The daily demise of several billions of normal cells from the human body goes virtually unrecognized by the immune system. This is important as the preservation of whole-body homeostasis involves the continuous turnover of multiple cellular compartments, and the activation of an immune response against dead cell-associated antigens would have catastrophic consequences1,2. Conversely, the death of only a few cells infected by an infectious pathogen can trigger a robust antigen-specific immune response. In this context, successful immune responses not only clear invading pathogens from the body, but also result in the establishment of long-term immunological memory2. What are the differences between these two instances of cellular demise? For decades, the 'self/non-self' model has been used as the sole framework to differentiate between homeostatic (that is, self and non-antigenic) and pathogen-driven (that is, non-self and antigenic) forms of cell death. However, work from the late 1990s unveiled the limitations of this model by demonstrating that endogenous entities can also initiate an immune response, at least under specific circumstances3,4. Such a paradigm shift revolutionized immunology as it pointed to the existence of at least one factor other than antigenicity that explains why some, but not all, forms of cell death are immunogenic.

Even before pathogens elicit adaptive immunity, specific microorganism-associated molecular patterns (MAMPs) are detected by sensors that are expressed by a wide variety of cells, including monocytes, macrophages, dendritic cells (DCs) and other components of the innate immune system5. Such MAMPs operate as natural adjuvants, and their interaction with pattern recognition receptors (PRRs) not only establishes a first line of defence against infection but also generates the ideal conditions for the initiation of antigen-specific immune responses5,6. MAMPs do not invariably exert net immunostimulatory effects, and in some instances they are actively involved in the establishment of immunological tolerance to microorganisms and symbiosis7. However, the mechanisms that determine whether specific MAMPs or combinations thereof mediate immunostimulatory versus immunosuppressive effects remain to be clarified7. Irrespective of this unknown, the activation of adaptive immunity against endogenous entities also relies on PRR signalling. In this latter scenario, PRRs are activated by damage-associated molecular patterns (DAMPs). Similar to their microbial counterparts, DAMPs produced by dying cells can act as adjuvants and communicate a state of danger to the organism8. However, these DAMPs are unable to initiate an adaptive immune response unless dying cells display an increased antigenicity — that is, they possess antigenic epitopes that have not previously elicited central or peripheral tolerance. Such neo-epitopes may be encoded by microbial genes, as well as by host genes that mutate in the course of oncogenesis and tumour progression2. Taken together, these observations indicate that the immunogenicity of cell death relies on a combination of antigenicity (provided by neo-epitopes) and adjuvanticity (conferred by specific MAMPs or DAMPs).

Corroborating the importance of both antigenicity and adjuvanticity for the engagement of antigen-specific immune responses by dying cells, human malignancies with a high mutational load show a superior response to immunotherapy with checkpoint blockers9 than tumours with a relatively low number of somatic mutations, and such a response mostly (if not entirely) depends on adaptive immunity. Cancer cells with innate or experimentally enforced defects in pathways that are required for cell death-associated DAMP release, including autophagy and the unfolded protein response (UPR), fail to die in an immunogenic manner in response to stimuli that would otherwise cause bona fide immunogenic cell death (ICD)10,11. Moreover, artificially boosting the availability of specific DAMPs efficiently converts non-immunogenic forms of cell demise into instances of ICD12.

Thus, in the presence of increased antigenicity, adjuvanticity must be limited for cell death to be overlooked by the immune system, and both pathogens and malignant cells evolve under such a selective pressure. This implies that DAMPs occupy a privileged position in the mechanisms that determine the immunogenicity of cell death, irrespective of whether cells succumb to exogenous or endogenous cues. Here, we discuss current knowledge on the mechanisms by which cell death is perceived as immunogenic, focusing on the stress response pathways that underlie DAMP emission by dying cells, the receptors and cells of the host that detect DAMPs, and the pathophysiological implications of these processes.

Forms of ICD

For a long time, cell death has been misleadingly classified in a dichotomic manner. Thus, although apoptosis (defined on the basis of morphological features) was considered to be a physiological, regulated and non-immunogenic (or even tolerogenic) instance of cell death, necrosis (also associated with specific morphological traits) was viewed as a pathological, incontrollable and immunogenic variant of cellular demise13. Now it has become evident that such clear-cut differences do not exist. Thus, cells may succumb to the activation of a genetically encoded molecular machinery (that is, in a regulated manner) while exhibiting a necrotic morphotype; regulated forms of necrosis participate in development and tissue homeostasis, and apoptotic cells can trigger an antigen-specific immune response13. One of the consequences of this conceptual revolution is that the gold-standard approach to determine whether cell death may be immunogenic no longer involves morphological or biochemical assessments on dying cells, but rather relies on vaccination experiments in which murine dying cells are injected into immunocompetent syngeneic mice14 (Box 1). No less than four types of ICD have been discovered so far, each of which relies on the emission and detection of a specific panel of DAMPs (Fig. 1; Table 1).

Figure 1: Differential requirements for the immunogenicity of cell death.
figure 1

a | Several processes have been linked to the immunogenicity of cell death, including the unfolded protein response and consequent exposure of calreticulin (CALR) and other endoplasmic reticulum (ER) chaperones on the cell surface; the activation of autophagy and consequent secretion of ATP; the release of interleukin-1β (IL-1β) upon inflammasome signalling; the activation of Toll-like receptor 3 (TLR3), resulting in a type I interferon (IFN) response that stimulates the production of CXC-chemokine ligand 10 (CXCL10); as well as the release of high-mobility group box 1 (HMGB1) and annexin A1 (ANXA1). b | Distinct variants of immunogenic cell death depend on (or are associated with) the emission of different sets of danger signals from dying cells. PDT, photodynamic therapy.

PowerPoint slide

Table 1 Danger signalling in immunogenic cell death

ICD driven by pathogens. Cell death constitutes one of the most ancient mechanisms of defence against invasion by pathogens. Although the delayed death of infected cells eventually favours pathogen spreading, both viruses and obligate intracellular bacteria such as Salmonella enterica require living and metabolically active cells to replicate, at least in the first phases of infection15,16. Early after (or even before) infection, cells can sense multiple MAMPs by means of specific PRRs, each of which operates to favour the disposal of the pathogen and to communicate the incipient danger to neighbouring cells. Thus, microbial components as diverse as lipopolysaccharide, lipoteichoic acid, flagellin, unmethylated CpG-containing oligodeoxynucleotides, and single- or double-stranded RNA are rapidly detected by dedicated Toll-like receptors (TLRs, which are located on the plasma membrane or within endosomal compartments), cytosolic DNA sensors, RIG-I-like receptors or NOD-like receptors (which are mostly cytoplasmic) to elicit an intracellular and microenvironmental danger response17,18,19,20. Intracellular danger signalling involves the activation of autophagy and the UPR21,22, whereas the microenvironmental danger response relies on the PRR-driven secretion of pro-inflammatory cytokines, including (but not limited to) tumour necrosis factor (TNF) and type I interferons (IFNs)23. In addition, in specific cell types, including macrophages, the extracellular battle against infectious challenges also depends on inflammasome activation within dying cells, culminating in the secretion of mature interleukin-1β (IL-1β) and IL-18 (Ref. 24).

Beyond its fundamental contribution to the preservation of cellular homeostasis in physiological conditions25, autophagy has a major role in the control of invading viral or bacterial pathogens26,27. Accordingly, defects in the molecular machinery for autophagy increase the susceptibility of mice to infection by several viruses and bacteria26,27. The UPR is coupled to the inactivating phosphorylation of eukaryotic translation initiation factor 2A (eIF2A), which results in prominent antiviral effects owing to the inhibition of CAP-dependent protein synthesis22. Thus, replacing endogenous eIF2A with a non-phosphorylatable variant (eIF2AS51A) abrogates the control of viral infection in mice28. Besides exerting antiviral and antibacterial effects at the intracellular level, autophagy and the phosphorylation of eIF2A are connected to the release of DAMPs by mouse and human cancer cells succumbing to chemotherapy- or irradiation-driven ICD10 (see below). However, to what extent autophagy and the UPR contribute to the immunogenicity of cell death triggered by viruses and intracellular bacteria has not yet been determined. Regardless of this unknown, when infected cells die, their corpses are rapidly taken up by professional antigen-presenting cells (APCs)23. Owing to the presence of viral or bacterial components, these corpses also contain an abundant repertoire of non-self antigenic epitopes (which by definition are not subjected to central or peripheral tolerance). In these conditions, cellular corpses and their debris are rapidly processed by DCs and presented on MHC class I and class II molecules to antigen-specific CD8+ and CD4+ T cells, respectively, resulting in the elicitation of potent responses associated with immunological memory, especially in the context of robust PRR signalling29,30 (Fig. 1).

ICD elicited by chemotherapeutics. Mouse cancer cells exposed to some chemotherapeutics that are currently used in the clinic, including doxorubicin, mitoxantrone, oxaliplatin and bortezomib, undergo bona fide ICD, as demonstrated by vaccination experiments in mice31,32. In mice, chemotherapy-driven ICD relies on the eIF2A phosphorylation-dependent exposure of endoplasmic reticulum (ER) chaperones such as calreticulin (CALR)33, protein disulfide isomerase family A member 3 (PDIA3; also known as ERp57)34, heat shock protein 70 kDa (HSP70; also known as HSPA1A)35 and heat shock protein 90 kDa (HSP90; also known as HSP90AA1)35 on the plasma membrane of dying cancer cells. Moreover, it involves the autophagy-mediated secretion of ATP36, the activation of a cancer cell-intrinsic type I IFN response and consequent secretion of CXC-chemokine ligand 10 (CXCL10)37, as well as the release of high-mobility group box 1 (HMGB1)38 and annexin A1 (ANXA1)39. Many of these manifestations of ICD have also been observed in human cancer cells succumbing to immunogenic chemotherapy10,11,35, although human cells cannot be characterized in vaccination assays. A panel of experimental interventions that interfere with the ability of mouse cancer cells to release these DAMPs as they succumb to ICD-promoting chemotherapeutics abrogate their capacity to vaccinate mice against a subsequent challenge with living cells of the same type; these include small-interfering RNA (siRNA)-mediated downregulation of CALR33 or HMGB1 (Ref. 38), short-hairpin RNA (shRNA)-mediated knockout of essential components of the autophagic machinery such as autophagy related 5 (ATG5) or ATG7 (Ref. 36), the overexpression of the extracellular ATP-degrading enzyme ectonucleoside triphosphate diphosphohydrolase 1 (ENTPD1; also known as CD39)36,40, as well as the inactivation of both copies of Ifnar1 (which encodes interferon α/β-receptor subunit 1) or Anxa1 (Refs 37,39) (Figs 1,2).

Figure 2: Mechanisms of chemotherapy-driven ICD.
figure 2

In response to inducers of immunogenic cell death (ICD), such as doxorubicin or oxaliplatin, malignant cells expose calreticulin (CALR) and other endoplasmic reticulum chaperones on their surface, secrete ATP, initiate a cell-intrinsic type I interferon (IFN) response culminating in the production of CXC-chemokine ligand 10 (CXCL10), and release high-mobility group box 1 (HMGB1) and annexin A1 (ANXA1). Upon binding to cognate receptors on the surface of myeloid or lymphoid cells, these damage-associated molecular patterns favour the uptake of cell corpses and debris thereof by antigen-presenting cells, including dendritic cells (DCs) in the context of robust immunostimulatory signals, eventually leading to the priming of an adaptive immune response involving both αβ and γδ T cells. In addition to being associated with the establishment of immunological memory, such a response has the potential to eradicate malignant cells that survive chemotherapy via an IFNγ-dependent mechanism. CXCR3, CXC-chemokine receptor 3; FPR1, formyl peptide receptor 1; IFNAR1; interferon α/β-receptor subunit 1; IL, interleukin; LRP1, LDL receptor related protein 1; P2RX7, purinergic receptor P2X7; P2RY2, purinergic receptor P2Y2; TLR4, Toll-like receptor 4.

PowerPoint slide

Along similar lines, chemotherapeutic agents that are intrinsically unable to promote the release of one or more of these DAMPs (or the activation of the underlying stress responses) a priori fail to promote ICD. As an example, cisplatin differs from its derivative oxaliplatin in its ability to trigger the UPR and the consequent translocation of CALR to the outer leaflet of the plasma membrane of dying cells41. Accordingly (and in spite of an otherwise similar activity profile), cisplatin-treated mouse cancer cells fail to vaccinate syngeneic hosts in conditions in which oxaliplatin-treated cells efficiently do so41. In this setting, the exogenous co-provision of a UPR inducer such as thapsigargin or tunicamycin efficiently restores the immunogenicity of cisplatin-elicited cell death, demonstrating that defects in the intracellular mechanisms that underlie the emission of ICD-associated DAMPs can be corrected for therapeutic purposes42. Such an intrinsic discrepancy between cisplatin and oxaliplatin also suggests that the ICD-promoting capacity of one specific intervention cannot be predicted based on structural or biochemical features but must be evaluated in vaccination assays14 (Box 1).

Importantly, the degree of antigenicity displayed by mouse and human cancer cells varies to a significant extent, reflecting the elevated heterogeneity of mutational load that is observed even across cancers of the same type9. It is therefore tempting to speculate that — besides affecting the response of patients with melanoma, non-small cell lung carcinoma or colorectal cancer to immunotherapy with checkpoint blockers43,44,45 (which fully relies on the activation or reactivation of a tumour-targeting immune response) — mutational burden may influence the propensity of cancer cells undergoing ICD to engage adaptive immunity. However, to the best of our knowledge, all malignant cells challenged so far with bona fide ICD inducers efficiently triggered protective immune responses in vaccination tests10. Thus, it is possible that even the lowest level of mutational load associated with oncogenesis generates sufficient antigenicity to support immunogenicity. Indeed, neoplastic cells express several neo-antigens that (at least in the initial phases of malignant transformation and tumour progression) constitute modified variants of self that are not subject to central and peripheral tolerance9 (Box 2).

ICD activated by physical cues. So far, three distinct physical interventions have been shown to trigger bona fide ICD of mouse cancer cells, as documented in vaccination assays: irradiation, hypericin-based photodynamic therapy (PDT) and high hydrostatic pressure46,47,48. Importantly, the induction of ICD by these interventions does not reflect the mere disassembly of the plasma membrane and consequent spillage of cytoplasmic content into the extracellular milieu49. Indeed, mouse cancer cells undergoing accidental necrosis in response to freeze–thawing or boiling are unable to activate DCs in vitro50 and fail to elicit protective immunity upon inoculation of syngeneic mice31,51. The molecular mechanisms that account for the immunogenicity of neoplastic cells exposed to irradiation, hypericin-based PDT and high hydrostatic pressure have been characterized less extensively than the pathways underlying chemotherapy-driven ICD. Nevertheless, it has been demonstrated that some modules of the machinery promoting ICD are universally required for the immunogenicity of dying cancer cells (at least in mice), whereas others are 'private' and contribute to ICD in a limited number of settings (see below) (Fig. 1).

The immunogenic potential of irradiation was initially recognized in the context of vaccination mediated by cancer cell lysates52. DCs loaded with irradiated malignant cells were found to elicit robust immune responses both in mice53 and in patients with cancer54. Soon thereafter, vaccination experiments unequivocally demonstrated that γ-irradiation as well as UVC light could kill mouse cancer cells in a manner that — in immunocompetent hosts — elicits an adaptive immune response associated with the establishment of protective immunological memory48,55. More recently, α-irradiation with 213Bi particles has been added to the list of bona fide ICD inducers, as per vaccination experiments with mouse MC38 colorectal carcinoma cells and syngeneic hosts56. The ability of irradiation to trigger ICD relies on UPR-dependent exposure of CALR48,55,57 and ATP secretion driven by autophagy57,58, in addition to type I IFN signalling59,60. Moreover, ICD induction by radiation therapy is accompanied by the exposure of HSP70 on the surface of malignant cells61, TLR3 signalling62, HMGB1 release57,61,63 and IL-1β release upon inflammasome activation64; however, the requirement of these processes for the engagement of adaptive immunity has not yet been tested formally. Importantly, ICD driven by irradiation and the consequent activation of a tumour-specific CD8+ T cell-dependent immune response are responsible for the abscopal effect65, especially when radiation therapy is combined with a checkpoint blocker like ipilimumab66,67,68. In this context, it should be noted that dose and schedule dramatically affect the immunogenicity of irradiation-driven cell death. Thus, whereas single-dose radiation therapy seems unable to induce innate immunity against dying cancer cells, fractionated radiotherapy results in optimal immunostimulatory effects (at least in mice)68. This observation lends further support to the notion that immune responses to cell death critically rely on the timely release of endogenous adjuvant (in the context of increased antigenicity). It is tempting to speculate that radiation therapy may also boost the antigenicity of cancer cells, at least to some extent, but this conjecture remains to be experimentally addressed (Fig. 1).

Similar to chemotherapy- and irradiation-driven ICD, the immunogenic demise of human and mouse cells exposed to hypericin-based PDT or high hydrostatic pressure is accompanied by the exposure of ER chaperones including CALR, HSP70 and HSP90 on the plasma membrane46,47,69, ATP secretion47,69 and HMGB1 release47,70. Accordingly, the exposure of DCs to human cancer cells succumbing to hypericin-based PDT or high hydrostatic pressure induces not only the upregulation of various DC activation markers, including CD80, CD83, CD86 and MHC class II molecules but also the secretion of several pro-inflammatory cytokines, such as IL-1β, IL-12 and TNF, resulting in the priming of tumour-specific CD8+ T cells47,69. Moreover, mouse cancer cells responding to hypericin-based PDT or high hydrostatic pressure efficiently protect syngeneic, immunocompetent mice against a subsequent challenge with living cells of the same type69,71 (Fig. 1).

The molecular mechanisms that underlie danger signalling in cancer cells that are exposed to chemotherapy, irradiation, hypericin-based PDT or high hydrostatic pressure are more heterogeneous than it might appear at first glance. Indeed, whereas CALR exposure on malignant cells succumbing to chemotherapy-driven ICD is accompanied by (and mechanistically depends on) the phosphorylation of eIF2A, the activation of caspase 8 and the co-exposure of PDIA3, neoplastic cells responding to hypericin-based PDT externalize CALR independently of eIF2A phosphorylation, caspase 8 activation and PDIA3 (Refs 34,46,72,73). Similarly, a proficient autophagic response is an absolute requirement for ATP secretion triggered by anthracyclines, oxaliplatin and irradiation36,58,73. Conversely, autophagy not only seems to be irrelevant for the secretion of ATP elicited by hypericin-based PDT in human cancer cells and mouse embryonic fibroblasts but also appears to inhibit the immunogenicity of this intervention owing to its capacity to limit oxidative damage, the UPR and consequent CALR exposure (which in this setting is initiated by reactive oxygen species)74. These observations indicate that not all danger signals may be universally required for the engagement of adaptive immunity in all scenarios. Moreover, they suggest that (at least in some cases) some intracellular mechanisms may functionally compensate for each other to ensure danger signalling. To what extent (if any) hypericin-based PDT and high hydrostatic pressures alter the antigenicity of neoplastic cells remains to be determined (Fig. 1).

Necroptotic ICD. Necroptosis is a form of regulated cell death that is precipitated by the receptor-interacting serine/threonine kinase 3 (RIPK3)-catalysed phosphorylation of the pseudokinase mixed lineage kinase domain-like (MLKL), which results in the rapid formation of MLKL oligomers that irreversibly permeabilize the plasma membrane75,76. In some (but not all) variants of necroptosis, RIPK3 is activated by a signalling cascade that emanates from TNF receptor superfamily member 1A (TNFRSF1A; also known as TNFR1) and depends on the RIPK3 homologue RIPK1, implying that it can be delayed by the chemical RIPK1 inhibitor necrostatin 1 (Ref. 13). Although necroptosis was soon recognized to be a highly pro-inflammatory form of cell death77, its ability to engage the adaptive arm of the immune system and to evoke an antigen-specific immune response has not been investigated until recently. Thus, mouse lung carcinoma TC-1 and EL4 lymphoma cells (which naturally express high levels of RIPK3) exposed to necroptosis-inducing conditions (that is, TNF plus Z-VAD-fmk and a SMAC mimetic) die as they expose CALR on the plasma membrane, secrete ATP and release HMGB1 — three manifestations of ICD that, in this setting, are abrogated by knocking out Ripk3 or Mlkl78. Accordingly, necroptotic TC-1 cells, but not TC-1 cells undergoing accidental necrosis upon freeze–thawing, efficiently vaccinate syngeneic C57BL/6 mice against a subsequent challenge with living cells of the same type78. TC-1 cells exposed to mitoxantrone also manifest markers of necroptosis, such as RIPK3 aggregation and the phosphorylation of MLKL, and die according to a kinetic that can be altered by the absence of Ripk3 or Mlkl78. Most importantly, necroptosis-deficient Ripk3−/− or Mlkl−/− TC-1 cells succumbing to mitoxantrone fail to elicit protective immunity in C57BL/6 mice in conditions in which wild-type cells efficiently do so — a defect that is linked to reduced ATP secretion and limited HMGB1 release78. Along similar lines, Ripk3−/− as well as Mlkl−/− TC-1 cells growing in C57BL/6 mice are less sensitive to mitoxantrone-based chemotherapy than their wild-type counterparts as they are poorly infiltrated by CD11c+CD86+ APCs and CD8+ T cells78. The local administration of an inhibitor of extracellular ATPases plus a synthetic TLR4 ligand restores the infiltration of necroptosis-deficient tumours by APCs and CD8+ T cells, hence re-establishing normal sensitivity to mitoxantrone-based chemotherapy78.

Comparable results have been obtained using mouse CT26 colorectal carcinoma cells (which do not naturally express high levels of RIPK3) engineered to express a variant of RIPK3 that can be activated by chemical-driven dimerization51. CT26 cells artificially driven into necroptosis secrete ATP, release HMGB1 and are rapidly phagocytosed by bone-derived mononuclear cells, which results in the upregulation of several activation markers including CD80, CD86 and MHC class II molecules51. Moreover, necroptotic CT26 cells (but not CT26 cells undergoing accidental necrosis in response to freeze–thawing) efficiently triggered an adaptive immune response driven by IFNγ-secreting CD8+ T cells and accompanied by robust immunological memory in syngeneic BALB/c mice51. No signs of UPR were detected in CT26 cells undergoing necroptosis upon RIPK3 dimerization51, which supports the notion that the adjuvanticity of ICD may be ensured by mechanistically distinct but functionally complementary pathways.

Further corroborating the hypothesis that necroptosis may constitute a functionally distinct variant of ICD, at least in some settings, the administration of Z-VAD-fmk (which favours necroptotic cell death by inhibiting caspases)79 has been found to considerably ameliorate the therapeutic effect of irradiation-based multimodal therapy (fractionated irradiation plus dacarbazine plus hyperthermia) in immunocompetent mice bearing syngeneic B16 melanomas, an effect that depended on TLR signalling and was abrogated by the co-administration of the ATP-degrading enzyme apyrase63 (see below). Z-VAD-fmk augmented the percentage of B16 cells dying with a necrotic morphology in response to multimodal therapy, resulting in increased HMGB1 release. This was associated with a superior adjuvanticity in vitro, as monitored by the expression of CD86, MHC class II molecules and TNF by APCs exposed to supernatants from dying B16 cells. In vivo, the presence of Z-VAD-fmk promoted tumour and lymph node infiltration by DCs and CD8+ T cells secreting IFNγ63. Together, these observations suggest that different variants of ICD may synergize in the induction of potent adaptive immune responses, at least in the presence of adequate antigenicity (Fig. 1).

Danger signalling in ICD

Once emitted by dying cells, DAMPs orchestrate antigen-specific immune responses by acting on both innate and adaptive components of the immune system11,80. Importantly, not all DAMPs are immunostimulatory, and immunosuppressive DAMPs like adenosine (a byproduct of extracellular ATP degradation) and prostaglandin E2 (an eicosanoid) seem to have a key function in the maintenance of tolerance in the course of physiological instances of cell death81,82. Irrespective of their importance for immunological homeostasis, immunosuppressive DAMPs are not discussed in further detail here.

CALR and ER chaperones. CALR, PDIA3, HSP70, HSP90 and possibly other ER chaperones that are exposed on the membrane of cells undergoing ICD act as 'eat-me' signals, hence promoting the uptake of cell corpses and debris by APCs83. The main docking site for CALR on the surface of cancer cells is provided by LDL receptor related protein 1 (LRP1; also known as CD91)69. In line with this notion, the immunogenicity of cancer cell death driven by hypericin-based PDT can be reduced by shRNA-dependent downregulation of LRP1 (Ref. 69). LRP1 is also the main ER chaperone receptor expressed by human and mouse myeloid cells84,85, and mouse macrophages lacking LRP1 exhibit limited phagocytic potential84,86. Moreover, the activation of DCs by human BC3 primary effusion lymphoma cells succumbing to bortezomib-driven ICD or human T24 bladder cancer cells treated with the putative ICD inducer capsaicin can be inhibited by a LRP1-targeting monoclonal antibody87 or by LRP1 silencing through RNA interference88, respectively. However, formal evidence indicating that LRP1 expression in the myeloid compartment is absolutely required for the initiation of adaptive immune responses by ICD is still missing. According to some investigators, ER chaperones are 'sticky' proteins and do not need receptors to bind membranes. Supporting this possibility, CALR-deficient cancer cells can be efficiently coated by CALR upon short incubation with the recombinant protein, which restores their uptake by phagocytes and hence reverses their intrinsically limited capacity to undergo productive ICD33,89. Robust expression of the CALR antagonist CD47 (which operates as a 'don't eat-me signal') has been linked to dismal prognosis in individuals affected by various cancers, including acute myeloid leukaemia, oesophageal carcinoma and ovarian cancer90,91,92. Conversely, high CALR levels in neoplastic cells (correlating with eIF2A phosphorylation) have recently been associated with improved clinical outcome in two independent cohorts of non-small-cell lung carcinoma patients93. Moreover, the monocytes of eight individuals with advanced melanoma progressing in an unusually slow manner were found to express elevated amounts of LRP1 compared with monocytes from eight patients who progressed normally94, which suggests that danger signalling through the CALR–LRP1 axis may influence the outcome of patients with cancer in settings in which immunosurveillance (be it natural or therapy-elicited) is involved (Fig. 3; Table 2).

Figure 3: Subversion of danger signalling.
figure 3

As they escape immunosurveillance, infectious pathogens and malignant cells acquire the ability to suppress danger signalling or the consequences thereof. In addition, clinical data obtained in patients with cancer indicate that multiple defects in the perception of damage-associated molecular patterns (DAMPs) by the host are associated with worsened disease outcome. ANXA1, annexin A1; CALR, calreticulin; CXCL10, CXC-chemokine ligand 10; CXCR3, CXC-chemokine receptor 3; DC, dendritic cell; FPR1, formyl peptide receptor 1; HMGB1, high-mobility group box 1; IFN, interferon; IFNAR1; interferon α/β-receptor subunit 1; IL-1β, interleukin-1β; IRF7, interferon regulatory factor 7; LRP1, LDL receptor related protein 1; PDT, photodynamic therapy; P2RX7, purinergic receptor P2X7; P2RY2, purinergic receptor P2Y2; SNP, single nucleotide polymorphism; STAT1, signal transducer and activator of transcription 1; TLR, Toll-like receptor; UPR, unfolded protein response.

PowerPoint slide

Table 2 Evasion of danger signalling by pathogens and cancer cells*

Extracellular ATP. Extracellular ATP mediates robust chemotactic and adjuvant-like effects by interacting with purinergic receptor P2Y2 (P2RY2) and purinergic receptor P2X7 (P2RX7), respectively, on APCs and their precursors40,95. Thus, the immunogenicity of cell death is abrogated not only when ATP fails to accumulate in the microenvironment of dying cells36,40,96 (see above), but also when P2ry2 or P2rx7 are absent from the myeloid compartment of the host40,95. In mouse DCs, purinergic signalling via P2RX7 promotes inflammasome activation coupled to the release of IL-1β95, which is required for the initial lymphoid response to ICD driven by IL-17-producing γδ T cells97. Accordingly, Nlrp3−/− mice (which lack an essential component of the inflammasome)98, Il17a−/− or Il17ra−/− mice (both of which are defective in the IL-17 system), as well as mice receiving an IL-1β-neutralizing antibody are unable to mount adaptive immune responses against syngeneic cancer cells succumbing to chemotherapy-driven ICD95. A loss-of-function polymorphism in P2RX7 has also been shown to negatively affect disease outcome in a cohort of patients with breast carcinoma treated with anthracycline-based chemotherapy95, suggesting that danger signalling has translational relevance in the context of cancer therapy (Fig. 3; Table 2).

HMGB1. The molecular mechanisms that underlie the release of HMGB1 from cells undergoing ICD remain to be elucidated. Regardless of this unknown, extracellular HMGB1 mediates robust adjuvant-like effects by binding to various distinct PRRs, including TLR2, TLR4 and advanced glycosylation end product-specific receptor (AGER; also known as RAGE)99. However, whereas Tlr2−/− and Ager−/− mice can be normally vaccinated by syngeneic cancer cells that undergo ICD in response to chemotherapy, the same does not hold true for Tlr4−/− mice or Myd88−/− mice (which lack myeloid differentiation primary-response protein 88, a key adaptor for TLR signalling)38. Thus, danger signalling through the HMGB1–TLR4–MYD88 axis appears to be required for adaptive immune responses against mouse cancer cells succumbing to ICD. Further supporting the translational relevance of ICD, loss-of-function polymorphisms in TLR4 have been associated with unfavourable disease outcome in cohorts of patients with breast carcinoma treated with anthracyclines38, patients with head and neck squamous cell carcinoma undergoing systemic chemotherapy100, and subjects with melanoma receiving a DC-based vaccine101 or other treatment modalities102. Moreover, the loss of HMGB1 from malignant cells has been noted to negatively affect prognosis in patients with breast cancer treated with anthracycline-based adjuvant chemotherapy103,104. Of note, HMGB1 has also been proposed to mediate immunosuppressive functions on binding to hepatitis A virus cellular receptor 2 (HAVCR2; also known as TIM3), at least in mice105. The significance of this signal transduction cascade for cancer therapy, however, remains to be elucidated (Fig. 3; Table 2).

Type I IFNs. Type I IFNs are secreted by virtually all cells upon viral or bacterial infection, reflecting the detection of multiple MAMPs by PRRs23. Upon binding to IFNAR1–IFNAR2 heterodimers, type I IFNs not only increase the resistance of neighbouring cells to infection, but also stimulate the activation of macrophages, DCs and natural killer (NK) cells, hence alerting them of pathogens23. Consistent with this notion, Ifnar1−/− mice are much more susceptible than their wild-type counterparts to several viruses, including respiratory syncytial virus, Zika virus and mouse hepatitis virus, as well as to intracellular bacteria such as S. enterica106,107,108,109. Type I IFNs are also produced by mouse neoplastic cells succumbing to chemotherapy-driven ICD, most likely reflecting TLR3 activation by cancer cell-derived RNA37. In this setting, however, the adjuvant effects of type I IFNs mainly stem from the activation of a cancer cell-autonomous signal transduction cascade resulting in CXCL10 secretion37. Thus, the efficacy of anthracycline-based chemotherapy in mice is reduced in the presence of an IFNAR1-neutralizing antibody, or when neoplastic cells (but not the host) lack Ifnar1, Ifnar2 or Tlr3. In all these models, therapeutic efficacy can be partially restored by the co-administration of recombinant CXCL10 (Ref. 37). Accordingly, neutralizing the main CXCL10 receptor, namely, CXC-chemokine receptor 3 (CXCR3), with a monoclonal antibody also impairs the efficacy of anthracycline-based chemotherapy in vivo37. Along similar lines, Ifnar1−/− mouse cancer cells killed by doxorubicin cannot vaccinate syngeneic hosts in conditions in which wild-type cells efficiently do so. Moreover, the rate of tumour-free mice in this setting is not affected by the absence of Ifnar1 in the host but is significantly reduced if vaccination is performed with wild-type mouse cells exposed to doxorubicin in the presence of monoclonal antibodies neutralizing type I IFNs or IFNAR1 (Ref. 37). Although CXCL10 is known to mediate chemotactic effects on T cells, it remains to be determined whether this is the mechanism that underlies its adjuvant activity in the context of chemotherapy-driven ICD. Irrespective of this, the expression of a type I IFN-related metagene has been shown to predict the pathological complete response of patients with breast carcinoma to anthracycline-based chemotherapy37. Reduced levels of interferon regulatory factor 7 (IRF7), one of the transducers involved in type I IFN signalling, have been linked to decreased metastasis-free survival in a large cohort of over 800 women with breast carcinoma110. Similarly, approximately one third of 161 human breast cancers displayed undetectable or very low levels of signal transducer and activator of transcription 1 (STAT1; another component of the type I IFN signalling pathway)111. Moreover, high levels of TLR3 in neoplastic lesions have been associated with improved disease outcome in breast carcinoma patients treated with a TLR3 agonist plus radiation therapy112. These observations lend further support to the idea that danger signalling is clinically relevant in the context of anticancer chemo- and radiotherapy (Fig. 3; Table 2).

ANXA1. ANXA1 belongs to a superfamily of proteins that bind acidic phospholipids in a Ca2+-dependent manner, and it is expressed at moderate-to-high levels by both myeloid and lymphoid cells as well as by many epithelial cell types113. Initially, ANXA1 was characterized for its key role in the resolution of inflammatory responses, which mainly reflects the ability of secreted or surface-exposed ANXA1 to elicit autocrine, paracrine or juxtacrine signalling via formyl peptide receptor 2 (FPR2)113. More recently, our group demonstrated that the therapeutic effect of anthracyclines or oxaliplatin in mice is lost when cancer cells lack Anxa1 and when the host immune system lacks Fpr1 (which encodes another ANXA1 receptor)39. Mouse Fpr1−/− APCs infiltrate malignant lesions treated with anthracyclines or oxaliplatin as well as their wild-type counterparts do, but they display a selective defect in the final approach to dying cancer cells, and hence cannot initiate an adaptive immune response39. A loss-of-function polymorphism in FPR1 has been associated with decreased time-to-metastasis and overall survival in a cohort of patients with breast carcinoma receiving anthracycline-based chemotherapy, as well as in a cohort of patients with colorectal carcinoma treated with oxaliplatin39. Importantly, the impact of FPR1 status could only be observed among patients with wild-type TLR3 and TLR4 (Ref. 39), which suggests that these three PRRs operate in the same pathway linking chemotherapy-driven ICD to the elicitation of a clinically relevant, tumour-targeting immune response (Fig. 3; Table 2).

Subversion of ICD

The concept of ICD has been developed in tumour models, in which adjuvanticity is solely dictated by DAMPs. This considerably simplified the molecular characterization of the phenomenon. However, the primordial function of the immune system is not anticancer immunosurveillance but rather the control of invading pathogens, suggesting that ICD may play a major role in the context of microbial infection. In favour of this hypothesis, it appears that pathogenic viruses and bacteria have developed multiple strategies to subvert the release or detection of DAMPs, presumably with the aim of escaping immune control.

Thus, several microorganisms including viruses and bacteria express functional orthologues of anti- or pro-apoptotic members of the BCL-2 protein family15, autophagy inhibitors16,114, proteins that favour the dephosphorylation of eIF2A115,116, inflammasome-inhibiting factors117, caspase blockers15, proteins that prevent the assembly of the necrosome118, and/or inhibitors of type I IFN signalling119, which often (if not always) are required for overt pathogenicity (Table 2). The existence of such a multipronged armamentarium strongly suggests that the selective pressure on limiting the adjuvanticity of infection-driven cell death has been high during (at least some stages of) the host–pathogen co-evolution. Controlling immunogenicity by limiting adjuvanticity (which is mainly determined by the host) seems indeed more straightforward than doing so by limiting antigenicity (which is mainly determined by the genetics of the pathogen).

Similar considerations can be made for malignant cells. Indeed, according to the immunosurveillance model, overtly malignant lesions only develop once neoplastic cells fully escape immune recognition and elimination120, and such a potentially lethal progression is intimately linked to the inhibition of various processes involved in DAMP emission and sensing, and/or the immunological consequences thereof121,122,123 (Table 2). Like pathogens, advanced tumours often exhibit an elevated antigenicity9, but they efficiently control immunogenicity by various mechanisms operating on (or downstream of) adjuvanticity121,122,123 (Box 2), which considerably impairs the efficiency of multiple chemo-, radio- and immunotherapeutic regimens124. Thus, cancer cells obtain an advantage not only from defective DAMP emission or sensing (see above), but also from the suppression of upstream adaptive mechanisms, including the UPR and autophagy, and from the subversion of the signal transduction cascades that precipitate cell death125 (Table 2).

Together, these observations reinforce the notion that restoring the immunogenicity of cancer cells constitutes a prominent, clinically-relevant goal, and that the most promising target in this sense is adjuvanticity (or processes downstream thereof).

Conclusions and perspectives

ICD stands out as a major initiator of adaptive immunity in the context of infectious and malignant diseases, both of which involve increased antigenicity and adjuvanticity. Adjuvanticity depends on the release or exposure of specific MAMPs or DAMPs, and both infectious pathogens and neoplastic cells have developed an arsenal of strategies to subvert danger signalling and hence avoid immune detection. Of note, danger signalling largely reflects the communication of a state of cellular stress to the organism. Thus, the stress-responsive processes that underlie danger signalling (including autophagy) operate to maintain whole-body homeostasis across the plasma membrane. Owing to its prominent cytoprotective activity, autophagy limits the intrinsic susceptibility of neoplastic cells to the cytotoxic effects of anticancer agents25. Accordingly, human and mouse autophagic-deficient malignant cells growing in vitro or in immunodeficient hosts are more sensitive than their wild-type counterparts to chemotherapy and irradiation25,58. However, cancers evolving in immunocompetent syngeneic mice are less sensitive to treatment if their autophagic proficiency is compromised by genetic interventions58. Similar considerations can be made for other mechanisms of cellular adaptation, such as the UPR22. This suggests not only that the functions of these processes (which are found in all eukaryotes) seem to have expanded from the preservation of cellular fitness to the maintenance of homeostasis at the organismal level, but also that protection of whole-body homeostasis generally dominates over the maintenance of cellular fitness.

Finally, it is tempting to speculate that (at least some) autoimmune disorders may originate from situations in which an unwarranted wave of cell death is mistakenly perceived as immunogenic. In this setting, robust DAMP signalling might drive an adaptive immune response even against innocuous antigens, possibly as a consequence of antigen spreading, and hence overcome tolerance to self. An intriguing finding in support of this notion is that various disorders with an autoimmune component, including rheumatoid arthritis126, coeliac disease127 and inflammatory bowel disease128, are associated with increased circulating levels of CALR or CALR-specific antibodies. Thus, it will be interesting to determine whether artificial exacerbation of DAMP signalling in the context of cell death (but in the absence of accrued antigenicity) can drive autoimmune disorders, and — if so — whether the antigenic target of such a pathological response is under defective tolerance.

In summary, DAMP signalling occupies a hierarchically superior position in the preservation of whole-body immunological and inflammatory homeostasis, hence constituting a promising target for the development of therapeutic agents with oncological and non-oncological applications.