Multilocus Association of Genetic Variants in MLL, CREBBP, EP300, and TOP2A with Childhood Acute Lymphoblastic Leukemia in Hispanics from Texas

Acute lymphoblastic leukemia (ALL) is a heterogeneous disease characterized by the predominance of immature hematopoietic cells, in which malignant cells express phenotypes of either T-cell or B-cell lineages (1). ALL accounts for 25% to 30% of all diagnoses of cancer in children (2, 3). The incidence of lymphoid leukemia has been found to be approximately 15% higher among Hispanic children than among White non-Hispanic children (4). No biological factors have been identified which account for this difference in risks among different ethnic or racially defined populations. Substantial evidence exists indicating that several risk factors (including gender, age, ethnicity, exposure to radiation, genetic syndromes) are involved in increased risk of ALL (5–9). Hispanic children with ALL have been found to have a higher incidence rate for secondary or therapy-related acute myeloid leukemia (t-AML) after treatment with topoisomerase II (topo II) inhibitors (10). Topoisomerase IIα (TOP2A) is a key enzyme in DNA replication and a molecular target for many anticancer drugs (11). In breast cancer, TOP2A gene amplification and deletion are associated with increased and decreased sensitivity to topo II inhibitors, respectively (12). A possible role of TOP2A in the development of leukemia has been suggested by Guerin and colleagues (13), who found overexpression of TOP2A in ALL cases. Because of the frequent finding of rearrangements of the mixed lineage leukemia gene (MLL) in therapy-related leukemias (14, 15), particularly those associated with previous therapies that included a topo II inhibitor (16), it was hypothesized that translocations could be caused by binding of topo II poisons to topo II cleavage sites. Such topo II consensus sites have been identified in the 8.3 kb breakpoint cluster region (BCR) of the gene (15), encompassing exons 8 to 14 [originally designated exons 5–11, where the breakpoints typically occur (17)]. The MLL gene encodes a histone H3 lysine 4–specific methyltransferase which is essential for normal hematopoiesis and mammalian development via the regulation of homeobox (HOX) genes (18). Previous studies have shown that overexpression of certain HOX genes can induce leukemia in mice (19).

Alternatively, chromatin structure and DNA accessibility have been suggested to be the critical determinants of cleavage within MLL resulting in translocations (20–22), a hypothesis that is corroborated by the findings of Khobta and colleagues (23), who observed specific histone modifications and acetylase/deacetylase within the BCR of MLL. In this context, the cyclin AMP response element binding protein gene (CREBBP, also known as CBP) located at 16p13.3 and its paralogue, the E1A-binding protein gene (EP300, also known as p300), located at 22q13 are of interest (24). Both genes have been characterized as fusion genes of MLL translocations in t-AML (25–28) or occasionally in de novo AML (29). CREBBP and EP300 are transcriptional coactivators that play an essential role in transcriptional regulation by chromatin remodeling through their own histone acetyltransferase activity and by interacting with transcription factors and other histone acetylators (30). A potential regulatory role for CREBBP/EP300 for protein acetylation in base mismatch repair and in maintaining genomic stability has been previously suggested (31, 32). A study by Shigeno and colleagues further found CREBBP/EP300 mutations in leukemias indicating that genetic alterations in these genes might be involved in leukemogenesis (30). Furthermore, Rubinstein–Taybi syndrome, characterized by mutations in CREBBP, is associated with increased risk to developing cancer, including leukemia (33). In mice, it was showed that EP300 and CREBBP function as suppressor of hematologic tumor formation (34).

Because of the higher risk in Hispanics to develop secondary leukemias with translocations involving 11q23 after treatment with topo II inhibitors, we sought to determine the potential role of genetic variants in TOP2A, MLL, CREBBP, and EP300 in the increased sensitivity of Hispanic children with ALL to topo II inhibitors. We investigated possible association between 52 tagged single nucleotide polymorphisms (SNP) covering the 4 genes in 241 Hispanic children with (n = 66) and without (n = 175) ALL and conducted a similar analysis in a White case–control sample to compare findings.

With a multicausal etiology of ALL, it is likely that multiple risk variants act simultaneously, synergistically or additively, to influence ALL susceptibility. We therefore determined associations of individual SNPs and higher-order SNP–SNP interactions. Our results suggest that common genetic variants in MLL and EP300, both individually and jointly, are associated with an increase in risk for ALL in Hispanics.

Consecutive case DNAs were obtained via blood draw from 66 Hispanic and 137 White non-Hispanic (referred to as “White”) unrelated children that were diagnosed with B-lineage ALL between 1990 and 2005 at the Children's Medical Center of Dallas (n = 196) and from a pilot study project between 2007 and 2009 at The University of Texas Health Science Center at San Antonio (n = 7). Collection of blood samples and clinical information from cases and controls was undertaken with signed informed consent from the parents of each child and ethical review board approval at both universities in accordance with the principles of the Declaration of Helsinki. A total of 414 age, gender, and ethnicity-matched unrelated control samples were obtained from surplus deidentified blood samples from children after emergency room treatment for diagnoses other than cancer at the laboratory services of Children's Medical Center including 175 Hispanics and 239 Whites. Exclusion criteria for the controls consisted of any history of malignancy as confirmed by review of medical records. Matching of the sample was done on frequency of the whole group, not individually. Ethnicity was self-reported as determined by parental report. The Hispanics were primarily children of Mexican descent, all with Hispanic surnames. To our knowledge, there was only 1 self-declared Hispanic case, which had 1 non-Hispanic parent. All samples underwent standard DNA preparation procedures and DNA was stored at 4°C before genotyping. Characteristics of the study group are summarized in Table 1.

SNPs were selected from the dbSNP database Build 129 (35) and SNPper (36). First, SNPs with potential functional effects and a minor allele frequency (MAF) more than 0.05 were defined from programs predicting functional annotation (F-SNP; ref. 37), PupaSuite (38), SNPinfo (39), and SNPs associated with leukemia, and in particular childhood ALL, as derived from literature (40), were determined. After this initial selection, tagging SNPs within each gene were identified by using Haploview (41) with the following criteria: (i) inclusion of preselected SNPs (see above), (ii) MAF more than 0.05 to gain more statistical power, (iii) r² threshold of 0.8, (iv) a log of odds threshold for multimarker testing of 3.0, (v) a minimum distance between tags of 60 bp, and (vi) the 2- and 3-marker haplotype tagging option. For each gene the search for tagging SNPs extended to a 10 kb region surrounding the gene. Because of limited amount of genotype data on Hispanics in dbSNP Build 129, the selection of tagged SNPs was based on the information on the European population. Genotyping was done with the GoldenGate assay of the VeraCode technology by using the BeadXpress reader system according to the manufacturer's protocol (Illumina). Duplicate samples were included as quality controls and the reproducibility rate was more than 99.8% for all SNPs. The observed error rate of duplicate samples was due to the fact that in 2 occasions 1 sample of a duplicate had too low a signal to being called by the program. An average 98% of the samples were successfully genotyped per SNP.

Allele frequencies among the cases and controls in each ethnic group were compared by using the χ² test. The OR and its 95% CI was estimated by logistic regression by using R statistical software version 2.9.1. All statistical tests were 2-sided and significance was set at P < 0.05. We used additive, dominant, and recessive models in the test analysis and chose the model with strongest association for presentation (i.e., model with smallest P value with minimum 5 individuals). The cumulative effect of combined genotypes was estimated by counting the number of genotypes associated with ALL on the basis of the best-fitting genetic inheritance from single SNP analysis. ORs and their 95% CIs were calculated for carriers of any combination of 1, or any combination of 2 or more alleles associated with ALL as compared with carriers of none of the risk genotypes by using the unconditional logistic regression analysis. The total number of risk alleles for each subject was treated as an additive risk factor, and P_trend is the P value for the association between the number of risk alleles and disease outcome. Logistic regression was used to calculate ORs of the haplotypes by using the method implemented in the haplo.ccs package in R (42). Only major haplotypes, with an estimated frequency of more than 5%, are considered in this report. The model was fit for each major haplotype so that the OR of each major haplotype was computed relative to a reference group consisting of all other haplotypes including rare haplotypes. All analyses were stratified by ethnicity and Native American ancestry was used as covariate in the analysis of the Hispanic sample. Haploview was used to measure linkage disequilibrium and to define the block structure between the tagging SNPs for each ethnicity. To correct for multiple testing in the SNP analysis, we used the method of Storey and Tibshirani on the basis of the concept of false discovery rate (FDR; ref. 43). The FDR estimation showed that for P < 0.05, the probability that the association is a true positive is more than 60% and for P < 0.006 the probability is more than 90%. The Bonferroni correction was used for multiple comparisons in haplotype analysis in which the adjusted P value is given by P at the 0.05 significance level divided by the number of major (frequency >5%) haplotypes.

Estimates of genetic ancestry of each individual in Hispanics were derived from 96 ancestral informative markers by using a maximum likelihood method implemented in the program Maximum Likelihood Individual Admixture Estimation, as described previously (Beuten and colleagues, in press; ref. 44). The ancestral proportions were estimated at 46% European, 47% Native American, and 7% West-African. To determine population stratification genetic ancestry was used as a covariate in the SNP association analysis and compared with the unadjusted outcome.

A total of 52 tagged SNPs were initially selected and genotyped in each sample. Ten SNPs were not polymorphic or failed genotyping and were left out from the study. All of the remaining 42 markers analyzed were in Hardy–Weinberg equilibrium in control samples. Table 2 shows the MAF distribution in the cases and the controls in each ethnic group. A significant difference in allele frequency between cases and controls was found for 5 SNPs in Hispanics; 2 (rs525549 and rs6589664) in the MLL gene and 3 (rs5758222, rs7286979, and rs20551) in the EP300. In Whites, none of the SNPs showed a significantly different allele frequency between cases and controls. Several markers showed a significant difference in allele frequencies across ethnic groups. In particular for variants within EP300 highly significant allele frequency differences were found between the control groups of Hispanics and Whites (Table 2). High pairwise linkage disequilibrium (LD) was found between SNPs within each of the 4 genes in both ethnicities (Supplementary Table S1). Several pairwise LD measures within each gene were significantly different between Hispanics and Whites. For example, perfect LD between rs9611502 and rs20551 or rs20552 in EP300 was found in Whites, whereas both combinations showed low LD in Hispanics.

Five SNPs showed a significant association with ALL in Hispanics, of which rs525549 (MLL) and rs20551 (EP300) were significant at α = 0.05 (FDR <40%) and rs65896642 (MLL), rs5758222, and rs7286979 (EP300) at α = 0.006 (FDR < 10%; Table 3). The strongest effect was seen for which homozygous AA carriers at rs525549 had a 3.35-fold increase in ALL risk (OR = 3.35; 95% CI = 1.37–8.18; P = 0.008). Results shown in Table 3 are not adjusted; however, inclusion of age or gender as covariate showed very similar results (data not shown). We also tested whether population stratification was confounding the outcome. Including individual Native American ancestral estimates as covariate in the regression analysis, indicated that significance remained for all 5 SNPs, although was slightly reduced for the 2 MLL variants (Table 3). A similar observation was found after correction for European ancestral proportions, whereas the outcome after adjusting for African ancestry did not differ from unadjusted results (data not shown).

Replication of the association findings in a White sample group showed that only rs65896642 within MLL was significantly associated with ALL (P = 0.029, Table 3). Of note is that for rs6589664 the A allele is the minor allele in White controls, whereas it is the major allele in Hispanic controls (the allele frequency of the controls is used as reference). However, G is the risk allele in both ethnic groups; in Whites, homozygous AA carriers have a 2.50-fold decrease in ALL risk whereas homozygotes for the G allele show a 2.32-fold increase in ALL risk in Hispanics. Results from logistic analysis for all SNPs by inheritance models and by ethnicities are shown in Supplementary Table S2.

Multivariate logistic regression on combinations of multiple risk alleles defined in the SNP analysis as compared with the combination with no risk allele as reference, showed cumulative risk effects. Because rs5758222 and rs7286979 in EP300 are in complete LD, we only included one of both SNPs in the model. An interaction was found for rs525549 and rs6589664 in MLL and rs5758222 in EP300, with a more than 16-fold increase in ALL risk (OR = 16.66; 95% CI = 4.48–61.9; P_trend < 0.0001; Table 4). It was noted that including rs7286979 instead of rs5758222 in the model, showed a similar outcome (data not shown). Moreover, adding rs20551 in the model also resulted in a significant cumulative effect. However, the 95% CI of the OR was much wider because of the absence of individuals with zero-risk genotypes (used as reference).

To examine the joint effect of risk alleles, the 5 SNPs that were found to be significantly associated with ALL from SNP analysis were included in haplotype analysis. A major haplotype A-G-G-A-A (6%), from the interaction between haplotypes of MLL and EP300, is significantly associated with ALL with an adjusted OR of 5.68 (95% CI = 2.82–11.44; P < 0.0001) and carries all 5 alleles associated with increased risk in SNP analysis (Table 5). This association remained significant after Bonferroni correction (the adjusted P value at the 0.05 significance level for 6 major haplotypes is 0.0083).

Because of our previously observed predilection for Hispanic children to develop secondary leukemias after treatment with topo II inhibitors, often showing 11q23 translocations, we focused on examining the association of variants within TOP2A, MLL, and 2 of its fusion partners, CREBBP and EP300, hypothesizing that variation in these genes could influence leukemia risk in general in this population, and estimated risks conferred by both individual SNPs and SNP–SNP joint effects. No association studies of genetic variants within these genes with ALL susceptibility have been reported to date. Only very few association studies of candidate gene in Hispanics have been reported so far (45–48). Being the fastest-growing cultural group and second-largest racial/ethnic group of the population in the United States, genetic studies in Hispanics, such as our report, are important steps in understanding the unique cancer risks in this susceptible population.

Genotyping of 52 haplotype tagged SNPs covering the 4 genes showed that 2 SNPs within MLL and 3 within EP300 were significantly associated with ALL (P = 0.001–0.04). The most significant association was found for rs525549 (OR = 3.52; P = 0.004), located in intron 11 of the MLL gene. The breakpoints for most MLL rearrangements, in particular in t-AML patients with previous exposure to drugs targeting topo II, cluster in the BCR encompassing exons 8 through 14 (15). The association finding for rs525549 in Hispanics is therefore of particular interest because this ethnic group has been shown to be at increased risk to develop secondary leukemias related to therapy with topo II inhibitors (10). The BCR is characterized by multiple topo II consensus sites and rs525549 is 2 base pairs upstream from a 90% matching consensus binding site for topo II. Moreover, rs525549 is less than 1.5 kb upstream and in high LD with the region encompassing exon 12 [or exon 9 according to an older numbering system (17)] containing a strong in vivo topo II consensus binding site (49) and less than 1.3 kb from a translocation hotspot (22) as well as a gene-internal promoter having a binding site for RNA polymerase II likely implicated in genetic (in)stability of the MLL gene (50). The association finding of rs525549 in Hispanics, therefore suggests that genetic variation in MLL could be of biological significance and be implicated in the increased risk of t-AML in Hispanics.

Replication in Whites revealed that only SNP rs6589664 (MLL) was found to be significant and the A allele is the protective allele for ALL in both ethnic groups. Rs6589664 is located downstream of the MLL gene and maps within exon 7 of the mRNA-transmembrane protein 25 gene (TMEM25), adjacent to MLL. TMEM25 is a member of the immunoglobulin superfamily, implicated in immune responses, growth factor signaling, and cell adhesion (51). No data have been reported on a potential role of TMEM25 in leukemogenesis and further studies are warranted to identify its contribution to ALL. Alternatively, high LD exists between rs6589664 and other SNPs within MLL, and rs6589664 therefore might be merely a marker for the true causal factor.

We also found 3 variants within EP300 (rs5758222, rs7286979, and rs20551) that were significantly associated with ALL in Hispanics. None of these variants have been previously reported in association studies with childhood ALL. The population-specific allele frequencies, showing significant differences in distribution between Hispanics and Whites, might explain the lack of statistical significance in Whites for genetic variants located within EP300. Of note is that perfect LD between rs9611502 and rs20551 or rs20552 was found in Whites, whereas both combinations showed low LD in Hispanics, suggestive for higher recombination rate in this region of the gene for the latter group.

Our study further showed a significant dosage effect with increasing numbers of potential high-risk genotypes. Combining 3 high-risk genotypes of MLL and EP300 in Hispanics showed synergistic effects; the risks conferred by multiple polymorphisms were elevated as compared with any of them alone (OR = 16.66; P < 0.0001). From these findings, we conclude that an interaction effect between different loci is operative and likely confers ALL risk. Such interaction was further confirmed by the interaction between haplotypes of both genes. Caution should be used when interpreting the results as to the biological meaning of these multilocus interactions; statistical significance does not necessarily correspond to true biological interactions between the specific genes studied (52). However, our findings are supportive for cooperative roles of EP300 and MLL in leukemogenesis and show for the first time a genetic interaction between MLL and EP300 in Hispanic children with ALL.

Significant heterogeneity exists among Hispanics in the United States and worldwide, however, the individuals designated self-described as Hispanic in this study virtually all derive from Mexico across the U.S. border in South Texas. Therefore, any possible race/ethnicity misclassification would be equally likely in cases and controls of each ethnic group and should therefore not have a substantial impact on our conclusion. In addition, after including the proportion of Native American ancestry as covariate, determined by information retrieved from ancestral informative markers, the associations did not change for variants within EP300 and were slightly reduced for variants in MLL. Therefore, these associations are true positives and the confounding effects because of admixture are insignificant.

Recently, genome-wide association studies in Europeans and Koreans have revealed that genetic variants play a role in childhood ALL (53–55). However, Hispanics have not been analyzed extensively at a genome-wide level. This may be explained by the fact that genotype information on Hispanics has only recently become more abundantly available in HapMap. For that reason, our selection of the 52 tagged SNPs was based on HapMap data of the European population. Because of the ethnic-specific LD patterns, it is possible that some tagged SNPs in Hispanics were omitted and the SNPs selected for our study may therefore not fully represent all tagged variants in this ethnic group. However, our selection of SNPs was enriched with potentially functionally important SNPs and thus high-priority SNPs were included in the analysis.

A limitation of the study is that we only evaluated pre-B ALL and did not consider other phenotypic subtypes neither did we examine interactions with nongenetic factors that may be involved in disease risk. In addition, the statistical power of this study is limited by the sample size, in particular for the analysis of the joint effects of risk alleles. The minimal detectable OR, corresponding to 80% power, is 2.05 considering a median MAF of 15% for our Hispanic sample of 66 cases and 175 controls. Therefore, replication of our findings in independent Hispanic samples is warranted to elucidate the role of these variants in ALL susceptibility and define their importance in the ethnic specific differences in ALL risk.

In summary, our results support the hypothesis that ALL develops through a variety of interrelated genes (polygenic model) and suggest that common genetic variants within MLL and EP300 genes are associated with ALL in Hispanics from Texas. The susceptibility defined by these genetic variants is population specific. To our knowledge, this is the first report to explore the effect of higher-ordered interactions between these genes and ALL risk. Our findings might help to improving prediction risk of an individual for ALL, or in explaining leukemia risks at the population level.

The authors have declared that no competing interests exist. The funders had no role in study design, data collection, and analysis, decision to publish, or preparation of the manuscript.

We thank family members for their participation and clinical staff for their skilled assistance.

This work was supported by the Texas Higher Education Coordinating Board Grant #010019–0105-2001, the Children's Cancer Fund Dallas, Texas and the Greehey Family Foundation (to G.E. Tomlinson), and the Commission on Higher Education Scholarship, Thailand (to D. Piwkham).

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.