The danger theory of immunity revisited

Introduction In 1994, Polly Matzinger provocatively introduced the ‘danger the- ory of immunity’1 (Box 1). According to her hypothesis, the immune response is not dictated by the nature of the antigen (signal 1), in contrast to the prevailing hypothesis at the time that autoantigens would be ignored, whereas xenoantigens would be recognized by T lymphocytes engaged in self–non-self-discrimination 2–4 . Instead, she proposed that the immune response is conditioned by the pres- ence of tissue stress, damage or infection that causes danger signals to activate antigen-presenting cells (APCs), which in turn provide a second signal (signal 2) to T lymphocytes. Signal 1 alone would fail to induce an immune response or even elicit immune tolerance, but the combination of signals 1 and 2 would induce an immune response. In this scenario, neither APCs nor T cells ‘decide’ whether an immune response will ensue. Instead, the danger signals emanating from the tissue govern this decision (Fig. 1). Thirty years have elapsed since the original formulation of the danger theory, which was initially regarded as highly unorthodox. However, following some controversy 5–8 and refinement 9,10 , the dan- ger theory has been integrated by the scientific community. Of note, in 2002, Polly Matzinger clearly pointed out the incompatibilities between her theory and that of self–non-self-discrimination11 . Later, in 2011, she introduced the idea that the tissue context would tailor the type of the immune response that is mounted, hence determining which subsets of myeloid or lymphoid cells accrue and which cytokines and other effector molecules are produced 10 . Thus, tissue not only decides whether an immune response ensues but also which type of response will arise. Since the initial introduction of the danger theory 1 , three dec- ades have witnessed spectacular progress in immunology, includ- ing the molecular identification of a vast array of damage-associated molecular patterns (DAMPs) which typically act on various pattern- recognition receptors (PRRs), many of which also recognize microorganism-associated molecular patterns (MAMPs) 12–15 . Furthermore, it has been discovered that in the absence of an intesti- nal microbiota, vaccination responses cannot be elicited, suggesting that MAMPs function as obligatory immunostimulants16 . Microbial virulence factors, or ‘effectors’, can be detected owing to the perturba- tions that they cause in the intracellular milieu, hence inducing stress responses that stimulate ‘effector-induced immunity’17 . Additionally, a class of stress-associated immunomodulatory molecules (SAIMs) with systemic immunomodulatory properties has emerged18 (Box 2). Progress in single-cell transcriptomics and proteomics has facilitated the establishment of an expanding catalogue of immune cell subsets, ranging from different myeloid cell types, including dendritic cell (DC) subclasses and tissue-resident macrophages, to lymphoid (for exam- ple, B, mucosal-associated invariant T (MAIT), natural killer (NK), NKT and T cells) subpopulations with distinct functions and increasingly nuanced ‘polarity’ 19–21 . These immune cells are not only engaged in potentially harmful defence responses against pathogens that can lead to collateral inflammation and autoimmunity but also have a major role in tissue homeostasis, as well as in the silent elimination of premalignant or senescent cells 22 . Expanding knowledge has allowed the distinction between immunity, which involves the generation of memory responses by clonally expanded T and B lymphocytes, and inflammation, which lacks such memory, and hence the differentiation of autoimmune pathologies from autoinflammatory pathologies 23–25 . As a result, we can apply the danger theory to immunity and examine how danger signals contribute to the induction of (auto)immune responses (Box 3).

In this Perspective, we summarize the literature on MAMPs, DAMPs and SAIMs to refine and extend the original danger theory. Special emphasis is placed on evidence from studies of human pathogens, cancers and on immunogenetic investigations that validate aspects of the danger theory in patients. Moreover, we update the danger theory with respect to recent insights into immune cell subsets as we attempt to render it compatible with older theories of tolerance achieved by clonal deletion, anergy and immunosuppression. DAMPs and sterile tissue damage A central question arising from the danger theory concerns the pre- cise molecular definition of the danger signals that emanate from damaged tissues. Here, we discuss danger signals in the context of sterile (non-infectious) damage, inflammation and immune responses. Much of the literature on sterile tissue damage is based on stud- ies with cancers that were treated by a vast array of drugs, including chemotherapeutics and so-called targeted agents, and by physical methods, including irradiation, microwave ablation, ultrasound and cryotherapy, as well as by combinations of chemical and physical methods, such as chemoembolization and photodynamic therapy26 . Although none of these treatments was initially conceived as an immu- notherapy, it appears that the most successful interventions are those that are capable of stimulating an antitumour immune response and

Box 1 | Historical context of the danger theory During the 1980s and 1990s of the twentieth century, immunologists largely focused on the molecular haracterization of antibodies and T cell receptors, their interaction with alien and self-antigens, as well as the mechanisms of positive and negative selection yielding the immune repertoire. In this context, the mechanisms accounting for the avoidance of autoimmunity or its pathogenic surge were investigated while focusing on the interaction between adaptive immune receptors and antigens. These interactions were thought to be regulated in a way that correct self–non-self-discrimination would be assured. Thus, it was postulated that autoantigen-specific T and B lymphocytes would be eliminated or silenced (that is, ‘tolerized’) by clonal selection or inactivation, respectively. In this context, unwarranted autoimmune responses would occur owing to the failure of selection/inactivation mechanisms or owing to the generation of ‘altered self’, that is, the formation of new autoantigens against which the system had not been tolerized 4,221,222. When the importance of innate immune effectors and receptors was discovered in the 1990s, it was speculated as well that the innate immune system would contribute to self–non-self-discrimination through the recognition of molecular patterns distinguishing infectious non-self from self 6,223. In this context, Polly Matzinger’s idea1,10,11 that the immune system would remain inert in healthy, unstressed tissue and become activated owing to the presence of danger and damage, even in the absence of infection, hence enabling an immune response against locally present antigens, constituted a frontal provocation against established self–non-self-discrimination theories.

hence convert treated tumour lesions into an in situ cancer vaccine22. Of note, immune responses are most efficiently induced when the tumour is not entirely lysed (including all its parenchymatous and stromal elements, as well as infiltrating immune cells), but when the damaged area is surrounded by a penumbra in which stressed or dying cells encounter recruited leukocytes26. Moreover, the greater therapeutic efficacy of neoadjuvant, as opposed to adjuvant, chemotherapies and immunotherapies against cancers22,27–29, supports the view that whole tissue rather than minimal residual disease owing to isolated transformed cells should be targeted for the induction of longlasting immune responses. Normal and cancerous cells respond to external cues, includ- ing hypoxia, deficient or excessive glucose concentrations, nutriment deprivation, osmotic or mechanical stress, hyperthermia or DNA damage, as well as to intracellular derangements, such as ionic imbal-ance, DNA replication stress, mitotic arrest, organellar dysfunction or the unfolded protein response. Such internal or external challenges induce homeostatic adaptations that can lead to the repair of damage, or, on the contrary, initiate a series of irreversible alterations that culminate in senescence or cell death30. Cell death can occur in a ‘violent’, accidental and hence unregulated or lytic fashion or in the form of ‘regulated cell death’ (RCD), in which the absence or presence of effector molecules and their regulatory factors dictates the mode of demise, which can be apoptotic, ferroptotic, necroptotic or pyroptotic, among others 31 . Moreover, stress, senescence, lysis and RCD are communicated by the affected cells to the environment through alterations at the cell surface, as well as by the release of factors into the extracellular environment30.

Fig. 1 | The danger theory of immunity in a nutshell. a, In normal, homeostatic conditions, antigen-presenting cells (APCs) remain in a resting state and only provide signal 1 in the form of MHC class I or MHC class II restricted peptides to T cells. This results in the absence of T cell response or the induction of immune tolerance. b, In stressed, damaged or infected tissues, danger signals trigger APC activation, enabling the induction of a genuine T cell response.

Microbial-associated molecular patterns (MAMPs) have been identified as bacterial, fungal or viral molecules that are chemically different from their host and are recognized by pattern-recognition receptors (PRRs). One of the first MAMPs to be characterized was bacterial lipopolysaccharide, which binds the PRR Toll-like receptor-4, which is expressed by macrophages and dendritic cells, leading to their activation211. Danger-associated molecular patterns (DAMPs) are usually hidden behind the intact plasma membrane of mammalian cells42. DAMPs can be exposed on the surface of stressed cells (as exemplified by calreticulin, CALR, an ‘eat-me’ signal for dendritic cells) or become accessible owing to the rupture of the plasma membrane (as exemplified by F-actin, a component of the cytoskeleton). lternatively, DAMPs can be actively secreted or passively released as soluble molecules by stressed and dying cells. Many DAMPs act on PRRs. 

A prominent example is high mobility group B1, a non-histone chromatin binding protein released by dead cells, that binds to Toll-like receptor-4 and other PRRs224. As a rule, both MAMPs and DAMPs predominantly mediate immunostimulatory effects42,211. Stress-associated immunomodulatory molecules are released in the context of systemic neuroendocrine stress responses, including during infection and sterile trauma18. These molecules comprise neuroendocrine factors such as corticosteroids, neurotransmitters, stress hormones and neuropeptides that allow short-term adaptations to acute systemic stress and often mediate immunosuppressive effects18. Hence, damage to tissues can generate local and systemic danger signals that influence the immune response