organ of Corti, and spiral ganglion have previously been implicated in CIO [28], with spindle cells, OHCs, and type I neurons residing in these respective structures. Although the role of supporting cells in CIO remains poorly understood, the observed decrease in abundance of these cells post-cisplatin treatment highlights the need to further investigate how these cells contribute to the development of CIO. In addition, it has been previously reported that cisplatin triggers inflammatory pathways, leading to the release of pro-inflammatory cytokines and the recruitment of immune cells such as macrophages [12], which is reflected by the observed increased abundance of macrophages post-cisplatin treatment. This inflammation can cause tissue damage and hearing loss [12] by compromising the structures responsible for auditory function. Macrophages, which are widely distributed throughout the cochlea, play a significant role in mediating this inflammatory response [14]. Lastly, cisplatin binds extensively to type I collagen in bone, creating a reservoir that gradually releases platinum, potentially leading to ototoxic effects [28]. By elucidating the contributions of these specific cochlear cells, our study significantly advances the understanding of CIO and offers novel insights into its underlying mechanisms, opening future avenues of research.

Study limitations

While our study has yielded valuable insights relating to the genetics underlying CIO, it is important to acknowledge several limitations which may contribute to the limited clinical utility of this score. First, given the potential incomplete overlap in the underlying genetic architecture between hearing loss and CIO, hearing loss GWAS may not capture all genetic factors that contribute to CIO risk. Although the PGSHL exhibited a significant association with CIO in the PCL cohort, the use of a future large-scale CIO GWAS as the base cohort for PGS development could provide deeper insights into the genetic determinants of CIO and enhance the accuracy and clinical applicability of this score. Further, many of the variants included in the initial PGSHL mapped to non-coding regions. Therefore, future studies should explore the integration of single nuclei multi-omics data, which includes single-nuclei assay for transposase-accessible chromatin sequencing (snATAC-seq) data, to allow for the annotation and inclusion of intergenic variants in the biologically informed PGS. Our study also recognizes that while the genomes of humans and mice are genetically similar, with approximately 90% conserved regions [44] and an 85% overlap in protein-coding regions [45], they are not identical, and the importance of age-related differences between mouse and human samples should also be acknowledged. Future studies should, therefore, investigate gene expression changes in older murine models, which align more closely with the age demographics of our human samples.

Importantly, the data used in this study included individuals who were predominantly of European descent. Consequently, it is unknown whether the results from this study are transferable across global populations. Therefore, it is crucial to include individuals with non-European ancestries in future studies. Recognizing this limitation, current initiatives such as the All of Us Program and Our Future Health are focusing on assembling larger and more diverse biobanks [46]. Using these diverse datasets in future research will enhance our ability to ensure that PGSs are applicable to a broader range of individuals. Finally, inherent differences in patient demographics and variations in the definition and measurement of CIO between the PCL and SJMB cohorts can have unknown consequences for PGSCIO [47]. The PCL cohort consists of childhood cancer survivors treated with or without carboplatin, alongside cisplatin, while the SJMB cohort focuses on medulloblastoma patients treated with cisplatin and CSI. Additionally, the PCL and SJMB cohorts employed different classification systems for CIO: the Muenster and Chang grading scales, respectively. This may lead to inconsistent classification of CIO cases, potentially impacting comparisons of PGSCIO performance across these cohorts.

Conclusion

Our study holds significant implications for understanding the genetics underlying CIO. By demonstrating the feasibility of integrating large-scale datasets, such as the hearing loss MTAG summary statistics data from the UKB cohort, with omics information, such as murine inner ear snRNA-seq data, we provide a novel approach for developing pharmacogenomic PGSs. Further, our identification of specific cochlear cell types involved in CIO sheds light on the intricate cellular pathways and processes underlying this ADR. This novel information will lay the groundwork for future research aimed at developing targeted therapeutic interventions to mitigate CIO risk. Overall, our study underscores the importance of interdisciplinary approaches to unravel the complexities of CIO and contributes to the growing body of literature that is improving our understanding of the etiology of this ADR.

Data availability

Data used in this study: MTAG summary statistics data for the base cohort was retrieved from the NeMo Archive (https://data.nemoarchive.org/other/grant/sament/sament/hearing_gwas/) on 26 March 2020. The summary statistics from the CIO GWAS were downloaded from the EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/publications/34262104#study_panel) on 18 May 2021. The LD reference panel for SBayesR was downloaded from https://zenodo.org/record/3375373#.YYQZEL3MLfa on 7 June 2022.

Code

Scripts that were used in this study are available on https://github.com/Drogemoller-Lab/Using-Genomics-to-Predict-and-Prevent-Cisplatin-Induced-Otoxicity, https://github.com/Drogemoller-Lab/CIO-PGS, and https://github.com/Drogemoller-Lab/scRNA-pipeline-UofM.

References

  1. Brown A, Kumar S, Tchounwou PB. Cisplatin-based chemotherapy of human cancers. J Cancer Sci Ther. 2019;11(4):97.

    PubMed  PubMed Central  Google Scholar 

  2. Han Y, Wen P, Li J, Kataoka K. Targeted nanomedicine in cisplatin-based cancer therapeutics. J Controlled Release. 2022;345:709–20.

    Article  CAS  Google Scholar 

  3. Einhorn LH. Treatment of testicular cancer: a new and improved model. J Clin Oncol. 1990;8(11):1777–81.

    Article  CAS  PubMed  Google Scholar 

  4. Rajput K, Edwards L, Brock P, Abiodun A, Simpkin P, Al-Malky G. Ototoxicity-induced hearing loss and quality of life in survivors of paediatric cancer. Int J Pediatr Otorhinolaryngol. 2020;138:110401.

    Article  PubMed  Google Scholar 

  5. Chattaraj A, Syed MP, Low CA, Owonikoko TK. Cisplatin-induced ototoxicity. A concise review of the b, prevention, and interception strategies. JCO Oncol Pract. 2023;19(5):278–83.

  6. Sheth S, Mukherjea D, Rybak LP, Ramkumar V. Mechanisms of Cisplatin-Induced Ototoxicity and Otoprotection. Front Cell Neurosci. 2017;11:338.

    Article  PubMed  PubMed Central  Google Scholar 

  7. Karasawa T, Steyger PS. An integrated view of cisplatin-induced nephrotoxicity and ototoxicity. Toxicol Lett. 2015;237(3):219–27.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  8. Marullo R, Werner E, Degtyareva N, Moore B, Altavilla G, Ramalingam SS, et al. Cisplatin induces a Mitochondrial-ROS response that contributes to cytotoxicity depending on mitochondrial redox status and bioenergetic functions. PLoS ONE. 2013;8(11):e81162.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Devarajan P, Savoca M, Castaneda MP, Park MS, Esteban-Cruciani N, Kalinec G, et al. Cisplatin-induced apoptosis in auditory cells: role of death receptor and mitochondrial pathways. Hear Res. 2002;174(1):45–54.

    Article  CAS  PubMed  Google Scholar 

  10. Mukherjea D, Whitworth CA, Nandish S, Dunaway GA, Rybak LP, Ramkumar V. Expression of the kidney injury molecule 1 in the rat cochlea and induction by cisplatin. Neuroscience. 2006;139(2):733–40.

    Article  CAS  PubMed  Google Scholar 

  11. Lynch ED, Gu R, Pierce C, Kil J. Reduction of acute cisplatin ototoxicity and nephrotoxicity in rats by oral administration of allopurinol and ebselen. Hear Res. 2005;201(1):81–9.

    Article  CAS  PubMed  Google Scholar 

  12. Ramkumar V, Mukherjea D, Dhukhwa A, Rybak LP. Oxidative stress and inflammation caused by Cisplatin Ototoxicity. Antioxidants. 2021;10(12):1919.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Li Y, Womer RB, Silber JH. Predicting cisplatin ototoxicity in children: the influence of age and the cumulative dose. Eur J Cancer. 2004;40(16):2445–51.

  14. Tan WJT, Vlajkovic SM. Molecular characteristics of Cisplatin-Induced ototoxicity and therapeutic interventions. Int J Mol Sci. 2023;24(22):16545.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Meijer AJM, Diepstraten FA, van den Heuvel-Eibrink MM, Bleyer A. Prevention of cisplatin-induced hearing loss in children: achievements and challenges for evidence-based implementation of sodium thiosulfate. Front Oncol. 2024;14:1336714.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  16. Chopra D, Rehan HS, Sharma V, Mishra R. Chemotherapy-induced adverse drug reactions in oncology patients: a prospective observational survey. Indian J Med Paediatr Oncol off J Indian Soc Med Paediatr Oncol. 2016;37(1):42–6.

    Article  Google Scholar 

  17. Xu H, Robinson GW, Huang J, Lim JYS, Zhang H, Bass JK, et al. Common variants in ACYP2 influence susceptibility to cisplatin-induced hearing loss. Nat Genet. 2015;47(3):263–6.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  18. Dille MF, Wilmington D, McMillan GP, Helt W, Fausti SA, Konrad-Martin D. Development and validation of a cisplatin dose-ototoxicity model. J Am Acad Audiol. 2012;23(7):510–21.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Deutsch BC, Collopy C, Kallogjeri D, Piccirillo JF. Validation of hearing loss Prediction Tool for Cisplatin Chemotherapy and Radiation in Head and Neck Cancer Treatment. JAMA Otolaryngol Neck Surg. 2021;147(2):182–9.

    Article  Google Scholar 

  20. Dolan ME, Newbold KG, Nagasubramanian R, Wu X, Ratain MJ, Cook EH, et al. Heritability and linkage analysis of sensitivity to cisplatin-induced cytotoxicity. Cancer Res. 2004;64(12):4353–6.

    Article  CAS  PubMed  Google Scholar 

  21. Thiesen S, Yin P, Jorgensen AL, Zhang JE, Manzo V, McEvoy L, et al. TPMT, COMT and ACYP2 genetic variants in paediatric cancer patients with cisplatin-induced ototoxicity. Pharmacogenet Genomics. 2017;27(6):213–22.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  22. Drögemöller BI, Brooks B, Critchley C, Monzon JG, Wright GEB, Liu G, et al. Further investigation of the role of ACYP2 and WFS1 pharmacogenomic variants in the Development of Cisplatin-Induced Ototoxicity in Testicular Cancer patients. Clin Cancer Res. 2018;24(8):1866–71.

    Article  PubMed  Google Scholar 

  23. Drögemöller BI, Wright GEB, Lo C, Le T, Brooks B, Bhavsar AP, et al. Pharmacogenomics of Cisplatin-Induced Ototoxicity: successes, shortcomings, and future avenues of Research. Clin Pharmacol Ther. 2019;106(2):350–9.

    Article  PubMed  Google Scholar 

  24. Wilke M. Using genomics to guide the development of strategies to predict and prevent cisplatin-induced ototoxicity. 2022 Dec 29 [cited 2023 Nov 15]; http://hdl.handle.net/1993/37057

  25. Kalra G, Milon B, Casella AM, Herb BR, Humphries E, Song Y, et al. Biological insights from multi-omic analysis of 31 genomic risk loci for adult hearing difficulty. PLoS Genet. 2020;16(9):e1009025.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  26. Wheeler HE, Gamazon ER, Frisina RD, Perez-Cervantes C, El Charif O, Mapes B, et al. Variants in WFS1 and other mendelian deafness genes are Associated with Cisplatin-Associated Ototoxicity. Clin Cancer Res off J Am Assoc Cancer Res. 2017;23(13):3325–33.

    Article  CAS  Google Scholar 

  27. Meijer AJM, Diepstraten FA, Langer T, Broer L, Domingo IK, Clemens E, et al. TCERG1L allelic variation is associated with cisplatin-induced hearing loss in childhood cancer, a PanCareLIFE study. NPJ Precis Oncol. 2021;5(1):64.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Breglio AM, Rusheen AE, Shide ED, Fernandez KA, Spielbauer KK, McLachlin KM, et al. Cisplatin is retained in the cochlea indefinitely following chemotherapy. Nat Commun. 2017;8(1):1654.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Yimit A, Adebali O, Sancar A, Jiang Y. Differential damage and repair of DNA-adducts induced by anti-cancer drug cisplatin across mouse organs. Nat Commun. 2019;10(1):309.

    Article  PubMed  PubMed Central  Google Scholar 

  30. Noda T, Meas SJ, Nogami J, Amemiya Y, Uchi R, Ohkawa Y, et al. Direct reprogramming of spiral ganglion non-neuronal cells into neurons: toward ameliorating Sensorineural hearing loss by Gene Therapy. Front Cell Dev Biol. 2018;6:16.

    Article  PubMed  PubMed Central  Google Scholar 

  31. Jean P, Wong Jun Tai F, Singh-Estivalet A, Lelli A, Scandola C, Megharba S et al. Single-cell transcriptomic profiling of the mouse cochlea: an atlas for targeted therapies. Proc Natl Acad Sci U S A 120(26):e2221744120.

  32. GitHub - hakyimlab. /summary-gwas-imputation: harmonization, liftover, and imputation of summary statistics from GWAS [Internet]. [cited 2024 Mar 11]. https://github.com/hakyimlab/summary-gwas-imputation

  33. Lloyd-Jones LR, Zeng J, Sidorenko J, Yengo L, Moser G, Kemper KE, et al. Improved polygenic prediction by bayesian multiple regression on summary statistics. Nat Commun. 2019;10:5086.

    Article  PubMed  PubMed Central  Google Scholar 

  34. Dann E, Henderson NC, Teichmann SA, Morgan MD, Marioni JC. Differential abundance testing on single-cell data using k-nearest neighbor graphs. Nat Biotechnol. 2022;40(2):245–53.

    Article  CAS  PubMed  Google Scholar 

  35. Love MI, Huber W, Anders S. Moderated estimation of Fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550.

    Article  PubMed  PubMed Central  Google Scholar 

  36. Yang Z, Paschou P, Drineas P. Reconstructing SNP allele and genotype frequencies from GWAS summary statistics. Sci Rep. 2022;12(1):8242.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  37. Miao DNR, Ladha F, Lyle SM, Olivier DW, Ahmed S, Drögemöller BI. Current Perspectives on Data Sharing and Open Science in Pharmacogenomics. Clin Pharmacol Ther [Internet]. [cited 2024 Feb 2];n/a(n/a). https://onlinelibrary.wiley.com/doi/abs/https://doi.org/10.1002/cpt.3115

  38. Ballou A. aballou16/geneticriskR [Internet]. 2023 [cited 2024 Mar 14]. https://github.com/aballou16/geneticriskR

  39. Tserga E, Nandwani T, Edvall NK, Bulla J, Patel P, Canlon B, et al. The genetic vulnerability to cisplatin ototoxicity: a systematic review. Sci Rep. 2019;9:3455.

    Article  PubMed  PubMed Central  Google Scholar 

  40. Khera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat Genet. 2018;50(9):1219–24.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Kullo IJ, Jouni H, Austin EE, Brown SA, Kruisselbrink TM, Isseh IN, et al. Incorporating a genetic risk score into Coronary Heart Disease Risk estimates: effect on LDL Cholesterol Levels (the MIGENES Clinical Trial). Circulation. 2016;133(12):1181–8.

    Article  PubMed  PubMed Central  Google Scholar 

  42. Johnson D, Wilke MAP, Lyle SM, Kowalec K, Jorgensen A, Wright GEB, et al. A systematic review and analysis of the Use of Polygenic scores in Pharmacogenomics. Clin Pharmacol Ther. 2022;111(4):919–30.

    Article  PubMed  Google Scholar 

  43. Turley P, Walters RK, Maghzian O, Okbay A, Lee JJ, Fontana MA, et al. Multi-trait analysis of genome-wide association summary statistics using MTAG. Nat Genet. 2018;50(2):229–37.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. Breschi A, Gingeras TR, Guigó R. Comparative transcriptomics in human and mouse. Nat Rev Genet. 2017;18(7):425–40.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  45. Why Mouse Matters [Internet]. [cited 2024 Jun 25]. https://www.genome.gov/10001345/importance-of-mouse-genome

  46. Xiang R, Kelemen M, Xu Y, Harris LW, Parkinson H, Inouye M, et al. Recent advances in polygenic scores: translation, equitability, methods and FAIR tools. Genome Med. 2024;16(1):33.

    Article  PubMed  PubMed Central  Google Scholar 

  47. Addressing the challenges of polygenic scores in human genetic research - ScienceDirect [Internet]. [cited 2024 May 3]. https://www.sciencedirect.com/science/article/pii/S0002929722004608

Download references

Author information

Authors and Affiliations

  1. Department of Biochemistry and Medical Genetics, Rady Faculty of Health Sciences, University of Manitoba, Winnipeg, MB, Canada

    Deanne Nixie R. Miao, MacKenzie A. P. Wilke, John Pham, Feryal Ladha, Mansumeet Singh & Britt I. Drögemöller

  2. Department of Internal Medicine, Max Rady College of Medicine, University of Manitoba, Winnipeg, MB, Canada

    Janilyn Arsenio

  3. Sunnybrook Research Institute, Toronto, ON, Canada

    Emilia Luca & Alain Dabdoub

  4. Department of Pharmacy and Pharmaceutical Sciences, St. Jude Children’s Research Hospital, Memphis, TN, 38105, USA

    Wejian Yang & Jun J. Yang

  5. Department of Oncology, St. Jude Children’s Research Hospital, Memphis, TN, 38105, USA

    Jun J. Yang

  6. Hematological Malignancies Program, St. Jude Children’s Research Hospital, Memphis, TN, 38105, USA

    Jun J. Yang

  7. CancerCare Manitoba Research Institute, Winnipeg, MB, Canada

    Britt I. Drögemöller

  8. Children’s Hospital Research Institute of Manitoba, Winnipeg, MB, Canada

    Britt I. Drögemöller

  9. Centre of Aging, University of Manitoba, Winnipeg, MB, Canada

    Britt I. Drögemöller

Authors
  1. Deanne Nixie R. Miao
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. MacKenzie A. P. Wilke
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. John Pham
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Feryal Ladha
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Mansumeet Singh
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Janilyn Arsenio
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Emilia Luca
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Alain Dabdoub
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Wejian Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Jun J. Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Britt I. Drögemöller
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

D.N.M , J.A., E.L. and A.D generated/provided training relating to the generation of the single nuclei data. W.Y and J.J.Y generated the genotype data. D.N.M, M.A.W, M.S and F.L performed analyses related to the polygenic scores. D.N.M and J.P performed analyses relating to the single nuclei data. B.I.D conceived the project and provided supervision. D.N.M wrote the first draft of the manuscript. All authors reviewed the manuscript.

Corresponding author

Correspondence to Britt I. Drögemöller.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Miao, D.N.R., Wilke, M.A.P., Pham, J. et al. Leveraging large-scale datasets and single cell omics data to develop a polygenic score for cisplatin-induced ototoxicity. Hum Genomics 18, 112 (2024). https://doi.org/10.1186/s40246-024-00679-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s40246-024-00679-5

Keywords

Human Genomics

ISSN: 1479-7364