From mice to humans: developments in cancer immunoediting

Elimination.

In the elimination phase, innate and adaptive immunity work together to destroy developing tumors long before they become clinically apparent. Although the elimination phase has not been directly visualized in vivo, studies have demonstrated that immunodeficient mice (deficient for effector molecules such as IFNs and perforin; recognition pathways like NKG2D; or cell types such as T and NK cells) displayed earlier onset or greater penetrance of carcinogen-induced and spontaneous cancers compared with that seen in WT mice (reviewed in refs. 1, 3–6).

Equilibrium.

Rare tumor cell variants not destroyed in the elimination phase can proceed into the equilibrium phase, where their outgrowth is prevented by immunologic mechanisms. In a 2007 study (7), WT mice treated with low-dose methylcholanthrene (MCA) were demonstrated to harbor occult cancer cells that adaptive immunity (e.g., T cells and IFN-γ) kept in check. A subsequent study (8) demonstrated that immune-mediated tumor dormancy was dictated by a balance between two opposing cytokines, IL-12 and IL-23 (9), and could last for much of the lifespan of a mouse. The existence of the equilibrium phase was additionally supported by observations in mice with p53-mutant tumors (8) and by two other studies in which a Th1 environment dictated the eventual outcome of dormant tumors (10, 11).

Escape.

When tumors circumvent immune recognition and/or destruction, they progress from the equilibrium to the escape phase, where they become clinically apparent. Tumors escape due to changes in their response to immunoselection pressures and/or to increased tumor-induced immunosuppression or immune system deterioration. The mechanisms of tumor cell escape can be classified into three categories, as shown in Figure 1. Over the past two decades, these pathways have been the subjects of intense investigation, with the aim of developing new cancer immunotherapies (reviewed extensively in refs. 1, 3, 5).

Regarding tumor editing in each of these three phases, tumors derived from immunodeficient mice were found to be more immunogenic than were similar tumors derived from immunocompetent mice (1). Although T cells were inferred to sculpt tumor immunogenicity, it was unclear whether antigens expressed by emerging nascent tumor cells were recognized by T cells and could then be subsequently modulated in response to selection pressure. A study using an exome-sequencing approach demonstrated that sarcoma cells derived from MCA-inoculated mice expressed immunodominant neoantigens that could be immunoedited by CD8⁺ T cells, leading to tumor escape (12). Thus, the outgrowth of tumor cells lacking these strong antigens represents one mechanism of cancer immunoediting.

Immune contexture.

The immune contexture of human cancers defines a complex immunologic tumor microenvironment in which the location, density, and functional orientation of infiltrating hematopoietic cells correlate with the clinical outcome of patients (reviewed extensively in ref. 16). In most cancers, a strong infiltration of memory CD8⁺ T cells and a Th1 orientation correlated with a favorable prognosis in terms of progression-free survival (PFS) and/or OS (ref. 16 and Figure 2). Other parameters such as immune cells, cytokines, chemokines, and other factors also influence the generation of a clinically efficient immune microenvironment (Figure 2). In tumors, all types of immune cells can be found in various proportions and locations. They include myeloid cells, B cells, and all subsets of T cells (refs. 16, 18, and discussed in further detail below).

Immunoscore.

The “immunoscore” is a scoring system derived from the immune contexture (16, 19, 20) and is a clinically useful prognostic marker (21) based on the enumeration of two lymphocyte populations (CD3 and CD8), in both the core of the tumor (CT) and in the invasive margin (IM) of tumors. The immunoscore provides a score ranging from immunoscore 0, in which low densities of both cell types are found in both regions, to immunoscore 4, in which high densities are found in both regions. Immunoscore classification was shown to have a prognostic significance in all stage I, II and III patients with colorectal cancer (CRC) that was superior to that of the American Joint Committee on Cancer/Union for International Cancer Control (AJCC/UICC) TNM (T stage, N stage, and tumor differentiation) classification system (22, 23). Tumor invasion was shown to be statistically dependent on the host’s immune reaction. It is believed that the ability of effector memory T cells to recall previously encountered antigens and traffic into tumors may lead to long-term immunity in human cancer (24, 25). Now, a clinical validation of the immunoscore with standardized procedures is necessary to reach clinical applicability for individual patients with CRC and other cancers (26, 27). The immunoscore also provides a tool and targets for novel therapeutic approaches, including immunotherapy (as recently illustrated in clinical trials boosting T cell responses with anti–CTLA-4, anti–PD-1, and anti–PD-L1) (28–31).

Tumor-adjacent TLSs.

In situations of chronic stimulation of the immune system, such as in autoimmune and inflammatory diseases or in transplanted organs, organized lymphoid structures can form in these pathological locations, displaying all characteristics of the immunity-generating sites of secondary lymphoid organs (reviewed in ref. 32). These TLSs can represent a very powerful means to generate effective, local immune reactions. In fact, they precede the appearance of lymph nodes in evolution (33), and lung viral infections can be eradicated by local TLSs in lymph node–deficient mice (34). TLSs are consistently found in the tumor microenvironment of most human cancers (32).

TLSs are composed of a T and a B cell zone, where T cells are in contact with mature DCs, whereas B cells interact with follicular DCs. These TLSs are surrounded by peripheral lymph node addressin–expressing high endothelial venules (HEVs), which are likely sites of entry of naive lymphocytes from the blood. All lymphocytes in TLSs express the peripheral lymph node addressin (PNAd) receptor CD62-L. Naive T cells are likely instructed by contacting DCs presenting TAAs and central memory and effector memory T cells, all of which are found in TLSs. In tumors, B cells are almost exclusively found in TLS germinal centers, where they express the somatic hypermutation and switch recombination enzyme activation-induced cytidine deaminase (AICDA). Plasma cells are seen at the periphery of the TLS and have been reported to produce antitumor-specific antigens in non–small-cell lung cancer (NSCLC) (35). One can therefore hypothesize that efficient antitumor reactions can be generated within TLSs, where DCs instruct lymphocytes, which, having entered directly through HEVs, escape the suppressive milieu of the tumor microenvironment (32).

TLSs appear to play a role in the anticancer immune response. Indeed, in melanoma, breast, lung, and colorectal carcinomas, high densities of TLSs have been reported to correlate with a favorable prognosis (36). Strikingly, TLSs are induced in cervical carcinomas (37) or pancreatic ductal cancer (38) upon vaccination with TSMAs. That TLSs are necessary for generating clinically efficient immune reactions is supported by the fact that a high density of CD8⁺ T cells is associated with longer patient survival when tumors display high TLS densities in lung (39), colorectal (40), and renal cell cancers (41). By contrast, they correlate with shorter survival in lung (39) and clear cell renal cell carcinoma (41) with low numbers of intratumoral TLSs. Recently, the presence of TLSs has been reported in some breast cancers, which suggests the presence of organized immunity and the generation of memory B cells within the tumor (42). Therefore, identification, enumeration, and structural analyses of TLSs are useful tools to predict patient prognosis and immunotherapy follow-up, but also to identify tumor-specific clones of T and B cells.

Neoantigen discovery.

While a number of TAAs have been identified since the 1990s (e.g., NY-ESO-1, MAGE), TSMAs have been harder to discover due to the labor-intensive process required to identify them. However, with advances in genomics and bioinformatics technology, a number of recent studies in mice and humans have identified TSMAs (59, 60). These studies further strengthen the idea that “passenger” mutations as opposed to “driver” mutations (which contribute directly to cancer initiation and progression) can function as neoantigens that induce tumor immunity before or after therapeutics such as ipilimumab. Using the same MCA-induced tumor model that Matsushita et al. used to study immunoediting of immunodominant neoantigens by CD8⁺ T cells (12), Gubin et al. showed that even escape tumors harbor TSMA-recognizing T cells, but are inhibited by immune checkpoints (PD-1 and CTLA-4) (59). Such TSMAs can act as therapeutic vaccines, as evidenced by their ability to cause established tumor rejection with an efficacy similar to that of immune checkpoint blockade therapy in tumor-bearing mice. Given the heterogeneity of tumors (61), it would be wise to vaccinate with a number of TSMAs, if possible, to reduce the emergence of escape variants. This is illustrated in a case report of a patient with melanoma who expressed a number of TAAs (NY-ESO-1, MAGE-C1, and Melan-A) prior to immunization with an NY-ESO-1 vaccine (62). In another case report, analyses of three consecutive lesions obtained within one year of a patient’s developing stage IV melanoma revealed a gradual loss of immunogenicity that culminated in complete T cell resistance of the tumor cells caused by an irreversible HLA class I–negative phenotype (63). Overall, these studies provide evidence that tumors can express antigens recognizable by CD8⁺ T cells and that they are critical in the immunoediting of tumors with an adaptive immune resistance phenotype.

Does cancer immunoediting occur in epithelial cancers?

Many of the principles of cancer immunoediting were proposed and demonstrated in mice using the chemically induced MCA sarcoma model (7, 64). These sarcomas are considered immunogenic and have a high mutational load similar to that of ultraviolet- and other carcinogen-induced human cancers (12). As such, a common criticism for the cancer immunoediting hypothesis is a lack of evidence that it holds true for epithelial cancers caused by simple drivers, which often have a lower mutational load (65). This may significantly limit the number of neoantigens capable of being presented on the tumor’s HLA class I alleles.

There is now increasing evidence to support the occurrence of cancer immunoediting in epithelial cancers. Recently, genetic findings provided evidence for immunoediting in tumors and uncovered mechanisms of tumor-intrinsic resistance to cytolytic activity (66). In a patient with metastatic cholangiocarcinoma, a cancer with a low mutational load, researchers detected the presence of a specific CD4⁺ T cell recognizing a mutation in the ERBB2-interacting protein (ERBB2IP)(67). Similarly, high levels of CD8⁺PD-1⁺ T cells have been observed in small numbers of patients with CRC-MSS, who generally have a lower mutational load, suggesting that good T cell neo-epitopes can be generated if the mutations are appropriately positioned (57). Epithelial ovarian cancer (EOC), classically thought to be a nonimmunogenic cancer, provides further examples of cancer immunoediting (68), including tumor-specific CD8⁺ T cells that recognize mutated oncogenes such as HER2 (69). An increasing number of large breast cancer data sets have reported the correlation of T cell infiltrates with better clinical outcomes, including in HER2-positive breast cancer patients, for whom an association between immunity and better survival after treatment with chemotherapy and anti-HER2 therapies has been reported (42). High levels of T cell checkpoint receptors have also been detected in HER2-overexpressing breast cancer at the mRNA level as well as in triple-negative breast cancer (TNBC) (70). Anti-HER2 and anti–PD-1 have been shown to be synergistic in mouse models (71), and recently, anti–PD-1 (pembrolizumab) demonstrated clinical activity in TNBC (72). This potentially opens up new treatment avenues for patients with breast cancer.

Myeloid cells.

In general, high densities of myeloid cells, i.e., macrophages and myeloid-derived suppressor cells (MDSCs), correlate with poor prognosis (73). When it has been characterized, it appears that the negatively impacting macrophages are of the M2 phenotype (74). In any case, the correlation between macrophage density and patient survival is less significant than that of T cells, particularly CD8⁺ T cells (18), and it has been proposed that clinical outcome could be better predicted by the ratio of the densities of CD8⁺ T cells/macrophages (75, 76). In contrast, a high density of mature DCs is generally associated with longer survival, as in NSCLC (77), breast cancer (78), or melanoma (78), although the density of myeloid cells expressing the DC-lysosomal–associated membrane protein (LAMP) marker was found to correlate with shorter survival in RCC (41, 79). In a mouse model of ovarian cancer, it was shown that phenotypically divergent DCs drive both immunosurveillance and accelerated tumor growth (80). The switch from one phenotype to the other coincided with an increasingly immunosuppressed tumor microenvironment. In addition to macrophages and DCs, emerging data have revealed a role of neutrophil extracellular traps (NETs) in cancer immunoediting. These are neutrophil-derived structures composed of extruded DNA, decorated with antimicrobial proteins. Although they are generally formed in response to infectious stimuli, they have been reported to sequester circulating tumor cells and promote metastases (81, 82).

CD4⁺ T cell subsets.

Compared with myeloid cells, the effects of T cell subsets in the tumor/immune microenvironment appear much less straightforward (16). The Th2 signature correlated with favorable prognosis in Hodgkin lymphoma (83) and breast cancer (84), but shorter survival in ovarian (85), pancreatic (86), and gastric cancers (87). Conversely, a Th17 signature was found to be deleterious in colorectal (88), lung (89), and hepatocellular carcinoma (90) and beneficial in ovarian (91), esophageal (92), and gastric (93) cancers. These findings highlight the fact that the cancer type may influence the clinical impact of the tumor/immune microenvironment on the apparent balance between Th2 and Th17 cells. Thus, when a Th2 signature was associated with poor prognosis, such as in ovarian, pancreatic, and gastric cancers, it was reported by independent groups that, in these cancer types, a Th17 signature correlated with a favorable clinical outcome. These observations suggest that the clinical impact of immune cell populations is highly dependent on the overall immune microenvironment, which is likely orchestrated by the cancer cell type (tissue of origin or organ) and by the genetics of the tumor. Intratumoral Tregs are even more complex, since they have been reported to be associated with poor or favorable prognosis, not only in different cancer types, but also in the same cancer (94, 95). It may reflect the heterogeneity of Treg subsets (96) in the tumor microenvironment or the selectivity of Treg markers (97).

Other immune cells.

A number of other types of lymphocytes also contribute to the immune contexture of cancer, cancer immunoediting, and patient outcome. Unlike what has been observed in mouse models (75), high densities of B cells and T follicular helper (Tfh) cells correlated with longer survival in NSCLC (35), breast cancers (98), and CRC (18). In contrast, the impact of NK cell density is either weakly favorable (79) or null (99), depending on the balance of the expression of activating and inhibitory receptors that generally regulate the effector function of NK cells (99, 100). The mouse lectin-like Ly49 family receptors, which are similar to human killer inhibitory receptors in their capacity to bind MHC class I molecules, largely control the self-reactivity of NK cells. The importance of Ly49 in NK cell–mediated tumor immunosurveillance and MHC I–directed tumor editing was recently demonstrated in a study using Ly49-deficient mice (101). Another study demonstrated that human tumors reduced activating NK cell receptor (NKp30) recognition by NK cells by shedding B7-H6 from their cell surface (102). Interestingly, the study observed significantly elevated levels of soluble B7-H6 in the sera of patients with melanoma compared with sera of healthy donors. There is also interest in what prevents NK cells from entering tumors. In mice, a recent study reported the importance of tumor-derived IL-17D in recruiting NK cells into tumors (103). Interestingly, IL-17D gene expression was decreased in metastatic prostate tumors compared with expression levels in primary prostate tumors and was decreased in more advanced gliomas. NK cells are generally recognized to protect mice from tumor metastasis; however, this role is less well established in humans. NK cell control of hematological malignancies is an area of active interest, given that NK cells normally crosstalk with other lymphocytes and myeloid cells.

Cytokines, chemokines, and metabolic factors.

A number of other soluble factors also control the composition and function of the cellular effectors and regulators in the tumor microenvironment. Intriguingly, a high density of CD8⁺ T cells correlated with shorter survival in RCC (79, 104), ocular melanoma (105), and Hodgkin lymphoma (106). Cellular and molecular analyses of the immune microenvironment of CRC and RCC suggest that the difference may be due to strong expression of immunosuppressive cytokines (TGF-β and IL-10), inflammatory cytokines (IL-6 and TNF), and angiogenic factors (VEGF) in these cancers (107). Notably, high levels of VEGF were also shown to negate the positive influence of infiltrating Th1/CD8⁺ T cells in ovarian and colorectal cancer (108, 109). In addition, evidence is now emerging that links hypoxia-induced angiogenesis with immune tolerance (110). Altered glucose metabolism (an emerging hallmark of cancer) (2) and acidification have also been linked with denaturation of IFN-γ and suppression of TNF production (111). In addition, recent progress in imaging technologies has highlighted an underappreciated role for the extracellular matrix (ECM) in modulating the effector function of T cells (112). A light reticular fiber network, rather than a dense cross-linked matrix, may facilitate T cell migration and immunosurveillance in the tumor stroma. Finally, recent studies have also uncovered a role for carcinoma-associated fibroblasts (CAFs) expressing fibroblast activation protein (FAP) and escape from immune surveillance through production of CXCL12 (113, 114).

The impact of the local education of T and B cells in CRC is enforced by the fact that in patients with tumors in which the gene encoding CXCL13 (a TLS-generating chemokine) was deleted, the tumors were less infiltrated by Tfh and B cells, and the patients had a shorter survival (18). IL-15 promotes lymphocyte survival and differentiation, and patients whose tumors had lost IL15 had shorter survival (115). In mice, it was shown that cutaneous tumors ceased CXCL9 production as a result of IFN-γ–mediated immunoediting (116). These tumors acquired resistance to T cell–mediated immunity and were less able to recruit T and NK cells. T cell effector function can also be suppressed by inhibitory metabolic enzymes such as indolamine 2′3′-dioxygenase (IDO), an IFN-γ–induced gene that is significantly upregulated in patients with CRC-MSI, (57). Overall, these data illustrate that immune surveillance of human cancers is complex and that many parameters can influence the effector function of CD8⁺ T cells.

Note added in proof.

In a recently published article, Le et al. demonstrated that anti–PD-1 induced an objective response rate (ORR) of 40% in CRC-MSI patients compared with a 0% ORR in CRC-MSS patients (120).