Polymorphisms in Genes Encoding Drugs and Xenobiotic Metabolizing Enzymes, DNA Repair Enzymes, and Response to Treatment of Childhood Acute Lymphoblastic Leukemia¹

ALL³ is the most common cancer found among children, accounting for approximately 25–30% of all childhood malignancies. It is characterized by the predominance of lymphoblasts or immature hematopoietic precursors, with malignant cells expressing diverse phenotypes and variable response to chemotherapy (1, 2). Treatment involving multiple chemotherapeutics has led to a remarkable improvement in disease outcome. Yet, 20–40% of the patients develop resistance to current therapeutic protocols (1, 3, 4). Intensive treatment also has significant long-term consequences, causing secondary malignancies and cognitive impairments (5, 6). There is thus a need to identify factors associated with both the risk of relapse and drug side effects. Although several prognostic indicators such as age, sex, WBC count, DNA ploidy, and immunophenotype are well known (1, 2), the molecular mechanisms underlying either the resistance to therapy or drug-related toxicity are poorly understood. Earlier studies focusing on the interindividual variation in activities along xenobiotic metabolism pathways showed that they played a role in the susceptibility to childhood ALL (7, 8, 9, 10, 11). Several of the participating enzymes not only metabolize environmental carcinogens but also process drugs used in the treatment of this malignancy (12, 13, 14, 15).

DNA variants of genes involved in phase I metabolism, such as CYP genes, are associated with a higher enzymatic inducibility, such as CYP1A1 (2), may have a reduced rate of substrate metabolism, such as CYP2D6 (12, 16, 17, 18), or may lead to a variable level of transcription (CYP2E1; Ref. 19). DNA polymorphism in genes encoding MPO, which also performs activation of toxic substances, leads to reduced MPO activity (20). Several variants in genes encoding detoxifying enzymes of phase II metabolism have been described (13, 14, 21, 22, 23, 24, 25, 26). Deletion polymorphisms are found in GSTM1 and GSTT1, leading to the absence of a functional enzyme (21, 22, 23). Two variants of the GSTP1 gene (GSTP1*B and GSTP1*C) result in a substrate-dependent alteration in the enzymatic activity (24, 25). Polymorphisms in genes coding for NAT1 and NAT2 are associated with different acetylation capacity, ranging from slow to fast metabolism of potentially toxic substances (13, 14, 27), whereas polymorphism in NQO1 results in reduction of NQO1 activity (26, 28).

GSTs and NQO1 participate directly in the metabolism of cytotoxic drugs. GSTs are involved in the metabolism of alkylating agents, anthracyclines, and steroids (29, 30, 31, 32). Genotypes resulting in lower GST activity may be advantageous for individuals undergoing chemotherapeutic treatment because a reduced detoxification may enhance the effectiveness of a cytotoxic drug. NQO1, on the other hand, performs a dual action, participating in both activation (33, 34) and inactivation of cytotoxic agents (28, 35).

DNA repair systems are important to correct DNA damage induced by carcinogens or anticancer drugs. The cancer risk as well as the responsiveness of cancer cells to these agents can be differentially affected by changes in repair efficiency (36, 37). Base excision repair, mismatch repair, and nucleotide excision repair were shown to contribute to the drug resistance phenotypes (38, 39). Recently, several common polymorphisms have been found in the coding sequences of hMLH1, hMSH3, XPF, XRCC1, and APE (40, 41, 42). Although little is known about the functional significance of these variants, some of them, such as those represented by XRCC1 alleles, were associated with the development of gastric and colorectal cancer (43, 44).

Our hypothesis is that polymorphisms of genes, the protein products of which influence drug or carcinogen metabolism, as well as DNA repair components, may alter the probability of relapse. Therefore, we studied the role of these variants in the outcome of childhood ALL.

Children with ALL, treated at the Sainte-Justine Hospital in Montreal between 1987 and 2001, with no other type of leukemia or hematological malignancy, were retrospectively included in this study (n = 320). The patients were treated with LAL protocols and protocols developed by the Dana-Farber Cancer Institute: DFCI 87-01, 91-01, and 95-01 (5, 45). The drugs administrated are summarized in Table 1. The patient characteristics and clinical prognostic factors [sex, age at diagnosis, presenting WBC count, immunophenotype (cell type), treatment protocol, risk group, and DNA index] are given in Table 2. The risk classification (standard, high, and very high) was performed at the time of the child’s enrollment in the treatment protocol. Children with at least one of the following characteristics were assigned to the high-risk group: high WBC count (>20 or >50 × 10⁹/l, depending on the protocol), age >10 or <1, T-cell type, Philadelphia chromosome positivity, biphenotypic leukemia, central nervous system disease, or presence of mediastinal mass. Patients without these factors were assigned to standard risk, whereas children enrolled in the treatment protocols preceding DFCI 91-01 with either WBC above 100 × 10⁹/l, Philadelphia chromosome or younger than 12 months were considered at very high risk of relapse. The children were considered to have an event (n = 68) if relapse or death attributable to the disease occurred during or after treatment. A consent letter was obtained for all patients participating in the study.

The genotyping of patients’ DNA from mouth swab or peripheral blood at remission was performed in 320 individuals. Because the genotyping of these loci was initially performed to study their role in ALL susceptibility, the number of individuals genotyped may vary from locus to locus, depending on the development of ongoing projects (see Table 5). The polymorphic deletion of the GSTM1 and GSTT1 genes was detected by PCR-based assays (8). PCR allele-specific-oligonucleotide hybridization assays was used to genotype polymorphisms in CYP1A1 (T6235C, A4889G, and C4887A), CYP2D6 (G1934A and a 1-bp deletion at position 2637), CYP2E1 (G-1259C), MPO (G463A), NAT1 (C559T, G560A, T640G, T1088A, and C1095A), NAT2 (T341C, C481T, G590A, A803G, and G857A), and NQO1 (C609T and C465T) as described previously (10, 46). The selected mutations in GSTP1 (A1578G and C2293T; Refs. 24, 25), hMLH1 (A676G), hMSH3 (G2835A and A3124G), XRCC1 (C26304T), APE (T2197G), and XPF (G1256A; Refs. 40, 41, 42) were also investigated by allele-specific oligonucleotide-based assays. These polymorphisms were used to define the variants of the tested genes (Table 3).

For the statistical analysis, variant genotypes were grouped on the basis of their phenotypic consequences as dichotomous variables. The analyses were performed for carriers of at least one CYP1A1*2A, CYP1A1*2B, CYP1A1*4, or CYP2E1*5 variant, CYP2D6 poor metabolizers (homozygous *3/*3 and *4/*4, as well as compound heterozygous *3/*4 individuals); carriers of MPO *2 allele, individuals with GSTM1 and GSTT1 null genotypes; carriers of at least one GSTP1*B or GSTP1*C variant; carriers of NAT1*10 with potentially higher acetylating capacity; NAT2 slow acetylators (carriers of two NAT2 variants other than NAT2*4 and NAT2*12A); and carriers of NQO1*2 or NQO1*3 variant. For the repair genes, the analysis were performed for carriers of hMSH3 Gln₉₄₀ or Ala₁₀₃₆, XRCC1 Trp₁₉₄, XPF Gln₄₁₅, and APE Glu₁₄₈ variants, as well as for hMLH1 Ile₂₁₉/Ile₂₁₉ individuals shown previously to influence ALL risk.⁴

As in other studies addressing the impact of various factors on ALL outcome (1, 2), we also use an event (a relapse or a fatal outcome attributable to the disease) as a dependent variable. The influence of tested variants on outcome was assessed by EFS probabilities estimated by Kaplan-Meier and Cox regression. In these analyses, patients with an event, considered as a group (n = 68), represent real observations, whereas children without event (n = 252) are encountered only as censored observations. Survival time corresponds to the time between diagnosis and the occurrence of event. For the children without an event, it corresponds to the time between diagnosis and the end of follow-up (84 months after diagnosis) or to the time between diagnosis and March 2001 for those that are receiving therapy or entered the follow-up period. Because the number of genotyped individuals may vary from locus to locus (see genotyping), the range and median for the length of follow-up of the non-event patients are given for each locus in Table 5. Kaplan-Meier curves were generated for graphical representation of survival. EFS differences between the patients with and without a given genotype were determined using log-rank test. HRs and accompanying Ps, with enclosure of clinical prognostic factors grouped as dichotomous or continuous covariates corresponding to the categorization given in Table 2, were estimated using the Cox proportional hazard model. Categories associated with better prognosis were considered as reference: WBC count below 50 × 10⁹/l, female sex, age between 12 and 120 months, B-cell type, DNA index >1.16, the most recent treatment protocol (95-01), and standard risk of relapse. The variants studied here show variation across different populations (47). We also found a slight difference in the frequency of CYP1A1*2B and CYP1A1*4 alleles in French-Canadians compared with other Caucasians (8); therefore, this population variable was also included in the multivariate analysis. To further reduce confounding effects of prognostic factors, the correlation between them and the genetic variants that had an impact on ALL was explored. This allowed adding to already existing factors of the multivariate model the interaction term between the variant and the factor for which the correlation was found (P′ in Table 5). The interaction terms between CYP1A1*2A and treatment protocols, as well as between CYP1A1*2B and cell phenotype, population, and treatment protocols were additionally introduced to calculate P′ in Table 5. Such interactions were not found for NQO1*2 variant (P = P′). To calculate HRs in stratified analysis given in Figs. 1 and 2, we included in the Cox regression model the main effects of variables (genetic variant and prognostic factor for which stratification was performed) as well as interaction term between them. Kaplan-Meier estimates were in such cases performed within each stratum. Finally, we assess the best predictors of the event in the children with ALL. For that purpose, using a stepwise routine, the estimations of HR were calculated by Cox regression analysis, the initial multivariate model of which included prognostic factors listed in Table 2 and genotypes that were shown to influence the risk of event (CYP1A1*2A and NQO1*2).

The influence of the patients’ characteristics on EFS was assessed by Kaplan-Meier and Cox regression analysis.

Our study included 320 ALL patients, among which 68 children had an event, which corresponded to relapse or fatal outcome. The impact of the polymorphisms of genes encoding xenobiotic metabolizing and repair enzymes (frequencies are given in Table 4) on the disease outcome was assessed by estimating EFS probabilities for ALL patients with and without the variant genotypes, using both Kaplan-Meier and Cox regression analyses (Table 5). Kaplan-Meier estimates showed that children with CYP1A1*2A, and NQO1*2 genotypes had shorter survival probabilities (P = 0.003 for both CYP1A1*2A⁺ versus CYP1A1*2A⁻ and NQO1*2⁺ versus NQO1*2⁻; Table 5 and Fig. 1,a). The effect of hMLH1 Ile₂₁₉/Ile₂₁₉ versus other hMLH1 genotypes was noticed but was of borderline significance (P = 0.08; Table 5). When the simultaneous effect of CYP1A1, NQO1, and hMLH1 variants on ALL outcome was assessed, a further increase in the risk was obvious for combination of hMLH1 Ile₂₁₉ and CYP1A1*2A variants. The difference between EFS of both of them versus none of these genotypes was 0.0002 (Fig. 2 a).

We further estimated HRs (Table 5) in Cox regression analysis for each of the tested variants in the presence of all prognostic factors listed in Table 2. In this analysis, the CYP1A1*2A and NQO1*2 retained their predictive values, whereas the effect of hMLH1 Ile₂₁₉ was attenuated. An additional impact of the CYP1A1*2B variant was observed (Table 5, P). To better understand the impact of variants with significant impact on ALL outcome (CYP1A1*2A, CYP1A1*2B, and NQO1*2), we built another multivariate model in which confounding effects that may arise through interactions between genetic variants and prognostic factors (see “Patients and Methods”) were additionally corrected (Table 5, P′). In this analysis, CYP1A1*2A and NQO1*2 retained their effects, suggesting that these two variants may independently influence the outcome of ALL. Therefore, we examined the influence of CYP1A1*2A and NQO1*2 on ALL outcome when stratified by risk (Fig. 1,b) and protocol (Fig. 1,c) groups that appeared to largely influence ALL prognosis (Table 2). The similar impact of these variants on ALL outcome was observed for both good and poor risk groups, whereas the difference of survival probabilities of individuals with and without variants was more apparent for DFCI protocols (Fig. 1,c). However, the HR for CYP1A1*2A and NQO*2 retained significance in multivariate analysis, which included stratification by treatment protocol (P = 0.005 and 0.05, respectively). The similar results of stratified analysis were observed for individuals with combined CYP1A1 and hMLH1 at-risk genotypes (Fig. 2, b and c).

We next determined the relative importance of CYP1A1*2A and NQO1*2 genotypes to predict the event. The initial Cox regression model, which included these variants together with all prognostic factors listed in Table 2, was applied to stepwise analysis. NQO1*2 and CYP1A1*2A variants appeared to be the best predictors of an event (P < 0.001 in both cases, satisfying the significance threshold level imposed by Bonferonni correction attributable to multiple testing). Respective HRs were 4.2 (95% CI, 2.1–8.6) and 3.3 (95% CI, 1.7–6.4).

The interpretation of any results presented here did not change if patients with only relapse or exclusively treated by homogeneous DFCI protocol were considered in the analyses.

We showed that the CYP1A1*2A variant was associated with a worse therapeutic outcome in children with ALL, both with (HR, 2.3; 95% CI, 1.2–4.9) and without (P = 0.003) enclosure of clinical prognostic factors. None of the drugs administrated in this patient population is metabolized directly by CYP1A1. However, it has been shown that synthetic glucocorticoids, an important component of ALL treatment protocols, can potentiate the induction and catalytic activity of CYP1A1 (48) acting through CYP1A1 glucocorticoid-responsive elements (49). Enhanced enzyme activity could promote activation and thus toxicity of CYP1A1 substrates if an individual is simultaneously exposed to such procarcinogens. Recently, we reported that the CYP1A1*2A variant can increase the risk of leukemia, especially when children are exposed to pesticides or cigarette smoke (8, 50, 51). In this regard, it is indeed plausible that children with this variant, particularly in the presence of substrates and/or inducers, not only have an enhanced susceptibility to develop ALL but also recurrent malignancy.

Two-electron reduction of a subset of quinone-containing drugs, performed by NQO1, leads to drug activation because generated hydroquinones are further processed to semiquinone and ROS (33, 34). However, the other quinone-based chemotherapeutics, among which is doxorubicin administrated in ALL treatment protocols, are not subjected to NQO1-mediated activation, either because they undergo one-electron reduction performed by P450 reductase or form stable hydroquinone with no or little ROS formation (34). Two-electron reduction of doxorubicin seems to lead to drug inactivation, because unstable and less active metabolites are formed (35). In addition to therapeutic agents, a whole range of quinone-containing carcinogens are substrates for NQO1, which exerts its detoxifying role through protection from free radicals generated by both drug and carcinogen metabolism (28, 52). Here we found that individuals with the NQO1*2 variant associated with reduced enzyme activity (26, 28) had worse therapeutic outcome (HR, 3.6; 95% CI, 1.7–7.4). Decreased NQO1 activity may enhance the drug effect but also ROS formation and drug-associated toxicity. Keeping in mind the detoxifying role of NQO1, it may be possible that impaired cell protection from free radicals that have arisen through drug and carcinogen metabolism may lead to the development of recurrent malignancies. The same variant has been shown previously to play a role in the susceptibility to childhood ALL (10) and therapy-related leukemia (53, 54, 55).

A reduced DNA mismatch repair efficiency was found to cause resistance to cytotoxic drugs, such as platinum-based drugs, methylating compounds, and doxorubicin by an inability to detect DNA damage, thus preventing cells to undergo DNA damage-induced apoptosis (38, 39, 56, 57). We found that the Ile₂₁₉ variant exerted its influence on ALL outcome only if combined with other at-risk genotypes such as the CYP1A1*2A variant (HR, 5.6; 95% CI, 1.9–16.8). The functional significance of hMLH1 Ile₂₁₉ is not known; however, we found recently that this variant contributes an additional risk of childhood ALL.⁴

The involvement of GST in drug detoxification might suggest their putative effect in drug resistance (29, 31, 32). We did not observe any association between GST genotypes associated with altered enzymatic activity and ALL outcome. On the other hand, a reduction of relapse in children with ALL was reported to be associated with the GST null genotypes (58), particularly GSTT1 null (59). In addition, numerous studies reported the correlation between GST expression and EFS in patients with both lymphoblastic and nonlymphoblastic leukemia (30, 60, 61, 62). However, several other studies reported a lack of such associations (63, 64). Differences from one study to another regarding design, number of patients, treatment protocol, and particularly changes of GST expression acquired through tumor development versus intrinsic enzymatic variability (Refs. 29, 30, 58, and 63, and this study) may account for observed discrepancies.

In conclusion, this is the first report that shows that specific genetic variants in CYP1A1, NQO1, and hMLH1 genes (the latter in combination with other at-risk genotypes) are associated with an increased risk of relapse in children undergoing treatment for childhood lymphoblastic leukemia. It is likely that their impact on outcome arise through interplay of gene-drug and gene-environment interactions, some of them already predisposing an individual to initial cancer development. Because these polymorphisms may interfere with drug activity, it would be also of interest to analyze their impact on toxicity after treatment. Although due care was taken to minimize the confounding effect of other factors, given the very high number of genes studied relative to the sample size, these preliminary results should be confirmed in larger studies. However, the finding of this study still suggests that determining an individual’s genotypes is important for prediction of disease outcome. At the same time, it justifies further analysis of polymorphisms of other genes relevant to therapeutic response in ALL. This might lead to an alternative to traditional chemotherapy protocols, which would allow drug dosing on an individual’s capacity to respond, thus leading to more efficient and less toxic treatment.

We are grateful to all patients and their parents who kindly agreed to participate in this study; to the physicians and the clinical staff for their collaboration; and Chantal Richer and Hugues Sinnett for technical assistance.