It is well established and generally accepted that six cell-intrinsic (cell-autonomous) phenomena determine early oncogenesis. Cancer cells characteristically provide their own growth signals, ignore growth-inhibitory signals, avoid cell death, replicate without limits, sustain angiogenesis, and invade tissues through basement membranes and capillary walls1. In addition, as recently proposed by Schreiber and colleagues2,3, avoidance of immunosurveillance might be the seventh hallmark of cancer. Cancer cells escape innate and adaptive immune responses — cancer immunosurveillance — by immunoselection (that is, selection of non-immunogenic tumour-cell variants, a process that is also known as immunoediting) or by immunosubversion (that is, active suppression of the immune response)4.

The importance of cancer immunosurveillance is still controversial, especially among non-immunologists and, in particular, among oncologists, who often think in cell-intrinsic terms. Although evidence that immunosurveillance has a fundamental role in cancer development and anticancer therapy has been accumulating in recent years, these findings have had little impact on the accepted theories of multistep carcinogenesis and have not influenced the way in which anticancer therapies are conceived and applied in the clinic. This Review summarizes recent studies that elucidate the numerous links between cell-extrinsic (immune-mediated) and cell-intrinsic mechanisms of suppression of tumours (Fig. 1), both of which need to be subverted for cancer to develop. We also suggest that the seventh hallmark of cancer (that is, avoidance of immunosurveillance) is mechanistically linked to the six established cell-intrinsic characteristics of cancer cells.

Figure 1: Relationship between cell-intrinsic and cell-extrinsic aspects of tumour progression.
figure 1

This figure illustrates the central concept that multistep carcinogenesis results from crosstalk of cancer-cell-intrinsic factors and host immune system (cell-extrinsic) effects.

Cancer immunosurveillance

The concept of cancer immunosurveillance predicts that the immune system can recognize precursors of cancer and, in most cases, destroy these precursors before they become clinically apparent (Fig. 2). It is well established that mice that lack essential components of the innate or adaptive immune system are more susceptible to the development of spontaneous or chemically induced tumours. This is the case for animals that lack the following components: recombination-activating gene 2 (RAG2), lack of which results in the absence of T cells, B cells and natural killer T (NKT) cells; the γ-chain of the T-cell receptor (TCRγ), lack of which results in the absence of γδ T cells; TCRβ and TCRγ, lack of which results in the absence of αβ and γδ T cells; the joining gene segment Jα281, lack of which reduces the number of invariant NKT cells expressing Vα14–Jα281- containing TCRs; interferon-γ (IFNγ) receptor 1 (Ref. 5); signal transducer and activator of transcription 1 (STAT1), a transcription factor that is required for IFNγ-induced signalling; perforin; or tumour-necrosis factor (TNF)-related apoptosis-inducing ligand (TRAIL) (reviewed in Ref. 4).

Figure 2: Cancer immunosurveillance.
figure 2

a | CD8+ cytotoxic T lymphocytes (CTLs) recognize and kill stromal and/or tumour cells in an MHC-restricted and perforin-dependent manner. In parallel, they secrete the anti-angiogenic cytokine interferon-γ (IFNγ). b | Activated CD4+ T cells recognize tumour-infiltrating macrophages in an MHC-class-II-dependent manner, converting interleukin-10 (IL-10)-producing M1 macrophages into IFNγ-producing M2 macrophages. Activated CD4+ T helper 1 (TH1) cells can also release IFNγ, which inhibits angiogenesis. TH2 cells produce IL-4 and block neo-angiogenesis indirectly, through an effect on stromal fibroblasts. c | IFN-producing killer dendritic cells (IKDCs) kill tumour cells in a TRAIL (tumour-necrosis factor (TNF)-related apoptosis-inducing ligand)- and perforin-dependent manner. IKDCs are also a major source of IFNγ. IKDCs might also cross-present tumour antigens to T cells. d | During chronic myeloid leukaemia, natural killer (NK) cells can be activated in an NKG2D (NK group 2, member D)-dependent manner by dendritic cells (DCs) with a BCRABL (breakpoint-cluster region fused with Abelson leukaemia-virus protein) translocation. During chronic inflammation and in the presence of IL-4 and IL-13, DCs upregulate TREM2 (triggering receptor expressed on myeloid cells 2), which is required for NK-cell triggering. Inhibitors of KIT promote NK-cell activation through cell–cell contact between DCs and NK cells. e | Natural killer T (NKT) cells recognize glycolipids bound to CD1d at the surface of antigen-presenting cells (APCs). Activated NKT cells secrete IFNγ in a CD40- and IL-12-dependent manner and can lyse tumour cells though an MHC-unrestricted, TRAIL- or perforin-dependent pathway. f | Tumour cells or APCs present the F1-ATPase–apolipoprotein-A complex or phosphoantigens to γδ T cells with a Vγ9Vδ2-containing T-cell receptor (TCR), and these γδ T cells produce cytokines or kill tumour cells in an NKG2D-dependent manner. g | Tumour-specific antibodies could have a tumoricidal role by promoting antiproliferative effects directly or by inducing complement-mediated lysis or antibody-dependent cytotoxicity by NK cells and macrophages expressing receptors for IgG (FcγRs). Question marks denote pathways that need to be confirmed by further studies. BrHDP, bromohydrin diphosphate (also known as BrHPP); CD40L, CD40 ligand; IDP, isopentenyl diphosphate (also known as IPP); RAE1, retinoic acid early transcript 1.

Such genetic experiments are supported by studies in which antibodies are used to deplete natural killer (NK) cells and NKT cells or to neutralize TRAIL6 or the activating receptor NKG2D (NK group 2, member D)7. By contrast, an immunostimulatory regimen designed to increase the number of NK cells and invariant NKT cells can reduce the development of malignant disease in mouse models8. Furthermore, a novel immunostimulatory regimen that comprises imatinib mesylate (Gleevec or Glivec; Novartis) plus interleukin-2 (IL-2) stimulates antitumour immune responses; in particular, this regimen elicits the expansion of a novel subset of dendritic cells (DCs), IFN-producing killer DCs (IKDCs), that can mediate the regression of tumours in vivo in a TRAIL-dependent manner9. In summary, from the current scientific literature, it is possible to identify several cell types and a range of effector molecules that are involved in cancer immunosurveillance (Fig. 2).

If immunosurveillance has an important role in the suppression of tumours, then it would be expected that patients with pre-malignant or early cancerous lesions mount vigorous immune responses. There are, indeed, several lines of evidence that indicate that this might be the case. For example, patients with monoclonal gammopathy, which is pre-malignant, mount strong T-cell responses to autologous pre-malignant B cells, whereas no such responses are found in patients with multiple myeloma, which is malignant10. In addition, the bone marrow of patients with operable breast cancer contains CD8+ T cells specific for peptides derived from the breast-cancer-associated proteins mucin-1 (MUC1) and ERBB2 (also known as HER2 and NEU), and such T cells can mediate the regression of autologous human tumours that have been transplanted into immunodeficient non-obese diabetic mice11. Similar results have been reported for patients with pancreatic cancer; these patients have bone-marrow cells that can reject tumour xenotransplants when transferred to the recipient animals12.

These examples and others that are discussed later indicate that, although tumours induce at least transient immune responses, cancer can still develop. Such cellular immune responses might be too inefficient to prevent the development of cancer either because tumour cells that evade the immune response are selected (known as immunoselection) or because tumour-antigen-specific immune tolerance is induced13. Nevertheless, tumour-specific antibody responses (also known as antibody signatures) can be used to detect cancers such as prostate cancer at early stages14. Moreover, antibodies specific for cyclin B1 (Ref. 15) or the tumour-suppressor protein p53 (Ref. 16) might be useful for detecting cancer at early stages. At present, serum cyclin-B1- and MUC1-specific antibodies are being assessed for their potential as prognostic biomarkers17.

In numerous cancers, the presence of tumour-infiltrating T lymphocytes (TILs) is a useful prognostic marker. This applies, in particular, to melanoma, ovarian carcinoma and colon carcinoma18. In patients with colorectal carcinoma, the presence of mRNA encoding molecules expressed by effector T helper 1 (TH1) cells (such as CD8α, granzyme B, granulysin, T-bet, IFNγ and IFN-regulatory factor 1) and effector memory T cells correlates with reduced metastatic invasion and increased survival of the patient19. By contrast, in patients with renal-cell carcinoma, the presence of a polymorphism in the IL-4 receptor that results in increased IL-4-receptor signalling and a bias towards TH2-cell responses is an independent indicator of adverse prognosis for the patient20. In patients with ovarian carcinoma or melanoma, the presence of CD4+CD25+ regulatory T (TReg) cells in the tumour is a predictor of reduced patient survival21,22. Accordingly, ex vivo depletion of TReg cells from TIL populations recovered from renal-cell carcinomas or melanomas increases the lysis of tumour cells by NK cells23. Similarly, the responses of TILs to prostate-cancer cells can be restored ex vivo by inhibiting two immunosuppressive enzymes that are produced by tumour cells: arginase-1 and nitric-oxide synthase 2 (NOS2; also known as iNOS)24. These findings indicate that, although human tumours are often infiltrated by TILs, the antitumour activity of these T cells seems to be inhibited.

Additional evidence that supports the idea that immunosurveillance has a role in the suppression of human cancer is provided by the finding that immunodeficiencies predispose patients to the development of cancer. Immunosuppressed transplant recipients have an increased risk of developing certain cancers, including cancers without a known viral aetiology, such as melanoma25,26. However, such patients have no increased risk of developing gastrointestinal, urogenital or respiratory cancers25, indicating that the mere suppression of the immune system by glucocorticoids and cyclosporin A (both of which mainly affect T cells) is not sufficient to increase the incidence of such cancers. Patients with Chediak–Higashi syndrome, an autosomal recessive disorder that is characterized by abnormal cytotoxic function of NK cells, have a 200-fold higher risk of developing malignancy than do individuals who do not suffer from this syndrome27. In addition, biallelic mutations of the gene that encodes perforin might predispose patients to the development of Hodgkin's lymphoma or non-Hodgkin's lymphoma28. Similarly, a polymorphism in CD95 ligand (CD95L; also known as FASL) pre-disposes patients to increased activation-induced cell death of T cells and to cervical cancer29. Carrying specific combinations of HLA alleles and alleles that encode the cognate killer-cell immunoglobulin-like receptors (KIRs), which are expressed by NK cells, also results in a predisposition to cervical cancer30. Moreover, the presence of certain human NKG2D alleles, which affect the natural cytotoxic activity of peripheral-blood lymphocytes, is associated with a risk of developing cancer at any site31,32. For example, two common NKG2D haplotypes (the low NK-cell-activity-related LNK1/LNK2 haplotype, and the high NK-cell-activity-related HNK1/HNK2 haplotype) correlated with the degree of spontaneous lysis of the tumour-cell line K-562, which is an NKG2D-dependent target32. The HNK1/HNK1 haplotype was associated with a lower risk of cancer than the LNK1/LNK1 haplotype33.

Together, the data from human studies support the existence of a cancer-immunosurveillance system that involves CD8+ T cells, TH1 cells and NK cells and is locally suppressed by TReg cells and tumour-cell products. Moreover, data from several studies31,32,33 support the concept that a two-hit process facilitates the development of cancer: the first hit being the cell-intrinsic oncogenic event, and the second being an innate or adaptive deficiency in immune recognition of, or effector function against, tumour cells. Nevertheless, whether immunosurveillance and its avoidance have a role in all types of human cancer or whether they influence only a few types requires further investigation.

Intrinsic and extrinsic barriers to carcinogenesis

There are numerous tumour-cell-intrinsic barriers that inhibit the development of cancer and that lead to the suppression of tumours34. For example, oncogene-driven tumour-cell division can trigger DNA damage associated with DNA replication, which then activates the DNA-damage response. This response involves the proteins ataxia-telangiectasia mutated (ATM), checkpoint kinase 1 homologue (CHK1) and p53, and it leads to cell-cycle arrest or apoptosis35,36. Therefore, evidence consistent with activation of the DNA-damage response, as well as increased apoptosis, is frequently observed in early superficial lesions (for example, in pre-malignant lesions of bronchi, intraductal breast carcinomas and dysplastic nevi), yet these processes do not occur after full development of carcinoma or melanoma, owing to inactivation of the ATM–CHK1–p53 pathway35,36. Intriguingly, the DNA-damage response can also activate the expression of ligands for NKG2D — such as MHC-class-I-polypeptide-related sequence A (MICA) and retinoic acid early transcript 1 (RAE1) — by tumour cells, rendering the tumour cells immunogenic and possibly susceptible to killing by NKG2D-expressing cells, such as NK cells, NKT cells, γδ T cells and some cytolytic CD8+ αβ T cells. The upregulation of expression of such NKG2D ligands involves the activation of ATM, ATR (ATM and Rad3 related) and CHK1 (Ref. 37), indicating that similar or overlapping pathways couple chronic activation of the DNA-damage response to cell-intrinsic and cell-extrinsic (immune-mediated) pathways of suppression of tumours (Fig. 3).

Figure 3: Hypothetical links between endogenous tumour suppression and immunosurveillance.
figure 3

Activation of the DNA-damage response by cell-intrinsic stimuli (oncogene activation or reactive oxygen species, ROS) or therapy activates the ATM (ataxia-telangiectasia mutated)–CHK1 (checkpoint kinase 1 homologue) pathway. This pathway leads either to p53-dependent apoptosis of tumour cells or to increased immunogenicity of tumour cells as a result of upregulation of NKG2D (natural-killer group 2, member D) ligands, which increases recognition of tumour cells by NKG2D-expressing NK cells, natural killer T (NKT) cells and T cells.

With regard to cell-extrinsic mechanisms of tumour suppression, neutralization of NKG2D by injection of a specific antibody increases the sensitivity of mice to the development of methylcholanthrene (MCA)-induced fibrosarcoma38. MCA-induced fibrosarcomas from perforin-deficient mice more frequently express the NKG2D ligand RAE1 than do fibrosarcomas from perforin-sufficient mice. Moreover, the fibrosarcomas that arise in perforin-deficient mice are immunogenic when transferred to wild-type syngeneic mice38, consistent with a role for perforin (and NKG2D) in the selection of non-immunogenic tumour-cell variants (that is, in immunoselection). In addition, DNA-based vaccines that encode syngeneic or allogeneic NKG2D ligands plus tumour antigens (such as survivin and carcinoembryonic antigen) elicit both NK-cell-mediated and CD8+ T-cell-mediated anticancer immune responses39, indicating that the recognition of NKG2D ligands might have a co-stimulatory effect on T cells. Human tumour cells often overexpress the NKG2D ligands MICA and MICB, and cell-surface expression of MICA is an indicator of good prognosis in patients with colorectal carcinoma40. Conversely, proteolytic shedding of MICA and MICB from tumour cells, as occurs in advanced human cancers41, might contribute to immunosubversion by ligating NKG2D and reducing its expression at the cell surface of lymphocytes, thereby inhibiting lymphocyte function. This might occur because soluble (in contrast to membrane-bound) NKG2D ligands cause internalization of NKG2D41. Constitutive expression of the mouse NKG2D ligand RAE1 (encoded by a transgene in normal epithelial cells), or transplantation of mice with RAE1+ tumours, causes downregulation of NKG2D expression by CD8+ T cells and NK cells (both those that are infiltrating the tumour and those present in the blood), and this effect was found to be coupled to a generalized defect in NK-cell-mediated cytotoxicity42. Importantly, Rae 1e-transgenic mice with an inactive NKG2D system are more susceptible to chemical-induced carcinogenesis than are wild-type mice42, underlining the role of NKG2D in immuno-surveillance at early stages of tumour development.

Although there seems to be a link between upstream events in the DNA-damage response (those involving ATM, ATR and CHK1) and expression of NKG2D ligands37, there are no data that establish such a link for the downstream effector of the DNA-damage response, p53. Nevertheless, in addition to the induction of apoptosis, p53 has been implicated in providing a cell-intrinsic barrier to the development of cancer by mediating cellular senescence: that is, by inducing a permanent arrest in the G0/1 (gap 0/1) phase of the cell cycle43,44. The impact of cancer-cell senescence on the immune response is unknown, although it has been inferred that senescent tumour cells might 'hide' from the immune system45. Indeed, one of the markers of senescent tumour cells, decoy receptor 2 (also known as TNFRSF10D), suppresses TRAIL-induced apoptosis46. Interestingly, it seems that type I IFNs (that is, IFNα and IFNβ), which have an important role in immuno-surveillance47, can upregulate the p53-mediated response of tumour cells48. Therefore, MCA-induced fibrosarcomas derived from IFNα-receptor-1-deficient mice are rejected in a lymphocyte-dependent manner when transplanted to wild-type mice. However, recent studies indicate that tumour cells (and by extension p53) are not important targets of endogenously produced type I IFNs. Instead, type I IFNs mediate their antitumour effects through host haematopoietic cells47. So the role of crosstalk between type I IFNs and p53 in mediating tumour suppression at the intersection between cell-intrinsic and cell-extrinsic (immune-mediated) pathways remains to be established. It is possible, however, that loss of p53 function, which is a frequent occurrence in developing cancers, might affect the recognition of tumour cells by the immune system. For example, for human keratinocytes, loss of p53 increases susceptibility to type-I-IFN-induced, TRAIL-dependent apoptosis49. Whether such a mechanism contributes to the increased susceptibility of cancers to TRAIL-induced apoptosis remains to be established.

Taken together, the evidence indicates that cell-intrinsic responses that participate in the suppression of tumours are linked to the expression of immunostimulatory NKG2D ligands and, perhaps, to the modulation of TRAIL-dependent apoptosis.

The six cell-intrinsic hallmarks of cancer

There are numerous molecular differences between cancer cells and healthy cells. These differences can be classified into six characteristic changes1. Here, we briefly discuss the links between these cancer-specific characteristics and immunosurveillance (Table 1).

Table 1 Impact of cancer-cell-intrinsic characteristics on the immune system

Self-sufficiency in growth signals. Although most soluble mitogenic growth factors are produced by one cell type to stimulate proliferation of another (a process known as heterotypic signalling), many cancer cells acquire the ability to synthesize growth factors to which they are responsive. This creates a positive-feedback signalling loop that is known as autocrine stimulation. Some of the autocrine growth factors that are produced by tumour cells not only provide growth-stimulatory signals but also subvert the immune response simultaneously (Table 1). For example, IL-4 and IL-10 are autocrine growth factors for thyroid carcinoma50, and both of these cytokines can polarize the T-cell response from a TH1-cell response to a TH2-cell response, thereby inactivating anticancer immunity, which often relies on TH1-cell responses. IL-6 is one of the most important autocrine growth factors in the pathogenesis of numerous cancers, such as prostate cancer, renal cancer and myeloma. Aberrant activation of STAT3, which is involved in signalling through the IL-6 receptor, has been shown to inhibit inflammatory responses and crosstalk between innate and adaptive immune responses in various human cancers, thereby favouring unrestrained tumour growth51. STAT3 can be activated by other growth factors, particularly by the oncogenic fusion protein of PAX3 (paired box protein 3) and FKHR (forkhead homologue 1, rhabdomyosarcoma; also known as FOXO1A)52, which is associated with the alveolar subtype of rhabdomyosarcoma. So PAX3–FKHR has a dual function: it favours the transformation of the cell while causing reduced expression of MHC molecules and increased production of IL-10 by tumour cells, to inhibit surrounding inflammatory cells and detection by the immune system52. IL-6 and IL-10 have also been found to stimulate a B7-H4-expressing T-cell-suppressive macrophage population in patients with ovarian carcinoma53.

Overexpression of growth-factor receptors might enable a cancer cell to become hyper-responsive to ambient amounts of growth factors that would not normally trigger proliferation, or such overexpression might cause ligand-independent signalling. For example, the epidermal-growth-factor receptor (EGFR; also known as ERBB1) is overexpressed in stomach, brain and breast tumours, whereas expression of the receptor ERBB2 is upregulated in stomach and breast carcinomas. Both EGFR and ERBB2 are important tumour antigens for the induction of T-cell responses, and patients with tumours that overexpress either of these growth-factor receptors often mount immune responses to EGFR- or ERBB2-derived peptides54. Another way in which growth signals can be provided is an activating mutation of signal transducers such as Ki-RAS, which is frequently mutated in colon carcinoma. Mutated Ki-RAS has been assessed as a potential antigen for incorporation in antitumour vaccines, and this approach is yielding promising clinical data55. However, to our knowledge, no cases of spontaneous immune responses to Ki-RAS have been reported in patients with cancer.

These findings indicate that tumour cells can produce soluble factors that affect the tumour and the immune system of the host, and that activating mutations in growth-stimulatory signal transducers can alter the antigenic profile of tumour cells.

Insensitivity to antigrowth signals. In a normal tissue, multiple antiproliferative signals operate to maintain cell quiescence and tissue homeostasis. Such signals include those mediated by both soluble growth inhibitors and immobilized growth inhibitors that are embedded in the extracellular matrix and on the surface of neighbouring cells. One of the most important antiproliferative signals is mediated by transforming growth factor-β (TGFβ), and many tumours disable components of the TGFβ-mediated signalling pathway (Table 1): for example, by mutation or loss of TGFβ receptors, by mutation of the transcription factor SMAD4 (mothers against decapentaplegic homologue 4), or by deletion of the loci that encode INK4B (also known as p15) and retinoblastoma protein. Moreover, TGFβ favours the epithelial to mesenchymal transition of established breast-cancer cell lines and therefore might function as an autocrine and paracrine factor that allows tumour-cell motility, invasiveness and metastasis56. In addition, TGFβ is one of the most potent known immunosuppressive agents and functions at several levels. Among other functions, TGFβ reduces the amount of antigen presentation by DCs, inhibits the activity of IFNγ, reduces the proliferation of T cells, suppresses the cytotoxic activity of NK cells and stimulates the proliferation of TReg cells57. Recent data also indicate that a specific subset of tumour-associated DCs can stimulate the proliferation of TReg cells in a TGFβ-dependent manner58 and that TReg cells that accumulate in the tumour bed express membrane-bound TGFβ23. By contrast, it is thought that TGFβ can also have immunostimulatory effects (in particular, on specific TH-cell subsets59), so it would be simplistic to suggest that TGFβ is always immunosuppressive. However, there are data that point to the in vivo relevance of endogenous TGFβ: adoptive transfer of TGFβ-desensitized CD8+ T cells that had been transfected with a dominant-negative TGFβ receptor 2 was found to be particularly efficient at eradicating prostate cancer in mice60.

Taken together, these data highlight that, by ignoring TGFβ-mediated growth-inhibitory signals, tumour cells can replicate in an immunosuppressive environment that is rich in TGFβ.

Evasion of apoptosis. The apoptotic programme is carried out through two main pathways: the mitochondrial pathway, which involves mitochondrial outer-membrane permeabilization (MOMP)61; and the death-receptor pathway, which involves ligation of a plasma-membrane receptor, leading to the formation of a death-inducing signalling complex (DISC)62. Both pathways culminate in the activation of caspases and caspase-independent cell-death pathways63. Defective apoptosis is not crucial for initial oncogenesis, but it contributes to the acquisition of resistance to chemotherapy and to immune effectors64. Tumour cells can develop resistance to the main apoptosis-inducing effector molecules of the innate and adaptive immune systems (Table 1). For example, aggressive tumour variants that are selected in vivo in mice that have been administered tumour-antigen-specific cytotoxic T lymphocytes (CTLs) are particularly resistant to IFNγ-induced transcriptional effects, inclu-ding IFNγ-induced expression of CD95 and killing induced by CD95 (Ref. 65).

Important endogenous inhibitors of MOMP are the anti-apoptotic proteins of the B-cell lymphoma 2 (BCL-2) family, which includes BCL-2, BCL-XL and myeloid-cell leukaemia sequence 1 (MCL1). It has been reported that patients with cancer at various sites have peripheral-blood CTLs that recognize peptides derived from these proteins, whereas healthy individuals lack such spontaneous immunity to BCL-2, BCL-XLand MCL1 (Ref. 66). Other relevant inhibitors of MOMP that do not belong to the BCL-2 family have recently been identified. The carboxy-terminal subunit of the epithelial-cell protein MUC1 localizes to mitochondria and blocks stress-induced activation of the mitochondrial apoptotic pathway67. MUC1 — which is a heterodimeric transmembrane glycoprotein that is overexpressed by most human carcinomas — is a T-cell antigen that is presented by various tumours, such as breast, colon, pancreatic, ovarian and lung carcinomas. Owing to its differential glycosylation in normal cells and tumour cells, MUC1 might be tumour-cell-specific antigen. In addition, MUC1 can have several immuno-suppressive effects: inhibiting the differentiation of monocytes into DCs, skewing the differentiation of DCs into cells with an IL-10hiIL-12low regulatory phenotype68, and functioning as a chemoattractant for immature DCs and then subverting their ability to stimulate TH1-cell responses69. Another inhibitor of MOMP that is relevant to cancer is survivin70, which is also an important tumour antigen71. For example, survivin induces upregulation of expression of the death-receptor ligand CD95L by colon carcinomas72. The expression of CD95L at the surface of tumour cells might contribute to the deletion of CD95-expressing activated T cells that invade the tumour73, although this possibility is controversial. CD95L might also elicit an inflammatory response74. Irrespective of whether survivin overexpression or other mechanisms account for the expression of CD95L by tumours, this expression is associated with a poor prognosis for patients with various conditions, such as Barrett's oesophagus75 and colon carcinoma76, for which CD95L expression has been associated with increased apoptosis of T cells.

Tumour cells often lose the expression of functional APAF1 (apoptotic-protease-activating factor 1), which is required for the apoptosome-dependent activation of caspases after MOMP. Moreover, tumours often overexpress inhibitor-of-apoptosis proteins (IAPs) — such as XIAP (X-linked IAP) and IAP1 — which inhibit caspases. Although the inactivation of APAF1 (Ref. 77) or the presence of IAPs is likely to have little impact on the survival of tumour cells (which remain susceptible to caspase-independent cell death)63, these factors are associated with a poor prognosis for patients with head and neck squamous-cell carcinoma (IAP1 overexpression)78, acute de novo myeloid leukaemia (XIAP overexpression)79, or melanoma or colorectal carcinoma (reduced APAF1 expression)80,81. Nevertheless, it is possible that the lack of caspase activation might affect the immunogenicity of cell death (Box 1). Indeed, activation of caspases might be required to elicit immunogenic tumour-cell death in an in vivo model of anthracycline-mediated antitumour chemotherapy82, and an adenovirus that encodes caspase-1 and IL-12 was found to improve the survival of mice with adenocarcinoma of the prostate83.

Evolving pre-malignant and malignant cell populations show chronic, widespread apoptosis and therefore suffer considerable cell attrition concomitant with cell accumulation. As a result, reducing the immunogenicity of apoptosis might be an important strategy that tumours use to avoid destruction by the immune system, not only after episodes of apoptosis-inducing chemotherapy but also during spontaneous cancer growth.

Limitless replicative potential. Replication without limits requires maintenance of the ends of the chromosomes, which are known as telomeres. Most (85–90%) malignant cells succeed in doing so by upregulating expression of the enzyme telomerase, which adds hexanucleotide repeats to the ends of telomeric DNA. The remaining 10–15% of malignant cells can activate a mechanism that is known as alternative lengthening of telomeres, which seems to maintain telomeres through recombination-based interchromosomal exchanges of sequence information. The catalytic subunit of human telomerase, TERT (telomerase reverse transcriptase), is a tumour antigen to which patients with malignant disease, such as breast cancer84 and chronic myeloid leukaemia85, mount immune responses. In addition, to ensure replicative potential, tumour cells can have mutations in, or lose expression of, senescence-inducing proteins, such as p53 (which can also function as a tumour antigen)86 (Table 1). In tumours that do not express p53, expression of the cell-cycle regulator cyclin B1 is increased, and cyclin B1 is also a potential tumour antigen15. However, there are no known immunomodulatory effects of cancer-cell immortalization, apart from the previously mentioned tumour-specific epitopes that can elicit spontaneous immune responses.

Sustained angiogenesis. Many tumour cells show higher expression of vascular endothelial growth factor (VEGF) and/or fibroblast growth factor 1 (FGF1) and FGF2 than their counterparts in normal tissue, and these factors are crucial for mediating an 'angiogenic switch' from the state of vascular quiescence. In addition to promoting angiogenesis, VEGF might inhibit antitumour immunity, both by inhibiting the activation of nuclear factor-κB in DCs, thereby preventing DC maturation87, and by suppressing the activation of T cells88 (Table 1). Consistent with these postulated immunosuppressive functions of VEGF, administration of VEGF-specific antibody increases the efficacy of immunotherapy in mouse models89.

Overexpression of the pro-inflammatory mediator cyclooxygenase-2 (COX2) is a common characteristic of various pre-malignant and malignant lesions of epithelial-cell origin in organs such as the colon, lungs, breasts, prostate, bladder, stomach and oesophagus. Moreover, inhibition of COX2 has been shown to have chemopreventive and, perhaps, chemotherapeutic effects in various human cancers, particularly in colon cancer. COX2 causes the local overproduction of prostaglandin E2 (PGE2), which is thought to have an important role in favouring angiogenesis, partly through the induction of VEGF production90. Moreover, COX2 and prostanoids, particularly PGE2, suppress antitumour immunity, by suppressing macrophage-mediated or T-cell-mediated tumour killing and by polarizing the balance of TH-cell responses towards TH2-cell responses90. Accordingly, selective inhibition of COX2 can restore the tumour-induced imbalance between IL-10 (a TH2 cytokine) and IL-12 (a TH1 cytokine) in mice with lung cancer91 and can re-establish impaired mononuclear-cell function in patients with head and neck cancer92.

Tissue invasion and metastasis. The settling of tumour cells in locations that are distant from the primary tumour — metastasis — is the cause of 90% of human deaths from cancer, and selective pressure is likely to guide cancer cells to niches where they can 'hide' from the immune response. The invasion of distant sites by tumour cells requires alterations in the proteins that are involved in the tethering of cells to their surroundings, as well as in the secretion of proteases. In terms of tethering, the tumour-suppressor gene TSLC1 (tumour suppressor in lung cancer 1), which is frequently inactivated in non-small-cell lung-cancer cells, encodes the protein nectin-like 2 (NECL2). NECL2 mediates epithelial-cell junctions though homotypic and/or heterotypic interactions with other nectins or nectin-like proteins. Moreover, NECL2 is recognized by CRTAM (MHC-class-I-restricted T-cell-associated molecule), which is expressed by NK cells and CD8+ T cells, and promotes the cytotoxicity of NK cells and production of IFNγ by T cells93. Therefore, inactivation of NECL2 expression has a dual oncogenic effect: it allows the disorganization of epithelial-cell layers, and it removes a protein that is recognized by NK cells and T cells (Table 1).

The secretion of proteolytic enzymes, such as cathepsins, is also involved in the local invasion of a tissue by tumour cells. One of the hallmarks of cancer is increased secretion of lysosome-resident proteases, a process that seems to be associated with increased amounts of heat-shock protein 70 (HSP70) in both the lysosome membrane and the plasma membrane94. In theory, this could increase the immunogenicity of tumour cells, because HSP70 is thought to be one of the main damage-associated molecular pattern molecules and can chaperone tumour peptides for the cross-priming of CD8+ T cells95,96. Moreover, HSP70hi tumour cells are killed more efficiently by NK cells than are their HSP70low counterparts. Importantly, colon- and pancreatic-carcinoma cells can release HSP70-bearing exosomes that stimulate the cytolytic activity of NK cells against tumours97.

Local invasion is also mediated by matrix metalloproteinases (MMPs) that are secreted by tumour cells and stromal cells. MMPs are also involved in the escape of cancer cells from immunosurveillance: for example, through MMP9-mediated cleavage of CD25, activation of TGFβ, and shedding of intercellular adhesion molecule 1 (ICAM1) and ICAM2 from tumour cells98.

These examples illustrate that the molecular features that determine the capacity of tumour cells to invade tissues and to metastasize also affect the immunogenicity of tumour cells.

The seventh hallmark of cancer

In humans, tumours develop a series of strategies to evade immunosurveillance, and these strategies are presumably unrelated to the other characteristics of carcinogenesis and result from the selective pressure exerted by the immune system. These strategies are known as immunoselection. One common strategy to elude a T-cell-mediated immune response is the downregulation or loss of expr ession of HLA class I molecules, as has been documented for a large range of epithelial-cell cancers and melanomas. Loss of HLA class I expression is common in lung cancer, and it has been postulated that only immunoselected tumour cells that lack HLA class I expression can escape immune attack and develop into cancer99. Similarly, tumours often downmodulate molecules that are involved in antigen processing and in antigen presentation by HLA class I molecules — including transporter associated with antigen processing 1 (TAP1), low-molecular-mass protein 2 (LMP2), LMP7 and tapasin — the expression of which is progressively lost during the development of colorectal carcinoma100 (Fig. 4a). Tumour cells can also develop mechanisms to avoid being killed by CTLs. Overexpression of the serine-protease inhibitor PI9 by tumour cells efficiently blocks the granzyme-B–perforin pathway of target-cell lysis101. Similarly, additional outcomes of immunoselection include downregulation or mutation of death receptors, methylation or mutation of the gene that encodes caspase-8, and overexpression of FLIP (caspase-8 (FLICE)-like inhibitory protein) or decoy receptors for TRAIL, all of which cause resistance to CTL-induced killing of tumour cells102.

Figure 4: Mechanisms of tumour escape from the immune system.
figure 4

a | As a result of immunoselection, tumour variants lose antigen-processing machinery, tumour antigens and sensitivity to immune effectors such as interferons (IFNs). b | Tumour-derived factors recruit myeloid suppressor cells (MSCs) and prevent their differentiation into mature dendritic cells (DCs), in a STAT3 (signal transducer and activator of transcription 3)-dependent manner. MSCs inhibit tumour-specific T cells through arginase-1 (ARG1) or nitric-oxide synthase 2 (NOS2). c | The interaction between CD80 or CD86, at the surface of DCs, and CTLA4 (cytotoxic T-lymphocyte antigen 4), at the surface of T cells or CD4+CD25+ regulatory T (Treg) cells, induces the production of IFNγ and the immunosuppressive factor indoleamine 2,3-dioxygenase (IDO) by DCs. This results in a reduction in the amount of tryptophan, which is T-cell tropic, and in the generation of kynurenines, which kill T cells. d | Plasmacytoid DCs (pDCs) activated by interleukin-3 (IL-3) and CD40 ligand (CD40L) promote the differentiation of naive CD4+ and CD8+ T cells into T helper 2 (TH2) cells and anergic IL-10-producing CD8+ regulatory T cells, respectively. This state of anergy (with respect to tumour-cell lysis) is mediated by IL-10, either directly (by interaction with cytotoxic T lymphocytes, CTLs) or indirectly (by inhibition of DCs). e | Repetitive stimulation of naive T cells with immature DCs results in T-cell anergy, together with IL-10 production and IL-10-independent, cell-contact-dependent regulatory activity. f | B7-H1 (and B7-H4) is expressed by some tumours (for example, in response to IFNγ), and it directly promotes T-cell apoptosis through programmed cell death 1 (PD1)-dependent pathways or PD1-independent pathways (which are mediated by IL-10 or CD95). g | CD4+ natural killer T (NKT) cells produce IL-13, which suppresses CTL-mediated tumour rejection through a pathway that involves the a-chain of the IL-13 receptor (IL-13Rα) and STAT6. IL-13 produced by NKT cells can also activate MSCs to produce transforming growth factor-β (TGFβ), which suppresses CTLs. h | Vascular leukocyte cells (VLCs) and pDCs are attracted to tumour beds through β-defensins and CXC-chemokine ligand 12 (CXCL12), respectively. The subsequent angiogenic effect is mediated by CXCL8 in the case of pDCs and by vascular endothelial growth factor (VEGF) in the case of VLCs, which can differentiate into endothelial cells or into bona fide DCs during acute inflammation. APC, antigen-presenting cell; IFNγR, IFNγ receptor; IL-3R, IL-3 receptor; LMP2, low-molecular-mass protein 2; M-CSF, macrophage colony-stimulating factor; TAP1, transporter associated with antigen processing 1; TCR, T-cell receptor; TNF, tumour-necrosis factor.

Several tumour products that are dispensable for cell-intrinsic cancer-cell characteristics (discussed earlier) might be involved in immunosubversion: that is, the active suppression of the immune response. For example, some tumours (or tumour-associated myeloid cells) overproduce nitric oxide and have increased arginase-1 activity, both of which can inhibit T-cell function103 (Fig. 4b). More importantly, tryptophan degradation by indoleamine 2,3-dioxygenase (IDO), which is constitutively expressed by human tumours (particularly by prostate, colon and pancreatic carcinomas, but also by interdigitating DCs), promotes resistance to immune-mediated rejection of the tumour cells104. Locally produced IDO can block the proliferation of CD8+ T cells at the tumour site104 (Fig. 4c), as well as promote the apoptosis of CD4+ T cells105. Expression of CD95L by the tumour itself or by locally activated T cells might also induce the death of CD95-expressing tumour-specific T cells73.

In mouse models, advanced cancer invariably subverts immune function. Typically, tumour-specific CD8+ T cells are activated at the stage of initiation of tumour growth, but these cells show a progressive loss of cytolytic function at the later stage of tumour expansion106. Similarly, tumour-specific CD4+ T cells progressively lose their antitumour activity107, whereas the number of TReg cells increases108. Indeed, the induction of tolerance to the tumour might even be a required factor for the initial steps of tumorigenesis13.

The exact molecular mechanisms by which tumours mediate immunosubversion are the subject of intense investigation. One possible explanation for how tumours subvert the immune response is to consider that the tumour is a 'false' lymphoid organ; therefore, T-cell priming in the tumour microenvironment is defective as a result of the presence of dysfunctional or tolerogenic antigen-presenting cells109. Indeed, tumour beds contain various factors (such as VEGF, IL-6, IL-10, TGFβ, macrophage colony-stimulating factor (M-CSF), NOS2, arginase-1, IDO, PGE2, COX2 and gangliosides) that can inhibit the differentiation, maturation and function of DCs109 (Fig. 4b,d,e). Accordingly, local DCs tend to mediate immunosuppressive, rather than immunostimulatory, effects and to promote TReg-cell differentiation58. In an ultraviolet-irradiation-induced tumour model, irradiation-induced immunosuppression was found to be mediated by CD1d-restricted NKT cells. These CD4+ NKT cells produced IL-13, which suppressed CTL-mediated tumour rejection. Moreover, IL-13 from NKT cells activated myeloid suppressor cells to produce TGFβ, which also suppressed CTL activity110 (Fig. 4g). In addition, tumour-associated macrophages mostly belong to the M2 class of macrophages, which produce arginase-1, IL-10, TGFβ and PGE2, and favour TH2-cell responses (whereas M1 macrophages, which are tumoricidal, produce NOS2, IL-12 and lymphotoxin-α)111. Moreover, in patients with ovarian carcinoma, plasmacytoid DCs (pDCs) have been shown to induce the clonal expansion of IL-10-producing CD8+ regulatory T cells112 (Fig. 4d).

Another possible explanation for tumour-mediated immunosubversion is based on a quantitative argument. Tumour characteristics that are immunostimulatory in small tumours can become immunosuppressive in large tumours. For example, the expression of NKG2D ligands (which stimulates an immune response at the initial stages of oncogenesis, as discussed earlier) seems to be immunosuppressive in larger tumours. NKG2D-ligand-expressing tumour cells (as well as soluble NKG2D ligands that are shed from tumour cells) can downregulate NKG2D expression by CD8+ T cells and NK cells or can uncouple NKG2D signalling from intracellular mobilization of Ca2+ or cell-mediated cytolysis, thereby contributing to suppression of the immune response113. Similarly, it could be argued that large tumours cause a general or specific downregulation of T-cell responses as a result of 'high-dose tolerance' to tumour antigens. Following successful systemic chemotherapy — for example, for ovarian carcinoma — CD8+ T-cell function can recover114, indicating that antitumour chemotherapies that have limited immunosuppressive side-effects can restore the normal immune response by abolishing tumour-mass-related immunosubversion. If this quantitative argument is correct, immunosubversion would be particularly important in advanced cancer.

In conclusion, it seems that there are numerous ways by which tumour cells can evade or 'paralyse' immunosurveillance. However, which of these multiple mechanisms affects oncogenesis and cancer progression in humans remains an open question.

Of note, in some cases, it is possible that, although an immune response to tumours is mounted, this response fails to eliminate the tumours or could even stimulate carcinogenesis and tumour progression, as a result of chronic inflammation. For example, in mouse models, CD4+ T cells have been shown to contribute to squamous-cell carcinogenesis induced by human-papillomavirus antigens115. Moreover, transplantation of allogeneic haematopoietic cells has been shown to increase microsatellite instability in transformed epith-elial cells from patients with squamous-cell cancer116. In patients with breast, prostate or bladder cancer, the secretion of IL-4 might upregulate the expression of anti-apoptotic proteins (such as FLIP and BCL-XL) by the tumour117. Although cytotoxic agents produced by immune cells can mediate tumour destruction, this leads to the release of tumour-cell antigens, which can function as tissue-specific chemoattractants and promote leukocyte migration. For example, tumour antigens (or self antigens) — such as gp100 (glycoprotein 100), MUC1 and carcinoembryonic antigen — can attract immature DCs (for example, through binding of gp100 to CC-chemokine receptor 2)118, but tumour antigens cannot promote the maturation of immature DCs at these chemotactic concentrations. Such increased trafficking of undifferentiated immature DCs to tumour beds might accelerate tumour growth in the absence of maturation stimuli119. Vascular leukocyte cells (VLCs) and pDCs can be recruited to tumour beds by pro-inflammatory mediators — such as β-defensins and CXC-chemokine ligand 12 (CXCL12), respectively — and contribute to angiogenesis. VLCs contribute to angiogenesis by differentiating into endothelial cells in the presence of local VEGF, and pDCs contribute by secreting the angiogenic cytokines TNF and CXCL8 (also known as IL-8)120,121 (Fig. 4h). These examples illustrate (without clarifying how) that an incomplete antitumour immune response might benefit the tumour but not the host.

Concluding remarks

Clinical oncologists tend to conceive of, and treat, cancer as a cell-intrinsic phenomenon, ignoring the contribution of the innate and adaptive immune systems to the therapeutic response. Given the data that we review here, it is plausible that evasion or subversion of cancer immunosurveillance is a central hallmark of many malignancies. It can be anticipated that progress in molecular profiling of cancers and in individually applied pharmacogenetic and immunogenetic approaches will facilitate patients' understanding of the cancer-immunosurveillance system. Comprehensive information on the tumour and the immune status of an individual could be expected to provide a precise picture of the ongoing evolution of the tumour (and therefore a useful tool for prognostic extrapolation), as well as to yield invaluable information about which strategy (surgery, chemotherapy, radiotherapy and/or immunotherapy) will result in an optimal therapeutic outcome.

Owing to the mainly nonspecific cytostatic and cytotoxic effects, most current chemotherapeutic regimens are immunosuppressive. However, some chemotherapeutic agents can stimulate an antitumour immune response, by having 'side-effects' on the immune system. For example, low doses of cyclophosphamide can selectively deplete and inhibit TReg cells, thereby restoring normal CTL and NK-cell function in patients with cancer122,123. Similarly, imatinib mesylate can stimulate the activity of NK cells, correlating with a favourable clinical outcome for patients with gastrointestinal stromal tumours124. Other chemotherapeutic agents can increase the antigenic properties of tumour cells. Therefore, intratumoral injections of anthracyclines and local irradiation of established tumours can promote immunogenic tumour-cell death82. Several cytotoxic drugs can stimulate tumour-cell-surface expression of proteins that facilitate tumour-cell lysis or tumour-cell recognition. As an example, numerous DNA-damaging agents induce the expression of death receptors such as TRAIL receptor 1 (TRAILR1), TRAILR2 and CD95 by tumour cells, thereby facilitating the lysis of these cells by TRAIL- or CD95L-expressing immune effector cells125. In addition, radiotherapy increases the expression of MHC class I molecules and improves antigen presentation126. Also, histone-deacetylase inhibitors induce hepato-cellular-carcinoma cells to express NKG2D ligands (such as MICA and MICB), thereby increasing the lysis of these cells by NK cells127. It is our hope that similar therapeutic strategies will improve the effcacy of antitumour therapies, especially if they are combined with immunostimulatory regimens.