Polygenic risk of paclitaxel-induced peripheral neuropathy: a genome-wide association study

Paclitaxel is a highly active chemotherapeutic agent used widely in the treatment of solid tumours [1]. However, chemotherapy-induced peripheral neurotoxicity (CIPN) is a major dose-limiting neurological side-effect of paclitaxel treatment that can persist long-term [2]. CIPN produces sensory and functional abnormalities leading to difficulties with fine motor and balance tasks, increased falls risk, and reduced quality of life [3]. Further, CIPN is a common cause of dose reduction and premature discontinuation, potentially affecting survival outcomes [4]. There are currently no neuroprotective measures to prevent the development of CIPN and no effective treatment options [5]. Importantly, understanding the mechanisms underlying CIPN and identifying which patients are most at-risk are critical to preventing long-term sequelae of treatment with paclitaxel.

Mechanistically, paclitaxel targets microtubules, inhibiting the dynamic assembly and disassembly of β-tubulin, leading to their stabilisation, cell-cycle arrest, and cell death [1]. While this mechanism has been proposed to produce neurotoxicity via disruption to axonal transport [6], growing evidence suggests a range of additional mechanisms, including disruption of neuronal cell metabolism, mitochondrial dysfunction, oxidative stress and neuroinflammation as underlying the development of CIPN [7]. Better understanding of underlying mechanisms of CIPN will be critical to the development of successful preventative and treatment strategies.

There have been substantial efforts to identify genetic profiles associated with heightened CIPN risk, with a range of single nucleotide polymorphisms (SNPs) in genes associated with neural development and structure, drug metabolism and neural repair associated with paclitaxel-induced CIPN [8]. However, there has been limited replication between studies, and there remains a lack of validated genetic associations with paclitaxel-induced CIPN. A key limitation is the lack of consensus regarding appropriate CIPN outcome measures [9]. It has been well documented that patients report greater neuropathy severity than clinicians [10] and that patient-reported outcomes and clinician-reported outcomes provide complementary but different information about CIPN [11]. However, despite this, there has been limited studies incorporating multimodal CIPN outcome measures with comprehensive phenotyping and patient-reported outcome measures.

Ultimately, identification and validation of genetic pathways involved in CIPN will enable characterisation of patients at-risk of significant, persistent toxicity. CIPN risk likely incorporates multiple genes [7] and polygenic models will be required to explain variability in CIPN outcomes rather than reliance on single SNPs. However, such models need to be developed in appropriately phenotyped cohorts. In the present study, we utilised comprehensive CIPN assessment and phenotyping using multiple assessment tools combined with genome-wide association studies (GWAS) and pathway analysis to provide an improved understanding of the genetic variants contributing to paclitaxel-induced CIPN.

We performed a GWAS on 183 paclitaxel-treated patients with comprehensive neuropathy phenotyping. Bioinformatic analyses, including pathway enrichment and polygenic risk score analysis, were used to identify mechanistic pathways of interest.

Chemotherapy-induced peripheral neurotoxicity is a significant adverse event of paclitaxel treatment that can lead to early treatment discontinuation, persistent functional disability and reduced quality of life [2, 3]. In this study, we performed a GWAS on 183 patients treated with paclitaxel to identify genetic variants associated with CIPN, as measured using multiple neuropathy outcome measures. We identified multiple SNPs with genome-wide significance associated with patient-reported neuropathy. Pathways analysis was used to identify mechanistic pathways involved in CIPN and a polygenic risk score was determined. Importantly, our findings highlight the potential role of axon development and regeneration pathways in paclitaxel-induced CIPN.

Our study identified 4 chromosomal regions (rs9846958, nearest gene GADL1 on chromosome 3; rs117158921, nearest gene AC124254.2 on chromosome 18; rs4560447, nearest gene LIMCH1 on chromosome 4; rs200091415, nearest gene FAM238B on chromosome 19) that passed genome-wide significance in the patient-reported neuropathy GWAS (P < 5 × 10^–8; Table 2). Prior GWAS (E5103 (26), CALGB 40101 [25, 27, 28] and a meta-analysis of two GWAS studies (CALGB 40502 and CALGB 40101[20]) on patients treated with paclitaxel have identified a range of SNPs associated with neuropathy, but none exceeded genome-wide significance. In our study, the potential impact of top associated variants on the function of non-coding RNAs was highlighted by VEP annotation (Fig. 1B, Additional file 2: Table S1). This is consistent with results from a transcriptomic study that identified dysregulation of long non-coding RNAs and mRNAs mediating neuroinflammation and pain in the spinal cord of a rat model of paclitaxel-induced peripheral neuropathy [30].

While functional annotations have traditionally focused on known genes, thousands of disease-associated SNPs are located within intergenic regions, making it difficult to understand their association with disease phenotypes. Recent analyses found that non-coding disease associated SNPs were frequently located in or approximate to regulatory elements, such as the binding sites for CCCTC-binding factors (CTCF) and enhancer elements that act distally to promote gene expression [31]. In our annotation of the top 54 associated SNPs (Additional file 2: Table S1), 4 were located within these regulatory elements. CADD scores are based on various genomic features derived from surrounding nucleotide sequences, gene model annotations, evolutionary constraints, epigenetic marks and functional predictions [32]. We observed that 2 intergenic SNPs had CADD scores greater than 10, that is they were ranked in the top 10% of all known variants likely to be deleterious (Additional file 2: Table S1). We also note that one of the top associated SNP (rs9846958) on chromosome 3 would be considered to be an intergenic SNP and currently lacks any functional annotation, but we have indicated the closet gene to be GADL1 (Fig. 1A).

In the present study, of the 4 SNPs with genome-wide significance, the associated gene LIMCH1 was most prominently expressed in the DRG, a key region implicated in CIPN pathogenesis. LIMCH1 has been identified as a key regulator of actin-cytoskeleton remodelling, involved in cell migration [22]. Due to its role in cell migration and adhesion, LIMCH1 has been associated with worse prognosis in multiple forms of cancer [33]. While LIMCH1 has not been directly associated with nerve function, actin-cytoskeletal frameworks are critical in neuronal development, and axonal growth and actin-binding LIM domain proteins are important in axonal regeneration [34]. Another actin-binding protein LIMK2, which acts to regulate cell proliferation and migration, has also been linked to paclitaxel-induced CIPN in a prior GWAS [27].

Further, there is substantial evidence highlighting the potential role of actin cytoskeleton and axonal guidance pathways in paclitaxel-induced CIPN [35]. Comparison of differences in signalling pathways and gene co-expression between paclitaxel-treated patients with and without CIPN provided molecular evidence of the involvement of cytoskeletal and axonal morphology pathways in neuropathy development [35]. This included a suite of genes previously associated with paclitaxel-induced neuropathy, including the EPHA gene family linked to receptors for axonal grown and neural development [27, 28] and FDG4, a F-actin binding protein [29]. Of note, although no candidate variants were independently replicated in our GWAS dataset, there was some support for EPHA5 (genotyped SNP rs3605041, P = 0.0021 for TNSc-GWAS), which encodes an ephrin receptor important in neurite growth during development [7].

In further support of the importance of axonal and cytoskeletal development pathways in paclitaxel-induced CIPN, a key gene-ontology pathway of interest from our analysis across multiple outcome measures was the axon development pathway. This underscores the results of previous analyses, which have highlighted this pathway as central to paclitaxel-induced PN [7, 36]. Consistent with our findings (Table 4), differential gene expression and pathway impact analysis identified significantly perturbed cytoskeleton- and axon morphology-related signalling pathways in patients treated with paclitaxel [35]. These pathways have recently been highlighted in conjunction with their links to Ras homolog family of guanosine triphosphate hydrolase (RhoGTPase) signalling pathways relevant to axon extension and cell mobility [7]. RhoGTPases are important in sensory neuronal development and outgrowth as well as axonal regeneration [37] and are linked to paclitaxel-induced PN development, including via LIM domain proteins [37].

Although our study has identified several variants with genome-wide significance, we did not independently replicate the findings of prior studies in our dataset, given the number of candidate variants examined. However, the top replicated variant was rs9332998 from the CIPN20-GWAS, a proxy for rs4646487 within the CYP4B1 gene. The gene is part of the CYP genes set that modulate paclitaxel pharmacokinetics and similar genes have been associated with paclitaxel-induced CIPN in prior analyses [9]. Replication studies have often failed to confirm genetic associations in CIPN, potentially related to a lack of standardisation in outcome measures, with different thresholds for CIPN case identification affecting findings [9].

We utilised multiple CIPN outcome measures, including patient reported symptoms, clinical grading scale and neurological assessment. While there remains no gold standard CIPN assessment tool, evidence suggests that multimodal CIPN assessment incorporating both patient report and clinician assessment may present the most comprehensive information about neuropathy status [11]. However, only a minority of prior genetic risk factor studies have utilised patient-reported outcomes [38]. Importantly, patients typically report greater severity of symptoms than reported by clinicians [10] and this has been demonstrated to affect the identification of genetic risk factors for paclitaxel-induced CIPN [38]. Conversely, there has been criticism of relying solely on patient-reported CIPN assessment for biomarker studies, as patient report may be more variable and lack a consistent benchmark of severity compared to clinical assessment [39]. Of note, in this study, SNPs with genome-wide significance were only identified in the GWAS using patient reported CIPN. This may be related to the sensitivity of patient-reported outcomes for neuropathy but may also reflect limitations in more objective outcomes which do not always match with patient report [11].

Another factor that complicates the search for genetic variants associated with paclitaxel-induced PN is the likely polygenic inheritance, with multiple variants contributing to the risk of PN [36, 40]. It is likely that a large number of SNPs each contribute a small, additive risk to the development of paclitaxel-induced PN [36, 40]. Importantly, the use of polygenic risk scores (PRS) which aggregate the effects of multiple genetic variants across the human genome into a single score, have recently been shown to have predictive value for multiple common diseases such as breast cancer and diabetes [41]. Further, the integration of genetic information with non-genetic risk factors has been demonstrated to enhance the sensitivity and specificity of PRS as a clinical tool [42]. In our dataset, a PRS calculated from 46 SNPs was highly correlated with patient-reported CIPN (Fig. 1C). Our PRS differs from scores calculated for idiopathic neurodegenerative diseases such as Alzheimer’s disease [43] which typically require > 100,000 SNPs and have poorer predictive values with r² < 0.1. This may reflect the fact that pharmacogenomic variants typically have stronger genetic effects compared with common disease-associated variants [44]. We also note that our PRS is calculated from patients of European descent, and validation of our PRS by other investigators should involve controlling for population stratification. This is especially important as the rate of severe CIPN may vary by ethnicity [9, 25]. Nonetheless, such an approach is likely to be beneficial for the prediction of CIPN and should form the basis for future genetic analyses of CIPN.

In conclusion, we have identified novel genetic loci associated with patient-reported paclitaxel- induced peripheral neuropathy and these findings provide further evidence for the involvement of axon development pathways in paclitaxel-induced CIPN. Our study highlights the importance of appropriate and patient-relevant CIPN outcome measures in defining the CIPN phenotype. In total, this study highlights the polygenic nature of CIPN risk, as definition of polygenic patterns of inheritance will be critical to ultimately enable genetic risk factors to become useful tools to predict patient risk in the clinic and improve patient quality of life following paclitaxel treatment.