In the last few years, immunotherapy has dramatically changed the treatment landscape of several types of solid tumors, such as melanoma and non-small cell lung cancer (NSCLC).1 Immune checkpoint inhibitors (ICI) targeting programmed cell death-1 (PD-1) and programmed cell death ligand-1 (PD-L1) axes are the most widely investigated drugs and the remarkable results obtained with these compounds have led to several US Food and Drug Administration (FDA) and European Medicines Agency (EMA) approvals across different cancer types and numerous indications.2 3 However, identifying tumors which will respond to immunotherapy has remained a key challenge.4

Overall, gastrointestinal (GI) neoplasia and more specifically colorectal cancers (CRC) have been always considered as ‘cold’ tumors, therefore, the vast majority of these patients do not benefit from an immunotherapy approach.5 To date, the only validated predictive biomarkers of response to ICIs used in clinical practice across all solid tumors, including GI cancers, is microsatellite instability high (MSI-H) or DNA mismatch repair deficiency (dMMR). In fact, in 2017 pembrolizumab (Keytruda, Merck)—an anti-PD-1 antibody—was FDA approved for the treatment of metastatic cancers that are characterized by MSI-H status or dMMR, regardless the site of tumor origin. In addition, few months later the FDA granted accelerated approval to nivolumab (Opdivo, Bristol-Myers Squibb) for the treatment of patients with dMMR and/or MSI-H metastatic CRC (mCRC) that has progressed following treatment with a fluoropyrimidine, oxaliplatin and irinotecan. Efficacy of PD-1-blockade in MSI-H tumors for patients with CRC, showing an objective response rate (ORR) of 36%, is similar to the efficacy in different cancer type (ORR 46% across 14 other cancer types). However, to date, there is no approved PD-1-blocking antibody for MSI-H cancers in Europe.6 Other predictive biomarkers to better select patients for immunotherapy are being currently studied, such as PD-L1,7–11 tumor mutational burden (TMB)12–14 or different combinations of these,15–17 though showing conflicting results in different tumor types, including GI cancers.

In this review article, we provide an overview of the role of immunotherapy biomarkers in these patients and the main clinical trials with ICI (i.e., anti PD-1/PD-L1) in CRC (table 1), esophageal and gastric (table 2) cancers, and well as other GI cancers. We further discuss the state-of-the-art of the known mechanisms of resistance to ICIs as well as the several methods to overcome resistance which are under investigation, with specific focus on GI cancers. Finally, we provide a perspective on possible future therapeutic approaches that may be implemented in clinical practice.

PD-L1, also called CD274 or B7 homolog 1, has been identified to be one of the binding and functional partners of PD-1, a cell surface coinhibitory receptor that induces immune inhibition and promotes tumor immune escape from cytotoxic T cells.18 19 Furthermore, PD-L1 has been also shown to collaborate with CD80, a functional ligand for cytotoxic T lymphocyte antigen-4 (CTLA-4), to mediate an immune-inhibitory signal. PD-1 and CTLA-4 mediated modulation of immune checkpoints is critical to regulate the duration and amplitude of immune responses and lead to the development of an exhausted T cell phenotype. PD-1 is expressed on a large proportion of tumor-infiltrating lymphocytes (TILs) from many cancer types, whereas PD-L1 is expressed in 20%–50% of human tumors.20 Hence, immune checkpoint blockade strategies targeting PD-1/PD-L1 have been developed to reduce the effect of tumor immune escape and rescue the cytotoxic cell-induced immune response. The role of PD-L1 as a biomarker of immunotherapy efficacy has been evaluated in several studies across different cancer types, showing conflicting results depending on tumor types21 22 (further discussed in the next paragraphs).

Initial results by Le et al, first published in 2015, reported for the first time the efficacy of PD-1 blockade in MSI-H tumors.23 MMR is a highly conserved DNA repair mechanism which is critical to recognize and correct erroneous nucleotide insertion, deletion and mis-incorporation during DNA replication and recombination, thus ensuring genomic integrity. dMMR determines the accumulation of DNA replication errors which translates into high frequency of frameshift mutations in microsatellite DNA (MSI), leading to a high somatic mutational burden (mutator phenotype).24 25 The role of MSI as a predictive biomarker is now well established, particularly in CRC.26 The prevalence of MSI in CRC varies depending on the tumor stage. The incidence is higher in early stages: about 20% in stage I–II and 12% in stage III, and lower in the metastatic setting: 4%–5%. About one-quarter of dMMR CRCs are due to Lynch syndrome, an autosomal dominant hereditary disease characterized by germline mutations in the MMR genes (ie, MLH1, MSH2, MSH6, PMS2 or EPCAM deletions leading to epigenetic inactivation of MSH2),27 while the remaining cases are defined as ‘sporadic’. The majority (80%–90%) of sporadic cases result from epigenetic silencing of the MLH1 gene promoter by hypermethylation,28–30 a phenomenon associated with a high CpG island methylation phenotype.31 32 The concomitant presence of BRAF V600E mutation can be identified in about 30% of dMMR CRC, limited to sporadic MSI.31 The MSI-H phenotype is characterized by distinct clinical and pathological features compared with those observed in microsatellite stable (MSS) CRC, such as prominent lymphocytic infiltrate, mucinous histology and poor differentiation, and right-sided colon location.33 MSI/dMMR testing is recommended by current international guidelines to assess the eligibility to treatment with ICI in mCRC and other metastatic GI cancers.

An emerging biomarker of response to anti-PD1/PDL1 therapies is the TMB34 35 which quantifies the number of somatic mutations in the tumor. However, tumors containing high mutational burden may exhibit variable responses suggesting that additional factors may contribute to antiPD1/PDL1 response. Lee and Ruppin36 evaluate systematically 36 different variables associated to anti-PD1/PDL1 response of 3 distinct classes: (1) tumor neoantigens, (2) tumor microenvironment and (3) checkpoint target. This analysis of multiomics data from the Cancer Genome Atlas cohort and ORRs to therapy data across 21 cancer types shows that estimated CD8 +T cell abundance is the most predictive biomarker, followed by TMB and the fraction of samples with high PD-1 gene expression. In a recent study within a large cohort of GI cancer, authors aimed to determine TMB, MSI-H and PD-L1 expression interrelationship in GI cancers.17 They found that the TMB-high rate varied widely among GI cancers. Although MSI-H is conceivably the main driver for TMB-high, other factors may be involved and higher PD-L1 expression was more likely to be seen in MSI-H compared with MSS tumors (20.6% vs 7.8%, p<0.0001).

On the other hand, research efforts are underway to identify biomarkers associated with resistance to ICI. The MDM2 proto-oncogene encodes a nuclear localized E3 ubiquitin ligase with the core function of inhibiting the tumor suppressor p53. MDM2 amplification has been reported in multiple tumor types and is a hallmark of tumorigenesis.37 Recently MDM2 amplification also has been implicated as a potential marker for accelerated tumor growth after checkpoint inhibitors treatment, a phenomenon known as hyperprogression, affecting approximately 9% of patients who receive PD-1/PD-L1 inhibitors.38 39 To date, hyperprogression after anti-PD-1/PD-L1 agents has been reported by at least four groups, however, the mechanisms that mediate this phenomenon remain unclear and the only markers that have been shown to correlate with this occurrence are MDM2 family gene amplifications and epidermal growth factor receptor (EGFR) alterations.40

The role of selected biomarkers according to different cancer types will be further addressed in the next paragraphs.

The immune anticancer response is modulated by a complex and dynamic interplay between tumor cells, the host immune system, and the associated microenvironment. ICI have been described to be more active in ‘hot’ tumors, characterized by a prominent lymphoid infiltration of the stroma and high density of TILs coupled with an inflammatory cytokine-rich microenvironment and a T cell-inflamed gene expression profile90 (figure 1). These features suggest the presence of a pre-existing immune response, which the tumor was able to escape employing several different mechanisms including: checkpoints activation leading to T-cell exhaustion; activation of oncogenic pathways; and interaction with the tumor microenvironment to promote an immunosuppressive status.91 An important characteristic of ‘hot’ tumors is the high TMB, which reflects an increased number of nonsynonymous, single nucleotide variants that may lead to an increased neoantigen production promoting the antitumor immune response. Indeed, cancer types harboring high somatic mutational burdens (eg, melanoma, NSCLC and bladder cancer) are among those with the highest response rates to ICI.14 34 92 Genetic instability due to alterations in DNA repair and replication genes can increase immunogenicity through high mutational burden with subsequent neoantigen formation.93 Patients with alterations in BRCA and additional DNA damage response genes, including ATM, POLE, FANCA, ERCC2 and MSH6, have recently shown correlation with high TMB and improved clinical outcomes to immune checkpoints inhibitors.94 Accordingly, dMMR tumors, which display exceptionally high numbers of somatic mutations, show an enhanced response to PD-1 blockade.23 44

On the other hand, cancers with low immune infiltrates are considered ‘cold’ tumors (e.g., pancreas, ovarian and prostate). These cancers are characterized by a tumor microenvironment that is enriched in myeloid-derived suppressor cells (MDSC) and regulatory suppressive T cells (T-regs), leading to a lack of immune response activation and TILs infiltration in the tumor (‘immune desert’).

A deeper understanding of the mechanisms behind ‘hot’ and ‘cold’ immune tumors is critical to overcome primary, adaptive and acquired resistance and to be able to boost a weak antitumor immunity.95 96 In fact, despite the encouraging success of ICI, roughly 60%–70% of advanced tumors are not responsive to single-agent treatment (primary resistance), and often those that initially respond become resistant to treatment over time (adaptive and acquired resistance).97 Extensive research is currently underway to develop novel strategies to overcome resistance to immunotherapies, which include fundamental efforts to transform immunologically ‘cold’ into ‘hot’ tumors.

Many mechanisms of resistance have been characterised to date, and more continue to be uncovered. By elucidating and targeting mechanisms of resistance, treatments can be tailored to improve clinical outcomes.93 Tumor-intrinsic mechanisms of immune evasion comprise genetic and epigenetic alterations (immunoediting) that influence the antitumor immune response, including loss of tumor antigen expression, loss of human leukocyte antigen) complex expression and alterations in the antigen processing machinery. For instance, loss of beta-2-microglobulin expression (due to deletions, point mutations, or loss of heterozygosity) results in impaired cell surface expression of major histocompatibility class I, which in turn impairs antigen presentation to cytotoxic T cells,98 thereby inducing resistance to PD-1 and CTLA-4 blockade.99 In addition, alterations in several key oncogenic signaling pathways, such as MAPK, PI3K, and WNT, can also lead to immune resistance.100 Loss of PTEN, which enhances PI3K signaling, has been found to be associated with resistance to immune checkpoint therapy.101 Constitutive PD-L1 expression and alteration of interferon (IFN) signaling are also considered mechanisms of resistance to immunotherapy.102–104 Finally, there is now emerging evidence that chromatin remodeling is involved in sensitivity and resistance to ICI, through mutations in SWI/SNF (SWItch/Sucrose Non-Fermentable) complexes, such as loss of ARID1A or inactivation of PBAF subunits.105 106

On the other hand, tumor-extrinsic mechanisms also participate in the establishment of immune resistance through processes driven by stromal and immune elements of the tumor microenvironment. Mounting evidence supports the role of the host microbiome in response to cancer treatment, with several recent studies demonstrating the influence of the gut microbiota on the response to immune checkpoint blockade across cancer types.107 108 Other tumor-extrinsic mechanisms include changes in immune checkpoint expression (eg, CTLA-4 and PD-1), T-cell exhaustion and phenotype change, recruitment of immune suppressive cell populations (eg, T-regs, MDSC and type II macrophages), and cytokine or metabolite release in the tumor microenvironment (eg, colony stimulating factor (CSF)-1, tryptophan metabolites, transforming growth factor-β and adenosine).109

Several approaches have been proposed and are currently under evaluation in the attempt to overcome the mechanisms of resistance to ICI. Combination strategies using multiple treatment modalities are emerging to surmount the lack of a pre-existing immune response, eventually turning cold tumors into hot tumors. A strong clinical and preclinical rationale support the use of molecularly targeted therapy in combination with immunotherapy. The most extensively studied cancer types treated with this strategy are melanoma and NSCLC, but, theoretically, it could apply to any cancer.110 This approach relies on targeted therapies either (1) inducing a rapid tumor death leading to the release of tumor-associated antigens or (2) modulating key cellular pathways that allow the cancer to maintain an adaptive resistance. For instance, there is accumulating evidence that bevacizumab, an anti-VEGF monoclonal antibody, has immunomodulatory properties, as VEGF exerts immunosuppressive functions through inhibition of dendritic cell maturation, reduction of T-cell tumor infiltration, and promotion of inhibitory cells in the tumor microenvironment. Hence, there is compelling rationale for the combination of ICI with anti-VEGF therapies; several clinical trials are evaluating such combinations across tumor types.111 Similar early favorable results have been reported with anti-BRAF, anti-MEK, anti-EGFR agents, among others. Less commonly targeted pathways, among those that could be involved in immunotherapy resistance, include PI3K-AKT-mTOR, hypoxia-inducible factor, adenosine, JAK/STAT and Wnt/β-catenin. Epigenetic modifications of cancer DNA are known to cause changes in immune-related gene expression, which can impact antigen processing and presentation as well as promote immune evasion. Therefore, demethylating agents may enable re-expression of immune-related genes, with the potential for therapeutic impact, especially in the setting of combination treatment with immunotherapy.112 113

Strategies to promote formation or presentation of suitable neoantigens in tumors with a non-inflamed, non-immune cell infiltrated tumor microenvironment are also being exploited through the use of chemotherapy and radiation (by inducing immunogenic cell death) or by stimulating innate immune responses and dendritic cell function (eg, type I IFN, Toll-like receptor ligands, LIGHT and oncolytic viruses).114 Over the two past decades, a growing body of preclinical and clinical evidence have supported the immunomodulatory effects of radiotherapy.115 116 In fact, the delivery of radiotherapy and the resulting induction of immunogenic cell death pathways can potentially convert the tumor into an in situ (and thus personalized) vaccine. This phenomenon of radiation shrinking the tumor locally while inducing an immune response systemically is known as the ‘abscopal’ effect. Several clinical trials are investigating the efficacy of immunotherapy in combination with radiation therapy across different tumor types, especially in the metastatic setting. Although many questions remain unresolved (eg, sequencing, timing and radiation fractionation) and data analysis is ongoing for this approach that is already being tested in the clinic, immunotherapy combination with radiation therapy may become a novel and effective approach to treat patients with cancer in the near future.

Cancer vaccines have shown promising results as a means of personalizing cancer immunotherapy and potentially enhancing immune memory in a minority of patients, such as NY-ESO-1 cancer vaccines.117 118 True success in this realm may be achieved with the identification of a pan-cancer antigen that can be targeted through vaccination.

Novel combined approaches to enhance immunotherapy exploits the combination of ICI with new negative checkpoint regulators (NCRs), for example, anti-lymphocyte activation gene 3, a coinhibitory receptor on T cells that, among other functions, enhances activity of regulatory T (Treg) cells and regulates T cell proliferation, differentiation and effector function. Other likely targets include additional coinhibitory receptors such as TIM3 (a marker for exhausted T cells); T cell immunoglobulin and ITIM domain (TIGIT, which counterbalances the costimulatory function of CD226, that is rapidly induced following T cell activation); B and T lymphocyte attenuator (also known as CD272), which is expressed by T cells and synergizes with herpesvirus entry mediator (also known as TNFRSF14), expressed on antigen-presenting cells; sialic acid-binding Ig-like lectin 9, which is upregulated in TILs and possibly determines a subclass of tumor-specific CD8 +TILs; and V-domain Ig suppressor of T cell activation (VISTA).119 VISTA is one of the most recently NCRs and can act as both a ligand and a receptor on T cells to inhibit T cell effector function and maintain peripheral tolerance.120 VISTA is produced at high levels in TILs, such as MDSCs and regulatory T cells, and its blockade with an antibody results in delayed tumor growth in mouse models of melanoma121 and SCC.122 A recent study gives a detailed analysis of immune infiltration in primary and metastatic pancreatic tumors compared with melanoma. These data indicate that human pancreatic tumors express CD68 +macrophages and highlight the inhibitory checkpoint molecule VISTA as a potential immunotherapeutic pancreatic cancer.123

Another strategy is the combination of ICI with costimulatory checkpoint molecules, such as anti-ICOS (inducible T cell co-stimulator), OX40 antigen (also known as TNFRSF4 (or CD134)), TNFRSF7 (also known as CD27), CD28, TNFRSF9 (also known as 4-1BB ligand receptor or CD137) and glucocorticoid-induced TNFR-related protein, all of which enhance T cell expansion and effector functions while controlling Treg cell suppressive functions.124 However, despite promising results, the use of costimulatory molecules could be clinically limited by systemic toxicity.

A proposed approach to turn cold tumors into hot tumors is to combine a priming therapy that enhances T cell responses (such as vaccines, adoptive T cell transfer (ACT), oncolytic viruses) with ICI. At present, only one oncolytic virus, talimogene laherparepvec, a genetically modified herpes simplex-1 virus expressing granulocyte-macrophage CSF, has been approved by the FDA for the local treatment of patients with unresectable melanoma. Local administration of therapeutic (oncolytic) viruses, that have a tropism for malignant cells, causes cancer cell lysis and/or death and can induce abscopal effects with reductions in the volume of non-injected lesions.125 Growing preclinical evidence indicates that non-virally induced oncolysis, defined as chemical or physical treatment (electrochemotherapy, locally injected cytotoxins, photodynamic therapy, laser therapy, microwave, radiofrequency or photothermal ablation, high-intensity focused ultrasonography and cryotherapy) administered locally to destroy malignant lesions, can promote a similar effect owing to the release of danger-associated molecular patterns that lead to the recruitment of immune cells, thus inducing a systemic response against tumor antigens that protects against local disease relapse and also mediates distant antineoplastic effects. This abscopal effect is a direct consequence of the immune response triggered by oncolysis and can be supported by localized or systemic immunotherapy.126

Among passive immunization strategies, ACT is becoming an increasingly important component of the antitumor arsenal: examples include TILs, transgenic T cell receptor and chimeric antigen receptor (CAR) T cells—which have been approved by the FDA for the treatment of B cell acute lymphoblastic leukemia and B cell non-Hodgkin's lymphoma.127 128 Several researchers are investigating the use of CAR T-cell therapy for the treatment of solid tumors. These studies require the prior identification of new antigens and the development of preclinical models of solid tumor in which to characterize therapies based on these antigens. Various types of CAR-T cells have been engineered to target HER2, EGFR and CEA, which are overexpressed on the surface of cells in GI tumors and are associated with the development and metastasis of tumors.129

Recent developments in the fields of nanotechnology and bioengineering afford new approaches that can dramatically improve not only the safety but also the efficacy of cancer immunotherapy by modifying their spatiotemporal release profiles.130

There is growing evidence that the gut microbiome plays an important role in cancer treatment. According to experimental studies several bacteria are involved in CRC carcinogenesis, including Fusobacterium nucleatum and certain strains of Escherichia coli and Bacteroides fragilis. The gut microbiota interacts with host cells to regulate many physiological processes, such as metabolism and immune response.131 The gut microbiome was initially found to modulate the response to immune checkpoint blockade in mouse models. Mice with different microbes had significant differences in response to treatment.132 133 Many of these findings have been verified in patients with metastatic melanoma receiving ICI. Several studies demonstrate the significant association between commensal microbial composition and clinical response to anti-PD-1-based immunotherapy.134 Specific bacteria are positively correlated with immunotherapeutic response, including Akkermansia muciniphila,135 Ruminococcus champanellensis, B. fragilis, Bifidobacterium, Eubacterium limosum, Faecalibacterium spp and Alistipes shahii.131 Analysis of the gut microbiota and of the immunological profiling in the tumor microenvironment shows that the expression of cytotoxic T cell markers and antigen processing and presentation are augmented in patients with favorable gut microbiota compared with patients with unfavorable gut microbiota.136 Another study has identified a consortium of 11 bacterial strains that could enhance the action of ICI to inhibit tumor growth in a subcutaneous tumor model derived from a mouse colon adenocarcinoma cell line.137 These findings have led to the possibility of detecting these core bacteria as predictive biomarkers for immunotherapy response.

Furthermore, the gut microbiota can also be associated with the adverse effects of immunotherapy, such as ICI-associated colitis.132 Increased representation of the Bacteroidetes phylum correlates with resistance to ICI-associated colitis, whereas a paucity of genes in polyamine transport and vitamin B biosynthesis pathways are associated with increased susceptibility to colitis.138 A report of two human cases has demonstrated successful treatment of refractory immunotherapy-associated colitis by fecal microbiota transplantation (FMT), with gut microbial changes correlating with complete resolution of colitis up to 53 days after one dose and up to 78 days after two doses of FMT, respectively. This study reports for the first time that modifying gut microbiota by fecal transplantation can ameliorate refractory colitis as an adverse effect of immunotherapy.139

An early phase 1 study (NCT04130763) has been designed to improve the response rate of anti-PD-1 among patients with anti-PD-1 resistant/refractory digestive system cancers (including the esophagus, stomach and intestine) through the intervention on their gut microbiota. Specifically, the study will identify healthy subjects that have the intestine similar to patients with GI cancers who has responded to anti-PD-1 therapy, then extract the gut microbiota of these healthy subjects to product FMT capsule, and perform a rechallenge with anti-PD-1 immunotherapy combined with FMT for GI system patients with cancer whose anti-PD-1 treatment failed.140

In the last few years, we witnessed the extraordinary impact that immunotherapy has generated for some patients with selected cancers. Nonetheless, to date, the vast majority of GI cancers do not benefit from this treatment approach, due to either primary or acquired resistance to immunotherapy. Several combination approaches are under investigation to develop novel treatment strategies to enhance immunotherapy efficacy and to expand the available treatment options for patients with GI cancer.