Background
The role of aberrant DNA methylation in the development of cancer is well recognized and documented. Tumor suppressor gene (TSG) inactivation by promoter region CpG island hypermethylation occurs in almost all cancer types and is an important mechanism of gene silencing in cancer. Unlike genetic changes in cancer, epigenetic changes are potentially reversible. Epigenetic therapy is a rapidly expanding field and a number of drugs that alter the epigenetic profiles of cancer cells are already in clinical trails. Two hypomethylating agents, 5-azacitidine (Vidaza) and 5-aza-2'-deoxycytidine (Decitabine) are currently in use and are approved therapies for myelodysplastic syndrome [
1,
2].
RASSF1A TSG is a classic example of a gene that is frequently methylated in the majority of adult and childhood cancers including epithelial and blood borne cancers [
3]. The
RASSF1 family of genes now includes 10 members (
RASSF1-10). We have recently described an epigenetic profile of the
RASSF1-10 genes in childhood acute lymphoblastic leukemia (ALL) [
4]. Our novel findings indicate that
RASSF6 and
RASSF10 are frequently and specifically methylated in ALL in contrast to
RASSF1A which is frequently methylated in epithelial cancers including lung, breast and kidney cancer but shows low frequency of methylation in childhood ALL. In addition we have demonstrated that methylation frequencies differ between B and T-ALL. Whilst
RASSF6 is methylated in the majority of (94%) B-ALL and less than half (41%) of T-ALL.
RASSF10 methylation frequency shows the opposite trend (16% in B-ALL vs 88% in T-ALL). Using a chromosome 3 Not1 array hybridization approach, we recently identified a number of genes frequently methylated in acute lymphoblastic leukemia [
5].
We have now used a recently developed high throughput approach, methylated CpG island recovery assay (MIRA) in combination with genome-wide CpG island arrays to identify epigenetic molecular markers in childhood ALL on a genome-wide scale. MIRA (methylated-CpG island recovery assay) is based on the high affinity of the methyl-CpG binding protein complex MBD2/MBD3L1 to methylated DNA [
6‐
8]. MIRA assay has several advantages for use as a tool for comprehensive analysis of DNA methylation patterns, for example it does not depend on having specific methylation-sensitive restriction sites in the target sequence, it does not depend on use of antibodies against 5-methylcytosine, and the ease of preparation of the recombinant GST tagged MBD2 and MBD3L1.
Discussion
We have combined the use of genome-wide CpG island arrays with a novel and sensitive method, the methylated-CpG island recovery assay (MIRA) to identify frequently methylated genes in acute lymphoblastic leukemia and other cancers. Global profiling of a small number (n = 5) of samples yielded a large number of methylation targets that were validated in a large cohort of clinical samples. Hence the MIRA-based CpG island microarray platform proved to be both efficient and effective. We identified and validated 30 genes that were methylated in 25% or more of ALL samples and 2 genes that were methylated at a frequency of 23%. Amongst these genes, 19 were newly identified methylation targets in cancer, whilst the remaining genes had been shown to undergo methylation in other cancers but this is the first report for methylation in ALL. The validated genes fell into several major functional categories, including transcription factors (
TFAP2A, TFAP2C, EBF2, TCF2, PAX6, PAX2, FOXF2), cell cycle control (
CDC14B, UBE2C), phosphatases (
EYA2, DUSP4), transforming growth factor-beta (TGFB) superfamily (
BMP2). G protein-coupled receptor (
GPR123), p53 gene target (
TP53I11), homeobox (
TSHZ3, NKX2-1, BARHL2, POU4F1), enzymes (
PTGS2), ion channels (
TRPC4), cadherin superfamily (
FAT1), Zinc finger (
PRDM12), nuclear receptor subfamily (
NR2E1,
NR4A2), ras association domain containing proteins (
MYO10,
ARHGAP20,
SSPN). Forty-four percent of the validated genes (and 45% of the short list) were targets of the polycomb complex in embryonic stem cells. This percentage is similar to what has been reported in other cancers including lung, breast, colorectal cancer and follicular lymphoma[
7,
12‐
14]. Our data provides further evidence that polycomb group proteins have an impact on the epigenetic programming of gene expression in a wide range of cancers.
Amongst the validated genes, three genes have been shown to undergo genetic inactivation events in sub types of leukemia and proposed to act as tumor suppressor genes.
GPR123 and
KNDC1 have recently been shown to be mutated in acute myeloid leukemia by using whole-genome sequencing methodology [
15], whilst
PRDM12 is located in a minimal commonly deleted region in chronic myeloid leukemia[
16].
A recent large-scale genome-wide study to identify genes methylated in adult ALL employing different high throughput approaches (MCA/RDA and MCA/array) validated 15 genes as showing frequent methylation in ALL (
GIPC2, RSPO1, MAGI1, CAST1, ADCY5, HSPA4L, OCLN, EFNA5, MSX2, GFPT2, GNA14, SALL1, MYO5B, ZNF382, MN1) [
17]. In our initial MIRA assay, we also identified 9 of the above 15 genes as frequent targets of methylation in childhood ALL, we did not analyze them any further since they had already been identified and validated in the Kuang
et al study [
17].
Chronic myeloid leukemia (CML) is a myeloproliferative neoplasm arising at the level of a pluripotent stem cell and consistently associated with the
BCR-ABL1 fusion gene. CML most commonly manifests in a chronic phase of the disease that progresses to advanced stage disease (blast crisis) that is resistance to therapy [
11]. Hence it is important to understand biological events involved in CML disease progression. We investigated the methylation status of the above genes (frequently methylated genes that were identified using the MIRA assay and validated in ALL samples) in a cohort of CML chronic phase and CML blast crisis samples. This led to the identification of two genes (
TFAP2A, EBF2) that showed a significant increase in methylation in CML-BC compared to CML-CP samples. We further determined that in paired samples from the same patient, BC samples showed methylation of multiple genes (including the above 2 genes) whilst the corresponding CP samples were mostly unmethylated. Hence we have identified genes that are likely to play a role in CML disease progression.
TFAP2A, a sequence specific DNA binding transcription factor has been demonstrated to be frequently methylated in large B-cell lymphoma, renal cell carcinoma and breast cancer [
18‐
20].
TFAP2A has been shown to act as a tumor suppressor gene and plays an important role in cancer cell chemosensitivity. In breast cancer cells it has been demonstrated that expression of
TFAP2A increased the chemosensitivity of cancer cells by sensitizing cells to undergo apoptosis upon chemotherapy[
20]. Methylated breast cancer cell lines treated with 5-aza-2'-deoxycytidine induced reexpression of
TFAP2A, resulted in apoptosis induction, increased chemosensitivity, decreased colony formation and loss of tumorigenesis upon chemotherapy. Amongst the 5 matched paired DNA samples (CML-CP and CML-BC from the same patient),
TFAP2A was unmethylated in all 5 CML-CP samples but was methylated in 3 out of the 5 corresponding CML-BC samples. Hence
TFAP2A may play an important role in CML-BC samples that become resistant to chemotherapy.
We also demonstrated that patients with methylation of
ATG16L2 (ATG16 autophagy related 16-like 2), an autophagy related gene, had a significantly decreased rate of major molecular response (MMR, defined as a BCR-ABL: ABL transcript ratio of 0.1% or less) at 12 and 18 months of imatinib treatment in comparison with patients with unmethylated
ATG16L2 gene (p = 0.013 and 0.017 at 12 and 18 months respectively). Other bona fide autophagy genes including
Beclin 1 have been shown to act as tumor suppressor genes in cancer.
Beclin 1 which is required for autophagy induction is monoallelically deleted in a high percentage of human breast, ovarian and prostate cancers, and its expression suppresses the tumorigenicity of human cancer cell lines [
21‐
24]. It likely acts as a haploinsufficient tumor suppressor gene. Beclin 1 heterozygous deficient mice have decreased autophagy and spontaneously develop tumors [
22]. Beclin 1 forms complexes with a range of proteins including UVRAG and Bif; these two proteins may also act as tumor suppressors [
25,
26]. UVRAG is frequently deleted (monoallelically) in colon cancers and overexpression leads to suppression of cell proliferation and tumorigenicity in human colon cancer cells [
25]. Whilst deletion of Bif in mice results in the development of spontaneous tumors [
26]. Recently ATG16L1 (ATG16 autophagy related 16-like 1) has been shown to be a bona fide autophagy protein [
27]. Another autophagy inducing gene
DAPK-1 is frequently silenced in human cancers by methylation and demonstrates tumor and metastasis suppressor properties [
28]. Future studies should be aimed at analyzing a larger series of CML samples for
ATG16L2 epigenetic inactivation and follow-up clinical parameters and at understanding the role it may play in CML development.
It would be very useful to identify epigenetic markers that could be utilized across several malignancies, to this end we determined the methylation status of the frequently methylated genes identified in ALL in six commonly occurring epithelial cancers (lung, breast, colorectal, kidney, brain and prostate). Ten genes (
BARHL2, EBF2, GPR123, NR2E1, PAX6, POU4F1, SALL3, TCF2, TFAP2A, TP53I11) demonstrated methylation frequencies of 50% or higher in 3 or more epithelial cancers. Although our data is from methylation analysis of epithelial tumor cell lines, one would expect that some if not many of the above genes showing 50% or higher methylation frequency in tumor cell lines would also show frequent methylation in primary tumors. Our preliminary data for transcription factor
EBF2 suggests that this indeed is the case (
EBF2 is methylated in >25% of lung and breast tumors, data not shown). In addition,
PAX6 methylation has previously been observed to occur frequently in colorectal and other cancers,
EYA2 is methylated in colorectal cancer,
NKX2-1 is methylated in thyroid cancer,
SALL3 is methylated in hepatocellular carcinoma,
TCF2 in ovarian cancer and
TFAP2A is methylated in large B-cell lymphoma, breast and kidney cancers [
18‐
20,
29‐
33]. We carried out a limited analysis of DNA from normal/control epithelial tissues (brain, breast and kidney), these were found to be unmethylated for the genes that were frequently methylated in the corresponding cancers.
Methods
Patient samples
Twelve leukemia cell lines (DND-41, CCRF-CEM (CEM), U937, Jurkat (JKT), TALL-1, NALM1, NALM6, NALM16, NALM17, THP-1, SUP-T1 and MOLT-4) and 64 primary childhood ALL comprising 52 B-cell ALL (B-ALL) and 12 T-cell ALL (T-ALL) were analyzed. Characteristics of the ALL samples have been described previously and see additional file
5 [
4,
5]. In addition
BCR-ABL positive CML samples consisting of 55 chronic phase and 8 blast crisis samples (including 5 matched CP and BC samples from the same patient) were also analyzed. DNA from a total of 10 normal peripheral blood lymphocytes and normal bone marrow (BM, AMS Biotechnology) sample were used as controls. All DNA samples from patients were obtained with informed consent and followed institutional guidelines.
Nine colorectal cancer cell lines (174T, DLD1, HCT116, HT29, LOVO, LS411, SW48, SW60, SW480), fourteen Lung cancer cell lines (A549, H1155, H1299, H1395, H1437, H157, H187, H1648, H1792, H187, H1993, H2171, H 2122, H460), nine breast cancer cell lines (HCC1143, HCC1395, HCC1419, HCC1437, HCC1806, HTB29, MCF7, T47D, MDA-MB- 231), seven Glioma cell lines (A172, H4, Hs683, T17, U87, U343, U373), twelve Renal cell carcinoma cell lines (768-O, CAKI, KTCL 26, KTCL 140, RCC4, SKRC18, SKRC39, SKRC45, SKRC47, SKRC54, UMRC2, UMRC3) and five Prostate cancer cell lines (22Rv1, DU-145, LNCaP, PC3 and VCap) were used in the Epithelial cancer cell line methylation panel. The collection of epithelial cancer cell lines have been described in our previous publications.
Imatinib response criteria for CML
Response to imatinib treatment was defined conventionally [
34]: complete cytogenetic response (CCR) = no Ph+ metaphases among at least 20 bone marrow metaphases or a BCR-ABL/ABL ratio of 1% or less; Major molecular response (MMR) = a BCR-ABL/ABL ratio of 0.1% or less.
Detection of DNA methylation changes by MIRA-assisted microarray platforms
Genomic DNA was fragmented by sonication and MIRA binding reaction was set up on 200 ng of sonicated DNA as described previously [
8]. The fraction representing the methylated DNA was collected from the binding reaction by Ni-NTA magnetic beads (Promega, Madison, WI) and washed 3 times with a 700-mM NaCl-containing buffer. Magnetic beads carrying the isolated fraction were picked up in 200 μl of TE buffer, mixed with one volume of phenol/chloroform and vortexed extensively. Magnetic beads were extracted by a magnet and the released methylated DNA fraction containing supernatant was ethanol precipitated after separation of the two phases in a microfuge. Isolated fraction was blunt-ended with T4 DNA polymerase (New England Biolabs), and a double-stranded adaptor was ligated onto the ends. Amplicons were created by LM-PCR. Labeling and array hybridization on human CpG island microarray platform (Agilent Technologies) were performed as described [
8]. MIRA-enriched control and MIRA-enriched ALL amplicons were hybridized onto the CpG microarray for detection of ALL-specific methylation changes. Microarray slides were scanned using an Axon GenePix 4000b scanner and images wee quantified by GenePix Pro 6 software.
Gene selection
The short list of 398 genes was generated as described in results section, also see figure
1.
Methylation analysis
Bisulfite modification of DNA was performed as described previously [
4]. The methylation status of all the CpG islands were determined by combined bisulfite restriction analysis (COBRA), semi-nested primers were designed to amplify regions of the CpG islands close to or overlapping the transcription start sites from bisulfite modified DNA (see additional file
6 for primer sequences). For enzymatic methylation detection ten microlitres of COBRA PCR product was incubated with 2U
BstUI restriction enzyme (CGCG) overnight at 37°C before visualisation on a 2% agarose gel. The methylation status of
ARHGAP20, CDC14B, CYP1B1, EYA2, FAT1, GPR123, KNDC1, MYO10, PTGS2, SALL3, TFAP2A, TFAP2C and
TRPC4 was also determined by bisulfite sequencing. Samples selected for bisulfite sequencing were cloned in the pGEM-Teasy vector according to manufacturer's instructions. Up to 10 individual colonies were chosen for colony PCR using the primers specific for the pGEM-Teasy vector back bone, forward 5'-TAATACGACTCACTATAGGG-3' and reverse 5'-ACACTATAGAATACTCAAGC-3'. Amplified PCR products were then sequenced to ascertain the methylation status of individual alleles and to determine the methylation index (MI). The MI was calculated as a percentage using the equation; number of CpG dinucleotides methylated/total number of CpG dinucleotides sequenced × 100.
Cell lines, 5azaDC treatment and RT-PCR
Leukemia cell lines were maintained in RPMI1640 (Sigma) supplemented with 10% FCS, 2 mM Glutamine, 20 mM HEPES, 1 mM Sodium Pyruvate and 12.6 mM Glucose Monohydrate at 37°C, 5% CO
2. Cells were treated with 5 μM of the DNA demethylating agent 5azaDC (Sigma) freshly prepared in ddH
2O and filter-sterilised. The medium (including 5 μM 5azaDC) was changed every day for 5 days. Cells were also treated on day 4 with 0.1 μM TSA for 24hrs. RNA was prepared using RNA bee (AMS biotechnology) according to manufacturers' instructions. cDNA was generated from 1 μg total RNA using SuperScript III (Invitrogen) and polyN primers. See additional file
7 for primer sequences. In all cases a
GAPDH control was included using conditions described previously [
4]. Gene expression was detected by amplification from 50 ng of cDNA using 0.8 μM of each primer, 2 mM MgCl
2, 0.25 mM dNTPs and 1U Fast start
Taq (Roche).
Statistical analysis
Statistical analysis was performed using Fisher's exact test or t test where appropriate. All reported P values were two-sided and P < 0.05 was taken as statistically significant.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
FL designed research, did statistical analysis, established all collaborations and wrote the paper. TD did majority of the bioinformatics analysis, gene methylation, sequencing and expression analysis and took part in writing the paper. TR and GPP did the MIRA and array hybridization. LH took part in gene methylation, sequencing/expression analysis. DG did tissue culture. LW and RC provided the CML samples together with relevant clinical information and took part in statistical analysis. DC provided the ALL samples. ERM was involved in statistical analysis and in overall study design. AD took part in bioinformatics analysis. All authors reviewed the paper.