Novel targets in the treatment of advanced gastric cancer: a perspective review

Abstract

Gastric cancer is responsible for a high burden of disease globally. Although more extensive use of chemotherapy together with the recent introduction of the two targeted agents trastuzumab and ramucirumab have contributed to marginal outcome prolongation, overall survival for patients with advanced stage disease remains poor. Over the last decade, a number of novel agents have been examined in clinical trials with largely disappointing results. Potential explanations for this are the absence of molecularly selected trial populations or weak predictive biomarkers within the context of a highly heterogeneous disease. In the recently published gastric cancer The Cancer Genome Atlas (TCGA) project a new classification of four different tumour subtypes according to different molecular characteristics has been proposed. With some overlap, several relatively distinct and potentially targetable pathways have been identified for each subtype. In this perspective review we match recent trial results with the subtypes described in the gastric cancer TCGA aiming to highlight data regarding novel agents under evaluation and to discuss whether this publication might provide a framework for future drug development.

Introduction

Gastric cancer (GC) is responsible for a high global burden of disease; it is the fifth most common cancer worldwide with 951,000 new cases annually, and is the third leading cause of cancer death with 723,000 events every year [Ferlay et al. 2015]. The highest incidence rates occur in Asia; whereas eastern Asia, central and eastern Europe and South America all have high GC-related mortality rates.

Approximately 50% of patients are diagnosed with GC when the disease is already locally advanced or metastatic. For early stage adenocarcinoma surgery remains the only potentially curative treatment, however more than 50% of patients experience disease relapse [Cunningham et al. 2006; Ychou et al. 2011]. Overall survival (OS) of patients with stage IV disease remains poor, between 3 months in the absence of any active treatment and 9–16 months in fit patients within the context of clinical trials [Cunningham et al. 2008; Bang et al. 2010; Wagner et al. 2010].

Treatment of advanced GC: cytotoxic chemotherapy as standard

Chemotherapy remains the backbone of treatment for advanced GC resulting in superior survival outcomes and symptom control and consequently improved quality of life compared with best supportive care [Wagner et al. 2010]. The chemotherapy regimen of choice varies geographically, however a platinum fluoropyrimidine based treatment is accepted worldwide as first-line therapy. In the UK NCRI REAL-2 clinical trial, oxaliplatin demonstrated noninferiority to cisplatin and capecitabine to infused 5-fluorouracil. Based on this study, a triplet regimen consisting of epirubicin, a platinum salt (cisplatin or oxaliplatin) and a fluoropyrimidine (capecitabine or 5-fluorouracil) is a standard first-line option for fit patients and results in a median OS between 9 and 11 months [Cunningham et al. 2008]. Similar survival outcomes have been achieved by adding docetaxel to the doublet cisplatin and 5-fluorouracil, although this regimen is associated with significantly increased toxicity [Van Cutsem et al. 2006]. Doublet regimens consisting of cisplatin or oxaliplatin and a fluoropyrimidine are a validated option for patients not fit for three drugs. For patients who are not suitable for platinum, irinotecan-containing regimens may provide a reasonable alternative [Guimbaud et al. 2014].

Following publication of the ToGA trial, the use of the monoclonal antibody trastuzumab targeting the human epidermal growth factor receptor 2 (HER2) in combination with cisplatin and fluorouracil has become a standard treatment for HER-2 positive GC (approximately 20% of patients in total) [Bang et al. 2010]. The role of second-line treatment with a taxane or irinotecan has recently also been accepted globally, as several trials have demonstrated a marginal (approximately 6 weeks) benefit in survival and quality of life [Thuss-Patience et al. 2011; Kang et al. 2012; Fuchs et al. 2014; Ford et al. 2014]. The antiangiogenic monoclonal antibody ramucirumab has recently been approved as a single agent or in combination with paclitaxel as second-line therapy, and for appropriate patients may increase OS in this setting to over 9 months when delivered in combination with paclitaxel [Fuchs et al. 2014; Wilke et al. 2014]. Thus it is clear that with the notable exception of trastuzumab, the introduction of molecularly targeted novel agents into clinical practice has lagged behind other cancers. Given the low median survival time for patients with metastatic GC, further research is required in this area.

The results of the recently published The Cancer Genome Atlas (TCGA) for GC have highlighted the complex heterogeneity of this disease [The Cancer Genome Atlas Research Network, 2014]. Using these results as a guide, in this perspective review we aim to identify and discuss data regarding novel agents directed to the targets which characterise each molecular GC subtype. In doing so, we will examine how the TCGA results provide a context to discuss previous trials in GC, and may also provide a framework for future drug development.

Molecular characterization from the TCGA analysis

Several previous GC classification systems have been described, however the generalizability of these was questionable due to the geographically discrete patient groups used and failure to validate results in external datasets [Shah et al. 2011; Tan et al. 2011; Deng et al. 2012; Lei et al. 2013; Smyth et al. 2013]. In order to overcome these challenges for the GC TCGA project, a total of 295 untreated surgically resected gastric adenocarcinomas were evaluated from a global population. Six distinct methodologies [array-based somatic copy number analysis, whole-exome sequencing, array-based DNA methylation profiling, messenger RNA sequencing, microRNA (miRNA) sequencing and reverse-phase protein array] were used to characterize patients, plus evaluation of microsatellite instability. The authors proposed a new classification in four different subtypes according to shared molecular characteristics. These were Epstein Barr positive tumours (EBV), microsatellite unstable tumours (MSI), genomically stable tumours (GS) and tumours with chromosomal instability (CIN). Although some characteristics of each subtype partially overlap, the prevalence of relatively distinct potentially targetable pathways has been identified for each subgroup (see Table 1).

Table 1.

Characteristics of the four diverse molecular subtypes (modified from TCGA report; total number: 295).

Characteristics		EBV (9%)	MSI (22%)	CIN (50%)	GS (20%)
Sex	Female	5 (19%)	36 (56%)	50 (34%)	22 (38%)
	Male	21 (81%)	28 (44%)	97 (66%)	36 (62%)
Lauren classification	Diffuse	5 (19%)	6 (9%)	18 (12%)	40 (69%)
	Intestinal	15 (57%)	48 (75%)	118 (80%)	15 (26%)
	Mixed	3 (11%)	3 (5%)	10 (7%)	3 (5%)
	not reported	3 (11%)	7 (11%)	1 (1%)	0
Main actionable targets		PD-L1/2 expression JAK2 expression PIK3CA mutation	Loss of HLA class 1 complexes PIK3CA mutation	HER2 amplification EGFR amplification MET amplification CCNE1, CCND1, CDK6 amplification VEGFA amplification FGFR2 amplification	VEGFA amplification FGFR2 amplification Cell adhesion pathwaysRHO-A?

CCND1, cyclin D1; CCNE1, cyclin E1; CDK, cyclin dependent kinase; CIN, chromosomal instability; EBV, Epstein Barr virus; EGFR, epidermal growth factor receptor; FGFR, fibroblast growth factor receptor; GS, genomically stable; HER2, epidermal growth factor receptor 2; HLA, human leukocyte antigen; JAK, Janus kinase; MSI, microsatellite unstable; OGJ, oesophagogastric junction; PD-L, programmed death ligand; PIK3CA, phosphatidylinositol 4,5 bisphosphate 3 kinase, catalytic subunit α; MET: mesenchimal epithelial transition; VEGFA, vascular endothelial growth factor A.

Targetable pathways in EBV tumours: immune-based therapies, phosphatidylinositol 4,5 bisphosphate 3 kinase, catalytic subunit α and Janus kinase 2

The EBV subtype (9% tumours in TCGA) is characterized by high EBV burden (determined using mRNA, miRNA, exome and whole genome sequencing) and high levels of DNA promoter hypermethylation (all EBV-positive tumours demonstrated CDKN2A promoter methylation). EBV-positive tumours were more frequently located in the gastric fundus or body and 81% of cases occurred in men. Relevant targetable pathways identified in this subtype were related to the elevated expression of programmed death ligand 1 and 2 (PD-L1 and PD-L2), Janus kinase 2 (JAK2) amplification and phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit α (PIK3CA) mutation.

An increasing enthusiasm for immunotherapy agents has been rapidly propagated from the treatment of advanced melanoma to several tumour types, including GC [Robert et al. 2011; Wolchok et al. 2013; Larkin et al. 2015]. Pembrolizumab, a highly selective immunoglobulin G4k humanized monoclonal antibody targeting the PD-1 receptor demonstrated activity in the phase Ib KEYNOTE-012 trial in a cohort of heavily pretreated Asian and non-Asian patients with GC, with acceptable toxicity profile [Shankaran et al. 2015]. On central review, overall response rate was 22% [95% confidence interval (CI) 10–39%] with median duration of response of 24 weeks (range 8–33 weeks). As most responses to chemotherapy in GC are short lived, this is of significant interest. A correlation between PD-L1 expression (defined as PD-L1 ⩾ 1% on archival tissue) and response has been subsequently demonstrated [Bang et al. 2015b]. A double checkpoint inhibition strategy targeting both the PD-1/PD-L1 and CTLA-4/B7 interaction which has already demonstrated efficacy in patients with melanoma is under evaluation in several GC trials such as nivolumab + ipilimumab [ClinicalTrials.gov identifier: NCI01928394] and MEDI4734 and tremelimumab [ClinicalTrials.gov identifier: NCI02340975].

The JAK pathway has to date been primarily a targeted strategy for myeloproliferative and inflammatory disorders and has only recently extended to solid tumours [Buchert et al. 2015]. With respect to gastrointestinal malignancies, ruxolitinib, a JAK1 and JAK2 inhibitor, has demonstrated preliminary efficacy in combination with capecitabine in pancreatic adenocarcinoma and is currently under evaluation in colorectal cancer in combination with regorafenib [ClinicalTrials.gov identifier: NCI02119676] [Hurwitz et al. 2014]. To our knowledge there are no trials ongoing in GC.

Activating mutations of the PIK3CA gene represent a further potentially actionable target in GC. In previous series the prevalence of this mutation has been reported to range between 5% and 16% [Barbi et al. 2010] and to be more frequent in white patients compared with Asian patients (15% versus 7%) [Chong et al. 2014]. In the TCGA report, 80% of EBV tumours and 42% of MSI tumours harboured a mutation in PIK3CA (although more commonly in the exon 20 kinase domain in non-EBV type tumours), thus supporting the rationale of pursuing the inhibition of this pathway in clinical trials.

Alternative mechanisms of targeting the PI3K pathway in cancer include inhibiting downstream effectors such as mechanistic target of rapamycin (mTOR) and AKT. Despite initially promising results with a disease control rate of over 50% in a phase II study, the oral mTOR inhibitor everolimus failed to improve survival compared with placebo in a phase III multicentre randomized clinical trial in patients with previously treated advanced GC, although a trend toward a reduced risk of death in non-Asian patients was noted [Ohtsu et al. 2013]. This trial was not biomarker selected and translational analyses are awaited. Several AKT inhibitors are currently being examined in GC: a randomized phase II trial of ipatasertib or placebo in combination with mFOLFOX (chemotherapy regimen for treatment of colorectal cancer, made up of the drugs: FOL – Folinic acid (leucovorin) F – Fluorouracil (5-FU) OX – Oxaliplatin (Eloxatin)) as first-line treatment for advanced GC recently completed recruitment while afuresertib in combination with paclitaxel in a second-line setting is currently under investigation in a phase IB study in an unselected population [ClinicalTrials.gov identifier: NCT01896531, NCT02240212]. In a molecularly selected population a further orally available AKT inhibitor, AZD5363, is currently being utilized in combination with paclitaxel in a second-line setting in patients with GC harbouring PIK3CA mutation or amplification [ClinicalTrials.gov identifier: NCT02451959].

Targetable pathways in MSI tumours

The MSI-high subtype represented 22% of the samples of the TCGA analysis and was characterized by elevated mutation rates and hypermethylation resulting in MutL homolog 1 (MLH1) silencing. This population was older, with a median age of 72 years and with a higher proportion of female patients (56%) compared with the other subtypes. Major histocompatibility complex class I gene alterations were common in this subtype. This, together with the increased number of tumour-specific neoantigens derived from hypermutated genes, suggest a potential additional role for immunotherapy for this category of tumour. Evidence of activity of pembrolizumab in a subset of patients with MSI-positive colorectal cancer has recently been presented; immune-related objective response and progression free survival rates were 40% and 78% respectively [Le et al. 2015]. With respect to the data presented on pembrolizumab in GC to date, neither MSI nor EBV status has been revealed at the time of writing. In addition to mutations in the kinase domain of PIK3CA, mutations in ERBB3 and KRAS among others were found in significant numbers in MSI tumours, which may represent other potential targetable abnormalities for future drug development.

One potential challenge in developing suitable candidate therapies for patients with MSI- and EBV-type tumours is that, paradoxically, these patients are likely to have improved survival following surgery compared with patients with other subtypes. Both MSI and EBV positivity have been validated as favourable prognostic factors in resected GC, and therefore may be present in lower proportions in the metastatic setting, with subsequent case finding difficulty hindering trial design [Camargo et al. 2014; Choi et al. 2014].

Targetable pathways in CIN tumours: receptor tyrosine kinase amplifications predominate

The largest subgroup within the gastric TCGA is that of chromosomally unstable tumours, comprising half of all patients (65% of junctional tumours) and correlating well with an intestinal pathological phenotype. Within this group, several targetable pathways have been identified, the majority of them related to receptor tyrosine kinase (RTK) gene amplifications, including HER2, epidermal growth factor receptor (EGFR), mesenchymal epithelial transition factor (MET), fibroblast growth factor receptor (FGFR) and vascular endothelial growth factor A (VEGFA). This group of cancers has been the subject of extensive clinical trials over the past decade.

The HER2 pathway has been well described in GC and trastuzumab is the sole validated targeted therapy in this disease [Bang et al. 2010] (see Table 2). Following ToGA, several anti-HER2 agents have been examined. The addition of pertuzumab (a monoclonal antibody blocking HER2/HER3 dimerization) to trastuzumab and docetaxel has already demonstrated a survival benefit in patients with breast cancer [Baselga et al. 2012]. To evaluate this combination in GC, a phase III multicentre international clinical trial of pertuzumab or placebo in combination with trastuzumab and cisplatin–fluoropyrimidine regimen is ongoing, with an estimated study completion date of September 2021 [ClinicalTrials.gov identifier: NCT01774786].

Table 2.

Key randomized phase III trial with targeted agents in previously untreated gastric cancer.

Trial name	Selected biomarker	Treatment	Number of patients	Primary endpoint	Hazard ratio (95% confidence interval)	p value	Median OS (months)
ToGA[Bang et al. 2010]	HER2 expression/amplification	CF/CXCF/CX +trastuzumab	290294	OS	0.74 (0.60–0.91)	0.004	13.811.1
LOGiC[Hecht et al. 2013]	HER2 expression/amplification	CapeOXCapeOx+lapatinib	273272	OS	0.91 (0.73–1.12)	0.35	10.512.2
REAL-3	all comers	EOX	275	OS	1.37 (1.07–1.76)	0.013	11.3
[Waddell et al. 2013]		EOX+panitumumab	278				8.8
EXPAND	all comers	CX	449	PFS	1.09 (0.92–1.29)	0.32	10.7
[Lordick et al. 2013]		CX+cetuximab	445				9.4
RILOMET[Cunningham et al. 2015]	MET positive by IHC HER2 negative	ECXECX+rilotumumab	305304	OS	1.37 (1.06–1.78)	0.016	11.5 9.6
AVAGAST	all comers	CX	387	OS	0.87 (0.73–1.03)	0.1	10.1
[Ohtsu et al. 2011]		CX+bevacizumab	387				12.1
AVATAR	all comers	CX	102	OS	1.11 (0.79–1.56)	0.5567	11.4
[Shen et al. 2015]		CX+bevacizumab	100				10.5

CapeOX, capecitabine, oxaliplatin; CF, cisplatin, fluorouracil; CX, cisplatin, capecitabine; ECX, epirubicin, cisplatin and capecitabine; EOX, epirubicine, oxaliplatin, capecitabine; HER2, human epidermal growth factor receptor 2; IHC, immunohistochemistry; OS, overall surivial; PFS, progression-free survival.

At this time, standard salvage treatment options for HER2-positive tumours are similar to those for HER2-negative disease, however the blockade of the HER2 pathway beyond trastuzumab progression is under investigation. The antibody–drug conjugate trastuzumab emtansine (T-DM1) has been evaluated in the second-line setting in a phase II clinical trial in previously treated patients with HER2-positive metastatic or advanced GC compared with docetaxel or paclitaxel. The primary objective of the study is OS and the results are anticipated in 2016 [ClinicalTrials.gov identifier: NCT01641939]. MM-111, a bispecific antibody fusion protein binding both HER2 and HER3 (the preferred dimerization partner of HER2) has been evaluated in a phase II clinical trial in combination with trastuzumab and paclitaxel [ClinicalTrials.gov identifier: NCT01774851], however this trial was terminated following inferior progression-free survival (PFS) results in the experimental arm.

Straddling the border between HER2 and EGFR inhibition, lapatinib, a dual inhibitor of EGFR and HER2, has been examined first line in combination with capecitabine and oxaliplatin and second line in combination with paclitaxel [Hecht et al. 2013; Satoh et al. 2014b]. Both phase III studies failed to meet their primary endpoint. In the TRIO013/LOGiC Trial, the primary endpoint of OS in patients with centrally confirmed harbouring HER2 amplification by fluorescence in situ hybridization (FISH) was not met [hazard ratio (HR) 0.91, 95% CI 0.73–1.12, p = 0.35]. However, a subgroup analysis demonstrated significant benefit in Asian patients and those younger than 60 years [Hecht et al. 2013]. Similarly, in the TyTAN trial, patients with previously treated HER2 FISH amplified tumours were randomized to paclitaxel + lapatinib/placebo. No significant benefit in OS was seen (HR 0.84, 95% CI 0.64–1.11, p = 0.1044). However, a significant benefit in response rate in patients with FISH amplification and 3+ immunohistochemistry (IHC) was noted [Satoh et al. 2014b]. Although the primary endpoint was not met in either study, subgroup analysis has provided proof of concept for some activity of lapatinib in HER2-amplified cancers. Investment in dual inhibitors of the ERBB family continues; at the recent American Society of Medical Oncology (ASCO) Annual Meeting 2015 an oral reversible tyrosine kinase inhibitor of the EGFR and HER2 named S-222611 demonstrated a 15% response rate (including one complete response) in HER2-positive GC [Deva et al. 2015].

EGFR gene amplification is the second most frequent RTK alteration reported in the GC TCGA study and is demonstrated in 10% of CIN subtype tumours. This finding was consistent with previous data, with rates between 4.9% and 7.7% being reported [Deng et al. 2012; Kiyose et al. 2012]. Unfortunately, disappointing results from two large randomized phase III trials have discouraged further investigation of anti-EGFR agents in molecularly unselected populations [Lordick et al. 2013; Waddell et al. 2013]. Panitumumab added to epirubicin, capecitabine and oxaliplatin as first-line treatment for metastatic or locally advanced oesophagogastric adenocarcinomas resulted in detrimental outcomes compared with chemotherapy alone [Waddell et al. 2013]. This was hypothesized to be related to a reduction in chemotherapy dose intensity due to overlapping toxicity (in terms of grade 3–4 diarrhoea), and potentially negative interactions between anti-EGFR agents and oxaliplatin-based regimens. Two other anti-EGFR agents, cetuximab and nimotuzumab, did not demonstrate benefit when added to standard chemotherapy. Cetuximab did not result in a survival advantage when added to the cisplatin–capecitabine regimen in previously untreated advanced junctional or gastric adenocarcinoma in the EXPAND trial [Lordick et al. 2013] and in two phase II randomized studies nimotuzumab failed to demonstrate benefit both in the first-line setting in combination with cisplatin and S-1, and in the second-line setting with irinotecan (even in a subset expressing EGFR 2+ or 3+ by IHC) [Satoh et al. 2014a; Du et al. 2015]. Additionally, two anti-EGFR tyrosine kinase inhibitors erlotinib and gefitinib were ineffective in a phase II trial and phase III trial compared with best supportive care in oesophageal and junctional cancers [Dragovich et al. 2006; Dutton et al. 2014]. However, intriguing retrospective biomarker analyses of the COG trial suggest that a subpopulation of tumours with EGFR copy number gain may benefit from anti-EGFR therapy, implying that refining the EGFR biomarker may yet yield positive results [Petty et al. 2014].

Disappointingly, both monoclonal antibodies and RTK inhibitors targeting the MET pathway have failed to meet expectations [Cunningham et al. 2015; Shah et al. 2015]. High variability of MET protein expression by IHC has been reported in case series, ranging from 22% to 90% [Ha et al. 2013; Scagliotti et al. 2013], whereas MET amplification has been more homogeneously reported in the literature as between 2% and 10% and confirmed in 8% of CIN subtype tumours in the TCGA series [Peng et al. 2014; The Cancer Genome Atlas Research Network, 2014]. Both MET overexpression and MET amplification have been validated as negative prognostic factors in GC and this pathway was therefore promising for pharmacological intervention [Peng et al. 2014].

Rilotumumab, a fully human monoclonal antibody targeting the hepatocyte growth factor (HGF), ligand of MET receptor, was associated with significantly longer PFS and OS when added to epirubicin, cisplatin and capecitabine chemotherapy in treatment naïve molecularly unselected patients with advanced gastric or oesophagogastric junction tumours in a multicentre phase II trial [Iveson et al. 2014]. For patients with MET-positive tumours, OS was longer in rilotumumab-treated patients compared with those on placebo (10.6 versus 5.7 months respectively). However, the randomized phase III trial recently presented by Cunningham and colleagues in patients with MET-positive tumours by IHC was terminated prematurely due to an imbalanced number of deaths in the experimental arm [Cunningham et al. 2015]. Concurrently, disappointing results were also presented regarding another anti-MET antibody, onartuzumab, in combination with mFOLFOX [Shah et al. 2015]. Compared with placebo in previously untreated HER2-negative advanced oesophagogastric adenocarcinomas in a randomized phase II study, onartuzumab did not improve the primary endpoint of PFS in an unselected population or in the MET-positive subgroup. The orally available MET kinase inhibitor AMG337 demonstrated encouraging preliminary activity in patients with oesophagogastric adenocarcinomas after failure of first-line treatment; of 10 patients with MET-amplified tumours, one patient had complete response, four had partial response and two had stable disease. However, in March 2015 after a prespecified interim review of efficacy and safety data, the study was terminated [Kwak et al. 2015].

In the GC TCGA dataset, 9% and 8% of CIN tumours and GS tumours respectively harboured amplification of the FGFR2 gene [The Cancer Genome Atlas Research Network, 2014]. Several orally available pan-FGFR inhibitors are under investigation in GC and other tumour subtypes. The results of a phase II randomized trial of AZD4547 (a selective inhibitor of FGFR1, 2 and 3) or paclitaxel in patients with FGFR2 amplification or polysomy in a second-line setting have recently been presented [Bang et al. 2015c]. In the population screened, the prevalence of FGFR2 amplification (9%) was in line with TCGA data. Following an interim analysis, in which PFS was no different between the experimental and paclitaxel groups, the study was halted prematurely. Biomarker analysis from the study has underlined high intratumour heterogeneity of FGFR2 amplification, an issue which has also been problematic in HER2 assessment in GC [Kim et al. 2011]. Interestingly, concurrently with this study a further phase II study with the same compound and the same target population identified high a FISH ratio (>8) and the presence of elevated FGFR2 copy number in free plasma DNA as potential biomarkers predictive of response to AZD4547, highlighting that thorough evaluation and refinement of the biomarker is a requisite for informed drug development [Smyth et al. 2015]. Two ongoing phase II studies are evaluating dovitinib (a multikinase inhibitor targeting FGFR) after failure of first-line treatment as a single agent in FGFR2-amplified GC (defined as copy number >3) or in combination with docetaxel in a biomarker unselected population [ClinicalTrials.gov identifier: NCT01719549, NCT01921673].

On a broader level the chromosomal instability which is characteristic of CIN subset GC may be exploited for therapeutic gain. Defects in DNA damage repair lead not only to point mutations but also to loss of part or whole chromosomes [Watkins et al. 2014; Marquard et al. 2015]. The presence of such genomic scars is associated with platinum sensitivity, and also sensitivity to poly ADP ribose polymerase (PARP) inhibitors. In ovarian cancer, the presence of large-scale homologous DNA repair defects (and not only BRCA mutations) has been associated with an increased benefit from PARP inhibition using rucaparib [McNeish et al. 2015; Shapira-Frommer et al. 2015]. As a randomized phase II trial of olaparib in combination with paclitaxel demonstrated an overall survival benefit in a nonbiomarker-selected second-line GC population, it will be interesting to observe whether molecular selection of patients using an HRD-LOH (Homologous Recombination Deficiency-Loss of Heterozygosity) signature or other measure of chromosomal instability will yield further incremental gains [Bang et al. 2015b].

Unique targetable pathways in GS tumours: RHOA

Tumours which did not meet the criteria for the previously discussed three subtypes, which are EBV negative, mismatch repair proficient and with low degree of aneuploidy, were classified as GS [The Cancer Genome Atlas Research Network, 2014]. This subtype represented 20% of the TCGA samples and has been associated with diffuse histology, earlier age at presentation (median 59 years) and distal localization. Several unique subtype specific molecular changes have been described for GS tumours. These include abnormalities of RHOA signalling encompassing mutations and CLDN18–ARHGAP26 fusions (approximately 15% for each aberration in a mutually exclusive fashion) which have been validated in external datasets [Kakiuchi et al. 2014; Yao et al. 2015]. As Ras homolog gene family, member A (RHOA) signalling has a role in cell motility, these changes may account for the discohesive, invasive nature of this GC subtype. Although not currently targetable, these pathways represent a valuable signpost for development of effective treatments for diffuse GC.

With respect to pathways currently under investigation in clinical trials, in common with the CIN subtype, the presence of FGFR2 and VEGFA amplification in GS tumours suggests angiogenesis an attractive targetable pathway. These amplifications also provide an important potential lesson regarding interpretation of the TCGA; although some pathways are quite unique to specific subtypes (RHOA dysregulation is an excellent example), others such as PIK3CA mutations and receptor tyrosine kinase amplifications are common to more than one subtype. For this reason, care is required when selecting patients for clinical trials so they are not excluded based on subtype alone: instead the underlying molecular aberration should be characterized.

Targets outside the TCGA subtypes: angiogenesis

The TCGA represents a new paradigm for the classification of GCs. However, it is not definitive and there are several potentially exploitable molecular targets which are not explicit components of this classification system. The most prominent of these is angiogenesis. As increased expression of angiogenic markers such as VEGF is associated with aggressive disease biology and inferior survival in patients with GC, angiogenesis is an attractive therapeutic target in this disease [Maeda et al. 1996; Song et al. 2002; Lieto et al. 2008]. However, despite significant survival benefits in other cancers when bevacizumab (a monoclonal antibody targeting VEGF) was added to standard first-line treatment in two large multicentre randomized phase III trials in GC, it failed to improve survival [Ohtsu et al. 2011; Shen et al. 2015]. The population in both studies was biomarker unselected; however the AVAGAST translational programme identified a survival benefit in patients with higher expression of VEGFA and low neuropilin 1 expression treated in the experimental arm in non-Asian regions [Van Cutsem et al. 2012].

As previously discussed, ramucirumab (a fully humanized monoclonal antibody targeting VEGFR2) recently became the second targeted agent approved for the treatment of GC. Ramucirumab has demonstrated significant benefit in OS, PFS and disease control rate as a single agent compared with placebo in the REGARD trial and in combination with paclitaxel compared with paclitaxel alone in the RAINBOW trial [Fuchs et al. 2014; Wilke et al. 2014] (see Table 3). However, the same success was not apparent when employed in a first-line setting in combination with mFOLFOX, suggesting a potential negative interaction with oxaliplatin-based regimens (or synergy with taxanes) [Yoon et al. 2014]. A further phase III randomized trial of ramucirumab or placebo in combination with cisplatin and capecitabine in a first-line setting in HER2-negative GC has recently commenced [ClinicalTrials.gov identifier: NCT02314117].

Table 3.

Key randomized phase III trial with targeted agents as salvage treatment for gastric cancer.

Trial name	Selected biomarker	Treatment	Number of patients	Primary endpoint	Hazard ratio (95% confidence interval)	p value	Median OS (months)
TyTAN[Satoh et al. 2014b]	HER2 amplification by FISH	PaclitaxelPaclitaxel+lapatinib	129132	OS	0.84 (0.64–1.1)	0.1044	8.911
REGARD[Fuchs et al. 2014]	All comers	BSCBSC+ramucirumab	117238	OS	0.776 (0.603–0.998)	0.047	3.85.2
RAINBOW[Wilke et al. 2014]	All comers	PaclitaxelPaclitaxel+ramucirumab	335330	OS	0.807 (0.678–0.962)	0.017	7.49.6
NCT01512745[Li et al. 2014]	All comers	PlaceboApatinib	92181	OS	0.71 (0.54–0.94)	0.016	4.66.4
GRANITE[Ohtsu et al. 2013]	All comers	BSCBSC+everolimus	217439	OS	0.90 (0.75–1.08)	0.124	4.35.4

BSC, best supportive care; FISH, fluorescence in situ hybridization; HER2, human epidermal growth factor receptor 2; OS, overall survival.

Apatinib (a RTK targeting VEGFR2) was associated with significant survival prolongation compared with placebo (140 versus 195 days, HR 0.71, p < 0.016) in a chemorefractory Chinese population, providing further evidence in supporting VEGFR2 as a valuable target in GC [Li et al. 2013]. However, as development of targeted therapy in GC has been hampered by regional differences in efficacy, it is unknown if these results would be replicated outside Asia. Another multikinase inhibitor targeting VEGFR2 and tyrosine kinase with Ig and EGF (epidermal growth factor) homology domains (TIE2) (regorafenib) has recently demonstrated a significant improvement in PFS compared with placebo (11.1 versus 3.9 weeks, HR 0.40, p < 0.0001) in patients with refractory advanced oesophago GC treated in an international phase II trial [Pavlakis et al. 2015]. Whether the efficacy demonstrated in any of the above trials of antiangiogenic therapy is related to the significant proportion of patients with VEGFA amplification is unknown, however as the other RTK amplifications do appear to confer sensitivity to target inhibition, this may well be true.

Alternatives to the GC TCGA classification system

A recent landmark publication from the Asian Cancer Research Group describing an alternative classification system to the GC TCGA has been thought provoking [Cristescu et al. 2015]. Using gene expression data on 300 Korean GC specimens, four distinct classes of tumour were identified and validated in two independent datasets. The four subgroups defined were MSI (23%), mesenchymal (15%) and microsatellite stable (MSS) p53 (26%) active and inactive (36%) tumours. Although the molecular characterization of this dataset lacks the complexity of the TCGA, it benefits from significantly longer follow up, and this classifier was prognostic and associated with distinct patterns of recurrence. In common with the GC TCGA, the presence of distinct molecular drivers is enriched in separate subgroups (e.g. HER2 amplification in MSS p53 inactive), however there were also differences in the proportion of clinical phenotypes between the two datasets (e.g. diffuse GC 45% Asian Cancer Research Group (ACRG) versus 24% TCGA)which are likely to have a knock on effect on the prevalence of underlying driver mutations or copy number alterations. As the TCGA dataset was derived from a global rather than an Asian dataset, it is used as the basis for this perspective review, however the ACRG classifier may also have significant clinical relevance.

Conclusion

GC remains a highly lethal disease worldwide. Although a large number of new drugs have been investigated, trastuzumab and ramucirumab are the only two targeted agents currently approved for the treatment of advanced disease. One of the many reasons to explain the lack of efficacy of targeted agents in large trials is the absence of biomarker-driven trials or indeed the methodology of biomarker selection. The recent analysis of the GC TCGA project has highlighted the variability of this disease with respect to underlying molecular drivers which is further complicated by intra-patient heterogeneity of biomarker expression [Kim et al. 2011; Smyth et al. 2015]. Although liquid biopsies could represent a possible solution to this complex matter, large-scale results of translational programmes in this field are required before implementing such practice changes.

The large amounts of data derived from molecular characterization panels are hypothesis generating; how to integrate them into clinical practice remains challenging. In this review, we matched clinical trial data reported in the recent literature with the different pathways highlighted in the four molecular subtypes identified by the TCGA analysis. We acknowledge that this is a somewhat artificial approach, as these subtypes do not at the current time define clinical practice or drug development. However, they may aid in the design of future clinical trials aiming to personalize treatment in several ways. First, by helping to identify the driving pathways sustaining tumour growth. Second, by identifying potential combination treatments targeting pathways with nonoverlapping toxicities. And third, by identifying predictable mechanisms of resistance with a view to designing sequential strategies.

A key caveat here is that as targetable molecular characteristics are enriched rather than invariably present in each subtype (with the exception of MSI), knowledge of the subtype alone may be insufficient to allocate a patient to a clinical trial. Instead and still usefully, it may identify a smaller population in which to screen for protocol eligible molecular aberrations. It is also crucial to remind ourselves that the functional implications of each genetic alteration may be context dependent, as demonstrated by the distinct pattern of the oncogenic effect of BRAF mutation in colorectal cancer and melanoma; preclinical validation of the effects of oncogenic driver aberrations may be useful prior to embarking on ambitious clinical trial programmes. Furthermore, these assumptions are made based on the translation of TCGA results into a simple molecular profile which is adapted to routine clinical use and this necessary work has not yet been completed. Finally, with the emergence of ever more refined patient groups, adaptive design and umbrella clinical trials will be required to develop personalized treatments in this highly heterogeneous disease.