Background Peripheral neuropathy is the dose limiting toxicity of paclitaxel, a chemotherapeutic drug widely used to treat solid tumours. This toxicity exhibits great inter-individual variability of unknown origin. The present study aimed to identify genetic variants associated with paclitaxel induced neuropathy via a whole genome approach.
Methods A genome-wide association study (GWAS) was performed in 144 white European patients uniformly treated with paclitaxel/carboplatin and for whom detailed data on neuropathy was available. Per allele single nucleotide polymorphism (SNP) associations were assessed by Cox regression, modelling the cumulative dose of paclitaxel up to the development of grade 2 sensory neuropathy.
Results The strongest evidence of association was observed for the ephrin type A receptor 4 (EPHA4) locus (rs17348202, p=1.0×10–6), and EPHA6 and EPHA5 were among the top 25 and 50 hits (rs301927, p=3.4×10–5 and rs1159057, p=6.8×10–5), respectively. A meta-analysis of EPHA5-rs7349683, the top marker for paclitaxel induced neuropathy in a previous GWAS (r2=0.79 with rs1159057), gave a hazard ratio (HR) estimate of 1.68 (p=1.4×10−9). Meta-analysis of the second hit of this GWAS, XKR4-rs4737264, gave a HR of 1.71 (p=3.1×10−8). Imputed SNPs at LIMK2 locus were also strongly associated with this toxicity (HR=2.78, p=2.0×10−7).
Conclusions This study provides independent support of EPHA5-rs7349683 and XKR4-rs4737264 as the first markers of risk of paclitaxel induced neuropathy. In addition, it suggests that other EPHA genes also involved in axonal guidance and repair following neural injury, as well as LIMK2 locus, may play an important role in the development of this toxicity. The identified SNPs could form the basis for individualised paclitaxel chemotherapy.
- Molecular genetics
- Peripheral nerve disease
Statistics from Altmetric.com
Paclitaxel is an antineoplastic drug frequently used in first line treatment for breast, ovarian, lung, and prostate cancers. This molecule binds the cellular microtubules through the β-tubulin subunit, promoting their stabilisation, preventing cell division and finally leading to apoptosis. Although paclitaxel is an effective treatment for various types of cancers, there are associated toxicities that lead to serious clinical limitations in its use. Peripheral neuropathy is the dose limiting toxicity of paclitaxel, with most patients treated with the drug developing this adverse effect.1 Peripheral neuropathy is predominantly sensory and is generally axonal, distal, symmetrical, debilitating, and painful.2 Although the causal mechanisms have not been well defined, it is clear that microtubule mediated axonal transport is affected.3 Paclitaxel neurotoxicity is dose dependent and there are various clinical factors that have been suggested to increase the risk of developing it, such as pre-treatment with neurotoxic agents, use of antiretroviral drugs, and a personal history of diabetes mellitus, chronic liver disease, alcoholism, hypothyroidism, nutritional deficits or hereditary or acquired polyneuropathies.1 However, despite these factors, there is great inter-individual variability in the neurotoxicity of patients receiving similar amounts of paclitaxel under similar protocols, with some patients suffering serious neuropathies which lead to dose reductions and suspensions. Peripheral neuropathy usually takes months to disappear, and in the most severe cases the damage to the peripheral nerves can be irreversible. Therefore, the identification of biomarkers to predict the risk of suffering severe neuropathy would be of great clinical utility.
Our group and others have applied candidate gene approaches, focused on the pharmacokinetic and pharmacodynamic pathways of paclitaxel, to identify genetic variants associated with peripheral neuropathy, in some cases observing contradictory results. In any case, the variants identified explain only a relatively small fraction of the variation in neuropathy outcome.4–8 Recently, the first genome-wide association study (GWAS) of paclitaxel induced neuropathy was published, and identified single nucleotide polymorphisms (SNPs) in genes not previously studied, such as EPHA5, XKR4, and FGD4,9 as putative markers of risk of this toxicity. However, replication of these results and/or meta-analyses including other GWAS are required to provide sufficient evidence to consider them for guiding clinical use of this critically important chemotherapeutic agent. In this study we performed a GWAS with 144 white European cancer patients homogenously treated with paclitaxel/carboplatin and for whom detailed neuropathy data were collected. We provide independent support of EPHA5-rs7349683 and XKR4-rs4737264 as the first markers of paclitaxel induced peripheral neuropathy and suggest common variants in other EPHA genes, and at LIMK2 locus, as potential additional markers of susceptibility to this toxicity.
Patients and methods
Patients and peripheral neuropathy assessment
Blood or saliva samples were collected from 144 cancer patients treated with paclitaxel in one Spanish and two Swedish centres: 48 patients (33%) from Hospital Universitario Fundación Alcorcón,7 63 (44%) from Karolinska University Hospital, and 33 (23%) from Linköping University Hospital.10 Eligible patients were over 18 years of age and had: received a chemotherapy regimen with paclitaxel 175 mg/m2 and carboplatin area under the time–concentration curve (AUC) 5–6 every 21 days to treat a histologically documented solid neoplasia; life expectancy of ≥12 weeks; Eastern Cooperative Oncology Group performance status of ≤2; no chemotherapy, hormonal therapy nor radiotherapy in the 4 weeks before treatment; taken contraception (fertile women only); adequate bone marrow, renal and hepatic function; and no previous history of neuropathy. Most patients had ovary or lung malignancies (70% and 19%, respectively) and paclitaxel/carboplatin chemotherapy was administered as first line treatment in the vast majority (94%, table 1). All patients were of white European origin. The collection of samples was approved by local internal ethical review committees and all patients gave written informed consent. Additional details related to the patients, tumours, and treatments are summarised in table 1.
Demographic, tumour, and treatment data were prospectively collected by medical chart review and stored in an electronic database. To guarantee that information about neuropathy was collected in a homogeneous manner, the three participating centres designed a common questionnaire based on the National Cancer Institute Common Toxicity Criteria (NCI-CTC) V.2 which included details of number of treatment cycles, sensory symptoms such as paresthaesia (in the feet only, or present in both the feet and hands), and functionally disabling neuropathies. Although a more extensive assessment of sensory and motor paclitaxel induced neurotoxicity was undertaken for a subset of 71 patients7 ,11 (see online supplementary methods), only NCI-CTC scoring was used for the analyses. Neuropathy symptoms at baseline and cumulative paclitaxel dose at first neuropathy event, at grade 2 sensory neuropathy, and at maximum neuropathy grade were also collected from all patients. Baseline neuropathy was recorded as zero, when the patients did not report symptoms of neuropathy previous to starting the chemotherapy. In addition, details of treatment delays, dose reductions and treatment suspensions, and the reasons for these, were recorded, together with the cumulative paclitaxel dose at any changes in treatment (table 2).
DNA isolation, genotyping, quality control, and SNP imputation
DNA was isolated from peripheral blood and from saliva using standard protocols. The final DNA concentration was quantified by PicoGreen (Invitrogen, Carlsbad, USA). Genotyping was performed on 250 ng of DNA using the Infinium BeadChip Human 660WQuad assay (Illumina, San Diego, USA) which consists of 657 366 markers in a four-sample array format. The GenomeStudio software package was used to extract genotype data from files created by the Illumina iScan System. One sample with a call rate <0.95, probably due to poor DNA quality, was excluded; all other samples had call rates >0.99. Non-diploid variants (eg, mitochondrial chromosomes), copy number variation (CNV) probes and SNPs deemed unreliable by Illumina (Tech Note: Infinium Genotyping Data Analysis, 2007) were excluded, leaving 559 348 SNPs. After also excluding SNPs with missing genotypes in >5% of samples, as well as those with minor allele frequency (MAF)<0.025, 518 577 SNPs were included in the association analysis. Genotyping for rs5749248 was performed using the KASPar SNP genotyping system (Kbiosciences, Hoddesdon, UK), and fluorescence was determined and alleles assigned by the sequence Detection System 7900HT (Applied Biosystems, Foster City, USA).
Genotypes were imputed for additional SNPs in or near EPHA4, EPHA6, EPHA5, EPHA8, and LIMK2 based on data from the 1000 Genomes Project (June 2011 release) using IMPUTE (V.2.0). Before imputation, genotyped SNPs were filtered and limited to those that were autosomal, with MAF ≥0.01 and p value for departure from Hardy–Weinberg equilibrium (HWE)≥0.001. Imputed SNPs with MAF<0.025 and call rate <0.90 were excluded.
Associations with risk of paclitaxel induced sensory neuropathy were tested for clinical factors and SNPs using Cox regression analysis, modelling the cumulative dose of paclitaxel up to the development of grade 2 peripheral sensory neuropathy. Patients with no or minimal adverse reaction (grade 0/1) were censored at total paclitaxel cumulative dose (mg). Associations with SNPs were assessed under an additive genetic model by multivariable analysis; age was included as a covariate because there was weak evidence that older patients were at increased risk of sensory neuropathy (HR per year=1.02, CI 95% 1.00 to 1.05; p=0.082). Although motor neuropathy was also recorded in the database, its incidence is lower than the sensory neuropathy (ie, only 6% of the patients developed motor neuropathy grade 2 or higher; while 48% of the patients developed sensory neuropathy grade 2 or higher) and it was not analysed due to low statistical power. Cox regression analyses were performed using the R statistical software (V.2.14.0); all other statistical analysis were carried out using PLINK (V.1.07). All SNPs with a nominal p value for association ≤10−4 were further evaluated for potential errors by visualising genotype cluster plots, comparing the estimated MAF with that reported by the HapMap Project for the CEU (Utah residents with ancestry from northern and western Europe) population (using Pearson's χ2 test) and considering failure rates and evidence for departure from HWE; details are provided in table 3. All SNPs except one (p>0.005) had p values >0.01 for departure from HWE, and none had missing genotypes. To account for possible differences in the ethnic origin of patients, analyses were repeated adjusting for country of origin, with no substantial differences observed in the results obtained. Also, using prior chemotherapy as an additional covariate in the analysis did not have an impact on the results (data not shown). Statistical analyses were also performed using cumulative paclitaxel doses in mg/m2 without finding substantial changes in the results obtained. The meta-analysis was performed using the R package meta in a fixed effect model, because of the similarities between studies (same phenotype measured on the same scale, ethnicity and genetic effects) and because both Cochran's Q measure of heterogeneity and I2 statistic showed low variation between them. Haplotype analysis was conducted for LIMK2 region using genotyped SNPs. Haplotype blocks were identified in HapMap CEU samples using the Haploview V.4.2 software, based on the method described by Gabriel et al12 and haplotypes were imputed using PHASE V.2.0. The association of each haplotype with peripheral neuropathy was assessed using Cox regression, adjusted for age.
The distribution of paclitaxel induced peripheral sensory neuropathy in the included patients is shown in table 2. The median cumulative paclitaxel dose at which patients developed peripheral sensory neuropathy was 2046 mg and this toxicity caused dose modifications in 25% of patients. There was no significant difference in the incidence of neuropathy between Spanish and Swedish patients (p=0.91). No association with paclitaxel induced sensory neuropathy was observed for gender, tumour type, tumour stage, previous chemotherapy or prior use of neurotoxic drugs (p≥0.29).
None of the SNPs genotyped in the array was associated with risk of sensory neuropathy at genome-wide statistical significance (defined as p<5×10−7 13), but p values <10−5 were observed for several SNPs (table 3). The strongest evidence of association was observed for rs17348202 (HR=4.85, 95% CI 2.57 to 9.13, p=1.02×10−6; table 3 and figure 1A), located downstream of EPHA4. Imputation of additional SNPs at the EPHA4 locus did not suggest any stronger association signals (data not shown).
Interestingly, a synonymous SNP in EPHA5 was the top association hit in a previous GWAS of paclitaxel induced neuropathy (rs7349683, HR=1.63, 95% CI 1.34 to 1.98, p=9.6×10−7).9 In our GWAS, two SNPs in complete linkage disequilibrium (LD), rs1159057 and rs12507286, located in a ∼50 kb LD block that included the EPHA5 gene, were among the top 50 hits based on p value (HR=2.01, 95% CI 1.43 to 2.84, p=6.84×10−5; table 4). SNP imputation revealed rs139491476 as having a slightly stronger association signal (p=3.95×10−5, see online supplementary figure S1). The previously reported EPHA5 marker, rs7349683, in high LD with rs1159057 (r2=0.79), had a p value for association of 3.33×10−4 in the present study (HR=1.83, 95% CI 1.32 to 2.55; figure 1B; see online supplementary figure S1). A meta-analysis of results for rs7349683 from both GWAS gave a genome-wide statistically significant p value of 1.4×10−9 (HR=1.68, 95% CI 1.42 to 1.99).
Given that our and the previous GWAS observed the strongest signals for EPHA4 and EPHA5, respectively, and that the EphA family is involved in nerve damage response,14–,18 we examined more closely the results for SNPs in genes from this family. An intronic SNP in EPHA6, rs301927, was among the 25 top hits (HR=2.35, 95% CI 1.57 to 3.53, p=3.44×10−5; table 3, figure 1C). Imputation of further SNPs at this locus revealed four SNPs, in complete LD, with a lower p value (p=2.87×10−5, see online supplementary figure S2). In addition, rs209709, 12 kb upstream of EPHA8, had a p value for association of 1.28×10−3 (table 4), and imputation revealed an EPHA8 intronic SNP, rs3754005, and a missense variant, rs606002, not in LD with rs209709 (r2<0.03), with p values of 9.83×10−4 and 3.36×10−3, respectively. No other SNPs in EPHA genes appeared to be associated with paclitaxel induced neuropathy.
The SNP with the second lowest p value for association was rs4141404 (HR=2.41, 95% CI 1.66 to 3.48, p=3.22 × 10−6; table 3), located in the 3′UTR of the LIMK2 gene. Analysis of the surrounding LD structure for the CEU population from the 1000 Genomes Project revealed a ∼500 kb region with r2>0.7 including eight additional genes (see online supplementary figure S3). An intronic SNP in LIMK2, rs2413045, but in relatively low LD with rs4141404 (r2=0.58), was also among the top 25 hits (table 3). SNP imputation identified two SNPs (rs5749227 and rs5749248, in complete LD) that showed a stronger association with neuropathy (p=6.38×10−8, see online supplementary figure S3). Direct genotyping of rs5749248 confirmed an association with paclitaxel induced peripheral neuropathy that reached genome-wide statistical significance (HR=2.78, 95% CI 1.89 to 4.08, p=1.98×10−7, figure 1D). Haplotype analysis of SNPs at this locus did not suggest any stronger associations with neuropathy than the single SNP analysis (data not shown).
In addition, it is interesting to note that rs3829306, rs4149023, and rs4149013 were among the 25 top associated SNPs. These SNPs in complete LD are located on introns and upstream of SLCO1B1, the gene encoding the paclitaxel hepatic uptake transporter OATP1B1 (table 3).
With respect to paclitaxel induced neuropathy markers proposed by Baldwin et al,9 we performed a meta-analysis of results for their eight top SNPs (see online supplementary table S1). In addition to rs7349683 in EPHA5, Baldwin's second hit rs4737264 in XKR4 gave a genome-wide statistically significant p value of 3.11×10−8 (HR=1.71, 95% CI 1.41 to 2.06).
We have performed a GWAS to identify genetic variants associated with paclitaxel induced peripheral sensory neuropathy, the dose limiting toxicity of this drug. These polymorphisms might serve as markers to define subsets of patients at high risk of neuropathy that could receive alternative chemotherapeutic regimens. Neuropathy can decrease not only the quality of life of patients but may also alter the efficacy of treatment, through early dose reductions and suspensions. Several candidate gene association studies have produced conflicting results,4 ,6–8 ,19 and a major part of the inter-individual variability in paclitaxel induced neuropathy remains unexplained. A recently published GWAS on this toxicity, based on the clinical trial CALGB 40101, identified putative novel susceptibility loci,9 although replication in independent studies is required to confirm these. Here we independently replicate the association of EPHA5 and XKR4 with paclitaxel induced peripheral sensory neuropathy reported by Baldwin et al, and propose additional genetic susceptibility loci in other EPHA genes and at LIMK2 locus.
There are nine EphAs in humans that are expressed in almost all tissue types during development and in most cell types of adults. EphA/ephrin-A signalling is crucial for nervous system development, tissue regeneration, and tumour progression.20 EphA5 knockout mice have shown that this receptor is essential in the initiation of the early phases of synaptogenesis,21 with EphA5 expression increasing in response to sciatic nerve lesions.14 EphA4 is directly implicated in the regulation of axonal regeneration,15 ,16 astrocyte responsiveness,22 and other pathways involved in the repair of neural injury.17 In addition, recent studies of amyotrophic lateral sclerosis identified EphA4 as a determinant of the vulnerability of neurons to axonal degeneration.18 EphA6 knockout mice were found to have behavioural deficits as well as learning and memory impairment,23 and findings in EphA8 knockout mice suggest that this receptor plays a role in axonal path finding during the development of the mammalian nervous system.24 Taken together, these results highlight the relevant function of EphAs in neurons and in pathways involved in the repair of neural injury.
In this study we replicate the finding that the EPHA5 synonymous variant rs7349683 is a marker of risk of paclitaxel induced neuropathy, with the association reaching genome-wide statistical significance in the meta-analysis (p=1.4×10−9) with an estimated HR of 1.68 (95% CI 1.42 to 1.99). Baldwin et al also included in their study a small replication set in which rs7349683 was genotyped; inclusion of results from this in the meta-analysis gave a HR estimate of 1.59 (95% CI 1.36 to 1.87) with a p value of 1.1×10−8. Approximately 48% and 16% of patients will be heterozygous and homozygous, respectively, for the rare allele of rs7349683, and have an estimated 1.68- and 2.82-fold higher risk of neuropathy, respectively, than wild-type homozygous patients. Interestingly, rs17348202 and rs301927, our first and 23rd top hits, are also located in or near EPHA4 and EPHA6, respectively, and an SNP near EPHA8 showed weaker evidence of association (table 4). On the whole, these findings suggest that polymorphisms in the EPHA genes may play a crucial role in the development of paclitaxel induced neuropathy and should therefore be studied further.
When we examined other paclitaxel induced neuropathy markers proposed by Baldwin et al,9 in addition to rs7349683 in EPHA5, rs4737264 in XKR4 also gave a genome-wide statistically significant p value in the meta-analysis (p=3.11×10−8). This is an intronic SNP in XKR4, a still uncharacterised protein expressed in the cerebellum, previously associated with attention deficit/hyperactivity disorder25 and to iloperidone and risperidone response.26 ,27 Among the top 25 most statistically significant associations were two independent SNPs located at LIMK2 locus (table 3), and direct genotyping of the imputed SNP rs5749248 confirmed a genome-wide statistically significant association (p=1.98×10−7, figure 1D). LIMK2 knockout mice exhibit minimal abnormalities, but LIMK-1/2 double-knockout mice have impaired excitatory synaptic function.28 In addition, knockdown of LIMK2 in cell lines results in significantly reduced neurite bearing cell count, neurite length, and cone extension growth rate.29 Furthermore, LIMK2 expression has been implicated in sensitivity to the microtubule binding agents vincristine and vinblastine in a neuroblastoma cell line.30 With respect to other genes at this locus, PIK3IP is a negative regulator of PI3K and PLA2G3 is involved in oxidative stress and is associated with Alzheimer's disease.
It is important to highlight that the 25 hits with the strongest evidence of association also included SLCO1B1, the gene encoding OATP1B1, which is a paclitaxel hepatic uptake transporter previously studied in paclitaxel related candidate gene approaches.7 ,31 The OATP1B1 amino acid change V174A, caused by rs4149056, is known to decrease the transport of several drugs including docetaxel.32 However, this variant is not correlated with rs3829306 (r2=0.012) and is not associated with neuropathy (HR=1.41, 95% CI 0.95 to 2.09, p=0.089).
As discussed previously, we have independently replicated two associations from a previous GWAS, confirming EPHA5-rs7349683 and XKR4-rs4737264 as markers of paclitaxel induced neuropathy.9 However, we did not observe evidence of association for other top hits from that study, such as FGD4 and FZD3. These discrepancies could be due to differences between the two studies. While the GWAS of Baldwin et al9 was based on 855 breast cancer patients treated with paclitaxel monotherapy at 175 mg/m2 every 2 weeks, we included mostly ovarian and lung cancer patients treated with paclitaxel 175 mg/m2 plus carboplatin AUC 5–6 every 3 weeks. We did not find differences in the incidence of neuropathy among the different types of tumours included, but breast cancer was not considered. Both GWAS included patients with European ancestry and the associations with risk of neuropathy were assessed using Cox regression, modelling the cumulative dose of paclitaxel up to the development of grade 2 peripheral sensory neuropathy.
It is important to highlight the inaccuracy of neuropathy assessment based exclusively on NCI-CTC grading reported by physicians.33 We applied the same criterion to assess neuropathy in the three participating hospitals, and half of the patients went through a thorough neurologic examination.7 ,10 However, the limited number of samples included in our study implies reduced statistical power, making it possible that additional true associations went undetected. It is also important to highlight that since all of our patients were treated with a combination of paclitaxel and carboplatin, even though paclitaxel is more neurotoxic than carboplatin,34 we cannot affirm that the neuropathy markers found correspond only to paclitaxel and not to carboplatin or a combination of both drugs. Despite these possibilities and the differences between the studies, both GWAS found an association for EPHA5-rs7349683 and XKR4-rs4737264, thus indicating that these are neuropathy markers valid for the different paclitaxel chemotherapy regimens considered. In addition, the nerve injury repair function of EphA5 suggests that this marker could also be valid for other neurotoxic drugs that depend on repair pathways similar to those of paclitaxel.35 ,36 Altogether, the results obtained by our group and others suggest that several common variants associated with paclitaxel sensorial peripheral neuropathy with moderate effect sizes are expected. The combination of these markers would improve the genetic prediction capability for this toxicity, facilitating implementation in the clinic. In this respect, different polygenic modelling methods have been proposed to explain a larger proportion of a trait under study.37–39
In summary, this study confirms EPHA5-rs7349683 and XKR4-rs4737264 as markers of risk of paclitaxel induced sensory neuropathy. In addition, it suggests that common variants in other EPHA genes and at the LIMK2 locus could also play a role in this toxicity. Together, these findings suggest that genes involved in the function and repair of peripheral nerves, which had not previously been studied in candidate gene approaches, could make a substantial contribution to the genetic susceptibility to this toxicity. EPHA5-rs7349683 and XKR4-rs4737264 appear to be the first markers which, when combined with others as they emerge, could be used clinically to classify patients according to their neuropathy risk—an important step towards individualised paclitaxel chemotherapy.
We thank the CNIO Bioinformatics Unit and the Spanish National Bioinformatics Institute Unit, especially Guillermo Comesaña for his help in the creation of an online clinical database via the EPIQuest tool (http://www.inab.org/wp-content/epiquestweb/index.html). We also thank Roger Milne for revising the manuscript.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
▸ Additional material is published online only. To view please visit the journal online (http://dx.doi.org/10.1136/jmedgenet-2012-101466).
EÅ-L, HG and CR-A acted as senior authors for this manuscript.
Contributors LJL-G, EA-L, HG and CR-A designed the study. LJL-G, EH, SL, CJ, XM, EA-L, HR and CR-A did the data collection. LJL-G, LI-P, GP, AG-N, MR, EA-L, HG and CR-A did the data analysis, data interpretation, and wrote the manuscript. CJ, EA-L and HG recruited patients. All authors critically reviewed the manuscript and approved the final version.
Funding This work was supported by projects from the Spanish Ministry of Science and Innovation (grant numbers SAF2006–01139 and SAF2009–08307), the Spanish Ministry of Economy and Competiveness (grant number SAF2012–35779), the Swedish Cancer Society, the Swedish Research Council, Fondkistan, Stiftelsen Sigurd och Elsa Goljes Minne and Markus Borgströms stiftelse, and The Cancer Research Funds of Radiumhemmet. Luis Javier Leandro-García was supported by a FIS fellowship (grant number FI08/00375).
Competing interests None.
Patient consent Obtained.
Ethics approval The collection of samples was approved by the local internal ethical review committees of the Hospital Universitario Fundación Alcorcón, Karolinska University Hospital and Linköping University Hospital Internal Ethical Review Committees.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.