Harnessing biomarkers of response to improve therapy selection in esophago-gastric adenocarcinoma

Abstract

Advanced esophago-gastric (OG) adenocarcinomas have a high mortality rate and new therapeutic options are urgently required. Despite recent advances in understanding the molecular characteristics of OG cancers, tumor heterogeneity poses a challenge in developing new therapeutics capable of improving patient outcomes. Consequently, chemotherapy remains the mainstay of systemic treatment, with the HER2 being the only predictive biomarker routinely targeted in clinical practice. Recent data indicate that immunotherapy will be incorporated into first-line chemotherapy, but further research is required to refine patient selection. This review will summarize the clinical strategies being evaluated to utilize our knowledge of predictive biomarkers with reference to novel therapeutics, and discuss the barriers to implementing precision oncology in OG adenocarcinoma.

Esophago-gastric (OG) adenocarcinomas represent a significant global health issue. Gastric and esophageal cancers are the sixth and ninth most common malignancies worldwide, yet these represent the second and fifth leading causes of cancer-related deaths respectively [1]. These divergences between incidence and mortality rates reflect the inherent aggressive biology of these highly heterogenous cancers, compounded by the lack of progress in the development of novel therapeutics relative to other malignancies.

Cytotoxic chemotherapy therefore remains the cornerstone of systemic therapy. In the curative setting, chemotherapy is a component of multimodal management in conjunction with surgery or radiotherapy. The majority of patients with OG adenocarcinoma present with de novo advanced disease. Within this paradigm, platinum and fluoropyrimidine-based combinations have historically been the global standard of care in the first-line setting for HER2-negative disease and are associated with a median overall survival (mOS) of 11 months [2]. Patients with HER2-positive disease benefit from the addition the monoclonal antibody trastuzumab with mOS extended to 14–16 months [3]. In addition to trastuzumab, the only targeted therapy currently licensed for use in advanced OG adenocarcinoma is ramucirumab, a VEGF inhibitor, in combination with paclitaxel in the second-line setting [4].

Beyond HER2, there has been a paucity of predictive biomarkers in OG adenocarcinomas successfully translated into clinically meaningful improvements in patient outcomes until recent studies have indicated the clinical value of PD-L1 expression using the combined positive score (CPS) as a predictor of response to immune checkpoint inhibitors (ICIs). Given the relatively static prognosis in this tumor type, particularly in advanced disease, the elucidation of biomarkers of response is an important strategy to consider. This review will outline the evidence relevant to predictive biomarkers currently under clinical evaluation within the field with a discussion of how these can better refine our use of both existing and emerging therapeutics to improve patient outcomes.

HER2

HER2, also known as ERBB2, is a member of the HER receptor family associated with tumor cell proliferation, apoptosis, adhesion, migration and differentiation. HER2 amplification and subsequent overexpression is found in a variety of malignancies including breast, colorectal, gastric, lung and bladder cancers [5]. It is currently an established therapeutic target in breast [6] and gastric cancers [5].

HER2 is overexpressed and/or amplified in approximately 20% of gastric cancers [7] and can be quantified by immunohistochemistry (IHC) or in situ hybridization (ISH). The ToGA study led to the establishment of trastuzumab in combination with a fluoropyrimidine/platinum chemotherapy backbone as the gold standard for first-line systemic therapy in patients with HER2-positive disease (Table 1) [3]. In this trial, HER2 positivity was defined as IHC 3+ (defined as moderate to strong, complete or basolateral membrane staining in >10% of tumor cells) or FISH amplification HER2/CEP17≥2 in locally advanced, recurrent or metastatic gastric or gastro-esophageal (GOJ) adenocarcinoma. HER2 positivity is more common in GOJ tumors (32.2%) than gastric cancers (21.4%) and in intestinal type tumors (31.8%) than diffuse type tumors (6.1%) [7].

Table 1. . Summary of HER2-targeted clinical trials in advanced esophago-gastric adenocarcinoma.

Trial	Phase	HER2 definition	Treatment arms	N	Primary end point	Results	Ref.
First-line therapy
ToGA	III	IHC 3+ and/or ISH-positive	Capecitabine or 5-FU, cisplatin +/- trastuzumab	594	OS	mOS: 13.8 vs 11.1 months (HR 0.74, 95% CI: 0.60–0.91, p = 0.0046)	[3]
TRIO-013/LOGiC	III	IHC3+ and/or ISH-positive	Capecitabine, oxaliplatin +/- lapatinib	545	OS	mOS: 12.2 vs 10.5 months (HR 0.91, 95% CI: 0.73–1.12, p = 0.32)	[8]
JACOB	III	IHC 3+ or IHC 2+ ISH-positive	Capecitabine or 5-FU, cisplatin, trastuzumab +/- pertuzumab	780	OS	mOS: 17.5 vs 14.2 months (HR 0.84, 95% CI: 0.71–1.00, p = 0.057)	[9]
HELOISE	IIIb	IHC 3+ or IHC 2+ ISH-positive	Cisplatin, capecitabine, trastuzumab 8 mg/kg loading dose + 6 mg/kg or 10 mg/kg maintenance dose	248	OS	mOS: 12.5 vs 10.6 months (HR 1.24, 95% CI: 0.86–1.78, p = 0.2401)	[10]
NCT02954536	II	IHC3+ and/or ISH-positive	Capecitabine, oxaliplatin or cisplatin, trastuzumab + pembrolizumab	37	≥26/37 patients progression-free at 6 months	26/37 patients (70%) progression-free at 6 months (95% CI: 54–83)	[11]
Second-line therapy
TyTan	III	ISH-positive	Paclitaxel +/- lapatinib	261	OS	mOS: 11.0 vs 8.9 months (HR 0.84, 95% CI: 0.64–1.11, p = 0.10)	[12]
GATSBY	II/III	IHC 3+ or IHC 2+ ISH-positive	Trastuzumab emtansine vs taxane	302	OS	mOS: 7.9 vs 8.6 months (HR 1.15, 95% CI: 0.87–1.51, p = 0.86)	[13]
T-ACT	II	IHC 3+ or IHC 2+ ISH-positive	Paclitaxel +/- trastuzumab	91	PFS	mPFS: 3.2 vs 3.7 months (HR 0.91, 80% CI 0.67–1.22, p = 0.33)	[14]
NCT02689284	II	IHC3+ and/or ISH-positive	Margetuximab + pembrolizumab	92	ORR	ORR: 19%	[15]
Third-line therapy
DESTINY-Gastric01	II	IHC 3+ or IHC 2+ ISH-positive	Trastuzumab deruxtecan vs physician's choice chemotherapy (irinotecan or paclitaxel)	187	ORR	ORR: 42.8% (95% CI: 33.8–52.3) vs 12.3% (95% CI: 5.2–24.1)	[16]
		Cohort 1: IHC 2+/ISH-negative Cohort 2: IHC 1+	Trastuzumab deruxtecan	Cohort 1: 20 Cohort 2: 24	ORR	Cohort 1 ORR: 26.3% Cohort 2 ORR: 9.5%	[17]

HR: Hazard ratio; IHC: Immunohistochemistry; ISH: In situ hybridization; ORR: Overall response rate; OS: Overall survival; PFS: Progression-free survival.

To date, further strategies to capitalize on HER2-positivity in OG adenocarcinoma produced results with limited clinical benefit. Unsuccessful approaches include large Phase III trials investigating dual HER2 blockade with trastuzumab and pertuzumab [9] and trastuzumab dose escalation [10] in the first-line setting (Table 1). The addition of tyrosine kinase inhibitors to a chemotherapy backbone [12] and the use of trastuzumab emtansine [13] as second-line therapy following initial treatment with trastuzumab have also undergone clinical trial evaluation and reported negative results (Table 1).

These findings have highlighted the biological distinctions in HER2 overexpression seen between OG and breast cancers, a tumor type where these treatment strategies have been effective and have widened the therapeutic options available for patients with HER2-positive breast cancer [6]. The membranous distribution of HER2 within breast cancer cells is predominantly circumferential. Conversely, HER2 protein expression OG cancer tends to spare the digestive luminal membrane, resulting in basolateral or lateral membrane staining [18]. This discrepancy has led to separate scoring systems for each tumor type. Furthermore, a greater degree of intratumoral HER2 heterogeneity has been observed in OG specimens, with the presence of variable focal areas of HER2 positivity seen [19]. HER2 expression can also change following trastuzumab administration with one series demonstrating loss of HER2 expression in 60.6% (n = 20/33) in repeat biopsies within 3 months of completing first-line therapy with trastuzumab, with the rate of HER2 3+ positivity via IHC reduced from 72.7 to 39.4% after treatment [20]. Dynamic HER2-expression has also been shown in an exploratory analysis of the Phase II T-ACT study, which randomized 91 patients with known HER2-positive OG adenocarcinoma determined prior to first-line therapy to receive second-line paclitaxel +/- trastuzumab following progression on platinum and fluoropyrimidine-containing chemotherapy with trastuzumab (Table 1) [14]. Eleven of 16 tumor samples acquired following disease progression on first-line chemotherapy were HER2-negative on reassessment. As such, direct extrapolation of the treatment approaches in breast cancer have not been successful in OG cancer and highlight the need for further clinical investigation using a more nuanced approach tailored to this tumor type.

In addition to these unique features associated with HER2-positivity in OG adenocarcinoma, challenges with targeting HER2-positive OG adenocarcinoma also arise from primary resistance mechanisms, such as EGFR/MET/KRAS/Pi3K mutations [21], and secondary resistance mechanisms that invariably develop with any targeted therapy [22]. Despite this, several therapeutic strategies for HER2-positive OG cancers are in development and will be described next.

Trastuzumab deruxtecan (T-DXd, DS-8201a)

Trastuzumab deruxtecan (T-DXd, DS-8201a) is an HER2-targeting antibody–drug conjugate consisting of an antibody component, MAAL-9001, covalently conjugated to a drug component, MAAA-1181a through an enzymatically cleavable tetrapeptide linker [23]. The antibody component is a recombinant humanized anti-human HER2 immunoglobulin monoclonal antibody, whereas MAAA-1181a is a topoisomerase-1 inhibitor permeable to the cell membrane and more potent than SN38, the active metabolite of irinotecan. Trastuzumab deruxtecan has a drug-to-antibody ratio (DAR) of 8, twice that of trastuzumab emtansine. This design allows for efficient delivery of a potent, cytotoxic payload to targeted cells while limiting off-target toxicity in normal cells. The payload is also membrane-permeable, allowing it to target neighboring tumor cells regardless of HER2-expression. Collectively, the high DAR and bystander effect achieved as part of the design of trastuzumab deruxtecan may overcome the challenges posed by HER2 heterogeneity seen in OG cancers [24]. Dose expansion Phase I evaluationconsisting of 40 patients reported a manageable safety profile consisting of anemia (30%) and decreases in neutrophil (20%), platelet (18%), and white blood cell (16%) counts. Pneumonitis was a noted treatment-emergent side effect in four patients (grade 2 in 3 cases, grade 1 in one patient). The confirmed ORR in HER2-overexpressing gastric cancer subjects in the dose expansion cohort was 43.2% (19/44 patients, 95% CI: 28.3–59.0) whereas the median duration of response (DoR) was 7 months (95% CI: 4.4–16.6). Median progression-free survival (PFS) was 5.6 months (95% CI: 3.0–8.3) [25]. DESTINY-Gastric01, a randomized Phase II evaluation in the third-line setting in an Asian population has recently reported an overall response rate (ORR) of 51% (n = 61/119) in patients who received trastuzumab deruxtecan, compared with 14% (n = 8/56) in patients who received physician's choice chemotherapy (irinotecan or paclitaxel) [26]. Median OS and median PFS were also significantly longer in the trastuzumab deruxtecan group (12.5 vs 8.4 months; HR: 0.59 [95% CI: 0.39–0.88] and 5.6 vs 3.5 months, HR: 0.47 [95% CI: 0.31–0.71]), indicating clinically meaningful results for patients with heavily pre-treated HER2-positive OG cancers. Although the mOS of 12.5 months was the longest ever to be reported in a randomized control trial in the third-line setting, these results will require further validation of efficacy in a larger, global patient population. Trastuzumab deruxtecan is also currently undergoing further Phase II evaluation as monotherapy following disease progression on trastuzumab (DESTINY-Gastric02, NCT04014075).

The bystander effect associated with trastuzumab deruxtecan has also posed potential clinical utility of this drug in HER2-low disease. The dose escalation component within the Phase I study of trastuzumab deruxtecan in heavily pre-treated gastric and breast cancers demonstrated anti-tumor activity even in low HER2-expressing tumors [27]. Further evaluation at Phase II level in exploratory cohorts of the DESTINY-Gastric01 study comprising of patients with HER2 IHC 2+/ISH- and IHC 1+ disease in the chemorefractory setting reported an ORR of 26.3% (n = 5/19) and 9.5% (n = 2/21) in these respective subgroups, providing preliminary evidence of clinical activity in HER2-low patients [17].

Preclinical data using trastuzumab deruxtecan monotherapy or in combination with an anti-PD-1 antibody in an immunocompetent murine model has also demonstrated anti-tumor activity and extended overall survival in comparison to monotherapy [28]. Additionally, major histocompatibility complex (MHC) class I expression on tumor cells and dendritic cell activation markers were upregulated with activation of adaptive immune response following treatment with trastuzumab deruxtecan. These results provide preliminary data to base further investigation into trastuzumab deruxtecan in combination with immune checkpoint inhibitors in HER2-overexpressing gastric cancers. DESTINY-Gastric03 (NCT04379596) is a Phase Ib/II trial currently in set-up and will aim to evaluate the safety, tolerability, immunogenicity and anti-tumor activity of trastuzumab deruxtecan as both as monotherapy and in combination with chemotherapy and/or durvalumab, in treatment-naive patients during the dose expansion phase.

HER2-directed immunotherapy

The use of trastuzumab in HER2-positive OG cancers may be exploited by concurrent use of immune checkpoint inhibition. In addition to exerting its anticancer effect by blocking multiple pro-survival cellular signalling pathways, trastuzumab also mediates antibody-dependent cellular cytotoxicity through activation of antibody-binding Fc receptors [29]. This has been shown to facilitate the uptake of tumor antigens by dendritic cells, which in turn stimulate the activation and expansion of tumor specific CD4 and CD8 T cells [30]. As HER2 antibody therapy boosts T-cell priming, combination strategies with other approaches to enhance T-cell activation are attractive [31], and murine models have suggested that immunotherapy could potentiate the immune-mediated cytotoxic effects of anti-HER2 targeting [30]. A Phase II study assessing the safety and activity of platinum, capecitabine and trastuzumab in combination with pembrolizumab, an anti-PD1 therapy, met is primary end point with 70% (n = 26/37) reported as progression free at 6 months [32]. The investigators reported an ORR of 91% (95% CI: 78–97) and median PFS of 13.0 months (95% CI: 8.6–not reached) in conjunction with a tolerable safety profile, providing a basis for Phase III investigation (KEYNOTE-811, NCT03615326).

Margetuximab is a monoclonal antibody derived from trastuzumab with increased affinity for lower and higher affinity forms of the Fc receptor CD16A, leading to enhanced tumor cell-directed antibody-directed cellular cytotoxicity (ADCC) against HER2-expressing and HER2 low-expressing cancer cells. In a Phase I study with patients with advanced HER2-positive solid tumors, 12% (n = 7/60) and 50% (n = 30/60) of patients achieved partial responses and stable disease respectively [33]. Seventy percent (70%) of these patients had documented disease progression on previous HER2-targeted therapies. Ex vivo analyses of patients' peripheral mononuclear cells showed that margetuximab augmented ADCC in comparison to trastuzumab. At Phase Ib/II level, margetuximab in combination with pembrolizumab in 92 patients with HER2-positive gastric or GOJ cancers previously treated with trastuzumab demonstrated tolerability and preliminary anti-tumor activity with an ORR of 19% and disease control rate (DCR) of 54%. Higher HER2 expression and PD-L1 positivity were identified as biomarkers of response to combination therapy [34].

ZW25

A bispecific antibody is engineered to recognize and bind two different antigens simultaneously. ZW25 is a bispecific antibody that binds to extracellular domain 4, the domain containing the trastuzumab binding site, and extracellular domain 2, the domain containing the pertuzumab binding site on HER2. Preclinical evaluation has shown high levels of anti-tumor activity across as a range of HER2 expression and induces more effective HER2 signalling inhibition than trastuzumab and pertuzumab in combination [35]. Single-agent activity in 33 patients with a variety of HER2-positive cancers recruited into a Phase I study, including 11 patients with OG cancer, showed good tolerance in a cohort of heavily pre-treated patients [36]. Additionally, an ORR of 43% and DCR of 56% was reported in the OG cohort. ZW25 is currently being investigated in a multicenter, Phase II trial in combination with physician's choice of doublet chemotherapy (CX, CF or mFOLFOX6) in the first-line setting for patients with HER2-positive advanced OG adenocarcinoma (NCT03929666).

FGFR amplifications

The FGFR family is composed of four highly conserved receptor tyrosine kinase receptors (FGFR1, FGFR2, FGFR3 and FGFR4). Binding of FGFs leads to receptor dimerisation, activation of the kinase domain and transphosphorylation of the intracellular domain. Downstream signalling routes utilizes the RAS, mitogen-activated protein kinase (MAPK) and phosphoinositide3-kinase-protein kinase (PI3K) pathways. In addition to driving cell proliferation, survival and migration, FGFR signalling can be context-dependent and promote cell differentiation [37]. Mutations in the FGFR pathway include gene amplifications, activating mutations and chromosomal translocations cause subsequent dysregulation of the FGFR signalling pathway, thus being recognised as an oncogenic driver in a variety of solid tumors [38].

FGFR2 amplifications have been described in 2–9% of gastric cancers [39–41], particularly in the diffuse subtype [42] and has been shown to be associated with higher tumor stage, the presence of lymph node metastases and poorer cancer-specific survival [43]. These features have made it a rational target for therapeutic development.

AZD4547 is an oral tyrosine kinase inhibitor selective for FGFR1, 2 and 3. Preclinical investigation has demonstrated that AZD4547 reduces cell proliferation and induces apoptosis in FGFR2 amplified gastric cancer cell lines and induced tumor regression in in vivo models [44]. The randomized Phase II trial, SHINE, evaluated the efficacy of AZD4547 in the second-line setting in patients with advanced gastric and GOJ adenocarcinoma with FGFR polysomy or gene amplification in comparison to single-agent paclitaxel [45]. This study failed to show a significant PFS difference in patients randomized to AZD4547 compared with paclitaxel (1.8 vs 3.5 months, p = 0.95) in the full analysis set. No significant differences in PFS were detected in the polysomy or amplified subgroups. Results from exploratory biomarker analyses reported significant intra-tumor heterogeneity for FGFR2 amplification in addition to poor concordance between FGFR2 amplification or polysomy with FGFR2 expression indicated the need for an alternative approach to predictive biomarker testing in these patients.

A basket Phase II study investigating AZD4547 in FGFR dysregulated tumors reported an ORR of 33% and mean DoR of 5.7 months in the gastric and GOJ adenocarcinoma cohort consisting of 138 patients with biopsy-proven FGFR2 amplification [46]. Data from an accompanying translational research study utilizing cell lines and patient-derived xenograft (PDX) models showed that only high-level FGFR2 amplification, defined as a ratio of FGFR2 gene to chromosome 10 centromer signals of >5, initiates an oncogene addiction phenotype characterized by FGFR2-mediated transactivation of alternative receptor kinases and bringing Pi3K/mammalian target of rapamycin (mTOR) signalling under FGFR control [47]. Although this suggests that screening for high-level FGFR2 amplification in circulating tumor DNA (ctDNA) may identify potential responders, its relatively low prevalence of 5% in gastric cancer (n = 7/135) [47] limits widespread clinical utility.

Other FGFR inhibitors undergoing evaluation in OG cancers include futibatinib (TAS-120), a highly selective covalent FGFR inhibitor was shown to be well-tolerated in a first-in-human study consisting of 36 subjects, which included six patients with gastric and esophageal cancers [48]. Preliminary results reported clinical responses in two patients, including one with FGFR2-amplified gastric cancer and in a patient with esophageal cancer. Bemarituzumab (FPA144) is a humanized monoclonal antibody (IgG1 isotype) specific to the human FGFR2b receptor. Upon binding to the FGFR2b receptor, bemarituzumab inhibits FGF signalling and receptor downregularion, internalization and degradation as well as enhancing ADCC. A Phase I study showed that bemarituzumab monotherapy had no dose-limiting toxicities with an ORR of 22% and DCR of 55.6% in patients with chemorefractory FGFR2b-overexpressing OG cancers [49]. Having demonstrated no dose-limiting toxicities when combined with FOLFOX6 [50], a randomized Phase III trial (FIGHT, NCT03343301) assessing bemarituzumab with FOLFOX6 as first-line treatment in patients with FGFR2b overexpressed and/or FGFR2 amplified advanced gastric and GOJ adenocarcinoma is currently underway [51]. Notably, patients will be deemed eligible for this study based on determination of IHC-derived FGFR2b overexpression determined or centrally performed ctDNA assay to detect FGFR2 gene amplification. This approach may provide further insight into the clinical capability of ctDNA as a diagnostic test in these patients. A third compound, derazantinib is a pan-FGFR inhibitor shown to have anti-tumor activity in gastric cancer murine models [52] which also inhibits the colony-stimulating factor-1 receptor kinase (CSF1R) [53], inducing tumor macrophage depletion [54] and enhancing responsiveness to immune checkpoint inhibition [55]. Derazantinib is due to be investigated in a forthcoming Phase Ib/II muti-cohort study as monotherapy and in combination with paclitaxel and ramucirumab or atezolizumab [56].

Claudin 18.2

Tight junctions form the apical junctional complex in epithelial and endothelial cellular sheets and are essential for the sealing of cellular sheets forming a luminary barrier and controlling the paracellular ion flux. A hallmark of cancer is that tight junction proteins lose their organisation in mutimeric structures, promoting loss of cell polarity, cohesion and differentiation [57]. Malignant transformation can therefore lead to exposure of tight junction molecules normally shielded, and therefore inaccessible to drug targeting, in the epithelia.

Claudin 18.2 (CLDN18.2) is an antigen expressed by differentiated gastric mucosa cells in the pit and base regions of gastric glands. Importantly, its expression is specific to normal gastric epithelial cells, and is retained upon malignant transformation in up to 70% of primary gastric adenocarcinomas [58]. Its relocation to the surface on malignant transformation makes it a putative therapeutic target. Based on The Cancer Genome Atlas (TCGA) data, fusion products between CLDN18 and ARHGAP were found in the genomically stable subgroup, which in turn were more frequent in the invasive phenotype of diffuse gastric cancer [59].

Zolbetuximab (IMAB362) is a first-in-class IgG1 antibody targeted against CLDN18.2 [60]. Preclinical studies have shown that zolbetuximab mediates ADCC against gastric CLDN18.2-positive gastric cancer cell lines and improved anti-tumor activity was observed in xenografted mice treated with zolbetuximab and chemotherapy as opposed to chemotherapy alone [61]. As monotherapy in chemorefractory advanced gastric adenocarcinoma with moderate-to-strong CLDN18.2 expression in at least 50% of tumor cells, zolbetuximab demonstrated an ORR of 9% (n = 4/43) and stable disease in an additional 15% (n = 6/43) of patients [62]. The most common toxicities were nausea, vomiting and fatigue. The FAST study is a Phase II randomized trial that assessed EOX chemotherapy (epirubicin, oxaliplatin and capecitabine) compared with combination therapy with zolbetuximab in advanced gastric, GOJ and esophageal adenocarcinoma expressing CLDN18.2 (≥2+ staining intensity on IHC using the anti-CLDN18 43–14A monoclonal antibody in ≥40% of tumor cells) [63]. Median PFS and OS in the combination therapy arm were significantly longer when compared with the chemotherapy arm (PFS: 7.5 vs 5.3 months, HR: 0.44, 95% CI: 0.29–0.67; p < 0.0005 and OS: 13.0 vs 8.4 months, HR: 0.56, 95% CI: 0.40–0.79; p = 0.0008) with a tolerable safety profile. Increased efficacy was seen in high expressors, defined as ≥2+ staining in ≥70% of tumor cells. Based on these promising results, two Phase III global, placebo-controlled studies evaluating zolbetuximab in patients with advanced esophago-gastric adenocarcinoma with moderate-to-strong membranous staining of CLDN18.2 in ≥75% tumor cells with mFOLFOX6 (SPOTLIGHT; NCT03504397) and CAPOX (GLOW; NCT03653507) [64] will provide further insight into the efficacy of zolbetuximab in first-line treatment.

MET

c-MET, a tyrosine kinase receptor, is activated when it is bound to its ligand HGF to form a complex involved in RAS-MAPK and PI3K-AKT signalling pathways that mediate cell migration, invasion and angiogenesis. Dysregulation in the MET/HGF pathway has been shown to be associated with a more aggressive phenotype and a poor prognosis, and MET activation stimulates tumor invasiveness. These features have made MET targeting a potential therapeutic target.

Rilotumumab is a fully humanized IgG2 monoclonal antibody against HGF which reported a longer PFS in patients randomized to receive rilotumumab in combination with ECX as first-line treatment independent of MET status in advanced OG patients [65]. In a pre-planned sub-group analysis of MET-positive tumors, defined as tumors with ≥25% membrane staining of ≥1+ on immunohistochemistry, patients who received rilotumumab had significantly longer PFS (HR 0.46, 95% CI: 0.25 - 0.85; p-0.013) and OS (HR 0.46, 95% CI: 0.24–0.87; p = 0.016) [65]. These promising results led to two Phase III trials assessing rilotumumab with ECX [66] and cisplatin and capecitabine [67] as first-line therapy in MET-positive patients. However, both trials were stopped early as due to a higher number of deaths being observed in patients receiving rilotumumab than placebo, with subgroup analysis failing to identify subsets of patients who may derive clinical benefit [66]. A second monoclonal antibody, onartuzumab, also failed to demonstrate a survival benefit in combination with modified FOLFOX6 in patients with treatment-naive MET-positive advanced OG cancer [68].

Novel compounds are currently being evaluated in MET-amplified OG cancer at preclinical and in early phase trials. For example, tepotinib has shown dose-dependent growth inhibition in c-MET gastric cancer cell lines, suggesting a therapeutic effects in MET-amplified OG cancer [69]. An on-going Phase I trial of GST-HG161, a potent and highly selective c-MET inhibitor has reported a tolerable safety profile to date in a cohort which includes gastric patients [70]. Nonetheless, these agents are at early stages of development and significant progress is required before MET amplification can be therapeutically relevant in the clinic.

DNA damage repair

Genomic instability is a hallmark of cancer which facilitates the acquisition of genetic events that ultimately precipitates oncogenic transformation. Genomic integrity is maintained by a network of DNA damage repair (DDR) pathways. Five major pathways exist: mismatch repair (MMR), base excision repair and nucleotide excision repair and direct repair for single-strand DNA breaks and homologous recombination and non-homologous end joining for double-strand breaks. Cancer cells often have DDR defects in one or more of these pathways, causing increased levels of replication stress and increased levels of endogenous DNA damage. As such, exploitation of these vulnerabilities pose an attractive therapeutic opportunity and OG cancers harbour certain molecular features that warrant the assessment of DDR-targeting agents in this tumor type.

One such molecular biomarker is ataxia-telangiectasia mutated (ATM) defects. ATM is a key activator of the DDR to double-strand breaks via homologous recombination. Between 13 and 22% of tumors from patients with advanced gastric cancer have low or undetectable expression of ATM (ATM_low) using immunohistochemistry [71]. Poly(ADP-ribose) polymerase (PARP) inhibitors block base excision repair and leads to cancer cell death through synthetic lethality in tumors with pre-existing homologous recombination repair deficiencies. A randomized Phase II study assessing olaparib 100 mg twice daily or placebo in conjunction with weekly paclitaxel in the second-line treatment of advanced OG cancer patients did not improve PFS, but reported that combination therapy significantly improved median OS in both the overall, biomarker-unselected population (13.1 vs 8.3 months, HR: 0.56, 80% CI 0.41–0.75; p = 0.005) and in the ATM_low population (mOS not reached vs 8.2 months, HR: 0.35, 80% CI 0.22–0.56; p = 0.002) [72]. Despite these positive results, the subsequent Phase III study GOLD failed to demonstrate a significant improvement in the overall cohort and ATM_low subgroup [73], although there was a numerical trend toward improved survival in patients randomized into the olaparib arm. The cause of these negative results are likely multifactorial [74]; for example, 51% of the Study 39 population were ATM_low due to prospective population enrichment, whereas ATM_low was represented by only 18% of the GOLD population. This difference may have led to inadvertent inflation of the survival outcomes seen in Study 39 which failed to be replicated at Phase III level. Lack of biomarker standardisation and heterogenous ATM expression may have also contributed to these discrepant results [74]. Nevertheless, these findings may indicate that, ATM loss in isolation may not be entirely predictive of PARP inhibitor sensitivity in OG cancers.

Another putative biomarker of response includes mutational signatures indicative of underlying homologous recombinant deficiencies. For example, in an analysis of 4,938,362 mutations from 7402 cancers, Alexandrov et al. described ‘signature 3’, a specific base-substitution signature characterized with substantial numbers of deletions with overlapping microhomology at breakpoint junctions, a pattern attributed to the use of error prone DDR pathways in place of compromised homologous recombination [75]. Although the majority of cases were observed in patients with germline BRCA mutant breast, ovarian and pancreatic cancers, signature 3 was also observed in a subset of non-BRCA mutant patients. When applied to 10,250 cancer genomes encompassing 36 tumor types, signature 3 was singularly identified in gastric cancers, suggesting that up to 12% of gastric cancers may benefit from PARP inhibition or platinum therapy [76]. Similarly, whole-genome sequencing of 129 cases of esophageal adenocarcinoma also derived a ‘DDR-impaired’ subgroup (18%) characterized by prevalent defects in homologous recombination and chromosome segregation pathways [77]. Dual exposure of a patient-derived gastric cancer cell line with a DDR-impaired profile to olaparib and the topoisomerase-1 inhibitor topotecan demonstrated in vitro sensitivity to this combination. Mutational signatures, therefore, offer a potential method of identifying patients who would benefit from DDR targeting agents and clinical investigation of this hypothesis has begun [78].

Predictive biomarkers to immune checkpoint blockade

Immune checkpoint blockade has been assessed as monotherapy and combination therapy in various settings in OG adenocarcinoma (Table 2). In contrast to the successes seen in tumor types such as non-small-cell lung cancer and melanoma, immunotherapy has yet to demonstrate significant clinical benefit in a biomarker unselected population in OG adenocarcinoma. The exception to this statement is the use of nivolumab in Asian patients with chemorefractory disease, based on the results of the ATTRACTION-02 trial which reported prolongation of median OS from 4.1 to 5.3 months (HR 0.63, 95% CI: 0.51–0.78; p < 0.0001) in patients treated with nivolumab and has led to regulatory approval of nivolumab in Japan, Korea and Taiwan for this indication. Although immunotherapy continues to be assessed in OG adenocarcinoma, it is clear that only a subset of patients respond to this treatment and further work is required to characterise these patients who will benefit from the sustained clinical benefit associated with immune checkpoint blockade.

Table 2. . Selected clinical trial data assessing immune checkpoint blockade and associated predictive biomarkers in advanced esophago-gastric adenocarcinoma.

Trial	Phase	Treatment arms	N	Primary end point	Biomarker	Results	Ref.
First-line therapy
CheckMate-649	III	Pembrolizumab + chemotherapy† vs chemotherapy†	1581	OS and PFS in CPS ≥5 population	PD-L1 (DAKO 28-8 pharmDx assay; CPS)	PD-L1 CPS ≥5 mOS: 14.4 vs 11.1 months (HR 0.71; 98.4% CI 0.59–0.86; p < 0.0001) mPFS: 7.7 vs 6.0 months (HR 0.68; 98% CI 0.56–0.81; p < 0.0001) PD-L1 CPS ≥1 mOS: 14.0 vs 11.3 months (HR 0.77; 99.3% CI 0.64–0.92; p = 0.0001) mPFS: 7.5 vs 6.9 months (HR 0.74; 95% CI: 0.65–0.85) All randomized mOS: 13.8 vs 11.6 months (HR 0.80; 99.3% CI 0.68–0.94; p = 0.0002) mPFS: 7.7 vs 6.9 months (HR 0.77; 95% CI: 0.68–0.87)	[79]
ATTRACTION-04	III	Nivolumab + chemotherapy‡ vs chemotherapy‡ + placebo	724	OS and PFS	PD-L1 (Dako 28-8 pharmDx assay; staining in ≥1% tumor cells)	mOS: 17.45 vs 17.15 months (HR 0.90; 95% CI: 0.75–1.08; p = 0.257) mPFS: 10.45 vs 8.34 months (HR 0.68; 98.51% CI 0.51–0.90; p = 0.0007)	[80]
KEYNOTE-590§	III	Pembrolizumab + chemotherapy¶ vs chemotherapy¶ + placebo	749	OS and PFS	PD-L1 (CPS; 22C3 pharmDx assay)	ESCC PD-L1 CPS ≥10 mOS: 13.9 vs 8.8 months (HR 0.57; 95% CI: 0.43–0.75; p < 0.0001) ESCC mOS: 12.6 vs 9.8 months (HR 0.72; 95% CI: 0.60 - 0.88; p = 0.0006) mPFS: 6.3 vs 5.8 months (HR 0.65; 95% CI: 0.54–0.78; p < 0.0001) PD-L1 CPS ≥10 mOS: 13.5 vs 9.4 months (HR 0.62; 95% CI: 0.49–0.78, p < 0.0001) mPFS: 7.5 vs 5.5 months (HR 0.51; 95% CI: 0.41–0.65; p < 0.0001) All patients mOS: 12.4 vs 9.8 months (HR 0.73; 95% CI: 0.62–0.86; p < 0.0001) mPFS: 6.3 vs 5.8 months (HR 0.65; 95% CI: 0.55–0.76; p < 0.0001)	[81]
KEYNOTE-062	III	Pembrolizumab vs chemotherapy# + pembrolizumab vs chemotherapy# + placebo	763	OS and PFS in CPS ≥1 population	PD-L1 (CPS; 22C3 pharmDx assay)	Pembrolizumab vs chemotherapy + placebo   • CPS ≥1     ○ mOS: 10.6 vs 11.1 months (HR 0.91; 95% CI: 0.74–1.10; p = 0.162)     ○ mPFS: 2.0 vs 6.4 months (HR 1.66; 95% CI: 1.37–2.01)   • CPS ≥10     ○ mOS: 17.4 vs 10.8 months (HR 0.69; 95% CI: 0.49–0.97)     ○ mPFS: 2.9 vs 6.1 months (HR 1.10; 95% CI: 0.79–1.51) Pembrolizumab + chemotherapy vs chemotherapy   • CPS ≥1     ○ mOS: 12.5 vs 11.1 months (HR 0.85; 95% CI: 0.70–1.03; p = 0.046)     ○ mPFS: 6.9 vs 6.4 months (HR 0.84; 95% CI: 0.70–1.02; p = 0.039)   • CPS ≥10     ○ mOS: 12.3 vs 10.8 months (HR 0.85; 95% CI: 0.62–1.17; p = 0.158)     ○ mPFS: 5.7 vs 6.1 months (HR 0.73; 95% CI: 0.53–1.00)	[82]
NCT02915432	Ib/II	Toripalimab vs CAPOX	18	ORR	PD-L1 (SP142, staining in ≥1% of tumor cells or any intensity in tumor-infiltrating immune cells)	Overall ORR: 66.7% (1 CR and 11 PR) PD-L1 results available for 18 patients 16.7% PD-L1 +ve ORR in both PD-L1 +ve and -ve patients 66.7%.	[83]
Maintenance therapy
JAVELIN Gastric 100	III	Avelumab vs continuation of chemotherapy or BSC	499	OS	PD-L1 (73–10 pharmDx Dako assay; staining in ≥1% of tumor cells) Exploratory analysis in PD-L1 +ve population using 22C3 pharmDx assay, CPS ≥1	Overall mOS: 10.4 vs 10.9 months (HR 0.91; 95% CI: 0.74–1.11; p = 0.18) PD-L1 results based on tumor cell positivity available in 438 patients 12.3% PD-L1 +ve mOS: 16.2 vs 17.7 months (HR 1.13, 95% CI: 0.57–2.23l p = 0.64) PD-L1 results using CPS available for 213 patients 64.3% PD-L1 +ve mOS: 14.9 vs 11.6 months (HR 0.72, 95% CI: 0.49–1.05)	[84]
Second-line therapy
KEYNOTE-061	III	Pembrolizumab vs weekly paclitaxel	592	OS and PFS in CPS ≥1 population	PD-L1 (CPS; 22C3 pharmDx assay)	CPS ≥1 mOS: 9.1 vs 8.3 months (HR 0.82; 95% CI: 0.66–1.03; p = 0.04) mPFS: 1.5 vs 4.1 months (HR 1.27; 95% CI: 1.03-1.57) CPS <1 mOS: 4.8 vs 8.2 months (HR 1.2; 95% CI: 0.89–1.63) CPS ≥10 mOS: 10.4 vs 8.0 months (HR 0.64; 95% CI: 0.41–1.02)	[85]
KEYNOTE-181††	III	Pembrolizumab vs physician's choice of taxane or irinotecan	628	OS in ITT OS in SCC OS in PD-L1 CPS ≥10	PD-L1 (CPS; 22C3 pharmDx assay)	mOS in ITT: 7.1 months in both groups (HR 0.89; 95% CI: 0.75–1.05, p = 0.056) 401 patients (63.9%) had SCC histology. mOS in SCC: 8.2 vs 7.1 months (HR 0.78, 95% CI: 0.63- 0.96, p = 0.0095) PD-L1 CPS ≥10 in 222 patients mOS in PD-L1 CPS ≥10: 9.3 vs 6.7 months (HR 0.69, 95CI% 0.52–0.93, p = 0.0074)   • SCC: 10.3. vs 6.7 months   • ACC: 6.3 vs 6.9 months	[86]
NCT02340975	Ib/II	Durvalumab + tremelimumab (Arm A) vs durvalumab (Arm B) vs.tremelimumab (Arm C)	63	ORR	PD-L1 (Ventana SP263 assay; staining in ≥1% of tumor cells) TMB	Overall ORR: 7.4 vs 0 vs 8.3% PD-L1 status available in 107 patients (including Arm D) 80.4% PD-L1 +ve No association between PD-L1 status and OS/PFS Median TMB 2.13 mutations/Mb (including Arm D) No correlation with OS/PFS or ORR when comparing TMB-H vs TMB-L based on median, upper tertile, and upper quartile cutoff points	[87]
Second-line therapy and beyond
NCT02915432	Ib/II	Toripalimab	58	ORR	PD-L1 (SP142, staining in ≥1% of tumor cells or any intensity in tumor-infiltrating immune cells) TMB (TMB-H: ≥12 mutations/Mb) EBV (EBV copy number >100)	Overall ORR: 12.1% (7 PR) PD-L1 results available for 55 patients 85.5% PD-L1 +ve (n = 47/55) ORR in PD-L1 +ve vs PD-L1 -ve patients: 37.5% vs 8.5%, p = 0.023 TMB results available for 54 patients 22.2% TMB-H (n = 12/54_ ORR in TMB-H vs TMB-L: 33.3% vs 7.1%, p = 0.017 EBV results available for 55 patients 0.07% EBV +ve (n = 4/55) ORR in EBV +ve patients: 25%	[83]
Third-line therapy/chemo-refractory
ATTRACTION-2 (ONO-4538-12)	III	Nivolumab vs placebo	192	OS	PD-L1 (Dako 28-8 pharmDx assay; staining in ≥1% tumor cells)	Overall mOS: 5 · 26 vs.4.14 months (HR 0.63, 95% CI: 0.51–0.78; p < 0.0001) mOS in PD-L1 +ve patients: 5.2 vs 3.8 months (HR 0.51, 95% CI: 0.21–1·25) mOS in PD-L1-ve patients 6.0 vs 4.2 (HR 0.72, 95% CI: 0.49–1.05)	[88]
KEYNOTE-059 (Cohort 1)	II	Pembrolizumab	259	ORR	PD-L1 (CPS ≥1; 22C3 pharmDx assay) T-cell inflamed GEP (18-gene)	Overall ORR 11.6% (95% CI: 8.0–16.1). 42.1% (n = 148/259) deemed PD = L1 +ve ORR in PD-L1 +ve vs PDL1 -ve patients: 15.5% (95% CI: 10.1–22.4) vs 6.4% (95% CI, 2.6–12.8) A higher GEP score was associated with improved propensity for response and PFS (p = 0.010 and p = 0.002, respectively).	[89]
JAVELIN Gastric 300	III	Avelumab vs physician's choice of weekly paclitaxel or two-weekly irinotecan or BSC	371	OS	PD-L1 (staining in ≥1% tumor cells)	Overall OS 4.6 vs 5.0 months (HR 1.1 [95CI 0.9–1.4]; p = 0.81} No significant difference in OS in subgroup analysis based on PD-L1 status.	[90]
CheckMate-032	I/II	Nivolumab 3 mg/kg (NIVO3) vs nivolumab 1 mg/kg + ipilimumab 3 mg/kg (NIVO1 + IPI3) vs nivolumab 3 mg/kg + ipilimumab 1 mg/kg (NIVO3 + IPI1)	160	ORR	PD-L1 (DAKO 28-8 pharmDx assay; staining in ≥1% tumor cells) MSI	Overall ORR: 12 vs 24 vs 8% No significant differences in ORR seen based on PD-L1 status	[91]
KEYNOTE-012 (Cohort D, PD-L1 +ve patients only)	Ib	Pembrolizumab	39	ORR	PD-L1 (22C3 pharmDx, staining in ≥1% of tumor cells and mononuclear inflammatory cells)	Overall ORR: 22% (95% CI: 10–39)	[92]
NCT02340975	Ib/II	Durvalumab + tremelimumab (Arm D)	25	ORR	PD-L1 (Ventana SP263 assay; staining in ≥1% of tumor cells) TMB	Overall ORR: 4.0% Biomarker results combined with Arms A-C (see ‘Second-line setting’)	[87]
		Durvalumab + tremelimumab with prospective screening for tumor-based IFN-γ gene signature (Arm E)	19	ORR	IFNγ 4-gene signature	ORR: 15.8% (95% CI: 3.4–39.6)

^†Capecitabine + oxaliplatin or FOLFOX.

^‡S-1 or capecitabine + oxaliplatin.

^§Includes esophageal squamous cell carcinoma (SCC: 548 patients; adenocarcinoma: 201 patients).

^¶5-fluorouracil + cisplatin.

^#Cisplatin + 5-fluorouracil/capecitabine.

^††Includes esophageal squamous cell carcinoma (SCC: 401 patients; adenocarcinoma: 227 patients).

BSC: Best supportive care; CAPOX: Capecitabine + oxaliplatin; CPS: Combined positive score; GEP: Gene expression profile; HR: Hazard ratio; IFN-γ: Interferon gamma; Mb: Megabase; ORR: Overall response rate; OS: Overall survival; PFS: Progression-free survival; TMB: Tumor mutational burden.

Microsatellite instability

Microsatellite instability (MSI) is arguably the most robust predictive biomarker for immune checkpoint blockade available. Pembrolizumab was granted regulatory approval for tumor-agnostic use in patients with mismatch repair deficiency (MMRd) and microsatellite instability-high (MSI-H) disease [93]. In advanced OG adenocarcinoma specifically, subgroup analyses of clinical trial results in patients with MSI-H tumors have confirmed superior outcomes in this population. For example, 174/259 patients recruited in the Phase II KEYNOTE-059 trial had MSI status available, of which 4% were deemed MSI-H. The ORR in MSI-H patients treated with was 57% (95% CI: 18–90), whereas the ORR in the overall study population was 12% (95% 8–16) [89]. The Phase III KEYNOTE-061 study investigating the efficacy of pembrolizumab against paclitaxel in the second-line setting also reported longer OS in MSI-H patients (n = 27) who were randomized to receive pembrolizumab instead of paclitaxel (mOS: not reached vs 8.1 months) [85]. Nevertheless, the incidence of MSI-H or MMRd in advanced OG cancer is estimated between 4 and 9% [94]; thus, limiting its clinical utility to a minority of advanced OG patients.

To compound this, heterogenous distribution of MSI-H has also been observed in OG adenocarcinoma [95]. This was described in the context of patients recruited into a Phase II trial where 61 patients with metastatic chemorefractory gastric cancer were treated with pembrolizumab. Of the seven patients deemed to be MSI-H, the one patient who failed to respond to pembrolizumab had heterogenous immunohistochemical staining of human mutL homolog 1 (MLH1), whereas MLH1 IHC-positive and -negative regions correlated with MSS and MSI-H, respectively. The remaining six MSI-H patients with documented response had homogeneous MSI status. This finding was validated in a separate cohort of MSI-H gastric cancer patients, where eight of 79 patients (8.9%) also demonstrated MSI-H heterogeneity. This demonstrates that MSI/MMR status itself, while useful as a tumor-agnostic predictive biomarker of immune checkpoint inhibition, has nuances in its clinical utility.

PD-L1

The majority of clinical trials investigating immune checkpoint inhibitor therapy in advanced OG adenocarcinoma have concurrently investigated PD-L1 expression in relation to patient clinical outcomes. PD-L1 is expressed in 25–65% of patients with OG adenocarcinoma and is associated with tumor size, lymph node metastasis and shorter overall survival. A key limitation to the use of PD-L1 expression as a predictive biomarker in OG adenocarcinoma is a lack of a universal cut-off to define PD-L1 positivity based on a variety of diagnostic assays used (Table 2). Approaches based on PD-L1 expression on tumor cells exclusively have failed to demonstrate PD-L1 as a discriminating biomarker of response [87,88,90], a method that has been effectively applied in non-small-cell lung and urothelial cancers. CPS, which considers PD-L1 positivity on both tumor and immune infiltration cells [96], seems to be more useful in advanced OG adenocarcinoma. Patients with a PD-L1 CPS ≥1 assessed with the 22C3 Dako assay in the single-arm Phase II KEYNOTE-059 trial had a higher ORR than PD-L1-negative patients (ORR: 15.5% [95% CI: 10.1–22.4] vs 6.4% [95% CI: 2.6–12.8]) when treated with third-line pembrolizumab [89]. These findings resulted in USFDA approval for the use of pembrolizumab in chemorefractory locally advanced or metastatic OG adenocarcinoma with PD-L1 CPS ≥1.

Although CPS has become more widely adopted by the OG community as the definition of choice for PD-L1 expression, the clinical applicability of this threshold remains unclear. In the KEYNOTE-061 trial which assessed the efficacy of pembrolizumab compared with second-line weekly paclitaxel, patients with a CPS ≥1 treated with pembrolizumab had comparable survival to patients who received paclitaxel (mOS: 9.1 vs 8.3 months, HR: 0.82, 95% CI: 0.66–1.03; p = 0.04) [85]. However, pembrolizumab prolonged survival in comparison to paclitaxel in patients with CPS ≥10 (mOS: 10.4 vs 8.0 months, HR: 0.64, 95% CI: 0.41–1.02). Updated results from this study following an additional 2 years of follow-up have reiterated that mOS is comparable between patients receiving pembrolizumab and paclitaxel at CPS ≥1 (HR: 0.81, 95% CI: 0.66–1.00), while the survival benefit conferred by pembrolizumab appeared to be more pronounced with increasing thresholds of CPS ≥5 (10.4 vs 8.3 months, HR: 0.72, 95% CI: 0.53–0.99) and CPS ≥10 (10.4 vs 8.0 months, HR: 0.69, 95% CI: 0.46–1.05) [97].

Similarly, results from KEYNOTE-062 that randomized treatment-naive patients with CPS of ≥1 to receive either pembrolizumab monotherapy, platinum-fluoropyrimidine doublet chemotherapy with pembrolizumab or chemotherapy with placebo reported similar OS in the pembrolizumab and chemotherapy arms (mOS: 10.6 vs 11.1 months, HR: 0.91, 95% CI: 0.74–1.10; p = 0.162), leading the authors to conclude that pembrolizumab is noninferior to chemotherapy in the context of first-line therapy [82]. Median OS in the pembrolizumab group improved to 17.4 months in the subset of patients with CPS ≥10, compared with 10.8 months in patients treated with chemotherapy (HR: 0.69, 95% CI: 0.49–0.97). While provocative, the exploratory nature of this analysis limits definitive conclusions from being drawn from these results. Furthermore, median PFS in the pembrolizumab arm was shorter than that seen in the chemotherapy arm at both CPS ≥1 and ≥10 thresholds (mPFS: 2.0 vs 6.4 months, HR: 1.66; 95% CI: 1.37–2.01) and 2.9 vs 6.1 months (HR: 1.10; 95% CI: 0.79–1.51) respectively, implying that the majority of patients treated with pembrolizumab have rapidly progressing disease regardless of CPS status and upfront chemotherapy in these patients may still represent the most appropriate treatment option. Separate analysis of the combination therapy and chemotherapy arms did not demonstrate a difference in OS at both CPS ≥1 and ≥10, although an improvement in ORR in patients with CPS ≥10 was seen with combination therapy (53%) when compared with chemotherapy alone (38%). Pooled analysis of 294 patients with CPS ≥10 from KEYNOTE-059, 061 and 062 reflecting chemorefractory, second- and first-line populations respectively have reported clinically meaningful efficacy and durable OS in comparison to chemotherapy [98].

More recently reported results from Phase III first-line studies further add to the debate surround the quantification of PD-L1 in advanced OG adenocarcinoma. In the global CheckMate-649 trial, 1581 patients were randomized between XELOX/FOLFOX with nivolumab or XELOX/FOLFOX alone. The trial met its co-primary end points of OS and PFS in the PD-L1 CPS ≥5 population, with an improvement of median OS of 3.3 months in patients receiving combination therapy in comparison to chemotherapy alone (14.4 vs 11.1 months, HR: 0.71, 98.4% CI: 0.59–0.86, p < 0.0001) [79]. A statistically significant OS advantage was also seen in the PD-L1 CPS ≥1 (14.0 vs 11.3 months, HR: 0.77, 99.3% CI: 0.64–0.92, p = 0.0001) and overall (13.8 vs 11.6 months, HR: 0.80, 99.3% CI: 0.68–0.94, p = 0.0002) populations. Similarly, median PFS was significantly longer in the combination therapy arm compared with chemotherapy (7.7 vs 6.0 months, HR: 0.68, 98% CI: 0.56–0.81). Nevertheless, 80% of patients recruited into the study were deemed to have PD-L1 CPS ≥1, which is higher than corresponding figures of approximately 60% in previous studies [85,99]. Additionally, 60% of the all patients randomized into CheckMate-649 were deemed to be PD-L1 CPS ≥5. The positive result of CheckMate-649 across all PD-L1 subpopulations may therefore have been driven by the proportionately higher than expected PD-L1-positive subgroup. This also evokes questions surrounding the magnitude of clinical applicability of upfront chemotherapy and immunotherapy in the real-world setting. A separate study recruited 717 patients with HER2-negative gastric adenocarcinoma from Asian countries (ATTRACTION-4) showed that the addition of nivolumab to S-1 or capecitabine with oxaliplatin (SOX) led to superior PFS (10.45 vs 8.34 months, HR: 0.68, 98.51% CI: 0.51–0.90, p = 0.0007) but no improvement in OS was seen with combination therapy (15.45 vs 17.15 months, HR: 0.90, 95% CI: 0.75–10.8, p = 0.257) when compared with SOX or CAPOX alone [80]. These results were derived from a nonbiomarker selected population although patients were stratified based on tumor cell PD-L1 expression (<1% vs ≥1%). Subgroup analyses failed to indicate that patients with positive tumor cell PD-L1 expression benefit from the addition of nivolumab to first-line chemotherapy.

The KEYNOTE-590 study randomized patients with treatment-naive both esophageal squamous cell carcinoma and adenocarcinoma in addition to Type 1 GOJ adenocarcinoma to receive cisplatin and 5-FU with pembrolizumab or placebo [81]. Of the patients recruited, 73.2% (n = 548/749) had squamous cell carcinoma and 26.8% (n = 201/749) had adenocarcinoma histology. While the OS benefit with combination therapy seen was most pronounced in patients with squamous cell carcinoma with a PD-L1 CPS ≥10, (13.9 vs 8.8 months, HR: 0.57, 95% CI: 0.43–0.75, p < 0.0001), a clinically meaningful OS benefit was also seen in all patients with PD-L1 CPS ≥10 (13.5 vs 9.4 months, HR: 0.62, 95% CI: 0.49–0.78, p < 0.0001) irrespective of histology. The breakdown according to histology in the PD-L1 CPS ≥10 cohort, however, remains unknown. The OS benefit seen in all squamous cell carcinoma patients (12.6 vs 9.8 months, HR: 0.72, 95% CI: 0.60–0.88, p = 0.0006) was comparable to that of the entire trial population (12.4 vs 9.8 months, HR: 0.73, 95% CI: 0.62–0.86, p < 0.0001). While it is difficult to draw direct conclusions on the degree of benefit that esophageal adenocarcinoma patients will derive from combination therapy from a trial population which consists predominantly of squamous cell carcinoma, these results indicate that PD-L1 CPS ≥10 may identify responders to immune checkpoint inhibition in esophageal adenocarcinoma.

Collectively, these data highlight that PD-L1 positivity determined using CPS, rather than tumor cell expression, can be used as a tool to inform patient selection to incorporate immune checkpoint inhibition into the systemic therapy armamentarium available in this disease. However, further investigation to identify a singular threshold that reliably discriminates responders from non-responders is clearly required. A separate issue is the interchangeability of PD-L1 assays in OG adenocarcinoma. Combined positive score assessment using the 22C3 assay was initially developed in tandem with pembrolizumab [100], whereas the 28–8 Dako assay used in CheckMate-649 has predominantly been used to determine tumor cell PD-L1 expression in OG adenocarcinoma, which has failed to discriminate between patients who will or will not benefit from immunotherapy [88]. Although both assays have shown analytical concordance in other tumor types [101], direct comparisons in the assessment of PD-L1 CPS in OG adenocarcinoma have yet to be published. Ascertaining PD-L1 status may be more complex in OG adenocarcinoma in comparison to other tumor types as it demonstrates spatial and temporal heterogeneity [102], and both tumor and immune cell PD-L1 expression are integral as evidenced by the clinical utility of PD-L1 CPS thus far. As such, cross validation of individual PD-L1 assays and further interrogation of PD-L1 thresholds capable of delineating subgroups who will benefit from ICI will be fundamental steps in developing a cohesive approach to PD-L1 testing that can be confidently translated into the clinic.

Epstein–Barr virus

Approximately 10% of gastric cancers are categorized as the Epstein–Barr virus (EBV)-positive using TCGA network classification, which was largely derived from early stage tumors [103]. As a subtype, EBV-positive tumors demonstrate a distinct molecular profile which includes extensive DNA protomer hypermethylation, recurrent PIK3CA mutations, amplifications of JAK2 and the chromosome 9 locus containing genes encoding for PD-L1 and PD-L2 and marked intra- or peri-tumoral immune cell infiltration, collectively suggesting potential sensitivity to immunotherapy.

In a single-center Phase II trial in Korea which treated 61 unselected metastatic gastric cancer patients with pembrolizumab, 9.8% of patients (n = 6) were confirmed to be EBV positive. All six patients achieved partial responses with a median duration of response of 8.5 months [95]. A separate series determined EBV status using Epstein–Barr encoding region in situ hybridization (EBER-ISH) in 26 metastatic OG adenocarcinoma patients who received immune checkpoint inhibitor therapy and identified one patient as EBV positive. This patient achieved a complete response lasting >30 months [104]. Prospective evaluation of EBV positivity as a predictor of response to immune checkpoint blockade was also carried out in a Phase Ib/II study of toripalimab determined using EBV DNA copy number. In a cohort of 55 patients tested for EBV, 4 patients had >100 copies which was considered EBV positive [83]. Of these patients, one patient achieved PR, two patients had stable disease and one patient demonstrated progressive disease on toripalimab. Although EBV positivity hold promise as a predictor of response to immune checkpoint blockade, the evidence supporting its predictive role is limited to small studies and its relatively low prevalence pose challenges to more robust interrogation of its predictive role.

Gene expression profiles

Gene expression signatures is a developing and comprehensive tool for predicting response to immune checkpoint inhibition. Among these signatures is the assessment of expression patterns relating to interferon-gamma (IFN-γ), a cytokine produced by activated T cells and natural killer cells. IFN-γ directly upregulates PD-L1 expression to promote cytotoxicity through tumor-infiltrating macrophage recruitment, cytotoxic T-cell proliferation and nitric oxide production. T-cell inflamed tumors show a high IFN-γ signature [103].

An 18-gene IFN-γ-driven pan-tumor gene expression profile (GEP) applied to 220 patients with 9 malignancies, including 33 patients with gastric cancer from KEYNOTE-012, showed a moderate association with clinical response to pembrolizumab [105]. This GEP was assessed in KEYNOTE-059 where a higher GEP score was associated with of radiological responses and longer PFS to pembrolizumab when compared with non-responders [89].

Another IFN-γ gene signature compromising of four genes (CD274, LAG3, CXCL9 and IFN-γ) shown to be predictive of enhanced clinical responses to durvalumab in non-small-cell lung and urothelial cancers [106] was prospectively evaluated in a Phase Ib/II study to determine its ability to identify advanced OG patients most likely to respond to dual immune checkpoint inhibitor therapy with durvalumab and tremelimumab in the second- and third-line setting [87]. The ORR in patients treated with this combination were higher in the IFN-γ group when compared with those seen in an unselected population (ORR 15.8 vs 4.0%), although survival outcomes remained comparable.

In contrast, an analysis of 40 patients recruited into CheckMate-032 to receive nivolumab ± ipilimumab evaluating seven distinct tumor immune infiltration and inflammatory signatures showed that responders had a higher signature scores in aggregate [107]. Specifically, the strongest association was seen in a four-gene BMS inflammatory signature (CD274, CD8A, LAG3 and STAT1). Despite these positive results, elevated expression of immune-related gene signatures may not necessarily be predictive of response to immune checkpoint inhibitors. For example, there is a lack of response to pembrolizumab in gastric tumors with mesenchymal subtype (n = 6), defined by elevated expression of an epithelial-to-mesenchymal transition (EMT) gene signature, despite concurrent elevated levels of immune signatures [95]. It should be noted, however, that the mesenchymal phenotype has previously been identified as a poor prognosticator in OG cancer [108]. Overall, inflammatory signatures warrant further investigation as predictive biomarker to immune checkpoint blockade.

Tumor mutational burden

Another putative biomarker of response to immune checkpoint inhibition is high tumor mutational burden (TMB), the total number of mutations per coding area of a tumor genome as measured by next-generation sequencing (NGS). The median TMB associated with OG adenocarcinoma ranges from five to six mutations/Megabase (Mb) [109–111]. TMB can be measured using either whole exome sequencing or comprehensive genomic profiling; data suggests good concordance between both methods [112]. Presently, a consensus on the definition of ‘high TMB’ specific to OG adenocarcinoma has yet to be established, and published studies estimate this threshold to be between 8 and 20 mutations/Mb based on results derived from comprehensive genomic profiling panels. However, TMB ≥10 mutations/Mb has been shown to be predictive of improved responses to pembrolizumab in multiple solid tumor types in the KEYNOTE-158 trial [113] Based on these results, pembrolizumab received FDA approval for the treatment of patients with unresectable or metastatic solid tumors with TMB ≥10 mutations/ Mb who have progressed following prior treatment with accompanying approval of the use of the FoundationOneCDx (F1CDx) assay as its companion diagnostic.

In single-center retrospective analysis of 101 advanced chemorefractory OG patients treated with anti-CTLA-4 and/or anti-PD1 or anti-PDL1 whose tumor samples underwent targeted next-generation sequencing using the MSK-IMPACT platform, a correlation between high TMB and improved survival was demonstrated using TMB as both a continuous variable and when categorized as quartiles [109]. The uppermost quartile (>8.78 mutations/Mb) was associated with a significant improvement in PFS (p = 0.007) and OS (p = 0.08) when compared with patients aggregated from the lower quartiles. Despite these results, the authors also reported patients who with low TMB who gained durable responses of >6 months from immune checkpoint inhibition.

In clinical trial settings, high tumor TMB using whole exome sequencing has been shown to correlate with enhanced clinical responses in a Phase Ib/II trial investigating the anti-PD1 inhibitor toripalimab (NCT02915432) in patients with treatment naive and pre-treated advanced OG adenocarcinoma [83]. High TMB (TMB-H) was defined as the top 20% of TMB results, which translated to a threshold of ≥12 mutations/Mb. Patients with TMB-H (n = 12) had a higher ORR than those with low TMB (TMB-L) (n = 42) (ORR 33.3% vs 7.1%, p = 0.017). A numerically longer PFS was reported in the TMB-H group in comparison to the TMB-L group (2.5 vs 1.9 months, HR: 0.51, 95% CI: 0.26–1.02; p = 0.055).

Additionally, an exploratory analysis assessing the effect of tumor TMB measured by whole exome sequencing with clinical outcomes based on patients enrolled in KEYNOTE-061 [114]. The clinical utility of tumor TMB as a predictive biomarker of response to pembrolizumab was prespecified ≥175 mutations/exome. Patients with ≥175 mutations/exome who received second-line pembrolizumab had a higher ORR (30%, 95% CI: 17–47) than those with <175 mutations/exome (ORR 8%, 95% CI: 5–14) and tumor TMB remained significantly associated with PFS and OS even after exclusion of MSI-H patients. Accompanying data investigating the evaluation of tumor TMB using the F1CDx panel with a prespecified threshold of ≥10 mutations/Mb also showed a positive association between tumor TMB and clinical outcomes [115]. These results were consistent with findings seen using whole exome sequencing and reinforce the notion that both WES and comprehensive genomic profiling panels may be used in clinical settings to inform patient selection.

Evidence also suggests correlation between tissue and circulating tumor DNA (ctDNA) mutational load. Kim et al. utilized the commercially available Guardant360 panel (73 genes) and showed good concordance (linear regression, r² = 0.54) in 23 patients where both tissue and plasma exome sequencing data were available [95]. High mutational load in both tissue and ctDNA showed a correlation improved radiological responses and PFS.

Tumor-infiltrating lymphocytes

The immune microenvironment may influence disease trajectory and modulate response to immune checkpoint inhibition. In colorectal cancer, the locations of tumor-infiltrating lymphocytes (TILs) confer prognostic significance independent of histopathological staging [116]. These patterns are less clear in OG cancer; although higher densities of CD8⁺ T cells and lower densities of FOXP3+ regulatory T cells seem to be favorable prognostic factors following curative surgery, many studies show contradictory conclusions [117], with more recent studies describing an association between higher densities of FOXP3+ with better prognosis in surgically resected chemotherapy-naive gastric cancer cohorts [118,119]. An ‘immune-inflamed’ tumor microenvironment characterized by CD8⁺ and CD4⁺ T cells within tumor parenchyma and at the invasive margin [120] is associated with improved responses and survival in patients with immune checkpoint inhibition across multiple tumor types [121,122], and there are early reports that composition, density and distribution of T-cell subsets in OG cancer determined via multiplex IHC may be associated with response to immune checkpoint blockade [123].

High TILs infiltration is also characteristic of MSI-H and EBV-positive subtypes of OG cancer. Despite this, a study which categorized 247 primary gastric cancers (43 EBV-positive, 79 MSI-H and 125 EBV-negative/microsatellite stable [MSS]) based on PD-L1 and CD8⁺ T-cell IHC expression has shown that approximately 40% of EBV-negative/MSS tumors have significant CD8⁺ T-cell infiltration [124]. Another analysis of 103 tumors from advanced OG patients also showed that while defective MMR and the presence of TILs were correlated, both parameters retained their independent prognostic role on multivariate analysis [125]. Although TILs infiltration has yet to be shown to be of predictive value to immune checkpoint inhibition in OG cancer, unravelling the interplay between tumor and immune milieu may help understand the influence of the tumor microenvironment on treatment outcome.

Conclusion

Improving biomarker-driven therapeutics and clinical outcomes in OG adenocarcinoma remains a significant challenge despite the extensive preclinical, translational and clinical investigations thus far. Further developments in our understanding of the underlying mechanisms of response and resistance, including the role of the microenvironment on tumor behavior are key, before we can form seamless transitions of these concepts into the clinic. The use of precision medicine to tailor individual patient strategies is an exciting prospect but necessitates access to cost-effective genomic and transcriptomic sequencing technologies, which currently pose a barrier to widespread use even in the face of promising research findings. Coupled with rational drug design and treatment strategies to overcome the challenges posed by the heterogenous nature of OG adenocarcinoma, we may be able to herald a new era in the management of these lethal cancers in the future.

Future perspective

Although the concept of predictive biomarkers has been proposed for decades, the discovery of specific genetic or protein biomarkers has been fundamentally complex because of the technical nature of comprehensive expression platforms, limitations in clinical assay development, low biomarker prevalence and, most significantly, an inadequate understanding of tumor biology. Nevertheless, how can we attempt to overcome these challenges for the future benefit of our patients?

Recent studies, such as those published by the TCGA research network [103] and Asian Cancer Research Group (ACRG) [108], have proposed that esophago-gastric cancers can be divided into molecular subtypes, shifting the paradigm of OG classification away from traditional histological categorization [126]. In addition to providing insight into the biology of OG cancer, these classifiers also provide an opportunity to develop a therapeutic approach dictated by molecular subtyping. The first hurdle limiting its translation to routine clinical use is the feasibility of prospective molecular analysis in a diagnostic setting. Ensuing efforts by the ACRG has led to the development of a 71-gene signature via the NanoString platform which displayed high concordance with the Affymetrix microarray platform initially used [127]. Separately, prospective use of the MSK-IMPACT targeted panel has been successfully applied to identify predictive biomarkers to contemporary systemic therapies in metastatic OG cancers and could be a means to define somatic gene mutations associated with individual molecular subtypes [104]. The second challenge currently restricting the clinical application of molecular subgroups to inform treatment selection is the identification of efficacious treatment options for each subtype. While MSI-H and EBV-positive patients have been shown to achieve durable responses to immune checkpoint blockade [104], the therapeutic opportunities may not be as straightforward in other subtypes as appropriate therapeutic options may not be available.

There have already been efforts to delve into precision oncology in the field of OG cancer. The VIKTORY basket trial performed targeted sequencing on 772 patients with advanced OG adenocarcinoma to identify eight targetable biomarkers (RAS mutation/amplification, TP53 mutation, PIK3CA mutation/amplification, MET amplification, MET overexpression, all negative, TSC2 deficient, or RIC-TOR amplification) with the aim to assign these patients into ten associated clinical trials in the second-line setting [128]. The authors reported that 14.7% of patients received biomarker-assigned drug treatment, with longer survival observed in this cohort in comparison to patients who received conventional second-line chemotherapy.

However, the spatial and temporal molecular heterogeneity seen in OG cancers implies that a single tumor biopsy is unlikely to be truly reflective of a patient's overall disease biology and adds a layer of complexity to the development of precision therapeutics in this tumor type. Taking this into account, the PANGEA study aimed to develop a personalised treatment strategy using monoclonal antibodies in combination with chemotherapy based on a predefined treatment algorithm over three lines of systemic therapy applied to tissue and cell-free DNA (cfDNA) profiles obtained at baseline and at disease progression [129]. Treatment reassignment according to metastatic tumor profiling over primary tumor profiles was also permitted. Using this strategy, PANGEA met its primary efficacy end point with a 1-year OS of 66% and mOS of 15.7 months. Comparative biomarker analysis of the primary tumor and metastatic tissue of patients enrolled into PANGEA demonstrated significant discordance in 36% of paired samples (n = 10/28), resulting in treatment reassignment in nine of these patients [130]. High concordance was seen between metastatic and cell-free DNA profiling. Moreover, five discordant cases had no actionable mutations detected in the primary tumor, yet actionable targets in ERBB2, MET, EGFR and FGFR2 were found in metastatic and cfDNA profiles.

These findings underscore the appeal of liquid biopsies in OG cancers, which confer obvious advantages to the patient experience due to its noninvasive nature and potential for longitudinal monitoring of tumor evolution secondary to therapeutic pressure. The evolution of somatic copy number aberrations, recognised as key drivers in OG adenocarcinoma, have been successfully tracked using low-coverage whole-genome sequencing of ctDNA throughout first-line chemotherapy, with chromosome 1q and 8p gains before treatment showing an association with chemotherapy responses [131]. A separate study with a small cohort of stage IV OG patients (n = 35) demonstrated that a decline in variant allele fraction of >50% was prognostic for superior survival. The investigators also assessed HER2 status in both primary and metastatic lesions as well as in ctDNA in seven HER2-positive cases and found that only two patients had concordant results in all three modalities. Additionally, acquired resistance mechanisms were also detected using ctDNA in patients with retained tumoral HER2 amplification following treatment with trastuzumab. These findings highlight the potential of liquid biopsies in providing prognostic and predictive information and offer a basis to refine the analytical methodologies required to further develop its role as an adjunct to clinical decision-making.

Finally, rational and innovative clinical trial design is crucial. Identification of potential predictive biomarkers should be reflected in biomarker-stratified clinical evaluation. Multiple Phase III trials assessing the efficacy of anti-EGFR antibodies in OG cancer have reported negative outcomes in biomarker-unselected populations [132–134], yet further investigation has indicated that patients with EGFR amplification and overexpression can derive clinical benefit from anti-EGFR therapy [135]. Given the relatively low biomarker prevalence of driver aberrations in OG adenocarcinoma, increasing use of platform and adaptive studies should also be considered to evaluate treatment strategies efficiently.

Executive summary.

• Esophago-gastric (OG) adenocarcinoma is associated with a poor prognosis due to a combination of late presentation, aggressive tumor biology and challenges in therapeutic development due to tumor heterogeneity.

• The use of chemotherapy remains underpins the management of advanced OG adenocarcinoma.

• HER2 is the only biomarker of response routinely targeted in the clinic using trastuzumab in combination with fluoropyridimine/platinum containing chemotherapy in the first-line setting. Novel therapeutic approaches, including the use of the antibody–drug conjugate trastuzumab deruxtecan, HER2-directed immunotherapy and bispecific antibodies are being developed.

• Other putative predictive biomarkers with relevant targeted therapies development include FGFR amplifications, claudin 18.2, MET and DNA damage repair pathway deficiencies.

• Immune checkpoint blockade has been assessed as monotherapy and in combination therapy in various settings in OG adenocarcinoma, and will likely be incorporated into first-line systemic therapy of advanced disease.

• The majority of clinical trials suggest that a subset of patients achieve durable responses from immunotherapy, although this subgroup of patients have yet to be fully characterized.

• In addition to microsatellite instability/mismatch repair deficiency, PD-L1 positivity measured using the combined positive score (CPS) is an increasingly recognised biomarker of response to immunotherapy, although there is yet to be consensus regarding the threshold indicating PD-L1 positivity.

• Other predictive biomarkers to immunotherapy in OG adenocarcinoma include high TMB, EBV positivity and gene expression profiles.

• Despite some progress in the molecular characterization of OG adenocarcinoma and translating predictive biomarkers into the clinic, an inadequate understanding of tumor biology continues to hamper therapeutic development.

• Utilizing molecular classifications to dictate therapeutic selection is currently limited due to the identification of efficacious treatments for individual subtypes.

• There is a growing body of evidence to support the use of next-generation sequencing to elucidate targetable biomarkers in OG adenocarcinoma. Due to the heterogenous nature of OG adenocarcinoma, treatment reassignment based on metastatic tumor profiling over primary tumor profiles should be considered.

• Liquid biopsies can enable longitudinal monitoring of tumor evolution secondary to therapeutic pressure and have the potential to provide prognostic and predictive information that could support treatment decisions.

• Successfully incorporating further targeted therapeutic strategies in OG adenocarcinoma will likely require a concerted effort to improve our insight into its underlying biology coupled with rational clinical trial design to facilitate any meaningful improvements in patient outcomes.