An Intron Splice Acceptor Polymorphism in hMSH2 and Risk of Leukemia after Treatment with Chemotherapeutic Alkylating Agents¹

t-AML³ was initially recognized >30 years ago in patients treated for multiple myeloma with melphalan (1). t-AML has since been reported subsequent to numerous primary malignancies, including Hodgkin’s disease, non-Hodgkin’s lymphoma, breast cancer, ovarian cancer, and testicular cancer (2, 3, 4, 5, 6, 7). This devastating condition now represents up to 10% of all AML cases.

The clinical characteristics of t-AML differ according to the nature of the therapeutic exposure. After treatment with DNA topoisomerase II inhibitors such as epipodophyllotoxins, t-AML has a relatively short latency period (typically up to 2 years) often with no preleukemic phase and is associated with translocations involving the MLL gene at 11q23 (8). In contrast, t-AML after treatment with chemotherapeutic alkylating agents such as nitrogen mustards, nitrosoureas, platinum-based compounds, and hydrazines is characterized by a longer latency period (up to 10 years) often presenting with myelodysplasia and is associated with loss or deletions of chromosome 5 and/or 7 and p53 mutations (9, 10). These observations suggest multiple genetic mechanisms by which t-AML may arise. The prevalence of MSI suggests that dysfunctional DNA mismatch repair may be one such mechanism of malignant transformation in t-AML (11).

The primary function of DNA mismatch repair is to correct replicative errors that escape DNA polymerase proofreading. In humans, the MutSα complex (composed of MSH2 and MSH6) and the MutSβ complex (MSH2 and MSH3) recognize and bind to single base mismatches and insertion/deletion loops in DNA (12, 13). Subsequent recruitment of MutLα (MLH1 and PMS2) facilitates degradation and resynthesis of the mispaired region in the newly synthesized daughter strand (14). Although defects in any of the proteins can impair the DNA mismatch repair system, the process is most greatly affected by abnormal MSH2 or MLH1 function (15, 16). Cells manifesting mutations as a result of dysfunctional mismatch repair are described as having a mutator phenotype. In human studies, the mutator phenotype has been demonstrated in a variety of sporadic cancers and inherited conditions, including HNPCC (17), and is associated with the presence of extensive MSI. Microsatellites are short, often polymorphic repeat sequences dispersed throughout the genome. Alterations in microsatellite length occur because of errors in replication, and instability at these sites is indicative of defects in DNA mismatch repair. MSI in hematological disorders is rare, but it has been observed in acute leukemia, including t-AML. Although Ben-Yehuda et al. (11) reported MSI in >90% of t-AML and t-MDS patients, subsequent studies have reported a lower incidence in t-AML (up to 50% of cases; Refs. 18, 19, 20, 21). Generally, the reported incidence of MSI in de novo AML (0–30% of cases) is comparatively lower than in t-AML (18, 20, 21). Critically, abnormal MSH2 expression has been reported in microsatellite unstable AML, including t-AML (19), providing additional evidence for dysfunctional mismatch repair in these cases.

Exon 13 of hMSH2 is located in the most conserved region of the gene and encodes part of the DNA binding domain (22, 23). Previous studies have identified a T>C transition polymorphism at position −6 of the 3′ splice acceptor site of exon 13 (22, 24). Polymorphisms within the 3′ splice junction consensus sequence between +1 and −16 may alter the efficiency of RNA splicing (25). The −6 exon 13 polymorphism lies within a short poly(T) tract forming an atypical 3′ splice acceptor sequence, which may weaken splice site recognition. The location of the −6 exon 13 T>C polymorphism has led to suggestions that the variant hMSH2 sequence may result in alternatively spliced mRNA. Although this has been reported (26, 27), including transcripts lacking exon 13, there appears to be no relation with the −6 T>C exon 13 polymorphism (26, 28). Thus, the functional significance of the hMSH2 −6 exon 13 polymorphism remains unclear.

Fishel et al. (22) observed heterozygosity for the −6 exon 13 transition polymorphism in the germ-line of affected individuals, but not unaffected individuals, from two chromosome 2-linked (hMSH2) HNPCC families, although this association remains unconfirmed. Subsequent studies have reported a significant association between heterozygosity for this polymorphism and susceptibility to sporadic colon cancer (24, 26, 28). In hematological studies, the −6 exon 13 transition polymorphism has been shown to be prevalent among human precursor T-cell lymphoblastic lymphoma cases (23). Also, the variant allele has been significantly associated with non-Hodgkin’s lymphoma (29).

The low frequency of the hMSH2 T>C splice acceptor polymorphism has precluded analysis of any potential disease association in the homozygous state. In this study, we have determined the distribution of the hMSH2 T>C −6 exon 13 polymorphism in a case-control series of 420 de novo AML cases and 837 matched unaffected controls. We have also determined genotype distribution in a case series of 91 individuals with t-AML. Where possible, MSI was also analyzed in corresponding samples as a marker of dysfunctional DNA mismatch repair and related to hMSH2 genotype and therapeutic exposure leading to t-AML.

The hMSH2 −6 exon 13 polymorphism has been examined in three groups of subjects; there were 91 t-AML cases, 420 de novo AML cases, and 837 unaffected controls (Table 1). For this study, t-AML is defined as AML after chemotherapy and/or radiotherapy diagnosed at least 2 months after the start of the initial cytotoxic therapy. All de novo AML cases, unaffected controls, and 24 t-AML cases were obtained as part of a United Kingdom population-based case-control study of adult acute leukemia that has been described elsewhere (30). Genotyping was performed using DNA extracted from presentation peripheral blood for all de novo AML cases and 24 t-AML cases. DNA extracted from presentation bone marrow was used for genotyping the remaining 67 t-AML cases. We acknowledge the remote possibility that blast contamination of peripheral blood or bone marrow samples may cause misgenotyping because of allelic loss in blast cells, which may cause a heterozygote to appear as a homozygote. However, we feel that this would be extremely unlikely for two main reasons: firstly, chromosome 2p (hMSH2) is rarely affected by deletion or any aberration in AML; secondly, even in patients with a high presentation blast content, we feel it is extremely unlikely that a peripheral blood or marrow sample would be completely devoid of normal constitutive DNA.

Additional DNA samples were obtained from 67 subjects with t-AML enrolled in the MRC AML trials 10, 11, and 12 (31). Eighty-nine of the 91 t-AML cases have been described in detail elsewhere (32). The remaining 2 t-AML cases were both female ages 55.4 years (FAB type M2) and 57.5 years (FAB type M5) at diagnosis of t-AML (and both had received radiotherapy only for a previous condition). Cytogenetic analysis was successful for the younger case only [50, XX, +8, +14, +21, +22 (Ref. 9)/46, XX (Ref. 1)]. All 420 de novo AML cases, 837 controls, and 69 of 91 t-AML cases were confirmed as United Kingdom Caucasians. Ethnicity data on the remaining 22 t-AML cases were not available. However, because the United Kingdom population is composed predominantly of Caucasians, it remains likely that most, if not all, of the remaining 22 t-AML cases were Caucasian. All 16 t-AML cases previously exposed to O⁶-guanine alkylating agents were confirmed as United Kingdom Caucasians. All subjects taking part in the Leukaemia Research Fund acute leukemia case-control study and the MRC AML trials 10, 11, and 12 gave informed consent.

Forward (5′-TGTGGGCAGGCTGTGGTTC-3′) and reverse (5′-CTCCCATATTGGGGCCTGCA-3′) primers were used to amplify the hMSH2 sequence flanking the −6 exon 13 polymorphism. The reverse primer was designed with a mispaired base at position 18 (corresponding to position −3 of exon 13) generating an artificial PstI restriction site upon amplification of the variant allele only (Fig. 1). PCR amplifications were performed using 0.5 units of AmpliTaq Gold (Perkin-Elmer), 10 pmol of each primer, 200 μm of each deoxynucleotide triphosphate, and 1.5 mm MgCl₂ in a 20-μl reaction mix containing 1× PCR reaction buffer and ∼50 ng of template DNA. PCR conditions were as follows: cycle 1, 15 min at 95°C; cycles 2–37, 1 min at 95°C; 1 min at 64°C, followed by 1 min at 72°C; and cycle 38, 10 min at 72°C. PCR products were digested with 20 units of PstI (New England Biolabs) at 37°C for 4 h. Digestion products were separated on a 3% agarose gel and visualized against a 100-bp ladder (Fig. 1).

To confirm that the RFLP-PCR was discriminating between polymorphic hMSH2 alleles, 60 samples were sequenced (14 TT homozygotes, 25 heterozygotes, and 21 variant CC homozygotes, as determined by the RFLP-PCR method). Firstly, DNA was amplified using forward (5′-GCAGGCTGTGGTTCTGCT-3′) and reverse (5′-TTGATTTACCTCCCATAT TGGG-3′) hMSH2 primers flanking the −6 exon 13 polymorphism, using the reaction condition described above at an annealing temperature of 60°C. Amplification products were then sequenced using the AmpliTaq Big Dye terminator cycle sequencing kit (Applied Biosystems). Sequencing electrophoresis was performed on the ABI Prism 377 sequencer, and results were analyzed using the ABI Prism 377 software. Of the 60 samples analyzed, 58 of the DNA sequencing results correlated with RFLP-PCR genotyping, confirming that the RFLP-PCR can discriminate between wild-type and variant alleles with a minimal error rate (3%). In both discrepant cases, heterozygotes by RFLP were subsequently sequenced as homozygotes (1 TT and 1 CC). Thus, to minimize the error frequency, all heterozygotes and homozygotes for the variant (C) allele and all t-AML cases irrespective of status were genotyped at least twice. In a few samples, a third assay was performed if the first two assays gave discrepant results. Thus, the final genotyping error frequency is likely to be <3%.

ORs and 95% CIs, adjusted for age and sex, were calculated using unconditional logistic regression (33). hMSH2 genotype was analyzed as a trichotomous (TT, TC, CC) or dichotomous (TT, TC + CC) variable. All analyses were conducted using Stata 1999 (Stata, College Station, TX).

Presentation bone marrow samples were available for 46 cases (Table 1), including 34 cases of t-AMLs and 12 cases of t-MDSs. Twenty-six of the t-AML cases were of United Kingdom origin and were part of the case series genotyped for the −6 exon 13 hMSH2 polymorphism. The blast content in bone marrow samples from 19 of the t-AML cases was ≥40% (median of 70%). Data regarding blast content was not available for the remaining 7 samples. Although leukemic blasts could not be separated from normal cells, a previous study shows that MSI can unambiguously be detected in samples containing low amounts of leukemic blasts (34). The remaining 8 t-AML and 12 t-MDS cases were of Finnish and North American origin and were included in prior epidemiological studies (5, 6, 10). One of the t-AML cases was male who had previously received chemotherapy for testicular cancer. The remaining t-AML and t-MDS cases were all female and had received platinum-based chemotherapy for ovarian cancer before diagnosis of t-AML or t-MDS. Data regarding blast content was available for 2 of the t-AML cases (40 and 50%) and 3 of the t-MDS cases (16, 15, and 11%) within the non-United Kingdom patient series.

Two mismatch repair-proficient (TK6 and SW480) and three mismatch repair-deficient (LoVo, MT1, and HCT116) human cell lines were used as negative and positive controls for MSI, respectively. LoVo and HCT116 have both alleles deleted for hMSH2 and hMLH1, respectively (35, 36), whereas MT1 is mutated in both copies of hMSH6 (37). DNA samples extracted from the peripheral blood of 29 subjects known not to have a malignant condition were used as controls (Table 1).

Primer pairs were synthesized using previously published sequences for BAT25 (chromosome 4q), BAT26 (chromosome 2p), and BAT16 (chromosome 7p) (38). One primer of each pair was 5′-end labeled with 6-carboxyfluorescein. PCR amplifications were performed using 0.5 units of TaKaRa LA TaqDNA polymerase (Takara Biomedicals, Shiga, Japan), 10 pmol of each primer, 0.4 mm of each deoxynucleotide triphosphate, and 2.5 mm MgCl₂, using ∼50 ng of template DNA in a 50-μl reaction containing 1× PCR reaction buffer. PCR conditions were as follows: cycle 1, 10 min at 94°C; cycles 2–37, 1 min at 94°C, followed by 1 min at 56°C for BAT16 or 55°C for BAT25 and BAT26 and then 1 min at 72°C; and cycle 38, 10 min at 72°C. Amplified PCR products were analyzed using an automated 373A DNA sequencer (Applied Biosystems), and results were analyzed using the Applied Biosystems Genescan software in accordance with manufacturers instructions. All three mononucleotide markers are quasimonomorphic, eliminating the requirement for constitutional DNA (38, 39, 40). MSI was diagnosed as the presence of a clearly discernable peak not present in the control population, as illustrated in Fig. 2. Normal cells could not be separated from blast cells and so a peak representing the stable constitutive alleles was always seen in the t-AML case DNA. In contrast, the mismatch repair defective cell lines LoVo and MT1 displayed only a single unstable peak (Fig. 2). All t-AML and t-MDS cases for whom bone marrow was available were examined for MSI at all three loci.

The hMSH2 T>C polymorphism at the −6 position of the splice acceptor site of exon 13 was not significantly associated with de novo AML (Table 2), although the de novo AML case series has a marginally higher percentage of individuals heterozygous and homozygous for the lower frequency variant (C) allele (15.7%), compared with controls (13.8%; OR, 1.17; 95% CI, 0.83–1.64; Table 2). hMSH2 polymorphic status was not associated with gender in either the control or de novo AML case series (data not shown) or with cytogenetic subgroup in the de novo AML case series (data not shown).

MSH2 is known to modulate apoptosis in response to alkylating agents, including several chemotherapeutic alkylating agents, and in particular those that alkylate the O⁶ position of guanine in DNA (41, 42). The t-AML case series described in this study has a high proportion of individuals whose leukemia developed subsequent to chemotherapeutic alkylating agents (32). Individuals heterozygous or homozygous for the hMSH2 variant allele were nonsignificantly overrepresented in the chemotherapy-related t-AML case-series, compared with controls (OR, 1.56; 95% CI, 0.75–3.21; Table 2). Sixteen of 91 t-AML cases had documented previous treatment with chemotherapeutic O⁶-guanine alkylating agents, which included procarbazine, dacarbazine, 1-(2-chloroethyl)-3-cyclohexyl-1-nitrosourea, and cyclophosphamide. Consistent with a role in conferring susceptibility to O⁶-guanine alkylating agent-induced myeloid leukemia specifically, 6 of these 16 (37.5%) cases were carriers of at least one variant hMSH2 allele, a frequency significantly higher than in both the control series (13.8%; OR, 4.02; 95% CI, 1.40–11.37; Table 2) and the de novo AML cases series (15.7%; OR, 3.10; 95% CI, 1.08–8.89). In contrast, 4 of 34 (11.8%) cases who received chemotherapy that did not include an O⁶-guanine alkylating agent were carriers of at least one variant hMSH2 allele, a frequency not significantly different from either the controls (13.8%) or the de novo AML case series (15.7%).

To investigate whether DNA mismatch repair dysfunction may be a characteristic feature of our t-AML case series, we examined three quasimonomorphic mononucleotide microsatellite markers for instability (BAT16, BAT25, and BAT26). These markers were chosen because they have been shown to be unstable in hereditary and sporadic colon cancer and some hematological malignancies (11, 40, 43, 44). Two of these markers, BAT25 and BAT26, are recommended for MSI analysis by the International Collaboration Group for HNPCC (40). Furthermore, in a study of 44 microsatellite markers and 107 colon cancers, Halford et al. (45) demonstrated that instability at BAT25 and BAT26 is 99 and 98% specific, respectively, for high-grade MSI (where >30% of all markers are unstable) and is indicative of DNA mismatch repair dysfunction.

MSI was detected in 13 of 34 (38%) t-AML cases for whom presentation bone marrow was available (Fig. 3, Table 3). In 12 of 13 MSI positive t-AML cases allelic shortening was found at BAT25 (Figs. 2 and 3). Two samples displayed instability at BAT26, one of which was also unstable at BAT25 (Table 3). Allelic shortening at BAT25 and BAT26 varied between 5 and 9 bases and 3 and 4 bases, respectively, consistent with published data (38, 39). MSI was not detected at BAT16 in any of the t-AML samples investigated. MSI was not detected at any locus in any of the 29 control DNA samples investigated or in either of the 2 DNA mismatch repair-proficient cell lines (TK6 and SW480) investigated. In contrast, MSI was detected in all three DNA mismatch repair-deficient cell lines. LoVo displayed MSI at BAT25 and BAT16. Amplification at BAT26 was not possible because of biallelic deletion. MT1 and HCT116 were both unstable at BAT25. MT1 was also unstable at BAT26. Instability was not detected at any of the three markers investigated in any of the 12 t-MDS cases.

All 13 MSI-positive t-AML cases were of United Kingdom origin and were genotyped for the hMSH2 −6 exon 13 polymorphism. Three of the 13 (23%) MSI-positive t-AML cases were carriers of at least one variant hMSH2 allele (Table 3), and 2 of these 13 cases (15.4%) were homozygous for the variant hMSH2 allele, a frequency considerably higher than in both the control population (1.6%) and de novo AML case population (2.5%). One case (case 56) had been treated with alkylating agent-based chemotherapy and radiotherapy for Hodgkin’s disease (Table 3). The other case (case 68) had been treated with 5-fluorouracil and radiotherapy for colorectal cancer (Table 3). Of the remaining 21 MSI-negative t-AML cases for whom presentation bone marrow was available, 13 were of United Kingdom origin and were genotyped for the hMSH2 −6 exon 13 polymorphism. Twelve of these 13 cases were homozygous for the common (T) allele, whereas the remaining case was heterozygous (data not shown). Thus, in the MSI-negative t-AML case population, the frequency of individuals with at least one copy of the variant hMSH2 allele (7.7%) was not significantly different to either the control (13.8%) or the de novo AML case population (15.7%).

A causal role for chemotherapeutic alkylating agents in t-AML etiology is well established. Furthermore, DNA mismatch repair status modulates apoptosis in response to alkylating agents, where loss of functional DNA mismatch repair confers tolerance to the apoptotic effects of some alkylating agents, particularly O⁶-guanine alkylating agents (15, 16, 41, 42). DNA mismatch repair-defective cells have a concomitantly high rate of both spontaneous and alkylation-induced mutagenesis (42, 46). Thus, dysfunctional mismatch repair confers not only tolerance to the killing effects but also sensitivity to the mutagenic effects of alkylating agents, consistent with the acute susceptibility to alkylation-induced tumorigenesis in DNA mismatch repair-defective mice (47). Our finding that the variant hMSH2 allele is significantly overrepresented in t-AML cases who specifically had previous treatment with chemotherapeutic O⁶-guanine alkylating agents is suggestive that this allele may confer moderate alkylation tolerance with concomitant susceptibility to t-AML. However, we cannot exclude the possibility that polymorphic hMSH2 may modulate susceptibility to the development of primary cancer or be involved in determining response to therapy for that cancer. Either of these may affect the prevalence of the hMSH2 polymorphism in the population at risk of subsequently developing t-AML.

The demonstration of MSI in 13 of 34 (38%) t-AML cases investigated confirms a role for compromised DNA mismatch repair in the pathogenesis of a subset of t-AML cases. These data are consistent with reports of MSI and aberrant MSH2 expression in other t-AML case series (11, 19, 21, 43). Furthermore, our data are consistent with that of Ben-Yehuda et al. (11), Olipitz et al. (43) and Gafanovich et al. (44), who also used the quasimonomorphic mononucleotide markers BAT25 and BAT26 and reported MSI at these markers in 30–60% of t-AML and t-MDS cases. Instability at these loci is highly predictive of extensive high-grade MSI (greater than 30% of marker instability; Ref. 45) and dysfunctional DNA mismatch repair and is very rarely seen in tumors that have low-level or no MSI (38, 39, 45). Thus, the high specificity of BAT25 and BAT26 (99 and 98%, respectively; Ref. 45) for high-grade MSI demonstrates dysfunctional mismatch repair in a significant proportion of the t-AML cases described in this study.

The high sensitivity of BAT25 and BAT26 (78 and 90%, respectively; Ref. 45) to detect high-grade MSI and the relatively poor concordance between these two markers in our study suggests that in our hands, BAT26 is underestimating MSI. As such, we cannot exclude the possibility that we have not been able to detect high-grade MSI in t-MDS, as reported by others (43, 48), because of insensitive MSI analysis and low or undefined presentation bone marrow blast content.

Of the 13 t-AML cases with MSI, 9 had prior exposure to chemotherapy, predominantly alkylating agents and base analogues, and 4 had received radiotherapy only. Moreover, 3 cases had an apparently normal karyotype, whereas the remaining included cases with balanced and unbalanced aberrations, including 6 cases with chromosome 5 and/or 7 aberrations. Thus, the clinical and cytogenetic heterogeneity of the MSI-positive t-AML cases supports the notion of multiple pathways by which MSI may arise in t-AML and suggests a role in hematopoietic progenitor cells for DNA mismatch repair in mediating response to mechanistically diverse genotoxic agents. Consistent with this, deletion of Msh2 in murine stem cells confers tolerance to the killing effects of chemotherapeutic alkylating agents, chemotherapeutic base analogues, and ionizing radiation (49, 50, 51). Our observations suggest a hypothesis whereby polymorphic hMSH2 weakly attenuates DNA mismatch repair function and confers moderate tolerance to the killing effects and concomitant susceptibility to the mutagenic effects of therapy, particularly O⁶-guanine alkylating therapy. As such, individuals carrying variant hMSH2 may be more susceptible to the induction of therapy-induced DNA mismatch repair-disabling mutations, giving rise to a true mutator phenotype and high-grade MSI. Homozygosity for the hMSH2 variant in 2 of 13 MSI-positive t-AML cases provides some support for this model.

We thank Rachel Clack (Clinical Trials Service Unit, Radcliffe Infirmary, Oxford, United Kingdom), William G. Thilly (Massachusetts Institute of Technology), David Swirsky (Hematological Malignancy Diagnostic Service, Leeds General Infirmary), and the staff of the Leukemia Research Fund Epidemiology and Genetics Unit at Leeds University for their expert assistance and provision of materials. We also thank Stephen Langabeer at the MRC DNA/RNA bank for his support, and University College London Hospital, a facility funded by the Kay Kendall Leukemia Fund.