Main

The successful incorporation of risk-associated factors in contemporary treatment regimens has rendered paediatric acute lymphoblastic leukaemia (ALL) to be one of the most curable forms of cancer, with cure rates exceeding 80% in developed countries (Pui et al, 2011). Presenting parameters, such as age, leukocyte counts and cytogenetics are widely used (Schultz et al, 2007). In addition, early response to therapy measured by minimal residual disease (MRD) levels after initial remission induction therapy is an independent predictor of the eventual outcome and is increasingly used for risk assignment in many contemporary protocols (Pui et al, 2009; Conter et al, 2010; Yeoh et al, 2012).

Clinically, inter-patient variations in treatment response are seen even in the same leukaemia subtype. For instance, up to 28% of patients with poor prognosis t(9;22)/BCR-ABL1-positive ALL can be cured (Arico et al, 2000), whereas up to 15% of patients with good-prognosis t(12;21)/ETV6-RUNX1 fusion or hyperdiploidy (>50 chromosomes) relapse on a contemporary treatment protocol (Pui et al, 2009; Pui et al, 2011; Yeoh et al, 2012). Host-related factors may account for a significant proportion of inter-patient variation in treatment response and toxicity. The polymorphic thiopurine methyltransferase (TPMT) gene is a well-known example that affects 6-mercaptopurine (6-MP) metabolism (Relling et al, 1999; McLeod et al, 2000; Stanulla et al, 2005a), a drug widely used in treatment of ALL.

Paediatric ALL protocols use multi-drug combinations, which may be affected by polymorphisms participating in the transport, metabolism as well as detoxification of the individual drugs (Krajinovic et al, 2002; de Jonge et al, 2005; Rocha et al, 2005; Costea et al, 2006; Cunningham and Aplenc, 2007; Davidsen et al, 2008; Xu et al, 2012). However, the impact of such genomic variations on treatment outcome, especially in relation to conventional risk assignment strategies, remains to be fully elucidated. Using a candidate gene approach targeting 20 germline polymorphisms in 11 genes implicated in treatment response in paediatric ALL, we evaluated their influence on treatment outcome in 463 children enrolled in the Malaysia-Singapore ALL 2003 study. The selected candidates (Supplementary Table S1) have been reported either to influence the metabolism of the standard chemotherapy drugs used in treatment of ALL (e.g. MTHFR 677C>T (de Jonge et al, 2005) and ABCB1 3435C>T (Jamroziak et al, 2004)) or affect MRD level at the end of induction (e.g. CCR5 246A>G (Davies et al, 2008) and IL15 SNPs (Yang et al, 2009)), or event-free survival (EFS, e.g. GSTM1 null (Rocha et al, 2005) and NQO1 609C>T (Krajinovic et al, 2002).

Materials and methods

Patients and samples

We studied a total of 463 consecutive patients who were enrolled from July 2002 to August 2009 (median follow-up of 5.7 years upon data analyses) in the Malaysia-Singapore (Ma-Spore) ALL 2003 study from four participating centers in Singapore and Malaysia (Yeoh et al, 2012). This smaller cohort has similar EFS (Supplementary Figure S1) as well as demographic composition (data not shown) compared with the entire Ma-Spore ALL patient group. Genotyping assays were carried out in patients who had sufficient DNA. The patients are primarily comprised of Chinese, Malays and Indians, according to self-reported ethnicity. The treatment protocol was a modification of the ALL IC-BFM 2002 backbone (Fronkova et al, 2008) with additional CCG-augmented BFM regimen for high-risk patients (Nachman et al, 1997). Risk assignment was based on clinical- and cytogenetic-presenting features as well as early response to therapy as determined by prednisolone response and molecular MRD at the end of remission induction and consolidation (Supplementary Table S2).

Mononuclear cells were separated and collected from bone marrow aspirates or peripheral blood at diagnosis and after remission, using a Ficoll-Paque (GE Healthcare, Little Chalfont, UK) density gradient centrifugation. DNA was isolated using TRIzol (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s recommendation.

The study was approved by the local institutional review boards at all participating institutions and informed consent was obtained from patients or their parents in accordance with the Declaration of Helsinki.

Cytogenetic subgroup, MRD, and genotype

Morphological examination and immunophenotyping were performed in all patients. Karyotyping, DNA index, and oncogene fusion screening by reverse transcription PCR were performed in National University Hospital or KK Women’s and Children’s Hospital in Singapore. For the present study, t(9;22)/BCR-ABL1, t(11q23)/MLL rearrangements or hypodiploid ALL (modal chromosomes <45 or DNA index <1.0) were classified as unfavourable cytogenetic subgroup, while t(12;21)/ETV6-RUNX1, t(1:19)/TCF-PBX1 or hyperdiploid ALL (modal chromosomes >50 or DNA index ⩾1.16) were classified as a favourable subgroup.

For MRD monitoring, patient-specific, leukaemia-associated markers targeting the IgH, TCRδ, and IgK-Kde rearrangements were quantitated using BIOMED-II primers and protocols (McClure et al, 2006; Fronkova et al, 2008), and results were interpreted according to the criteria of the EuroMRD Study Group (Pongers-Willemse et al, 1999; van der Velden et al, 2006).

GSTM1/T1 present/null and TYMS 28-bp enhancer repeat (rs34743033) were visualised on agarose gel directly after conventional PCR. Genotyping assays for other loci were performed using in-house allele-specific primer extension by real-time PCR. A qualified genotyping result must satisfy the following criteria: (a) at least one allele signal appears before the cycle threshold (CT) of 25; (b) for a homozygous call, CT difference between two allele signals exceeds four; and (c) for a heterozygous call, CT difference should be less than one. All bi-allelic loci studied were in Hardy–Weinberg equilibrium in each ethnic group (not applicable for GSTM1/T1 loci).

Statistical analyses

Homozygosity of the major allele in the single-nucleotide polymorphism (SNP) was used as the reference genotype unless otherwise specified. For each SNP, genotype was pooled as a dichotomous variable if applicable, and tested under both dominant and recessive models (1 degree of freedom). The grouping that yielded the smaller P-values was retained. To control for multiple testing while maintaining sufficient statistical power, we applied Benjamini–Hochberg Step-up FDR-controlling procedure for the results of preliminary test on individual SNP. Only the candidates with adjusted P-values of ⩽0.05 in multivariate tests were selected for further analyses. Subsequent investigation on the associations between shortlisted SNPs and various clinical end points was exploratory, where nominal P-values of ⩽0.05 were considered significant.

Patients who abandoned treatment for more than six continuous weeks were excluded from all analyses provided that no event occurred before the abandonment. The EFS and OS were estimated by both Kaplan–Meier (PKM by log rank test) and Cox regression (PCOX by proportional hazards model) using PASW Statistics 18.0 (IBM, New York, NY, USA). The time-to-event was calculated from diagnosis until the occurrence of resistance (morphological blast count ⩾5% at the end of remission induction), relapse (at any site), or death (from any cause), whichever occurred first, in EFS analysis (N=74; the time-to-event for resistance and induction death was registered as 1 day), or until death in OS analysis (N=39). For patients in continuous complete remission (CCR), survival times were censored at the date of last follow-up or until 31 December 2012. Cumulative incidence of relapse (including resistance; N=51) was estimated in both univariate (PG by Gray’s test) and multivariate analyses (PFG by Fine and Gray hazards model) using R (version 3.0.1) packages according to published guidelines (Scrucca et al, 2007, 2010), where death was considered as a competing event. Clinical-presenting features, that is, the race, sex, lineage, NCI risk group (the risk for infantile cases was set to ‘high’ in statistical analyses), and cytogenetic subgroup, were used as co-variables in all multivariate analyses. The genotypes were examined together with clinical features in a full model (that is, non-stepwise mode). The particular genotypes that were associated with poorer EFS (termed as risk genotypes) were also tested for their additive effect on EFS, with an increasing number of such risk genotypes. When appropriate, Day 33 (end-of-induction) MRD was further added in the regression model to test the robustness of significant results discovered.

To investigate potential sampling bias, patients enrolled were split into two halves according to the time of enrolment at each participating center, and the additive effect of risk genotypes was re-examined. In addition, as a validation we performed re-sampling to generate 100 test sets each comprising 50% of events and 50% of continuous complete remission randomly selected from the original cohort and evaluated the additive effect of risk genotypes in each set. In both scenarios, clinical co-variables were included in analyses.

Binary logistic regression was used to identify the genotypes associated with failure to achieve complete remission (that is, resistance or induction death; N=30) and treatment-related mortality (defined as the death during remission induction or in complete remission without prior leukemic relapse; N=23) after adjusting for clinical and cytogenetic features. The genotype frequencies in different categorical groups were compared by using the Pearson Chi-square test.

Haplotype frequencies of ABCB1 and IL15 genes were calculated using Haploview (version 4.2) (Barrett et al, 2005).

Results

Identification of polymorphisms of ABCB1 and IL15 as predictors of EFS

The clinical and cytogenetic features at presentation and the prevalence of alleles studied are summarised in Table 1, Supplementary Tables S3a and S3b, respectively. All bi-allelic loci were in Hardy–Weinberg equilibrium in each ethnic group (not applicable for GSTM1/T1 loci). Of the presenting features examined, unfavourable cytogenetics was associated with a significantly poorer EFS (PCOX<0.001; Table 2), while NCI high-risk status had borderline significance (PCOX=0.091; Table 2); self-reported ethnicity, gender, and cell type were not significantly related to EFS. The complete genotype data as well as patient characteristics are available for research purposes upon request.

Table 1 Distribution of patients according to demographic and prognostic categories
Full size table
Table 2 The influence of ABCB1 2677G>T/A/3435C>T and IL15 67276493G>C genotypes on EFS for patients enrolled in the Malaysia-Singapore ALL 2003 Study
Full size table

Among the 20 polymorphisms examined individually in preliminary test, polymorphisms in ABCB1 (rs2032582 and rs1045642) and IL15 (rs17015014) were found eligible for further analyses (adjusted P=0.038 and 0.048, respectively; Supplementary Table S4). They were subsequently analysed together in a Cox regression model adjusting for presenting clinical and cytogenetic features (Table 2). The homozygosity of either ABCB1 2677T/T or 3435T/T was still significantly related with a poorer EFS compared with other genotypes at these two loci (PKM=0.001; PCOX=0.003, HR=2.11, 95% CI=1.29–3.45; Figure 1A). There was moderate linkage disequilibrium (LD) between 2677T and 3435T (details shown in later section). IL15 67276493G/G was also associated with significantly poorer EFS compared with 67276493G/C and C/C (PKM=0.050; PCOX=0.022, HR=1.84, 95% CI=1.09–3.08; Figure 1B). Adjusting for MRD at the end of remission induction (Supplementary Figures S2 A–B) or excluding the cases of resistance from EFS analyses (Supplementary Table S5 and Supplementary Figures S3A–B) did not alter the significance of these polymorphisms identified.

Figure 1
figure 1

Plots of EFS probability for 413 patients enrolled in the Malaysia-Singapore ALL 2003 Study. Regression was adjusted for patients’ race, sex, lineage, NCI risk group, and cytogenetic subgroup. The patients were stratified by respective genotypes of (A) ABCB1 2677G>T/A/3435C>T, (B) IL15 67276493G>C, and by (C) the number of risk genotypes.

Full size image

The risk genotypes, that is, ABCB1 T/T at 2677 or 3435 or both as well as IL15 67276493G/G, exhibited an additive effect (PKM<0.001; PCOX<0.001, HR=1.97, 95% CI=1.38–2.82) demonstrating progressively poorer EFS in patients with an increasing number of risk genotypes (Table 2 and Figure 1C). Compared with the patients without any risk genotype, the hazard of adverse events raised 2.2 folds in patients with one risk genotype (PKM=0.005; PCOX=0.003, HR=2.20, 95% CI=1.30–3.70) and 3.6 folds in those with two risk genotypes (PKM=0.002; PCOX=0.001, HR=3.64, 95% CI=1.68–7.92). After adjusting for MRD at the end of remission induction (Supplementary Table S6 and Supplementary Figure S2C) or excluding the cases of resistance (Supplementary Table S7 and Supplementary Figure S3C), this additive effect remained significant. There was no significant EFS difference between patients with only ABCB1 risk genotype and those with only IL15 risk genotype (data not shown).

To further investigate the impact of ABCB1 2677G>T/A/3435C>T and IL15 67276493G>C genotypes on a specific event, we examined their influence on the cumulative incidence of relapse (including resistance), failure to achieve complete remission as well as treatment-related mortality. Homozygosity of either ABCB1 2677T/T or 3435T/T significantly increased the risk of relapse in both univariate (PG=0.001; Supplementary Figure S4A) and multivariate models after adjusting for clinical and cytogenetic features (PFG=0.011, HR=2.17, 95% CI=1.19–3.94; Supplementary Table S8). The risk of relapse in patients with at least one T/T was 20.4% compared with 9.0% in those without any T/T. Taking the end-of-induction MRD level into account did not alter the significance (PFG=0.027, HR=2.01, 95% CI=1.08–3.73). The risk of relapse also became higher with increasing number of risk genotypes (PG<0.001; Supplementary Figure S4B; PFG=0.011, HR=1.84, 95% CI=1.15–2.93; Supplementary Table S8). The trend remained significant after adjusting for the end-of-induction MRD level (PFG=0.026, HR=1.69, 95% CI=1.07–2.68).

IL15 67276493G/G was associated with higher risk of failure to achieve complete remission when compared with G/C and C/C genotypes (P=0.009, OR=3.14, 95% CI=1.34–7.36; Supplementary Table S9). G/G homozygosity also correlated with poorer overall survival (PKM=0.050; PCOX=0.005, HR=2.74, 95% CI=1.35–5.58; Supplementary Table S10 and Supplementary Figure S5), primarily due to more than 3-fold higher risk of treatment-related mortality in patients with IL15 67276493G/G (P=0.015, OR=3.40, 95% CI=1.27–9.11; Supplementary Table S11).

We also examined the associations of identified risk genotypes with known risk factors. Neither T/T homozygosity at ABCB1 2677 or 3435 nor IL15 67276493G/G was found to correlate with cytogenetics and MRD level after remission induction (Supplementary Table S12).

Prognostic significance of ABCB1 and IL15 in patient subgroups

In patients with favourable cytogenetics, T/T homozygosity at either ABCB1 2677 or 3435 position was associated with a significantly poorer EFS (PKM=0.011; PCOX=0.011, HR=4.02, 95% CI=1.38–11.75; Figure 2A). The EFS in patients carrying ABCB1 2677T/T or 3435T/T was 78.2%±7.6%, compared with 94.4%±1.9% in those carrying other genotypes. Such an association became even stronger after adjusting for MRD level (PCOX<0.001, HR=19.26, 95% CI=4.43–83.72; 80.6%±8.6% vs 98.9%±0.8%). In contrast, the impact of IL15 67276493 genotype was more pronounced in patients who do not carry chromosomal abnormalities of known prognosis, showing a significantly worse EFS in G/G carriers than the others (PKM=0.038; PCOX=0.022, HR=2.20, 95% CI=1.12–4.30; Figure 2C), with EFS of 73.9%±5.6% for G/G carriers and 86.7%±2.8% for G/C and C/C carriers, respectively. Adjustment for MRD level did not moderate the significance (PCOX=0.022, HR=2.30, 95% CI=1.13–4.70; 78.6%±5.4% vs 90.2%±2.7%).

Figure 2
figure 2

Plots of EFS probability for patients without high-risk cytogenetic lesions. Regression was adjusted for patients’ race, sex, lineage, and NCI risk group. The patients were stratified by (A) ABCB1 2677G>T/A/3435C>T (N=184) and (B) the number of risk genotypes (N=174) in favourable cytogenetic subgroup, and (C) IL15 67276493G>C (N=209) and (D) the number of risk genotypes in the subgroup without common chromosomal abnormalities (N=208).

Full size image

In both cytogenetic subgroups, the additive effect of risk genotypes was found (PKM<0.001 and PCOX<0.001, Figure 2B; and PKM=0.013 and PCOX=0.009, Figure 2D) and remained significant after adjusting for MRD level (PCOX<0.001 and=0.008 respectively). In addition, we also examined the influence of risk genotypes in the two major ethnic groups, the Chinese and Malays, respectively. In both races, the additive effect of risk genotypes was also found (PKM<0.001 and PCOX=0.001 in the Chinese, Supplementary Figure S6A; PKM=0.075 and PCOX=0.036 in the Malays, Supplementary Figure S6B) and remained significant after adjusting for MRD level (PCOX=0.003 and=0.006, respectively).

Validation tests

Patients were split into two halves based on the time of enrolment at each participating centre. In both cohorts, the additive effect of T/T homozygosity at ABCB1 2677 or 3435 and IL15 67276493G/G still remained significant (Supplementary Figure S7A, C), even after adjusting for MRD levels at the end of induction (Supplementary Figure S7B, D). In addition, we generated 100 random test sets by re-sampling. Of these 100 test scenarios, the significance was achieved for 86 test sets (Supplementary Figure S8A), and all results consistently showed that increasing number of risk genotypes corresponded to a poorer EFS (i.e., HR >1; Supplementary Figure S8B).

Linkage disequilibrium

As multiple loci in ABCB1 and IL15 were genotyped, we calculated the linkage disequilibrium (LD) among different alleles. Two loci, ABCB1 2677 (rs2032582) and ABCB1 3435 (rs1045642), were in a moderate LD block. The LD was slightly stronger in the Chinese (r2=0.58 for entire patient cohort and r2=0.74 for the Chinese; Supplementary Figure S9), with two major haplotypes, 2677G-3435C (52.3%) and 2677T-3435T (36.2%). IL15 67276493G>C (rs17015014) was not in LD with any other studied loci of the IL15 gene (Supplementary Figure S10).

Discussion

In childhood ALL, there is significant inter-individual variation in response to chemotherapy, which is not fully predicted by presenting features such as age and leukaemic cytogenetics. Newly described markers such as IKZF1 and CRLF2 have demonstrated prognostic impact in many studies (Mullighan et al, 2009; Buitenkamp et al, 2012; Chen et al, 2012; Feng and Tang, 2013), but their values still remain inconclusive in the Malaysia-Singapore ALL 2003 study according to our pilot investigations (data not shown). Here we showed that host germline variants of ABCB1 and IL15 genes are significant predictors of treatment outcome, suggesting that the characterisation of these genes’ variants at diagnosis could contribute to risk-stratification strategies. In our study, the ABCB1 2677T/T or 3435T/T genotypes were associated with an increased risk of relapse/resistance. ABCB1 2677G>T/A introduces an Ala-to-Ser/Thr change in exon 21 while 3435C>T is a synonymous polymorphism in exon 26, which affects transport by altering substrate specificity (Kimchi-Sarfaty et al, 2007). Various studies have shown that not all synonymous SNPs are considered silent; they can cause changes in protein expression, conformation or function and are increasingly implicated in human disease risk and other complex traits (Sauna and Kimchi-Sarfaty, 2011).

Large studies in patients with acute myocardial infarction on oral anti-coagulant clopidogrel, a substrate of P-glycoprotein (P-gp), reported lower drug levels and poorer cardiovascular outcomes in patients with ABCB1 3435T/T homozygotes (Simon et al, 2009; Mega et al, 2010). Oral dexamethasone, used extensively in paediatric ALL treatment including our Ma-Spore ALL 2003 protocol, is also a known substrate for P-gp. Patients with ABCB1 2677T/T or 3435T/T may have lower systemic dexamethasone exposure due to higher ABCB1 expression on the apical villi of enterocytes (Nakamura et al, 2002), causing enhanced extrusion of drugs into the intestinal lumen. Supporting this hypothesis, Yang et al (2012) recently reported two ABCB1 variants (rs10264856 and rs4728709) that affected oral dexamethasone clearance and were associated with increased risk of relapse in the St Jude and COG cohorts, although we did not observe any LD between rs10264856/rs4728709 and rs2032582/rs1045642 in our study as well as the data from 1000 Genomes Project (the maximum r2 is only 0.14). Taken together with our findings on ABCB1 2677 and 3435, these strongly suggest the importance of ABCB1 variants in treatment of childhood ALL.

However, Jamroziak et al (2004) and Stanulla et al (2005b) previously reported lower EFS (N=111) and higher CNS relapse (N=34), respectively, in ABCB1 3435C/C carriers treated on the BFM-ALL 86/90/95 protocols. Differences in the type of steroid used during induction therapy (prednisolone in the BFM-ALL 86/90/95 compared with dexamethasone in Ma-Spore ALL 2003) may account for this observed difference. In addition, the lower frequency of ABCB1 2677T and 3435T alleles among African Americans (10.0% and 20.2%, respectively) (Kroetz et al, 2003) compared with our studied population (43.6% and 40.5%, respectively) may mask the effect of ABCB1 variants on treatment outcome in the previously reported St Jude Total XIIIB protocol (Rocha et al, 2005).

The manner that ABCB1 variants affect the P-gp expression on the apical villi of enterocytes remains unresolved (Lin and Yamazaki, 2003; Leschziner et al, 2007; Hodges et al, 2011; Wolf et al, 2011). Nakamura et al (2002) reported that healthy Japanese subjects with 3435T/T had significantly higher duodenal ABCB1 mRNA expression and lower digoxin plasma level compared with C/C or C/T genotype carriers, while Hoffmeyer et al (2000) and Efferth et al (2003) disagreed with such a correlation. The effect of ABCB1 polymorphisms on P-gp activity is likely dependent on the substrates as well as the tissues (or cell types) studied, thus becoming a confounding factor when studying P-gp genotype–phenotype relationship. Further studies in correlating dexamethasone clearance and ABCB1 polymorphisms are required.

Considering the drug efflux capacity of P-gp, it is rational to speculate that the impact of ABCB1 genotype on the risk of relapse/resistance originates from its effect on MRD level. Although not statistically significant in our patients, Day 33 MRD high risk (i.e., ⩾10−2) tended to be enriched in the ABCB1 T/T group compared with the others (14.9% vs 8.8%, P=0.088, Supplementary Table S12). Published data on the relationship between ABCB1 gene and MRD are controversial. For instance, an earlier report showed that the median ABCB1 expression at diagnosis of childhood ALL were roughly comparable between the MRD-positive group and MRD-negative group at the end of induction therapy (Fedasenka et al, 2008). This finding was challenged by a very recent study reporting that mRNA expression of ABCB1 higher than a certain cutoff would raise the risk of MRD positivity by 12-fold after 1-year treatment in children with ALL (Rahgozar et al, 2014). Such discrepancies are likely due to (but not limited to) the diverse grouping criteria, the quantitation methods applied, the time points investigated, and sampled populations in different studies. As the determinants on early treatment response are known to be multifactorial, a clear correlation between ABCB1 genotype and MRD level may not be expected.

The IL15 67276493G/G genotype was associated with a poorer EFS and OS in our study, and carried a higher probability of treatment-related mortality. IL15 regulates T- and natural killer cell activation and proliferation, having central roles in cell-mediated immunity against microbes (Yoshikai and Nishimura, 2000). Administration of exogenous IL15 in animal models causes severe but reversible neutropenia due to redistribution of neutrophils out of the circulation (Berger et al, 2009; Waldmann et al, 2011). This SNP resides in the 3′UTR region of the gene, which exerts a negative regulatory effect on the expression of IL15; polymorphism at the 3′UTR may reduce this negative regulation, resulting in an enhanced expression. It is noteworthy that genetic variants in IL15, albeit different IL15 SNPs, have been previously inferred to significantly correlate with the MRD level at the end of induction (Yang et al, 2009), a strong surrogate marker to predict treatment outcome.

ABCB1 3435T/T or 2677T/T and IL15 67276493G/G showed a surprising additive effect in predicting a poorer EFS probability, even after adjusting for MRD levels at the end of remission induction. P-glycoprotein mediates in transmembrane transport of cytokines. It has been reported that ABCB1 2677T/T–3435T/T are associated with lower secretion of cytokines in PHA-stimulated lymphocytes (Pawlik et al, 2005). Conversely, in HIV-infected individuals, P-glycoprotein expression and function are decreased and restored only after IL15 stimulation (Chang et al, 2000). Therefore, we speculate that ABCB1 2677T/T or 3435T/T may alter secretion of IL15, which affects cellular immune responses and treatment outcome consequently.

In summary, host genetic variations in ABCB1 and IL15 genes influence treatment outcomes in paediatric ALL in this predominantly Asian population treated on the Ma-Spore ALL 2003. Inclusion of host variants in ABCB1 and IL15 genes may help future risk assignment strategies and deserves further study in other treatment cohorts.