Clinical Development of PD-1/PD-L1 Immunotherapy for Gastrointestinal Cancers: Facts and Hopes

Cancers of the gastrointestinal (GI) tract are among the most deadly malignancies, with a high mortality to incidence ratio. Esophagogastric, pancreatic, liver, and colorectal cancers account for more than 2,894,000 deaths per annum (1). Although serial incremental survival benefits have been made with cytotoxic chemotherapy with metastatic esophago-gastric and colorectal cancers, a plateau of achievement has been reached. Applying modern integrative genomic technology, distinct molecular subgroups have been identified in GI cancers. This not only highlighted the heterogeneity in tumors of each primary anatomical site but also identified novel therapeutic targets in distinct molecular subgroups and might improve the yield of clinical success. Molecular characteristics of tumors and their interaction with the tumor microenvironment would further affect development of combination therapy, including immunotherapy.

Currently, immune checkpoint blockade especially targeting programmed death-1 (PD-1) and programmed death ligand (PD-L)1 attracts the most intense research. PD-1 is a cell surface, co-inhibitory receptor expressing in T cells, B cells, monocytes, and natural killer (NK) cells. It has two known ligands—PD-L1 and PD-L2. PD-L1 is upregulated by tumor cells and by cells in the tumor microenvironment. PD-1 interaction with its ligands inhibits T-cell receptor signaling and downregulates T-cell responses. Inhibition of PD-L1 could restore T-cell activity against tumor cells, thereby preventing cancer metastasis and reducing tumor volume (2). This review focuses on the current and future approach of immunotherapy and its interface with recent genomic data from GI cancers.

There have been several large-scale research efforts to ascertain molecular subgrouping for gastric cancer. Notably, The Cancer Genome Atlas (TCGA) identified four subgroups:Epstein–Barr virus (EBV) infected (9%), microsatellite instability (MSI, 22%), genomically stability (GS, 20%), and chromosomal instability (CIN, 50%; ref. 3). The Asian Cancer Research Group (ACRG) also described four subgroups: MSI (22.7%), microsatellite stable (MSS)/epithelial-to-mesenchymal transition (15.3%), MSS/TP53 positive (26.3%), and MSS/TP53 negative (35.7%; ref. 4). However, of note, the four subgroups described by the TCGA did not carry any prognostic effect, although this might be partly due to the tumor samples deriving from operable esophagogastric cancers with limited follow-up (3). Furthermore, several molecular aberrations overlapped between different subgroups, and thus these might not be completely distinct subgroups. For example, PIK3CA mutations were frequently observed in the EBV subgroup but were also found, albeit less frequently, in the MSI, GS, and CIN subgroups. In contrast, the four subgroups identified by the ACRG did have statistically significant survival differences (4). This prognostic difference between TCGA and ACRG was not necessarily related to the limited follow-up of the TCGA. In addition, the semi-supervised analysis used by the ACRG with the incorporation of clinical characteristics might have contributed to this difference (3, 4). Within the EBV-infected and MSI gastric cancer described in the TCGA, there were significantly higher expressions of PD-L1 in both the tumor cells and immune cells compared with other subgroups (5, 6). Furthermore, IFNγ gene set enrichment was also more frequently seen in the EBV-infected and MSI subgroups, although there was no association between IFNγ signature and total number of mutations (5). These subgroups might be particularly sensitive to PD-L1 blockade and of enhanced relevance, especially MSI gastric tumors, which might have a negative prognostic impact when treated with cytotoxic chemotherapy (7).

The initial enthusiasm in targeting PD-L1 in gastric adenocarcinoma came from the results of pembrolizumab in PD-L1–positive gastric cancer in the KEYNOTE-012 study (8). In this study, patients from both Asian and non-Asian countries were enrolled. Forty percent of the screened population were found to be PD-L1 positive in a relatively heavily pretreated patient population. An objective response rate (ORR) of 22% on central review and durable responses were seen with median duration of responses of 40 weeks. The six-month progression-free survival (PFS) rate was 26%, and impressively, the median overall survival (OS) was 11.4 months, with a 12-month OS rate of 42%. Based on these data, several randomized controlled trials (RCT) have been/are being performed. Table 1 shows selected ongoing randomized studies evaluating PD-(L)1 antibodies in GI cancers. In the second-line KEYNOTE-061 RCT, patients were not initially preselected for tumor PD-L1 expression, but PD-L1–positive patients were enriched at a latter part of the study. In the first-line KEYNOTE-062 study, only patients with PD-L1–positive esophago-gastric cancer are being recruited, as this three-arm RCT has a pembrolizumab-alone treatment arm without any cytotoxic chemotherapy. Data from KEYNOTE-012 were on PD-L1–positive gastric cancer alone.

Nivolumab has also been evaluated in a number of studies in gastric cancer. In the CHECKMATE 032 study, both nivolumab monotherapy and the combination of nivolumab plus ipilimumab were tested in gastric cancer patients. The combined PD-L1 and CTLA-4 targeting were first found to be valuable in malignant melanoma (9) but more recently in other tumors such as small cell lung cancer (10). CHECKMATE 032 gastric cohort recruited 160 patients with gastric cancer who were allocated nonrandomly to nivolumab (3 mg/kg) monotherapy (nivo alone; n = 59) and two dose schedules of nivolumab plus ipilimumab—nivolumab 1 mg/kg and ipilimumab 3 mg/kg (nivo 1, ipi 3; n = 49) or nivolumab 3 mg/kg plus ipilimumab 1 mg/kg (nivo 3, ipi 1; n = 52; ref. 11). Similar to KEYNOTE-012, a heavily pretreated patient population was recruited with 79% of patients who had ≥2 prior therapies. However, unlike KETNOTE-012, patients were enrolled irrespective of PD-L1 expression status, and all patients were of Western population. The ORR was 14% (nivo alone), 26% (nivo 1, ipi 3), and 10% (nivo 3, ipi 1). The median duration of response was 7.1 months, 5.6 months, and not reached, respectively. Six-month PFS rates were 18%, 24%, and 9%, and 12-month OS rates were 36%, 34%, and not available, respectively. There was some correlation between ORR and PD-L1 expression for nivolumab alone but less so with the combination of nivolumab and ipilimumab, similar to the observation in malignant melanoma (9, 11).

Most recently, a phase III, placebo-controlled RCT was reported for nivolumab in third- or subsequent-line therapy. The ONO12 (ATTRACTION-2) study recruited patients in Korea, Japan, and China only and thus consisted entirely of Asian population (12). In this large study, 493 patients were randomized in a 2:1 fashion to nivolumab or placebo. Nivolumab resulted in statistically superior OS [hazard ratio (HR), 0.63; 95% confidence interval (CI), 0.50–0.78; P < 0.0001], PFS (HR, 0.60; 95% CI, 0.49–0.75; P < 0.0001), and ORR. Twelve-month OS rates were 26.6% and 10.9%, and ORR rates were 11.2% versus 0% for nivolumab and placebo, respectively. Whereas the ONO-12 and CHECKMATE 032 studies showed similar efficacy for nivo alone in both Asian and Western populations, it has been previously shown that the Asian and non-Asian gastric cancer might exhibit distinct gene signatures related to inflammation and immunity (13). In particular, immune T-cell expression signatures were enriched in non-Asian gastric cancers including both CD28 and CTLA-4 signaling with supportive immunohistochemistry data showing T-cell markers (CD3, CD45R0, and CD8) significantly enriched in Caucasian compared with Asian gastric cancer. The exception was the immunosuppressive T-regulatory cell marker FOXP3, which was significantly enriched in the Asian population. These immune-related differences were however unrelated to EBV infection and mismatch repair (MMR) status.

Avelumab, an anti–PD-L1 antibody, had been evaluated in the phase IB expanded cohort JAVELIN study in two different settings—maintenance after first-line therapy and second-line treatment (14). In the second-line setting, ORR was similar to nivolumab and pembrolizumab. Maintenance setting has so far not been explored by other PD-1 antibodies and forms the current registration strategy for avelumab in gastric cancer (Table 1).

When one interrogated the integrated molecular description of gastric cancer in the TCGA, both JAK2/PD-L1/2 and VEGF-A were altered in the CIN subgroup. Targeting angiogenesis is now an established treatment option in gastric cancer (15, 16). VEGFR2 pathway activation by VEGF-A might suppress antitumor T-cell activation by (i) blocking the maturation of dendritic cells disrupting tumor antigen presentation; (ii) inducing the expression of PD-L1 on dendritic cells; and (iii) enhancing regulatory T cell, which could inactivate antitumor immune cells (17, 18). In addition, preclinical evidence has shown low vascular-normalizing doses of antiangiogenics such as DC101 (murine parent antibody to ramucirumab targeting VEGFR2) reprogrammed the tumor microenvironment from immunosuppressive to immunosupportive and potentiated immunotherapy (19). In contrast, high-dose DC101 might prune tumor blood vessels and promote immunosuppressive tumor microenvironment. Furthermore, there were further data suggesting synergistic inhibitory effect of DC101 with anti–PD-1 antibody in a colon adenocarcinoma murine model (20).

With this background, a phase I study was initiated combining pembrolizumab and ramucirumab in PD-L1 unselected multitumor patient cohorts. Second-/third-line gastric cancer cohorts were examined in two different dose schedules (low-dose or high-dose ramucirumab plus fixed-dose pembrolizumab; ref. 21). A further first-line chemo-naïve cohort was also being explored. The safety profile of ramucirumab combined with pembrolizumab was favorable allowing administration of each drug at full dose. Some antitumor activity was observed in previously treated gastric adenocarcinoma, but data were immature for survival endpoints (21). No results are currently available for the chemo-naïve cohort. Similar to targeting angiogenesis, another combination strategy of targeting the tumor microenvironment with immune-oncology (IO) compounds would be against matrix metalloproteinase (MMP)-9 and PD-L1. A randomized phase II study of nivolumab with or without andecaliximab (GS-5745) is ongoing (Table 1).

There are a number of adaptive-designed, phase II studies either recruiting or being planned to combine different IO agents in gastric cancer. For example, FRACTION (a phase II, Fast Real-time Assessment of Combination Therapies in Immuno-Oncology study in patients with advanced gastric cancer) is currently randomizing nivolumab plus ipilimumab with nivolumab plus BMS-986016 (anti-LAG3 antibody), with further combination IO compounds to be added in future (Table 1).

TCGA also recently reported the genomic characterization of esophageal carcinoma (22). As one would expect, squamous cell carcinoma (SCC) and adenocarcinoma were molecularly distinct. Esophageal adenocarcinoma strongly resembled the CIN variant of gastric adenocarcinoma, although DNA hypermethylation occurred disproportionately in esophageal adenocarcinoma. No esophageal adenocarcinomas were positive for MSI or EBV. In contrast, SCC revealed frequent alterations in cell-cycle regulators with inactivation of CDKN2A and amplification of CCND1. Furthermore, EGFR amplification or mutation was seen in 19% and alterations of PIK3CA, PTEN, or PIK3R1 in 24% of tumors. TCGA divided esophageal SCC into three molecular subtypes: ESCC1 was characterized by alterations in the NRF2 pathway and more closely resembled lung and head and neck SCC; ESCC2 showed higher rates of mutation of NOTCH1 or ZNF750 and greater leukocyte infiltration; and ESCC3 sustained alterations predicted to activate the PI3K pathway. With the success of nivolumab in SCC head and neck (23) and lung (24), ESCC1 might be a subgroup more susceptible to PD-1 targeting.

The first study of targeting PD-1 in SCC esophagus was recently published (25). This was conducted entirely in Japan with 65 patients enrolled, and it was unselected for tumor PD-L1 positivity. An ORR of 17% on central review and median OS of 10.8 months were observed. Interestingly, immune-related ORR was 25%.

Similar to esophagogastric cancer, there have been several large-scale efforts to define molecular subgroups in colorectal cancer. Indeed, a consensus molecular subgroup (CMS) has been proposed (26). CMS1 (MSI immune), CMS2 (canonical), CMS3 (metabolic), and CMS4 (mesenchymal) had different molecular characterization. Of particular interest, CMS1 had MSI and thus hypermutation with immune infiltrate activation. This subgroup had worse survival after relapse. For CMS4, mesenchymal, there was stromal infiltration, TGFβ activation, and angiogenesis. This subgroup had worse relapse-free survival and OS.

The immune landscape of these CMS has also been explored (27). CMS1 and CMS4 had high expression of lymphoid as well as myeloid cell–specific genes, thus exhibiting a strong immune and inflammatory contexture. However, the poor prognostic CMS4 differed from CMS1 with higher expression of endothelial cell and fibroblast genes. In addition, functional relevant immune genes were also upregulated in CMS1 and CMS4. CMS1 exhibited a high expression of genes coding for T-cell–attracting chemokines or involved in the formation of tumor-adjacent tertiary lymphoid structures, which are all associated with good prognosis in colorectal cancer. In contrast, CMS4 exhibited high expression of myeloid chemokine, angiogenic factors, and immune-suppressive molecules (27). CMS2 and CMS3 might potentially be “immune deserts” consisting up to 50% of colorectal cancer cases, whereas CMS1 and CMS4 resembled more of “immune paradise,” although CMS4 also had inflammatory and angiogenic components (Fig. 1).

In the early development of anti–PD-(L)1 antibodies, only marginal, if any, benefit was seen in metastatic colorectal cancer (mCRC). Much more encouraging efficacy was seen in a small subgroup of patients with MMR-deficient mCRC. This subgroup constituted only approximately 4% to 5% of all patients with mCRC. ORR of 62% was reported in 13 patients with MMR-deficient colorectal cancer using pembrolizumab, whereas no response was observed in 25 patients with MMR-proficient colorectal cancer (28). In this study, 85% of patients with MMR deficiency were of Lynch syndrome families. Preliminary PFS and OS were all superior in the MMR-deficient compared with MMR-proficient patients (P = 0.001 and 0.03, respectively) when treated with pembrolizumab (28). In an updated analysis, a total of 53 patients (28 MMR deficient and 25 MMR proficient) were treated. ORR was 50% for MMR-deficient colorectal cancer and 0% for MMR-proficient colorectal cancer, respectively. For MMR-deficient colorectal cancer, the PFS rate was 61% at 24 months, and the OS rate was 66% at 24 months (29). The number of somatic mutations was significantly higher in the MMR-deficient tumors compared with the MMR-proficient tumors, and this correlated with objective response (28). Mutation rate and neoantigen load might contribute to sensitivity to anti–PD-1 antibodies (30, 31), although this might not be a universal phenomenon in GI cancers.

A further study was recently reported evaluating nivolumab ± ipilimumab for MMR-deficient mCRC (32, 33). A larger number of patients were recruited in this study (n = 74), and ORR of 20% was observed for nivolumab monotherapy. Twelve-month PFS and OS rates were 48% and 74%, respectively. The proportion of Lynch syndrome families in this study was 31%. Responses were seen regardless of tumor PD-L1 expression, abundance of PD-L1–expressing tumor-associated immune cells, BRAF mutation, or Lynch syndrome. In addition, there was improvement of quality of life observed after nivolumab, approaching the population norm (32). The combination of nivolumab plus ipilimumab resulted in an ORR of 33.3%. Six-month PFS and OS rates were 66.6% and 85%, respectively, with the combination (33). Similar to the pembrolizumab study, only one response was seen out of 20 patients with MMR-proficient tumors.

In mCRC, the challenge remains how to convert the MSS tumors and immune desert–like CMS2 and CMS3 to be more responsive to immunotherapy. There was preclinical evidence to suggest that MEK inhibition alone could result in intratumoral CD8⁺ effector T-cell accumulation and MHC 1 upregulation. This synergized with an anti–PD-1 agent to promote durable tumor regression (34). With this rationale, a phase I study of cobimetinib (MEK inhibitor) with atezolizumab (PD-L1 antibody) was conducted with an expanded cohort in mCRC (n = 23). An ORR of 17% was observed in all mCRC patients and 20% in a KRAS-mutant colorectal cancer cohort (n = 20; ref. 35). This contrasted with almost 0% response seen in patients with MSS treated with PD-1 antibodies. Based on this, a phase III trial (COTEZO) has completed recruitment for mCRC patients at third- or subsequent-line treatment randomizing to regorafenib (control), atezolizumab, or cobimetinib plus atezolizumab (Table 1).

Aside from PD-1 blockade, other immunotherapy strategies are being actively pursued in mCRC. Bispecific T-cell engager (BiTE) are molecules that recruit and engage T cells through simultaneous binding to the CD3ϵ subunit of the T-cell receptor complex and a tumor surface antigen, which results in T-cell cross-linking (36). One FDA-approved BiTE is blinatumomab, a CD19/CD3 BiTE used for acute lymphoblastic leukemia. For mCRC and GI cancers generally, a carcinoembryonic antigen (CEA) would be a suitable tumor surface antigen. MEDI-565/AMG 211 was a CEA and CD3 bispecific single-chain antibody, and it was found in a preclinical model that it mediated T-cell–directed killing of CEA-positive cells, and this activity was independent of the mutational status of cancer cell lines such as KRAS/BRAF/PI3KCA mutation, loss-of-function mutation in TP53, or PTEN loss (37). This preclinical activity was also seen in patients' CEA-positive colorectal cancer specimens (38). A phase I study was initiated for MEDI-565 for GI adenocarcinoma. Thirty-nine patients were recruited, with the majority having mCRC. The dose-limiting toxicities seen were hypoxia and cytokine release syndrome (39). Unfortunately, no objective response was seen, and it was unclear how many patients had CEA-positive tumors.

Another new CEA T cell–bispecific antibody (TCB) has been developed (RG7802), and this CEA TCB had a bivalent binding to CEA and monovalent binding to the CD3ϵ subunit of the T-cell receptor (36, 40). Preclinically, CEA TCB was found to mediate efficient T-cell–dependent tumor cell lysis by inducing stable cross-linking of multiple T cells to individual tumor cells. It demonstrated efficacy in noninflamed and poorly T-cell–infiltrated tumors. It increased T-cell infiltration in tumors converting noninflamed, PD-L1–negative tumors to highly inflamed, PD-L1–positive tumors, leading to a more inflamed tumor microenvironment (36, 40), which also paved the way to future combination of CEA TCB and PD-(L)1 antibodies. Cergutuzumab amunaleukin (RO6895882) is another strategy of immunocytokine that consists of a variant of IL2 that targets CEA. A combination study of RO6895882 and atezolizumab is being performed in various CEA-expressing solid tumors including colorectal and pancreatic cancers.

Another approach that has been evaluated in mCRC was activating the innate immune response with Toll-like receptor (TLR) agonists (41). TLRs are key components of the innate immune system and are essential for the recognition of pathogen-associated molecular pattern and/or damage-associated molecular pattern (DAMP) molecules. The release of DAMP resulted from a noninfectious inflammatory response. Whereas host inflammatory cells would attempt to destroy malignant cells if this acute inflammatory response was insufficient to fully destroy the developing tumor, a dysregulation of the immune system could occur, resulting in a chronic inflammatory response typified by production of large numbers of certain innate immune cells ultimately promoting the growth and progression of cancer (42, 43). Currently, 10 human TLRs have been identified. TLR-9 is located in the plasmacytoid dendritic cells but also expressed in the majority of innate and adaptive (CD4⁺, CD8⁺, NK T, and γδ T) effector cells and in B cells. During tumor cell death, mitochondrial DNA (mtDNA) and mitochondrial formyl peptides are released, which may act as DAMP and potentially result in the dysregulation of TLR-9.

The first-generation TLR-9 agonist developed was a CpG oligodeoxynucleotides (CpG-ODN) called PF3512676. Unfortunately, when added to standard chemotherapy in non–small cell lung cancer, no survival benefit was seen (44, 45). The second-generation TLR-9 agonist also halted development due to toxicity and lack of efficacy. The next-generation TLR-9 agonist, lefitolimod (MGN 1703), underwent further structural changes that might improve efficacy and safety. Lefitolimod has been found to be much more potent than CpG-ODN, with evidence of immune activation in heavily pretreated patients with solid tumors and mCRC in particular. Both innate and adaptive immune responses were seen in vivo (41). Lefitolimod was tested in a small, randomized, placebo-controlled trial as maintenance therapy. Patients who had completed first-line therapy with oxaliplatin or irinotecan/fluoropyrimidine ± bevacizumab were randomly allocated to lefitolimod or placebo (46). There was a trend toward better PFS with lefitolimod from randomization and statistically significant better PFS from start of induction therapy. In addition, the greatest benefit of lefitolimod appeared to be in patients with relatively low tumor burden. Therefore, IMPALA, the phase III RCT of lefitolimod, is recruiting 540 mCRC patients with at least partial responses to first-line therapy as maintenance therapy. The addition of pembrolizumab to a TLR-4 agonist is also being evaluated in other tumors such as follicular lymphoma. A phase Ib/II study is designed to evaluate intratumoral G100 (TLR-4 agonist) plus local radiation and pembrolizumab versus G100 plus local radiation alone in patients with follicular lymphoma (NCT02501473).

As prognostication, the Immunoscore has gained much recent attention. Taking into account the proportion of cytotoxic and memory T cells as well as their location—tumor center or invasive margin (47, 48)—patients with high Immunoscore had a much more favorable time to recurrence in a recent multinational validation project (49). The challenge is whether Immunoscore could be incorporated into predictive biomarker for checkpoint inhibitors in mCRC.

Integrated analysis of genomic, epigenomic, and transcriptomic characteristics in pancreatic ductal adenocarcinoma (PDAC) has also identified four molecular subtypes: squamous, pancreatic progenitor, aberrantly differentiated endocrine exocrine, and immunogenic (50). Yet checkpoint inhibition in PDAC has been disappointing thus far. The immune environment of PDAC might be particularly hostile. This was characterized by (i) sparse intratumoral cytotoxic CD8⁺ effector T cells, (ii) RAS oncogene driving inflammatory program, (iii) CD3⁺ T-cell sequestering preferentially at tumor margin, and (iv) excessive immunosuppressive leukocytes in the tumor microenvironement (51). In a recent pooled analysis of TCGA and International Cancer Genome Consortium PDAC data (52), it was found that T cells were present but inactive. Robust tumor-infiltrating lymphocytes were present, but the T-cell activation signature was absent. Contrary to other tumors, high mutation load in PDAC was inversely related with T-cell activity (52).

PD-L1–positive PDAC had less favorable prognosis, although expression was sparse (53). In a phase II study of ipilimumab in metastatic pancreatic cancer, no objective response was observed out of 23 patients, although one patient had immune-related response (54). In another study of ipilimumab with or without GVAX, a vaccine based on allogeneic pancreatic tumor cells genetically modified to produce GM-CSF, again, no responses was seen, although a decrease in CA19-9 marker level was observed when GVAX was added (55). Furthermore, in a phase I study of PD-L1 antibody, there were no responses seen out of 14 patients with PDAC (56). In MMR-deficient pancreatic cancer, objective responses have been seen with pembrolizumab, but patient number was extremely limited (57).

The largest study evaluating an immunotherapy approach in advanced PDAC was the TELOVAC study (58). GV1001, a human telomerase reverse transcriptase catalytic subunit class II peptide vaccine, was given either sequentially after chemotherapy or concomitantly with chemotherapy. Gemcitabine plus capecitabine was given both as control arm and as chemotherapy with GV1001 in the experimental arms. TELOVAC was probably the largest RCT ever conducted in advanced PDAC, with 1,002 patients randomized to these three arms. Unfortunately, no survival benefit was seen with this immunotherapeutic approach (58).

The immune desert of pancreatic cancer represented a major therapeutic challenge. Interestingly preclinically, similar to mCRC, there was evidence that combining MEK and PD-1 inhibition exhibited greater inhibitory effect on tumor growth compared with PD-1 inhibition or MEK inhibition alone (59). Myeloid cells protected pancreatic tumor cell viability by blocking CD8⁺ T-cell–mediated antitumor response. This was achieved by activating PD-1/PD-L1 checkpoint through EGFR/MAPK signaling. Depletion of myeloid cells in the microenvironment arrested tumor growth or induced tumor regression. An increased level of immunosuppressive leukocytes and desmoplastic stroma, which formed a barrier to T-cell infiltration, represented critical obstacles to immunotherapy in PDAC. Focal adhesion kinase (FAK) might be important for regulating fibrotic PDAC tumor microenvironment. FAK inhibition (FAKi) induced tumor stabilization, decreased fibrosis without accelerating tumor progression, and decreased immunosuppressive cell population in tumors (60). FAKi improved survival the most with FAKi plus gemcitabine plus anti–PD-1/anti–CTLA-4 therapy in mice bearing transplantable KRAS-mutated pancreatic tumors (60).

Primary biliary tract cancers (BTC) consists of cholangiocarcinoma–intrahepatic (ICC) and extrahepatic as well as gallbladder cancer. They arise from malignant transformation of biliary epithelium, typically occurring in the setting of chronic inflammation (61). Expression of PD-1 and PD-L1 is upregulated in ICC tumor tissues (62). Increased expression of PD-L1 was associated with poor differentiation and stage of ICC, whereas increased CD8⁺ T cells in tumors was associated with better tumor differentiation.

In a series of resected ICC, the majority of tumors expressed PD-L1 on tumor cells located either within the tumor front or on tissue-associated macrophages (63). PD-L1 expression was found to be associated with nodal metastasis and larger number of lesions. PD-L1 expression within the tumor front was associated with worse survival, suggesting the PD-1 pathway might be suppressing the host immune response in ICC (63). In another study, all resected ICC expressed PD-1, which was present on tumor-infiltrating lymphocytes but was not detected on ICC cells (64). Although immune infiltrate was present in all tumors analyzed, the proportion of CD8⁺ T cells was significantly higher in the fibrous septa compared with tumor lobules (64). ICC tumors with downregulation of HLA class I antigen expression presented with more advanced stage. Therefore, this microenvironment might potentially be attractive for a PD-(L)1 antibody therapy.

There is paucity of clinical data in the use of anti–PD-(L)1 antibody in advanced BTC. Pembrolizumab was tested in the multicohort, phase IB KEYNOTE–028 (65). Only patients with PD-L1–positive tumors were recruited. Of the screened advanced BTC population, 42% were PD-L1 positive. Twenty-four patients were recruited, and an ORR of 17% was observed (65). In the aforementioned phase I study evaluating pembrolizumab plus ramucirumab in patients with gastric cancer, there was a separate cohort of patients with advanced BTC that has completed recruitment, but no results are available yet.

Hepatocellular carcinoma has a distinct immune environment that might represent a potential immune paradise for checkpoint inhibition. However, a separate review article will focus on this particular subject.

There are still much data with immunotherapy in GI cancers not discussed in this review—active immunotherapy with dendritic cell vaccine and viral vector vaccine, and passive immunotherapy such as chimeric antigen receptor T-cell therapy and other checkpoint modulators such as LAG3, OXO40, KIR, and TIM-3. Combining radiotherapy (RT) with PD-(L)1 antibodies to enhance antitumor T-cell response and augment abscopal effect is also being actively pursued in GI cancers. This abscopal effect, where a nonirradiated site regresses after RT to the primary tumor, might stem from an immune-related mechanism, and synergistic effect of RT and PD-(L)1 antibodies has been observed in preclinical models. The immune landscape of GI cancer is wide-ranging, with immune paradise and immune desert. Integrating information collected from genomic analysis and the immune microenvironment would hopefully turn these into an immune oasis and provide greater opportunities in immunotherapy for the greater benefits of our patients.

I. Chau is a consultant/advisory board member for Bayer, Bristol-Myers Squibb, Eli Lilly, Five Prime Therapeutics, MSD, Roche, and Sanofi Oncology, reports receiving commercial research grants from Janssen-Cilag, Merck-Serono, Novartis, and Sanofi Oncology, and reports receiving honoraria from Amgen, Eli Lilly, Gilead Science, Pfizer, and Taiho.

I. Chau would like to acknowledge National Health Service funding to the National Institute for Health Research Biomedical Research Centre at the Royal Marsden NHS Foundation Trust and The Institute of Cancer Research. I. Chau would also like to acknowledge the receipt of the Swedish Society of Gastrointestinal Oncology (Gastrointestinal Onkologisk Förening) Second Bengt Glimelius' Award for Outstanding Research in GI cancer 2016.