Relationship between Single Nucleotide Polymorphisms and Haplotypes in DPYD and Toxicity and Efficacy of Capecitabine in Advanced Colorectal Cancer Free

Colorectal cancer is the second most frequent cause of cancer-related death in the Western world (1, 2). At early stages, resection with a curative intent is the first choice of therapy, but chemotherapy remains the backbone of treatment of irresectable, metastasized colorectal cancer (3). Commonly used chemotherapeutics are oxaliplatin, irinotecan, and fluoropyrimidines such as capecitabine and 5-fluorouracil (5-FU; refs. 4–9). Recently, the addition of targeted agents against the VEGF or EGFR (epidermal growth factor receptor) has shown to improve survival, and the current first-line standard treatment of metastatic colorectal cancer is fluoropyrimidine-based chemotherapy plus bevacizumab (10–12).

Although fluoropyrimidine drugs are generally well tolerated, approximately 10% of the patients suffer from severe fluoropyrimidine-induced toxicity. This may lead to prolonged hospitalization periods for recovery and undesired treatment delays (13). One plausible explanation for this intolerance in a subgroup of patients is interindividual variability in activity of proteins related to the pharmacokinetics (PK) and pharmacodynamics (PD) of fluoropyrimidine drugs. Especially the polymorphically expressed enzyme dihydropyrimidine dehydrogenase (DPD), encoded by DPD gene (DPYD), plays a crucial role in the pharmacology of fluoropyrimidines, as it inactivates up to 85% of 5-FU to 5,6-dihydro-5-fluorouracil (14). Genetic polymorphism in DPYD has shown to be potentially responsible for lethal toxicity after capecitabine-based chemotherapy (15). Knowledge of the clinical impact of polymorphisms in DPYD and in other genes involved in the PK/PD of fluoropyrimidines may provide opportunities for patient-tailored chemotherapy, resulting in decreased incidence of severe side effects, reduced numbers of treatment delays or cessations, and possibly increased survival probability.

The clinically most relevant polymorphism in DPYD is DPYD*2A (IVS14+1G>A), a single nucleotide substitution at the invariant splice donor site of intron 14 that leads to skipping of exon 14 during pre-mRNA splicing. As a consequence, a truncated protein is formed with absent DPD activity (16, 17). Indeed, DPD enzyme activity in heterozygous individuals for IVS14+1G>A is on average reduced by approximately 50% compared with the population average (18–20). Moreover, multiple case reports (21–27), retrospective investigations (28–30), and prospective pharmacogenetic trials (31–34) have shown a significantly increased inborn risk for severe, potentially lethal toxicity for IVS14+1G>A hetero- and homozygotic patients, when given standard doses of fluoropyrimidine drugs, although a few studies could not confirm this association (35). Besides IVS14+1G>A, more than 50 other polymorphisms in DPYD have been identified to date (36, 37). Although very few of these polymorphisms have been associated with increased risk for toxicity, the clinical relevance for the majority of these polymorphisms is low or unclear. Furthermore, data on the association of DPYD single nucleotide polymorphisms (SNP) with survival and dose modifications of capecitabine are scarce.

Therefore, the purpose of this study was to determine whether polymorphisms in DPYD are associated with toxicity of capecitabine in patients with metastatic colorectal cancer receiving capecitabine-based chemotherapy plus targeted agents. Secondary aims were to assess the effect of DPYD polymorphisms on dose modifications of capecitabine and on progression-free survival and overall survival. In addition, DPYD haplotypes were tested for these associations.

Blood samples were obtained from patients enrolled in a randomized, multicenter phase III trial, the CAIRO2 study of the Dutch Colorectal Cancer Group (38). We refer to this article for detailed study descriptions. Briefly, 736 eligible patients with metastatic colorectal cancer were randomly assigned to 3 weekly cycles of capecitabine (1,000 mg/m² bid. for 14 days), oxaliplatin (130 mg/m² on day 1), and bevacizumab (7.5 mg/kg body weight on day 1), without (CB group) or with (CBC group) cetuximab (400 mg/m² on day 1 of the first treatment cycle, followed by 250 mg/m² weekly thereafter). To reduce the incidence of peripheral sensory neurotoxicity, oxaliplatin was administered for a maximum of 6 cycles, and from course 7 onward the dose of capecitabine was increased to 1,250 mg/m². No previous chemotherapy for metastatic disease was allowed, and no adjuvant chemotherapy was allowed within 6 months before randomization. Full recovery from previous adjuvant chemotherapy was required for study participation. Adverse events were graded according to the National Cancer Institute Common Toxicity Criteria, Version 3.0. Dose reductions due to toxicity were done for each agent as specified in the study protocol. Tumor evaluation was done every 9 weeks according to RECIST (Response Evaluation Criteria in Solid Tumors) Version 1.0 (39). The study was approved by 1 central and all local institutional review boards. All patients provided written informed consent before study entry, including for translational research. Germline DNA was obtained from 568 eligible patients prior to start of therapy. All clinical results were blinded by genotype.

To retrospectively explore the effect of polymorphisms in DPYD on outcome of capecitabine-based chemotherapy, we began with a nested case–cohort study (40), followed by analysis of selected SNPs of interest in all individuals of the CAIRO2 study whose DNA was obtained (n = 568).

Cases were selected on the basis of the presence of typically capecitabine-induced toxicity events that occurred during the first 2 treatment courses, that is, diarrhea, dehydration, nausea/vomiting, stomatitis, hand-foot syndrome, leukopenia, and febrile neutropenia. Specific selection criteria for cases were at least 1 of these events grade 4 or more, or at least 2 events grade 3 or more. This led to a total of 45 cases. Although the cases were selected on specific toxicity criteria, the cohort consisted of 100 patients randomly selected from the CAIRO2 population. The frequencies of DPYD variants in the cohort population thereby reflect the incidences in the overall population, to which the polymorphism incidences in the cases were then compared with. Because of random selection of the cohort, 8 subjects selected for the case population belonged to the cohort population as well.

The entire DPYD coding region (23 exons including their flanking intronic regions) and the 3′ untranslated region (UTR) were sequenced in the 45 cases and 100 cohort controls. Genotype frequencies of observed polymorphisms were calculated. Most discriminating SNPs (P < 0.1), or nonsynonymous SNPs with a genotype frequency of more than 1% were analyzed in all individuals of the CAIRO2 study by using SNP-specific TaqMan assays. SNPs assessed in the entire CAIRO2 population that had a genotype population prevalence of more than 2% were included in the DPYD haplotype estimation, using Phase v2.1 software (41, 42). This software provides the most likely haplotype allele pair for each individual. Haplotype pairs were then associated with outcome parameters. Rarely occurring haplotype pairs (frequency <5%) were grouped for the association tests into one group consisting of patients having 1 wild-type allele and another of the rare haplotype alleles, as well as into a group of patients that had both haplotype alleles mutated. Haploview v4.1 (http://www.broad.mit.edu/mpg/haploview) was used for the analysis and visualization of linkage disequilibrium (LD). SNPs assessed in the entire CAIRO2 population and DPYD haplotype pairs were tested for association with toxicity, survival, and dose modifications of capecitabine.

Genomic DNA was isolated from peripheral blood cells by using the MagNA Pure Total Nucleic Acid Isolation Kit I on MagNA Pure LC (Roche Diagnostics). After DNA amplification by PCR, the PCR products were purified, and both DNA strands were sequenced on an Applied Biosystems 3730 DNA analyzer. Seqscape v2.5 (Applied Biosystems) was used for sequence alignment. PCR amplifications were done in a reaction volume of 50 μL containing ∼20 ng of genomic DNA, final concentrations of 0.2 μmol/L forward and reverse primers, 0.2 mmol/L dNTPs and 1.5 mmol/L MgCl₂, 1 U AmpliTaq Gold, 5 μL 10X PCR buffer II (Applied Biosystems), and water on a PTC-200 thermocycler (MJ Research, Inc.).

All PCR reactions started with denaturation and activation of the Taq enzyme for 9 minutes at 95°C, followed by 39 cycles of 1 minute at 95°C, 1 minute at the appropriate annealing temperature (Supplementary Table SA2), and 1 minute at 72°C. Exon 23 and the adjacent 3′UTR were amplified in 3 reactions (23A, 23B, and 23C) because of the large fragment size.

PCR products were purified using 2 μL ExoSAP-IT (Amersham Biosciences) and 10 μL PCR-product. The mixture was incubated for 15 minutes at 37°C and inactivated by heating to 80°C for 15 minutes. After purification, DNA cycle sequencing was carried out essentially as described by the manufacturer (Applied Biosystems) in 20 μL reactions on a PTC-200 thermocycler (MJ Research, Inc.), using identical forward and reverse primers as in the PCR amplifications. Primer sequences are given in Supplementary Table SA2.

Real-time PCR (RT-PCR) reactions were done in an end volume of 5 μL containing ∼10 ng genomic DNA, water, 0.25 μL Assay Mix consisting of SNP-specific primers, and FAM and VIC dye-labeled TaqMan MGB probes and 2.5 μL 2x TaqMan Universal PCR Fast Master Mix No AmpErase UNG on a 7500 Fast RT-PCR system (all Applied Biosystems). Primer and probe sequences for 2846A>T were 5′-TGAATTGAGCAACGTAGAGCA-3′; 5′-TGTAGCATTTACCACAGTTGATACACA-3′; VIC-TGGCTATGATTGATGAAGAA-MGB; 6-FAM-TGGCTATGATTGTTGAAGAA-MGB, and for IVS14+1G>A 5′-CATATTGGTGTCAAAGTGTCACTGAA-3′; 5′-CAACTTATGCCAATTCTCTTGTTTTAGA-3′; FAM-AGACAACGTAAGTGTGATTTA-MGB; VIC-AGACAACATAAGTGTGATTTA-MGB, respectively. Primers and probes for 1236G>A, 85T>C, 496A>G, and 2194G>A were designed by Applied Biosystems and therefore unknown to the investigators. Cross-validation by sequencing confirmed that the right polymorphism of interest were amplified for all RT-PCRs. Appropriate nontemplate, wild-type, heterozygous and homozygous control samples were included in all reactions, except for 2846A>T for which no homozygous control was available. As a quality control, 10% of all RT-PCRs were done in duplicate. The concordance rate between duplicated reactions and the primary reactions was 100%.

Six of the 8 SNPs had a population frequency of more than 2% and were used for the haplotype reconstruction. This led to the identification of 15 haplotype alleles (H1–H15; Table 5). The wild-type haplotype allele (H1) was most frequently present, with an allele frequency of 52.3%. On the basis of the most likely haplotype allele pair combination for the individual patient which was provided by the software, patients were categorized into 6 haplotype pairs (HP1–HP6): HP1 for patients with 2 wild-type haplotype alleles; HP2–HP4 for patients with 1 wild-type and 1 of the 3 most common variant haplotype alleles (H2–H4, respectively); HP5 for patients with 1 wild-type and 1 of the rare (<3%) variant haplotype alleles (grouped as H5–H15); and HP6 for patients with 2 variant haplotype alleles (grouped as H2–H15; Table 5).

Hardy–Weinberg equilibrium was evaluated using the χ² test. SNPs in the case–cohort study part were analyzed using uni- and multivariate logistic regression, dichotomized as wild type (wt) versus homozygote/heterozygote (hom/het). Association tests of SNPs assessed in the entire CAIRO2 population and DPYD haplotypes with incidences of grade 3 or more diarrhea, hand-foot syndrome grade 2 or more, and any grade 3 or more toxicity (i.e., any hematologic or nonhematologic toxicity) were done using Fisher's exact tests. The SNPs were tested and dichotomized twice to calculate the positive predictive values, both as wt versus hom/het and as wt/het versus hom, but only when there were 15 or more patients in each level of the dichotomized SNP variables. For survival endpoints, HR were calculated and tested using log-rank tests. Dose modifications of capecitabine were tested using mixed-effect modeling. The set of P values were combined and false discovery rates (FDR) were calculated in accordance with the method of Strimmer (43). Associations significant at the 5% level with FDR less than 0.3 were considered strong markers, with FDR 0.3 ≤ x < 0.4 intermediate, and with FDR 0.4 or more were considered weak markers.

Patient and treatment characteristics are shown in Table 1. Germline DNA and full clinical data were available from 568 patients (77% of the total CAIRO2 population). Patient demographics and clinical characteristics were consistent with those of the entire trial population (38).

To determine the association of polymorphisms in DPYD with toxicity of capecitabine, all 23 exons in DPYD including their flanking intronic regions and the 3′ UTR were sequenced in 45 cases and in the cohort of 100 randomly selected patients. By this method, 29 SNPs were identified: 9 nonsynonymous, 2 synonymous, 11 intronic, and 7 3′UTR SNPs (Table 2). No previously unidentified polymorphisms were detected. SNP detection by sequencing was successful in 1,245 of 1,305 (95%) and 2,840 of 2,900 (99%) of the investigated polymorphisms in the case and cohort population, respectively. All SNPs were in Hardy–Weinberg equilibrium (P > 0.05) except the variant 4079T>C, which deviated in both the case and cohort population from Hardy–Weinberg equilibrium (reason unknown); 4079T>C was therefore excluded for further analysis. LD plot and LD values are provided in Supplementary Figure SA1. The prevalence of IVS14+1G>A was higher in the case than in the cohort population (11% vs. 1%; P = 0.004) and was significantly associated with capecitabine-related toxicity (Table 2). The second and third strongest associations were observed for IVS9-51T>G (14% vs. 5%; P = 0.07) and 1236G>A (13% vs. 5%; P = 0.08), which showed high LD. When the exonic SNP 1236G>A was included in the multivariate analysis including IVS14+1G>A, gender and treatment arm, both SNPs were significantly associated with toxicity (Supplementary Table SA1). Furthermore, 2 exonic (2567C>T and 2846A>T) and 3 intronic SNPs (IVS4+66G>C, IVS11-181C>A, IVS11-119A>G) were detected only heterozygously in the cases (n = 1, 3, 1, 1, and 2 patients, respectively) but not in the cohort. On the other hand, 1679C>T and 2303C>A were only detected each in 1 individual from the cohort population. The single patient heterozygous for 1679T>G experienced no severe toxicity. Finally, 8 of the 29 observed SNPs were selected for further analysis in the entire CAIRO2 population (Table 2). Because IVS9-51T>G was in strong LD with 1236G>A, only the exonic SNP was analyzed.

Of 29 detected SNPs in the case–cohort section, 8 were assessed in the entire CAIRO2 population (n = 568). Table 3 lists the associations of these SNPs with diarrhea, hand-foot syndrome, and any (hematologic or nonhematologic) toxicity. In Table 4, the sensitivity, specificity, and the positive and negative predictive values of these associations are provided. IVS14+1G>A and 1236G>A were strongly associated (P < 0.05; FDR < 0.3) with grade 3 to 4 diarrhea, whereas 2846A>T and 2194G>A showed an intermediate association (P < 0.05; FDR 0.3 ≤ x < 0.4). Positive predictive values were 71%, 50%, 63%, and 41%, respectively. A weak association (P < 0.05; FDR ≥ 0.4) was observed for 496A>G with grade 3 to 4 diarrhea, and with grade 2 to 3 hand-foot syndrome, however, with low positive predictive values. None of the other tested SNPs was associated with hand-foot syndrome grade 2 to 3.

The probability of developing any hematologic or nonhematologic grade 3 to 4 toxicity for any genotype was 85% in this study population, which probably explains why no significant association with any grade 3 to 4 toxicity was observed for the tested SNPs. With regard to IVS14+1G>A, all patients carrying the variant allele developed grade 3 to 4 toxicity, of which 1 patient died during the third cycle of treatment that was possibly related to capecitabine, oxaliplatin, and cetuximab.

Seven patients developed febrile neutropenia, of which 3 patients seemed wild type for all 8 tested SNPs. The other 4 patients, though, were carriers of DPYD variants: 1 patient was heterozygous polymorphic for 85T>C, 496A>G, and 2846A>T; 1 patient heterozygous for 1601G>A; 1 patient heterozygous for 85T>C and 496A>G; and 1 patient homozygous for 85T>C plus heterozygous polymorphic for 1627G>A.

The cumulative administered dose of capecitabine for the first 6 courses was calculated and expressed as a percentage of the planned dose according to the protocol. The cumulative dose of capecitabine per course was significantly reduced in patients heterozygous for IVS14+1G>A (P < 0.0001) or 2846A>T (P = 0.005; Fig. 1). An average dose reduction of up to 50% was applied in IVS14+1G>A and 25% in 2846A>T variant allele carriers compared with 10% in wild-type patients, thereby treatments could be safely continued. Other SNPs in DPYD were not significantly associated with dose modifications of capecitabine.

Table 5 provides the results from the haplotype allele estimation and construction of haplotype allele pairs, and Table 3 lists the associations of haplotype allele pairs with toxicity. The commonly occurring HP3 (wild type at all SNP loci except heterozygous for 85T>C) was strongly associated with a decreased risk for grade 3 to 4 diarrhea (P < 0.05; FDR < 0.03). Conversely, patients with 1 rare variant haplotype allele (HP5) and patients with 2 variant haplotype alleles (HP6) showed a strongly increased risk for developing grade 3 to 4 diarrhea. None of the haplotype pairs were associated with grade 2 to 3 hand-foot syndrome, any grade 3 to 4 toxicity, or with dose modifications of capecitabine.

SNPs in DPYD were not significantly associated with overall or progression-free survival in this patient population. On the other hand, an intermediate association was observed for HP5 with increased overall survival [HR (95% CI) = 0.57 (0.35–0.95; P = 0.03; FDR 0.3 ≤ x < 0.4)] and a trend with increased progression-free survival [HR = 0.71 (0.50–1.0); P = 0.06; Fig. 2A and B]. A trend toward increased overall survival was observed for wild-type (HP1) versus mutated haplotype patients (not HP1; Fig. 2C). Other haplotype alleles were not associated with overall or progression-free survival.

We show that SNPs in DPYD are associated with toxicity of capecitabine in patients with metastatic colorectal cancer treated with capecitabine-based chemotherapy plus targeted agents. We show that variant allele carriers of IVS14+1G>A, 1236G>A, 2846A>T, 2194G>A, or 496A>G are at significantly increased risk for developing severe diarrhea. In addition, all patients who were heterozygous polymorphic for IVS14+1G>A developed any grade 3 to 4 toxicity, of which 1 died because of complications that was possibly related to capecitabine treatment. IVS14+1G>A has previously been associated with severe and potentially lethal fluoropyrimidine-induced toxicity, although its positive predictive value in comparable study populations has shown to range widely from 46% to 100% (30–34, 44). Explanations for this observed variability besides additional genetic variations are nongenetic factors including disease status, comorbidity, and age, which were not tested for in this study, but perhaps more important fluoropyrimidine dose intensity and concomitant chemotherapy. Notwithstanding this, the results from this study support the importance of routine screening for IVS14+1G>A prior to start of therapy. Future studies will have to resolve whether this strategy is cost-effective.

Schwab and colleagues noted a pronounced gene–sex interaction for IVS14+1G>A in their study in patients receiving various schedules of 5-FU monotherapy (34). Herein, male gender increased the prediction rate for severe toxicity whereas female gender did not. In our study, this unexpected gene–sex interaction could not be confirmed; in contrast, all females (43% of all patients with the IVS14+G>A variant genotype) developed grade 3 to 4 toxicity, which required treatment delays, followed by significant dose reductions in subsequent treatment cycles.

The observed relationships of 2846A>T and 1236G>A with toxicity are in line with recent findings from others (31, 43, 45). It is known that 2846A>T affects the DPD activity through direct interference with cofactor binding and electron transport (29, 46). However, the functional effect of the silent SNP 1236G>A that is in high LD with 2 intronic SNPs (Supplementary Fig. SA1), though, is unknown. Although 2194G>A and 496A>G were significantly associated with grade 3 to 4 diarrhea, their positive predictive values being 41% and 33%, respectively, were rather low and as such their clinical relevance and consequently usefulness is limited. It is noteworthy, however, that in a previous smaller study, 496A>G has been associated with toxicity in patients with breast or gastroesophageal cancer, but not in patients with colorectal cancer (47).

We started off with a nested case–cohort design to identify polymorphisms of possible clinical relevance by sequencing the entire coding region of DPYD. Sequencing has the advantage of possibly identifying new polymorphisms. Instead of taking matched control patients, we compared SNP frequencies observed in the cases to those in a cohort consisting of randomly selected patients. Strengths of this strategy are that the cohort is a nonbiased control population and the number of individuals is easily increased to a higher number than there are cases. Thus, statistical power is increased without facing the possibility of running out of well-matched control patients (40, 48).

A drawback of this study is that only a subselection of SNPs that were identified in the nested case–cohort analysis was analyzed in the entire population set. Thereby, SNPs of possible clinical relevance might have been missed. For example, 2567C>T, IVS4+66G>C, IVS11-181C>A, and IVS11-119A>G were detected only in the cases but not in the controls, though not further analyzed because they were either intronic or the genotype prevalence was too low. Additional studies are required to determine whether these SNPs are possibly predictive for severe toxicity.

To our knowledge, this is the first study that investigates the relationship between dose modifications of capecitabine and DPYD genotypes. Patients heterozygous for IVS14+1G>A and 2846A>T required large dose reductions in subsequent courses due to severe toxicity. Nonetheless, these SNPs were not associated with reduced overall or progression-free survival. We therefore speculate that in patients treated with this type of treatment regimen, initial 50% and 25% dose reductions of capecitabine in patients heterozygous polymorphic for IVS14+1G>A and 2846A>T, respectively, followed by further dose titration on clinical tolerability, would significantly decrease the risk for severe toxicity in this subgroup of patients while maintaining the likelihood of response to treatment. This however, requires prospective testing. Ideally, personalized dosing of capecitabine would be additionally guided using PK monitoring. For example, Gamelin and colleagues showed that individual pharmacokinetically guided dosing in 5-FU–treated patients with metastatic colorectal cancer resulted in a significantly improved response rate, fewer severe side effects, and a trend toward increased overall survival (49). However, this has not yet been shown to be applicable for patients treated with capecitabine. Another possibility to improve patient safety of fluoropyrimidine therapy would by to administer a 5-FU test dose and to monitor the rate of 5-FU metabolism (50). A strong advantage of this strategy is that there is a lower risk of generating false-negative results, because herewith also nongenetically induced, DPD-deficient patients may be identified. It is, however, less convenient and more time consuming for the patients than genotyping approaches. Another method that has shown to potentially increase patient safety of 5-FU–based chemotherapy is the use of the endogenous uracil/dihydrouracil (U/UH₂) plasma ratio as a measure for DPD activity. In patients with head and neck cancer, the initially administered dose of 5-FU was reduced depending on the endogenous U/UH₂ plasma ratio, thereby the rate of severe toxicity was strongly reduced compared with a historical control data set treated with the standard dose of 5-FU (51).

Patients with the common HP3 haplotype showed a decreased risk for developing severe diarrhea. Patients carrying the HP3 haplotype are heterozygous for the clinically nonsignificant SNP 85T>C and wild type at the other loci. On the basis of this genotype and the decreased risk for severe diarrhea, a normal DPD enzyme activity can be assumed for HP3 patients. The observed inverse relationship between risk and toxicity is a new finding and in line with phenotypic findings in which 5-FU–treated patients with normal to high DPD activities had reduced risk for developing severe toxicity (52). In contrast to HP3, patients heterozygous for rare haplotypes and patients with 2 variant haplotype alleles (HP5 and HP6, respectively) experienced increased risk for severe diarrhea. These data suggest that besides the use of SNPs, DPYD haplotypes might additionally assist in patient-tailored chemotherapy with fluoropyrimidines; however, this warrants additional research. This study population consisted of mostly Caucasian patients. However, as regional differences are observed in the tolerability of fluoropyrimidines (53), it would be interesting to test for differences in haplotype frequencies in various geographic populations.

This is the largest study to date that has associated DPYD polymorphisms with survival in fluoropyrimidine-based chemotherapy regimens. Our initial hypothesis that polymorphisms in DPYD might beneficially affect survival, as they may lead to reduced DPD enzyme activities, seemed not true. For haplotypes, on the other hand, we observed a significant association with overall survival for HP5 (patients with 1 rare haplotype allele) and a trend toward significance for HP1. In contrast to the high numbers of SNPs and haplotype alleles that were significantly associated with toxicity, the number of associations with survival was rather low. The additional chemotherapy and targeted agents in this treatment regimen might have overshadowed or at least hampered the ability to show such relationships, if they do exist. This idea is supported by the observation that the survival curves overlap during the first 6 months after treatment randomization, as well as for HP1 versus non-HP1, and for HP5 versus non-HP5, but split up thereafter. Interestingly, oxaliplatin was limited to a maximum of 6 cycles (i.e., ∼4.5 months of therapy), after which capecitabine was increased to its usual dose as monotherapy of 1,250 mg/m² according to the study protocol. This might possibly explain why these survival curves started to split from 6 months onward. Clearly, however, whether these relationships of DPYD haplotypes with survival do exist has to be confirmed in additional patient populations.

In summary, we conclude that SNPs and DPYD haplotypes are useful predictive markers for toxicity in patients with metastatic colorectal cancer treated with capecitabine-based chemotherapy. The data suggest that initial dose reductions of 50% in IVS14+1G>A and 25% in 2846A>T variant allele carriers with further dose titration would significantly reduce the total number of severe toxicity events, thereby separate validation is indicated.

The authors declare no conflict of interest.

The study was funded by the Netherlands Cancer Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.