Genomic Analyses and Precision Oncology in Gastroesophageal Cancer: Forwards or Backwards?

Cancers of the stomach and esophagus, collectively termed gastroesophageal adenocarcinomas (GEA), are leading causes of global cancer mortality. GEA is recognized as a highly heterogenous disease among individuals, and strategies for subclassifying GEAs into more homogenous subtypes may improve patient outcomes by allowing subtype-specific treatment regimens, a strategy termed “precision oncology.” Motivated by this concept, previous studies have attempted to classify GEAs according to their molecular profiles, including The Cancer Genome Atlas and the Asian Cancer Research Group. However, none of these genomic classifications to date have been incorporated into standard treatment algorithms, and biomarkers currently used to guide clinical decisions in metastatic GEA include only HER2 for trastuzumab and microsatellite instability (MSI) or PD-L1 status for pembrolizumab. Other agents, such as ramucirumab, which targets the VEGF pathway, and nivolumab (anti–PD-1; in Asia), do not require biomarker selection. Currently, biomarker testing in GEA has typically involved testing a single biopsy or a few biopsies from the primary tumor (PT). However, recent studies have suggested that significant heterogeneity may exist between PTs and their metastatic lesions, and even within the same PT, which may affect clinical responses. As such, it is critical to explore two questions: First, is genomic characterization of a single PT biopsy sufficient to capture the majority of targetable alterations in a patient with metastatic GEA, and, if not, are there alternative approaches? Second, will treating patients with GEA according to their pretreatment biomarker profiles result in tangible improvement in clinical outcome, or should other factors be considered, such as co-occurring alterations?

In this issue of Cancer Discovery, two studies are reported addressing these questions (1, 2). In the first study, Pectasides and colleagues applied various different genomic platforms to distinct patient cohorts to quantify levels of mutational concordance between primary GEA specimens and patient-matched metastatic lesions, including levels of intratumor heterogeneity in the same PT (1). For some cohorts, circulating cell-free DNA (cfDNA) analysis was also performed to capture the mutational profiles of PTs and metastases. In the second study, Janjigian and colleagues conducted a single-center cohort study of patients with metastatic GEA, whose pretreatment biopsy specimens were analyzed by targeted sequencing (∼340 genes) and correlated with biomarker-defined treatments and clinical outcomes (2). The two studies are highly complementary, and in some cases conceptually similar analyses were performed, providing a platform for contrast and comparison. Taken collectively, these two studies reaffirm the notion that GEAs are highly heterogeneous and that tackling this heterogeneity will be essential for the success of GEA precision oncology.

To define the extent of GEA heterogeneity in metastatic patients, Pectasides and colleagues (1) analyzed paired PTs and metastatic lesions (including lymph-node metastasis) from multiple patient cohorts. The results generally suggest a strikingly high level of mutational discordance between PTs and metastatic sites in GEA, from approximately 40% for single-nucleotide variants and insertion–deletion elements (indel), to approximately 60% for amplified genes such as HER2, CDK4/6, EGFR, and KRAS. Importantly, discordant profiles were observed even in patients who had not received prior systemic therapy, suggesting that high GEA heterogeneity is likely a natural feature of this malignancy, rather than a consequence of clonal selection arising from therapeutic pressure. Interestingly, a conceptually similar assessment of GEA heterogeneity performed by Janjigian and colleagues reported a “high” level of genomic concordance between pre– and post–trastuzumab-treated biopsies, although numerical values were not explicitly provided (2). This may be due to differences in technical and study design, such as the use of different sequencing panels (340 cancer genes in Janjigian and colleagues vs. whole-exome sequencing in Pectasides and colleagues), or differences in the methods used to calculate concordance levels or when comparing synchronous versus metachronous lesions.

Recent studies have shown that circulating tumor cfDNA can harbor genetic abnormalities reflective of both the original PT and also metastatic lesions, released in the bloodstream through cancer cell necrosis or apoptosis (3). Comparing cfDNAs, PTs, and metastatic lesions from the same patients, Pectasides and colleagues observed concordances but also discordances in clinically relevant genomic aberrations (e.g., HER2, FGFR2). Demonstrating that cfDNA can identify new targetable alterations absent in solid-tissue biopsies, patients with GEA were identified exhibiting CDK6 amplifications in cfDNAs not observed in PTs. Reciprocally, however, in other cases, alterations such as FGFR2 and HER2 were observed to be present in PTs but not matched cfDNAs, suggesting that not every alteration in a patient with GEA can be detected by cfDNA. These results suggest that cfDNAs likely provide a “summarized” snapshot of the majority tumor profile, rather than an exhaustive comprehensive catalog. The observation of better concordance between cfDNA and metastasis profiles may have therapeutic implications, as in a preliminary clinical trial cohort (PANGEA) in Pectasides and colleagues, and knowledge of metastatic profiles led to a reassignment in treatment selection in 32% of patients, with anecdotal clinical responses.

Besides tumor heterogeneity, another important question is the extent to which outcomes of patients with GEA are indeed improved by treating patients according to their underlying tumor molecular profiles. In this regard, Janjigian and colleagues found that approximately 50% of patients with GEA exhibit a potentially targetable alteration as defined by the OncoKB database, and they explored GEA responses to three broad therapeutic classes: chemotherapy, targeted therapies (trastuzumab), and immune checkpoint inhibition. In the Janjigian and colleagues cohort, patients with MSI-high (MSI-H) GEAs progressed significantly more rapidly on cytotoxic therapy compared with non–MSI-H GEAs, a finding that has also been independently observed (4). However, the team was unable to identify specific genetic abnormalities predictive of GEA responses to first-line platinum/fluoropyrimidine combination therapy. Specifically, in other tumor types, homologous recombination defects (HRD) have been proposed to predict responses to platinum therapy, and in an analysis of over 10,000 cancer genomes across 36 tumor types, about 10% of GEAs were identified with possible HRD based on the presence of a mutational signature (Signature 3; ref. 5). However, Janjigian and colleagues were unable to correlate biomarkers of HRD, including mutations in DNA repair genes, with treatment response, putting into question the contribution of HRD in GEA therapeutic response. It should be noted, however, that Janjigian and colleagues used large-scale transitions (LST) to detect HRD status, which as a single modality may not fully capture HRD levels compared with combination scores integrating LST with other measurements (telomeric allelic imbalance and loss of heterozygosity), or using methods such as HRDetect (6, 7).

Janjigian and colleagues also found strong correlations in HER2 copy-number gains inferred by next-generation sequencing (NGS) or clinical methods such as IHC or FISH. Reaffirming the role of HER2 copy-number levels as a critical predictor of therapeutic sensitivity, the few discordant cases (8%) labeled HER2-negative by NGS had significantly poorer outcomes on trastuzumab. Importantly, Janjigian and colleagues demonstrated that besides HER2 levels, co-occurring genomic alterations, particularly in the MAPK or PI3K pathway, also modulated primary or acquired resistance to anti-Her2 therapy. Such findings thus highlight the importance of measuring multiple and not just single biomarkers in one individual. Clonal selection with loss of HER2 amplification or secondary alterations in the HER2 gene were also identified as mechanisms of acquired resistance; drawing knowledge from Pectasides and colleagues, it remains possible that studies of cfDNA in these patients may identify markers of resistance to HER2 therapy earlier in the course of treatment, allowing for timely switches in treatment modalities.

Although nearly a third of the patients (28%) treated with immune checkpoint inhibition in the study by Janjigian and colleagues responded, only 13% (n = 5) had a prolonged response. This is disappointing considering that patients in this cohort were preferentially directed to this therapy based on NGS results or MSI status, and is similar to large phase III studies in heavily treated gastric cancer, without biomarker selection (8). Notably, 2 patients (one EBV-positive) with low tumor mutational burden and non–MSI-H had long and durable responses to immunotherapy, whereas not all MSI-H tumors responded. Tumor mutational burden appeared to be a good predictor of response to immunotherapy, with patients with tumors exhibiting more than approximately 10 mutations/Mb having superior outcomes. In one anti–PD-1 responder, a new B2M mutation was detected in the posttreatment sample. B2M alterations have been described as resistance mechanisms to immunotherapy in melanoma (9), suggesting that mechanisms associated with immunotherapy resistance may be conserved across different tumor types.

In summary, both teams should be congratulated for remarkable studies that deepen our understanding of GEA. Pectasides and colleagues' study has revealed the humbling truth of roadblocks faced by GEA precision oncology in the face of vast genomic tumoral heterogeneity (Fig. 1), whereas the Janjigian and colleagues study gives us hope that with a deeper understanding of the disease and the addition of immunotherapy and other novel therapeutic agents, steady gains will be eked out in the battle against this dreadful disease. Finally, the importance of genomic drivers notwithstanding, future work should also undoubtedly focus on understanding tumor heterogeneity at other levels, such as those related to epigenetics, the tumor microenvironment, and the tumor microbiome, that collectively interact to drive gastroesophageal malignancy (10).

No potential conflicts of interest were disclosed.

R. Sundar is supported by a National Medical Research Council (NMRC) MOH Healthcare Research Scholarship, Singapore. P. Tan is supported by Duke-NUS Medical School and the Biomedical Research Council, Agency for Science, Technology and Research.