1. Enhancing the Efficacy of Checkpoint Blockade Through Combination Therapies

CTLA-4 (Cytotoxic T Lymphocyte Antigen 4) was the first T cell co-receptor identified as inhibitory, and has a critical role in maintaining immune tolerance (Tivol et al. 1995; Waterhouse et al. 1995; Brunet et al. 1987). CTLA-4 also was the first coinhibitory receptor to show therapeutic promise in mouse models of cancer when blocked with a monoclonal antibody (Leach et al. 1996). Accordingly, CTLA-4 was the first coinhibitory receptor to be targeted in clinical trials and eventually approved for clinical use (Hodi et al. 2010). It is perhaps therefore surprising that the mechanism(s) by which CTLA-4 controls T cell responses remains controversial.

CTLA-4 is inducibly expressed upon activation of naive T cells (CD4⁺FoxP3⁻ and CD8⁺) and constitutively expressed on suppressive regulatory T cells (CD4⁺FoxP3⁺, Tregs) (Alegre et al. 1996). CTLA-4 has both cell-intrinsic and cell-extrinsic functions (Grosso and Jure-Kunkel 2013). It binds to the same ligands (B7-1 and B7-2) as the costimulatory receptor CD28, but with a higher affinity (van der Merwe et al. 1997; Freeman et al. 1992; Freeman et al. 1993a; Freeman et al. 1993b). CTLA-4 inhibits the activation of naïve T cells and is a critical mediator of Treg cell suppressive function (Wing et al. 2008). CTLA-4 can inhibit T cell activation intrinsically, either by outcompeting CD28 or by recruiting phosphatases to the cytoplasmic domain of CTLA-4 upon ligation, leading to decreased TCR and CD28 signaling (Marengère et al. 1996; Grohmann et al. 2002). CTLA-4 also may inhibit activation of other T cells in a cell extrinsic fashion by leading to downregulation of B7-1 and B7-2 on antigen-presenting cells (APCs), either indirectly (through cytokines such as IL-10) or directly (transendocytosis) (Grohmann et al. 2002; Chen et al. 1998; Qureshi et al. 2011). Whereas other coinhibitory molecules seem to exert their influence primarily at the site of immune response, CTLA-4 is critical for initial activation of T cells in secondary lymphoid organs. Mice that lack CTLA-4 develop a fatal systemic inflammatory phenotype within 2–4 weeks of birth, characterized by uncontrolled T cell expansion (Tivol et al. 1995; Waterhouse et al. 1995). Analogously, humans with heterozygous CTLA-4 mutations have increased susceptibility to severe immune dysregulation (Kuehn et al. 2014; Schubert et al. 2014; Topalian and Sharpe 2014). Therefore, although questions remain about mechanisms, it is clear that CTLA-4 plays a critical role in attenuating T cell responses.

The seminal work demonstrating that CTLA-4 blockade can promote antitumor immune responses in mouse tumor models was performed in James Allison’s lab in the mid-1990s (Leach et al. 1996). Since then, the basic science exploring the specific mechanisms underlying this efficacy has proceeded in parallel with the clinical development of antibodies that have a similar effect. Blockade of CTLA-4 on both effector T cells and regulatory T cells is necessary for optimal antitumor immunity, suggesting multiple modes of action (Peggs et al. 2009). In some, but not all mouse models, CD4⁺ T cells are necessary for therapeutic benefit with CTLA-4 blockade, whereas CD8⁺ T cells appear to always be necessary (Hurwitz et al. 1998; van Elsas et al. 1999). This is likely because cytotoxic CD8⁺ T cells are responsible for killing tumor cells, an important mode of antitumor immunity. Recent reports suggest that depletion of intratumoral regulatory T cells (the highest expressors of CTLA-4) may be a critical part of the therapeutic mechanism (Bulliard et al. 2013; Selby et al. 2013; Simpson et al. 2013).

The relevance of many of these findings to human cancer patients is an active area of investigation. Because of the challenges of obtaining fresh tumor samples from patients, many clinical studies have focused on peripheral blood samples. A recent study of blood samples from patients undergoing anti-CTLA-4 therapy showed an increase in new tumor-reactive T cell clones, but no change in preexisting clones (Kvistborg et al. 2014). This suggests that anti-CTLA-4 has a major effect on T cell priming (i.e., in the lymph nodes). Although studies such as this are informative, the phenotype of tumor-infiltrating lymphocytes (TILs) can differ drastically from peripheral sites (draining lymph node, blood, etc.) (Ahmadzadeh et al. 2009). Investigators are now designing protocols to obtain serial tumor biopsies before and during treatment. Initial studies of this sort show that anti-CTLA-4 increases T cell responses within the tumor (Cooper et al. 2014). In addition, mutational load in tumors prior to therapy has been correlated with response to anti-CTLA-4 in patients, suggesting that this therapy requires an immunogenic tumor (i.e., it can be recognized by T cells) (Snyder et al. 2014).

It is important to note that, while CTLA-4 blockade can induce striking antitumor immunity in some patients, many patients also experience immune-related adverse events (irAEs), generally manifesting as tissue-specific inflammation (Teply and Lipson 2014). This can be managed by administration of immunosuppressive drugs, and temporally or permanently stopping anti-CTLA-4 therapy. A better understanding of the mechanism of action of anti-CTLA-4 antibodies is needed to minimize the associated toxicities and increase the fraction of responding patients.

The PD-1 (programmed death 1) pathway plays a critical role in regulating peripheral T cell tolerance and protects healthy tissues from inflammatory tissue damage (Topalian et al. 2015; Flies et al. 2011; Francisco et al. 2010a; Ostrand-Rosenberg et al. 2014). Similarly to CTLA-4, PD-1 is induced on naïve T cells upon activation, and engagement by either of its ligands PD-L1 (B7-H1) or PD-L2 (B7-DC) sends inhibitory signals into T cells (Keir et al. 2008). This can attenuate the initial activation of T cells in the lymphoid organs, or it can diminish effector T cell functions in tissue (Keir et al. 2007). In the setting of an acute viral infection, PD-1 is upregulated on T cells but returns to baseline upon viral clearance (Petrovas et al. 2006). In contrast, in the setting of a chronic infection and cancer, T cells that are continuously exposed to antigen express high levels of PD-1 and eventually become dysfunctional (Wherry 2011). PD-1 is also constitutively expressed on Tregs, and affects their suppressive capacity (Francisco et al. 2010b). Expression of PD-1 ligands dampens the local immune response and controls resolution of inflammation after clearance of microbes. PD-L1 is constitutively expressed on many immune cell types and its expression can be induced on many nonimmune cells (including epithelial cells, vascular endothelial and stromal cells) by pro-inflammatory cytokines (e.g., Type I and II IFNs, TNFα, VEGF) (Keir et al. 2007). PD-L2 is expressed primarily by APCs, induced by many of the same cytokines as PD-L1, but IL-4 and GM-CSF are the most potent stimuli for PD-L2 expression (Francisco et al. 2010b). Tumor cells can express one or both PD-1 ligands, as do many of the other cells in the tumor microenvironment (e.g., fibroblasts, endothelial cells, immune cells) (Flies et al. 2011). Multiple and reinforcing mechanisms drive PD-L1 and PD-L2 expression on tumor cells, including oncogenic signaling (e.g., AKT), amplifications and/or translocations of chromosome 9p24 (which contains PD-L1 and PD-L2), and Epstein Barr Virus latent membrane protein 1 (LMP1). This constitutive expression of PD-L1 is termed “innate immune resistance.” In addition, PD-L1 expression can be induced on tumor cells in response to T cells producing immunostimulatory cytokines (such as IFNs) (Spranger et al. 2013). This is termed “adaptive immune resistance,” and represents a mechanism by which tumor cells attempt to evade immune mediated antitumor responses (Spranger et al. 2013; Spranger et al. 2015; Taube et al. 2012). Thus, there are multiple means by which PD-1 signaling may contribute to the immunosuppressive tumor environment.

Ligation of PD-1 negatively regulates T cells in several ways. First, engagement of PD-1 diminishes the signals downstream of TCR stimulation, and facilitates the downregulation of the TCR itself, leading to decreased activation and cytokine production (Karwacz et al. 2011; Sheppard et al. 2004). Second, PD-1 ligation can induce genes that impair T cell proliferation and cytokine production (e.g., the transcription factor batf) (Quigley et al. 2010), and can decrease anti-apoptotic gene expression and increase pro-apoptotic gene expression (e.g., bcl-xl), reducing T cell survival (Gibbons et al. 2012; Parry et al. 2005). Third, PD-1 signaling can decrease the production of cytotoxic molecules by T cells, decreasing their killing capacity (Azuma et al. 2008; Barber et al. 2006). Fourth, the PD-1 pathway can promote the induction of regulatory T cells from naïve or Th cells (iTregs), which suppress effector T cells (Francisco et al. 2009). Therefore, PD-1 signaling can modulate T cells in multiple ways that synergize to suppress immune responses.

Many important questions remain about the mechanisms that underlie antitumor immunity induced by PD-1 blockade. As a monotherapy, PD-1 blockade works in ~20–50 % of patients, but we do not understand the molecular nature of an effective antitumor response, nor know how to identify patients who will respond to PD-1 blockade. Furthermore, we do not understand why patients fail to respond to anti-PD-1/PD-L1. This knowledge is needed for rational combination of PD-1 pathway blockade with other therapies to treat patients who do not respond to anti-PD-1 or anti-PD-L1 alone.

Recent work has begun to address these questions. Several papers have noted that PD-1 is more frequently expressed on tumor-specific TILs than on the bulk TIL population, and that the TILs from patients who respond to PD-1 blockade are more clonal prior to therapy than in patients that do not respond (Ahmadzadeh et al. 2009; Gros et al. 2014; Tumeh et al. 2014). TIL clonality is often used as a measure for antigen-specific T cell responses, as T cells proliferate in response to their cognate antigen (e.g., clonal expansion). Similar to CTLA-4 blockade, multiple studies have shown that a higher mutational burden—specifically, non-synonymous mutations—is associated with a better response to PD-1 blockade (Messina et al. 2012; Le et al. 2015a). These nonsynonymous mutations likely give rise to neoantigens and induce tumor-specific T cells. Similarly, multiple studies have shown that response to PD-1 pathway blockade is more frequent in the setting of PD-L1 expression in the tumor microenvironment (on either tumor cells or nontumor cells, or both) (Larkin et al. 2015; Taube et al. 2012; Garon et al. 2015). It is important to note, however, that responses to PD-1 checkpoint blockade can occur even when PD-L1 expression is not observed, and that PD-L1 expression is not needed for response to combined PD-1 and CTLA-4 therapy. It is likely that a high mutation rate and PD-L1 expression are in fact linked, as a high mutation rate correlates with a higher CD8⁺ T cell infiltration into tumors, which produce cytokines that can lead to increased PD-L1 expression in the tumor (i.e., adaptive immune resistance) (Spranger et al. 2013). Indeed, initial observations support the link between mutation rate and PD-L1 expression, but further work is needed to explore this concept (Le et al. 2015a).

The success of anti-CTLA-4 and anti-PD-1 in the clinic has led to the search for other checkpoint molecules that can be targeted, especially in cases where anti-CTLA-4 or anti-PD-1 immunotherapy has little or no effect. There are many known checkpoint molecules (Table 1.2), and understanding the immunoregulatory roles of these molecules in cancer is an active area of investigation. Therapeutic targeting of several of these molecules has progressed to clinical trials (Shin and Ribas 2015).

One promising target is LAG-3 (lymphocyte-activation gene-3) (Goldberg and Drake 2011). LAG-3 is expressed on effector and regulatory T cells, B cells, NK cells and plasmacytoid dendritic cells, and binds with high affinity to MHCII to negatively regulate T cell responses (Shin and Ribas 2015). Whereas mice deficient in LAG-3 alone or PD-1 alone are generally healthy, mice lacking both LAG-3 and PD-1 develop systemic autoimmunity, highlighting the synergy between these two pathways in controlling T cell tolerance (Okazaki et al. 2011; Woo et al. 2012). T cells in tumors can co-express PD-1 and LAG-3, and combined blockade of PD-1 and LAG-3 in tumor models has a greater therapeutic benefit than blockade of either alone (Woo et al. 2012; Matsuzaki et al. 2010). These findings have given impetus to clinical trials with anti-LAG-3 antibodies (Shin and Ribas 2015).

Another promising target is TIM-3 (T-cell immunoglobulin and mucin-domain containing-3), which is expressed on T cells (Th1 and Tc1 cells) as well as on monocytes, macrophages and dendritic cells (Anderson 2014). TILs that express PD-1 can also express TIM-3; the CD8+PD-1⁺TIM-3⁺ population is thought to be more dysfunctional than the CD8+ PD-1⁺TIM-3⁻ population (Fourcade et al. 2010; Sakuishi et al. 2010). TIM-3 binds to galectin 9, HMGB1, and phosphatidylserine (Chiba et al. 2012; DeKruyff et al. 2010; Zhu et al. 2005). Tumor cells and other cells in the tumor microenvironment, such as myeloid-derived suppressor cells (MDSCs), can express galectin-9 (Compagno et al. 2013). Signals through TIM-3 can lead to T cell death or tolerance (Zhu et al. 2005). In addition to impairing effector T cell function, TIM-3 expression on Tregs may enhance their suppressive capacity (Gautron et al. 2014) and TIM-3 on dendritic cells may decrease their ability to present antigens to stimulate T cells (Nakayama et al. 2009). Thus, TIM-3 may regulate innate and adaptive cells in the tumor microenvironment. Blockade of TIM-3 alone can induce antitumor immunity in some mouse models, and has shown potent synergy with PD-1 blockade (Sakuishi et al. 2010).

Another emerging coinhibitory receptor in cancer immunotherapy is TIGIT (T cell immunoreceptor with Ig and ITIM domains), which is expressed on T cells upon activation, as well as on NK cells (Shin and Ribas 2015). TIGIT shares the ligands CD155 and CD112 with CD226, but TIGIT ligation leads to T cell inhibition while CD226 ligation leads to T cell stimulation (Lozano et al. 2012). A recent study showed that co-blockade of TIGIT and PD-L1 led to tumor regression in a tumor model whereas blockade of either alone had minimal effect, suggesting a synergy between these pathways (Johnston et al. 2014).

The major goal of combining checkpoint blockade with other therapies is to increase the fraction of patients who respond to therapy, while maintaining the durability of response and minimizing the toxicity. In order to analyze the infinite space of potential combination strategies (i.e., therapy, timing, dosing) with a finite set of resources (i.e., money and clinical trials), it is imperative to take a principled approach to testing combinations. It is likely that the optimal combination will differ between different cancers and even between patients with the same clinical diagnosis. Thus, it is critical to gain mechanistic insights into why certain combinations succeed or fail, rather than simply evaluating response. These insights will allow clinicians to confidently choose specific combinations for each patient.

The goals of checkpoint blockade are to either stimulate new antitumor T cells to attack tumors or to relieve suppression of existing antitumor T cells, or both (Fig. 1.1). In cases where there is a lack of clinical response to checkpoint blockade monotherapy, there is likely a failure in one or both of these goals, or it may be that antitumor T cells alone are not sufficient to control the tumor. This may be the case in particularly proliferative tumors, such that the growth rate outcompetes the rate of killing by T cells, in tumors with a relatively low mutation burden so there are few antigens to drive T cell responses, or in tumors that create a particularly suppressive microenvironment. Thus, combining checkpoint blockade with therapies that overcome these three possible failure modes may lead to synergies (Fig. 1.2). Therapies to combine with checkpoint blockade should: (1) increase the number of antitumor T cells (i.e., increase antigenicity of tumor, induce immunogenic cell death), (2) enhance the intrinsic function of antitumor T cells (i.e., block other suppressive pathways, upregulate stimulatory pathways), (3) extrinsically support antitumor T cell responses to tumors (i.e., stimulate other immune cell types that can promote antitumor immunity), and/or (4) simply debulk the tumor to allow the immune response to catch up to tumor growth.

Over the last decade, many studies have examined the effect of therapies not traditionally thought to affect the immune system on immune responses (Table 1.3). Many therapies can affect the immune system including targeted therapies (epigenetic modifiers, antiangiogenic agents, small molecules targeted at specific mutations), radiation therapy, and chemotherapy. These findings suggest that multiple types of cancer therapies have the potential to synergize with checkpoint blockade. In this section, we will discuss the principles behind specific combinations that are being studied, as well as summarize insights gained from preclinical and clinical studies of these combinations.

One approach is to combine checkpoint blockade with other immunotherapies. Multiple immunotherapies (e.g., vaccines, T cell stimulatory agents, T cell transfer) have been extensively studied, and many may target one or more of the three failure modes of checkpoint blockade. However, without understanding the precise mechanisms underlying checkpoint blockade, it is currently difficult to predict whether these therapeutics will be synergistic or redundant, or perhaps even negate one another. Importantly, targeting multiple axes of the immune response may lead to an increased frequency or severity of autoimmune side effects. Thus, careful preclinical evaluation of these combinations is critical.

It has become increasingly clear that therapies initially thought to target tumor cells specifically also can profoundly affect immune responses (Table 1.3). This understanding provides rationale for combining these therapies with checkpoint blockade in certain settings. In addition, since checkpoint blockade can take weeks to months to have a clinical effect, in some situations it is necessary to start with another therapy that can debulk the tumor, or at least slow tumor growth.

Many combination checkpoint therapies are making their way to the clinic. Due to the large number of FDA-approved cancer therapies and the vast number currently in clinical trials, it is imperative that the choice of combination therapies to test in patients be rational. A failure to do so may lead to numerous disappointing trial results and exhaust resources and patient populations necessary for finding effective combinations. Towards this goal, we suggest an approach for testing combinations that is based on data from currently ongoing immunotherapy studies and grounded in a sound molecular understanding of the pathways to be targeted: (1) identify the patients most likely to benefit from combination therapy, (2) characterize the mechanisms by which tumors resist immunotherapy, and (3) determine strategies to target these resistance pathways effectively.

Optimal combination therapies will require analyses of the tumor microenvironment of each patient. The emerging principle that checkpoint blockade likely has efficacy only where there is already an ongoing (albeit failing) anti-immune response pre-therapy suggests that this is a primary aspect of the tumor that should be studied. Analytic approaches are needed to distinguish between a lack of an immune response and an active suppression of an immune response. The former type of patients would likely benefit from therapies that promote an immune response (e.g., vaccination, induction of immunogenic cell death), whereas the latter would benefit from therapies that remove suppression (e.g., combination checkpoint blockade, depletion of suppressive cells). These analytic strategies may also benefit patients who develop resistance to checkpoint blockade.

A major impediment to the principled combination of checkpoint blockade antibodies with other therapies is an incomplete understanding of their mechanisms of action. Rapid progress is being made in this area. In the clinical setting, tumor biopsies and resections obtained before and after treatment from the same patient are more valuable than equivalent samples obtained from different patients (Cooper et al. 2014). Collaborative efforts (between clinicians and scientists) have the potential to maximize the information obtained from these samples. In the preclinical setting, care must be taken in selection of animal models for studying therapeutic mechanisms of action. Many mouse tumor models do not respond to checkpoint blockade monotherapy (Grosso and Jure-Kunkel 2013). It is important to understand the differences between models that respond and those that do not, and utilize the appropriate models to study mechanism of response and resistance to checkpoint blockade.

For CTLA-4 and PD-1 checkpoint blockade, several fundamental questions remain. The relative effects of anti-CTLA-4 and anti-PD-1 on effector versus Treg cells versus other PD-1 expressing cells (NK cells, myeloid cells) are incompletely understood, and this knowledge may suggest approaches for combination therapy. For example, whereas blockade of PD-1 enhances the function of effector T cells, genetic loss of PD-1 in mouse models actually increases the suppressive capacity of at least a subset of Tregs (Sage et al. 2013), which could in turn inhibit antitumor immunity. Similarly, dissecting the role of PD-L1 and PD-L2 on specific cell types (e.g., immune cells versus tumor cells versus stromal) may lead to mechanistic insights as well as biomarkers that predict response. Future studies also should address the relative effects of checkpoint blockade in the periphery (lymph node, blood) versus the tumor microenvironment, as this information not only will inform the search for biomarkers, but also suggest combination therapeutics. Finally, a molecular understanding of the changes that occur in T cells following checkpoint blockade will greatly refine the combination therapies being explored to enhance T cell function.

Multiple checkpoint blockade antibodies are now available to clinicians but to optimize their therapeutic impact, a better understanding of the similarities and differences between coinhibitory pathways, as well as mechanisms of synergy between coinhibitory pathways is needed. Studying biopsy samples for the relative expression of coinhibitory molecules on T cells and their ligands in the tumor microenvironment will likely help inform the choice of which agent(s) to use, although how to make this choice is still an open question. Insights may come from determining whether synergies between coinhibitory pathways (e.g., PD-1 plus CTLA-4 vs. PD-1 plus LAG-3 vs. PD-1 plus TIM-3) affect similar or different molecular pathways.

While the majority of studies have focused on mechanisms of therapeutic efficacy, relatively little is known about mechanisms of durability and thus questions remain about the necessary length of therapy. Ipilimumab is given in only four doses over 12 weeks, but anti-PD-1 agents are given either every 2 weeks or every 3 weeks indefinitely. Despite the finite length of therapy, the response to ipilimumab seems to be durable, as responders from the original clinical trial remain in remission or with stable disease for many years (Schadendorf et al. 2015). However, because blockade of CTLA-4 and PD-1 likely work by different mechanisms, the durability seen with CTLA-4 blockade does not imply that blockade of PD-1 will have a similar durability. Further work is needed to understand how CTLA-4 and PD-1 blockade affect the generation, function and maintenance of memory T cell subsets. Updates on patients from early clinical trials will begin to answer the question of durability.

The identification of resistance mechanisms to checkpoint blockade and the development of strategies to circumvent resistance are important steps toward increasing the proportion of patients that respond to therapy. Studies in mouse tumors models in which checkpoint blockade induces tumor regression in certain settings but not others (e.g., larger tumor burden) should provide insights into general principles of therapeutic nonresponsiveness. The potential of such work is illustrated by a recent study showing that elevated beta-catenin signaling in melanoma tumors suppresses spontaneous T cell infiltration into tumors, which is likely a prerequisite for a strong response to checkpoint blockade monotherapy (Spranger et al. 2015). Temporal studies of patient samples (tumor biopsies from patients with recurrence and blood samples from all patients) obtained at multiple time points during checkpoint blockade (during response versus resistance) will be valuable for investigating resistance mechanisms.

Moreover, analyses of immune alterations within human tumor cells and the tumor microenvironment have the potential to identify resistance mechanisms. For example, recently published data from The Cancer Genome Atlas (TCGA) project suggest that tumors may harbor mutations that increase the likelihood of resistance to immunotherapy. Tumors with an increased infiltration of cytolytic effector cells (CD8⁺ T cells, NK cells) are also more likely to harbor mutations in MHC-I complex subunits, which may impair antigen presentation or in molecules of the extrinsic cell death pathway, which may limit effectiveness of cytolytic effector cells (Rooney et al. 2015). Mutations such as these may be driving the incomplete responses observed in patients and would thus represent a tumor in which another therapy should be pursued.

Additionally, immunophenotyping of patient tumors that fail to respond completely could reveal the increased presence of suppressive cell populations that are preventing the successful reinvigoration of T cells. For example, class-I HDAC inhibitors deplete myeloid-derived suppressor cell (MDSC) populations disproportionately both in vitro and in vivo and synergize with checkpoint blockade, thus representing a strategy for specifically depleting suppressive populations in the tumor microenvironment (Kim et al. 2014).

Since immune cell types in addition to T cells can modulate the tumor microenvironment, there is great potential for strategies that combine manipulation of these immune cell types with checkpoint blockade. Here we will discuss principled combination of dendritic cells (DCs), natural killer (NK) cells, and myeloid cell modulation with checkpoint blockade.

Studies of DCs have focused mainly on their use in vaccines, as well as approaches to manipulate endogenous DC populations. The goal of both approaches is a more potent CD8⁺ T cell response to the tumor, as discussed above. Recent advances in rapid prediction and synthesis of tumor-specific peptide sequences following sequencing of a patient’s tumor should enable development of more effective tumor vaccines (Rajasagi et al. 2014). Further work is needed to develop strategies to enhance recruitment and maturation of DCs following peptide vaccination, as well as to improve tumor peptide formulations to promote epitope spreading, the process by which an ongoing immune response to a particular antigen results in “spreading” of the targeted immune response to other antigens. One recent advance involved the preconditioning of a vaccine site with tetanus toxoid in order to enhance DC migration to the lymph node upon vaccination (Mitchell et al. 2015). This approach improved survival of patients with glioblastoma multiforme in a clinical trial. Approaches such as these should be considered in tumors with relatively few infiltrating T cells to induce an immune response that can be supported by combination with checkpoint blockade therapy.

Natural killer (NK) cells are another immune population under intense investigation since NK cells can directly kill tumor cells and mediate the effects of antibody-dependent cellular cytotoxicity. NK cells are particularly important in tumors that downregulate MHCI in an attempt to avoid CD8⁺ T cell responses, as NK cells attack cells with low MHCI expression. Although NK cells express many of the same coinhibitory receptors as T cells, little is known about the contribution of NK cells to the tumor immunity induced by checkpoint blockade. Intriguingly, a recent study demonstrated that Cbl-b was responsible for regulating NK cell mediated antitumor immunity and that knockdown of Cbl-b induced impressive spontaneous rejection of tumors (Paolino et al. 2014). Further work is need to determine how to modulate NK cell proliferation and effector function alone or in combination with antibodies designed to enhance ADCC or with T cell immunotherapies.

Myeloid cells are the most common immune cell subset in many tumors and often associated with an immunosuppressive M2 macrophage phenotype or myeloid derived suppressor cell phenotype (Lemke and Rothlin 2008; Ruffell et al. 2012). Myeloid derived suppressor cells are a Gr-1⁺ immature macrophage or dendritic cell subset characterized by their expression of ARG1 and iNos (Youn and Gabrilovich 2010). Myeloid cells in the tumor microenvironment can exert suppressive effects through the expression of PD-L1 as well as the production of immunosuppressive molecules such IL-10, TGFβ, and nitric oxide (Youn and Gabrilovich 2010). There are a number of strategies under investigation to overcome immunosuppressive effects of myeloid cells including repolarizing them to an M1 phenotype (as is seen in the setting of CSF1R blockade) (Zhu et al. 2014), or blocking their recruitment to the tumor (e.g., by disrupting the CXCR2: CXCL8 axis) (Highfill et al. 2014). In addition, macrophages have an underappreciated role in interfacing with endothelial cells, angiogenesis, and the maintenance of tumor cell vasculature. Tie2⁺ macrophages promote the production of new endothelial cells, which aids in the process of angiogenesis. Perturbation of these interactions decreases tumor cell vasculature and the ability of tumor cells to thrive in a hypoxic environment. Macrophages also can scavenge dead/stressed tumor cells in a non-immunogenic fashion, and strategies to increase the immunogenicity of the tumor through the accumulation of tumor-derived HMGB1 and ATP, two well known immunostimulatory signals in the tumor microenvironment, are under investigation. Lastly, the modulation of the local humoral response may provide a distinct means to promote tumor cell killing by skewing B cells to produce antibodies that participate in complement fixation, ADCC, or FcR-mediated phagocytosis. These approaches are likely to synergize with PD-1 blockade given the distinct mechanisms of action.

The effects of non-immunotherapeutic approaches on the immune response must be closely examined in order to determine how to optimally combine them with checkpoint blockade. Chemotherapy, targeted therapies, and radiotherapy are often not as effective at inducing tumor regression in the absence of functional immunity (Cooper et al. 2014; Zitvogel et al. 2008). It is clear that innate and adaptive immunity play a critical role in mediating the effects of various non-immunologic cancer therapies (Table 1.3). Traditional therapeutics that elicit immunogenic cell death and increase priming of adaptive immunity, such as chemotherapy with anthracyclines or radiotherapy may synergize with checkpoint blockade. Surgical resection can relieve systemic immunosuppression and perhaps restore T cell function (Danna et al. 2004; Salvadori et al. 2000). A better understanding of the effect of surgical resection of accessible lesions on the immune response to other lesions is needed.

Current data indicate that when a patient responds to checkpoint blockade therapy, the response is most often durable (McDermott et al. 2014). However, across different cancer types, sizeable fractions of patients do not respond (Table 1.2). Thus, it is imperative to develop biomarkers that help stratify patients and predict whether a given patient is likely respond to monotherapy, should receive some type of combination therapy, or receive other therapies entirely. To date, three types of biomarkers have been studied intensively, particularly in the context of PD-1 blockade: a high degree of immune infiltrate in the tumor—specifically CD8⁺ T cells, a high mutational burden/predicted neoepitopes, and expression of PD-L1 by tumor cells or tumor immune cell infiltrates (Snyder et al. 2014; Tumeh et al. 2014; Le et al. 2015a; Herbst et al. 2014; Rizvi et al. 2015a).

Work of Jerome Galon demonstrated the benefit of the immunoscore on patient survival (Anitei et al. 2014). The immunoscore is being investigated for its predictive value for a patient’s response to checkpoint blockade. Moreover, with the development of multicolor multiplexed IHC, it is now possible to evaluate CD8⁺ T cell numbers, as well as their functional status and location within the tumor simultaneously. This approach may lead to identification of biomarkers that will predict whether checkpoint blockade monotherapy will be sufficient to induce an objective response. Ectopic lymphoid follicles are highly correlated with an increase in overall survival in melanoma, breast, and colorectal cancer patients (Messina et al. 2012; Bindea et al. 2013; Gu-Trantien et al. 2013; Huang et al. 2015). The association of these structures with an objective response to checkpoint blockade remains to be investigated.

A higher mutational burden within tumor cells correlates with a better response to checkpoint blockade (Snyder et al. 2014; Le et al. 2015a; Rizvi et al. 2015a). This was initially observed by comparing response rates and mutational burden across different types of cancers, but also is seen in patients with the same type of cancer (e.g., non-small cell lung cancer, colorectal cancer), but with different mutational burdens (Le et al. 2015a; Rizvi et al. 2015a). Together, these data further the hypothesis that tumors that elicit a strong CD8⁺ T cell response to tumor antigens (e.g., neoantigens) respond better to checkpoint blockade monotherapy than tumors without many mutations and therefore without many tumor antigens. If this is indeed the case, then analysis of the mutational burden of tumor cells, as well as the T cell repertoire in the tumor prior to therapy, may be useful for identifying patients more likely to respond to checkpoint blockade and patients who likely need another intervention prior to checkpoint therapy.

High expression of PD-L1 on tumor cells increases the likelihood of a response to PD-1 pathway blockade, but durable responses are seen in patients with little or no PD-L1 expression. It remains unclear whether optimization of PD-L1 immunohistochemistry will enable better predictive value of PD-L1 expression. Further work is needed to investigate intratumoral heterogeneity of PD-L1 expression, as well as relative expression of PD-L1 in primary and metastatic tumors.

New biomarkers may result from other approaches under investigation. Quantification and characterization of circulating tumor cells in the blood might allow for less invasive analysis of tumor cells, along with real-time monitoring or response to therapy. Similarly, more sensitive tumor DNA detection techniques allow for the real time monitoring of tumor burden in plasma (Anderson 2014). These methods may provide a means to probe the tumor microenvironment via a blood-based assay and enable rapid determination of the efficacy of checkpoint blockade in patients.

The major goal of combining checkpoint blockade with other therapies is to increase the fraction of patients that have a durable response while minimizing irAEs. The increased rate of irAEs seen with the combination of nivolumab (anti-PD-1) with ipilimumab (anti-CTLA-4) may limit this combination to only otherwise healthy patients, despite the increased durable response rate seen with this therapy. New strategies that target multiple immunosuppressive pathways in parallel and that target these pathways in specific cells, as well as new strategies to study the antitumor immune response are under development and may overcome these limitations.

One potential method for reducing irAEs that result from systemic activation of the immune response is to use targeted antibodies, such as bispecific antibodies or antibodies with modified affinity, both with the goal of targeting specific immune cell populations. Bispecific antibodies have two targets rather than one (as with traditional antibodies). Bispecific antibodies have been used to bring T cells and tumor cells together, but they could be designed to bring T cells and dendritic cells together, draw NK cells to the tumor, and to more specifically target a cell (e.g., a CD8β/PD-1 bispecific antibody that blocks PD-1 specifically on CD8⁺ T cells to increase their effector function). Another avenue for more specific antibody targeting of immune cells in the tumor involves development of antibodies with a modified affinity such that they only bind cells expressing very high levels of a target molecule.

Many efforts are underway to target multiple coinhibitory pathways (e.g., PD-1, CTLA-4, TIM-3, and LAG-3 in pairwise combinations) in preclinical mouse models and clinical trials. A better understanding of the molecular pathways triggered by coinhibitory receptors is needed to determine shared and unique signaling nodes. This knowledge may reveal new therapeutic targets and strategies. Small molecule inhibitors that target a shared node might potentially replace administration of multiple checkpoint antibodies; this approach may efficiently invigorate dysfunctional CD8⁺ T cells and potentially decrease the side effects associated with multiple checkpoint inhibition, in addition to being easier to administer and likely less expensive. Furthermore, in engineered T cells (e.g., CAR T cells or adoptive cell transfer) these nodes could be targeted genetically (e.g., using the CRISPR/Cas9 or TALEN systems). In addition, CD8⁺ T cells could be manipulated to recruit other immune cell populations to the tumor, break down the stromal architecture of the tumor, disrupt tumor vasculature, or potentially support formation of memory CD8⁺ T cells.

Another way to limit systemic activation of immune cells is to take advantage of advances in drug delivery: both delivery to specific sites and to specific cells. Polymer scaffolds engineered to recruit effector cells and prime them with proper cytokines are being tested in animal models. Such scaffolds could potentially also incorporate checkpoint antibodies. In addition, intratumoral injection approaches may provide a means to stimulate an immune response locally. The potential for this approach is illustrated by studies showing that intratumoral injection of anti-CTLA-4, anti-OX40 plus CpG led to a curative immune response and required 1/100 of the dosage of antibody in mouse tumor models. Directed delivery of checkpoint antibodies to specific cells by targeted nanoparticles or bispecific antibodies may provide even greater specificity by targeting effector cells.

Finally, improvements in cellular and tissue analysis may enable finer methods for evaluating the tumor microenvironment and developing more meaningful immunoscores to more accurately predict if a patient will respond to checkpoint blockade. CyTOF technology coupled with barcoding approaches provides a novel means to identify the spatial location of cells within the tumor and to reconstruct tissue architecture computationally overlaid with quantitation of up to 50 proteins. This approach should make possible functional characterization of immune cell infiltrates in the tumor while maintaining spatial relationships in the tumor. Laser capture microscopy followed by single cell RNAseq provides a complementary approach to investigate the entire transcriptome while preserving tissue structure. These two techniques are powerful tools for investigating clonal heterogeneity of both the immune cells and tumor cells during tumor evolution and determining why tumors regress or progress. This knowledge should aid development of better biomarkers for predicting which patients will respond to checkpoint therapy, monitoring responses to checkpoint blockade and determining if additional interventions are needed to promote tumor eradication.