Molecular and Translational Classifications of DAMPs in Immunogenic Cell Death

1Augmenting the immunogenicity of cancer cells to improve the efficacy of cancer therapy is a paradigm that has gained significant momentum over the past 5 years. Researchers have realized that besides therapeutically exploiting innate or adaptive immune cells directly (e.g., through dendritic cell (DC)-based vaccines or adoptive T-cell transfer) and/or improving the effector functions of T cells (through checkpoint-blocking therapies), cancer cells also need to be made immunogenic. This has diverted attention toward studying the interface between stressed or dying cancer cells and the immune system, in the hope of efficiently exploiting it for therapeutic purposes.

2Early indications regarding immune system-driven tumor control emerged in the eighteenth century, when feverish infections in cancer patients were circumstantially associated with tumor remission. The first evidence that immunotherapy can be applied to achieve tumor regression emerged from the work of William Coley, who in the 1890s achieved tumor regression in some sarcoma/lymphoma patients upon the intra-tumoral injection of streptococcal cultures (provided by Robert Koch). In the following 43 years, Coley injected nearly 900 (mostly sarcoma) patients with his bacterial preparation (achieving a cure rate >10%), which later became known as “Coley’s toxin”. However, the Coley’s toxin came under intense scrutiny owing to an elevated toxicity and some difficulties in reproducing remission rates. Eventually, the first experimental evidence that virus-unrelated tumors can indeed be recognized by the host immune system emerged in the 1940s, and by the 1960s, coupled with the discovery of T cells, it was proposed that the human immune system may also react against tumors. The ability of anticancer therapies to enhance the immunogenic potential of malignant cells gained some appreciation by the 1970s. It was recognized that if specific treatments are applied (e.g., radiotherapy, the bacillus Calmette–Guerin, or some chemotherapeutics), the immunogenicity of malignant cells increases enough to induce durable anti-tumor immunity. By the 1980s, researchers started to report more specific observations regarding the therapeutic impact of cancer cell immunogenicity, e.g., the ability of curative hyperthermia to cause the (heat-shock based) generation of circumstantial anti-tumor immunity, the fact that the immunogenicity of cancer cells influences patient prognosis after radiotherapy, and the increase in tumor immunogenicity due to hydrostatic pressure. However, these early studies (especially those published before the 1980s) had several issues linked to a lack in consensus. For instance, due to early controversies on the existence of tumor-associated antigens (TAAs), the target of tumor-specific immune responses was unclear, and the mechanism of action of some therapies came under scrutiny. Moreover, such therapies could operate by directly modulating immune effector cells rather than improving the immunogenic potential of tumors. In particular, the death of cancer cells exposed to therapy was never suspected to drive anti-tumor immunity, since it was considered to be a relatively “silent” process in terms of immunogenicity. Moreover, the classical “self/non-self” theory was unable to explain the possibility that dying cancer cells could elicit an immune response.

3By the early 1990s, the molecular characterization of mice and human TAAs clarified the entities targeted by anti-tumor immune responses. Similarly, the so-called “danger theory” started to emerge, challenging the classical model of “self/non-self” immune recognition, especially in a diseased or damaged tissue. This model proposed that the immune recognition is not restricted to “non-self” entities, but rather discriminates between “dangerous” and “safe” entities, irrespective of source. Indeed, “dangerous” entities include pathogens as well as injured, infected, diseased and necrotic tissues, or cells undergoing non-physiological cell death which emit danger signals (or alarmins) with pro-inflammatory activity. These danger signals are now collectively referred to as “damage-associated molecular patterns” (DAMPs). DAMPs are endogenous molecules that are concealed intracellularly in normal conditions, but are exposed or released upon stress, injury, cell death, thereby becoming able to bind cognate receptors on immune cells. Table 1 summarizes the most prominent DAMPs characterized to date and their mode of emission, the cell death pathway they are associated with, and their known cognate receptors. It is important to consider that not all DAMPs may act as immunogenic danger signals. Several DAMPs exist that are crucial for the maintenance of tissue homeostasis, and the avoidance of auto-immune responses, as they exert immunosuppressive effects, including phosphatidylserine (PS), annexin A1 (ANXA1), death domain 1α (DD1α), B-cell CLL/lymphoma 2 (BCL2) and some extracellular matrix-derived molecules (Table 1). Accordingly, the blockade of these anti-inflammatory DAMPs accentuates the immunogenic potential of dying cells, or renders immunogenic otherwise tolerogenic forms of cell death. Moreover, some danger signals are not always involved in the immunogenicity of cell death, but act as “bystanders.” This is the case for heat shock protein 90 kDa alpha (cytosolic), class A member 1 (HSP90AA1, best known as HSP90) exposed on the cell surface after melphalan treatment. Last (but not least), several DAMPs may be subjected to post-translational modifications (e.g., oxidation, reduction, citrullination) that may potentially neutralize, increase, or change their immunogenic properties – a process that is still incompletely understood.

4Despite these advances, the overall role of regulated cell death (RCD) in augmenting cancer immunogenicity remained obscure. Initial observations involving the immunogenicity of cell death in the efficacy of cancer therapy were published between 1998 and 2004, when it was proposed that the non-apoptotic demise of malignant cells (within the context of the so-called “immunogenic death”) could be associated with the emission of the danger signal heat shock 70 kDa protein 1A (HSPA1A, best known as HSP70) (Table 1), enhancing the immunogenic potential of dying cancer cells in vivo. The dogmatic view that only necrotic or non-apoptotic (as postulated by the “immunogenic death” concept) cancer cells are characterized by an elevated immunogenic potential started to be questioned by a series of studies published between 2005 and 2007. These publications outlined that cancer cells undergoing apoptosis in response to specific anticancer therapies are immunogenic [a subroutine termed immunogenic cell death (ICD)], as long as they emit precise DAMPs in a spatiotemporally defined fashion. Cells succumbing to ICD are sufficient for the elicitation of durable anti-tumor immune responses. ICD is indeed paralleled by the redirection and emission of DAMPs, owing to the stimulation of distinct danger signaling pathways occurring in synchrony with cell death signaling. Table 2 summarizes the main signaling pathways that play a role in the trafficking and emission of DAMPs. ICD-associated DAMPs and other immunostimulatory factors released by cells destined to undergo ICD favor the establishment of a productive interface between dying cancer cells and innate immune cells (like DCs or macrophages), thereby leading to the initiation of a therapeutically relevant adaptive immune response (Figure 1). In some contexts, DAMPs may regulate the function of specific innate immune cell subsets, e.g., following anthracycline treatment, extracellular adenosine triphosphate (ATP) assists in recruitment and differentiation of CD11c+Cd11b+Ly6Chigh cells into CD11c+CD86+MHCII+ DCs; similarly, necrosis associated F-actin exposure activates an immune response by directing the dead cell debris to specifically CD8α+ DCs. Indeed, DCs and other antigen-presenting cells exposed to cancer cells succumbing to ICD can then prime CD4+ T cells (and polarize them into TH1, TH17, or TH1/TH17-like phenotype), CD8+ cytotoxic T lymphocytes (CTLs) and γδ T lymphocytes against one or several TAAs (Figure 1). Of note, residual cancer cells that survive ICD inducers can also show some enduring immunogenic characteristics that make them susceptible to immunological control by CTLs.

5Over the past few years, a number of single-agent ICD inducers have been discovered, encompassing conventional chemotherapeutics, targeted anticancer agents and various other ­biological and physicochemical therapies. Table 3 summarizes single-agent ICD inducers characterized so far, as per consensus guidelines, and the spectra of DAMPs and other immunostimulatory signals associated with them. For combinatorial therapeutic strategies capable of achieving ICD, readers may want to refer to other recent publications. It is clear that a general structure–function relationship capable of clustering all existing ICD inducers and predicting new ones does not exist, an issue that makes discovering new ICD-inducing therapies based on cheminformatic analyses challenging, if not impossible. A peculiar characteristic of most, if not all, ICD inducers is their ability to induce reactive oxygen species (ROS)-based/associated endoplasmic reticulum (ER) stress, as first delineated for anthracyclines. This peculiarity was exploited for the targeted discovery of hypericin-based photodynamic therapy (Hyp-PDT) – a therapeutic modality that can trigger ICD through the induction of ROS that target the ER. Along with an ever more precise characterization of the links between ROS, ER stress, and ICD induction, it became clear that the more “focused” ER stress is, the higher the probability of inducing ICD. These observations paved way for a classification system based on how ICD inducers engage ER stress for cell death and danger signaling. Based on this classification, Type I ICD inducers are defined as anticancer agents that act on non-ER proteins for the induction of cell death, but promote collateral ER stress for danger signaling, thereby operating on multiple targets, while Type II ICD inducers are anticancer agents that target the ER for both cell death induction and danger signaling. Table 4 summarizes the classification of current ICD inducers into Type I and Type II, and their cell death/danger signaling targets. Such a classification suggest that while Type I ICD inducers can be discovered through various approaches (e.g., DAMP-based drug screening platforms), putative Type II ICD inducers can be characterized rapidly on the basis of their ability to selectively or predominantly target the ER. Recent findings comforted the purpose and usefulness of this classification system, as two novel Type II ICD inducers [i.e., PtII N-heterocyclic carbene complex and Newcastle disease virotherapy (NDV)] were identified based on the notion that they induce predominant ROS-based ER stress. Nevertheless, as more ICD inducers and features are discovered, this classification system is expected to evolve or be substituted by a more refined one.

6Since its discovery, a plethora of molecular and immunological components responsible for ICD have been discovered (Figure 1). Table 5 summarizes the molecular and immunological determinants of ICD characterized so far, as well as the models of ICD in which they operate (in a positive, negative or dispensable manner). Anthracyclines and oxaliplatin are the most common ICD inducers employed in experimental settings, followed by Hyp-PDT. According to current understanding, cancer cell-associated determinants of ICD can be subdivided into those that are common to all ICD inducers (i.e., “core” signaling components), and those that operate in an ICD inducer-dependent manner (i.e., “private” signaling components). Thus, eukaryotic translation initiation factor 2-alpha kinase 3 (EIF2AK3, best known as PERK) and the ER-to-Golgi secretory machinery are considered “core” signaling components on the cancer cell side. Similarly, from the immune system side, a general role for (IFNγ-producing) CD4+ and CD8+ T cells has been confirmed for most, if not all, ICD inducers (Table 5). Interestingly, some components that are required for ICD induction by some agents (like autophagy for anthracyclines and oxaliplatin) might be either dispensable for ICD induction by other agents, e.g., autophagy for NDV and phosphorylation of eukaryotic translation initiation factor 2α (eIF2α), caspase-8 (CASP8) activation or cytosolic Ca2+ levels for Hyp-PDT; or even negatively regulate ICD in some settings, e.g., autophagy in case of Hyp-PDT (Table 5). Thus, it will be important to expand our molecular knowledge of ICD to as many experimental settings as possible.

7The relevance of ICD has been verified in a number of rodent models, with a variety of chemical and physicochemical ICD inducers. Table 6 summarizes the most prominent mouse or rat models used so far for the characterization and study of ICD. For the moment, ICD has been mostly investigated in heterotopic syngeneic subcutaneous models. Within such models, inter-species differences (mouse versus rats), inter-strain differences (among BALB/c, C57BL/6, C3H and KMF mice), and inter-cell line differences, as well as differences in therapeutic setups (prophylactic versus curative) have been amply accounted for (Table 6). Nevertheless, there is predominance in the use of cancer cells derived from carcinogen-induced tumors and transplanted subcutaneously (Table 6). In very few cases, ICD has been characterized in either orthotopic (for NDV) or spontaneous (for anthracyclines) tumor murine models (Table 6). This has been questioned as a prominent Achilles’ heel of ICD research. While this criticism is valid, it has to be recognized that no rodent model is perfect at all immunological levels.

8As a recent systematic review summarized, heterotopic murine models suffer from a number of caveats, including the inability to recapitulate the early interaction between transformed cells and the immune system and the incompatibility between the cancer type and the site-of-transplantation. Orthotopic murine models are useful as they overcome the cancer cell-tissue type incompatibility issue. While genetically engineered tumor murine models (GEMMs) overcome most of the issues mentioned above, they come with their own set of shortcomings, including a limited genetic mosaicism, a low tumor heterogeneity, a lack of well-defined immunogenic TAAs, the presence of unintended “passenger” genetic modifications, and a reduced mutational spectrum. Many of these parameters are critical for responses to immunotherapy/ICD. For instance, the lack of well-defined immunogenic TAAs was the reason why preliminary results obtained in spontaneously developing murine tumors disputed the very existence of TAAs. Similarly, a high mutational spectrum (which produces considerable amounts of neo-antigens) has been found to be mandatory for the clinical efficacy of checkpoint blockers. Last (but not least), laboratory rodent models in general are associated with some critical issues, including the fact that a high level of inbreeding (which produces a number of shortcomings e.g., homozygous recessive defects) reduces the general immunological fitness, responsiveness and diversity in these models. Moreover, numerous immunological differences between mouse and humans tend to affect the translational relevance of the findings obtained. Also, the time frames of tumor growth rates between rodent models and humans are relatively divergent. This further complicates clinical translation of immunotherapeutic paradigms since the level of immunosurveillance and immunoediting experienced by human tumors can be much higher than any rodent tumor model.

9In summary, it would be ideal to test ICD across as many different rodent models as possible, in order to determine the features that can be exploited for therapeutic purposes in humans. Moreover, if ICD fails in a specific experimental model, active effort should be made to characterize the mechanisms behind such failure, since resistance phenotypes can have profound clinical implications. This emerges from various studies summarized in Table 7. Indeed, several ICD resistance mechanisms exist operating at both the cancer cell and the immune system level, which have been characterized in different experimental models. Several of these resistance mechanisms have also been identified in cancer patients, thereby justifying further studies along these lines Table 7.

10A considerable amounts of clinical findings support the relevance of ICD or ICD-related signatures in (at least subsets of) cancer patients. As summarized in Table 8, various ICD-linked (specific) parameters have been associated with the prognosis of cancer patients treated with clinically relevant ICD inducers (like anthracyclines, oxaliplatin, paclitaxel, or radiotherapy). Moreover, it is becoming clear that ICD-related or ICD-derived (immunological) genetic signatures (e.g., a MX1-centered metagene, a CXCR3-PRF1-CASP1-centered metagene, an ASAH1-centered metagene) can be positively associated with good prognosis in patients affected by various neoplasms, including breast, lung, and ovarian malignancies. These observations indicate that ICD or ICD-relevant parameters may have prognostic or predictive relevance in at least a subset of cancer patients. It will be important to characterize new and more specific ICD-associated parameters linked to patient prognosis as well as biomarkers that may predict improved disease outcome in cancer patient treated with ICD inducers. Of note, considering the current clinical experience with immunotherapies, the patients with an increased likelihood to benefit from ICD inducers are probably those that display pre-existing (baseline) immune reactivity against cancer cells. This may depend on the ability of ICD to reboot and/or revive pre-existing TAA-directed immunity rather to prime de novo immune reactivity. In future, it would be crucial to characterize biomarkers that allow clinicians to delineate patients with reduced baseline immune reactivity against malignant cells so that proper combinatorial therapies involving ICD inducers can be implemented.

11It is well-established that the response of cancer patients to immunotherapy relies on the activity of effector T cells [that employ their T-cell receptors (TCRs) for recognizing TAAs]. However, these TAA-targeting T cells may also constitute obstacles for effective anti-tumor immunity. As opposed to T lymphocytes recognizing pathogen-associated antigens (PAAs) (Figure 2), indeed, T cells directed against some TAAs (derived from non-mutated proteins that are source of self or near-to-self antigens) are developmentally subjected to negative selection in the thymus and peripheral lymphoid organs (Figure 2). As a result, T cells bearing TCRs with high affinity for self antigens (including some TAAs) are clonally deleted to avoid auto-immunity (Figure 2). However, some “leakiness” in this process allows TAA-specific T cells possessing TCRs with low affinity to escape deletion and persist, although at low precursor frequencies (Figure 2). Unfortunately, as compared to PAA-specific T cells, which bear high-affinity TCRs (Figure 2), TAA-specific T cells exhibit limited effector and memory functions. Coupled with the tendency of progressing tumors to generate a highly immunosuppressive microenvironment, this renders the insurgence of lifelong protective immunity nearly impossible. Of note, central and peripheral tolerance may not affect T cells reactive toward neo-tumor-specific antigens (neo-TSAs) e.g., tumor-specific neo-antigens that are generated de novo in the course of tumor progression because of mutational events. However, the extent to which such neo-TSAs can elicit consistent “immunodominant” T cell reactivity is still a matter of investigation. Nevertheless, in this context, inefficient T-cell stimulation can be overcome through the ICD-based improvement of effector T-cell functions. ICD can be further combined with checkpoint-blocking therapies, which potently reverse immunosuppression. However, the lifelong maintenance of anti-tumor T cells remains a particularly hard challenge.

12In the clinical reality, anticancer agents are administered to patients in a limited number of cycles. Even if these therapeutic regimens may attain optimal efficacy in terms of ICD induction, they are unlikely to ensure the lifelong persistence of TAA-directed T cells with low-affinity TCR. This probably reflects the contraction of TAA-targeting T cells occurring once the immunostimulatory stimulus provided by ICD ceases, owing to peripheral tolerance mechanisms. Clinically, it may not be feasible to administer ICD inducers repeatedly over time, since many of them can cause lymphopenia (which negatively affects disease outcome), or are associated with other side effects. It has been proposed that active immunization with ICD-based anticancer vaccines (which are associated with robust immunogenicity) given in a repetitive manner may achieve this goal (Figure 2). Thus, it will be important to test whether the long-term administration of ICD-based anticancer vaccines can sustain the effector function of TAA-specific T cells bearing low-affinity TCRs, hence, ensuring lifelong disease-free survival. Of note, in the case of hematological malignancies, this issue could be overcome upon the adoptive transfer of CTLs expressing chimeric antigen receptors (CARs). However, whether CAR-expressing CTLs generate protective immunological memory in the absence of considerable side effects remains to be determined. Moreover, the use of this therapeutic strategy against solid malignancies is relatively challenging owing to lack of well-defined “unique” TAAs.

13The model of ICD has been considerably refined since the initial identification of a cell death modality manifesting apoptotic features but able to induce an adaptive immune response. This model strives to integrate several phenomena observed throughout the second half of the twentieth century in one therapeutically relevant platform. However, as discussed above, several challenges still need to be addressed. First, comprehensive testing should be performed in advanced experimental settings like GEMMs or orthotopic tumor models. Second, ICD resistance mechanisms should be characterized with precision. Third, various issues linked to the successful translation of ICD to cancer therapy will have to be resolved, including (but not limited to) treatment schedules, dosages, and combinatorial strategies. This translational drive also needs to be coupled with effective strategies for the discovery of new and effective ICD inducers. Drug screening programs are often complicated by the possibility of false-positive (due to bystander presence of DAMPs) or false-negative (due to limited number of biomarkers used for screening) hits. This issue can only be ironed out by discovering new and common regulators of ICD, and integrating them into existing screening platforms. Last, but not least, it will be important to identify new ICD-related/derived biomarkers that can be used to improve current protocols of patient stratification and clinical decision making. We are positive that all these objectives are at reach.

14ADG did the literature study, data collection, as well as conceived and wrote the manuscript. PA provided senior supervision and guidance, conceived the paper, helped in writing, and critically revised the manuscript. LG improved and edited the manuscript. JMBSP helped with the preparation of figures. All authors participated in the critical reading of the manuscript (wherever applicable), approved content and conclusions, as well as helped in ensuring the accuracy of cited literature.

15Akseli Hemminki is shareholder in Targovax AG and TILT Biotherapeutics Ltd. The remaining authors have no conflict of interest to declare.