The Role of the TP53 Pathway in Predicting Response to Neoadjuvant Therapy in Esophageal Adenocarcinoma

Esophageal adenocarcinoma is an aggressive malignancy with poor prognosis owing to its overall resistance to existing therapies (1). According to the National Comprehensive Care Network guidelines, standard treatment for locally advanced EAC involving the lower esophagus and gastroesophageal junction comprises neoadjuvant chemoradiotherapy followed by radical esophagectomy (2, 3). The best foundational evidence for this trimodality approach comes from the CROSS (ChemoRadiotherapy for Oesophageal Cancer followed by Surgery Study) trial, conducted with five cycles of neoadjuvant carboplatin and paclitaxel concurrent with 40.4 Gy of radiotherapy (4). However, the more recent Cancer and Leukemia Group B (CALGB) 80803 and FLOT3 trials suggest greater efficacy using a combination that includes 5-fluorouracil, leucovorin, and oxaliplatin, especially when administered using a PET-directed approach during induction therapy (5, 6).

Several prospective and retrospective studies have established that response to neoadjuvant therapy, particularly with respect to presence or absence of residual nodal disease, is an independent predictor of overall and disease-free survival among patients with locally advanced esophageal carcinoma treated with trimodality therapy (7, 8). Our previous work demonstrates that overall survival and disease-free survival are optimized in patients with ≥90% treatment response in the primary tumor bed in the absence of residual pathologic nodal disease (9). A major pathologic response, as defined by this threshold, was seen in approximately 40% of patients. However, the majority of patients were classified as nonresponders with significant residual primary tumor and/or nodal disease after neoadjuvant therapy and 5-year survival likely similar to patients treated with surgery alone (10). At present, robust clinical or molecular predictors of response to neoadjuvant therapy are lacking, and, as a result, clinicians are unable to predict nonresponders upfront. Meanwhile, these patients are in urgent need of alternative therapeutic regimens to enhance the success of surgical resection and long-term outcomes.

Therefore, the current study aims to identify biomarkers of poor pathologic responses to neoadjuvant chemoradiotherapy so that genomic information from their tumors can be leveraged to develop novel therapeutic strategies. We believe that this approach represents, by far, the most significant opportunity to improve outcomes for patients with locally advanced esophageal carcinoma. By merging high-quality clinicopathologic and genomic data from a clinically validated broad-panel next-generation sequencing assay, we investigated genomic features, somatic alterations, and oncogenic pathways associated with treatment response among 237 patients with locally advanced lower esophageal and junctional adenocarcinoma treated with curative intent.

We used the Memorial Sloan Kettering (MSK) cBioPortal to identify all patients with adenocarcinoma of the lower esophagus or gastroesophageal junction treated with multimodality therapy with curative intent at our institution from 2014 through 2020 who also underwent prospective sequencing by MSK-IMPACT (MSK-Integrated Mutation Profiling of Actionable Cancer Targets; refs. 11, 12). Clinical annotations were then obtained via cross-referencing our manually curated, prospectively maintained institutional database. All patients provided written informed consent for targeted sequencing under protocol NCT01775072, approved by the MSK Institutional Review Board. Tumor tissue for sequencing was obtained from either primary or metastatic sites at the time of biopsy or surgery. Tumor purity was assessed by histopathologic review of specimens by an expert pathologist from the MSK Molecular Diagnostics Service. For patients who had more than one sample sequenced, the sample with the higher tumor purity was selected for inclusion.

The MSK-IMPACT next-generation sequencing assay, which is currently FDA authorized, was performed as part of routine clinical assessment in a Clinical Laboratory Improvement Amendments–compliant laboratory, as described previously (13). Briefly, genomic DNA was extracted from tumor tissue and patient-matched blood samples to generate barcoded libraries. After capture of exons and selected introns of the genes included in the sequencing panel, pooled libraries were sequenced on the Illumina HiSeq 2500 system. Samples were sequenced using MSK-IMPACT v1 (341 genes), MSK-IMPACT v2 (410 genes), or MSK-IMPACT v3 (468 genes).

Sequencing files were processed using stringent quality-control criteria and analyzed using an optimized informatics pipeline to identify somatic mutations, copy-number alterations, and select structural rearrangements. Full details regarding the performance and validation of the MSK-IMPACT assay have been reported (14). Utilizing the OncoKB database, we excluded variants of unknown significance (15, 16). Copy-number alterations were identified by comparing targeted regions of the tumor sample with the matched diploid normal sample. The log ratio coverage values for segments were calculated and compared between the tumor and normal samples. A fold change threshold of <−2 and FDR-corrected P value < 0.05 was used to determine whole gene loss, or deep deletion/homozygous deletion, while a fold change threshold of >2 was used to determine whole gene amplification. Alterations (oncogenic mutations, copy-number alterations, structural rearrangements, or fusions) were considered for analysis only if present in 8% of patients in the cohort after eliminating microsatellite instability (MSI)-high cases. Number of oncogenic drivers was calculated for each patient as the total number of driver alterations present.

MSI status was assessed using the MSI-sensor algorithm, which calculates the percentage of microsatellite loci covered by the MSK-IMPACT assay that are unstable in the tumor as compared with the patient's matched normal sample (17). Samples with a score ≥10 were classified as MSI-high. To calculate tumor mutation burden, we determined the total number of somatic nonsilent protein-coding mutations in the sequenced genes and normalized it to the exonic coverage of the respective MSK-IMPACT panel in megabases. Tumor mutation burden calculations using this panel are strongly associated with those assessed by whole-exome sequencing (18). The fraction of genome altered was defined as the fraction of log₂ copy-number variation (gain or loss) >0.2, divided by the size of the genome whose copy number was profiled. Fraction of genome altered was corrected for tumor purity, ploidy, and clonal heterogeneity using the FACETS algorithm (19). Presence or absence of whole-genome doubling was estimated using the probability model as described previously (20). Mutual exclusivity or co-occurrence of genomic alterations was analyzed using a Fisher exact test (21). Large-scale transition scores were inferred on the basis of copy-number data (22).

Biallelic inactivation of TP53 was determined to have occurred if no copies of the wild-type (WT) allele were present after accounting for the total copy number of TP53, the number of mutant copies present, and whether loss of heterozygosity occurred as assessed by the FACETS algorithm. The number of copies of the mutant allele was determined as described previously (23).

We evaluated 11 canonical cancer-related signaling pathways as defined by The Cancer Genome Atlas (TCGA) PanCancer Atlas Project (24). The pathways analyzed were p53, cell cycle, Hippo, Myc, Notch, NRF2, PI3K, receptor tyrosine kinase (RTK)/RAS/MAPK, TGFβ, Wnt, and DNA damage response (DDR). A tumor was considered altered in a specific pathway if at least one gene belonging to that pathway was altered, and pathways were considered for analysis if at least 8% of patients harbored an alteration in that pathway. Number of pathways altered was calculated for each patient as the total number of altered pathways out of the 11 pathways specified above.

All surgical specimens were reviewed by a board-certified expert gastrointestinal pathologist at our center. In posttreatment specimens, if grossly viable tumor (mass, ulcer, or polyp) was present, five or more representative sections of the tumor were evaluated; if grossly viable tumor was absent, the scar-like lesions at the primary tumor site were submitted in their entirety for histopathologic evaluation. Pathologic response in the primary tumor bed was quantified as the percentage treatment effect (TE) in terms of residual viable carcinoma in relation to areas of fibrosis or fibroinflammation within the gross lesion, as well as by use of College of American Pathologists tumor regression grade (CAP-TRG) guidelines (25). A CAP-TRG score of 0 corresponds to complete response or 100% TE, a score of 1 corresponds to near-complete response or 90%–99% TE, a score of 2 corresponds to partial response or 50%–89% TE, and a score of 3 corresponds to poor or no response or <50% TE. Pathologic complete response (pCR) was defined as 100% TE (i.e., CAP-TRG score 0) and pN0. Both clinical and pathologic staging were performed in accordance with the 8th Edition of the American Joint Committee on Cancer Staging Manual. Patients were considered to be pathologically downstaged if their pathologic stage grouping was lower than their clinical stage grouping (i.e., from clinical stage 3 to pathologic stage 2).

Clinicopathologic characteristics were summarized using frequency and percentage for categorical variables and median and interquartile range (IQR) for continuous variables. Genomic and/or pathway alterations were counted as either present or absent. Association of genomic driver alterations, pathways, features, and clinicopathologic factors with treatment response was evaluated using the Wilcoxon rank-sum test for continuous variables or a Fisher exact test for categorical variables. Stage-adjusted ORs were calculated using a multivariable logistic regression model for clinicopathologic variables in conjunction with clinical stage. P values <0.05 were considered statistically significant; however, FDRs (q-values) using the Benjamini–Hochberg procedure were reported for analysis of genomic factors to account for multiple hypothesis testing. Multivariable logistic regression models were then utilized to evaluate the association of treatment response with genomic variables that were found to be significant on univariable analysis at a threshold of q < 0.1, controlling for significant clinicopathologic (P < 0.05) variables. For analyses of long-term outcomes stratified by a genomic alteration of interest, progression-free survival (PFS) was estimated using the Kaplan–Meier method and compared using the log-rank test. PFS was measured from the date of completion of neoadjuvant therapy to the date of disease progression, death, or last follow-up. Similarly, the cumulative incidence of progression (1 minus Kaplan–Meier estimate) measured all instances of disease progression, recurrence, or death after completion of neoadjuvant therapy. All statistical analyses were performed using R (version 3.6.3; R Foundation for Statistical Computing).

All study data are freely available on cBioPortal: https://www.cbioportal.org/study/summary?id=egc_msk_tp53_ccr_2022.

In total, 237 patients treated with curative intent were included in our study. Patient characteristics are summarized in Supplementary Table S1. The majority of patients (82%) were male; median age was 61 years (IQR, 55–68). Clinical staging was based on PET/CT and either endoscopic ultrasound or endoscopic mucosal resection as indicated, and most patients (186; 78%) were diagnosed with clinical stage III disease. The primary tumor was sequenced in 192 patients (81%), while pretreatment tissue adequate for sequencing was available in 110 patients (46%).

Neoadjuvant chemotherapy regimens included carboplatin and paclitaxel (Carbo-Taxol) in 115 patients (49%), and either 5-fluorouracil, leucovorin, and oxaliplatin (FOLFOX) or 5-fluoruracil, leucovorin, oxaliplatin, docetaxel (FLOT) in 77 patients (32%). Of note, 151 patients (64%) were treated with induction chemotherapy using a PET-directed strategy, as described previously (26), and as a result 38 patients (16%) received both Carbo-Taxol and FOLFOX due to a postinduction switch in chemotherapy regimen. A total of 213 patients (90%) received radiotherapy concurrent with neoadjuvant chemotherapy. In addition to chemotherapy, 10 patients received neoadjuvant trastuzumab based on clinical Her2 status in the context of either a clinical trial (4 patients) or limited metastatic disease (6 patients) followed by durable response on imaging to warrant surgical resection. Similarly, 33 patients received neoadjuvant immune checkpoint blockade in either the context of a clinical trial (30 patients) or limited metastatic disease (3 patients). In total, 10 patients with clinical stage IVB or limited metastatic disease were included in this study; these patients were treated with curative intent based on multidisciplinary tumor board discussion.

Following neoadjuvant therapy, 210 patients (88%) underwent surgical resection via either esophagectomy and/or gastrectomy. Of the remaining patients, 14 were found at surgical exploration to have unresectable disease, and 13 had progression of disease on restaging PET that precluded surgery altogether. The rate of pCR in this cohort was only 12%. The median treatment effect in the primary tumor bed was 80% (IQR 30–98), and more than half of patients (126; 53%) had evidence of residual nodal disease. Among the patients who underwent surgical resection, 12 (6%) had positive margins.

The primary outcome of interest in this study was response to neoadjuvant therapy. Because of previously observed differences in both clinical and genomic behavior (27), we eliminated MSI-high patients (n = 14; 6%) from analyses of treatment response and focused only on microsatellite stable patients with chromosomal instability characteristics. Patients were classified as either neoadjuvant therapy responders (nT-R) or nonresponders (nT-NR) based on our definition of major pathologic response (9). Specifically, nT-Rs were patients who had met both of the following criteria: (i) absence of residual nodal disease (pN0), (ii) treatment effect ≥ 90% (TRG score 0 or 1). On the other hand, nT-NRs were patients with residual nodal disease, those with treatment effect <90% (TRG score >1), or those with progression of disease either on restaging PET or surgical exploration. Figure 1A provides a flow chart of how patients were classified as either nT-R (n = 64) or nT-NR (n = 159) after elimination of MSI-high patients from the cohort.

Patients classified as nT-R had significantly longer PFS than those classified as nT-NR (Fig. 1B). Furthermore, our definition of nT-R heavily overlapped with other clinically relevant metrics of treatment response, such as pathologic downstaging and metabolic response on PET [defined as ≥35% decrease in standard uptake value (SUV) measured in the primary tumor between pretreatment and posttreatment PET scans] (Fig. 1C).

First, we evaluated relevant clinicopathologic factors for their association with pathologic treatment response by comparing the nT-R and nT-NR groups (Table 1). Because higher clinical stage was associated with lack of treatment response, we report clinical stage-adjusted ORs for the remainder of variables. In addition to clinical stage, both tumor grade and neoadjuvant chemotherapy regimen were significantly associated with treatment response. In particular, high-grade or poorly differentiated tumors were associated with worse treatment response [OR, 0.52; 95% confidence interval (CI), 0.28–0.94; P = 0.03]. In comparison with Carbo-Taxol, FOLFOX was associated with a significantly better treatment response (OR, 2.38; 95% CI, 1.20–4.76; P = 0.013). Notably, metabolic response on PET was not significantly associated with pathologic treatment response in our cohort.

Targeted sequence analysis of our study group identified eight genes harboring recurrent oncogenic driver alterations (cross-validated using OncoKB and TCGA databases) at ≥8% prevalence: TP53 (79%), CDKN2A (22%), SMAD4 (10%), ARID1A (14%), ERBB2 (16%), KRAS (19%), MDM2 (9%), and CCNE1 (8%; Fig. 2A). In particular, ERBB2, KRAS, MDM2, and CCNE1 were predominantly affected by oncogenic amplifications. Similarly, eight oncogenic pathways were identified with a ≥8% prevalence of alteration in our cohort: TP53, cell cycle, RTK-RAS, WNT, TGFβ, PI3K, Notch, and DDR. We then assessed whether any genomic alterations, pathways, or features were significantly associated with response to neoadjuvant therapy by comparing the nT-R and nT-NR groups (Supplementary Table S2). Of note, neither tumor purity, location of sample sequenced, nor whether any treatment was received prior to sampling varied significantly by treatment response group.

Among individual genomic alterations, only MDM2 amplification was significantly associated with the nT-NR group (P = 0.01, q = 0.09). We then constructed a multivariable logistic regression model including both MDM2 and significant clinicopathologic variables: clinical stage, tumor grade, and neoadjuvant chemotherapy regimen. All variables except for clinical stage retained statistical significance, and thus MDM2 amplification was found to be independently associated with poor treatment response in our cohort (OR, 0.10; 95% CI, 0.01–0.55; P = 0.032).

Among oncogenic pathways, alterations in the TP53 pathway were enriched in the nT-NR group, but this association did not meet the threshold for statistical significance after multiple hypothesis testing (P = 0.05, q = 0.27; Supplementary Table S2). Interestingly, no other genomic alterations, pathways, or features such as tumor mutation burden, whole-genome doubling, fraction of genome altered, or large-scale transition scores were associated with response to neoadjuvant therapy in our cohort.

Given the results above with respect to the TP53 pathway, we next investigated whether the specific type of TP53 mutation was associated with treatment response, or whether MDM2 amplification was the lone driver of TP53 pathway–level effects. We hypothesized that biallelic truncating mutations, which lead to abolished TP53 function, may be associated with worse treatment response and outcome in comparison with other types of TP53 mutations, such as monoallelic truncating mutations or missense or splice mutations, which often lead to incomplete TP53 dysfunction. As shown in Fig. 3A, TP53 and MDM2 were, by far, the most frequently altered members of the TP53 pathway altered in this dataset, while ATM and CHEK2 mutations were rare. Therefore, we instead subdivided the TP53 pathway into four categories reflecting the severity of alteration in terms of TP53 dysfunction: (i) MDM2 amplification, (ii) TP53 truncating biallelic mutation (TP53 TruncBi), (iii) other TP53 pathway mutation (TP53 other), and (iv) TP53 pathway WT. On univariable analysis, this modified TP53 pathway was significantly associated with treatment response (P = 0.004, q = 0.07). Furthermore, on multivariable analysis, MDM2 amplification was associated with worse treatment response in comparison with WT (OR, 0.06; 95% CI, 0.00–0.39; P = 0.01), while TP53 TruncBi mutation trended similarly (OR, 0.31; 95% CI, 0.08–1.03; P = 0.06; Fig. 3B). Tumor grade and neoadjuvant chemotherapy regimen were also significantly associated with treatment response in this model, while clinical stage was not, as TP53 status was also correlated with nodal status.

Mutual exclusivity between TP53 mutation and MDM2 amplification has been reported (28), and in our dataset TP53 TruncBi mutations and MDM2 amplification were almost entirely mutually exclusive (q < 0.05; Fig. 3B). Moreover, tumors with MDM2 amplification and those with TP53 TruncBi mutations displayed remarkably similar clinical behavior and outcomes with respect to treatment responses (Table 2), cumulative incidence of progression, and PFS. Compared with the WT group, both MDM2 amplification and TP53 TruncBi groups had a significantly lower pathologic treatment effect seen in the primary tumor (Fig. 3D), and consequently a significantly higher cumulative incidence of progression and shorter PFS (Fig. 3E). In particular, the 24-month cumulative incidence of progression was 61% (95% CI, 28–79) among patients with MDM2 amplification and 67% (95% CI, 44–80) among patients with a TP53 TruncBi mutation. By comparison, TP53 pathway WT patients had a 24-month cumulative incidence of progression of 42% (95% CI, 19–58). Overall, the median length of follow-up among survivors was 20.9 months (range, 1.6–98.4) and the median PFS was 15.9 months (95% CI, 13.8–20.9).

Among the 21 patients with MDM2-amplified tumors, 1 patient showed disease progression on imaging and was no longer a candidate for surgery, 3 patients underwent surgical exploration and were found to be unresectable, and 1 patient was resected with a positive proximal margin despite the fact that the cervical esophagus was divided as high as possible to still allow for reconstruction with a gastric pullup (Table 2). Eight patients underwent a switch in chemotherapy postinduction due to lack of response on postinduction PET CT, and only 3 of these ultimately went on to have an improvement in PET response, though all but one underwent surgical resection with negative margins. Most patients with MDM2-amplified tumors (16/21; 76%) were noted to have either clinical or pathologic nodal disease. Five of the patients with MDM2 amplification had a PFS greater than 20 months, and 4 of these 5 had concurrent TP53 mutations. While 8 patients were pathologically downstaged, only a single patient with an MDM2-amplified tumor was classified in the nT-R group. This patient had a 95% treatment effect (TRG score 1) in the absence of residual nodal disease and remains free of recurrence at 9 months of follow-up. Notably, this patient presented with a clinical stage III junctional tumor and was treated with neoadjuvant FLOT chemotherapy alone followed by complete resection.

Among the 37 patients with TP53 TruncBi-mutant tumors, 4 had disease progression on posttreatment imaging, and 6 underwent surgical exploration and were deemed unresectable. Furthermore, while only 15 TruncBi patients (41%) were clinical node positive, 25 patients (68%) had pathologic evidence of residual nodal disease.

Prospectively applying broad-panel clinical next-generation sequencing to patients with esophageal adenocarcinoma treated with curative intent, we discovered that MDM2 amplification is independently associated with response to neoadjuvant therapy, irrespective of important clinicopathologic factors including chemotherapy regimen, clinical stage, or tumor grade. Moreover, patients with MDM2 amplifications or with truncating mutations that lead to biallelic TP53 loss appear to face similar outcomes: poor response to neoadjuvant chemotherapy, and consequently, a higher cumulative incidence of progression and shorter PFS in comparison with patients with TP53 pathway WT tumors.

Esophageal adenocarcinoma is widely known to be resistant to systemic therapy. This trait is attributed primarily to chromosomal instability with consequent increased intratumoral heterogeneity and rapid tumor evolution (29, 30). Furthermore, loss of TP53 tumor suppressor function is known to be a primary driver event in esophageal adenocarcinoma tumorigenesis. To our knowledge, this is the first study to identify individual driver alterations within the TP53 pathway that are strongly associated with poor treatment response and outcome in esophageal adenocarcinoma. More specifically, a greater degree of TP53 loss and dysfunction in esophageal adenocarcinoma appears to correspond to worse treatment response and shorter PFS. A total of 57 patients, or 26%, in our cohort harbored either an MDM2 amplification or TP53 TruncBi mutation, accounting for a substantial proportion of non-responders. Conversely, TP53 pathway WT patients consistently had the best treatment responses and long-term outcomes in our cohort.

Interestingly, we did not detect any relationship between genomic features known to be correlated with higher levels of chromosomal instability, such as whole-genome doubling or fraction of copy-number genome altered, and treatment response (Supplementary Fig. S1A). However, rates of whole-genome doubling were higher in tumors with MDM2 amplification and significantly higher in tumors with TP53 TruncBi mutation in comparison to those with TP53 pathway WT status (Supplementary Fig. S1B). Intratumoral heterogeneity has been a proposed driver of treatment resistance, but unfortunately this characteristic is difficult to measure using panel-based sequencing. Because of this limitation of our methodology, we cannot draw any definitive conclusions regarding the contribution of intratumoral heterogeneity to treatment resistance. Therefore, the mechanism by which either MDM2 amplification or TP53 TruncBi mutation confers treatment resistance remains unclear. Their strong mutual exclusivity, but closely related clinical behavior, suggests functional overlap in that both alterations lead to critical, if not absolute, TP53 dysfunction.

While this is the first study to elucidate the role of MDM2 amplification in predicting treatment response in esophageal adenocarcinoma, MDM2 amplification has been previously shown to be associated with poor response to both chemotherapy and radiation in other cancer types. Kondo and colleagues were among the first to recognize this and suggested that MDM2 overexpression inhibits cisplatin-induced apoptosis in human glioblastoma cell lines via TP53 downregulation (31). Similar effects have been seen with other chemotherapeutic agents as well, such as doxorubicin and 5-fluorouracil (32). MDM2 is known to tightly regulate TP53 via a negative feedback loop that involves TP53 directly binding MDM2, followed by ubiquitination and degradation of TP53. These effects of MDM2 are likely to be more potent on WT TP53 than on mutant TP53 (33). This idea may, in part, explain why the few patients with concurrent MDM2 amplification and TP53 mutation had marginally better long-term outcomes. Though the mechanism of MDM2-related treatment resistance in esophageal adenocarcinoma is likely to be dependent on this TP53 negative feedback loop, additional studies are needed to test this hypothesis.

Reduced or abolished TP53 function has been linked with resistance to a variety of chemotherapeutic and targeted agents (alkylating agents, anthracyclines, antimetabolites, antiestrogens, and tyrosine kinase inhibitors). Therefore, discovery and testing of small molecules that can restore WT TP53 conformation and function represent important areas of future investigation in efforts to overcome treatment resistance (34). While a few agents, that is, PRIMA-1 and related thiosemicarbazones, have demonstrated early promise in preclinical studies, their efficacy in tumors with truncating TP53 mutations may be diminished, given that their mechanism of restoring TP53 function is by correcting misfolded TP53 proteins. Therefore, we anticipate that TP53 truncating biallelic mutations may be challenging to target.

On the other hand, several MDM2 inhibitors, predominantly derivatives of Nutlins, are currently under either preclinical or clinical investigation. In the setting of WT or functional TP53, MDM2 antagonists lead to reactivation of TP53, which suggests that if they are used in combination with other chemotherapeutic agents, treatment response may be greatly enhanced. For example, in preclinical studies, idasanutlin administered in combination with chemotherapy exhibited potent anticancer activity against xenograft models of fibrosarcoma and acute myeloid leukemia expressing functional TP53 (35). Furthermore, early clinical trials in patients with acute myeloid leukemia demonstrated reasonable safety and tolerability of idasanutlin combined with cytotoxic chemotherapy (36). Therefore, MDM2 inhibitors in combination with chemotherapy may represent an important therapeutic breakthrough in the approximately 9% of patients with esophageal adenocarcinoma harboring MDM2 amplifications, most of whom currently exhibit poor responses to systemic therapy.

Consistent with initial results of the CALGB 80803 trial (5), we also observed better responses to neoadjuvant FOLFOX in comparison with Carbo-Taxol, which was used in the CROSS trial. We also did not discern any alterations differentiating responders vs. non-responders to either regimen, though interestingly, a higher proportion of patients with MDM2-amplified and TP53 TruncBi-mutated tumors were treated with Carbo-Taxol in comparison with FOLFOX. Moreover, as previously shown (37), metabolic response on PET did not correlate well with response to neoadjuvant therapy. The pCR rate in this study was low at 12%, which is likely related to bias in patient selection, for example, from the availability or purity of tissue for sequencing. Patients in our cohort tended to have more advanced clinical stages of disease and poor response to treatment, and these patients also tend to have increased availability of adequate tumor tissue for next-generation sequencing. Conversely, inclusion of patients with pCR depended on adequate pretreatment endoscopic biopsy tissue; if this was unavailable, their tumors could not be sequenced.

In conclusion, MDM2 amplification was found to be independently associated with poor response to neoadjuvant therapy in patients with esophageal adenocarcinoma treated with curative intent. Given the dearth of actionable targets in esophageal adenocarcinoma, MDM2 inhibition, in combination with cytotoxic chemotherapy, may represent an important therapeutic strategy to overcome treatment resistance and improve outcomes in these patients. Furthermore, TP53 status was strongly associated with response to neoadjuvant therapy in esophageal adenocarcinoma, with worsening TP53 dysfunction corresponding to worse outcomes.

M. Hsu reports grants from NIH/NCI Cancer Center Support Grant P30CA008748 during the conduct of the study. Y.Y. Janjigian reports other support from AstraZeneca, Basilea Pharmaceutica, Bayer, Bristol Myers Squibb, Daiichi Sankyo, Eli Lilly, Imugene, Merck, Merck Serono, Michael J. Hennessy Associates, Paradigm Medical Communications, Pfizer, Rgenix, Seagen, Zymeworks Inc., Cycle for Survival, Department of Defense, Fred's Team, Genetech/Roche, and NCI outside the submitted work. S.B. Maron reports personal fees from Natera, Bicara, Daiichi Sankyo, Novartis, and Basilea; non-financial support from Guardant Health, Genentech, and Bayer outside the submitted work; and prior equity in Calithera. A. Wu reports grants from CivaTech Oncology, personal fees from Nanovi A/S, and other support from Simphotek, Inc. outside the submitted work. D.R. Jones reports other support from AstraZeneca and Merck outside the submitted work. D.B. Solit reports personal fees from Pfizer, Loxo/Lilly Oncology, Scorpion Therapeutics, Vividion Therapeutics, Fore Therapeutics, and BridgeBio outside the submitted work. M.F. Berger reports personal fees from Eli Lilly and PetDX outside the submitted work. D. Molena reports other support from Johnson & Johnson, Boston Scientific, UroGen, and AstraZeneca outside the submitted work. No disclosures were reported by the other authors.

S. Sihag: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S.C. Nussenzweig: Data curation, formal analysis, methodology, project administration. H.S. Walch: Data curation, software, formal analysis, visualization, methodology. M. Hsu: Software, formal analysis, validation, investigation, methodology. K.S. Tan: Software, formal analysis, validation, investigation, methodology. S. De La Torre: Data curation, project administration. Y.Y. Janjigian: Investigation, writing–review and editing. S.B. Maron: Investigation, writing–review and editing. G.Y. Ku: Investigation, writing–review and editing. L.H. Tang: Formal analysis, visualization. P.M. Shah: Investigation, writing–review and editing. A. Wu: Investigation, writing–review and editing. D.R. Jones: Resources, supervision, writing–review and editing. D.B. Solit: Resources, supervision, writing–review and editing. N. Schultz: Resources, software, supervision, validation, investigation, writing–review and editing. K. Ganesh: Conceptualization, validation, investigation. M.F. Berger: Resources, software, supervision, validation. D. Molena: Conceptualization, resources, formal analysis, supervision, validation, project administration, writing–review and editing.

This work was supported, in part, by the NIH/NCI Cancer Center Support Grant P30 CA008748. S. Sihag is supported by a grant from the Fiona and Stanley Druckenmiller Center for Lung Cancer Research at Memorial Sloan Kettering Cancer Center, an Association of Women Surgeons Ethicon Research Fellowship, and the David J. Sugarbaker Research Scholarship from the American Association for Thoracic Surgery.

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.