The hallmarks of successful anticancer immunotherapy

See allHide authors and affiliations

Science Translational Medicine  19 Sep 2018:
Vol. 10, Issue 459, eaat7807
DOI: 10.1126/scitranslmed.aat7807


Immunotherapy is revolutionizing the clinical management of multiple tumors. However, only a fraction of patients with cancer responds to immunotherapy, and currently available immunotherapeutic agents are expensive and generally associated with considerable toxicity, calling for the identification of robust predictive biomarkers. The overall genomic configuration of malignant cells, potentially favoring the emergence of immunogenic tumor neoantigens, as well as specific mutations that compromise the ability of the immune system to recognize or eradicate the disease have been associated with differential sensitivity to immunotherapy in preclinical and clinical settings. Along similar lines, the type, density, localization, and functional orientation of the immune infiltrate have a prominent impact on anticancer immunity, as do features of the tumor microenvironment linked to the vasculature and stroma, and systemic factors including the composition of the gut microbiota. On the basis of these considerations, we outline the hallmarks of successful anticancer immunotherapy.


Over the past decade, several immunotherapeutic agents have become available for the routine clinical management of cancer (1). These include (but are not limited to) various immune checkpoint blockers (ICBs) targeting cytotoxic T lymphocyte–associated protein 4 (CTLA4), programmed cell death 1 (PDCD1; best known as PD-1) or its main ligand (CD274; best known as PD-L1) (2), as well as autologous T cells engineered to express a CD19-targeting chimeric antigen receptor (CAR) (3). ICBs used as stand-alone therapeutic interventions elicit durable objective responses in patients affected by a variety of cancers, including melanoma, non–small cell lung carcinoma (NSCLC), urothelial carcinoma, and Hodgkin’s lymphoma (47). Along similar lines, CD19-specific CAR T cells have pronounced clinical activity in patients with B cell acute lymphoblastic leukemia (ALL) exhibiting innate or acquired resistance to standard treatment, as well as in adult patients with relapsed or refractory large B cell lymphoma (8). Other anticancer immunotherapies are currently being investigated in clinical trials, encompassing a multitude of immunostimulatory monoclonal antibodies (mAbs), small molecules that reverse cancer-associated immunosuppression, as well as multiple tumor-targeting therapeutic vaccines (9). Many of these agents are in relatively advanced phases of clinical development, suggesting that the number of immunotherapeutic strategies licensed for the clinical management of cancer will increase in the near future.

The fraction of patients who respond to ICBs used as stand-alone therapeutic interventions is generally around 20% for the most common solid tumors, although it varies greatly in different oncologic indications (46). Moreover, ICBs are associated with moderate-to-severe immunological toxicities that—at least in some cases—require treatment discontinuation and/or active management, generally with glucocorticoids (10). Finally, the cost of ICB-based immunotherapy has been estimated at 100,000 to 250,000 USD per patient (depending on the specific ICB, administration protocol, and treatment duration) (11). The fraction of patients responding to CD19 CAR T cells used as an on-label intervention is remarkably high (around 80%) (12). However, these patients may experience relatively severe neurological events or cytokine release syndromes that need to be managed with the interleukin 6 receptor (IL-6R) blocker tocilizumab (8). Moreover, CD19 CAR T cells are currently provided at more than 350,000 USD per full treatment (although some companies charge responders only) (8). Similar considerations of toxicity and cost apply to many (if not all) anticancer immunotherapies in clinical development. A variety of combinatorial approaches is being investigated to increase the fraction of cancer patients responding to immunotherapy and/or to limit side effects (13). In addition, considerable efforts are being devoted to the identification of biomarkers with robust predictive value (14). Here, we attempt to define the hallmarks of successful anticancer immunotherapy as we analyze the promise and obstacles for these features to become part of clinical decision-making in the near future.


The success of anticancer immunotherapy depends to a large degree on the features of cancer cells that determine (i) their intrinsic potential to initiate a tumor-targeting immune response, (ii) their ability to establish an immunosuppressive tumor microenvironment (TME), and (iii) their sensitivity to immune effector mechanisms (Fig. 1).

Fig. 1 Malignant cells in the regulation of tumor-targeting immune responses driven by immunotherapy.

Several aspects of the biology of malignant cells affect the likelihood of anticancer immunotherapy to elicit robust clinical responses. During initiation (left), the abundance of tumor neoantigens (TNAs), which to some extent depends on mutational load, their quality (notably their resemblance to viral antigens), and the ability of malignant cells to emit danger signals as they die have a major influence on the elicitation of anticancer immunity by dendritic cells (DCs) and other antigen-presenting cells. Moreover, cancer cells compete for nutrients with immune effector cells and express co-inhibitory ligands and other factors including CD73 and lactate that mediate local immunosuppression (regulation; middle). Finally, during execution (right), the ability of cancer cells to properly present tumor neoantigens, respond to interferon gamma (IFN-γ) and granzyme B (GZMB), undergo regulated cell death (RCD), or mount cytoprotective autophagic responses determines their sensitivity to immune effectors. IFNGR, interferon gamma receptor; TME, tumor microenvironment.



Successful adaptive immune responses against cancer cells often target tumor neoantigens, which can arise as a consequence of somatic, tumor-specific nonsynonymous DNA mutations (15). In line with this notion, a high mutational load (which increases the likelihood for the emergence of neoantigens) as well as an increased abundance of predicted tumor neoantigens have been associated with improved sensitivity to ICB-based immunotherapy in a variety of clinical settings (1618). In addition, high mutational burden has recently been attributed robust predictive value for clinical responses to ICB-based immunotherapy in patients with lung cancer (19, 20). Along similar lines, elevated degrees of genomic instability, such as those displayed by tumors with high microsatellite instability (MSI-H) or defects in mismatch repair (MMR-D), have been linked to improved disease outcome in cancer patients treated with ICBs (21), presumably reflecting the increased propensity of these neoplasms to accumulate nonsynonymous mutations. Notably, this latter discovery led to the first-in-history U.S. Food and Drug Administration (FDA) approval of pembrolizumab (a PD-1–targeting ICB) for the treatment of MSI-H or MMR-D tumors irrespective of anatomical site (21).

That being said, the quality of tumor neoantigens—a parameter linked to their dissimilarity from “self” (often along with some degree of homology to microbial antigens), and hence to their likelihood to be recognized by the mature T cell receptor (TCR) repertoire—appears to be more important than their abundance in determining the success of anticancer immunotherapy, at least in some settings (22, 23). Indeed, T cell clones potentially recognizing tumor neoantigens that are highly homologous to self-antigens are likely to undergo negative selection within the thymus (and therefore be missing from the peripheral TCR repertoire) (24). Adaptive immune responses driven by immunogenic chemotherapy and radiation therapy also appear to rely on (at least some degree of) pathogen mimicry. In this setting, cancer cells succumbing to treatment in the context of stress responses release a panel of endogenous molecules that operate as adjuvants, including (but not limited to) adenosine 5′-triphosphate (ATP) downstream of autophagy activation, calreticulin (CALR) downstream of endoplasmic reticulum stress, and type I interferon (IFN) downstream of Toll-like receptor 3 (TLR3) or cyclic GMP-AMP synthase (CGAS) signaling (25). Preclinical and clinical data suggest that the acute and robust activation of these pathways, which are intimately involved in the control of viral replication (26), plays a major role in the establishment of anticancer immunity (27, 28). Accordingly, autophagy inhibition downstream of cancer-germline antigen expression has recently been associated with clinical resistance to CTLA4 blockade in melanoma patients (29). However, biomarkers of chronic type I (and type II, see below) IFN signaling have also been linked with limited tumor sensitivity to ICBs in the clinic (30). Moreover, reflecting the robust antiviral effects of this pathway, accumulating evidence suggests that the success of oncolytic virotherapy could be compromised by proficient type I IFN responses (31). These considerations demonstrate two points that must be taken into careful consideration as predictive biomarkers for cancer immunotherapy are developed and interpreted: (i) The acute versus chronic activation of some signaling pathways may have opposite effects on anticancer immune responses, and (ii) the influence of a specific process on distinct forms of immunotherapy may exhibit a large degree of heterogeneity.


Cancer cells establish local and systemic immunosuppression via a variety of mechanisms, all of which can affect the clinical success of immunotherapy. Perhaps the most studied of these mechanisms is the expression of surface molecules that drive tumor-infiltrating CD8+ T cells and natural killer (NK) cells into exhaustion, such as PD-L1 (13). In some indications including NSCLC patients treated with ICBs, PD-L1 expression on malignant cells bears robust predictive information (5, 32). Accordingly, the immunohistochemical assessment of membranous PD-L1 levels on cancer cells is approved by the FDA as a companion diagnostic for PD-1–targeting ICBs (33). However, the actual predictive value of PD-L1 expression by neoplastic cells in other clinical settings remains to be elucidated. Similarly, whether PD-L1 expression should be harnessed as a direct mechanistic target for treatment versus a surrogate biomarker of ongoing anticancer immunity has not been conclusively demonstrated yet. PD-L1 expression by cancer cells is particularly sensitive to interferon gamma (IFNG), which is one of the major effectors of tumor-targeting immune responses (34). This suggests that PD-L1 expression (by cancer cells or immune cells) may have predictive value (at least in patients affected by some tumors) only when measured before treatment.

Another example of the mechanisms whereby malignant cells establish local immunosuppression involves 5′-nucleotidase ecto (NT5E; best known as CD73), a plasma membrane protein that participates in the conversion of extracellular ATP, which has robust chemotactic effects and potently activates DCs, into adenosine, which inhibits immune responses by a variety of mechanisms, hence controlling the balance between immunostimulatory adrenergic signaling and its immunosuppressive adenosinergic counterpart (35). In line with this notion, high expression of CD73 has been associated with poor disease outcome in triple-negative breast cancer patients receiving adjuvant immunogenic chemotherapy (36). Preclinical data supporting the importance of proficient adrenergic signaling in the TME and pointing to CD73 as a potential target for novel anticancer immunotherapies are accumulating (37, 38). In this context, it will be important to determine the relative contribution of CD73 to immunosuppression in settings in which malignant cells express high levels of PD-L1 and other ligands for co-inhibitory receptors on T cells.

Activated lymphocytes resemble cancer cells in their high metabolic demands to support intensive proliferation (39). Accordingly, cancer cells and lymphocytes engage in a metabolic competition that influences the overall immunological status of the TME (and hence the likelihood for efficient anticancer immune responses) (40). Such a competition not only is centered around glucose, whose availability in the TME is critical for cytotoxic T lymphocyte (CTL) expansion and antitumor efficacy (40, 41), but also involves multiple amino acids including glutamine and arginine (42). Several other factors that influence the competition for nutrients in the TME have just begun to emerge (42). Malignant cells also secrete a variety of metabolites and cytokines that support local and systemic immunosuppression, including (but not limited to) lactate, IL-10, and transforming growth factor beta 1 (TGFB1; best known as TGF-β1) (4345). However, the most prominent source of immunosuppressive cytokines is often tumor-infiltrating immune cells and/or stromal cells (see below).


The susceptibility of cancer cells to immune effector mechanisms is essential for the success of anticancer immunotherapy. Accordingly, inactivating mutations of beta 2 microglobulin (B2M), encoding a core component of the major histocompatibility complex (MHC) class I antigen presentation machinery that is required for CD8+ T cells to recognize and kill cancer cells, as well as specific MHC class I genotypes have been linked with reduced susceptibility to ICB-based immunotherapy in cohorts of melanoma and NSCLC patients (4650). Along similar lines, genetic and epigenetic alterations affecting the interferon gamma receptor (IFNGR) signaling pathway in malignant cells, including (but not limited to) JAK1, JAK2, and APLNR mutations, have been associated with resistance to ICB in a variety of preclinical and clinical settings (48, 5154). Finally, proficient autophagic responses appear to limit the sensitivity of malignant cells to NK cell–dependent cytotoxicity, at least in part owing to the ability of autophagy to degrade the cytolytic molecule GZMB (55, 56). Although no immunotherapy specifically targeting NK cells is currently approved for clinical use, this latter observation may be of significance for agents that were conceived as targeted therapies but eventually turned out to engage innate immune effectors, such as the FDA-approved agents rituximab and trastuzumab (57). Allele-specific MHC loss as well as genetic defects in B2M and caspase 8 (CASP8), which encodes one of the mediators of apoptotic cancer cell death induced by CD8+ T cells, are common driver events in NSCLC (49, 58). This suggests that developing tumors are advantaged by mutations that render them less susceptible to CTL-mediated killing also in the absence of treatment. However, genetic alterations in MHC class I–coding loci as well as B2M and CASP8 mutations have been linked with increased mutational load and an immunological score that predicted clinical responses to ipilimumab among melanoma patients (59, 60). Moreover, CASP8 mutations have recently been associated with high amounts of intratumoral leukocytes across all cancers (61). Thus, it is tempting to speculate, but remains to be formally elucidated, that tumors evading cell-intrinsic oncosuppression and the effector arm of immunosurveillance may accumulate mutations at an increased pace. That said, how CD8+ T cells reactivated by ipilimumab would recognize and kill malignant cells bearing B2M or CASP8 mutations is an open conundrum. As a possibility, the inhibition of CTLA4 in the context of robust infiltration by CD8+ T cells (such as in tumors with CASP8 mutations) (61) may enable some degree of nonspecific activation, reminiscent of the mechanism of action of bispecific T cell engagers, (62), perhaps linked to indirect, IFNG-driven cytotoxicity via myeloid cells. This hypothesis has not been addressed experimentally yet.

As part of the identification of predictive biomarkers for anticancer immunotherapy, great attention should be given to signaling pathways and therapeutic paradigms that have long been believed to influence cancer cell–intrinsic processes only. Many of these highly investigated processes (for example, autophagy and apoptotic cell death) and interventions (for example, some forms of chemotherapy and radiation therapy and targeted anticancer agents) act at the interface between cancer cells and the immune system. Large amounts of samples and/or data are available for (re)evaluation and/or for the generation of testable hypotheses with respect to the immunological aspects of cytotoxic therapies.


Three aspects of the immunological tumor infiltrate have a major impact on the success of anticancer immunotherapy (and several forms of conventional therapy): (i) composition, (ii) localization, and (iii) functionality (Fig. 2).

Fig. 2 Impact of the tumor infiltrate on clinical responses to immunotherapy.

Three interrelated but conceptually distinct aspects of the immunological tumor infiltrate determine the likelihood of cancer patients to respond to immunotherapy: (i) the relative abundance of effector versus suppressor cells (left), (ii) the localization of immune cells with respect to their malignant counterparts (middle), and (iii) the activation status of immune effectors (right).



Tumor-infiltrating CD8+ T cells, T helper 1 (TH1)–polarized CD4+ T cells, and CD103+ DCs are widely considered to be central players in the initiation and execution of anticancer immune responses driven by multiple (immuno)therapeutic agents, including ICBs, oncolytic virotherapy, adoptive T cell transfer, radiation therapy, and immunogenic chemotherapy (6365). Thus, high intratumoral amounts of these cells have been consistently associated with improved disease outcome (66). Conversely, extensive tumor infiltration by CD4+CD25+FOXP3+ regulatory T (Treg) cells or M2-polarized macrophages, both of which mediate robust immunosuppressive effects, has generally been linked with limited sensitivity to a variety of (immuno)therapeutic regimens (66). In line with this notion, accumulating evidence suggests that repolarizing tumor-associated macrophages toward an immunostimulatory (so-called M1) functional state represents a valid approach to enhance CTL-driven anticancer immunity (67, 68). Tumor-infiltrating NK cells have long been underappreciated as mediators of tumor-targeting immune responses, potentially owing to the constitutive activation of the recently uncovered co-inhibitory receptor single Ig and TIR domain containing (SIGIRR; best known as IL-1R8) (69). However, accumulating preclinical evidence suggests that NK cells drive anticancer immunity in multiple settings (70, 71) as it defines therapeutically relevant links between NK cells and other components of the tumor infiltrate including DCs (72). Along similar lines, the role of intratumoral plasma cells and less mature cells from the B lineage in anticancer immune responses has just begun to emerge (73). In this setting, it is important to keep in mind that tumor infiltration by immune cells that drive anticancer immune responses is often accompanied by a compensatory increase in immunosuppressive cells, which can limit (at least in some settings) the predictive value of absolute quantifications of single cell types. As an example, the relative abundance of intratumoral CD8+ T cells over Treg cells often conveys improved informative value as compared to the abundance of each population taken individually (66). However, the adequate quantification of the immunological tumor infiltrate for predictive purposes calls for robust normalization procedures that are not always implemented. When specific populations of immune cells are studied, flow cytometry data are often normalized to the CD45+ infiltrate (all immune cells) or to even more restricted populations (for example, CD3+ T cells), which does not provide information on the actual quantity of such cells within tumors. In addition, it remains unclear whether de novo peripheral expansion of tumor-targeting lymphocytes is an absolute requirement for cancer patients to respond to all forms of immunotherapy. However, at least in some settings including melanoma patients, reinvigoration of preexisting immunity in the context of limited disease burden has been linked to improved disease outcome upon pembrolizumab treatment (74).


Not only the magnitude but also the localization of the immune infiltrate affects the likelihood of cancer patients to respond to immunotherapy. In particular, the so-called immune-excluded phenotype, in which high levels of T cells and other immune cells accumulate at the tumor margin but cannot invade malignant cell nests, is generally linked to poor disease outcome, as compared to the “inflamed” or “hot” phenotype, in which intratumoral immune cells are abundant and get into direct apposition with neoplastic cells (75). Similarly, the so-called immune-desert or “cold” phenotype, characterized by a global paucity of T cells, has been associated with reduced sensitivity to immunotherapy in a variety of clinical settings (75). Notably, the physical localization of tumor-infiltrating cells was attributed prognostic and predictive value in patients with colorectal carcinoma treated with conventional treatment when modern immunotherapy had yet to be translated into a clinical reality (76). These findings inspired a worldwide and ongoing effort aimed at standardizing the spatial quantification of the immune infiltrate toward an “immunoscore” that would resolve the inherent limits of immunohistochemical methods (77). Although the development of imaging platforms for clinical samples based on mass cytometry is still in relatively early phases of development, and only a few such instruments are available worldwide (78), this approach has raised great expectations as a possible means to quantitatively, spatially, and qualitatively characterize the cancer immune infiltrate in the same sample at improved resolution. It will be interesting to see whether such expectations will be met and imaging mass cytometry will eventually replace immunohistochemistry in this setting.


Additional features of the immune infiltrate influence the success of anticancer immunotherapy, including its functional status. Features of an active multi-epitope memory T cell response including a diversified TCR repertoire as well as surface features such as Ki67 positivity, CD69 expression, and CD45RO-to-CD45RA switch have been associated with improved responses to immunotherapy in preclinical and clinical settings (74, 79, 80). Increased expression of ectonucleoside triphosphate diphosphohydrolase 1 (ENTPD1; best known as CD39)—the enzyme that operates upstream of CD73 in the degradation of extracellular ATP—on Treg cells has been linked with robust immunosuppression and limited sensitivity to treatment in various preclinical and clinical settings, even in the context of Treg cell death upon oxidative stress (8183). An abundant preclinical literature has linked elevated levels of indoleamine 2,3-dioxygenase 1 (IDO1) and arginase 1 (ARG1) in tumor-infiltrating DCs with accelerated tumor growth and increased resistance to immunotherapy (84, 85). Similar considerations apply to many other immunosuppressive enzymes and surface molecules produced by tumor-infiltrating cells, including IL-10 (73), as well as to exhaustion markers including multiple co-inhibitory receptors (86).

In this setting, it is important to note that although high intratumoral levels of immunosuppressive molecules, co-inhibitory receptors, or their ligands generally reflect an exhausted immune response accompanied by disease progression, they may also identify targets for immunotherapeutic interventions. Thus, although high PD-L1 levels are expected to quench anticancer immune responses upon PD-1 binding, robust PD-L1 expression on immune cells has been associated with an increased probability for patients with urothelial carcinoma to respond to atezolizumab (a PD-L1–targeting ICB) (6, 87). Considerable efforts are being focused on the identification of such targets in support of clinical decision-making, especially in the context of neoplasms with innate or acquired resistance to first-line (immuno)therapy (88). Clinical data from a variety of indications and treatment paradigms are urgently awaited to potentially implement such an approach to interventions other than ICBs. One potential obstacle in this sense is represented by the limited availability of biopsies other than the diagnostic specimen, and hence the limited possibility to dissect the tumor infiltrate in a longitudinal manner (for example, pretreatment versus posttreatment). To circumvent this issue, considerable efforts are being devoted to the development of clinical-grade reagents that may enable longitudinal, noninvasive imaging studies of the immune infiltrate in patients, including radiolabeled PD-L1–targeting antibodies (89). However, although some of these reagents are already being tested in the clinic (, the number of biomarkers that can be simultaneously assessed by this approach is reduced (90), and spatial resolution is low (91). Thus, despite the logistical and personal difficulties linked to collecting sequential biopsies from the same individual, we believe that efforts should be made for raising awareness among patients and funding sources on the importance of this approach to obtain an improved understanding of the tumor infiltrate over time.


The tumor stroma and vasculature are less investigated than tumor-infiltrating immune cells for their role in tumor-targeting immune responses. However, accumulating evidence suggests that these two nonmalignant components of the TME can have a considerable impact on the success of anticancer immunotherapy (Fig. 3).

Fig. 3 Regulation of anticancer immunity by the tumor stroma and endothelium.

The tumor endothelium expresses FAS ligand (FASL) downstream of vascular endothelial growth factor A (VEGFA) and prostaglandin E2 (PGE2) signaling (1), hence favoring the preferential extravasation of CD4+CD25+FOXP3+ Treg cells (2) as a consequence of cytotoxic T lymphocyte (CTL) elimination. Cadherin 5 (CDH5) also contributes to the limited permeability of the tumor endothelium to immune effector cells (3). CD4+ TH1 cells secreting IFN-γ appear to participate in a bidirectional cross-talk with the tumor endothelium, resulting in vascular normalization and immune cell reprogramming (4). Cancer-associated fibroblasts (CAFs) secrete high amounts of TGF-β (5), resulting in the establishment of a dense stromal reaction commonly known as desmoplastic stroma. Protein tyrosine kinase 2 (PTK2; best known as FAK) activation in malignant cells and CAFs have been involved in this process (6). However, although TGF-β is involved in immune exclusion, the actual impact of the desmoplastic stroma on anticancer immunity remains to be clarified. CAFs can also delete tumor-targeting T cells via a contact-dependent mechanism involving Fas ligand (FASL) and programmed cell death 1 ligand 2 (PDCD1LG2; another PD-1 ligand best known as PD-L2; 7) and secrete matrix metalloproteinases (MMPs) that generate soluble NK cell-activating receptor (sNCAR) ligands (8). Hypoxia, lactate accumulation, and (at least in some tumors) the local microbiota also contribute to immunosuppression via a variety of mechanisms (9).


Some solid tumors including pancreatic ductal adenocarcinoma (PDAC), cholangiocarcinoma, and (less frequently) melanoma can establish a robust stromal reaction characterized by abundant cancer-associated fibroblasts and extensive extracellular matrix (ECM) deposition coupled to fibrosis, which is commonly known as desmoplastic stroma (92). The desmoplastic stroma is generally viewed as a physical barrier to successful tumor eradication, not only by the effector arm of the immune system but also by chemotherapy and targeted anticancer agents (93). Thus, considerable efforts have been focused on the development of therapeutic strategies that would loosen the desmoplastic stroma, including particle radiation therapy and pegylated human hyaluronidase, with variable degrees of success (93). Multiple immunotherapeutic interventions with (at least some degree of) efficacy in rodent models of PDAC or PDAC patients, including the adoptive transfer of NK cells activated in vitro by IL-15 (94), an immunostimulatory mAb specific for CD40 (95), and an irradiated allogeneic whole-cell vaccine secreting colony-stimulating factor 2 (CSF2; best known as GM-CSF) (96), appear to enable PDAC infiltration by immune cells along with a partial resolution of microenvironmental fibrosis. Along similar lines, inhibiting protein tyrosine kinase 2 (PTK2; best known as FAK)—which has been involved in the establishment of the desmoplastic stroma in PDACs—with a small molecule not only prolonged the survival of KPC mice (a genetic model of PDAC) but also rendered their tumors susceptible to adoptive T cell transfer and PD-1 blockade (97).

That being said, NK cells appear to limit metastatic dissemination by promoting production of the ECM protein fibronectin 1 (FN1) from melanoma cells, and hence ECM stiffening, as a consequence of IFNG secretion (98). Moreover, depletion of CAFs from mouse PDAC models accelerates disease progression along with the establishment of an immunosuppressive TME, and low levels of CAFs correlate with reduced survival in patients with PDAC (99). Likewise, patients with desmoplastic melanoma appear to be particularly sensitive to ICBs targeting PD-1 or PD-L1 irrespective of the extensive fibrotic reaction that characterize these tumors (100). Surprisingly, desmoplastic melanomas exhibit higher infiltration by PD-L1+ cells, which may be responsible for their improved sensitivity to PD-1 or PD-L1 blockade (100). Although the difference in mutational burden may explain why desmoplastic melanomas are infiltrated by immune cells while PDACs are not, the increased amounts of PD-L1+ cells observed in desmoplastic versus nondesmoplastic melanomas suggest that other factors are involved. Additional work is required to elucidate the impact of the desmoplastic stroma on anticancer immunity.

Besides depositing collagen and other components of the fibrotic ECM that characterize the desmoplastic stroma, CAFs are a major source of TGF-β1 and other members of the TGF-β protein family (101). Upon microenvironmental activation, TGF-β supports the establishment of the desmoplastic stroma (by propagating a wave of profibrotic signaling in the TME), which contributes to the sequestration of immune cells within CAF- and collagen-rich stromal areas of the tumor (102, 103). Suggesting a key role for TGF-β in the immune-excluded phenotype of some tumors and consequent resistance to immunotherapy, targeting TGF-β with a variety of strategies (that is, neutralizing mAbs, receptor inhibitors, and ligand traps) has recently been shown to sensitize a variety of tumors to ICB-based immunotherapy in preclinical settings along with the recovery of an inflamed phenotype (102104). Moreover, a gene signature representative of TGF-β signaling in CAFs has been specifically associated with stable or progressive disease (as opposed to partial and complete responses) among immune-excluded (but not immune-desert or inflamed) tumors in a large phase 2 clinical trial testing atezolizumab in patients with metastatic urothelial carcinoma (102). These observations corroborate the notion that TGF-β constitutes a promising target for the development of novel immunotherapeutic regimens to inflame immune-excluded tumors, and suggest that indicators of TGF-β signaling in CAFs may predict the responsiveness of immune-excluded tumors to (at least some forms of) immunotherapy. Because (i) TGF-β is secreted in an inactive form that needs to be released from the ECM to become bioavailable and (ii) the TGF-β system is rather pleiotropic (101), measuring TGFB1 levels by RNA sequencing (RNA-seq) or extracellular TGF-β levels by immunohistochemistry does not necessarily reflect functional TGF-β status. In this setting, indicators of downstream TGF-β signaling, including the phosphorylation of SMAD family members, may provide superior informational value.

CAFs also mediate immunosuppressive effects by releasing metalloproteinases that can cleave NK cell–activating receptor (NCAR) ligands such as MHC class I polypeptide-related sequence A (MICA) and MICB from the surface of cancer cells, generating inhibitory decoys (57, 105). Moreover, CAFs can promote tumor infiltration by immature myeloid cells (106) and delete cancer-specific CD8+ T cells from the TME (107). This latter mechanism appears to involve the ability of CAFs to acquire tumor-derived antigens and cross-present them to CD8+ T cells along with lethal signals delivered by programmed cell death 1 ligand 2 (PDCD1LG2; another PD-1 ligand best known as PD-L2) and Fas ligand (FASL), an activator of extrinsic apoptosis (107, 108). Accordingly, blockade of PD-L2 or FASL has been shown to mediate stand-alone therapeutic effects in rodent models of melanoma (107). Although the actual impact of PD-L2 signaling in the PDAC microenvironment remains to be elucidated, CAFs from PDACs express high levels of PD-L2 (94), and this has been linked to durable disease response in a PDAC patient receiving pembrolizumab (109). It will be important to delineate the impact of PD-L2 signaling on the success of immunotherapy for PDAC patients.

As compared to normal vasculature, the tumor vasculature is disorganized, has limited pericyte coverage, and is characterized by loose interendothelial cell junctions (110). This results not only in impaired microcirculatory flow and consequent hypoxia but also in increased interstitial pressure, which constitutes a physical barrier to infiltration by immune cells (110). Moreover, the endothelial lining of tumor-associated (but not normal) blood vessels expresses FASL, which has been associated with limited tumor infiltration by CD8+ T cells (which express high levels of FAS) and accumulation of Treg cells (which express comparatively lower FAS levels) (111). In this setting, endothelial cells were shown to express FASL downstream of PGE2 and VEGFA, which are known to drive Treg-dependent immunosuppression (112). Accordingly, inhibition of prostaglandin E2 (PGE2) or vascular endothelial growth factor A (VEGFA) signaling mediated therapeutic effects accompanied by restored CD8+ T cell infiltration of malignant lesions (111). Along similar lines, increasing the permeability of tumor vasculature to immune cells by inhibiting cadherin 5 (CDH5) in endothelial cells improved the recruitment of CD8+ T cells to the parenchyma of mouse melanomas and colorectal carcinomas, as it mediated stand-alone therapeutic effects that were amplified by PD-1 blockade (113). Thus, the immunological status of the TME is considerably influenced by the local vasculature. CD4+ TH1 cells secreting IFN-γ have been shown to play a major role in vessel normalization, especially upon activation with ICBs (114), and this appears to correlate to sensitivity to treatment (115). Moreover, activation of the complement system on the endothelial lining is necessary downstream of TH1 cytokines for optimal immune cell extravasation and tumor control, at least in mouse models of lung carcinoma (116). These observations suggest that the relationship between the tumor vasculature and immune infiltrate is bidirectional. This point should be taken into consideration for the development of novel immunotherapies, especially when efficient extravasation is an absolute requirement for efficacy (for example, cold tumors and adoptive cell therapy). Notably, the role of tumor-efferent lymphatics in anticancer immunity remains to be fully characterized. An intact dermal lymphatic network appears to be required for optimal responses to a melanoma-targeting vaccine in mice, yet it impairs the ability of adoptively transferred T cells to control disease (117). Moreover, lymphatics are dispensable for the priming of tumor-targeting immune responses by immunogenic cell death inducers (118). Tumor-efferent lymphatics have also been shown to express immunomodulatory molecules including PD-L1, potentially contributing to the establishment of local immunosuppression (119). The actual relevance of this observation, however, remains to be clarified.

The TME is generally hypoxic (owing to vasculature defects) and contains excess lactate (owing to the metabolic reprogramming of malignant cells), resulting in the activation of multiple signaling cascades with immunosuppressive effects (43, 120, 121). Moreover, it has recently been shown that the stroma of some tumors (notably PDACs) contains a microbial component that influences the immunological profile of the TME as a consequence of TLR signaling in tumor-infiltrating monocytes (122). In this setting, bacterial ablation with antibiotics resulted in reprogramming of the myeloid compartment followed by the establishment of a TH1 CD4+ T cell response culminating with CTL activation and partial disease control (122). In support of the clinical relevance of these observations, the presence of Fusobacterium spp. in human PDACs has been associated with negative prognostic value in a cohort of 283 patients (123). Although the role of the intratumoral (as opposed to mucosal) microbiota has just begun to emerge, these findings suggest that investigating the microbial component of human cancers may bear prognostic or predictive value. Data from large clinical studies designed to investigate this possibility are urgently awaited for clarifying the actual impact of intratumoral bacteria on the success of anticancer immunotherapy.


Cancer control by immunotherapy is enabled and regulated by systemic (circulating and noncirculating) factors, some of which are being investigated as targets for novel immunotherapeutic interventions. Moreover, active anticancer immunity can, at least in some settings, manifest with variations in circulating parameters, which are attracting considerable interest as potential predictors of response, largely reflecting accessibility considerations (Fig. 4).

Fig. 4 Systemic hallmarks of successful anticancer immunotherapy.

The likelihood of cancer patients to obtain clinical benefits from immunotherapy depends on their global immunological competence, which can be influenced by cancer- and treatment-unrelated factors including variations in genes involved in the elicitation, regulation, and execution of tumor-targeting immunity; by viral infections or pharmacological agents resulting in systemic immunosuppression; and by the composition of the gut microbiota. In addition, developing tumors (and some forms of treatment) can change the systemic immunological microenvironment by favoring the generation of immature myeloid cells with immunosuppressive activity, by altering the gut microbiome (dysbiosis), and by releasing cytokines and other factors that quench anticancer immune responses. Many of these changes are instrumental for tumors to resist to treatment, hence constituting promising targets for the development of novel therapeutic interventions. These and other circulating factors including indicators of the functional competence of immune effector cells may be harnessed as biomarkers to assist clinical decision-making.


Tumor-targeting immune responses rely on an intact and functional immune system, implying that any systemic issue potentially causing immunological defects is likely to limit the responsiveness of patients to (immuno)therapy. This applies not only to rather generalized problems such as cytomegalovirus infection (124) or transplantation-associated pharmacological immunosuppression but also to single-nucleotide polymorphisms affecting very specific facets of cancer recognition and/or eradication. For instance, polymorphisms in FCGR3A that affect the competence of ICB-based therapies to elicit antibody-dependent cell-mediated cytotoxicity (ADCC) have been shown to predict ipilimumab efficacy among melanoma patients (125), as well as trastuzumab efficacy in patients with early ERBB2/HER2-positive breast cancer, corroborating the notion that trastuzumab also engages innate immune effectors (see above) (126). Similarly, loss-of-function polymorphisms affecting multiple pattern recognition receptors (PRRs) that sense endogenous adjuvants released by dying cancer cells are associated with worsened disease outcome among breast carcinoma patients receiving neoadjuvant chemotherapy with immunogenic cell death inducers (25). Moreover, an unanticipated degree of interindividual variability in specific DC subsets has recently been uncovered (127). These observations exemplify the fact that individuals differ in their intrinsic immunologic proficiency, due to both inherent (genetic) and acquired (environmental) causes, which affects the success of cancer immunotherapy. Lending further support to this notion, it is now clear that the composition of the gut microbiome has a major impact not only on intestinal tumorigenesis but also on the susceptibility of extraintestinal tumors to chemo- and immunotherapy (128). In particular, the abundance of specific bacterial species or signs of an immune response targeting them have been shown to correlate with sensitivity to ICBs or chemoimmunotherapy in patients affected by advanced solid tumors including melanoma as well as lung and ovarian carcinoma (129133). Accumulating evidence indicates that this reflects the impact of the gut microbiota and its products translocating across the intestinal wall on the activation of anticancer immune responses (134136). The microbiome of wild mice has recently been shown to improve the resistance of laboratory mice to intestinal tumorigenesis (137). Together, these observations should raise awareness regarding (i) the possibility that currently used laboratory animals may be particularly susceptible to tumorigenesis owing (at least in part) to their microbiome, which may also influence how these animals respond to treatment, and (ii) the need to consider the variability in the microbiome of mice from different vendors or facilities as potential contributors to differential sensitivity to immunotherapy recorded. It is paramount for the advancement of the field to define standardized measures to homogenize the gut microbiota in rodent models.

Although they can change during tumor progression and with treatment, the systemic factors described above are generally present (and hence influence sensitivity to tumorigenesis and immunotherapy) at baseline. There are also factors altering the efficacy of immunotherapy that become manifest as tumors form or even when established tumors are challenged with treatment. For instance, progressing tumors can alter hematopoiesis and often drive the accumulation of circulating myeloid cells with a rather immature phenotype and potent immunosuppressive activity, which are commonly known as myeloid-derived suppressor cells (MDSCs) (138). MDSCs can infiltrate malignant lesions to mediate local immunosuppression, and their circulating abundance is generally associated with poor disease outcome and limited sensitivity to immunotherapy (138, 139). Along similar lines, circulating CD73 levels have been shown to increase in patients with solid tumors including melanoma, a setting in which abundant CD73 has been linked to decreased sensitivity to nivolumab (a PD-1–targeting ICB) and poor overall survival (140). Similar observations have been made for circulating NCAR ligands in melanoma patients treated with ipilimumab plus fotemustine (141), as well as for soluble NCAR ligands found in the ascites of ovarian carcinoma patients treated with surgery (142). Preclinical data suggest that mAbs neutralizing soluble NCAR ligands may exert beneficial effects against some tumors including melanoma and prostate carcinoma, even when used as stand-alone therapeutic interventions (105, 143). Platelet-derived metalloproteinases seem to contribute to the shedding of NCAR ligands from neoplastic cells, and this cleavage activity appears to be higher in platelets from cancer patients (as compared to platelets from healthy individuals) (144). These observations exemplify a number of immunosuppressive circuits established in cancer patients that manifest systemically, at least to some degree. To what extent these systemic pathways can be modulated or monitored for therapeutic or predictive purposes remains to be elucidated.

The peripheral blood can also bear indicators of tumor-targeting immune responses, and at least in some settings, these biomarkers have predictive value. For instance, the immunological profiling of peripheral blood from patients with stage IV melanoma before and after treatment with pembrolizumab revealed that clinical insensitivity to treatment is associated with an imbalance between the reinvigoration of circulating CD8+ T cells by treatment (as documented by multiple parameters including Ki67 positivity, PD-1 expression, and clonal expansion) and tumor burden (74). Moreover, the percentage of circulating FAS+CD4+ or PD-L1+CD4+ T cells has been reported to negatively correlate with response to adjuvant ipilimumab in a cohort of melanoma patients, linked to a decrease in overall survival (145). Conversely, detectable TNF receptor superfamily member 9 (TNFRSF9; best known as CD137) on circulating CD8+ T cells has been associated with improved disease-free survival in stage III metastatic melanoma patients receiving adjuvant ipilimumab plus nivolumab (145). Along similar lines, the surface phenotype of circulating NK cells and their ability to secrete IFN-γ upon challenge have been linked to differential sensitivity to the immunostimulatory anticancer agent imatinib in patients with gastrointestinal stromal tumors (146, 147). Together, these findings indicate that, at least in some settings, the features of specific populations of circulating immune cells may be representative of the immunological landscape of the TME. Similar results have been reported for the circulating levels of some cytokines, including IL-15 (148), and other soluble immunomodulators, including soluble PD-L1 (149). As with virtually all circulating biomarkers excluding tumor-derived DNA (which can be differentiated on the basis of mutational status), however, the actual value of circulating immunomodulators as biomarkers for disease detection, progression, or responsiveness to immunotherapy may be limited by possible nonneoplastic sources, including subclinical infections. This point must be taken into attentive consideration when novel assays based on non-necessarily tumor-specific biomarkers measured in the periphery are developed.


Over the past decade, considerable efforts have been dedicated to the identification of actionable hallmarks of successful anticancer immunity that may either be pharmacologically targeted to increase therapeutic efficacy or monitored to prospectively identify responders. Such an intensive wave of investigation has unveiled several obstacles that still hamper the straightforward translation of preclinical and clinical data into novel immunotherapeutics and/or decision-making protocols. First, little consideration has been given so far to the elevated genetic, epigenetic, and immunological heterogeneity of most (if not all) solid tumors within the neoplastic lesion itself, as well as to the extraordinary capacity of malignant lesions to evolve (150, 151). Sampling the TME at multiple locations and ideally at multiple time points in the course of treatment might enable an improved characterization of malignant lesions with respect to both heterogeneity and evolutionary behavior, and hence provide superior feedback to clinicians. Second, multiple processes that are relevant for the success of anticancer immunotherapy have been widely investigated from a cancer cell–intrinsic perspective, with little attention to (i) the role of these processes in other compartments of the TME and (ii) their cancer cell–extrinsic effects (152). It will be crucial to investigate in detail how such processes, which include bioenergetic metabolism, autophagy, cellular senescence, and immunomodulation, operate in the stroma, endothelium, and immune infiltrate and how these compartments are connected together in the regulation of antitumor immunity (107, 153). In addition, it will be important to concomitantly interrogate the innate and adaptive arms of the immune system, as accumulating evidence suggests that only a properly orchestrated immune response is capable of eradicating established tumors in response to immunotherapy. Third, at least some of the factors that influence the success of anticancer immunotherapy have been investigated with emerging techniques that are unable to grasp the entire complexity of the TME. For instance, although RNA-seq is a powerful technique for the unsupervised characterization of the overall transcriptional profile of the TME, transcriptomic data are intrinsically unable to reflect the abundance of proteins subjected to robust posttranslational control, including PD-L1 (154), TGF-β (101), and IL-1β (155). Although immunohistochemical tests can partially circumvent such issues for intracellular or membrane-bound proteins, the actual abundance of secreted factors in the TME remains challenging to be tested. Fourth, the interactions between conventional forms of treatment (that is, surgery, chemotherapy, radiation therapy, and targeted therapy) and immunotherapy have been incompletely investigated, especially in clinical settings. To develop safe and efficient combinatorial regimens, it will be critical to delve deeper into how conventional therapeutic approaches influence the efficacy of immunotherapy (156). Finally, the impact of the gut microbiome on the efficacy of multiple forms of treatment, including immunotherapy, has just begun to emerge. On the basis of the robust preclinical and clinical findings of the last few years, this aspect of anticancer immune responses deserves additional attention (157). It is therefore imperative not only to monitor the microbiome of patients enrolled in clinical trials but also to define harmonized procedures for preclinical investigation.

In summary, although the search for hallmarks of efficient anticancer immune responses still presents several obstacles, considerable progress has been made, enabling the identification of factors that are actionable from therapeutic or predictive purposes. That said, we suspect that none of the individual hallmarks discussed here will allow for the implementation of truly personalized immunotherapeutic protocols, calling for the development of multiparametric assessments with superior informational value. We surmise that addressing the obstacles delineated here will contribute to the development of next-generation immunotherapeutic regimens with improved clinical efficacy and foster the use of precision medicine in immuno-oncology.


Acknowledgments: We apologize to the authors of several high-quality articles dealing with the hallmarks of successful anticancer immunotherapy that we were not able to discuss and cite owing to space limitations. Funding: L.G. was supported by a startup grant from the Department of Radiation Oncology at Weill Cornell Medicine (New York) and by donations from Phosplatin (New York), the Luke Heller TECPR2 Foundation (Boston), and Sotio a.s. (Prague, Czech Republic). J.D.W. was supported by the Ludwig Institute for Cancer Research, the Parker Institute for Cancer Immunotherapy, Swim Across America, National Cancer Institute P30CA008748, and Ono Pharmaceuticals. G.K. was supported by the Ligue contre le Cancer (equipe labellisee); Agence National de la Recherche (ANR)–Projets blancs; ANR under the frame of E-Rare-2, the ERA-Net for Research on Rare Diseases; Association pour la recherche sur le cancer (ARC); Canceropole Ile-de-France; and Chancelerie des universites de Paris (Legs Poix), Fondation pour la Recherche Medicale (FRM); a donation by Elior; the European Commission (ArtForce); ERA Network on Cardiovascular Diseases (ERA-CVD, MINOTAUR); the European Research Council (ERC); Fondation Carrefour; Institut National du Cancer (INCa); INSERM (HTE); Institut Universitaire de France; LeDucq Foundation; the LabEx Immuno-Oncology; the RHU Torino Lumiere; the Seerave Foundation; the SIRIC Stratified Oncology Cell DNA Repair and Tumor Immune Elimination (SOCRATE); the SIRIC Cancer Research and Personalized Medicine (CARPEM); and the Paris Alliance of Cancer Research Institutes (PACRI). Author contributions: L.G. and A.L.-S. conceived the paper. L.G. wrote the first version of the manuscript; designed the figures; integrated inputs from T.A.C., G.K., J.D.W., and A.L.-S.; and centralized the revision of the paper based on the feedback from reviewers. All authors approved the final version of the article. Competing interests: L.G. provides remunerated consulting to AstraZeneca, VL47, and OmniSEQ, is a member of the Scientific Advisory Board of OmniSEQ, and receives research funding from Lytix and Phosplatin. T.A.C. is a cofounder of Gritstone Oncology and holds equity and receives research funding from AstraZeneca, Bristol Myers Squibb, and Illumina. G.K. is a cofounder of everImmune, is a member of the Scientific Advisory Boards of Lytix and Macrophage Pharma, provides remunerated consulting to Bayer Pharma, and has research contract with Lytix and PharmaMar. J.D.W. is a consultant to Bristol-Myers Squibb, Merck, MedImmune, Celgene, Genentech, BeiGene, Potenza Therapeutics, Tizona Therapeutics, Imvaq, Apricity, Surface Oncology, Polaris, Polynoma, Array, Ascentage, Puretech, Chugai, FStar, Amgen, Sellas Life Sciences, Serametrix, Neon, Eli Lilly, and Kleo Pharmaceuticals, is a cofounder of Potenza Therapeutics, Tizona Therapeutics, and Imvaq, holds equity in Potenza Therapeutics, Tizona Therapeutics, Adaptive Biotech, Imvaq, Apricity, and BeiGene, and receives research funding from Bristol-Myers Squibb, Merck, Genentech, and MedImmune. A.L.-S. has no relevant conflicts of interest to disclose.

Stay Connected to Science Translational Medicine

Navigate This Article