CYP3A genetic variation and taxane-induced peripheral neuropathy: a systematic review, meta-analysis, and candidate gene study

Background: Taxane-induced peripheral neuropathy (TIPN) is an important cause of premature treatment cessation and dose-limitation in cancer therapy. It also reduces quality of life and survivorship in affected patients. Genetic polymorphisms in the CYP3A family have been investigated but the findings have been inconsistent and contradictory. Methods: A systematic review identified 12 pharmacogenetic studies investigating genetic variation in CYP3A4*22 and CYP3A5*3 and TIPN. In our candidate gene study, 288 eligible participants (211 taxane participants receiving docetaxel or paclitaxel, and 77 control participants receiving oxaliplatin) were successfully genotyped for CYP3A4*22 and CYP3A5*3. Genotyping data was transformed into a combined CYP3A metaboliser phenotype: Poor metabolisers, intermediate metabolisers and extensive metabolisers. Individual genotypes and combined CYP3A metaboliser phenotypes were assessed in relation to neurotoxicity, including by meta-analysis where possible. Results: In the systematic review, no significant association was found between CYP3A5*3 and TIPN in seven studies, with one study reporting a protective association. For CYP3A4*22, one study has reported an association with TIPN, while four other studies failed to show an association. Evaluation of our patient cohort showed that paclitaxel was found to be more neurotoxic than docetaxel (p < 0.001). Diabetes was also significantly associated with the development of TIPN. The candidate gene analysis showed no significant association between either SNP (CYP3A5*3 and CYP3A4*22) and the development of TIPN overall, or severe TIPN. Meta-analysis showed no association between these two variants and TIPN. Transformed into combined CYP3A metaboliser phenotypes, 30 taxane recipients were poor metabolisers, 159 were intermediate metabolisers, and 22 were extensive metabolisers. No significant association was observed between metaboliser status and case-control status. Summary: We have shown that the risk of peripheral neuropathy during taxane chemotherapy is greater in patients who have diabetes. CYP3A genotype or phenotype was not identified as a risk factor in either the candidate gene analysis or the systematic review/meta-analysis, although we cannot exclude the possibility of a minor contribution, which would require a larger sample size.

Background: Taxane-induced peripheral neuropathy (TIPN) is an important cause of premature treatment cessation and dose-limitation in cancer therapy. It also reduces quality of life and survivorship in affected patients. Genetic polymorphisms in the CYP3A family have been investigated but the findings have been inconsistent and contradictory.
Methods: A systematic review identified 12 pharmacogenetic studies investigating genetic variation in CYP3A4*22 and CYP3A5*3 and TIPN. In our candidate gene study, 288 eligible participants (211 taxane participants receiving docetaxel or paclitaxel, and 77 control participants receiving oxaliplatin) were successfully genotyped for CYP3A4*22 and CYP3A5*3. Genotyping data was transformed into a combined CYP3A metaboliser phenotype: Poor metabolisers, intermediate metabolisers and extensive metabolisers. Individual genotypes and combined CYP3A metaboliser phenotypes were assessed in relation to neurotoxicity, including by meta-analysis where possible.
Results: In the systematic review, no significant association was found between CYP3A5*3 and TIPN in seven studies, with one study reporting a protective association. For CYP3A4*22, one study has reported an association with TIPN, while four other studies failed to show an association. Evaluation of our patient cohort showed that paclitaxel was found to be more neurotoxic than docetaxel (p < 0.001). Diabetes was also significantly associated with the development of TIPN. The candidate gene analysis showed no significant association between either SNP (CYP3A5*3 and CYP3A4*22) and the development of TIPN overall, or severe TIPN. Meta-analysis showed no association between these two variants and TIPN. Transformed into combined CYP3A metaboliser phenotypes, 30 taxane
TIPN symptoms generally manifest within several weeks of taxane commencement, though both docetaxel and paclitaxel may incite acute neuropathic pain in the first week following the initial dose (Loprinzi et al., 2007;Loprinzi et al., 2011;Reeves et al., 2012;Tanabe et al., 2013;Fernandes et al., 2016;Starobova and Vetter, 2017). Classically, TIPN manifests in a 'glove-and-stocking' distribution, starting in the fingers and toes and may progress to the hand/wrist and lower leg. Sensory symptoms are most common and include numbness and tingling, dysaesthesia and paraesthesia, and may progress to painful and burning sensations. Motor symptoms are less common, manifesting as muscle weakness, and impaired fine motor movement, and are usually associated with higher doses (Chaudhry et al., 1994;Freilich et al., 1996). Motor impairment can result in significant loss of functional abilities including balance (Tofthagen et al., 2013;Miltenburg and Boogerd, 2014;Ibrahim and Ehrlich, 2020), with potential for mobility-related disability (Hile et al., 2010). Cranial nerve involvement has been reported anecdotally (Velasco and Bruna, 2015). Uncommon autonomic symptoms include digestive, sexual, and urinary disruption (Mols et al., 2016;Zajaczkowska et al., 2019). Development of TIPN is an indication for dose reduction or discontinuation, both resulting in sub-optimal chemotherapy regimens with potential to affect cancer treatment outcomes and overall survival.
Despite recent developments in identifying predictive and blood biomarkers of PN using ultrasensitive protein assays (Rossor and Reilly, 2022), there are currently no universally-accepted gold standard diagnostic or assessment tools for CIPN (McCrary et al., 2017;Ibrahim and Ehrlich, 2020). The NCI-CTCAE is the most commonly used assessment tool in clinical practice (Tan et al., 2019;Li et al., 2020;Selvy et al., 2021), yet a 2017 systematic review identified 117 distinct CIPN assessment tools (McCrary et al., 2017).
The primary aims of our study were to i) perform a systematic review, evaluating the current literature for potential CYP3A4*22 and CYP3A5*3 involvement as genetic risk factors for TIPN and ii) using a candidate genotype approach, interrogate potential associations between these pharmacokinetic genetic variants, CYP3A4*22 and CYP3A5*3, and susceptibility to TIPN, evaluating the combined influence of these 2 SNPs as a genotype-derived combined CYP3A metaboliser phenotype as a genetic risk factor for TIPN.   Frontiers in Pharmacology frontiersin.org 03 and Meta-Analyses) (Page et al., 2021). Searches were not limited by date restrictions. During the review process of our manuscript, a further study was identified (published May 2023) and subsequently included in the systematic review prior to final submission.

Study selection and review
English language articles investigating an association between single nucleotide polymorphisms in CYP3A4*22 and CYP3A5*3 and peripheral neuropathy outcomes in adult populations receiving anticancer taxane chemotherapy regimens were included. Search results were identified, compiled, and screened (LMc). Irrelevant articles were excluded in the first instance by title, then by abstract. Full text examination of remaining articles assessed eligibility and identified manuscripts for inclusion (LMc). Reference lists from reviews and eligible studies were also screened for suitable articles. The article identification and selection process is outlined in the schematic (Figure 1).
The following relevant clinical and methodological information was extracted from the manuscripts: Allele, chemotherapy (treatment), phenotype definition, ethnicity, participant information (sample size and cancer type) and main findings. The data is presented in Table 2.

Candidate gene study 2.2.1 Participants, recruitment and consent
Participants were recruited retrospectively using a multi-centre ethics approved protocol, Molecular Genetics of Adverse Drug Reaction (NIHR Portfolio Study ID 8630). All patients were given a Patient Information Leaflet (PIL) and a signed consent form was required for recruitment and to be included in the study. If a patient lacked capacity to consent, a personal consultee or a nominated consultee was approached. Recruitment took place across five sites located throughout England: Clatterbridge Cancer Centre NHS Foundation Trust (CCO), St Helens and Knowsley NHS Trust (SHK), The Christie NHS Foundation Trust (CHR), Wrightington, Wigan and Leigh NHS Foundation Trust (WWL), and The Shrewsbury and Telford Hospital NHS Trust (RSO).
Eligible patients, aged >18 years and meeting the case or control definitions, were invited to participate in the study. Identification of eligible patients was via the following methods: patient lists, generated by coding departments, of individuals having received the relevant chemotherapy regimens; patient consultants informing the trials/study team; and identification on chemotherapy day units. Participants receiving oxaliplatin represented a control cohort in   Figure 2 (Continued on following page) Frontiers in Pharmacology frontiersin.org 05 this study. The following were the criteria for categorisation as either case or control: Case definition: • No pre-existing peripheral neuropathy at the time of starting chemotherapy. • Grade 2 or greater peripheral neuropathy leading to dose reduction, dose delay, or early cessation of chemotherapy.

OR.
Any grade 3 or 4 peripheral neuropathy developing during or within 6 weeks of completion of chemotherapy (this was also defined as a severe case).
• Chemotherapy regimens included were oxaliplatin, 3 weekly paclitaxel, and any three weekly docetaxel regimen. • No subsequent treatment with cisplatin or vinca alkaloid.
Control definition: • Has received at least 6 cycles of oxaliplatin containing chemotherapy OR 6 cycles of 3-weekly paclitaxel chemotherapy, OR 6 or more cycles of docetaxel 75 mg/m 2 OR 4 or more cycles of docetaxel 100 mg/m 2 .
• No pre-existing peripheral neuropathy at the time of starting chemotherapy. • No peripheral neuropathy during chemotherapy OR maximum of grade 1 peripheral neuropathy. • No subsequent treatment with cisplatin or a vinca alkaloid.
Peripheral neuropathy grading was confirmed using the National Cancer Institute Common Terminology Criteria for Adverse Events NCI-CTCAE v4.0 (https://ctep.cancer.gov/ protocolDevelopment/electronic_applications/ctc.htm) (National Cancer Institute, 2010). Maximal symptoms and symptoms at recruitment, including trips or falls due to PN, were recorded. Information on family history of PN and pre-existing PN was also captured.
Treatment information details were obtained from patient case notes, including chemotherapy regimen, cumulative dose of index drug, oncology treatment medications and alterations to planned chemotherapy regimen (if any): dose delays, reductions, and premature cessation of treatment. ADR data was recorded: date of onset of reaction, resolution of ADR, and if resolved, date thereof. Concurrent medications/chemicals (at time of reaction) were recorded. All case report form (CRF) data was transcribed to 'OpenClinica' eCRFs (electronic CRFs) facilitating clinical data management (OpenClinica LLC, Waltham, MA, USA). Biomek Corp., Ontario, Canada) or DNA Genotek Oragene ® DNA OG-500 kits (DNA Genotek Inc., Ontario, Canada). Blood samples were stored at −20°C prior to DNA isolation. Saliva samples were stored at 4°C prior to DNA isolation.

Statistics and data analysis
Statistical analysis was undertaken using the SPSS ® statistical package (IBM Corp. Released 2021. IBM SPSS Statistics for Windows, Version 28.0. Armonk, NY: IBM Corp). Continuous variables were analysed by student's (independent) t-test, and categorical variables by the chi-square test.
Univariate analysis analysed SNP association between case, severe case, and control groups. Clinical factors with significant differences (p = <0.1) were carried forward for inclusion in the binary logistic regression model.
All studies were published between 2011 and 2023. Across all 12 studies, assessment of PN was made using NCI-CTCAE, although the version used differed between studies as shown in Table 2. In addition to evaluation of toxic effects using NCI-CTCAE, one study also used the World Health Organization grading scale to assess neurotoxicity (Hu et al., 2016). Nine studies were conducted in Europe. These studies reported ethnicity information as European or Caucasian (>95%) as described in Table 2. One study was conducted in China (Hu et al., 2016), with exclusively Chinese participants, and two in the United States (Schneider et al., 2015;Shen et al., 2023): one of these US studies analysed patients of European ancestry (Shen et al., 2023). Seven studies focussed on single cancer types: five studies on breast cancer (Bosó et al., 2014; Frontiers in Pharmacology frontiersin.org 07 Eckhoff et al., 2015a;Schneider et al., 2015;Ciruelos et al., 2019;Shen et al., 2023), and two on ovarian cancer (Lambrechts et al., 2015;Hu et al., 2016). One study recruited both breast and ovarian cancer patients (Apellaniz-Ruiz et al., 2015), and another recruited gynaecological cancers (ovarian, fallopian tube and peritoneal cancers) (Bergmann et al., 2012). The remaining studies included participants with various cancer types (Leskelä et al., 2011;de Graan et al., 2013;Di Francia et al., 2017). Considering chemotherapy regimens, eight studies were taxane-only therapies (Leskelä et al., 2011;de Graan et al., 2013;Bosó et al., 2014;Eckhoff et al., 2015a;Apellaniz-Ruiz et al., 2015;Schneider et al., 2015;Di Francia et al., 2017;Ciruelos et al., 2019). Three studies recruited patients treated with paclitaxel plus carboplatin combination therapy (Bergmann et al., 2012;Lambrechts et al., 2015;Hu et al., 2016). The final study recruited patients treated with adjuvant paclitaxel (Shen et al., 2023). in Spanish participants receiving paclitaxel chemotherapy reported a trend towards higher treatment modifications in carriers of CYP3A4*22 (p = 0.066), but no statistically significant differences were observed for neuropathy grade and treatment modifications due to neuropathy (Apellaniz-Ruiz et al., 2015).

Candidate gene study
Twenty-four participants were excluded due to genotyping failures despite repeat genotyping. A total of 288 eligible participants with full clinical data were successfully genotyped and included in the final analysis. All participants were Caucasian.
Paclitaxel was significantly more neurotoxic than docetaxel (p ≤ 0.001; chi-square). Diabetes was also found to be significantly associated with neuropathy: 4% of control patients had diabetes, whilst 13% of case patients had diabetes (p = 0.024). In severe PN cases, 19% of patients had diabetes (p = 0.012). No significant differences were observed between cases and controls in BMI, alcohol consumption (units/day), hepatic impairment or CYP3Ainteracting concurrent medications. Cases tended to be older although this was not significant (p = 0.065).
For the taxane cohort, both SNPs were confirmed to be in Hardy-Weinberg equilibrium (CYP3A4*22, p = 0.25; CYP3A5*3, p = 0.40).  No significant associations were observed between either SNP and development of peripheral neuropathy or severe peripheral neuropathy (Table 4). Combined CYP3A4 and CYP3A5 genotypes were transformed into combined CYP3A metaboliser phenotype classifications (Sim et al., 2018), as shown in Supplementary Tables  S2 and S5. No significant associations were observed between metaboliser phenotype and case-control status or severe casecontrol status (Tables 4 and 5). No significant association was observed between either SNP and the development of peripheral neuropathy in the taxane-treated participants in univariate or multivariate analysis after adjustment for relevant clinical factors.
For the oxaliplatin cohort, both SNPs were confirmed to be in Hardy-Weinberg equilibrium (CYP3A4*22, p = 0.86; CYP3A5*3, p = 0.12). As expected, no significant association was observed between either SNP and development of peripheral neuropathy or severe peripheral neuropathy (Table 4). No significant associations were observed between genotype-derived combined CYP3A metaboliser phenotype (CYP3A4/CYP3A5) and development of peripheral neuropathy or severe peripheral neuropathy (Table 5).

Meta-analysis
The results of both Forest Plots are shown in Figure 2. Ethnicity information for all studies is described in Table 2. In our study, all participants were European Caucasians.
We undertook a systematic review, and also genotyped a cohort of patients treated with taxanes and oxaliplatin. In relation to CYP3A4, although de Graan et al. had identified an association between CYP3A4*22 carriers and PN (de Graan et al., 2013), other studies failed to demonstrate an association ( Table 2). The results of our candidate gene study also failed to demonstrate an association between CYP3A4*22 and development of PN or PN severity ( Table 4). The absence of subjects displaying a homozygous  (B). Association between CYP3A5*3 and taxane-induced peripheral neuropathy. Analysis of *3 homozygous carriage (*3/*3) vs. non-carriage and heterozygous carriage (*1/*1 and *1/*3).
Frontiers in Pharmacology frontiersin.org CYP3A4*22 genotype, however, is noteworthy. It is important to acknowledge that a potential risk associated with homozygous CYP3A4*22 genotypes cannot be completely ruled out due to our limited sample size. Further investigations with a larger, more diverse cohort, are warranted to comprehensively assess potential implications of homozygous CYP3A4*22 genotypes in TIPN.
Similarly, an assessment of the current evidence failed to show a significant association between the CYP3A5*3 variant and development of PN or severity of PN (Bergmann et al., 2012;Bosó et al., 2014;Eckhoff et al., 2015a;Lambrechts et al., 2015;Schneider et al., 2015;Hu et al., 2016;Shen et al., 2023). Our candidate gene analysis also failed to demonstrate an association between CYP3A5*3 and development of PN or PN severity. For both variants, the lack of an association was confirmed by our meta-analyses. Combined CYP3A4 and CYP3A5 genotypes were transformed into a CYP3A metaboliser phenotype classification with expected metabolic activities as previously defined (Sim et al., 2018). Carriers of CYP3A4*22 and CYP3A5*3 variant alleles have reduced metabolic activity and theoretically, would be at greater risk of toxicity. However, our data suggest that CYP3A metaboliser phenotype (CYP3A4/ CYP3A5) is not a risk factor with a large effect size for taxaneinduced peripheral neuropathy. However, we cannot exclude the possibility of a minor contribution, which would require a larger sample size.
We also studied a cohort of patients who had received oxaliplatin chemotherapy, which could be regarded as a separate "control" cohort. Oxaliplatin is not metabolised by CYP3A and therefore we did not expect to find any association with the CYP3A genetic polymorphisms. Indeed, this was borne out in our analysis with no association identified with either the individual alleles (Table 4) or the metaboliser status (Table 5).
Conventionally, paclitaxel is considered to be more frequently associated with peripheral neuropathy than docetaxel (Markman, 2003;Lee and Swain, 2006;Shimozuma et al., 2012;Kudlowitz and Muggia, 2013;Bhatnagar et al., 2014;Miltenburg and Boogerd, 2014). Our results support this. Our data suggest that development of peripheral neuropathy during chemotherapy is a greater risk in patients who have diabetes in keeping with previous literature Molassiotis et al., 2019). Advanced age has been a reported a risk factor in the literature (Tanabe et al., 2013;Schneider et al., 2015;Bao et al., 2016;Hershman et al., 2016;Tanabe et al., 2017;Bandos et al., 2018;Miaskowski et al., 2018;Molassiotis et al., 2019;Sánchez-Barroso et al., 2019;Hiramoto et al., 2022;Rattanakrong et al., 2022), and our data show a trend towards age being a risk factor, but this was not significant presumably because of a lack of power. Sex was also significantly associated with development of PN: 91% of cases were female.
The size of our candidate gene study is a limitation. A larger sample size may identify minor contributions from genetic variants predisposing individuals to increased susceptibility to TIPN, and allow for analysis of polygenic risk scores (PRS), rather than investigating individual alleles. Our study also lacks ethnic diversity: all participants were Caucasian. This is not rare. Genomic studies are commonly Euro-centric, with 97% of existing GWAS data from participants of European ancestry (Mills and Rahal, 2020;Pirmohamed, 2023). Our systematic review also showed an under-representation of non-European ancestry populations: 9 out of 12 studies were conducted in Europe, with an additional US study exclusively analysing patients genetically determined to be of European ancestry (Shen et al., 2023). However, evaluation of CYP3A4*22 and CYP3A5*3 allele frequencies in global populations (Supplementary Table S3) shows that the loss-of-function alleles are more common in European ancestry populations, and if an association exists with these variants and TIPN, it is more likely to be identified in this population. Concerning CYP3A5*3, there exists a significant challenge in establishing associations with TIPN in Caucasian populations due to the high frequency of the CYP3A5*3 variant. Further investigations in African ancestry populations, with a greater proportion of CYP3A5*1 alleles, may have greater power to detect differences in genetic predisposition to TIPN. Thus, future work should aim to improve ethnic diversity, serving to help address issues concerning heath disparities in genomic medicine and promote health equity. It would also identify novel loci predisposing to TIPN especially if genome-wide approaches are adopted.
In future studies, it will be important to consider not only the genetic factors, but also individual clinical factors with a view to generating multi-modal algorithms. For instance, multimorbidity, a recognised risk factor in CIPN, increases the prevalence of polypharmacy (Masnoon et al., 2017) and with that, the likelihood of drug-and gene-based interactions (Turner et al., 2020). With increasing cancer incidence and the increasing number of cancer survivors, together with changing population demographics, it is likely that the prevalence of CIPN will increase. It is therefore important that further research is conducted in this area, not only to identify predisposing factors, but also to develop biomarkers (Rossor and Reilly, 2022) which may allow earlier detection of toxicity, and offer insight into underlying mechanisms, pathogenesis and treatment (Rodwin et al., 2022).
It is also interesting to note that preventive strategies that can protect nerves from being damaged by taxane-based chemotherapy are being trialled (see Supplementary Table S4 which summarises trials registered on clinicaltrials.org). Some of these trials are investigating the use of neuroprotective agents, including vitamin E (Anoushirvani et al., 2018;Samuels and Ben-Arye, 2020;Chen et al., 2021) and acetyl-L-carnitine (Di Stefano et al., 2019), which have been shown to display some efficacy in preclinical studies. Alpha-lipoic acid has also been studied as an effective intervention for the treatment of diabetic neuropathy (Abubaker et al., 2022). Other studies are evaluating dose modifications (Sharma et al., 2020), and exercise programme (Kleckner et al., 2018;Müller et al., 2021;Chung et al., 2022) to prevent or manage PN. Despite some encouraging results, given that we do not fully understand the mechanisms of CIPN/TIPN, whether these intervention strategies will be successful is unclear, and continuing further research is needed.
Another factor which is important for future studies is to ensure that phenotypic assessment of patients is standardised. Standardization of phenotype definitions and consensus guidelines is an approach which has been employed with other ADRs (Aithal et al., 2011;Pirmohamed et al., 2011;Behr et al., 2012;Alfirevic et al., 2014;Carr et al., 2017;Nicoletti et al., 2021) as championed by the Phenotype Standardization Project (Pirmohamed et al., 2011). NCI-CTCAE is the most widely-used assessment tool for CIPN in clinical practice (Tan et al., 2019;Li et al., 2020;Selvy et al., 2021), yet despite this, a review identified 117 distinct CIPN assessment tools (McCrary et al., 2017).
Currently, oncology is considered the most evolved field in personalised medicine. Despite this, use of PGx testing in taxanebased therapies is limited, in part due to speculation over positive impact on health outcomes, cost-effectiveness, and contradictory findings. A validated PGx panel assay for the prevention of neurotoxicity has been proposed (Di Francia et al., 2017), using pharmacogenomic profiles to stratify predicted treatment outcomes and optimize pharmacotherapy, but is not implemented, perhaps because of the contradictory genetic findings and lack of prospective evaluation. The importance of a robust study design in demonstrating the clinical utility of panel pharmacogenetic testing has been demonstrated recently in the PREPARE trial, where a 12-gene panel was able to reduce adverse drug reactions to a variety of compounds by 30% (Swen et al., 2023). Looking forward, as nextgeneration sequencing (NGS) technologies become more commonplace in research and clinical practice, the importance of low frequency variants (minor allele frequencies between 5% and 1%) and rare variants (minor allele frequency <1%) in drug response should be explored and evaluated using large-scale population studies supplemented by robust electronic health records or population biobanks with linked genomics data (Pirmohamed, 2023).
In conclusion, we did not demonstrate an association between TIPN and genetic polymorphisms in CYP3A4 and CYP3A5. Our systematic review also shows some contradictory findings, but overall is consistent with our candidate gene analysis in failing to show an association, confirmed by the meta-analysis. Our study is limited by a small sample size, and so we cannot exclude a smaller effect size, and thus larger studies should be undertaken. However, in pharmacogenomics, this is not an easy task because of the need to identify patients treated with the same drug who have been accurately phenotyped. Given the narrow therapeutic index of many anti-cancer agents, and the wide range of toxicities which have been reported, this is an area which needs further study to improve the benefit-risk ratio. Although some successes in pharmacogenomics in relation to chemotherapy have been reported (Pirmohamed, 2023), much more work remains to be undertaken.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

Ethics statement
The studies involving human participants were reviewed and approved by Research Ethics No: 09/H1005/28. The patients/ participants provided their written informed consent to participate in this study.

Author contributions
All authors contributed to the conception and design of the study. LMc was responsible for acquisition of data. LMc, DC and MP drafted the paper; All authors contributed to the article and approved the submitted version.

Funding
The work was supported by the MRC Centre for Drug Safety Science. We wish to thank the MRC Centre for Drug Safety Science for support (Grant Number: MR/L006758/1). Eckhoff