Clinical Implications of The Cancer Genome Atlas Molecular Classification System in Esophagogastric Cancer

Henry S. Walch', Raktim Borpatragohain?, Justin Jee?, Waleed Chatila, Christopher Fong4, Steven B. Maron?, Geoffrey Y. { \mathsf { K U } } ^ { 5 } David H. Ilson', Yelena Y. Janjigian', Abraham J. Wu, Pari Shah', Daniel G. Coit?, Manjit S. Bains?, Valerie W. Rusch?, Bernard J. Park?, Matthew J. Bott?, Katherine Gray?, David R. Jones?, Michael Berger', Nikolaus Schultz', Vivian E. Strong, Daniela Molena?, and Smita Sihag?

ABsTrACT

Purpose: The Cancer Genome Atlas (TCGA) project defined four distinct molecular subtypes of esophagogastric adenocarcinoma: microsatellite instable (MSI), Epstein-Barr virus (EBV)-associated, genomically stable (GS), and chromosomally instable (CIN). However, an association between molecular subtypes and clinical outcomes has not been clearly demonstrated. Given few actionable biomarkers, we investigated the clinical relevance of TCGA classification system.

Experimental Design: We identified all patients with esophagogastric adenocarcinoma whose tumors underwent prospective next-generation sequencing using the Memorial Sloan KetteringIMPACT assay from 2014 to 2023. We classified all tumors in accordance with TCGA methodology and correlated molecular subtypes with high-quality clinicopathologic data.

Results: Among 1,438 included patients, 941 had CIN, 344 had GS, 103 had MSI, and 50 had EBV tumors. Accounting for the clinical stage and tumor grade, molecular classification was independently associated with overall cancer-specific survival ( P < 0 . 0 0 1 ) on Cox multivariable analysis. Furthermore, genomic signatures, patient demographics, pathologic responses to neoadjuvant therapy, patterns of recurrence, and metastatic organotropism differed significantly by molecular subtype. Although most distal esophageal and gastroesophageal junction tumors were CIN, up to 2 5 % of these included GS, MSI, or EBV subtypes in contrast to TCGA. Random forest machine learning demonstrated that the molecular subtype is more influential in predicting response to treatment than tumor location.

Conclusions: Molecular classification is independently prognostic and may warrant inclusion in future staging and treatment guidelines. Routine molecular profiling is clinically feasible and may play a role in the management of patients to help guide appropriate treatment selection and clinical trial enrollment in the place of tumor location.

Introduction

Esophageal and gastric cancers are among the top 10 cancers worldwide in incidence and mortality (1). In 2015, The Cancer Genome Atlas (TCGA) defined four distinct molecular subtypes of gastric cancer: microsatellite instable (MsI), Epstein-Barr virus (EBV)-associated, genomically stable (GS), and chromosomally instable (CIN; ref. 2). The genomic features that characterize each subtype include hypermutation in MSI, global hypermethylation in EBV, focal high-level amplifications of receptor tyrosine kinases and aneuploidy in CIN, and enrichment of RHOA and somatic CDH1 mutations in GS. Subsequent work by TCGA demonstrated that distal esophageal and gastroesophageal junction (GEJ) tumors fit into a larger landscape of esophagogastric adenocarcinomas in which esophageal and gastric CIN tumors share overwhelming similarity, and the tumor location is strongly correlated with the relative frequency of molecular subtypes (3).

The precision oncology promise of the landmark TCGA studies was to provide a roadmap for patient stratification to guide treatment approaches and clinical trials of novel therapeutics in esophagogastric adenocarcinoma. However, an association between the molecular subtype and clinical outcome has not been demonstrated in TCGA cohort. Concurrent work by the Asian Cancer Research Group (ACRG) applied TCGA classification to the ACRG study cohorts and found that, although the MSI group experienced improved survival, the association was not statistically significant (4). Other studies have attempted to correlate molecular subtypes with clinicopathologic characteristics and survival outcomes, but these have either generated alternative molecular classification schemes with potential prognostic relevance (4-6) or been limited by small sample sizes or methodologic challenges in accurately classifying tumors in accordance with TCGA methodology (7, 8). Moreover, comparison of pathologic treatment responses by molecular subtype has not been widely feasible because of the inclusion of treatmentnaive patients in many of these sequencing efforts.

Translational Relevance

The Cancer Genome Atlas previously defined four molecular subtypes of esophagogastric adenocarcinoma with the aim of creating a roadmap for effective patient stratification to guide treatment approaches and clinical trials, but an association between the molecular subtype and clinical outcome was not demonstrated in The Cancer Genome Atlas cohort. Given relatively few actionable biomarkers, tumor location has instead primarily determined treatment strategy and clinical trial eligibility. In this study, we report the largest known integrative analysis of genomic and clinicopathologic data in 1,438 patients with esophagogastric adenocarcinoma. We demonstrate that molecular classification is independently prognostic and may warrant inclusion in future staging guidelines. Furthermore, genomic signatures, patient demographics, treatment responses, patterns of recurrence, and metastatic organotropism differed significantly by molecular subtype, suggesting that routine molecular profiling may play a role in the management of patients with esophagogastric cancer to help guide appropriate treatment selection and clinical trial enrollment.

As a result, none of these genomic classification strategies have been incorporated into clinical algorithms, and esophageal and gastric treatment paradigms remain largely separate. Current American Joint Committee on Cancer (AJCC) staging and National Comprehensive Cancer Network (NCCN) treatment guidelines delineate the boundary of esophageal and gastric cancers at a somewhat arbitrary threshold of 2 {cm } from the cardia (9, 10). Furthermore, clinical trials have historically assigned eligibility criteria by tumor location-in particular, coupling GEJ tumors with either gastric or esophageal cancers, without specific justification beyond proximity (11-13). Therefore, our objective was to investigate the clinical relevance of TCGA molecular classification scheme in a large, well-annotated cohort of patients with esophagogastric adenocarcinoma who underwent prospective broad-panel next-generation sequencing (NGS). We hypothesized that molecular classification of esophagogastric tumors is feasible using a targeted clinical molecular profiling assay and that successful correlation with clinicopathologic outcomes may have significant implications for future staging and treatment guidelines, as well as for the design analysis and interpretation of clinical trials.

Materials and Methods

Patients and samples

In this retrospective cohort study, we used the Memorial Sloan Kettering (MSK) cBioPortal to identify all patients with esophagogastric adenocarcinoma who underwent prospective sequencing using the MSK-IMPACT NGS assay from 2014 to 2023 (Supplementary Methods S1; refs. 14, 15). As of 2017, a large-scale effort to routinely sequence tumors from all patients with esophagogastric cancer was initiated. All patients provided written informed consent for targeted sequencing (protocol NCT01775072; approved by the MSK Institutional Review Board and in compliance with the US Common Rule). Tumor tissue for sequencing was obtained from primary or metastatic sites at the time of biopsy or surgery. Tumor purity was assessed by histopathologic review of specimens by an expert pathologist at MSK. Clinicopathologic annotations were obtained using natural language processing algorithms to automatically extract variables from the MSK Institutional Database and electronic medical records, including radiology and histopathology reports, as well as procedural notes and oncologic treatment summaries, as described previously (16), as part of the MSK Cancer Data Science Initiative. Algorithms were initially developed and trained on the Project GENIE Biopharma Collaborative dataset of the American Association for Cancer Research and validated using the MSK-Biopharma Collaborative patient subsets. All natural language processing-extracted data were then manually reviewed and verified.

Molecular classification strategy

Our molecular classification strategy is depicted in Fig. 1A. EBV tumors were identified using Epstein-Barr encoding region ISH (17). MSI status was assessed using the MSI-sensor algorithm, which calculates the percentage of microsatellite loci covered by the MSKIMPACT assay that are unstable in the tumor compared with the matched normal sample. Samples with a score >= 1 0 were classified as MSI (18). For EBV-negative microsatellite-stable tumors, the FACETS algorithm (19) was used to calculate a clonal deletion score (CDS) and define a threshold to distinguish between CIN and GS subtypes on the basis of TCGA approach (Supplementary Methods S1; ref. 20). Tumors were classified as CIN if they had either a CDS { >= } 0 . 1 5 6 or evidence of high-level focal oncogenic amplifications present on evaluation of genomic segmentation data. Tumors with a CDS < 0 . 1 5 6 in the absence of any focal or high-level amplifications were classified as GS.

Histopathologic assessment of treatment responses

All surgical specimens in the curative-intent cohort were reviewed by an expert gastrointestinal pathologist. In posttreatment specimens, pathologic response in the primary tumor bed was quantified as the percentage treatment effect in terms of residual viable carcinoma in relation to areas of fibrosis or fibroinflammation within the gross lesion, as well as according to College of American Pathologists tumor regression grade (CAP-TRG) guidelines (21). Pathologic complete response (pCR) was defined as 1 0 0 % treatment response (i.e., CAP-TRG score 0 and ypT0N0), whereas pathologic major response (pMR) was defined as 9 0 % treatment response (i.e., CAP-TRG score 0 or 1 and ypN0), as previously described (22). Patients without a pCR or pMR were considered to have a nonresponse. Clinical staging and pathologic staging were performed in accordance with the eighth edition of the AJCC Staging Manual for Esophageal or Gastric Cancer. The tumor location reflected the tumor epicenter as reported on baseline endoscopy in conjunction with axial imaging and/or surgical pathology reports. Pathology reports from all patients with tumors localized to the distal esophagus and GEJ were screened for the presence of Barrett's/intestinal metaplasia or dysplasia to annotate the Barrett's positive versus negative cohort.

Statistical analysis

Clinicopathologic characteristics were summarized using frequency and percentage for categorical variables and median and IQR for continuous variables. The association between molecular subtypes and genomic driver alterations, pathways, features, and clinicopathologic factors was evaluated using the Wilcoxon ranksum test (continuous variables) or the Fisher exact test (categorical variables). P < 0 . 0 5 was considered statistically significant; these values were corrected using the Benjamini-Hochberg procedure to account for multiple hypothesis testing where appropriate. For analyses of long-term outcomes stratified by molecular classification, overall cancer-specific survival (OCss) and recurrence-free survival (RFS) were estimated using the Kaplan-Meier method and compared using the log-rank test. OCsS was measured from the date of diagnosis to the date of esophagogastric cancer-related death or last follow-up, and RFS was measured from the date of curative-intent surgery to the date of esophagogastric cancer recurrence. The cumulative incidence of recurrence (1-minus Kaplan-Meier estimate) measured all instances of disease recurrence after the date of surgery, accounting for the competing risk of death. Multivariable Cox proportional hazard models were used to evaluate the association between molecular subtypes, relevant clinicopathologic variables, and OCss. Random forest machine learning algorithms were used to predict treatment responses to neoadjuvant therapy, incorporating independent variables found to be associated with pMR (Supplementary Methods S1). All statistical analyses were performed using R (version 3.6.3; R Foundation for Statistical Computing).

Data availability

Data are available on Zenodo at the following link: https:// zenodo.org/records/14833157. Please contact the corresponding author for any further data requests.

Results

Genomic landscape of molecular subtypes

In total, 1,438 patients were included. Of these, 50 had EBV tumors, 103 had MSI tumors, 941 had CIN tumors, and 344 had GS tumors (Fig. 1B). The genomic landscape of each molecular subtype in our cohort was strongly concordant with results published by TCGA, validating our molecular classification approach using MSKIMPACT. The CIN subtype was associated with the highest frequency of TP53 mutations ( 8 1 % ) and oncogenic amplifications involving cell-cycle and RTK-RAS pathway oncogenes and the highest overall fraction of genome altered (Fig. 1C and D; Supplementary Fig. S1A). These tumors predominantly localized to the lower esophagus and GEJ (Fig. 1E). GS tumors were associated with a lower frequency of TP53 mutations ( 5 5 % ) , instead harboring alterations in ARID1A and CDH1 genes, without any robust pathwaylevel signal, and these tumors had the lowest frequency of clinically actionable alterations (Fig. 1C and D; Supplementary Fig. S1B). MSI tumors were characterized by hypermutation and alterations in several different genes and oncogenic pathways-most notably, the DNA mismatch repair pathway (Fig. 1C and D, Supplementary Fig. S1C). EBV tumors had the lowest frequency of TP53 alterations ( 3 3 % ) and were enriched in ARID1A and PIK3CA pathway gene alterations (Fig. 1C and D). As expected, GS, MSI, and EBV tumors were predominantly gastric (Fig. 1E), although these subtypes comprised up to 2 5 % of tumors that localized to the distal esophagus and GEJ as well.

Association between molecular subtypes and clinicopathologic characteristics

Patient characteristics stratified by molecular subtype are summarized in Table 1 and Supplementary Table S1. Demographic characteristics, such as sex and race, were representative of esophagogastric adenocarcinoma in the United States and varied significantly by molecular subtype (Supplementary Table S2; Fig. 2A). Overall, patients with EBV or MSI tumors were diagnosed at earlier stages and therefore were more likely to receive curativeintent treatment, including surgery, whereas patients with CIN or GS tumors were more likely to have advanced disease at diagnosis and undergo palliative therapies (Fig. 2A). Identification of Barrett's metaplasia or dysplasia within the biopsy or surgical specimen was not associated with a particular subtype, whereas Lauren's diffuseor mixed-type classification and poorly differentiated signet-ring histology was correlated with the GS subtype (Fig. 2B). A significantly higher proportion of EBV and MSI tumors had a PDL1 combined positive score >= 5 compared with CIN and GS tumors, and GS tumors were largely PD-L1 negative (Fig. 2B).

Association between molecular subtypes and overall survival

Kaplan-Meier analysis demonstrated that the molecular subtype was associated with OCss in our cohort (Fig. 2C). Patients with EBV or MSI tumors had significantly better OCSS than patients with CIN or GS tumors, both of which were associated with simi larly poor OCss. When subtypes were stratified by treatment intent, this association remained significant in both the curative-intent and palliative-intent cohorts (Supplementary Fig. S2A and S2B). We then investigated the effect of immunotherapy as a possible underlying factor for better OCSS in patients with EBV or MSI tumors. We observed that although immunotherapy may have contributed to better outcomes among patients with MSI tumors in the palliativeintent cohort (Supplementary Fig. S2C and S2D), comparisons were limited by the fact that relatively few patients in the palliative-intent cohort did not receive any immunotherapy throughout their treatment course and relatively few patients in the curative-intent cohort did. Next, we performed a Cox univariable analysis to identify all prognostic clinicopathologic variables in our cohort (Supplementary Table S3). As expected, the clinical stage, race, tumor grade, curative surgery, treatment intent, and molecular subtype were associated with OCSs. Subsequent Cox multivariable analysis revealed that molecular subtypes remained prognostic when independent variables associated with OCss, such as the clinical stage and tumor grade, were accounted for (Fig. 2D).

Association between molecular subtypes and treatment response

To assess whether molecular subtypes can guide treatment selection, we analyzed the curative-intent cohort ( n = 6 0 6 to evaluate pathologic response rates to different neoadjuvant therapy regimens in locally advanced disease by molecular subtype ( n = 4 3 3 ). Most patients with CIN tumors received neoadjuvant chemoradiotherapy, whereas most patients with GS, MSI, and EBV tumors received neoadjuvant chemotherapy alone (Supplementary Table S4) in accordance with evolving NCCN guidelines for esophageal, GEJ, and gastric tumors during the study period. Neoadjuvant regimens infrequently included immune checkpoint inhibitors and/or targeted therapies, such as trastuzumab, except in the context of ongoing clinical trials. Overall, the rate of pCR to neoadjuvant therapy was lowest in patients with GS tumors at 8 % and the pCR rate in patients with CIN tumors was 1 6 % (Table 2; Fig. 3A). The pCR and pMR rates after neoadjuvant chemotherapy alone in patients with CIN or GS tumors were similar, at { ~ } 6 % to 9 % and { ~ } 1 5 % to 1 7 % signifying that nearly 8 5 % of patients had no meaningful response to neoadjuvant chemotherapy alone. However, patients with CIN tumors responded to the addition of neoadjuvant radiation to a much higher degree than patients with GS tumors, with pCR rates of 2 1 % versus 1 3 % and pMR rates of 3 9 % versus 1 9 % ( P < 0 . 0 1 ) Although relatively few patients with MSI or EBV tumors received neoadjuvant therapy, patients with MSI tumors had the lowest rate of response to chemotherapy alone ( 9 3 % nonresponse rate), whereas patients with EBV tumors had the highest ( 2 2 % pMR rate). Among a small number of patients with MSI tumors who received immune checkpoint blockade in combination with either neoadjuvant chemotherapy or neoadjuvant chemoradiotherapy, 0 of 3 responded to neoadjuvant chemoimmunotherapy, and 5 of 5 responded to neoadjuvant chemoimmunoradiotherapy. The addition of immunotherapy to neoadjuvant chemoradiotherapy also led to an improvement in pMR rates, to ~ 5 0 % in patients with CIN or GS tumors.

We then used random forest machine learning to directly quantify the influence of molecular subtypes on the probability of pMR (Fig. 3B) and identified a strong association P < 0 . 0 0 0 1 { { \ : } } Other factors associated with the probability of pMR included the treatment regimen, clinical stage, tumor location, and tumor grade (Supplementary Fig. S3A-S3D). The relative importance of each of these variables in the model indicates that the contribution of the molecular subtype is just less than 2 0 % but remains higher than the tumor location (Supplementary Fig. S3E). In combination, the model achieved a receiver operator area under the curve score of 7 2 % as pMR remains a relatively infrequent outcome for these patients (Supplementary Fig. S3E). Of note, removing the tumor location from the model led to a slight improvement in its predictive capacity, whereas removing the molecular subtype from the model led to a decline (Supplementary Table S5).

Association between molecular subtypes and recurrence

Despite more robust responses to neoadjuvant therapy, patients with CIN tumors had a significantly higher cumulative incidence of recurrence and shorter RFS than patients with GS tumors (Fig. 3C and D).Patients with MSI or EBV tumors had a lower cumulative incidence of recurrence and better outcomes in the event of recurrence, as nearly all were treated with immune checkpoint inhibitors in this setting. Furthermore, we observed significant variability in the pattern of disease recurrence by molecular subtype in terms of organotropism related to the first site(s) of metastatic disease (Fig. 3E). MSI and EBV tumors had a high propensity to recur in either locoregional or distant lymph nodes, whereas GS tumors were more likely to spread to the peritoneum, causing carcinomatosis. CIN tumors were less preferential, given their high metastatic potential, and were the only subtype to be associated with brain metastases.

Discussion

Molecular classification of esophagogastric tumors, as defined by TCGA, is both clinically feasible and valuable to potentially guide treatment selection and optimize patient outcomes. This is the first large, clinically well-annotated cohort study to demonstrate that TCGA molecular classification is independently prognostic. Therefore, the inclusion of this system in future AJCC staging guidelines may be warranted. Overall, prognostic differences in our cohort were largely driven by better outcomes for MSI and EBV subtypes. Both groups are known to exhibit exceptional responses to immune checkpoint inhibitors (23), and we observed this in our cohort. On the basis of our analysis, however, it is not clear whether the survival benefit seen in patients with MSI or EBV tumors is directly attributable to treatment with immunotherapy. Survival outcomes among immunotherapy-naive patients treated in the ACRG, CLASsIC, and MAGIC trials suggest that the underlying biology of both MSI and EBV tumors, related to heightened immune surveillance, may play a dominant role in determining their favorable prognosis and low potential for metastatic proliferation (4, 24). This idea is consistent with the fact that patients with MSI or EBV tumors in our cohort were more likely to have localized disease at the time of diagnosis and experienced excellent long-term outcomes after resection.

Among patients treated with neoadjuvant therapy, those with EBV tumors had the best response to chemotherapy alone, whereas those with MSI tumors had the worst response, as shown previously (25). In a small number of patients with MSI tumors, the addition of immunotherapy to chemotherapy did not seem to change outcomes; however, the addition of immunotherapy to chemoradiation had greater effect. By contrast, initial results from prospective clinical trials evaluating the efficacy of chemotherapy plus immunotherapy for junctional and gastric tumors suggest better outcomes potentially for both MSI and microsatellite-stable patients alike (26-28). Current guidelines support use of combined CTLA-4 and PD-1 inhibition in MSI patients (29, 30), although only one patient in this cohort underwent surgery following this regimen (others were closely observed per protocol). Thus, the optimal neoadjuvant regimen in patients with MSI tumors remains to be fully determined.

Whereas both EBV and MSI status can be effectively assessed in _ { < 1 } week using techniques from molecular pathology, including Epstein-Barr encoding region ISH and mismatch repair protein IHC, differentiation between CIN and GS tumors ideally requires analysis of NGS data. Arguably, until now, there has been little clinical justification to pursue further molecular classification of HER2-negative, microsatellite-stable tumors. Our observations indicate that, although overall survival (OS) remains similarly poor, treatment responses may vary significantly between CIN and GS tumors, especially with respect to the benefit of neoadjuvant radiotherapy. In particular, the role of neoadjuvant radiotherapy in the treatment of locally advanced GEJ adenocarcinomas has been a matter of debate during the past decade. The results of CROsS and CALGB 80803 trials support the use of neoadjuvant chemoradiotherapy, with higher rates of pathologic downstaging, whereas the more recent FLOT4 and now ESOPEC trials show better OS with perioperative FLOT chemotherapy compared with most prior regimens (11, 13, 31, 32). Moreover, dedicated randomized controlled trials and meta-analyses comparing neoadjuvant chemotherapy and chemoradiotherapy have failed to show an OS advantage for the addition of radiotherapy, despite improvement in local controls (33-35). Data from the present study suggest that were a post hoc subgroup analysis performed for many of the abovementioned clinical trials by molecular subtype, such an analysis may reveal that corresponding populations of treatment responders or nonresponders are heavily driving reported outcomes.

To date, the tumor location has been used as a surrogate for molecular classification to determine treatment algorithms and clinical trial enrollment, with the assumption that tumors within 2 \ {cm } of the gastric cardia share similar cellular origin and pathogenesis. In contrast to TCGA data, our data from a much larger cohort of patients indicate that although molecular classification and the tumor location are closely related, they are not perfectly overlapping. In fact, up to 2 5 % of distal esophageal and GEJ tumors seem to be subtypes other than CIN. Furthermore, our machine learning model demonstrates that the molecular subtype is more influential than the tumor location in predicting a response to neoadjuvant therapy. Therefore, routine implementation of targeted NGS with a clinically actionable turnaround time may be feasible and justifiable for this cancer type to guide optimal treatment selection and clinical trial enrollment in the place of the tumor location. As relatively few biomarkers are currently available to direct therapy, the results of the present study lay the groundwork for further integration of TCGA molecular classification system into current staging and treatment guidelines and in the design and interpretation of clinical trials in esophagogastric cancer.

Authors' Disclosures

S.B. Maron reports personal fees from Daiichi Sankyo, Bicara Therapeutics, Novartis, Amgen, Elevation Oncology, Purple Oncology, PineTree Therapeutics, Bolt Biotherapeutics, and OneCell Diagnostics, grants from Paige.ai and AstraZeneca, and nonfinancial support from Guardant Health outside the submitted work. G.Y. Ku reports grants and personal fees from AstraZeneca, Bristol Myers Squibb, Daiichi Sankyo, I-MAB, Jazz Pharmaceuticals, Merck, Oncolys BioPharma, Pieris Pharmaceuticals, and Zymeworks, grants from CARsgen Therapeutics and Triumvira Immunologics, and personal fees from Bayer and DAVA Oncology during the conduct of the study. Y.Y. Janjigian reports research funding from Astellas Pharma Inc., AstraZeneca, Arcus Biosciences, Bayer, Bristol Myers Squibb, Cycle for Survival, Department of Defense, Eli Lilly and Company, Fred's Team, Genentech/Roche, Inspirna, Merck, NCI, Stand Up To Cancer, and Transcenta; membership in advisory boards and employment as a consultant in AbbVie, AmerisourceBergen, AskGene Pharma, Inc., Arcus Biosciences, Astellas Pharma Inc., AstraZeneca, Basilea Pharmaceutica, Bayer, Boehringer Ingelheim, Bristol Myers Squibb, Clinical Care Options, Daiichi Sankyo, eChinaHealth, ED MedResources (OncInfo), Eisai, Eli Lilly and Company, Geneos Therapeutics, GlaxoSmithKline, Guardant Health, Inc., H.C. Wainwright & Co., iMedX, Imugene, Inspirna, Lynx Health, Master Clinician Alliance, Merck, Merck Serono, Mersana Therapeutics, Michael J. Hennessy Associates, Paradigm Medical Communications, PeerMD, PeerView Institute, Pfizer, Physician's Education Resource, LLC, Research to Practice, Sanofi Genzyme, Seagen, Silverback Therapeutics, Suzhou Liangyihui Network Technology Co., Ltd., Talem Health, TotalCME, WebMD, LLC, and Zymeworks; and ownership of stock options in Inspirna and Veeda LifeSciences, Inc. A.J. Wu reports other support from Simphotek, Inc. outside the submitted work. V.W. Rusch reports unpaid membership in the Data Safety and Monitoring Committee of MARS II and RAMON trials and Cancer Research, UK. D.R. Jones reports other support from AstraZeneca and grants from Merck outside the submitted work. M. Berger reports personal fees from AstraZeneca, Eli Lilly and Company, and Paige.ai outside the submitted work, as well as intellectual property rights from SOPHiA Genetics. V.E. Strong reports personal fees from AstraZeneca and Merck outside the submitted work. D. Molena reports personal fees from AstraZeneca, Medtronic, Johnson & Johnson, Boston Scientific, Bristol Myers Squibb, and OncLive outside the submitted work. S. Sihag reports personal fees from AstraZeneca, Intuitive Surgical, and Medtronic and grants from Delfi Diagnostics outside the submitted work. No disclosures were reported by the other authors.

supervision, validation, methodology.V.E.Strong: Supervision, investigation, writing-review and editing. D. Molena: Supervision, investigation, writing-review and editing.S. Sihag: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, writing-original draft, project administration.

Authors' Contributions

H.S. Walch: Data curation, formal analysis, methodology, writing-original draft. R. Borpatragohain: Data curation, formal analysis, visualization. J. Jee: Data curation, software. W. Chatila: Data curation, software. C. Fong: Data curation, software. S.B.Maron: Methodology, writing-review and editing. G.Y. Ku: Writing-review and editing. D.H. Ilson: Writing-review and editing. Y.Y. Janjigian: Resources, project administration, writing-review and editing. A.J. Wu: Writing-review and editing.P. Shah:Writing-review and editing.D.G.Coit: Writing-review and editing. M.S. Bains: Writing-review and editing. V.W. Rusch: Writing-review and editing. B.J. Park: Writing-review and editing. M.J. Bott: Writing-review and editing. K. Gray: Writing-review and editing.D.R. Jones: Resources, supervision, project administration. M. Berger: Resources, supervision, methodology, project administration. N. Schultz: Resources, software,

Acknowledgments

This study was supported, in part, by the NIH/NCI (Cancer Center Support Grant P30 CA008748). S. Sihag is supported by an American Association for Thoracic Surgery Research Scholarship and Department of Defense Career Development Award (CA220781).

Note

Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

Received October 22, 2024; revised December 3, 2024; accepted March 12, 2025; posted first April 29, 2025.

References

1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2021;71: 209-49.
2. Cancer Genome Atlas Research Network. Comprehensive molecular characterization of gastric adenocarcinoma. Nature 2014;513:202-9.
3. Cancer Genome Atlas Research Network; Analysis Working Group: Asan University; BC Cancer Agency; Brigham and Women's Hospital; Broad Institute; Brown University, et al. Integrated genomic characterization of oesophageal carcinoma. Nature 2017;541:169-75.
4. Cristescu R, Lee J, Nebozhyn M, Kim KM, Ting JC, Wong SS, et al. Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med 2015;21:449-56.
5. Jeong YS, Eun YG, Lee SH, Kang SH, Yim SY, Kim EH, et al. Clinically conserved genomic subtypes of gastric adenocarcinoma. Mol Cancer 2023;22:147.
6. Li B, Zhang F, Niu Q, Liu J, Yu Y, Wang P, et al. A molecular classification of gastric cancer associated with distinct clinical outcomes and validated by an XGBoost-based prediction model. Mol Ther Nucleic Acids 2023;31:224-40.
7. Sohn BH, Hwang JE, Jang HJ, Lee HS, Oh SC, Shim JJ, et al. Clinical significance of four molecular subtypes of gastric cancer identified by the Cancer Genome Atlas Project. Clin Cancer Res 2017;23:4441-9.
8. Nshizirungu JP, Bennis S, Mellouki I, Sekal M, Benajah DA, Lahmidani N, et al. Reproduction of the Cancer Genome Atlas (TCGA) and Asian Cancer Research Group (ACRG) gastric cancer molecular classifications and their association with clinicopathological characteristics and overall survival in Moroccan patients. Dis Markers 2021;2021:9980410.
9. Rice TW, Patil DT, Blackstone EH. 8th edition AJCC/UICC staging of cancers of the esophagus and esophagogastric junction: application to clinical practice. Ann Cardiothorac Surg 2017;6:119-30.
10. Ajani JA, D'Amico TA, Bentrem DJ, Chao J, Corvera C, Das P, et al. Esophageal and esophagogastric junction cancers, version 2.2019, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw 2019;17:855-83.
11. van Hagen P, Hulshof MC, van Lanschot JJ, Steyerberg EW, van Berge Henegouwen MI, Wijnhoven BP, et al. Preoperative chemoradiotherapy for esophageal or junctional cancer. N Engl J Med 2012;366:2074-84.
12. Wilke H, Muro K, Van Cutsem E, Oh SC, Bodoky G, Shimada Y, et al. Ramucirumab plus paclitaxel versus placebo plus paclitaxel in patients with previously treated advanced gastric or gastro-oesophageal junction adenocarcinoma (RAINBOW): a double-blind, randomised phase 3 trial. Lancet Oncol 2014;15:1224-35.
13. Al-Batran SE, Homann N, Pauligk C, Goetze TO, Meiler J, Kasper S, et al. Perioperative chemotherapy with fluorouracil plus leucovorin, oxaliplatin, and docetaxel versus fluorouracil or capecitabine plus cisplatin and epirubicin for locally advanced, resectable gastric or gastro-oesophageal junction adenocarcinoma (FLOT4): a randomised, phase 2/3 trial. Lancet 2019;393:1948-57.
14. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, et al. The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov 2012;2:4014.
15. Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal 2013;6:pl1.
16. Jee J, Fong C, Pichotta K, Tran TN, Luthra A, Waters M, et al. Automated realworld data integration improves cancer outcome prediction. Nature 2024;636: 728-36.
17. Fan H, Gulley ML. Molecular methods for detecting Epstein-Barr virus (Part I): in situ hybridization to Epstein-Barr virus-encoded RNA (EBER) transcripts. Methods Mol Med 2001;49:301-11.
18. Middha S, Zhang L, Nafa K, Jayakumaran G, Wong D, Kim HR, et al. Reliable pan-cancer microsatellite instability assessment by using targeted nextgeneration sequencing data. JCO Precis Oncol 2017;2017:1-17.
19. Shen R, Seshan VE. FACETS: allele-specific copy number and clonal heterogeneity analysis tool for high-throughput DNA sequencing. Nucleic Acids Res 2016;44:e131.
20. Liu Y, Sethi NS, Hinoue T, Schneider BG, Cherniack AD, Sanchez-Vega F, et al. Comparative molecular analysis of gastrointestinal adenocarcinomas. Cancer Cell 2018;33:721-35.e8.
21. College of American Pathologists. Protocol for the examination of specimens from patients with carcinoma of the esophagus; 2016 [cited 2024 Nov 27]. Available from: https:/documents.cap.org/protocols/cp-esophagus-2016-v3200.pdf.
22. Sihag S, Nobel T, Hsu M, De La Torre S, Tan KS, Janjigian YY, et al. Survival following trimodality therapy in patients with locally advanced esophagogastric adenocarcinoma: does only a complete pathologic response matter? Ann Surg 2022;276:1017-22.
23. Kim ST, Cristescu R, Bass AJ, Kim KM, Odegaard JI, Kim K, et al. Comprehensive molecular characterization of clinical responses to PD-1 inhibition in metastatic gastric cancer. Nat Med 2018;24:1449-58.
24. Pietrantonio F, Miceli R, Raimondi A, Kim YW, Kang WK, Langley RE, et al. Individual patient data meta-analysis of the value of microsatellite instability as a biomarker in gastric cancer. J Clin Oncol 2019;37:3392-400.
25. Janjigian YY, Sanchez-Vega F, Jonsson P, Chatila WK, Hechtman JF, Ku GY, et al. Genetic predictors of response to systemic therapy in esophagogastric cancer. Cancer Discov 2018;8:49-58.
26. Janjigian YY, Al-Batran SE, Wainberg ZA, Van Cutsem E, Molena D, Muro K, et al. LBA73 pathological complete response (pCR) to durvalumab plus 5-fluorouracil, leucovorin, oxaliplatin and docetaxel (FLOT) in resectable gastric and gastroesophageal junction cancer (GC/GEJC): interim results of the global, phase III MATTERHORN study. Ann Oncol 2023;34:S1315-6.
27. Shitara K, Rha SY, Wyrwicz LS, Oshima T, Karaseva N, Osipov M, et al. Neoadjuvant and adjuvant pembrolizumab plus chemotherapy in locally advanced gastric or gastro-oesophageal cancer (KEYNOTE-585): an interim analysis of the multicentre, double-blind, randomised phase 3 study. Lancet Oncol 2024;25:212-24.
28. Lorenzen S, Gotze TO, Thuss-Patience P, Biebl M, Homann N, Schenk M, et al. Perioperative atezolizumab plus fluorouracil, leucovorin, oxaliplatin, and docetaxel for resectable esophagogastric cancer: interim results from the randomized, multicenter, phase II/III DANTE/IKF-s633 trial. J Clin Oncol 2024;42:41020.
29. Andre T, Tougeron D, Piessen G, de la Fouchardiere C, Louvet C, Adenis A, et al. Neoadjuvant nivolumab plus ipilimumab and adjuvant nivolumab in localized deficient mismatch repair/microsatellite instability-high gastric or esophagogastric junction adenocarcinoma: the GERCOR NEONIPIGA phase II study. J Clin Oncol 2023;41:255-65.
30. Pietrantonio F, Raimondi A, Lonardi S, Murgioni S, Cardellino GG, Tamberi S, et al. INFINITY: a multicentre, single-arm, multi-cohort, phase II trial of tremelimumab and durvalumab as neoadjuvant treatment of patients with microsatellite instability-high (MSI) resectable gastric or gastroesophageal junction adenocarcinoma (GAC/GEJAC). J Clin Oncol 2023; 41:358.
31.Goodman KA Hall N, Bekaii-Saab TS, Ou F-S, Twohy E, Meyers MO et al. Survival outcomes from CALGB 80803 (Alliance): a randomized phase II trial of PET scan-directed combined modality therapy for esophageal cancer. J Clin Oncol 2018;36(Suppl_15):4012.
32. Hoeppner J, Brunner T, Lordick F, Schmoor C, Kulemann B, Neumann UP, et al. Prospective randomized multicenter phase III trial comparing perioperative chemotherapy (FLOT protocol) to neoadjuvant chemoradiation (CROSS protocol) in patients with adenocarcinoma of the esophagus (ESOPEC trial). J Clin Oncol 2024;42:LBA1.
33.Elliott JA, Klevebro F, Mantziari S, Markar SR, Goense L Johar A, et al. Neoadjuvant chemoradiotherapy versus chemotherapy for the treatment of locally advanced esophageal adenocarcinoma in the European multicenter ENSURE study. Ann Surg 2023;278:692-700.
34. Stahl M, Walz MK, Stuschke M, Lehmann N, Meyer HJ, Riera-Knorrenschild J, et al. Phase III comparison of preoperative chemotherapy compared with chemoradiotherapy in patients with locally advanced adenocarcinoma of the esophagogastric junction. J Clin Oncol 2009;27:851-6.
35. Sjoquist KM, Burmeister BH, Smithers BM, Zalcberg JR, Simes RJ, Barbour A, et al. Survival after neoadjuvant chemotherapy or chemoradiotherapy for resectable oesophageal carcinoma: an updated meta-analysis. Lancet Oncol 2011; 12:68192.