Homozygous Deletion of MTAP Gene as a Poor Prognosticator in Gastrointestinal Stromal Tumors

Gastrointestinal stromal tumors (GISTs) are characterized by constitutive activation of receptor tyrosine kinase (RTK) resulting from gain-of-function mutations of KIT and PDGFRA genes (1). Although NIH consensus scheme proved prognostically useful (2,5), the associations of tumor location and RTK genotypes with outcomes underscore the need to assess new adjuncts to better risk-stratify GISTs (4, 613). Furthermore, identification of novel genes altered in tumor cells is critical for better understanding tumorigenesis, developing diagnostic tests, and designing effective therapies.

The prospect of genome-wide approaches is now encouraging efforts to characterize cancer genomes. Decreased gene dosage by hemizygous and/or homozygous deletion can lead to inactivation of tumor suppressor genes (TSGs), which may open a new avenue for derived targeted therapies (14). Nonrandom losses of chromosome 9p and/or 9q have been recurrently observed in advanced GISTs (15,20). To pinpoint potential TSGs associated with tumor progression in chromosome 9, we used oligonucleotide-based array comparative genomic hybridization (aCGH) with ultrahigh-resolution for profiling copy number alterations in 22 GISTs. There is mounting evidence that DNA losses of chromosome 9 in cancers, either spanning the entire chromosome or restricted areas of p arm, oftentimes encompass 9p21 that harbors several candidate or established TSGs, such as CDKN2A (also known as p16^INK4A/p14^ARF), CDKN2B (also known as p15^INK4B), and MTAP, etc. (20 26).

MTAP (methylthioadenosine phosphorylase) encodes a key enzyme in the catabolism of methylthioadenosine, which is a by-product of polyamine biosynthesis in the methionine salvage pathway (22, 23, 25, 27, 28). Recently, MTAP has been proposed as a functional TSG, prompting us to select this gene for further validation of aCGH findings (26 29). Using large independent cohorts of GISTs, we assessed MTAP protein expression by tissue microarray (TMA)based immunohistochemistry and precisely quantified MTAP gene dosage by coupling laser capture microdissection (LCM) with real-time quantitative PCR. The MTAP gene status, dichotomized as homozygously deleted or not, was successfully determined in 146 primary localized GISTs. In addition, promoter hypermethylation of MTAP gene was identified in three of seven cases deficient in protein expression but devoid of homozygous deletion. As a major mechanism to deplete protein expression, MTAP homozygous deletion was associated with important prognosticators and independently predictive of worse outcomes.

The institutional review board had approved this study (97-2211A3). We first used aCGH to profile somatic copy number alterations of 22 fresh GIST specimens (screening set), including 3 low-risk, 9 intermediate-risk, and 10 high-risk tumors. To independently validate aCGH results in formalin-fixed tissues, MATP gene dosage of LCM-isolated tumor cells and MTAP immunoexpression were successfully determined for 146 and 306 primary GISTs by using real-time quantitative PCR and TMA technologies, respectively. The mutation variants of RTK genes were successfully genotyped for 181 cases. All patients enrolled for quantitative PCR assay, immunostain, and RTK genotyping were independent of the screening set and did not receive imatinib treatment before disease relapses. Details on clinicopathologic characteristics of cohorts for aCGH analysis and for validation of MATP gene status and immunoexpression were summarized in Supplementary Tables S1 and S2, respectively.

For each fresh sample, 1 g of genomic DNA was extracted for hybridization against oligonucleotide microarrays. Each chip has 385K probes with a median spacing of 6 kb and variable lengths to achieve a melting temperature of 76C (NimbleGen Systems). The reference DNAs were obtained from normal lymphocytes of gender-matched donors. The procedures of DNA labeling, hybridization, normalization of oligonucleotide arrays, window averaging of contained probes, and data acquisition were done by the facility of manufacture, as reported (30). The Cy3 and Cy5 signal intensities were normalized to one another using Qspline normalization. The circular binary segmentation algorithm proposed by Olshen et al. was used for segmentation of the averaged log₂ ratio data (31). Each segment was then assigned a log₂ ratio that was the median of the contained probes, and the data were centered by the tallest mode in the distribution of the segmented values. To finely delineate the breakpoints of whole array probes, we defined gains and losses as log₂ ratios of +0.20 or 0.20, respectively. To unravel candidate TSGs showing copy number alterationdriven deregulation associated with disease progression, we searched for hemizygous (log₂ ratio between 0.20 and 0.50) and/or homozygous (log₂ ratio<0.50) deletions at identical DNA segments among cases, which were recurrently present in at least 40 of high-risk samples but not in >25 of nonhigh-risk samples. Java TreeView software¹⁰

The TMA sections were prepared as previously described for antigen retrieval (3), followed by overnight incubation with the primary antibody of MTAP (MGC:31876, 1:100, Proteintech) and detection with ChemMate EnVision kit (K5001, DAKO). The percentage of tumor cells with cytoplasmic immunoreactivity was scored for multiple cores from the same patient and averaged to obtain a mean labeling index (LI). MTAP LI of <10 was used to define aberrant loss of protein expression in keeping with various cutoff values adopted previously (23, 27, 32). Most TMA blocks had also been previously used for Ki-67 staining in our earlier report (3), wherein we used a mean LI of >5 to define Ki-67 overexpression.

Approximately 3,000 tumor cells in each sample were microdissected by Veritas LCM machine (Arcturus Engineering) to extract genomic DNAs as previously described (33). Given that aCGH showed very few copy number alterations spanning PFKL gene at 21q22.3, this gene was chosen as the reference to measure MTAP gene dosage. Because the comparative computed tomography method may result in skewed values when analyzing low-copy templates (e.g., LCM-isolated cells from paraffin-embedded tissue; refs. 34, 35), we first constructed standard curves for the reference PFKL and target MTAP genes for real-time PCR assays as detailed in Supplementary Method S1 and Supplementary Fig. S1. The sequences of primer pairs and probes targeting PFKL and exon 8 of MTAP gene were described in Supplementary Method S1. We adopted a cutoff of average MTAP/PFKL ratio at <0.2 to define homozygous deletion in two independent assays with duplicate samples for each case, based upon the assumption that the normal tissue contamination and intratumoral heterogeneity should collectively account for <20 of experimental deviation.

We had previously described the methods of DNA extraction, PCR amplification, direct sequencing of KIT exon 11, and screening by denatured high-performance liquid chromatography for exons 9, 13, and 17 of KIT gene and exons 12 and 18 of PDGFRA gene with confirmatory sequencing (11).

Following manufacturer's instructions, we used Puregene kit (Gentra Systems) to extract genomic DNA from formalin-fixed tissue, 1 g of which was then modified with sodium bisulfite for each sample using Herman's method (36). To maximize success in the amplification of bisulfite-treated DNA, consensus outer primers and inner primers specific for methylated and unmethylated promoters of MTAP gene (Supplementary Method S2) were designed using MethPrimer software. DNA from healthy donors' peripheral lymphocytes was treated with or without M.SssI methyltransferase (New England Biolabs) to serve as methylated and unmethylated controls, respectively.

Statistical analyses were done using SPSS 14 software package. Association and comparisons with various parameters were evaluated by ², Wilcoxon rank sum, and Student's t tests as appropriate for MTAP gene status. Follow-up data were available in 306 and 146 cases with informative MTAP immunohistochemical scores (median, 49.9 mo) and MTAP gene dosage ratio (median, 56.3 mo), respectively. The end point was disease-free survival (DFS), which would not be confounded by imatinib therapy for patients with disseminated disease. RTK genotypes were dichotomized into two prognostically different groups, as detailed in our previous report (11). Kaplan-Meier curves were plotted to compare prognostic differences by log-rank tests. The 146 cases with informative MTAP gene status formed the basis for multivariate survival analysis. Significant prognosticators at univariate level were analyzed by Cox regression model except for the strong dependent covariate of MTAP gene status (i.e., MTAP immunoexpression) and component factors of NIH risk scheme (i.e., tumor size, mitotic activity; ref. 2). For all analyses, two-sided tests of significance were used with P < 0.05 considered significant.

As shown in Fig. 1A, complete and/or partial loss in both 9p and 9q was recurrently detected in 7 of 10 high-risk GISTs, including two cases showing a number of interstitial deletions. However, another high-risk case revealed complete loss of 9p alone. Conversely, among 12 nonhigh-risk GISTs tested, there was only one intermediate-risk case showing almost complete loss of chromosome 9, whereas no large-scale alterations were detected in the remaining cases.

The recurrent regions of DNA losses on 9p were mapped to five major deletion cores and collectively spanned 11 hemizygously and/or homozygously deleted candidate TSGs (Fig. 1A). These included DOCK8, ANKRD15 (also known as KANK1), and SMARCA2 at 9p24.3; PTPRD at 9p24.1; SH3GL2 at 9p22.2; MTAP, CDKN2A, CDKN2B, TUSC1, TOPORS at 9p21; and RECK at 9p13.3. As for the nonrandom losses on 9q, there were six potential TSGs with hemizygous and/or homozygous deletions within three major deletion cores where TLE1 (9q21.32) and DAPK1 (9q21.33) at 9q21.3; DEC1 (9q33.1), DBC1 (9q33.1), and DAB2IP (9q33.2) at 9q33; and MRPL41 at 9q34.3 were identified. Noticeably, one intermediate-risk (11) and seven high-risk (70) GISTs clearly showed deletions at the chromosome band 9p21.3. The latter encompassed the most frequent loci of homozygous losses (Fig. 1B), affecting MTAP in three cases (two high-risk cases, one intermediate-risk case) and CDKN2A/CDKN2B in four (three high-risk cases, one intermediate-risk case). Moreover, two cases were homozygously codeleted at both loci.

The cohort for MATP immunohistochemical analysis (Fig. 2; Supplementary Table S2) consisted of 306 GISTs from 153 males and 153 females, including 115, 96, and 95 cases classified as very low or low risk (Fig. 2A), intermediate risk (Fig. 2B), and high risk (Fig. 2C) based on the NIH consensus scheme, respectively. MTAP immunoexpression displayed a wide variation in LI from 0 to 100 and was aberrantly lost (LI < 10) in 58 cases (19; Fig. 2D-F). Among cases with immunohistochemical results, normalized MTAP gene dosage was successfully determined in 146 GISTs, consisting of 44 very low-risk or low-risk, 49 intermediate-risk, and 53 high-risk cases. Twenty-five GISTs (17) showed homozygous deletion with MTAP/PFKL ratio of <0.2 (Table 1). No essential difference in the clinicopathologic characteristics and RTK genotypes was found between the cohorts examined for immunoexpression and gene status (Supplementary Table S2).

In this study, analysis of RTK genes was successfully done in 181 GISTs with mutations detected in 158 cases (87). The trends of survival curves for individual RTK genotypes were generally in keeping with our, as well as others', earlier reports (4, 6, 8, 9, 11). According to our previous grouping rationale (11), these 181 cases were further dichotomized as favorable (n = 88) versus unfavorable (n = 93) genotypes as detailed in Supplementary Fig. S2. The prognostically favorable genotypes comprised (a) PDFGRA mutation involving exon 12 or 18 in nine cases, (b) 3 tandem insertion of KIT exon 11 with or without point mutation in 11 cases, and (c) single-point mutation of KIT exon 11 in 68 cases. The group of unfavorable genotypes included (a) Ala502-Tyr503 insertion of KIT exon 9 in three cases, (b) wild type for both KIT and PDGFRA genes in 23 cases, and (c) 5 deletion of KIT exon 11 with or without point mutation in 67 cases. Of the KIT exon 11deleted subgroup, there were 42 and 26 cases with deletions involving codons 557 to 558 and other codons, respectively. No mutation of KIT exon 13 or 17 was detected in this series.

As shown in Table 1, a strong correlation was substantiated between MTAP homozygous deletion and loss of protein expression (P < 0.001) in GISTs. In cases showing homozygous deletion, the vast majority (92, 23 of 25) showed MTAP protein deficiency, whereas this aberrant loss was only observed in 6 (7 of 121) of cases without homozygous deletion. MTAP homozygous deletion, more frequently involving GISTs that occurred in the nongastric location (P = 0.041), showed presence of epithelioid histology (P = 0.003) and displayed higher mitotic rate (P = 0.008). More interestingly, it was also highly related to larger tumor size (P < 0.001), increasing NIH risk levels (P < 0.001), and higher proliferative index (P = 0.007; Fig. 2G-J). The findings of strong associations between homozygous deletion of MTAP gene and several adverse clinicopathologic prognosticators suggested its crucial role in tumor progression of GISTs. However, the correlation of this genetic aberration with RTK genotypes could not be confirmed.

Correlations of clinical outcomes with various clinicopathologic, immunohistochemical, and molecular parameters at the univariate level are shown in Table 2 and Fig. 3. Inferior DFS was significantly associated with nongastric location (P = 0.0132). In addition, presence of epithelioid histology (P < 0.0001), larger tumor size (P < 0.0001), higher mitotic count (P < 0.0001), increasing NIH risk levels (P < 0.0001), high proliferative index (P < 0.0001), and unfavorable RTK genotypes (P < 0.0001; Fig. 3A) were all highly predictive of adverse outcomes. Of these, GISTs with favorable RTK genotypes had a 10-year DFS of 78.5, whereas this rate declined to 31.1 in those with unfavorable RTK genotypes. More importantly, both MTAP protein deficiency (P = 0.0018, DFS rate 28.1 versus 59.6 at 10 years; Fig. 3B) and homozygous deletion of MTAP gene (P < 0.0001, DFS rate 20.9 versus 62.9 at 10 years; Fig. 3C) were strongly adverse prognosticators of inferior DFS.

In multivariate analysis (Table 2), homozygous deletion of MTAP gene remained prognostically independent (P = 0.0369), along with high NIH risk level (P = 0.0001), high Ki-67 index (P = 0.0106), and nongastric location (P = 0.0416). Furthermore, MTAP homozygous deletion also identified patients at >2-fold higher risk of relapsed disease, whereas histologic type and RTK genotypes lost statistical significance.

Approximately one fourth (7 of 30; Table 1) of MTAP proteindeficient GISTs were not homozygously deleted in MTAP gene, suggesting the operation of alternative molecular mechanism(s) and potential prognostic heterogeneity within this subgroup. Therefore, methylation-specific PCR was adopted to substantiate whether promoter hypermethylation of MTAP gene was present in seven cases that were devoid of protein expression but not homozygously deleted in genomic DNA. We found that promoter hypermethylation was present in three nonhigh-risk (two low-risk case, one intermediate-risk case) cases but undetected in another four high-risk counterparts (Fig. 4). More intriguingly, GISTs with homozygously deleted MTAP behaved more aggressively with significantly shorter DFS than those MTAP proteindeficient cases without homozygous deletion (P = 0.0493; Fig. 3D). The latter was actually not different from MTAP-expressing GISTs in the DFS rate (P = 0.9695).

Whereas oncogenic mutations of RTK genes are primary events in GIST tumorigenesis, chromosomal imbalances are considered to play pivotal roles in promoting clinical aggressiveness (18, 37). Previous studies using conventional CGH and/or array-based genomic profiling indicated that loss of chromosome 9, especially 9p, is a specific indicator of unfavorable outcomes in GISTs (16,18). We corroborated this association by using ultrahigh-resolution aCGH, showing a striking predilection of chromosome 9 deletions, mostly involving both arms, in high-risk GISTs (70). Compared with previous series (15,19), the higher prevalence of 9p losses in high-risk GISTs may be attributable to better resolution of the current aCGH platform in detecting subtle interstitial deletions. Specifically, GISTs with chromosome 9 losses almost all involved the probe sets within 9p21.3 where p16^INK4A, the leading TSG of CDKN2A/2B loci, was recurrently inactivated in aggressive GISTs by various mechanisms, including the most common homozygous deletion (20, 38). Moreover, the major deletion cores on both arms, also preferentially affecting high-risk cases, were found to span the breakpoints of several other candidate TSGs. Many of these were recently reported to implicate other cancers, such as PTPRD located at 9p24.1 in neuroblastomas and cutaneous squamous cell carcinomas (30, 39); MTAP, TUSC1, and TOPORS at 9p21 in lung or colorectal cancers (26, 40, 41); DAPK1 at 9q21.3 in renal cell carcinomas (42); and DEC1, DBC1, and DAB2IP at 9q33 in esophageal (43), lung (44), and breast carcinomas (45), respectively.

Prior studies on tumor specimens and/or cell models have shown that MTAP depletion, rendering cells dependent on de novo synthesis of purine derivatives, is not uncommon in human malignancies (21, 22, 2529), such as melanoma (27, 29), nonsmall cell lung cancers (26), and T-cell acute lymphoblastic leukemia (21). Nevertheless, there exist conflicting thoughts on whether loss of MTAP activity is indeed pathogenically relevant, given frequent codeletion of MTAP with proximate CDKN2A and/or CDNKN2B. However, evidence has emerged that MTAP may represent as a genuine TSG with unique tumor-suppressive effects and functional basis. First, MTAP can be lost independently of CDKN2A in nonsmall cell lung cancers, with an even higher rate of homozygous deletion in the former (38 versus 18; ref. 26). Second, reexpression of MTAP in MTAP-deleted breast cancer cells (MCF-7) enables dramatic inhibition of anchorage-independent growth in vitro and tumorigenicity in vivo because of decreased ornithine decarboxylase and polyamines, such as putrescine (28). Third, through accumulation of 5-deoxy-5-(methylthio)adenosine (the MTAP substrate), MTAP depletion may enhance the invasive and vasculogenic capability of melanoma cells by inducing expression of matrix metalloproteinases and angiogenic growth factors, respectively (29).

Integrating various methodologies, we have first unraveled the gene status and protein expression of MTAP at 9p21.3 in GISTs and provided compelling evidence of MTAP homozygous deletion as a critical event in disease progression. Unfavorable RTK genotypes, especially KIT exon 11 deletions, were recently found to predict aggressiveness of GISTs (4, 6, 8, 9, 11). Nevertheless, these genotypic variations did not correlate with MTAP homozygous deletion, further supporting that the latter was secondarily acquired during cytogenetic evolution rather than being driven by activated RTKs.

In GISTs, MTAP homozygous deletion showed strong concordance with loss of immunoexpression, confirming this aberration (77) as the main inactivating mechanism leading to MTAP deficiency. As a candidate TSG, MTAP was epigenetically silenced via promoter hypermethylation in three cases, whereas the underlying cause of MTAP deficiency was not identified in another four MTAP-immunonegative GISTs. However, alternative genetic defects contributing to MTAP downregulation could not be completely excluded, such as point mutations or small deletions involving more 5 exons and/or promoter of MTAP. Recently, promoter hypermethylation was found preponderant in MTAP-deficient hepatocarcinomas, implying histotype-dependent selectivity in its inactivating mechanisms (46, 47).

MTAP homozygous deletion in GISTs not only correlated with larger size and higher mitotic rate, Ki-67 index, and NIH risk level but also independently predicted worse DFS with >2-fold increased risk. These findings corresponded with the similar prognostic effect observed in mantle cell lymphomas (48), indicating that this genomic loss may provide a survival or proliferation advantage. Although only a small proportion (17) of cases were homozygously deleted for MTAP, approximately a third of high-risk GISTs carried this aberration and would be eligible for alternative MTAP-directed therapy (e.g., l-alanosine) to completely switch off AMP supply to tumor cells (21, 25, 48, 49) once resistance to imatinib occurs. However, in vitro studies could not validate the proliferation-promoting effect of MTAP deficiency in melanoma- and hepatoma-derived cell lines with promoter hypermethylation (27, 47). More intriguingly, MTAP promoter hypermethylation, albeit a small proportion, was all seen in nonhigh-risk cases, consistent with occurrence of this epigenetic alteration even as early as in precancerous liver cirrhotic tissues (46). Furthermore, nonhomozygously deleted, MTAP-deficient GISTs behaved as indolently as MTAP-expressing GISTs, unlike those aggressive, homozygously deleted tumors. Accordingly, it is possible that the timing of MTAP gene abrogation may be different among various inactivating mechanisms in the evolution of GISTs, thereby culminating in different prognostic effect.

In summary, MTAP homozygous deletion, present in 17 of GISTs, represents the predominant mechanism to deplete protein expression. Notably, it not only correlates with important adverse parameters but also provides independent prognostic effect, further highlighting its role in disease progression. Our findings may warrant prospective validation in future studies to justify the potentiality of MTAP-directed agents as an alternative therapy in high-risk, imatinib-resistant GISTs devoid of MTAP expression.

No potential conflicts of interest were disclosed.

We thank GeneChannel, Inc., Taipei, Taiwan, for critical technical assistance.