The Next Decade of Immune Checkpoint Therapy

Metastatic cancers remain incurable in most patients as conventional therapies that target tumor cells usually cannot provide a durable response. The discovery of immune checkpoints that regulate immune responses transformed cancer care, offering long-term clinical responses and the possibility of cure in many more patients with metastatic cancer (1). In 2011, the FDA approved the first immune checkpoint targeting agent, ipilimumab—a mAb targeting CTLA4—which opened the field of immune checkpoint therapy (ICT). Subsequently, mAbs blocking other checkpoints such as PD-1 or PD-L1 have received FDA approvals to treat a number of tumor types alone and in combination with other agents.

Despite the unprecedented durable clinical responses observed in subsets of patients, most patients do not respond, and some patients develop resistance to therapy after initial response. Furthermore, ICTs can result in life-threatening toxicities, known as immune-related adverse events (irAE). Therefore, research studies are ongoing to understand pathways of resistance to ICT and mechanisms of irAEs. In this review, we outline the fundamentals of ICT and the current state of ICT in the clinic. We propose that development of effective ICT combinations will require a reverse translational approach, in which hypotheses are generated from in-depth immune monitoring studies of patients' samples and then tested in appropriate preclinical models, thereby yielding important insights to guide rational therapeutic strategies in subsequent clinical trials. The last ten years provided an understanding of how T cells can be targeted with anti-CTLA4 and anti–PD-1/PD-L1 to provide clinical benefit. Bolstered by the latest technologies, in the next decade we will better understand how to combine ICT with agents targeting multiple other cell subsets and pathways in the innate and adaptive immune system. The future of ICT offers ongoing promise to improve outcomes in patients with cancer.

T cells are the soldiers of the immune response. Activation of T cells requires a coordinated and stepwise process involving two primary signals (Fig. 1; ref. 2). T cells express the T-cell receptor (TCR), which signals via the CD3 complex upon the interaction of the TCR with MHC plus its cognate peptide antigen on antigen-presenting cells (APC). T cells also constitutively express a costimulatory receptor, CD28, which binds to the B7 family of costimulatory molecules that are predominantly expressed on professional APCs. To activate or turn T cells “on,” both the TCR signal via CD3 (signal 1) and the CD28 costimulatory signal via B7 interactions (signal 2) are required (Fig. 1A). Interaction of APCs with T cells that bear the specific TCR for a given antigen leads to activation of T cells and T cell–mediated responses. T cells then differentiate and clonally expand, with subsequent formation of memory T cells (3).

CD4 and CD8 T cells constitute the majority of T cells in the immune system. CD4 T cells differentiate into a variety of subtypes: Th1, Th2, Th17, follicular Th (Tfh), T regulatory (Treg), and Th9 cells. These subsets have distinct transcription factors that regulate their function and cytokine profiles, which affect immune responses that range from effector cell responses for tumor rejection to regulatory cell responses to suppress effector cells. These transcription factors include T-bet for Th1 cells, BCL6 for Tfh, RORγ for Th17 cells, and FOXP3 for Tregs. Although CD4 T cells are known as helper cells, accumulating evidence suggests a cytotoxic function demonstrated by certain CD4 T cells (4). CD8 T cells are typically regarded as a uniform population with cytotoxic function; however, recent studies have highlighted the diversity of the CD8 T-cell population (5–9).

In general, Th1 CD4 cells and CD8 T cells are the effector cells responsible for driving antitumor immune responses. However, these immune responses are controlled by multiple mechanisms. Upon T-cell activation via the TCR and CD28, T cells proliferate and produce cytokines, but also upregulate inhibitory molecules that attenuate T-cell activation (Fig. 1B). Therefore, T cells have limited time to function before they are reined in to prevent damage to normal cells. Multiple inhibitory mechanisms exist, both cell-intrinsic and cell-extrinsic, to control T-cell responses. The signals that control antitumor immunity and immune suppression are comprised of a tightly regulated process that forms the yin and yang of the immune response.

CTLA4 is a canonical inhibitory molecule that is expressed immediately following engagement of TCR, with peak expression around 48 to 72 hours following T-cell activation. It bears structural homology to CD28 and binds to B7-1 (CD80) and B7-2 (CD86) molecules on APCs with higher affinity than CD28, leading to competitive inhibition of costimulatory CD28 signaling, thereby dampening T-cell signaling (10, 11). In early studies, an anti-CTLA4 antibody prevented CTLA4 from engaging B7, leading to prolonged T-cell responses and eradication of cancer cells (1). Similarly, the in vivo administration of anti-CTLA4 antibodies enhanced the antitumor effect of tumor-infiltrating T cells, leading to regression of established tumors and long-lived immunity in tumor models (1).

Antibody blockade of CTLA4 was proposed as a way to enhance T-cell responses in the setting of cancer. By removing the brakes from T cells, it was hypothesized that T cells would have a longer time to function, which would enable tumor rejection. Anti-CTLA4 opened an entire new field termed “ICT.” The success of anti-CTLA4 led to the identification of other pathways that regulate T-cell responses, including both costimulatory and coinhibitory pathways (Fig. 1C; refs. 1, 12).

The primary biological function of PD-1 is to maintain peripheral tolerance. PD-1 engages with its ligands PD-L1 and PD-L2, widely expressed on nonlymphoid tissues, primarily dampening T-cell activation in the periphery. PD-1 interacts with an SHP2 tyrosine phosphatase domain, leading to dephosphorylation of signaling molecules downstream of the TCR and inhibiting TCR-mediated activation of IL2 production and T-cell proliferation (13). Therefore, whereas CTLA4 inhibits signal 2 (CD28-B7), PD-1/PD-L1 inhibits signal 1 (TCR).

PD-L1 expression has been described on a variety of tumor types, and is directly correlated with poor prognosis in several cancers (14, 15). Preclinical models demonstrated that PD-1 blockade results in tumor rejection through reinvigoration of exhausted CD8 T cells that express PD-1, LAG3, and TIM3 and expansion of CD8 T cells, a hallmark of response to anti–PD-1 therapy (16).

Although both CTLA4 and PD-1 are inhibitory molecules that attenuate T-cell activation, they have different mechanisms of action, and inhibition of CTLA4 and PD-1 leads to distinct immune responses. Anti-CTLA4 primarily functions in T-cell priming and expands clonal diversity. Anti-CTLA4 predominantly affects CD4 T cells, with multiple studies identifying increased frequency of a Th1 subset of CD4 T cells that express inducible costimulator (ICOS) as a result of anti-CTLA4 therapy (17–19). Increased frequency of ICOS⁺ CD4 T cells in peripheral blood acts as a pharmacodynamic biomarker of anti-CTLA4 therapy (20), whereas increased frequency of ICOS⁺ CD4 T cells in tumor tissues, possibly in the context of tertiary lymphoid structures, tends to correlate with improved clinical outcomes with ICT (21). Anti-CTLA4 can also promote T-cell trafficking into immunologically “cold” tumors. Investigation into the mechanisms of antitumor activity of CTLA4 blockade demonstrated the contribution of both the effector T-cell and Treg compartments in preclinical models (22, 23). Of note, although CTLA4 blockade was shown to cause selective depletion of intratumoral Tregs through antibody-dependent cell-mediated cytotoxicity (ADCC) in murine models and ex vivo experiments on human tumor samples (24–26), subsequent studies comparing pretreatment and posttreatment tumor samples from patients who received anti-CTLA4 therapy, either as tremelimumab (IgG2 antibody) or as ipilimumab (IgG1 antibody), did not observe a decrease in intratumoral Tregs after therapy (27). It is possible that the expression of specific Fc receptors on APCs or other cell types may be limited in some human tumors, thereby affecting Treg depletion by anti-CTLA4 therapy in patients with cancer. In future studies, it may be important to ascertain the expression of specific Fc receptors within the tumor microenvironment (TME) to determine whether ICT antibodies will mediate ADCC to affect Treg frequency. In general, anti-CTLA4 expands CD4 effector T cells (Teff) with subsequent increase in the Teff/Treg ratio in responding tumors (22).

Anti–PD-1/PD-L1 antibodies predominantly affect exhausted CD8 T cells. Anti–PD-1/PD-L1 also does not appear to expand clonal diversity nor promote T-cell trafficking into tumors. Preclinical studies have shown that concurrent targeting of CTLA4 and PD-1/PD-L1 can improve therapeutic efficacy when compared with each alone (28–30). mAbs targeting CTLA4, PD-1, and PD-L1 have formed the foundation of effective ICT in patients and are now in widespread clinical use across a number of tumor types (31–34).

There are now more than 50 FDA approvals for ICTs in human cancers, which are briefly discussed in this section and summarized in Table 1. The clinically approved ICTs have been comprehensively reviewed in other sources (35, 36).

Ipilimumab (anti-CTLA4) was the first ICT to gain FDA approval in 2011 for the treatment of metastatic melanoma, on the basis of phase III trials demonstrating improvements in overall survival compared with vaccine or chemotherapy (37, 38). Initial studies were challenging because of the paradoxical phenomenon of pseudoprogression, in which patients initially experienced tumor growth before subsequent regression. The benefit of ICT would not have been adequately captured by traditional metrics of tumor measurement (39–41). Survival therefore proved to be the most important indicator of ICT benefit. Indeed, approximately 20% of patients who received ipilimumab achieved long-term survival greater than three years, with survival of 10 years noted for some patients (38, 42). Although ipilimumab remains the only anti-CTLA4 therapy to have approval in many countries, including FDA approval, other mAbs are being studied, including tremelimumab and quavonlimab (43). New forms of anti-CTLA4 therapies are also under development, including glycoengineered antibodies that seek to enhance ADCC and prodrug forms with reduced systemic activity to improve safety (44).

In 2014, the first PD-1 inhibitors were approved for the treatment of unresectable or metastatic melanoma (45, 46). The PD-1/PD-L1 inhibitors are now approved across multiple tumor types, including skin, genitourinary, lung, head and neck, breast, lymphoma, gynecologic, and gastrointestinal cancers. These approvals are summarized in Table 1. The PD-1 mAbs include nivolumab (fully human IgG4), pembrolizumab (humanized IgG4), and cemiplimab (fully human IgG4). The PD-L1 mAbs include durvalumab (fully human IgG1κ), atezolizumab (humanized IgG1), and avelumab (fully human IgG1). Anti–PD-1 therapies have also received approval for treatment of patients with cancer who are diagnosed with tumors that have microsatellite instability–high (MSI-H)/mismatch-repair deficiency (dMMR; refs. 47, 48) or high tumor mutational burden (TMB-high) regardless of tumor type (49). In addition, anti–PD-1/PD-L1 therapies were also approved by the FDA as maintenance therapy, which is administered after initial treatment of locally advanced or metastatic disease. Clinical benefit with maintenance ICT in specific settings was demonstrated in patients with non–small cell lung cancer (NSCLC; ref. 50), extensive-stage small cell lung cancer (51, 52), and bladder cancer (53), leading to FDA approvals in these tumor types.

Despite the promising responses with ICT monotherapies, these responses are observed in only 20% to 30% of treated patients. Therefore, combination therapy was pursued in the clinic based on preclinical data demonstrating that anti-CTLA4 and anti–PD-1/PD-L1 function through distinct pathways, and combined inhibition enhanced antitumor responses in murine models (28).

Multiple clinical trials with combined blockade of CTLA4 and PD-1/PD-L1 demonstrated higher response rates than either monotherapy, in different cancer types including melanoma (54, 55), hepatocellular carcinoma (56), renal cell carcinoma (RCC; refs. 57, 58), NSCLC (59), malignant pleural mesothelioma (60), and colorectal cancer (MSI-H/dMMR; ref. 61). Long-term follow-up has demonstrated durable survival benefit for a subset of patients with melanoma and RCC, and combination therapy with ICT is under active investigation for a number of tumor types (55, 58).

The quest to improve response rates has also prompted combinations with other established cancer drugs, for example, chemotherapy, which is hypothesized to kill immunosuppressive cells in the TME while increasing exposure to tumor antigens via the induction of tumor cell death, thereby enhancing recognition and elimination of tumor cells by the host immune system (62). Toward this end, combinations of ICTs with concurrent and/or sequential chemotherapy are under extensive investigation, with phase III clinical data demonstrating overall survival and/or progression-free survival benefit in NSCLC (63–66), small cell lung cancer (51, 52), triple-negative breast cancer (67), bladder cancer (53), and head and neck squamous cell carcinoma (68). However, the impact of different chemotherapy agents on the immune response within the TME has not been fully elucidated. Similarly, improved clinical responses have been demonstrated with the combination of ICT and targeted therapy. Combinations of VEGF-targeted agents with anti–PD-1 have demonstrated clinical benefits in endometrial carcinoma (pembrolizumab plus lenvatinib; refs. 69, 70), RCC (pembrolizumab plus axitinib; nivolumab plus cabozantinib; avelumab plus axitinib), and hepatocellular carcinoma (atezolizumab plus bevacizumab; refs. 71–75). Recently, the addition of atezolizumab to targeted therapy was also shown to improve progression-free survival in BRAF-mutated melanoma (76).

Although these combination therapies offer the possibility of improving cancer-specific outcomes, they remain limited to specific tumor types and disease settings. For the field of cancer immunotherapy to move forward successfully, with improved clinical benefit for many more patients, we will require rational combination therapies, which should be based on a reasonable understanding of the mechanism of action of each agent and its impact on immune responses.

As combination strategies are pursued, there is an increased risk of development of irAEs. Mitigation strategies, therefore, will require a deeper mechanistic understanding of the toxicities. These irAEs are distinct from the predictable side effects associated with conventional cancer therapies. The irAEs can affect any organ, most commonly skin, gastrointestinal tract, and endocrine system (including the thyroid, adrenal, and pituitary glands). The severity of irAEs can range from mild to even fatal. Clinical guidelines for the management of irAEs have been reviewed elsewhere (77). Although guidelines recommend high-dose corticosteroids for severe irAEs, often with prolonged tapers, this approach can be associated with serious iatrogenic side effects, highlighting the need for mechanistically informed, steroid-sparing immunosuppressive approaches. Early recognition and treatment of irAEs is essential to prevent progression to severe events, and major efforts are under way to develop biomarkers for earlier predictions of irAEs, including blood-based biomarkers such as CD8 T-cell clonal expansion, serum autoantibodies (reflective of B-cell response), and technology-based approaches with patient symptom reporting (78–81).

Published data indicate that nonfatal irAEs may be associated with improved outcomes in patients with cancer (82–96). Therefore, the development of strategies that can enable patients to continue on ICT, including the use of selective immunosuppressive agents, will be important as the field of ICT moves forward. For example, the recent development of a murine model of myocarditis, one of the most feared complications of ICT, provided a rationale for the use of CTLA4-Ig (abatacept), which blocks T-cell costimulation, although an open question remains as to whether this will interfere with antitumor immunity (97, 98). Similarly, in a study of patients with melanoma who developed colitis after ICT, single-cell RNA sequencing (scRNA-seq) and TCR sequencing identified tissue-resident colitis-associated CD8 T cells, consistent with the clinical observation that colitis symptoms tend to occur early after initiation of ICT, and also provided support for the use of the α4β7 antibody vedolizumab to decrease inflammatory immune responses in ICT-associated colitis (99).

Overall, as multiple mechanisms related to irAEs become clear, including both T cell– and B cell–mediated pathways, a major question will be to address whether the mechanisms of irAEs can successfully be decoupled from those of antitumor efficacy. Future ICT clinical trials may include components of selective, prophylactic immunosuppression for high-risk, biomarker-selected patients to prevent severe irAEs, or possibly include administration of targeted immunosuppression to enable continuation of ICT in patients.

To obtain a greater understanding of mechanisms of response and resistance to ICT, we propose that reverse translational studies on individual agents be undertaken, whereby each agent can be studied in small cohorts of patients, with longitudinal samples collected and analyzed thoroughly to generate hypotheses regarding immunologic mechanisms that can be tested in appropriate preclinical models, and with translation of the preclinical data to new clinical trials. These data sets may provide relevant information that can help with the design of rational clinical trials as well as avoid the pitfalls of failed clinical trials. We have adopted this integrated strategy to understand the pathways regulating immune responses and to guide rational combinatorial strategies, as outlined below.

Prostate cancer is considered a “cold” tumor due to the fact that it has few mutations, as compared with “hot” tumors such as melanoma and NSCLC. As a result of a low number of mutations, and therefore few antigens for T-cell recognition, as noted by scarce tumor-infiltrating T cells, prostate cancer has been predominantly resistant to treatment with ICT (100–102). The mechanisms by which prostate tumors exclude T cells or fail to elicit robust infiltration by T cells represent primary resistance mechanisms to ICT. Without tumor-infiltrating T cells, the blockade of a T cell–inhibitory pathway such as PD-1/PD-L1 is unlikely to provide clinical benefit. For example, a recent phase III clinical trial of the androgen receptor antagonist enzalutamide plus anti–PD-L1 failed to meet its primary endpoint of improved overall survival (103). Previous studies on human prostate cancers demonstrated few tumor-infiltrating T cells, with minimal to no expression of PD-1 or PD-L1, and lack of significant clinical benefit with anti–PD-1 monotherapy in patients with metastatic prostate cancer (104, 105). Therefore, the phase III combination study would have benefited from a smaller clinical trial to truly determine whether enzalutamide was capable of increasing tumor-infiltrating T cells with subsequent increase in PD-1 and PD-L1 expression within the TME (106). Similarly, a phase III combination study of anti-CTLA4 plus radiotherapy in patients with metastatic prostate cancer also failed to meet its primary endpoint (100, 107). The combination was conducted without understanding the immunologic impact of anti-CTLA4 or radiotherapy on the prostate TME, including sites of metastatic disease, which are predominantly bone metastases in patients with metastatic prostate cancer.

To address the immunologic impact of anti-CTLA4 on the prostate TME, a small clinical trial evaluated pretreatment and posttreatment tumor samples, which demonstrated that anti-CTLA4 was capable of increasing tumor-infiltrating T cells (105). However, anti-CTLA4 also led to an increase in compensatory inhibitory mechanisms, also known as adaptive resistance mechanisms, including increased frequency of cells expressing PD-1/PD-L1 and VISTA within the TME (105). VISTA is an inhibitory checkpoint that is highly expressed on myeloid cells [including macrophages, dendritic cells (DC), monocytes, neutrophils], and moderately expressed on T cells (108). This yin and yang of the immune response, where therapy may drive antitumor immune responses but also increase immunosuppressive mechanisms, is an important concept to take into consideration as new immunotherapeutic strategies are developed.

The data from this small clinical trial generated hypotheses regarding combination therapies with anti-CTLA4 plus anti–PD-1 and/or anti-VISTA. These hypotheses were tested in preclinical studies, and the data led to a new study with anti-CTLA4 plus anti–PD-1 (109). The new combination clinical trial elicited antitumor responses in some patients, but these responses were predominantly in patients with soft-tissue metastases as opposed to bone metastases (109). Immune monitoring data of posttreatment tumor samples collected from patients indicated that soft-tissue metastases were infiltrated by Th1 cells, whereas bone metastases were infiltrated by Th17 cells (110). The increased Th1 responses provided an explanation for why soft-tissue metastases responded better than bone metastases to ICT.

But why did ICT drive Th1 responses in soft-tissue metastases and Th17 responses in bone metastases? A preclinical model was designed to answer this question. Data from these experiments demonstrated high levels of IL6 and TGFβ in bone metastases as opposed to soft-tissue metastases, which are important cytokines for skewing T-cell response toward Th17 instead of Th1 (110). These preclinical data are now being used to develop a new combination therapy clinical trial with ICT plus an agent to target TGFβ.

On the basis of these data, the concept of organ-specific immune microenvironments will need to be addressed as the field of ICT moves forward in the next decade. Treatment of patients with metastatic disease will need to take into consideration whether certain organs, such as bone and liver, have distinct immune microenvironments that will require different immunotherapeutic strategies (111).

The immune microenvironment is unique to each tumor type and plays a critical role in response and resistance to ICT. Therefore, it is important to understand the composition and phenotypic states of intratumoral immune cells of different tumor types to develop a tumor-specific ICT strategy. Analysis of the immune microenvironment of five different tumor types, including those that respond relatively well to ICT and those that do not, revealed that ICT-sensitive and ICT-resistant tumor types have distinct immune microenvironments. Interaction of tumor cells with the immune microenvironment guides the constitution and phenotypes of immune subpopulations.

Immunologically cold tumors, such as prostate cancer and glioblastoma multiforme (GBM), have few tumor-infiltrating T cells but higher frequency of immune-suppressive myeloid cells. Comparison of tumor-infiltrating lymphocytes (TIL) from GBM and multiple other tumor types led to the identification of a GBM-specific CD73^hi myeloid cell population, which persists even after treatment with anti–PD-1 (112). CD73 is an extracellular ectonucleotidase that plays a role in the adenosine immunosuppressive pathway, inhibiting T-cell activation and proliferation. Preclinical studies with CD73 knockout mice showed improved survival with ICT compared with control mice, demonstrating a direct role for CD73 (112). These series of studies helped in the identification of CD73 as a specific immunotherapeutic target to improve antitumor immune responses to ICT in GBM, generating a combination strategy for a future clinical trial in GBM.

Other myeloid-specific targets, including chemokine receptors (CXCR4, CCR2, and CSF1R), intracellular mediators (IDO1 and PI3Kϒ), and proangiogenic agents (VEGF receptors, semaphorin 4D and angiopoietin-2), are also being explored in an attempt to limit the function of immunosuppressive (M2) macrophages while promoting the function of M1 macrophages, which facilitate antitumor immune responses (113).

The differentiation of macrophages into M1 or M2 subsets is a complex biological process that is being investigated. Many different signals, including cytokine and epigenetic signals, appear to play a role in driving macrophage differentiation. Interestingly, the process of phagocytosis, which is required for clearing tumor cells, also seems to play a role in driving M2 differentiation in an attempt to suppress immune responses and avoid further cell death (114), indicating the complexity of how immune responses are regulated. Detailed understanding of the signals and biological processes involved in the differentiation of macrophages will be needed to effectively target these cells.

As outlined previously, the functional and phenotypic states of immune subpopulations play an important role in determining response to ICT. Cellular plasticity is often driven by epigenetic mechanisms. Although the role of the epigenetic machinery in regulating cancer cell plasticity has been characterized (115), we lack mechanistic insight into the epigenetic regulation of immune subsets, especially in the context of ICT.

It was demonstrated that ipilimumab induces the expression of EZH2, a key epigenetic enzyme that plays an important role in cellular plasticity (115–118). CD28 signaling leads to increased EZH2 expression on T cells (119). Anti-CTLA4 increases CD28 signaling, thereby leading to increased EZH2 expression on T cells. On the basis of these observations, a series of preclinical studies were designed that identified a potential role for EZH2 in driving adaptive resistance to anti-CTLA4 therapy. The combination of EZH2 inhibition with anti-CTLA4 therapy led to improved survival in the preclinical studies (116). These studies formed the basis for a clinical trial of the rational combination of an EZH1/2 inhibitor (DS3201) with ipilimumab in patients with genitourinary malignancies with primary resistance to ICT (NCT04388852). This represents the tip of the iceberg in terms of epigenetic regulation of immune cells, providing the foundation for the combination of epigenetic modulators with ICT (120).

It is clear that we currently have the tools at our disposal to interrogate human immune responses in great depth and we will need to take advantage of these tools, and combine the human data with appropriate preclinical models, to identify rational targets for combination strategies that will improve clinical outcomes for patients. Patient selection will also play an important part in developing effective treatments, and we will need to consider multiple factors, including both tumor-related factors and immune-related factors, to identify appropriate populations.

One of the major challenges in the field of ICT is the lack of a robust predictive biomarker for optimal patient selection. Traditionally, biomarkers have focused primarily on tumor-intrinsic factors or single immune-specific markers. As our understanding of determinants of antitumor immunity has improved, we appreciate that single biomarkers in isolation are insufficient. Combinatorial biomarker strategies that capture attributes of the host and tumor–immune ecosystem will be essential (Fig. 2; refs. 121–124).

High TMB is associated with response to ICT in numerous cancer types, most notably melanoma and NSCLC (49). High TMB may contribute to the formation of neoantigens that are recognized by the immune system (125, 126). On a tissue-agnostic basis, anti–PD-1 (pembrolizumab) received FDA approval for the treatment of advanced pediatric and adult solid tumors that have high TMB (≥10 mutations/megabase). This is a promising step on which to build, and future studies on TMB as a predictive biomarker will address threshold validation as well as provide a deeper understanding of the immunogenicity of different mutational patterns (127).

Although high TMB led to FDA approval, it should be noted that certain tumor types with moderate TMB carry relatively high response rates to ICT, such as clear cell RCC (128, 129). In addition, low TMB does not preclude generation of effective antigen responses, nor does high TMB guarantee response to ICT (130–132). For example, it was shown that anti-CTLA4 therapy can induce effective neoantigen-specific T-cell responses in patients with prostate cancer, even in the background of low TMB (131). Therefore, one approach to enhance antigen-specific T-cell responses may be the combination of ICT with personalized neoantigen vaccines, which have been shown to modulate T cells in the TME and enhance responses to ICT, including in both immunologically “cold” tumors such as glioblastoma and “hot” tumors such as melanoma (133–135). As vaccine therapies evolve to encompass appropriate formulations, adjuvants, and delivery methods that can elicit effective antitumor immune responses, it may be possible to take advantage of tumor mutations to generate neoantigen vaccines in combination with ICT for the development of successful therapeutic strategies.

There are fundamental links between DNA damage and immunogenicity. Tumors with DNA damage repair (DDR) defects, exemplified by MSI-H/dMMR status, were predicted to have high rates of neoantigen formation due to their high mutational load (approximately 10-fold compared with that of MMR-proficient tumors) and therefore higher rates of response to ICT (136). An initial study in patients with colorectal cancer demonstrated significantly higher rates of response to pembrolizumab in dMMR patients compared with MMR-proficient patients (47), a finding that was later extended to multiple tumor types (48). These data led to the tumor-agnostic approval of pembrolizumab for MSI-H/dMMR solid tumors (48).

Additionally, homologous repair defects, such as pathogenic mutations in BRCA1/2, can lead to enhanced immune response by multiple pathways including the release of cytosolic DNA, promotion of type I IFN signaling, and upregulation of PD-L1 (137–139). In a study of patients with urothelial carcinoma, higher response rates to anti–PD-1/PD-L1 antibodies were observed in patients with known or likely deleterious DDR alterations (140). Emerging evidence indicates that specific DDR mutations have differential effects on the tumor–immune microenvironment, and therefore mutations will need to be selected carefully as candidate predictive biomarkers (141). In all, an integrated approach incorporating events in specific driver genes and mutational signatures will be critical to develop additional genetic biomarkers of response (142).

IHC evaluation of PD-L1 expression on tumor and/or immune cells is frequently performed to select patients for treatment with anti–PD-1/PD-L1 antibodies. A correlation between PD-L1 expression on tumor cells and clinical response is seen in certain circumstances (e.g., in NSCLC); however, in other cases, substantial responses can be observed in some patients with PD-L1–negative tumors, which highlights the difficulty of using PD-L1 as a single biomarker (34, 143, 144). PD-L1 expression is dynamic due to multiple factors, including T-cell responses and IFNγ signaling. Therefore, variations in the timing of biopsy of the tumor as well as the site of biopsy may influence whether PD-L1 expression is detected or not. Although a range of assays have been approved, there remains significant variation in antibodies, tissue handling techniques, and thresholds, which limits the uniform benefit of PD-L1 expression as an isolated biomarker, and efforts are under way to integrate PD-L1 expression with other biomarkers (described below; ref. 127).

TILs have a long history in cancer immunotherapy, with demonstration in the 1980s that adoptive transfer of TILs derived from patients' melanoma tumors can lead to cancer regression (145). It was later shown that TILs, including CD8 T cells, correlated with clinical outcomes in multiple tumor types, including ovarian cancer and bladder cancer (146, 147). Subsequent research on the correlates of response of anti-CTLA4 therapy demonstrated that the extent of ICT-induced tumor necrosis was directly related to the ratio of intratumoral CD8 T cells to FOXP3⁺ Tregs (148). TILs are therefore under active investigation as a biomarker to guide selection of patients for treatment with ICT, although significant variability exists in definitions and thresholds (149, 150). In a prospective study, the AMADEUS trial (NCT03651271, ongoing) was designed to evaluate the benefit of classifying tumors into immunologically “hot” and “cold” tumors based on CD8⁺ T-cell density (≥15% and <15%, respectively), with anti–PD-1 (nivolumab) monotherapy administered as treatment for “hot” tumors, whereas anti-CTLA4 (ipilimumab) plus anti–PD-1 (nivolumab) combination is administered as treatment for “cold” tumors in an attempt to take advantage of the ability of anti-CTLA4 to drive T cells into “cold” tumors. In addition, the Immunoscore is an approach that aims to standardize and quantify CD3 and CD8 T cells infiltrating and surrounding a tumor (149, 151). However, due to variability in immune cell infiltration in different tumor types, its applicability across cancers is challenging. Multiple studies indicate that the location of T cells in the tumor (whether at the invasive margins or in tumor parenchyma) and the phenotype of the T cells (whether cells are memory-like, exhausted, and/or activated) will need to be taken into consideration (152–154).

In the same vein, T-cell exclusion is a mechanism of resistance to ICT, exemplified by “cold” tumors such as prostate cancer and pancreatic cancer, which have few T cells in the TME. Certain oncogenic pathways may enable tumors to exploit this mechanism of immune evasion. For example, tumor cell–intrinsic WNT/β-catenin activation has been shown to lead to exclusion of T cells within the TME (123, 155, 156). In addition, PTEN loss is also associated with increased signaling through the PI3K–AKT pathway and correlated with exclusion of CD8 T cells and poor clinical responses to ICT (124). Thus, patient selection for treatment and strategies to overcome T-cell exclusion in the TME will need to be considered in future ICT trials.

IFNγ is a key cytokine produced by activated T cells, as well as natural killer (NK) and NK T cells, and is critical for effective antitumor immunity. Preclinical work using IFNγ-blocking antibody and murine models lacking essential IFNγ signaling molecules demonstrated a key role for IFNγ in immune surveillance and tumor rejection (157, 158). In addition, knockdown of IFNγ-receptor 1 led to impaired antitumor responses with anti-CTLA4 in a murine model of melanoma (159). Consistent with the preclinical data, patients with melanoma harboring genomic defects in IFNγ pathway genes did not respond to anti-CTLA4 (160). Also, patients with melanoma tumors harboring loss-of-function mutations in JAK1 and JAK2, downstream IFNγ signaling molecules, were resistant to anti–PD-1 therapy, highlighting the utility of IFNγ as a potential biomarker.

IFNγ induces a host of target genes that have been used to develop gene-expression signatures, which correlate with responses to ICT. The gene-expression signatures can be retrospectively labeled as “high” or “low” based on median cutoff values. Multiple studies have reported that tumors identified as having “high” expression of IFNγ-related gene signatures have better clinical responses to ICT (131, 161, 162). However, the gene panels used to design the IFN-gene signatures were not consistent across these studies. In addition, consistent with the yin and yang concept of immune responses, IFNγ drives expression of proteins such as PD-L1 and IDO1, which may act to suppress immune responses in the TME. Additional studies are ongoing to develop reproducible assays for detection of IFNγ-related genes and signatures that may be used as a prospective biomarker for selection of patients for ICT.

Tertiary lymphoid structures (TLS) have recently emerged as biomarkers of response to ICT across several tumor types, including lung cancer, melanoma, RCC, sarcoma, and urothelial cancer (163–167). TLS or TLS-like structures comprised of T cells, B cells, DCs, and other APCs may play a significant role in antitumor T-cell responses, likely due to further priming and activation of T cells within the TME, which may represent another important event in the cancer immunity cycle. In contrast to other biomarkers of response which give a snapshot of a single time point and single cell type, TLS are dynamic structures comprised of multiple cell types. Multiple chemokines and cytokines are involved in attracting these cell types and forming the TLS, including CXCL13 (B cells) and IL2 (T cells; ref. 168).

TLS represent sites of interactions between multiple immune cell types and capture an ongoing antitumor response (168). Future studies using newer technologies will be essential to understand which cells are interacting within the TLS, and how they differ from the typical structures within lymph nodes. Finally, because TLS appear to be associated with responses to ICT across multiple tumor types, it will be important to define the specific mechanisms driving TLS formation to determine whether these mechanisms can be exploited to induce the formation of TLS in otherwise nonresponding tumors.

Although many of the biomarkers described above, including tumor-intrinsic and host-specific variables, correlate with response to ICTs, many of the studies have been retrospective and need further validation through prospective clinical trials. Furthermore, there is a critical need to bridge preclinical mechanistic studies with clinical investigations and vice versa to develop reproducible predictive biomarkers. There is growing recognition that single biomarker strategies do not incorporate the dynamic and complex interaction of the tumor and host immune system and that combinatorial biomarkers will be essential to optimize patient selection. With inherent inter- and intratumor heterogeneity, a complex interplay of diverse cell types, and the continuously evolving tumor immune microenvironment, it is clear that the utility of isolated biomarkers is limited. A combination approach integrating multiple data sets across different platforms from diverse studies will be essential for developing more sensitive and robust biomarkers of response.

Recent studies demonstrated that the combination of existing biomarkers has better predictive capacity than single biomarkers. For example, TMB plus a T cell–inflamed gene-expression profile or TMB plus PD-L1 expression showed an improved prediction performance compared with single biomarkers (130). A recent study of more than 1,000 ICT-treated patients across seven tumor types showed that a multivariable predictor of response outperformed TMB alone (169). Such combinatorial biomarkers, tested in prospective clinical trials, will be key for optimal patient selection.

It was also recently demonstrated that two novel biomarkers correlate with response to anti–PD-1/PD-L1 therapies. A retrospective analysis identified mutations in ARID1A in tumor tissues and expression of the chemokine CXCL13 as predictive biomarkers for ICT (170). Preclinical studies demonstrated association of ARID1A and CXCL13 with antitumor immunity in mice (170). Importantly, the combination of ARID1A mutation plus CXCL13 showed improved predictive capacity compared with either of the single biomarkers alone (170). This retrospective study highlighted the importance of integrating tumor mutational status with the immune microenvironment to predict responses to ICT, which led to a prospective clinical trial to test ARID1A mutation plus CXCL13 as a combinatorial biomarker strategy for patient selection.

The field of ICT has expanded across many different tumor types, with improved benefit for patients with cancer. In addition, as the field continues to highlight other pathways that can be targeted, there are now a vast number of combinatorial approaches than can be tested in patients. Therefore, it will be important to efficiently design future studies with attention to the mechanisms of action of each agent as well as an understanding of resistance mechanisms that evolve over time to control immune responses. The field of ICT is still growing, but as we gain knowledge in multiple areas, as outlined below, the field will mature to provide clinical benefit for even more patients with cancer (Fig. 3).

Over the past decade, ICT has become established as the fourth pillar of cancer therapy, along with chemotherapy/targeted therapy, radiotherapy, and surgery. We have seen some success in combining ICT with chemotherapy and targeted therapy. The next decade will provide scientific insights into integrating ICT with these pillars of cancer care to benefit more patients. Exciting preclinical data on radiotherapy plus ICT should provide insight into translating it to clinical practice. Although surgical options in the metastatic setting are extremely limited, we are exploring combinations of ICT with surgery in the metastatic setting. These studies will broaden the field of ICT across all aspects of oncology as a unifying foundation for rational combination strategies with chemotherapy, targeted therapy, radiotherapy, and surgery. In addition, a number of other potential candidates such as other immune checkpoints and costimulatory receptors, epigenetic modulators, and metabolic targets hold promise for improving the durable clinical responses seen with ICT.

Treatment of patients with earlier stages of disease, with therapy given in either the neoadjuvant or adjuvant setting, has the potential to prevent disease relapse and lead to improved survival. Neoadjuvant ICT offers the possibility of surgical downstaging and provides ample tissue for examination of an intact TME, potentially yielding insight into posttreatment changes that can inform subsequent clinical trials. The first neoadjuvant ICT trial was conducted in 2006 and provided safety data indicating that ICT can be given prior to surgery (17). The study was conducted in patients with localized bladder cancer who received anti-CTLA4 before proceeding to cystectomy (17). This study was also the first ICT trial in patients with bladder cancer, which provided critical data demonstrating antitumor responses in patients with bladder cancer, thereby leading to large clinical trials with ICT in patients with metastatic bladder cancer, with subsequent FDA approval. In addition, the immune monitoring studies performed on the trial highlighted the reverse translational approach and identified ICOS⁺ CD4 T cells as a novel subset of T cells associated with anti-CTLA4 therapy (17, 18, 171, 172). The trial also led to the first combination ICT neoadjuvant trial with anti-CTLA4 plus anti–PD-L1 in patients with bladder cancer, which demonstrated clinical responses and correlation of TLS with responses (21). Clinical trials with ICT in the neoadjuvant setting have now demonstrated clinical benefit in multiple tumor types, including melanoma (173, 174), Merkel cell carcinoma (175), NSCLC (176), MMR-deficient colorectal cancer (177), urothelial carcinoma (21, 178), and breast cancer (179–181). Additional studies are being planned, especially as investigators consider whether to implement survival or other clinical endpoints, including pathologic complete responses, which will enable faster readout and review for possible approval (182–188).

Administration of ICT in the neoadjuvant setting enables T and B cells to have access to the antigens that are present on tumor tissues, which is in contrast to adjuvant studies where ICT is given after tumor tissues have been removed, thereby limiting the amount of tumor antigens that T and B cells encounter. Nonetheless, adjuvant ICT has been successful in high-risk melanoma, which led to FDA approval (189–193). A number of additional studies are currently ongoing and will provide additional insight into the use of ICT in the adjuvant setting.

Although much attention has traditionally focused on adaptive immunity, there is growing interest in the role of innate immune responses to boost overall antitumor immunity. The innate immune system is composed of NK cells, dendritic cells, macrophages, monocytes, neutrophils, eosinophils, basophils, and mast cells. Therapeutic approaches include stimulating innate responses with microbial peptides or oncolytic viruses, which have previously been shown to enhance local and systemic antitumor immunity in combination with ICT and may be able to improve payload to enhance agonist signals (194–196). The presence of tumor-infiltrating DCs expressing the transcription factor BATF as well as the CD103 marker has been shown to be essential for presentation of neoantigens and activation of cytotoxic T cells in the TME (197, 198). Therefore, approaches that enhance and/or activate these DCs may improve local antitumor responses. Systemic approaches to stimulating innate immunity, such as via agonism of stimulator of interferon response cGAMP interactor 1 (STING) or Toll-like receptors (TLR), have been challenging (199–201). Therefore, intratumoral delivery rather than systemic delivery of agents to target innate immune cells may provide an avenue to ensure presentation of tumor-specific antigens, enable effective T-cell responses, and also reduce systemic toxicities. Intratumoral agents under investigation in clinical trials include mRNAs encoding CD70, CD40 ligand, and TLR4 (202) and oncolytic viruses such as talimogene laherparepvec, or T-VEC (203, 204), which is FDA- approved for melanoma and is being explored in a number of clinical trials. However, successful DC activation in a suppressive TME continues to remain a challenge and needs to be further investigated.

Furthermore, as our understanding of host effects on the immune system has grown, it is clear that the gut microbiome exerts influence over the effectiveness of ICTs (127). Responses to ICTs have been correlated with commensal gut flora, and there appear to be substantial differences in the diversity and composition in patients who respond to ICT versus nonresponders (205, 206). Challenges in investigating the gut microbiome include differences in sampling, analysis platforms, and population heterogeneity, though these can be addressed with the development of standard sequencing and analytic tools. Manipulation of the gut microbiome, such as through fecal microbiota transplantation, antibiotic therapy, or diet, may serve as an adjunct strategy to enhance responses to ICT. These strategies are still in early stages of being explored but will provide important data that will affect treatment with ICT.

On the basis of our understanding of T-cell activation, there is growing interest in driving agonist signals via costimulatory receptors. However, clinical experience with a CD28 superagonist that resulted in severe toxicities has led the field to proceed with caution as additional agonist antibodies are evaluated (207). Both OX40 and 4-1BB/CD137 agonists have been tested in clinical trials. Preclinical work with OX40 and 4-1BB showed efficacy; however, a phase Ib study of an OX40 agonist combined with atezolizumab demonstrated safety but limited efficacy in patients (208–210). In addition, phase I studies with CD137-directed mAb urelumab caused significant elevation of liver enzymes (211), and a second anti-CD137 antibody, utomilumab, was well tolerated but had limited efficacy (212). These studies highlight significant knowledge gaps regarding the use of agonist antibodies in terms of understanding expression of the target proteins, sequencing of the agonist antibody with ICT and choice of ICT agent (anti-CTLA4 vs. anti–PD-1/PD-L1). Currently, a clinical trial with anti-CTLA4 plus an agonist anti-ICOS antibody is ongoing (NCT03989362) based on preclinical data demonstrating enhanced antitumor immune responses in murine models of melanoma and prostate cancer (172). In addition, the ICOS agonist GSK3359609 is also under study in a phase III clinical trial (NCT04128696; ref. 213).

The development of immune agonists faces a number of challenges, including target affinity, epitope selection, receptor occupancy, Fc gamma receptor interactions, antibody isotype, and appropriate in vitro assays to clearly distinguish between an antagonist antibody with activity against a costimulatory molecule expressed on immunosuppressive cells versus an agonist antibody with activity against the same costimulatory molecule expressed on effector cells (214). As a result, the development of an agonist antibody to target T-cell responses may require additional experimental steps and careful immune monitoring studies of patients to help design appropriate combination trials with current ICT agents.

As the field builds on current data with agonist antibodies, novel approaches are being developed to deliver agonist signals in the TME. For example, bispecific antibodies are being explored, including a bispecific antibody to target CD28 plus a tumor-associated antigen (NCT03972657; ref. 215). This approach of targeting the costimulatory pathway within the TME, as opposed to systemic targeting, may provide a safer way to provide agonist signals to T cells. As these processes are optimized, costimulatory receptor targeting is becoming a feasible strategy to enhance the efficacy of ICT.

In the future, spatiotranscriptomic tools such as Digital Spatial Profiling and CO-Detection by indEXing (CODEX) may allow us to decipher the cell types and their interactions within the tumor. In addition, the incorporation of single-cell high-throughput technologies into clinical and preclinical studies will enable better characterization of cell subsets regulating antitumor responses. Assessing single-cell transcriptomics, proteomics, and chromatin state through scRNA-seq, Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq), and single-cell Assay for Transposase-Accessible Chromatin sequencing (scATAC-seq) will enable identification of cellular identity at a single-cell level. Furthermore, integration of various platforms assisted by novel bioinformatic tools will allow us to understand the regulatory pathways governing response and resistance to ICT. In addition, integration of longitudinal blood-based assays such as circulating-tumor DNA (ctDNA; ref. 216), which are easier to implement and less invasive as opposed to longitudinal tumor biopsies, is under investigation as a method to monitor responses to ICT.

Additionally, the development of better bioinformatics algorithms that help with neoantigen prediction will enable identification of more relevant therapeutic targets. Progress in the prediction of shared MHC class I antigens in clinical samples has been exciting. A recent study developed a pipeline to determine tumor epitope immunogenicity that might enable prediction of effective antitumor responses with much higher accuracy than conventional methods (217). However, although developing prediction algorithms for class I antigens is necessary, it has recently been shown that MHC class II neoantigens are also involved in effective tumor responses and will need to be considered (218).

The successful implementation of these technologies will require close collaboration between computational biologists and immunologists to avoid simplification of the complex biological and functional processes of immune responses, thereby maintaining fidelity to the dynamic nature of cell subsets within the immune system. For example, PD-1 expression on a T cell may represent an “activated” T cell or an “exhausted” T cell. The simple identification of PD-1 through a bioinformatics approach may not provide sufficient understanding regarding the complexity of the T-cell response. Therefore, appropriate functional studies or other immunologic assays may be necessary to identify additional biomarkers that should be included in the bioinformatic analyses for a more accurate definition of a given immune cell state. Careful attention to these types of details will be necessary for advancing the field of ICT.

Over the next decade, the field of cancer immunotherapy will need to grow vertically, to gain a deeper understanding of different immune cell subsets and pathways that regulate their responses, as well as horizontally, to integrate with other fields, including cancer biology, epigenetics, computational biology, and many others. ICT will also need to expand across all of oncology for the development of combination strategies, including combinations with surgery, radiotherapy, chemotherapy, targeted therapy, and other immunotherapy agents such as novel immune checkpoints and chimeric antigen receptor T cells. As the field continues to grow, it will need to remain grounded in the strength of basic immunology while it fosters close collaborations with other scientific disciplines. The future of ICT is promising, with ample opportunity to further improve outcomes for patients with cancer.

P. Sharma reports personal fees from Jounce Therapeutics, Oncolytics, Earli, Glympse Bio, Lava Therapeutics, Lytix, Phenomics, Codiak, Constellation, Achelois, Dragonfly, BioAtla, Infinity Pharma, Hummingbird, and Polaris outside the submitted work; in addition, P. Sharma has a patent for ARID1A plus CXCL13 as combinatorial biomarker pending and a patent for combination of anti-CTLA4 plus agonist targeting of ICOS for improved antitumor immune responses licensed to Jounce Therapeutics. S.K. Subudhi reports personal fees from Dendreon, Kahr Medical, Ltd., Cancer Expert Now, Inc., Bayer Healthcare Pharmaceuticals, Apricity Heatlh, LLC, Dava Oncology, Bristol-Meyers Squibb, Amgen, Inc., and Exelixis, Inc. outside the submitted work. J.P. Allison reports personal fees from Jounce Therapeutics, Codiak Biosciences, Achelois, Lava Therapeutics, Lytix, Earli, Phenomics, Dragonfly, Hummingbird, ImaginAb, Forty-Seven, and Polaris outside the submitted work. No disclosures were reported by the other authors.