Development of Immunotherapy Combination Strategies in Cancer

For most of the last century, cancer research was focused on exploiting the antitumor effects of different classes of therapeutics exclusively on cancer cells (1). Over the past decade, however, we have gained a deeper understanding of the role of the immune system in mediating such responses. This has led to a revolution in preclinical, translational, and clinical efforts dedicated to harnessing the immune system for the development of novel immunotherapeutics in cancer medicine (2). Recent studies have led to a vast range of immunotherapies being assessed in clinical trials and subsequently achieving FDA approval.

Immunotherapy has allowed the field of oncology to turn a critical corner, where long-term survival and even durable cures are achievable for patients with metastatic solid tumors. The current reality, however, is that the majority of patients enter care with immune “cold” tumors which respond poorly, if at all, to existing checkpoint therapies (ref. 3; Fig. 1). Immune suppression in these cancers resists reversal with checkpoint blockade due to its multimodal nature encompassing suppressive cytokines, lack of antigen presentation, apoptotic triggering of T cells, and hostile metabolic states and nutrient deprivation. These additional layers of tumor immune privilege must be peeled back therapeutically in order to reveal the benefits of T-cell checkpoint blockade and drive tumor regression. Thus, combinations of multiple immune interventions are necessary to reverse the “cold” tumor state, yet most of the existing “backbone” immunotherapies already approach the ceiling of tolerability even when used at doses that are clearly below their maximum efficacy (4–6).

The limitations in efficacy observed with the first wave of approved immunotherapies, primarily involving immune-checkpoint inhibitors, have illustrated a need to improve our understanding of the mechanistic underpinnings of the immune system and thereby develop more robust preclinical and clinical drug development strategies (2). A better understanding of primary and secondary resistance is also required to improve patient outcomes to single-agent immunotherapy strategies (7). Improving our insights into mechanisms of response and resistance is the crucial next step for the future development of immunotherapeutics.

In this review, we begin by detailing the successes observed to date with FDA-approved immunotherapy combinations in different tumor indications, before reviewing promising strategies currently in clinical trial testing. We then describe the use of preclinical models to optimize the development of rational combinations. Finally, we discuss the development of high-order immunotherapy combination strategies and novel biomarker and clinical trial strategies to support the development of combination approaches.

A number of immune checkpoint–based combination treatments are now approved by the FDA (Table 1; Fig. 2), with nivolumab (anti–PD-1) and ipilimumab (anti-CTLA4) the earliest immuno-oncology (IO) combination to receive FDA approval in September 2015 for the first-line treatment of metastatic melanoma (8, 9). Although the treatment of metastatic melanoma had already been transformed by single-agent immune-checkpoint blockade, the nivolumab–ipilimumab combination improved objective response rates (ORR) to 58% and median progression-free survival (mPFS) to 11.5 months compared with nivolumab monotherapy (ORR 45%; mPFS 6.9 months) or ipilimumab alone (ORR 19%; mPFS 2.9 months). Despite a high rate of discontinuation due to toxicities, a survival benefit was apparent even for patients who discontinued treatment, with median overall survival (mOS) not reached at 60 months (10, 11).

Following approval for metastatic melanoma, the combination of nivolumab and ipilimumab has been similarly successful for other tumor sites. Although single-agent immune-checkpoint blockade has been strikingly effective in mismatch repair–deficient (dMMR) cancers (12), this is improved by a combination approach. Nivolumab with lower-dose ipilimumab (1 mg/kg) for the second-line treatment of metastatic dMMR colorectal cancer showed a 60% ORR (13, 14) compared with 31% ORR with nivolumab alone (15). There was a relatively lower toxicity rate than that seen with higher-dose ipilimumab (3 mg/kg) in melanoma, with 32% patients experiencing grade 3 to 5 toxicities.

Similarly, in the first-line treatment of intermediate- or poor-risk advanced clear-cell renal cell cancer, nivolumab–ipilimumab studied in CheckMate 213 resulted in improved patient outcomes, with an ORR of 42%, in contrast to 29% with sunitinib (16, 17). As 1 mg/kg ipilimumab appeared to reduce toxicity without losing efficacy, clinical trials are investigating this dose of ipilimumab with nivolumab (3 mg/kg) in metastatic melanoma, with early data indicating comparable efficacy and lower grade 3 to 5 toxicity rates (18).

Multikinase inhibitors have been the only effective and approved treatments in advanced unresectable hepatocellular cancer (HCC) until the recent approval of immune-checkpoint inhibitors (19). The CheckMate 040 study included comparison of nivolumab–ipilimumab combinations as second-line treatment, comprising arm A (nivolumab 1 mg/kg plus ipilimumab 3 mg/kg every 3 weeks for 4 doses), arm B (nivolumab 3 mg/kg plus ipilimumab 1 mg/kg every 3 weeks for 4 doses followed by maintenance nivolumab every 2 weeks), or arm C (nivolumab 3 mg/kg every 2 weeks plus ipilimumab 1 mg/kg every 6 weeks). ORR was similar between arms A, B, and C at 32%, 31%, and 31%, respectively. Arm A experienced the longest survival at 23 months (95% CI, 9–NA), versus 12 months (95% CI, 9–15), and 13 months (95% CI, 7–33) with arms B and C, respectively. Arms B and C, however, at 1 mg/kg ipilimumab carried the fewest grade 3 to 4 toxicities (29% arm B/31% arm C vs. 53% arm A; ref. 20). FDA approval has been granted to combination IO using the arm A dosing.

In the same trial in HCC, the triplet combination of cabozantinib, nivolumab, and ipilimumab was compared with cabozantinib and nivolumab, including in the first-line setting. Although mOS has not been reported, the triplet arm had an mPFS of 6.8 months versus 5.5 months in the doublet arm, albeit with higher grade 3 to 4 toxicities of 71% versus 42%, respectively (21). Although results from this study are awaited, a number of IO plus other anticancer treatments have already been granted FDA approval, as discussed below.

Combinations with chemotherapy have, to date, been the most numerous approved regimens in various indications. A number of studies in non–small cell lung cancer (NSCLC) have demonstrated a survival advantage when immune-checkpoint blockade is administered in combination with chemotherapy. Pembrolizumab given in combination with pemetrexed and platinum chemotherapy as first-line treatment of nonsquamous NSCLC without EGFR or ALK mutations resulted in an mOS of 22.0 months (95% CI, 19.5–25.2) versus 10.7 months (95% CI 8.7–13.6; refs. 22–24). Similarly, in squamous NSCLC, pembrolizumab in combination with carboplatin and taxane (paclitaxel or nab-paclitaxel) chemotherapy demonstrated an improved mOS at 17.1 months (95% CI, 14.4–19.9) compared with chemotherapy alone (11.6 months; 95% CI, 10.1–13.7; refs. 25, 26). Moreover, following consolidation chemoradiotherapy for stage III NSCLC, durvalumab was shown to lead to a significantly improved mPFS (17.2 vs. 5.6 months) and OS [not reached (NR) vs. 29.1 months], respectively (27–29).

Although combination IO treatments were early successes in the treatment of NSCLC, small cell lung cancer (SCLC) initially proved challenging—studies of maintenance anti–PD-1 and anti-CTLA4–chemotherapy combination treatment did not result in improved PFS (30, 31). However, a study of the PD-L1 inhibitor atezolizumab in combination with carboplatin and etoposide for extensive-stage SCLC resulted in improved mOS of 12.3 months (10.8–15.8) versus 10.3 months (9.3–11.3) in the placebo arm (32, 33), leading to FDA approval in the first-line setting. Similarly, the CASPIAN phase III trial assessed durvalumab in combination with platinum and etoposide chemotherapy as first-line treatment of extensive-stage SCLC, demonstrating an improvement in OS when compared with platinum etoposide alone (13.0 months vs. 10.3 months, HR: 0.73; P = 0.005; ref. 34).

Successful approaches have similarly been reported in breast cancer. Triple-negative breast cancer (TNBC) is the most immunogenic of breast cancers, with high tumor-infiltrating lymphocytes particularly in tumors with features of “BRCAness” (35–37). Initially, however, single-agent anti–PD-1 treatment was disappointing, with a response rate of only 5.3% (38). More recently, atezolizumab in combination with nab-paclitaxel resulted in improved OS (25.0 vs. 18.0 months) in PD-L1–positive tumors (with PD-L1 ≥ 1% on tumor-infiltrating immune cells), although not in unselected TNBC (39). Interestingly, a similar approach with paclitaxel (as opposed to nab-paclitaxel, a regimen that does not require steroid immunosuppression for potential allergic reactions) did not improve patient outcomes (https://www.fda.gov/drugs/resources-information-approved-drugs/fda-issues-alert-about-efficacy-and-potential-safety-concerns-atezolizumab-combination-paclitaxel). These conflicting results highlight the need for robust preclinical studies and the development of rational combinations.

Head and neck squamous cell carcinoma (HNSCC) was an early success in single-agent checkpoint studies, with combination approaches now approved in the first-line HNSCC setting (40). Recurrent or metastatic HNSCC has a dismal prognosis of six to nine months, and standard chemotherapy combination regimens lead to significant comorbidity and involve taxing treatment schedules (41). Pembrolizumab plus chemotherapy was compared with standard-of-care 5-fluorouracil plus platinum chemotherapy with cetuximab given as first-line therapy for incurable or metastatic HNSCC. The immunotherapy combination resulted in improved mOS of 13.0 versus 10.7 months in the standard-of-care group (HR, 0.77; P = 0.0067; ref. 42). This OS benefit was observed despite no significant change in PFS, highlighting the importance of this clinical endpoint in trials of immunotherapy.

Antiangiogenic agents, targeting vascular endothelial growth factor (VEGF) or its receptor (VEGFR), are promising agents in combination with immune-checkpoint blockade. The increased density of high endothelial venules (typically surrounded by tertiary lymphoid structures) in the tumor microenvironment promotes T-cell trafficking to the tumor and subsequent response to anti–PD-1 therapy (43). This has been borne out in the clinical setting, with many combinations of antiangiogenics and immune-checkpoint blockade now approved. Accelerated FDA approval was granted to the combination of pembrolizumab plus lenvatinib for the treatment of patients with advanced endometrial carcinoma that is not microsatellite instability–high (MSI-H) or dMMR. Efficacy was demonstrated in 108 patients with metastatic endometrial carcinoma that had progressed following at least one prior line of therapy on the KEYNOTE-146 (NCT02501096) trial (44, 45). The overall response rate at 24 weeks was 38.0% [63.6% in patients with MSI-H tumors (n = 11) and 36.2% in patients with microsatellite-stable (MSS) tumors (n = 94)]. For previously treated patients, the overall median duration of response (DOR) was 21.2 months, mPFS was 7.4 months and mOS was 16.7 months. Of note, grade 3 or 4 treatment-related adverse events occurred in 83 of 124 (66.9%) patients.

A quadruple combination regimen [atezolizumab, bevacizumab, carboplatin, and paclitaxel (ABCP)] with the antiangiogenic bevacizumab with atezolizumab (anti–PD-L1) and carboplatin–paclitaxel chemotherapy in nonsquamous NSCLC demonstrated improved mPFS compared with bevacizumab–carboplatin–paclitaxel (8.3 vs. 6.8 months). This study also reported a PFS benefit in patients with EGFR- or ALK-mutant NSCLC, which is recognized to have a lower benefit rate from single-agent PD-1/PD-L1 inhibitors (46–48). The FDA approval, however, did not include the EGFR- and ALK-mutant patient subgroup.

Successful strategies for treating renal cell cancer (RCC) have focused on the sensitivity of RCC to immune-checkpoint inhibitors and antiangiogenic agents, especially given its insensitivity to cytotoxic chemotherapy (49). The combination of bevacizumab with atezolizumab in RCC did not lead to an improvement in OS at interim analysis (33.6 months vs 34.9 months) despite demonstrating a superior mPFS versus sunitinib alone (11.2 months vs. 8.4 months, P < 0.02; ref. 50). However, single-agent checkpoint inhibition has also been added to the oral small-molecule axitinib in two phase III trials. Compared with sunitinib, the combination of pembrolizumab and axitinib led to a benefit in 12-month OS of 89.9% versus 78.3% (HR, 0.53; 95% CI, 0.38–0.74; P < 0.01; ref. 51). Axitinib has also been assessed in combination with avelumab in untreated RCC and compared with sunitinib. Although OS for this combination is yet to be reported, there was a PFS benefit of 13.8 months (95% CI, 11.1–NR) versus 8.4 months (6.9–11.1) with sunitinib (52). Both combinations have subsequently been approved by the FDA.

In the IMBrave150 open-label phase III randomized trial in patients with resectable HCC naïve to systemic therapy, the combination of the angiogenesis inhibitor bevacizumab added to atezolizumab was compared with sorafenib in a 2:1 fashion. Although the mOS has not been published, the OS at 12 months was 67.2% for the IO combination versus 54.6% for sorafenib. mPFS was improved in the IO group [6.8 months (95% CI, 5.7–8.3) versus 4.3 months (95% CI, 4.0–5.6)], respectively. Grade 3 or 4 adverse events were comparable between both arms, at 56.5% with the IO combination and 55.1% with sorafenib (53).

Also in HCC, the phase Ib/II trial combining lenvatinib and pembrolizumab regardless of PD-L1 status resulted in an ORR of 36.4% (8/22) but with 91% experiencing grade 3 or 4 toxicities (54). This combination is now undergoing evaluation in a randomized phase III study (NCT04199104).

The combination of BRAF and MEK inhibitors for the treatment of BRAF-mutant metastatic melanoma transformed outcomes in this aggressive subgroup, achieving rapid disease control, and is now also approved in the adjuvant setting (55–57). In IMspire150, vemurafenib and cobimetinib given in combination with atezolizumab improved PFS for patients with unresectable or metastatic melanoma (15.1 months; 95% CI, 11.4–18.4) versus vemurafenib and cobimetinib with placebo (10.6 months; 95% CI, 9.3–12.7). DOR was also improved with the addition of checkpoint inhibition [21.0 (95% CI, 15.1–NE) vs. 12.6 months (95% CI, 10.5–16.6)] with, importantly, little change in the grade 3 to 4 toxicity rate (79% vs. 73%). FDA approval was granted in July 2020, with a first cycle of vemurafenib and cobimetinib, and atezolizumab from cycle 2 onward. Interestingly, a subsequent phase III trial of the anti–PD-1 antibody spartalizumab in combination with the BRAF inhibitor dabrafenib and MEK inhibitor trametinib in patients with BRAF-mutated metastatic melanoma, reported at the 2020 European Society for Medical Oncology (ESMO) Annual Meeting, failed to reach its primary endpoint and raised questions regarding the utility of triplet combinations in melanoma (58).

These encouraging results with regulatory approval in multiple indications with combination immunotherapy strategies have provided the continued impetus for a large number of combination studies currently in preclinical trial testing, and selected studies are detailed in Table 2.

The generally safe and durable responses observed with single-agent PD-1/PD-L1 inhibitors have established this class of agents as the therapeutic backbone of the majority of immune-checkpoint inhibitor combinations (59). The tantalizing promise of durable antitumor responses has understandably fed enthusiasm for further combination studies, with 2,251 active PD-1/PD-L1 inhibitor–based combination trials of 295 targets in 2019 (www.cancerresearch.org; Fig. 2).

In metastatic melanoma, high-dose anti-CTLA4 agent ipilimumab at 10 versus 3 mg/kg was found to have increased mOS (15.7 months vs. 11.5 months; P = 0.04) at a significant cost of increased treatment-related adverse events (37% vs. 18%; ref. 5). Conversely, the anti–PD-1 pembrolizumab has been found to have similar efficacy with similar rates of toxicity at two different weight-based dosing schemes in NSCLC (47) and melanoma (60). Exposure-response modeling in melanoma for nivolumab at doses of 0.1–10 mg/kg every two weeks has shown linear pharmacokinetics with time-varying clearance and without exposure being a significant predictor of response or survival. The FDA has since recommended a flat-dosing scheme for both nivolumab and pembrolizumab.

PD-L1 antibodies have been found to more effectively block PD-1 signaling in vivo (61) and have less PD-L2 interaction (62). This difference has led some to propose differential efficacy and adverse events, although prospective head-to-head comparisons have not been made. A meta-analysis of 11,379 clinical trial patients found improved survival and PFS with anti–PD-1 inhibitors compared with anti–PD-L1, whereas adverse events were similar (63). PD-1 inhibitors have been associated with increased risk of grade 3 to 4 colitis (0.85% vs. 0.34%, relative risk; 2.52) compared with PD-L1 inhibitors (64).

In preclinical studies, the combination of anti-CTLA4 and anti–PD-1 results in an approximate doubling of the tumor rejection rate compared with anti–PD-1 alone. In the tumor microenvironment of immune-checkpoint inhibitor combination-treated tumors, increased CD8⁺ cytotoxic T-cell and CD4⁺ T effector (Teff) cell tumor infiltration can be observed, along with reduced immunosuppressive T regulatory cells (Treg) and myeloid-derived suppressor cells (MDSC; refs. 65, 66). As discussed, this combination regimen has led to improved response rates and OS in metastatic melanoma, dMMR colorectal cancer, and HCC.

Building on the principle of dual checkpoint inhibition with anti-CTLA4 and anti–PD-1/PD-L1 to enhance responses, targeting other checkpoints in combination with PD-1/PD-L1 inhibition has been an area of intense investigation with the aim of improving responses and optimizing toxicity rates. The immune checkpoint indoleamine 2,3-dioxygenase 1 (IDO1) results in suppression of Teff and natural killer (NK) cells in the tumor microenvironment, and increases Treg and MDSC activity, leading to immunosuppression (67, 68). The combination of anti-CTLA4 and IDO1 inhibition in mice bearing B16 melanoma resulted in improved survival compared with anti-CTLA4 alone; however, single-agent IDO1 inhibition did not demonstrate antitumor effects in vivo (69). Similarly, in clinical trials assessing IDO1 inhibitors, single-agent therapy led to limited or no responses (70, 71). A number of phase II studies of IDO1 inhibitors in combination with anti–PD-1 therapy across solid tumors demonstrated promising response rates and reduced toxicities (13%–24% grade 3–4 toxicity) compared with CTLA4 plus PD-1/PD-L1 inhibitor combinations (72). However, research enthusiasm was significantly reduced when a phase III trial in checkpoint-naïve metastatic melanoma demonstrated no PFS benefit of the IDO1 inhibitor (epacadostat) in combination with pembrolizumab versus the PD-1 alone (mPFS: 4.7 vs. 4.9 months; ref. 73). This study also highlighted the need for a better understanding of the underlying biology, and for improved biomarkers for therapeutic efficacy and patient selection.

T cells in a chronically exhausted state can coexpress multiple checkpoints, suggesting that the targeting of a number of these, concurrently or sequentially, may be required to invoke an antitumorigenic response (74). Studies have reported that the upregulation of the TIM3 immune checkpoint correlates with resistance to anti–PD-1 agents in both in vivo and clinical samples (75–77). Early results from the combination of anti-TIM3 anti–PD-1/PD-L1 therapies have been promising. The AMBER trial of a TIM3 inhibitor in combination with an anti–PD-1 inhibitor reported partial responses in 4 of 31 patients with advanced NSCLC who had previously progressed on PD-1/PD-L1 inhibitors (78). A trial of the anti-TIM3 agent LY3321367 given as monotherapy reported one partial response out of 23 in a patient who had previously progressed on CTLA4 plus PD-1 inhibitor combination therapy, with results from the combination of LY3321367 and the PD-L1 inhibitor LY3300054 awaited (79).

The immune checkpoint LAG3 mediates the suppressive activity of Tregs and regulates T-cell expansion and homeostasis, with blockade of LAG3 resulting in an antitumorigenic T_H1 phenotype (80–82). In preclinical models, the combination of PD-1 and LAG3 blockade resulted in improved tumor rejection compared with PD-1 inhibition alone (83–85). A phase II combination trial of LAG525, targeting LAG3, and the PD-1 inhibitor spartalizumab in 121 heavily pretreated patients with a range of solid tumors demonstrated one complete response and 11 partial responses (9.9%) with combination therapy, although no responses were observed with single-agent LAG3 blockade (86). Another LAG3 inhibitor, relatlimab, in combination with nivolumab demonstrated a response rate of 11.5% in 61 patients with melanoma whose disease had progressed on prior anti–PD-1/PD-L1 therapy. This response rate appeared to be 3.5-fold higher in patients with LAG3 expression of at least 1% versus those with less than 1% LAG3 expression (87). This relatlimab combination is now being assessed in a phase II/III trial in patients with advanced melanoma in the first-line setting. In a smaller study of 15 patients with the anti-LAG3 agent MK-4280 with pembrolizumab, 4 of 15 patients achieved a partial response (88). The lack of single-agent LAG3 inhibitor activity has led to caution; however, there is a clear and urgent clinical need to understand mechanisms of resistance to current immunotherapy regimens so as to develop new options for patients who develop progressive disease on immunotherapy. The PLATforM study aims to address this critical question with the combination of LAG525 and spartalizumab, as well as spartalizumab in combination with c-MET or CDK4/6 inhibition or anti-IL1β in patients with advanced metastatic melanoma (89). This adaptive trial design represents a rational approach and may be a model for future clinical trials in the post-immunotherapy space.

The concept of dual checkpoint inhibition may lead to higher-order combinations and will be expanded upon later in this review. An emerging approach that promises further specificity in targeting multiple checkpoints is the development of bispecific antibodies (90). This evolving technology, which allows a monovalent antibody that can target two or more receptors, opens up multiple potential approaches to drug development in this space. Specifically, some of the first emerging bispecific antibodies are targeting two checkpoints such as bispecifics to PD-1 and CTLA4, allowing a more specific targeting of the CTLA4 receptor on PD-1–positive cells and promises to uncouple efficacy from toxicity of this combination. The interesting characteristic of this approach is that which receptor the bispecific has higher affinity to (e.g., PD-1) will define which cells will be targeted for inhibition of the second receptor (e.g., CTLA4), and sometimes at hundreds of fold higher affinity (ref. 91; NCT03517488).

From an immunotherapy resistance perspective, using bispecifics can allow targeting specific mechanisms of resistance such as TGFb or MET inhibition (92). For instance, a particularly interesting molecule that has a PD-L1 antibody and TGFβ-TRAP as the second part of the bispecific allows TGFβ levels—a well-described mediator of resistance in the tumor microenvironment—to be modulated favorably. This approach is currently in clinical testing alone or in combination (ref. 93; NCT 04349280). Virtually every mechanism of resistance described in this review could potentially be amenable to targeting by bispecific antibodies. Bispecifics also permit the targeting of specific cells as well. For example, T-cell engagers or BiTEs are bispecifics bearing CD3 motifs that can attach a T cell directly to a tumor cell and initiate an immune response that is MHC-independent (93). Trispecifics are also being developed with a focus on NK cells in so-called trispecific NK-cell engager therapeutics (TriNKET). Those novel agents are entering the therapeutic armamentarium and will likely be used alone or in combination with other therapeutics, including chemotherapy and TKIs, allowing multipronged targeting of immunotherapy resistance.

Delivering combination immune-checkpoint inhibition in the neoadjuvant setting takes advantage of the in situ tumor to stimulate immune responses, resulting in increased clonal expansion of T cells compared with adjuvant IO (94). Systemic immunosuppression is observed in the metastatic setting (95), and patients with metastatic disease have often experienced a prior treatment course including immunosuppressive treatments. Therefore, it could be hypothesized that treating at the earliest stage possible not only results in increased exposure to tumor antigens, but also utilizes an active, responsive immune system that may improve long-term systemic immunity.

In keeping with this, the combination of nivolumab–ipilimumab results in improved pathologic complete response (pCR) rates in resectable melanoma compared with nivolumab alone (45% vs. 25% in a 23-patient study), although with an increased toxicity cost (73% grade 3–4 toxicities vs. 8%; ref. 96). A randomized trial examined three different doses and schedules of the combination of ipilimumab and nivolumab in the neoadjuvant therapy of melanoma and concluded that two doses of ipilimumab at 1 mg/kg and nivolumab at 3 mg/kg intravenously 3 weeks apart gave the optimal outcome of more than 70% pCR rate and 30% toxicity and is the currently accepted neoadjuvant regimen (97). Similarly, in resectable colorectal cancer, neoadjuvant nivolumab–ipilimumab for 4 weeks resulted in pathologic responses in 100% of MSI cancers, and 27% of MSS disease (98). Grade 3 to 4 toxicities were experienced by 13% with one patient (out of 40) requiring infliximab for checkpoint-induced colitis. In head and neck cancer, where palliative immunotherapy has been effective, neoadjuvant nivolumab–ipilimumab resulted in responses in 69% of patients, with limited toxicity from this short course of treatment (99). Neoadjuvant combinations with chemotherapy or other standard-of-care agents are also promising, with combinations in breast cancer discussed below.

Activating the immune system using costimulatory molecules in combination with anti–PD-1/PD-L1 has been shown to lead to improved outcomes by enhancing antitumorigenic activity. For example, activation of IL2 signaling via CD122 results in activation of naïve CD8⁺ T cells and also increases NK-cell activity (100, 101). The combination of the CD122 agonist NKTR-214 and anti–PD-1 demonstrated promising results in preclinical models of NSCLC and colorectal cancer (102). In treatment-naïve metastatic melanoma, the combination of NKTR-214 and nivolumab resulted in an ORR of 59.5% (22/37), with 7 (18.9%) patients experiencing a complete response (103). This combination has now been granted the breakthrough therapy designation from the FDA, which would expedite this combination through ongoing phase III studies. Other costimulatory molecules, such as 4-1BB, OX-40, and GITR, similarly result in increased CD8⁺ and NK-cell activity and reduced Treg-mediated immunosuppression, with multiple phase I/II trials ongoing in combination with immune-checkpoint blockade (104–106).

Innate immune-stimulating agents, such as TLR9 or stimulator of interferon genes (STING) agonists, delivered intratumorally can prime the tumor microenvironment prior to immune-checkpoint therapy. A number of TLR9 agonists (CpG-oligodeoxynucleotides) are in ongoing phase I to III studies in combination with anti–PD-1 and anti–CTLA4 immune-checkpoint inhibition. A phase Ib study of the TLR9 agonist SD-101 in combination with pembrolizumab resulted in an ORR of 78% in IO-naïve patients with melanoma and 15% of checkpoint (anti–PD-1) refractory disease (107), with CD8⁺ and NK-cell trafficking to the tumor site. A similar approach with lefitolimod and ipilimumab is ongoing in an all-comer study (108). Following promising results in phase II, with 38% ORR in anti–PD-1–refractory melanoma (109), the combination of tilsotolimod and ipilimumab is now under investigation in the phase III study ILLUMINATE301 (110). Initial studies of intratumoral STING agonists have also demonstrated the ability of these agents to activate local immune responses and synergize with anti–PD-1 therapy (111, 112), although mixed results from trials to date have demonstrated the need for patient selection for these approaches, and further preclinical studies to understand potential resistance mechanisms.

Chimeric antigen receptor (CAR) T cells directly target tumor-specific antigens, independent of MHC expression, with successful outcomes in a number of hematologic malignancies (113). However, replicating this success in solid tumors has proved challenging. One potential resistance mechanism is the upregulation of immune-checkpoint expression common to solid tumors, causing CAR-T cells to become exhausted and ineffective. In mesothelioma, which has already demonstrated positive efficacy data with nivolumab–ipilimumab combination therapy (114), CAR-T cells targeting mesothelin followed by anti–PD-1 resulted in an ORR of 50% (7/14 patients; ref. 115). CAR-T cells can also be manipulated using CRISPR/Cas9 to disrupt PD-1 expression. The advantages of this approach include a potential reduction in systemic effects of PD-1 inhibition, with highly specific PD-1 knockdown only in tumor-targeted T cells, and avoiding repeated administrations of systemic anti–PD-1 (116). Both PD-1–disrupted CAR-T cell therapy and combinations with systemic anti–PD-1 therapy are the subject of an ongoing study in the early-phase setting (117).

The role of the immune system in response to chemotherapy has been noted since the 1970s, when studies assessing the mechanisms of action of anthracyclines demonstrated an improved response in immunocompetent murine models (118). Immunogenic cell death (ICD) has been proposed as a key mechanism resulting in an immune-infiltrated tumor microenvironment favorable for immunotherapy, suggesting that ICD-inducing regimens may synergize with immune-checkpoint inhibitors (119). Ongoing chemotherapy–immunotherapy combination trials are discussed in greater detail in other review articles (120). The combination of platinum chemotherapy and immune-checkpoint blockade has been successful in SCLC (22), with the immunogenic response generated by chemotherapy overcoming the lower response rates observed with anti–PD-1 treatment alone in low PD-L1–expressing tumors (31). DNA-damaging chemotherapy additionally activates the innate immune cGAS–STING pathway via cytosolic DNA, resulting in an immune-rich microenvironment that may predict responses to checkpoint blockade (121–123).

Although taxanes do not directly cause ICD, they decrease intratumoral immunosuppressive MDSCs and Tregs, augmenting cytotoxic T-lymphocyte activity (124, 125). The combination of nab-paclitaxel and atezolizumab is approved for PD-L1–positive metastatic TNBC, but was not effective in the PD-L1–negative population (126). Alternative chemotherapy regimens may further improve patient outcomes, as illustrated by the ISPY-2 studies in neoadjuvant breast cancer with pembrolizumab and standard anthracycline-based neoadjuvant chemotherapy (127). In these studies, neoadjuvant pembrolizumab and paclitaxel followed by anthracycline–cyclophosphamide resulted in pathologic complete response (pCR) rates of 60% in TNBC compared with 22% without pembrolizumab, and 30% versus 13% in hormone receptor–positive/HER2-negative disease (128). In a similar setting, four cycles of pembrolizumab in combination with paclitaxel–carboplatin chemotherapy, followed by anthracycline—cyclophosphamide, in TNBC resulted in a pCR of 64.8% versus 51.2% without pembrolizumab. In this study, patients received adjuvant pembrolizumab or placebo for a further nine cycles following resection, with follow-up ongoing (129).

DNA damage by radiotherapy is associated with a range of immunogenic effects, including activation of the cGAS–STING pathway, increasing neoantigen expression, and upregulation of PD-L1 expression (130–132). Sequential chemoradiotherapy–IO has been shown to lead to significantly improved patient outcomes in stage III NSCLC (27, 28), whereas a phase II study of concurrent anti–CTLA4 plus radiotherapy in NSCLC suggested improved responses in patients previously resistant to anti–CTLA4 given alone or in combination with chemotherapy (133). Although more than 100 trials of anti–PD-1/PD-L1 therapy in combination with chemo/radiotherapy are ongoing (134), questions remain over the optimal dose and schedule of radiotherapy–immunotherapy combinations. For example, serial low doses of radiation have been shown to activate a STING-mediated immune response, whereas higher doses (i.e., ≥12–18 Gy) instead activate TREX1, removing cytosolic DNA and preventing STING activation (135). Concurrent radiation may result in enhanced responses, but this may be dependent on the target (CTLA4 vs. PD-1/PD-L1), dose, and delivery of the radiation using proton beam or other approaches (136). Although the abscopal effect of radiotherapy may be enhanced by immune-targeting therapies, clinical evidence for this approach has thus far been largely confined to case reports rather than large-scale trials (137, 138).

DNA damage response (DDR) deficiency results in activation of the innate immune system via the critical cGAS–STING pathway, which is required for interferon expression in response to cytoplasmic DNA (124, 139, 140). A number of studies have now demonstrated cytoplasmic DNA and tumor cell–intrinsic activation of the STING pathway following treatment with agents targeting the DDR, in particular PARP inhibitors (141–144). PARP inhibition leads to activation of intratumoral dendritic cells and increased CD8⁺ T-cell infiltration via activation of the STING pathway. Interestingly, PARP inhibition also reduces the PARylation of STAT3, while increasing STAT3 transcriptional activity and expression of PD-L1 (145). These studies suggest that PARP inhibition acts synergistically with immune-checkpoint blockade in both DNA repair–deficient (e.g., BRCA1/2-mutant) and DNA repair–proficient cancers (146).

Early-phase trials of PARP inhibitor plus immune-checkpoint inhibitor combinations have been promising, with activity demonstrated in patients with advanced castration-resistant prostate cancer (147, 148), TNBC (149), and ovarian cancer (150). This combination has demonstrated antitumor activity in BRCA1/2 wild-type TNBC and platinum-resistant ovarian cancer, populations that are typically resistant to single-agent PARP inhibitors. Importantly, bone marrow suppression typical of PARP inhibitor toxicity was shown to occur at similar rates to that observed with single-agent therapy, and no significant increase was found in immune-related toxicities.

Targeting the ATR/CHK1 pathway also results in activation of a STING-dependent immune response (143), and ATR inhibitors have demonstrated promising clinical activity in combination with durvalumab (151). There is also growing evidence for a number of DDR inhibitors in combination with immune-checkpoint inhibitors, including those targeting WEE1 (152), ATM (153, 154), and DNA-PK (155). With improved understanding of the impact of these novel DDR targeting agents on the immune response in specific patient subgroups, it is tempting to propose that future DDR inhibitor combination therapies will focus on targeting specific patient populations, based on the DNA-repair capacity identified in individual tumors, in combination with immune-targeting agents designed to appropriately modulate the tumor microenvironment.

A number of targeted therapies are recognized to increase the immunogenicity of tumors and are therefore predicted to work synergistically with immune-checkpoint blockade. For example, inhibitors of the MAPK pathway result in increased expression of MHC I on the tumor cell surface, with subsequent increase in antigen-specific T effector cells and T cell–mediated cytotoxicity (121, 156). This combination has been hampered in early clinical trials by associated increases in toxicity rates, with a trial of anti-CTLA4 and the BRAF inhibitor vemurafenib terminated due to a high rate of hepatotoxicity (157). The addition of the MEK inhibitor trametinib, while reducing hepatotoxicity, resulted in colitis and bowel perforation in two of seven patients receiving this triplet combination (158). More promisingly, a phase II study of BRAF and MEK inhibition with anti–PD-1/PD-L1 therapy in treatment-naïve BRAF-mutant melanoma reported tolerability (58% grade 3–5 toxicities vs. 27% in patients receiving BRAF and MEK inhibition alone) and improved PFS (12 month PFS 59% vs. 45%; ref. 159). A phase III randomized trial in patients with unresectable stage IIIC to IV BRAF^V600-mutant melanoma assessed vemurafenib plus cobimetinib with atezolizumab versus placebo with atezolizumab. In the triplet therapy group, PFS was superior versus the reference arm (15.1 vs. 10.6 months; HR 0.78; 95% CI, 0.63–0.97; P = 0.025), with minimal increases in grade 3 to 4 toxicities (79% vs. 73%), and this study resulted in FDA approval of the triplet combination in the first-line therapy of BRAF-mutant metastatic melanoma (160). However, a phase III study of MEK inhibition with atezolizumab in heavily pretreated MSS colorectal cancer did not result in improved survival compared with standard-of-care regorafenib therapy alone (161).

Epigenetic modification by histone deacetylase (HDAC) inhibitors reduces the number of MDSCs in the tumor microenvironment in preclinical models, resulting in improved response to anti–PD-1 blockade (162, 163). In addition, treatment with HDAC inhibition leads to upregulation of the chemokines CXCL10 and CCL5 in the tumor microenvironment that, in turn, drives recruitment of antitumorigenic CD8⁺ T cells (164). A phase Ib/II trial of the HDAC inhibitor entinostat with pembrolizumab in patients with metastatic melanoma who had previously progressed on immune-checkpoint therapy demonstrated responses in 10 (18.9%) of 53 patients, including one complete response (165). Treatment with demethylating agents such as azacitdine upregulates MHC I on tumor cells and also unmasks endogenous retroviruses, resulting in cytoplasmic double-stranded RNA (dsRNA) that activates innate immune responses, increasing tumor immunogenicity (37, 166, 167). A combination study of azacitidine and nivolumab demonstrated responses in relapsed or refractory acute myeloid leukemia (168). However, a phase II study of azacitidine and pembrolizumab in MSS colorectal cancer demonstrated only one partial response out of 31 patients to the combination therapy (169).

Given our current superficial knowledge of the therapeutic effects of different combinations on the immune system, multiple preclinical tumor models have been developed to provide a deeper understanding of the mechanisms of action and to help predict the efficacy and safety of different immunotherapeutic interventions in patients with cancer. As summarized in Table 2, many proof-of-principle preclinical studies involving a wide range of immunotherapy combinations have been conducted with different murine models. The results from these studies provide a compendium of immunotherapy combinations for future clinical assessment. Although none of these murine models completely mirror the heterogeneity and adaptability of the antitumor immune network in patients with cancer, each one has its unique value in the cancer immunotherapy research field. Therefore, understanding the pros and cons of each model can guide the optimal selection of appropriate murine tumor models for preclinical assessment of different immunotherapy combinations. Current murine tumor models may be summarized into three categories based on the types of immune systems involved: syngeneic, chimeric, and humanized tumor models.

Mouse models in this category are fully immunocompetent. Tumor development in these mice is induced either by inoculation of murine cell lines or by tissue-specific activation of oncogenic pathways. Due to the general ease of use, inoculating wild-type mice with established syngeneic tumor cell lines is by far the most commonly used approach to generate syngeneic tumor models. Several inbred strains such as C57BL/6, BALB/c, 129/sv, and DBA/2 have been used to generate a variety of murine cell lines to represent different cancer types, including but not limited to B16 (melanoma; ref. 170), MCA205 (sarcoma; ref. 171), A20 (lymphoma; ref. 172), P815 (mastocytoma; ref. 173), 4T1 (breast cancer; ref. 174), 344SQ (lung cancer; ref. 175), ID8 (ovarian cancer; ref. 176), MB49 (bladder cancer; ref. 177), MC38 (colon cancer; ref. 178), and CT26 (colon cancer; ref. 179). The majority of these tumor cell lines display a certain degree of genome instability, which results in the expression of tumor-specific antigens that may be recognized by the adaptive immune system, particularly T cells (180, 181). Because mice challenged with these tumor cells have intact murine immune systems, a “physiologically relevant” tumor immune microenvironment can be achieved in these tumor-bearing mice.

However, host mice used in syngeneic cell line models are commonly young and healthy inbred mice, which are housed in specific pathogen-free conditions. In addition, there is mounting evidence supporting a high divergence in immune responses among inbred mouse strains and cell lines. For example, the C57BL/6 strain is T_H1-biased, whereas BALB/c and DBA/2 strains are T_H2-biased (182). Discrepancies in the antitumor effects of IO combinations may thus be observed depending on the murine tumor cell line model chosen. Therefore, to increase the likelihood of successful clinical translation, the use of multiple murine tumor cell line models on a range of genetic backgrounds is encouraged. Moreover, introducing fundamental cancer-driven mutations into tumor cell lines or perturbing environmental factors in housing conditions, such as diet and microbiome, should be used to more broadly recapitulate interpatient heterogeneity.

Recent advances in technologies of in vivo genetic manipulation allow us to precisely control the timing, duration, and tissue type of gene expression. Using these approaches, genetically engineered mouse models (GEMM) have been applied in immunotherapy studies to better mimic the heterogeneity of the natural steps of tumor development in patients with cancer. The Mouse Tumor Biology Database (http://tumor.informatics.jax.org) provides a comprehensive list of GEMMs for a variety of human cancer types (183). As tumor development in GEMMs is driven mainly by one or two genetic alterations in dominant oncogenes, these models generally have a low tumor mutation burden. Furthermore, de novo tumor growth in GEMMs provides sufficient time to trigger a broad range of immunosuppressive mechanisms within the tumor microenvironment. This lack of neoantigens and the profound immunosuppression make GEMMs typically highly resistant to cancer immunotherapy, sometimes even more so than clinically equivalent malignancies. For example, melanomas derived from the Tyr gene promoted BRAF-mutant PTEN-loss model (184) have been demonstrated to be resistant to immune-checkpoint inhibitors (185). Depleting Tregs also failed to control the development of tumors with transgenic expression of the RET receptor tyrosine kinase gene, suggesting that other immunosuppressive cells play an important role and replace immunosuppressive, tumor-promoting functions of Tregs (186). Therefore, effective immunotherapy strategies should consider including the inhibition of Treg migration into the tumor combined with neutralization of other immunosuppressive cells and factors in the tumor microenvironment (186). These models have now been used to address the challenges identified in patients with cancer with low immunogenic tumors or immune-resistant tumors. Peng and colleagues utilized the BRAF-mutant PTEN-loss GEMM to demonstrate that inhibition of the oncogenic activation of the PI3K pathway by PTEN loss can sensitize tumors to immune-checkpoint blockade therapy (187). Despite the expensive and time-consuming procedures required, GEMMs are currently the optimal models to use to evaluate the potential of IO combinations to overcome immune resistance, particularly resistance associated with genetic alterations.

Although syngeneic murine models provide an opportunity to characterize the immunologic changes associated with IO combinations in the presence of an intact immune system, they cannot be used to test the human version of IO reagents due to cross-species differences. To circumvent this limitation, chimeric murine tumor models have been developed by knocking in human immune-related genes. In these immunocompetent tumor models, the majority of immune compartments are murine-based, whereas the knocked-in human genes allow a portion of their immune system to be controlled by certain human immune factors. The first human knock-in model was the HLA-A2 murine strain. Mice carrying a human MHC class I gene (HLA-A2.1) were created to evaluate the antitumor activity of HLA-restricted vaccination strategies (188). Since then, further models have been generated to study clinical-grade immunomodulatory agents, including anti-human CTLA4, OX40, or Tim3 (189–191). These studies successfully confirmed the in vivo efficacy of immunomodulatory agents alone or in combination with other treatment modalities. Chimeric models are, therefore, valuable in validating combination effects of clinical-grade reagents targeting novel immune regulators in vivo, with commercial services now available to generate customized models.

To better represent tumor heterogeneity within patients with cancer, efforts have been made to generate humanized tumor models. In these models, immunocompromised mice are partially reconstituted with the human immune system by transplantation of CD34⁺ hematopoietic progenitor cells (HPC) from human umbilical cord blood, bone marrow, or peripheral blood. The implantation of tumor tissues or cell lines derived from patients with cancer in humanized host mice is used to induce tumor development. The NSG strain, which carries the NOD/SCID IL2Rγ chain knockout and lacks functional T, B, and NK cells, is most frequently used as the host strain for humanized models. Additional genetic modifications have recently been performed in NSG and other immunodeficient mice to express human cytokines, including stem cell factor, M-CSF, GM-CSF, and/or IL3, which are essential for the growth and differentiation of HPCs and to support human myeloid reconstitution (192, 193). The expression of human cytokines in these new strains, namely NSG-SGM3, NOG-ExL, MSTRG, and MISTRG, significantly increases the engraftment rate of the human immune system (194). The efficacy of pembrolizumab in a CD8⁺ T cell–dependent manner was successfully reproduced in these humanized models (195, 196).

However, antitumor immune responses in these models come largely from allogenic immune cells. Compared with natural tumor rejection through tumor-associated antigens, the antitumor immune responses mediated by allogenic rejection may be more intense. Moreover, the cost of humanized tumor models is dramatically higher than the other two types of models described above. Therefore, humanized tumor models are more suited for validating the efficacy and safety of the human versions of immunotherapy combinations than discovering novel combinations.

Highly active antiretroviral therapy (HAART) consisting of four-drug combinations that completely suppress human immunodeficiency virus (HIV) replication has evolved to become so efficacious yet tolerable that it can even be used as a routine prophylactic in “at risk” populations. HIV therapy began, however, with single, toxic drugs with limited efficacy, which gained efficacy but even greater toxicity in two-drug combinations. It was only with the inception of truly virus-specific therapeutics that high-order, high-efficacy combination therapies became practical. In cancer, high-order (three or more) immunotherapies have now begun to enter the clinic with promising initial signs against “cold” tumors. Much like HAART, however, the development of widely applicable and broadly efficacious multiple IO combinations will necessitate the development of more tumor-specific modulators of host immunity.

Due to the toxicities inherent in high-order combinations of existing immunotherapies, these trials have thus far been limited to patients with metastatic “cold” cancers (Fig. 1). In preclinical models of pancreatic ductal adenocarcinoma (PDAC), for example, only high-order combinations of chemotherapy, T-cell checkpoint blockade, and antigen-presenting cell activation via CD40 agonism could promote durable benefit and tumor regression (197, 198). This combination of checkpoint blockade, CD40 agonist antibody, and chemotherapy has recently reported promising phase I clinical trial data in PDAC, and final data are eagerly awaited (199). In this case, existing agents were combined successfully through thoughtful sequencing in order to avoid intolerable levels of additive toxicity.

In many cases, the ideal setting for high-order IO combinations would necessitate neither dose reduction to suboptimal levels nor sequencing of component therapies. To this end, multiple immunomodulatory antibodies have entered the clinic having been engineered to act selectively within the tumor microenvironment, thus sparing the host from adverse events. These novel approaches can be broadly classified into conditionally active versus bispecific antibodies. The most advanced example of conditionally active antibodies are “probodies,” in which the antigen-binding region of the antibody is blocked by a peptide that is tethered to the antibody with a linker containing multiple cleavage sites for tumor-selective proteases (200). In a phase I trial of the PD-L1 probody CX-072, only 6% of the 72 patients treated experienced a grade 3 or greater adverse event and no adverse event–related discontinuations occurred (201). The alternative approaches to tumor selectivity are bispecific antibodies in which one arm of the antibody, usually the higher affinity, binds to a tumor or tumor microenvironment selective antigen, thus localizing the effect of the lower affinity active arm. The CD137 (4-1BB) and HER2 bispecific antibody PRS-343, in which 4-1BB activation is sequestered in the tumor microenvironment by HER2 binding, is one of the most advanced examples of this approach (202). Although CD137 activation can cause liver toxicity in certain patients (4), this construct safely sequesters its activity to the tumor. PRS-343 was safe and well tolerated in the first-in-human phase I trial involving 53 HER2⁺ patients with advanced solid tumors. Two patients achieved radiologic responses, whereas other patients had stable disease. Importantly, PRS-343 demonstrated a potent increase in CD8⁺ T-cell numbers and proliferative index in the tumor microenvironment of responders, suggestive of 4-1BB agonism on T cells. By localizing the effect of checkpoint blocking and costimulatory agonist antibodies to the tumor, these approaches set the stage for tolerable combinations of multiple synergistic immunotherapies, for example, CTLA4, PD-1, and 4-1BB inhibitor combination therapy.

An alternative emergent approach to multimodal immunotherapy circumvents systemic toxicity concerns by administering some or all therapeutics directly into the tumor itself. Innate immune agonists such as activators of Toll-like receptors or the STING pathway are injected intratumorally in a single lesion to generate T-cell responses, which can then traffic to and eradicate distal sites of cancer (abscopal effect; ref. 203). Although clinical combinations with these therapies pair them with a single systemic checkpoint antibody, preclinical studies have demonstrated that as many as three immunomodulatory antibodies can be coinjected with a STING agonist, resulting in superior capacity to eliminate noninjected sites of disease (204, 205). With this type of nontoxic scaffold to build upon, current efforts can focus on identifying systemic therapies to weaken the immune resistance of uninjected sites of cancer, making it easier for the injection-mobilized T cells to eradicate them.

Although novel, less toxic immunotherapies will form the backbone of most emergent high-order IO combinations, the effect of these therapies may be augmented and extended through the addition of FDA-approved, even over-the-counter, drugs that can contribute to antitumor immunity without significant toxicities. Extracellular adenosine, for example, plays a clearly established role in immune suppression in the tumor microenvironment (206). Although still pending approval in the United States, the adenosine receptor blocking compound istradefylline is approved in Japan for the treatment of Parkinson's disease, where it has shown high tolerability (207). Such adenosine receptor antagonists may potentially be a promising addition to any immunotherapy combination (208). Multiple studies have demonstrated antitumor efficacy of inhibition of the renin–angiotensin system through decreased suppressive polarization and activation of tumor stroma (209). These drugs are commonly used to treat cardiovascular disease and would be a rational addition to immunotherapy combination regimens. Platelets have also been implicated in tumor immune suppression, which can be blocked by clopidogrel and over-the-counter aspirin (210). PDE5 inhibitors may also have value through their capacity to inhibit MDSCs (211). Further study is warranted to assess the capacity of these nontoxic, widely available drugs to augment tumor immunity; however, one or more could likely become key components of multimodal immunotherapy combination therapy.

Finally, the greatest challenge to development of these high-order combinations may not be toxicity or engineering, but rather finding a system in which the most effective combinations can be screened, so that only the most promising ones enter clinical trials. As discussed above, there are more than 2,250 trials evaluating various PD-1/L1 inhibitor combinations, a number that is increasingly straining resources and the availability of appropriate patients to evaluate potential immunotherapies. The number of possible permutations involved in developing five-drug IO combination therapies makes primary screening in patients impractical, if not impossible. In this setting, drugs must first be prioritized based on single-agent efficacy, underlying biological rationale for combination, and lack of additive toxicity. Potential combinations involving such agents can then be screened through appropriate murine tumor models of cancer to identify the most promising candidates for study in patient trials. Using such a defined development and screening approach and leveraging existing therapies, novel engineering, and selective FDA-approved drugs may provide a path to multimodal IO combinations that approach the efficacy and tolerability of HAART regimens for HIV.

Drug development to reverse resistance to immunotherapy needs to take into account multiple streams of knowledge being generated in the preclinical, translational, and clinical spaces. For instance, PD-L1 testing seems to predict a better response to immunotherapy across the board. However, due to its poor ability to discriminate antitumor efficacy in cancers such as melanoma, it has not entered clinical practice as a useful test. However, given our increasing abilities to pathologically, genomically, and transcriptomally characterize cancers, we have a mandate to now hone drug development into better-defined patient populations where therapeutic benefit can be augmented.

A recent effort initiated by the Society for Immunotherapy of Cancer (SITC) provided updated and improved clinical definitions of drug resistance to harmonize and standardize drug development efforts in the PD-1 inhibitor resistance space (212). Specifically, the SITC Resistance Taskforce defined three main clinical patterns of resistance to single-agent anti–PD-L1 antibodies: (i) Primary resistance that occurs while on treatment with single-agent therapy; this pattern requires patients to have adequate exposure to the agent (at least six weeks), to have shown no evidence of therapeutic benefit [no complete response (CR) or partial response (PR) or stable disease (SD) > 6 months], and to have a confirmatory scan to confirm progression (and predominantly to rule out pseudoprogression). (ii) Secondary resistance is defined as tumor progression after an initial clinical benefit (achievement of radiologic CR, PR, or SD > 6 months), and also requires a confirmatory scan. (iii) The third pattern defined in this effort is recurrence or progression of the tumor after halting therapy. This definition assigns the pattern to be similar to either primary or secondary resistance depending on the setting of therapy (adjuvant, neoadjuvant, or advanced disease) and the time from when therapy was discontinued (<12 weeks or >6 months).

Another clinical pattern of resistance is site-specific resistance, where patients may experience control of the disease in most of their body but have progressive lesions in one or two organs. The central nervous system (CNS) is a particularly relevant location and has emerged as a major therapeutic challenge. For instance, in melanoma, it has been reported that the CNS is the most frequent site of first progression on single-agent checkpoint blockade (213). Moving forward, it will be important to include this population of patients with progressing brain metastases in trials of novel agents, so as to enhance our understanding of the intracranial activity of immunotherapeutic agents, as well as to develop more brain-penetrant agents and approaches that tackle this most devastating complication of cancer progression. This has been recognized by the FDA, which has issued guidance to increase the representation of patients with CNS metastases on clinical trials (214).

Improved patient selection on immunotherapy combination clinical trials may seem like an aspirational goal at this time, but all the elements now appear to be falling into place, and a concerted effort between academia and industry partners could indeed accelerate this approach to achieve improved patient outcomes. Some of the examples listed above provide context to this evolving immunotherapeutic landscape. For instance, the failure of the IDO inhibitor epacadostat in a large phase III melanoma study, despite demonstrating somewhat promising data in early-phase trials, has been attributed largely to the fact that this was an unselected population, and this agent may indeed still have useful activity in a subset of patients (73, 215, 216). It is therefore encouraging to see how the development of other immunotherapy combination trials seems to be informed by that experience. An example is the anti-LAG3 antibody relatlimab and nivolumab combination already described above. The phase I data of this combination in the second-line setting indicated an ORR of 11% in unselected patients with melanoma, but was 18% in LAG3-positive patients (87). Subsequently, a phase II/III trial of relatlimab was initiated in the first-line setting in melanoma, which, although all first-line patients are still eligible, does require screening of patient tumor samples obtained within a three-month period before enrollment and for evaluable results from PD-L1 and LAG3 IHC staining to be available prior to patients being randomized (NCT03743766). Additionally, the study is stratified for both of these biomarkers, as well as BRAF mutations, which will be incredibly valuable at the time of study analysis because one can start gleaning a specific effect of the combination in a particular subset of patients. In general, such biomarker-driven approaches will of course have to be carefully designed by taking into account all the statistical caveats of subgroup analyses.

This trend is now becoming increasingly visible across the immunotherapeutic landscape, and still needs to be furthered, perhaps even more so in the immunotherapy-refractory setting in cancers where immune-checkpoint blockade has become a first-line approach. Translational and preclinical studies are continually identifying novel mechanisms of resistance to different immunotherapy agents on an ongoing basis. Investigators and pharma should be emboldened to conduct studies in molecularly characterized populations of patients where the agent in use specifically targets the biomarker used for patient selection. An example of such a strategy is the use of selective PI3Kβ inhibitors in patients harboring tumors with PTEN loss. Such a thoughtful biomarker-driven approach takes into consideration that certain key signaling networks that can be potently targeted in tumors, such as the PI3K pathway, are also critical for T-cell activation and function. Being able to identify specific molecularly targeted agents that may have a differential impact on tumor versus T cells is the new quest for a “therapeutic window.” This is especially true of novel metabolism-directed agents, such as glutaminase inhibitors (217).

There are also opportunities for the development of agents with pleotropic effects where patient selection may be challenging but have the potential to affect multiple nodes of immune activation and tolerance. Examples of those approaches are agents targeting epigenetic modulation, such as hypomethylating agents (218), HDAC inhibitors (219), and bromodomain and extraterminal domain (BET) inhibitors (220), which have genome-wide effects and affect both tumor cells and T cells.

In 2017, the FDA approved pembrolizumab for the treatment of MSI-H or dMMR regardless of primary tumor site (221), which has subsequently been supported by a randomized trial of pembrolizumab utilized for 27 different tumor types (222). However, MSI-H may have been a surrogate for high tumor mutation burden (TMB-high; ref. 223), which has subsequently been used as a biomarker for selection in both tissue- and blood-based whole-exome sequencing (224). A meta-regression analysis of 117 clinical trials for response in the TMB-high group, defined as ≥10 mutations per megabase, found single-agent PD-L1, CTLA4, and combination to significantly affect ORR (225). Prospectively, the multicohort phase II KEYNOTE-158 found an improved overall response rate of pembrolizumab in the TMB-high group compared with the TMB-low group (29% vs. 6%), leading to approval of single-agent IO for the TMB-high group (226).

In addition, there is an increasing appreciation of the role of host factors on the immune response unrelated to the intrinsic biology of tumor development and resistance. For instance, as gender and diet differences have long been observed to affect patient outcomes to cancer therapy, the microbiome has emerged as a critical element in modulating the immune response. There have been many high-impact publications indicating that the diversity and composition of the gut microbiome can indeed predict for higher response rates and better outcomes to therapy. Functional studies, as well as in vivo data, offer solid evidence that this is a causal effect resulting from the intricate interaction of gut bacteria with the immune system (227, 228). This has emerged as a novel therapeutic target, and multiple studies currently under way, in which modulation of the microbiome is being utilized with fecal microbiota transplantation, are showing early evidence of activity, including in patients with PD-1–resistant disease (229). Other interesting approaches utilizing diet to modulate the microbiome favorably are also under way.

Biomarker development continues to evolve at a rapid pace, and emphasis should continue to be placed on the incorporation of tumor biopsy and blood collection from patients on clinical trials, especially at the time of disease progression for the assessment of predictive biomarkers of resistance. This could yield a deeper understanding on how the tumor has evolved clonally and perhaps activated other pathways that could then be subsequently targeted. Although there are of course logistic, and occasionally even ethical, challenges, the data obtained from such clinical trial biopsies may also be incredibly valuable in identifying areas where the studied agent has failed in exerting its effects, offering potential pharmacokinetic and pharmacodynamic strategies to improve pharmacologic outcomes. Comprehensive and longitudinal profiling along the patient journey, within and outside a clinical trial, should become the norm rather than the exception.

T.A. Yap reports grants from Artios, Constellation, Cyteir, and Eli Lilly, Forbius, GlaxoSmithKline, Genentech, ImmuneSensor, Ipsen, Jounce, Karyopharm, Kyowa, Novartis, Ribon Therapeutics, Regeneron, Sanofi, Scholar Rock, Tesaro, and Vertex Pharmaceuticals, grants and personal fees from AstraZeneca, Bayer, Clovis, EMD Serono, F-Star, Merck, Pfizer, Repare, and Seattle Genetics, personal fees from Almac, Aduro, Atrin, Axiom, Bristol Myers Squibb, Calithera, Cybrexa, Guidepoint, Ignyta, I-Mab, Janssen, Roche, Rubius, Schrodinger, Varian, and Zai Labs outside the submitted work. E.E. Parkes reports personal fees from Boehringer Ingelheim outside the submitted work. J.T. Moyers reports nonfinancial support from Astellas Pharmaceuticals outside the submitted work. M.A. Curran reports personal fees from ImmunoGenesis, Inc., Agenus, Inc., Alligator Bioscience, Inc., Ammunix, Inc., Aptevo, Inc., and ImmunOs, Inc., Oncoresponse, Inc., Pieris, Inc., Nurix, Inc., Servier, Inc., Kineta, Inc., Xencor, Inc., and Salarius, Inc., and grants and personal fees from ImmunoMet, Inc., during the conduct of the study; in addition, M.A. Curran has a patent for antibodies that bind PD-L1 and prevent its binding to PD-1 pending and licensed to ImmunoGenesis, Inc., a patent for dual specificity antibodies which bind both PD-L1 and PD-L2 and prevent their binding to PD-1 pending and licensed to ImmunoGenesis, Inc., a patent for antibodies which bind PD-L2 and prevent its binding to PD-1 pending and licensed to ImmunoGenesis, Inc., and a patent for methods and composition for localized secretion of anti-CTLA4 antibodies issued and licensed to multiple licensees. H.A. Tawbi reports grants and personal fees from Bristol Myers Squibb, Merck, Novartis, and Genentech, grants from GSK, Celgene, and personal fees from Eisai and Iovance outside the submitted work. No other disclosures were reported.