Tumor Genotype Is an Independent Prognostic Factor in Primary Gastrointestinal Stromal Tumors of Gastric Origin: A European Multicenter Analysis Based on ConticaGIST

Gastrointestinal stromal tumors (GIST) are rare mesenchymal neoplasms of the gastrointestinal tract, characterized by the expression of KIT protein, detectable in about 95% of tumors. Constitutively activating mutations in the genes coding for receptor tyrosine kinase KIT (70%–75% of cases) or platelet-derived growth factor receptor alpha (PDGFRA; 10%–14% of cases) play a crucial role in the biology of these tumors, and are critical therapeutic targets (for review see ref. 1). The mutations of the KIT gene in GIST occur most frequently in KIT exon 11 (juxtamembrane domain), followed by KIT exon 9 (extracellular domain); less often, primary mutations in the ATP-binding pocket (exon 13) or activation loop (exon 17) are found. Primary PDGFRA mutations are usually present in the activation loop (exon 18; 90% of cases); among them, PDGFRA p.D842V mutation is the most common. Approximately 10% to 15% of adult GIST and 85% of pediatric GIST, defined as KIT/PDGFRA “wild type” (WT), lack KIT or PDGFRA mutations. These tumors are highly heterogeneous and profoundly different from KIT/PDGFRA-mutated GIST in terms of clinical behavior and molecular profiles, and they are now considered as separate pathologic entities (2).

The most common organ site for primary GIST is stomach (50%–60%), followed by small intestine (20%–30%), colon and rectum (10%), and esophagus (5%). Rarely, the tumors appear without attachment to gastrointestinal tract (so-called extragastrointestinal GIST). Complete surgical resection is possible in most localized GIST and constitutes the current standard of care. Although a significant proportion of patients will be cured with surgery alone, approximately 40% to 50% will eventually relapse, usually within the first 5 years (3, 4). The malignant potential in GIST varies from negligible in minute gastric GIST tumorlets to aggressive cancer (5, 6). The treatment of recurrent GIST is the first-line tyrosine kinase inhibitor, imatinib mesylate, if surgery is not an option. Over 80% of patients with advanced GIST benefit from imatinib treatment. Assessment of the risk of recurrence is important in the management of operable GIST, and needs to be estimated also when considering the use of adjuvant imatinib therapy, which is a widely accepted treatment standard in patients at higher risk of relapse (7). Among risk-stratification tools, the most widely used are the Armed Forces Institute of Pathology (AFIP) classification (8) and the modified NIH consensus (9). Primary tumor size, mitotic count, and tumor site are well-established risk factors for disease-free survival (DFS); tumor rupture has later been associated with the higher risk of relapse (9, 10). Moreover, in the last years the potential role of mutational status as a prognostic factor has been examined in a number of retrospective studies. In particular, it has been shown that deletions of KIT exon 11, especially those involving codon 557 and/or codon 558 (designated as KITdel-inc557/558), are associated with malignant behavior (11–14). Conversely, KIT/PDGFRA WT GIST and most PDGFRA-mutated GIST generally have a lower potential for malignancy (2, 15). However, the available data are still insufficient to be incorporated into routine clinical risk assessment.

In the present study, we assess the role of the tumor KIT and PDGFRA genotype along with key prognostic factors for DFS in localized, operable GIST in the so far largest dataset from 1,056 patients registered in the European ConticaGIST database.

We assessed the information from patients that were included in the ConticaGIST registry (https://conticagist.sarcomabcb.org) and diagnosed between January 1985 and April 2012 in 13 contributing institutions from four European countries. The majority of cases, i.e., all those diagnosed from 2001, were studied prospectively (83% of all). Each patient was assigned a registry code prior data transfer to ensure data anonymization. The study was approved by Ethics Boards of participating institutions. A small subset (n = 127) of the present cohort has been previously published (16). Only patients with primary GIST diagnosis who underwent surgery with curative intent (R0 or R1) were eligible. Pre- or postoperatively, the patients were not treated with any chemotherapeutic agents, including imatinib, until disease recurrence. From 1,625 patients with data transferred from the pooled series, 559 did not fulfill these eligibility criteria and were excluded from the study. Because of the small number of patients, we have also excluded 10 pediatric GISTs (age of diagnosis <18). Finally, a total of 1,056 patients were included for further analysis. DFS, defined as the duration from surgery to relapse (local recurrence or metastasis) or last follow-up, was used for prognosis evaluation. Relapse was determined based on routine imaging assessment and/or biopsy. The follow-up was updated on January 15, 2014. If patients were relapse free at the time of the last follow-up, they were censored for a disease recurrence.

All samples were reviewed in one of the four reference centers [Institut Bergonié, Bordeaux, France (n = 448); M. Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Warsaw, Poland (n = 279); University Hospitals Leuven, Belgium (n = 221); and GIST and Sarcoma Registry North Rhine-Westphalia Germany (n = 108)] by expert sarcoma pathologists. The diagnosis of GIST was based on histology, immunohistochemistry, and molecular pathology. Eligible patients were required to have tumor morphology compatible with GIST and positive immunostaining for KIT protein (CD117). Approximately 3% of tumors did not reveal KIT immunopositivity, and in such cases, the diagnosis was considered to be confirmed if a mutation was found in either KIT or PDGFRA genes. Mitotic index (MI) was evaluated per 50 high power fields (HPF), by examination of hematoxilin and eosin staining (H&E)-stained specimens. Risk stratification was defined according to the modified NIH classification (9).

Mutational analysis of KIT and PDGFRA was performed based on DNA isolated from formalin-fixed, paraffin-embedded (FFPE) or frozen material. DNA was extracted after histological review, from tumor areas containing at least 80% of tumor cells. Mutation of KIT exons 9, 11, 13, and 17 and PDGFRA exons 12, 14, and 18 was identified by Sanger sequencing of PCR products, according to the routine protocols available in participating institutions. The institutions, which performed over 95% of mutational analysis, were included in external, international quality control programs (17, 18). Mutation nomenclature followed the recommendations of the Human Genome Variation Society (www.hgvs.org).

For univariate analysis, the χ² or Fisher exact tests and Mann–Whitney test were used to compare categorical and continuous variables, respectively. Survival analysis was performed with the Kaplan–Meier method, and survival between groups was compared using the log-rank test. Univariate and multivariate Cox proportional hazard models were used to determine associations between variables of interest (tumor size, MI, tumor rupture, and tumor genotype) and DFS. The strengths of associations were summarized with hazard ratio (HR) and corresponding 95% confidence interval (CI).

The value of P < 0.05 was interpreted as statistically significant. Analyses were carried out with STATISTICA v.12 software package (StatSoft Inc).

Information about primary tumor size, MI, anatomical site, and mutational status was available from 996, 974, 1,050, and 1,056 patients, respectively (Table 1). The DFS data were available from 998 patients.

The majority of tumors were localized in the stomach (55.6%), followed by small intestine (28.1%), duodenum (5.8%), and colon and rectum (5.5%). The median tumor size was 6.0 cm (mean, 7.4 ± 5.3 cm; range, 0.2–40 cm), and 20.6% of tumors were >10 cm at initial diagnosis. The median number of mitoses was 4 per 50 HPF (mean, 13 ± 27/50 HPF; range, 0–300/50 HPF), with a mitotic rate >5/50 HPF documented in 38.8% of cases. Perioperative tumor rupture was recorded in 6.3% of patients. Using the modified NIH classification, the risk of relapse could be assessed in 957 cases: 7.7%, 36.4%, 17.5%, and 38.4% were very low, low-, intermediate-, and high-risk tumors, respectively. Patient's demographic and clinicopathologic characteristics are provided in Table 1.

Overall, mutations were identified in 899 (85.1%) cases (Table 1). The KIT or PDGFRA mutations were found in 751 (71.1%) and 148 (14.0%) samples, respectively. Mutations localized within exons 9 and 11 of KIT and exon 18 of PDGFRA accounted for 861 (96.8%) of all the mutations. All but one KIT exon 9 mutations were duplications, whereas KIT exon 11 mutations were mainly deletions (34%), followed by substitutions (19.7%) and duplications (6.6%). The KITdel-inc557/558 or specific p.W557_K558del were detected in 226 (21.4%) and 66 (6.3%) of all cases, respectively. Among other most frequent KIT mutations, six were substitutions (p.V559D, p.V560D, p.W557R, p.L576P, and p.V559D, and p.K642E) and one a deletion (p.V560del). Among PDGFRA mutations, the most prevalent was the p.D842V substitution (9.8% of all mutations and 65.2% of the PDGFRA exon 18 mutants). The 10 most frequent mutations were correlated with clinicopathologic data (Table 2).

As expected, the PDGFRA p.D842V mutation was associated mainly with primary gastric localization (93.2%), whereas KIT exon 9 duplication was found prevalently in nongastric tumors (90.9%). When compared with KIT p.W557_K558del, KIT p.A502_Y503dup had a lower median mitotic rate (6 vs. 16/50 HPF; P = 0.005), whereas there was no difference in tumor size for these mutants (7 vs. 8 cm; P = 0.257). Notably, KIT p.W557_K558del mutants equally segregated in gastric and nongastric sites (55% vs. 45%). Moreover, KIT p.W557_K558del was more frequently identified in patients ≤ 60 years (59% vs. 42.4%; P = 0.01), in tumors >5 cm (84.5% vs. 57.7%; P = 0.0001), with MI >5/50 HPF (68.9% vs. 39.4%; P < 0.0001), and classified as high risk (70.2% vs. 38.9%; P < 0.0001), when compared with other KIT exon 11 mutants. The median tumor size was 8.0 cm (mean, 9.1 cm; range, 1–30 cm), and median mitotic rate was 16/50 HPF (mean, 30.6/50 HPF; range, 0–300), and both values were significantly higher than in other KIT exon 11 mutants (P < 0.0001 for both). Notably, the clinicopathologic characteristics of tumors bearing KIT p.W557_K558del were comparable with the group of tumors with KITdel-inc557/558, within which tumor size, mitotic rate, and fraction of high-risk tumors were also significantly higher than in tumors with other KIT exon 11 mutants (P < 0.001; Table 3).

During a median follow-up period of 52 months (range, 1–304 months), progressive disease was observed in 369 patients (34.9%). The median DFS in this group was 17.6 (range, 1–252) months.

For further analysis, only the most frequent tumor genotypes (KIT exon 9, KITdel-inc557/558, other KIT exon 11 and PDGFRA exon 18 mutants) were taken into account. By the survival analysis, there was a significant difference when patients were grouped according to the tumor size, MI, location of the primary GIST, or mutational status (all P < 0.001; Fig. 1A–D). Overall, both KIT exon 9 and KITdel-inc557/558 mutants were associated with relatively equal and inferior DFS (median DFS in both groups, 45.5 months; 5-year DFS, 37.9% and 33.1%, respectively), whereas PDGFRA exon 18 mutation correlated with very favorable disease outcome (median DFS not reached; 5-year DFS, 75%) in comparison with other mutants. There was no significant difference in DFS of PDGFRA p.D842V versus other PDGFRA exon 18 mutants (P = 0.21; data not shown). Of note, however, within the dominant PDGFRA mutants of gastric origin, the vast majority that progressed (11 of 14) carried exon 18 PDGFRA-D842V substitution.

In univariate analysis, tumor size, MI, primary location, rupture, and mutational status were strongly associated with DFS (Table 4). In the multivariate analysis, all prognostic factors but tumor rupture remained significant for DFS (Table 4).

With regard to tumor genotype, KIT exon 9 and KITdel-inc557/558 were associated with an increased risk for tumor progression [HR, 1.47; 95% CI, 0.9–2.5 and 1.45 (95% CI, 1.0–2.2), respectively], whereas PDGFRA exon 18 mutation (HR, 0.23; 95% CI, 0.1–0.6) was associated with a better DFS.

Given the relatively high frequency of KITdel-inc557/558 mutants (overall 21.7%) with the equal distribution in gastric and nongastric sites, we have further explored the possible impact of this genotype on DFS, depending on the anatomical site of tumor origin. Importantly, in clear contrast with other KIT exon 11, KIT exon 9, and PDGFRA exon 18 mutations, the poor prognosticator of KITdel-inc557/558 on patients' survival was only significant in GIST localized in the stomach (P < 0.001; Fig. 1E), but not in tumors with nongastric origin (P = 0.26; Fig. 1F). The same associations were also evident when comparing KITdel-inc557/558 mutants with other exon 11 KIT mutations, overall (P < 0.0001; Supplementary Fig. S1A) or comparing gastric (P < 0.0001; Supplementary Fig. S1B) and nongastric (P = 0.599; Supplementary Fig. S1C) GIST.

This phenomenon might be related to the observation that gastric GIST with KITdel-inc557/558 had a larger size (7 vs. 5.8 cm; P < 0.0001) and a higher mitotic rate (6 vs. 4/50 HPF; P < 0.0001) when compared with other KIT exon 11 mutants. Consequently, the fraction of patients with gastric GIST harboring KITdel-inc557/558 who relapsed 5 years after surgery was twice as high as patients with other KIT exon 11 mutations (61% vs. 29%, respectively; P < 0.0001).

Remarkably, even in tumors classified as non–high risk [(very)low and intermediate] according to modified NIH criteria, and originating from the stomach (n = 320), the presence of KITdel-inc557/558 remained an important prognosticator for poor outcome in comparison with other KIT exon 11 mutations, KIT exon 9 and PDGFRA exon 18 mutations (P = 0.002; Fig. 1G). As indicated on Supplementary Fig. S1D, the same was true when comparing KITdel-inc557/558 with other exon 11 KIT mutations in non–high-risk tumors from the stomach (P = 0.002). This relationship was not observed in nongastric tumors (P = 0.45; data not shown).

Interestingly, when considering nontypical primary tumor locations, only one of the seven gastric GIST with KIT exon 9 mutation relapsed during the median 56.9 months of follow-up. Similarly, the frequency of relapse was lower in gastric versus nongastric PDGFRA exon 18 mutants [11.8% (14 of 119) vs. 25% (2 of 8), respectively]. However, the number of the tumors with KIT exon 9 and PDGFRA exon 18 mutations originating from nontypical anatomical sites was too low for a conclusive analysis.

In the present work, the prognostic value of the genotype in localized GIST, along with the already known prognostic factors for DFS, was investigated in a large prospective series of patients with GIST, diagnosed in four different European countries with the longest follow-up time published to date. On multivariate analysis, relapse was predicted by MI >10 mitoses/50 HPF, tumor size >10 cm, primary nongastric tumor location, and independently by a specific tumor genotype. In particular, the effect of tumor size (HR, 6.75) and mitotic count (HR, 5.7) was remarkable, and also evident for tumor location (HR 1.64). Surprisingly, tumor rupture has not been shown to be an independent predictor of outcome on multivariate analysis. The reasons for the latter remain speculative, but the completeness of reporting of tumor rupture might have had a role, because this parameter was frequently not reported in the older cases and was lacking in nearly 20% of patients. In addition, the intrinsic characteristics of our cohort, which included relatively lower number of tumors ≤2 cm in size than other pooled series (only 7.8% vs. 12.3–18.2%, respectively; ref. 4), could play a role.

Most importantly, our results demonstrate the independent significance of the KITdel-inc557/558 mutation, which represents an adverse relapse indicator for gastric GIST, while confronting this factor with the site of tumor origin and other commonly used risk classification criteria (modified NIH classification). Several previous reports have documented the increased risk of recurrence in GIST harboring a KIT exon 11 deletion as compared with another type of exon 11 mutation, KIT exon 9 mutation, or KIT/PDGFRA WT genotype (14, 16, 19–21). In particular, the KIT exon 11 deletions involving codons 557 and/or 558 were associated with metastasis and aggressive disease (12, 13, 22). Nevertheless, the association of KITdel-inc557/558 mutation as poor prognosticator specifically in GIST of gastric origin was hitherto not reported. Most likely, previous studies were underpowered to reveal this correlation. Interestingly, although the KITdel-inc557/558 mutations were equally common in tumors of gastric and nongastric origin in our cohort, they were significantly more frequently identified in patients ≤60 years, in tumors >5 cm, with mitotic rate >5/50 HPF, and consequently in high-risk GIST when compared with other KIT mutations. Furthermore, patients with KITdel-inc557/558 had significantly shorter DFS as compared with those with other KIT exon 11 mutations. These characteristics are most likely related to the inherently higher aggressive potential (possibly more robust KIT signaling) of KITdel-inc557/558 mutants. Importantly, the presence of a KITdel-inc557/558 mutation was associated with poorer outcomes both in high-risk and non–high-risk patients with gastric GIST. This indicates its additional prognostic value for patient selection for adjuvant therapy, especially because this mutation is known to be sensitive to imatinib (23).

Markedly, KIT exon 9 duplication was also associated with inferior DFS in the multivariate analysis of our study. In our cohort, KIT exon 9 mutation was observed mostly in nongastric GIST, as previously reported, and is known to have poor clinical outcome (24, 25). However, it is well known that compared with gastric tumors, small intestinal tumors with similar size and mitotic activity have a markedly worse prognosis (8). In our series, GISTs localized outside of the stomach were larger, had a higher mitotic rate, and had worse 5-year DFS (34.2% vs. 58.1%) in comparison with gastric tumors. Nonetheless, comparison between tumors with KIT exon 9 and KIT exon 11 mutations (both, KITdel-inc557/558 and other KIT exon 11) of nongastric origin did not show differences in tumor clinical behavior as assessed by survival analysis. Thus, we conclude that in extragastric sites, the worse prognosis of KIT exon 9 mutants is related to the tumor location rather than to an intrinsic aggressive biologic nature of this mutation, similarly to what was suggested by others (26, 27). In support of this notion, the vast majority of gastric KIT exon 9 mutants in our study (6 of 7) belonged to the non–high-risk category, and only one of them progressed with relatively long DFS (56 months).

PDGFRA mutations are reported in 1.6% to 2.7% of advanced GIST treated in phase III clinical trials (27, 28) and up to 12.9% to 16% of primary tumors in population studies (15, 16, 30). In our study, PDGFRA was mutated in 14% of cases, and these mutants were almost exclusively (90.5%) of gastric origin, as previously reported (15, 26, 31, 32). Also in agreement with earlier reports, the most prevalent genotype was the p.D842V substitution (59.5% of all PDGFRA mutants). A better outcome and a lower chance of metastasis of GIST carrying PDGFRA mutations were suggested before (15, 31). In our study, the identification of PDGFRA mutations in GIST was a strong predictor of good clinical outcome on its own by multivariate analysis, and this molecular factor could therefore add a significant value to the current consensus risk criteria used for GIST stratification. In addition, given that these mutations comprise the majority of cases with the imatinib-resistant p.D842V subtype (28, 29, 33), mutational testing is highly relevant, particularly to avoid overtreatment of gastric tumors with imatinib in the adjuvant setting. Recent prospective ACOSOG Z9001 adjuvant imatinib GIST trial reports an excellent survival of patients with PDGFRA mutations in placebo group, providing a good argument that these patients may not need the adjuvant treatment at all (33). This issue warrants further studies, possibly using pooled cohorts stratified for the mutational status from the available prospective clinical trials in the adjuvant setting.

The main limitation of this study is that it is not a population-based analysis. Yet, the main clinicopathologic and molecular characteristics, and the frequency of most common types of KIT or PDGFRA mutations were comparable with what has been reported in the MolecGIST prospective population-based study of 492 patients (30). Particularly, the proportion of patients with intermediate/high-risk tumors (in our study 52.1% vs. 55.0% in MolecGIST) and the frequencies of KIT exon 11, KIT exon 9, and PDGFRA mutations (our study 71.1%, 7.4%, and 14.0% vs. MolecGIST 62.3%, 5.5%, and 15%, respectively) were reasonably similar. In addition, pathologic and molecular features of our cohort matched the data from the placebo arm of the ACOSOG Z9001 adjuvant imatinib GIST trial, including tumor primary location (55.5% vs. 62.4% of gastric GIST, respectively), median tumors size (6.0 cm vs. 6.5 cm, respectively), and MI (4/50 HPF vs. 3/50 HPF, respectively; ref. 21). Likewise, the KIT and PDGFRA mutations frequency in ACOSOG Z9001 trial was similar to what we found in our series (8.5% vs. 7.4% KIT exon 9, 66.3% vs. 61.3% KIT exon 11, 10.5% vs. 14% PDGFRA mutants; ref. 21). The ACOSOG trial design excluded patients with tumors ≤3 cm, however, which may explain the slightly higher prevalence of KIT exon 11 mutations and lower frequency of PDGFRA mutants as compared with our study.

The ability to predict GIST outcome constitutes a key element for the proper counseling and management of patients. Particularly, accurate prognostication is essential to identify high-risk tumors, for which an efficient adjuvant systemic treatment is available. The most commonly used routine risk-stratification schemes categorize tumor size and MI, which might cause incorrect risk estimation when both of these variables are close to a cutoff value. Mitotic count is a strong prognostic factor in GIST, but its reliability is controversial. The recent report on 506 patients who underwent surgery for a localized GIST (34) revealed that the risk of relapse (evaluated according to AFIP classification) was underestimated in more than 30% of the cases. Subsequently, these patients were undertreated and had a significant rate of relapse compared with patients that benefited from an accurate estimation of the risk of relapse. The U.S. National Comprehensive Cancer Network now recommends consideration of adjuvant imatinib for at least 3 years for patients with a high risk of GIST recurrence (35). Thus, intermediate-risk patients are ineligible for adjuvant imatinib, although 15% to 20% of these patients will progress. Our findings have an important consequence for the selection of patients with intermediate gastric GIST as candidates for adjuvant treatment. Our practical proposal of using mutational analysis when considering adjuvant therapy is presented in Fig. 2. It remains to be shown whether patients with GIST with KITdel-inc557/558 genotype should not be kept longer (or even lifelong) on adjuvant imatinib treatment than standard 3 years.

In conclusion, our results underscore the importance of routine genotyping for an optimal management of GIST, either in the adjuvant or advanced disease setting. Evaluation of tumor genotype has especially prognostic value in GIST of gastric origin. Our findings contribute to the identification of a subset of patients with localized gastric GIST with higher risk of relapse irrespectively of standard clinical prognostic factors. The presence of a KITdel-inc557/558 or PDGFRA exon 18 mutations could be used as an additional parameter for more accurate selection of patients for adjuvant therapy.

A. Wozniak and E. Wardelmann report receiving speakers bureau honoraria from Novartis. P. Rutkowski reports receiving speakers bureau honoraria from Novartis and Pfizer and is a consultant/advisory board member for Bayer and Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Wozniak, P. Rutkowski, P. Schöffski, J.-Y. Blay, E. Wardelmann, J.-M. Coindre, M. Debiec-Rychter

Development of methodology: A. Wozniak, H.-U. Schildhaus, J.-Y. Blay, J.A. Siedlecki, J.-M. Coindre, M. Debiec-Rychter

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Wozniak, P. Rutkowski, P. Schöffski, I. Ray-Coquard, H.-U. Schildhaus, A. Le Cesne, E. Bylina, J.-Y. Blay, J.A. Siedlecki, E. Wardelmann, R. Sciot, J.-M. Coindre, M. Debiec-Rychter

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Wozniak, P. Rutkowski, P. Schöffski, I. Ray-Coquard, I. Hostein, H.-U. Schildhaus, A. Le Cesne, J.-Y. Blay, M. Debiec-Rychter

Writing, review, and/or revision of the manuscript: A. Wozniak, P. Rutkowski, P. Schöffski, I. Ray-Coquard, I. Hostein, H.-U. Schildhaus, A. Le Cesne, J. Limon, J.-Y. Blay, E. Wardelmann, R. Sciot, J.-M. Coindre, M. Debiec-Rychter

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Wozniak, E. Bylina, J.-Y. Blay, J.-M. Coindre

Study supervision: A. Wozniak, P. Rutkowski, J.-M. Coindre, M. Debiec-Rychter

The ConticaGIST database was supported by CONTICANET, FP6 Network of Excellence (FP6-018806), and KU Leuven Concerted Action Grant 2011/010 (to M. Debiec-Rychter and R. Sciot).

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.