Predictors of survival and toxicity in patients on adjuvant therapy with 5-fluorouracil for colorectal cancer

The present study aimed at investigating whether the simultaneous evaluation of pharmacokinetic, pharmacogenetic and demographic factors could improve prediction on toxicity and survival in colorectal cancer patients treated with adjuvant 5-fluorouracil (5FU)/leucovorin therapy. One hundred and thirty consecutive, B2 and C Duke's stage colorectal cancer patients were prospectively enrolled. 5FU pharmacokinetics was evaluated at the first cycle. Thymidylate synthase (TYMS) 5′UTR and 3′UTR polymorphisms and methylenetetrahydrofolate reductase (MTHFR) C677T and A1298C polymorphisms were assessed in peripheral leukocytes. Univariate and multivariate analyses were applied to evaluate which variables could predict chemotherapy-induced toxicity, disease-free survival (DFS) and overall survival (OS). Multivariate analysis showed that: (a) low 5FU clearance was an independent predictive factor for severe toxicity (OR=7.32; P<0.0001); (b) high-5FU clearance predicted poorer DFS (HR=1.96; P=0.041) and OS (HR=3.37; P=0.011); (c) advanced age was associated with shorter DFS (HR=3.34; P=0.0008) and OS (HR=2.66; P=0.024); (d) the C/C genotype of the MTHFR C677T polymorphism was protective against grade 3–4 toxicity (P=0.040); (e) none of the TYMS polymorphisms could explain 5FU toxicity or clinical outcome.

consisting of a G4C base change at the twelfth nucleotide, which makes the transcriptional activity of the 3R allele as low as that of the 2R allele (Mandola et al, 2003). Third, a 6 bp deletion may occur in the 3 0 -untranslated region (3 0 -UTR) of TYMS, associated with decreased mRNA stability and lower TS expression (Mandola et al, 2004). Two functionally important single nucleotide polymorphisms (SNP) have also been detected in the MTHFR gene (MTHFR), C677T and A1298C, resulting in decreased MTHFR activity (Ulrich et al, 2002).
Despite these theoretical and experimental premises, clinical studies which have evaluated the effect of TYMS (Gusella and Padrini, 2007) and MTHFR (Robien et al, 2005) polymorphisms on the colorectal cancer response/toxicity to 5FU have produced inconsistent results.
An additional determinant of 5FU activity is the metabolic rate of dihydropyrimidine dehydrogenase (DPD), the enzyme responsible for 5FU inactivation (van Kuilenburg, 2004). Variability in DPD activity is only marginally explained by genetic factors, as deficient variants of DPD gene are extremely rare (van Kuilenburg et al, 2001). Hence, DPD activity is best determined by direct in vivo measurement of 5FU plasma clearance (CL).
The aim of our work was to evaluate the clinical outcome and toxicity of colorectal cancer patients on adjuvant treatment with 5FU in relation with demographic characteristics, known TYMS and MTHFR polymorphisms, and individual 5FU CL.

Patients and treatment
A total of 130 consecutive B2 and C Duke's stage colorectal cancer patients on adjuvant therapy were prospectively studied between 1999 and 2008. Chemotherapeutic treatment consisted of the 5FU þ leucovorin combination according to the Mayo administration schedule (O'Connel et al, 1997) (2 min i.v. bolus administration of 425 mg m -2 5FU þ 20 mg m -2 folinic acid daily for 5 days, for 6 consecutive cycles every 4 weeks). The starting dose could be reduced by 25% in the oldest patients. If toxicity of grade X2 occurred, the next doses were reduced by 25 -50% or stopped, according to toxicity level. The study was approved by the Ethics Committee of the Azienda -ULSS 18 Rovigo, and all patients gave their written informed consent.

Clinical evaluation
Toxicity was recorded and graded according to World Health Organization (WHO) criteria and was dichotomously classified for statistical analysis in none-to-moderate (grade 0 -2) and severe (grade 3 -4).
The outcome variables were disease-free survival (DFS) and overall survival (OS). For DFS, an event was defined as the first occurrence of a tumour relapse that was not preceded by a second primary cancer. For OS, the outcome was death related to cancer and patients who died for other causes were treated as censored in survival analysis. The duration of follow-up was measured from the date of starting chemotherapy. Follow-up visits were scheduled every 3 months in the first year, every 6 months in the second and third years, and every year thereafter.

Pharmacokinetic analysis
5FU pharmacokinetics was determined on day 2 of the first therapy cycle, using an HPLC method and a limited sampling strategy described earlier (Gusella et al, 2002). Plasma concentration decay was analysed with a mono-exponential function, so that intercept (A) and slope (k) were obtained and used to calculate the main pharmacokinetic parameters: area under the plasma concentration-time curve: AUC ¼ A/k; plasma clearance: CL ¼ Dose/AUC.
With the limited sampling methods used, bias and accuracy for AUC estimate are 3.4 and 5.1%, respectively (Gusella et al, 2002).

Genetic analysis
Genomic DNA was extracted from peripheral blood leukocytes with a commercial kit (Promega Corporation, Madison, WI, USA) and stored at 41C until genotyping. 0.2 mg per sample of genomic DNA was used to carry out the PCR using specific primers in a 50 ml final volume; genotyping was carried out according to earlier reported methods for TYMS 5 0 UTR VNTR and G4C substitution (Gusella et al, 2006a), TYMS 3 0 UTR 6 bp insert-deletion (Ulrich et al, 2000), MTHFR C677T and MTHFR A1298C polymorphisms (Miao et al, 2002).
Repeat genotyping of a 10% random sample yielded a 100% agreement.
Classification of TS expression based on combined 5 0 UTR and 3 0 UTR genotypes Variable number of tandem repeats and G4C SNP polymorphisms in the 5 0 UTR were classified according to Kawakami and Watanabe (2003) in low expression (2R/2R, 2R/3RC, 3RC/3RC) and highexpression TS genotypes (2R/3RG, 3RC/3RG, 3RG/3RG). 5 0 UTR and 3 0 UTR were also combined in two different ways, based on the expected TS expression level (see Table 1). The simplest combination (combination A) takes into account VNTR and ins/del 3 0 UTR polymorphisms and classifies TS activity as 'high' or 'low', according to Hitre et al (2005). The more complex combination (combination B) also includes G4C SNP in the 5 0 UTR and divides TS activity into three expression groups: 'high', 'intermediate' and 'low'.

Classification of MTHFR activity based on combined C677T and A1298C genotypes
Similarly, MTHFR activity was classified as 'high', 'intermediate' and 'low', based on the expected MTHFR metabolic rate of the combined C677T and A1298C polymorphisms (Ulrich et al, 2002) (see Table 2).

Statistical analysis
Univariate logistic regression was used to identify the baseline prognostic factors related to toxicity. The factors significant at the 0.20 level were then introduced in a multivariate logistic regression  Survival and toxicity determinants of adjuvant fluorouracil M Gusella et al model, and a backward-elimination approach was adopted.
Interactions were tested at the 0.10 level of significance. The results are presented as odds ratio (OR) and 95% confidence interval. The hazard ratios (HRs) for disease recurrence and mortality, were estimated for individual covariates with the univariate Cox proportional-hazards model. Variables significant at the 0.20 level in the univariate analysis were entered into multivariate Cox proportional-hazards models in a backward selection procedure. Interactions were tested at the 0.10 level of significance. Proportionality assumption was tested including time-dependent covariates in the final model at a significance level of 0.10. The results are presented as HR and 95% confidence interval. Covariates considered were age (two groups divided by the median value), sex, Dukes' stage (B2 and C), TYMS VNTR polymorphism (three genotypes), TYMS G4C SNP polymorphism (low-and high-expression genotypes), TYMS ins/del polymorphism (three genotypes), TYMS 5 0 UTR and 3 0 UTR combined genotypes (combination A: two groups; combination B: three groups; see Table 1), MTHFR C677T polymorphism (three genotypes), MTHFR A1298C polymorphism (three genotypes), MTHFR 677 and 1298 combined genotypes (three groups; see Table 2), 5FU clearance per Kg (two groups divided by median value) and true cumulative 5FU dose administered, expressed as a percentage of the theoretical dose (continuous).
Survival curves were estimated using the Kaplan -Meier method and differences between individual curves were assessed by logrank test. Data are presented as means ± s.d., unless otherwise specified. The w 2 -test and t-test for unpaired data were applied to compare frequencies and means, respectively. A one-way ANOVA (followed by the post-hoc linear-trend test, when needed) was used to compare groups. All tests were two-sided with a 5% significance level. Data were analysed with version 9.1.3 of SAS software for Windows (SAS Institute Inc., Cary, NC, USA).
Hardy -Weinberg equilibrium and linkage disequilibrium were evaluated by PyPop on line free software (University of California, Berkley, USA) (Lancaster et al, 2003).

Clinical data
A total of 130 patients (46 women and 84 men) were the study population. All had a Karnofsky performance status score between 100 and 80. Mean age was 63.7 ± 10.3, ranging from 34 to 84 years. The primary tumour site was right colon in 47 cases and left colon in 83 cases. Duke's tumour stage was B2 in 51 patients and C in 79 patients. Stage C was more frequently found in patients aged X65 years than in those aged o65 (80.3 vs 42.2%; Po0.0001).
A total of 116 patients were treated for all six planned cycles; 10 patients stopped the treatment prematurely because of toxicity, mainly at the first or second cycle, and four patients stopped because of disease progression. The true cumulative dose was, on average, 74.1±20.2% (range: 14.4 -101.6%) of the theoretical one, and dose reduction was significantly greater in patients aged X65 (30.5 ± 19.0%) than o65 (21.1 ± 20.5%) (Po0.0075).

Pharmacokinetic data
The mean 5FU starting dose was 724.5±88.2 mg (range: 350 -920 mg) and the related pharmacokinetic parameters were broadly variable among patients (AUC range: 146-1236 mg l -1 Â min; CL range: 0.0084 -0.0705 l min -1 Kg -1 ). 5FU clearance measured at the first cycle was inversely correlated with the percent dose reduction during the following cycles: the lower the CL, the greater the reduction in the cumulative dose, with a linear trend across the CL quartiles (ANOVA, Po0.0001; linear trend, Po0.0001) ( Figure 1).

Survival analysis
Survival analysis was restricted to the 122 patients who completed at least three chemotherapy cycles, that is, those who received at least 50% of the planned cumulative dose, in order to exclude the patients whose clinical outcome could not be reasonably ascribed to 5FU treatment. Univariate analysis showed that advanced age and Dukes' stage C had a significant negative prognostic effect on  (Table 4). Conversely, sex, cumulative dose reduction and onset of severe toxicity had no predictive value. Also, no significant association was observed between DFS or OS and any TYMS or MTHFR polymorphisms, either analysed separately or in combination.
When patients were stratified according to CL and age in four groups (Figure 2), those with high CL and aged X65 survived less (HR: 6.50; 95% CI: 1.44 -29.36; P ¼ 0.015) and relapsed earlier (HR: 5.99; 95% CI: 1.95 -18.51; P ¼ 0.0018) than those with low CL and aged o65, who had the best outcome. A significant difference in DFS was also observed between old/high CL and young/high CL patients (log-rank test: P ¼ 0.025) and between old/low CL and young/low CL patients (P ¼ 0.047). In addition, OS of patients aged 465 and with high CL was significantly worse than that of any other group (P range: 0.012 -0.009). The comparisons between the other groups did not reach the significance level.
Frequency of Duke's C stage was lower in the younger/low CL group than in the older/high CL group (42.3 vs 81.5%, P ¼ 0.0033), whereas severe toxicity was more frequent in the former group than in the latter (42.3 vs 18.5%, P ¼ 0.059). Nevertheless, as stated above, in the overall population neither C Duke's stage nor severe toxicity were markers of OS/DFS, as their predictive value in the multivariate analysis was probably included in age and CL, respectively.

DISCUSSION
The main clinical implication emerging from our study is that high-5FU CL (X0.017 l min -1 Kg -1 ), measured at the beginning of the adjuvant chemotherapy course in colorectal patients, is the strongest single predictor of reduced toxicity and poor clinical outcome in terms of both DFS and OS. Multivariate analysis also showed that age X65 was associated with shorter DFS and OS but not with more severe toxicity, suggesting that preventive dose reduction in elderly patients is not advisable. As a whole, the association of high-5FU CL with advanced age had the worse prognostic meaning for both DFS and OS. Furthermore, patients homozygous C/C for C677T MTHFR, who are presumed to havehigh enzyme activity, had a lower-toxicity risk.
With respect to 5FU tolerability, our findings are in good agreement with earlier literature data, as high AUC or low CL have invariably been associated with more severe 5FU toxicity after any administration schedule (Young et al, 1999;Di Paolo et al, 2002;Jansman et al, 2002;Ychou et al, 2003;Codacci-Pisanelli et al, 2005;Gusella et al, 2006b;Capitain et al, 2008). A recent, large, prospective trial (Schwab et al, 2008) on cancer patients treated with 5FU monotherapy confirmed a significant --although weak --link between mutant MTHFR C677T genotypes and FU toxicity but, at odds with our results, also found a significant correlation with the 2R/2R TYMS genotype. A correlation between 5FU toxicity and the VNTR TYMS polymorphism has also been reported by Lecomte et al (2004) and Pullarkat et al (2001), who found less toxicity in patients with the 3R/3R genotype. Indeed, our univariate analysis also suggests a non-significant trend toward a decreased toxicity risk in patients carrying the 3R/3R genotype in comparison with carriers of the 2R/2R TYMS genotype (OR ¼ 0.55). A significant correlation with C677T MTHFR genotype, but not with TYMS genotypes, was also found by Sharma et al (2008) in colorectal patients treated with capecitabine, a 5FU prodrug.
With respect to factors correlated with therapeutic outcomes, age and 5FU CL turned out to be the only significant predictors of DFS and OS in our study. The association between age and clinical outcome may be partly explained by the greater prevalence of stage C in our older patients. However, other studies found that advanced age is an independent risk factor (Gill et al, 2004;Schippinger et al, 2007). Similarly, an association between low 5FU Survival and toxicity determinants of adjuvant fluorouracil M Gusella et al CL or high AUC and improved 5FU response has already been reported in cancer patients treated with either continuous 5FU infusion (Milano et al, 1994;Gamelin et al, 1996) or Machover's bolus administration schedule (Di Paolo et al, 2008).
DPD governs 5FU catabolism in both normal and tumour cells, and a significant correlation has been shown between DPD expression in colorectal tumours and the surrounding intestinal mucosa (Tanaka-Nozaki et al, 2001;Amatori et al, 2006). Thus, on  (Popat et al, 2004;Ciaparrone et al, 2006;Jensen et al, 2006;Lassmann et al, 2006;Popat et al, 2006;Yamada et al, 2008), whereas no consistent correlation has been shown with TS tumour expression (Kamoshida et al, 2004;Aguiar et al, 2005;Ciaparrone et al, 2006;Jensen et al, 2006;Lassmann et al, 2006;Popat et al, 2006;Soong et al, 2008;Yamada et al, 2008). Indeed, in our study, none of the three TS polymorphisms, considered alone or in combination (VNTR, G4C substitution, 6 bp insertiondeletion), could predict the clinical outcome. This negative results may be because of an insufficient power of our study to detect genetic associations or to some inaccuracy in grouping the genotypes. However, our population size was comparable with that of other similar studies and genotype classifications reflected the best available information on gene biological activity. In this respect, literature data are inconclusive in that the same TS polymorphisms have been correlated with both favourable and unfavourable effects, and no correlation has even been reported (Dotor et al, 2006;Gusella and Padrini, 2007;Gosens et al, 2008;Fernandez-Contreras et al, 2009 Figure 2 Disease-free (A) and overall survival (OS) plots (B) of patients stratified in four groups according to age (o or X65 years) and 5FU clearance (o or X0.017 mg l À1 Kg À1 ). Open circles indicate censored data points. Log-rank test was significant for both disease-free survival (DFS) (P ¼ 0.0012) and overall survival (OS) (P ¼ 0.0036).
differences between clinical trials, a number of biological factors may have obscured the link between TYMS polymorphisms and 5FU activity. First, it is known that 5FU can not only inhibit TS but also affect mRNA function. Sobrero et al (1997) have suggested, and others (Kubota et al, 2002;Matsusaka et al, 2003;Hoshino et al, 2005) have experimentally confirmed, that 5FU incorporation into colorectal tumour mRNA is significantly greater after bolus than after continuous infusion. If this is true, RNA-mediated effects may prevail over TS inhibition, and TS expression may play only an ancillary role in the antitumour activity of 5FU bolus administration. Second, translation of TS mRNA is inhibited by TS protein in an autoregulatory manner (Chu et al, 1991), so that the influence of gene polymorphisms may be blunted. Third, it has been reported that 5FU treatment can induce a variable, compensatory increase in TS mRNA levels in colorectal tumours (Ukida et al, 2001;Tanaka-Nozaki et al, 2003) and that patients with no or minor mRNA increase show a better response to therapy (Ukida et al, 2001). Fourth, the 5 0 -UTR polymorphism is in linkage disequilibrium with the 3 0 -UTR polymorphism, in that the þ 6/ þ 6 bp and À6/À6 bp genotypes are associated with 2R/2R and 3R/3R genotypes, respectively (Mandola et al, 2004). It is therefore likely that the coexistence in the same gene of high-and low-expression genotypes may attenuate functional consequences. Fifth, TS tumour levels may be simultaneously markers of tumour aggressiveness and tumour sensitivity to 5FU. In this regard, a recent, large study (Soong et al, 2008) reported that low-TS expression was correlated with worse prognosis in colorectal patients treated with surgery alone whereas no significant effect on survival was noted in the 5FU-treated group. As it is generally acknowledged that low TS-expressing tumours are more sensitive to 5FU (Gusella and Padrini, 2007), the lack of prognostic significance of low TS levels in 5FU-treated patients may be explained by a concomitant harmful effect on tumour growth. The prognostic role of the MTHFR polymorphism in the response to 5FU did not emerge from our study. No studies with adjuvant therapy are available for a comparison, although contradictory results come from studies in other clinical settings (Cohen et al, 2003;Etienne et al, 2004;Jakobsen et al, 2005;Terrazzino et al, 2006;Ruzzo et al, 2007;Zhang et al, 2007). The reasons for such discrepancies are unclear.
In conclusion, our study makes a contribution to the controversial issue of which colorectal patients are more likely to gain benefit from 5FU adjuvant chemotherapy. One strength of our investigation is the simultaneous evaluation of an index of drug exposure (5FU CL) and of several TYMS and MTHFR polymorphisms possibly related to tumour sensitivity, in a group of patients homogeneously treated with a repeated bolus regimen. The finding that patients' demographic and pharmacokinetic characteristics are more predictive of clinical outcome than some putative genetic markers suggests that the traditional, phenotypic approach to clinical cancer research should not be abandoned and that the genetic approach requires further work to prove its clinical usefulness. Feasibility and utility of a pharmacokinetic test to assess 5FU CL in colorectal patients starting adjuvant therapy need to be evaluated in a prospective trial.