Marc Monestier. Encyclopedia of Life Sciences: Supplementary Set. Volume 23. Chichester, UK: Wiley, 2007.
The conventional understanding of the immune system is that it can differentiate self from nonself. In contrast, the danger theory proposes that the immune system does not react to foreign substances but instead responds to situations that are potentially harmful.
For many years, immunologists have viewed the ability to differentiate between self and nonself as one of the essential characteristics of the immune system. This property should be essential as it enables the immune system to discriminate between a potentially harmful infectious agent and self antigens, the integrity of which must be preserved. However, self-nonself discrimination does not necessarily differentiate between a pathogenic agent and a harmless microbe. The body is constantly exposed to large numbers of microbial agents, many of them commensal, that pose no danger to the host. It would be wasteful for the immune system to develop a fully fledged response to a harmless bacterium. The main proposition of the danger theory is that the immune system has been selected during evolution for its ability to recognize a potentially dangerous situation, usually an infection, rather than to differentiate between self and nonelf.
The Danger Theory and Danger Signals
The danger model was originally proposed by Polly Matzinger, who has described it extensively in several articles (Matzinger, 1994; Matzinger, 1998; Fuchs and Matzinger, 1996; Gallucci and Matzinger, 2001). The theory contains some novel and controversial elements, but also combines previously known concepts under a new umbrella. As a result, this model has elicited a vigorous debate in the immunological community (Langman and Cohn, 1996; Vance, 2000). This article depicts the essential aspects of the theory and reviews some of the major criticisms of the model.
According to the danger theory, the immune system does not merely distinguish between self and nonself. Rather than detecting whether an entity is foreign or not, the immune system determines whether it will cause damage to the body. Antigen-presenting cells are probably the principal responders to danger signals because of their pivotal role in controlling the fate of an immune response. More specifically, dendritic cells must be critical in this process because of their ability to initiate a primary immune response and to activate naive T cells (Mellman and Steinman, 2001). According to the classical two-signal model, when dendritic cells are activated, they must provide two signals which in turn activate T cells. The first signal is provided by the antigen (ordinarily a peptide resulting from antigen processing) in association with major histocompatibility complex (MHC) class II molecules. This first signal can crosslink the T-cell receptor molecules, but a second signal is required to trigger cell division and differentiation by naive T cells. This second signal is usually a costimulatory molecule such as B7-1 or B7-2 which is expressed by the antigen-presenting cells and can bind a receptor, such as CD28, on the surface of the T cells. Without co-stimulation, signal 1 alone will induce anergy (tolerance) in the T cells rather than activation. In the absence of a danger signal (i.e. under normal homeostasis) dendritic cells will provide only signal 1 and maintain tolerance towards self antigens, whereas in a danger situation signal 2 will become available. Therefore, a critical issue is to understand how the immune system defines danger or rather what makes a danger signal.
The danger theory proposes that the signals that initiate immune responses are endogenous rather than exogenous. These endogenous signals are emitted by cells that are in a dangerous situation, such as stress, injury or necrotic death, but not by cells undergoing programmed cell death or apoptosis. Several groups have reported evidence that dendritic cells can be activated by necrotic but not by apoptotic cells, and that capture of apoptotic cells by dendritic cells can in fact be tolerogenic (Gallucci et al., 1999; Steinman et al., 2000). Nevertheless, as will be discussed below, these data have not been confirmed by all investigators and this issue remains an area of controversy.
The danger theory proposes that these danger signals fall into two main classes: constitutive and inducible (Matzinger, 1998; Gallucci and Matzinger, 2001). A cell may die without having the time to synthesize an inducible signal (obviously, an extreme danger situation), hence the need for constitutive or prepackaged warning. Theoretically, any intracellular molecule might work as such, provided it is not normally found externally and that antigen-presenting cells have a receptor for it. The nature of these intracellular components is, however, uncertain.
Mitochondria, as well as mannose or nucleic acids (deoxyribonucleic acid (DNA), ribonucleic acid (RNA)) have been proposed, but there is yet no convincing evidence that such molecules can initiate or potentiate an immune response. Intracellular nucleotides, adenosine triphosphate (ATP) or uridine triphosphate (UTP), can induce maturation of and cytokine production by dendritic cells by binding to a purinergic receptor that is found on their surface (Marriott et al., 1999; Schnurr et al., 2000).
We know more about inducible signals that can be produced by cells that find themselves under a stress situation. An obvious category is represented by heat shock proteins, which represent a large family of soluble proteins whose different members are found in various cellular compartments. Heat shock proteins overlap the constitutive and inducible categories of danger signals because some of them are produced constitutively whereas others are upregulated in response to stress. Heat shock proteins can bind to dendritic cells and macrophages via a variety of surface receptors, including scavenger receptors, to activate them (Chenet al., 1999; Binder et al., 2000; Singh-Jasuja, 2000; Todryk et al., 2000).
Among the inducible danger signals, Gallucci and Matzinger (2001) differentiate between two categories: those corresponding to existing molecules that are modified during cellular injury and those that are newly synthesized. The components of the first category are less well defined and may include lycosaminoglycan degradation products, which could bind directly to dendritic cells and induce activation and cytokine production (Kodaira et al., 2000). In contrast, there are a few better characterized candidates for newly synthesized danger signals. Type I interferons are virally induced molecules that can activate resting dendritic cells (Gallucci et al., 1999). Stressed cells may release reactive oxygen species that can upregulate various dendritic cell surface markers involved in interaction with T cells, including costimulatory and MHC class II molecules (Rutault et al., 1999). MICA is a stress-inducible surface molecule that is distantly related to MHC class I molecules. MICA (and its relative MICB) can be recognized by NKG2D, a receptor expressed on natural killer (NK) cells, CD8 T cells and γδ T cells. Engagement of NKG2D by its MICA ligand results in the activation of these cells and an increase in their cytolytic activity (Bauer et al., 1999).
Gallucci and Matzinger (2001) also propose that danger signals are primal or feedback molecules. Primal danger signals are a direct consequence of cellular trauma and their availability does not require any role from the antigen-presenting cells. In contrast, feedback signals (which are also usually primal signals) are produced by activated dendritic cells or by other cells that have been stimulated by dendritic cells. Many of the examples described above belong to the primal category, that is, they are produced by cells that are not part of the immune system, even though these primal signals will eventually activate dendritic cells. Some signals belong to both the primal and feedback categories, such as CD40 ligand, tumour necrosis factor (TNF) α and interleukin (IL) 1β. These molecules can be upregulated by nonimmune cells, but may also be induced in dendritic cells, thereby providing a positive feedback mechanism.
The Danger Theory in Common Physiological or Pathological Circumstances
The following three examples describe how the danger theory applies to frequently encountered clinical situations (Matzinger, 1998).
It would seem that the growth of a tumour represents extreme danger because uncontrolled cancer proliferation will result in the death of the host. However, the immune system may not perceive the tumour growth as dangerous because the cancer does not send any alarm signals. Indeed, as long as the tumour expands in a relatively ordered manner, it will offer signal 1 without signal 2. However, in practice, tumours do not progress following an ordered pattern of expansion, but often grow anarchically with inadequate neovascularization. As a consequence, necrotic cell death will occur in the tumour and should elicit a danger signal. However, there are many tumours that will pursue their growth while containing large amounts of necrotic tissue. The danger theory suggests an explanation to this apparent contradiction. Most tumours that contain necrotic segments are relatively large in size. Although the necrotic cancer material may result in dendritic cells activating the immune system against antigens from the tumour, these cells will not be sufficient to destroy the entire tumour, while at the same time the non-necrotic part of the tumour will induce tolerance by providing signal 1 without signal 2. Therefore, in such a situation, when there is conflict between two different messages, the tumour is likely to win as long as the growing mass of healthy (i.e. cancerous but non-necrotic) tissue maintains a relative advantage.
The application of the danger theory has led Polly Matzinger (1998) to propose an interesting design for cancer treatment. Various experimental approaches to cancer immunotherapy have been attempted by many groups. Most of these can be efficient if a vaccine is administered before the injection of tumour cells in animal systems. In humans, such treatments can induce temporary remissions or improvements (decrease in tumour size) but rarely result in a complete cure. The danger theory proposes an explanation for this lack of efficacy: the initial response induced by the treatment is not sufficient to eliminate the entire tumour and whatever is left of it will induce tolerance to the tumour (by providing signal 1 without signal 2). Therefore, the response to the tumour is normally downregulated because the immune system does not receive danger signals any more. Thus, an improved approach to cancer therapy would be to administer the tumour vaccine repeatedly so as to provide a danger signal constantly and reactivate the immune system.
The classical view of tolerance is based on the ability to distinguish between self and nonself. The danger theory views immune tolerance in a different light. A tenant of the danger theory is that the immune system does not make an intrinsic difference between self and nonself. A critical situation could develop during an infection, for instance, when both microbial and self antigens are presented by antigen-presenting cells with co-stimulation. It seems unavoidable that, during the course of an infection associated with danger signals, antigen-presenting cells will present both microbial antigens as well as antigens from cells that died during the inflammatory process, for instance. This situation may be referred to as ‘bystander autoreactivity’. Therefore, why is self-tolerance not broken each time we suffer an infection?Autoimmune disease; Autoimmune disease: pathogenesis; Autoimmune disease: aetiology and pathogenesis
There are several possible mechanisms to explain why an autoimmune disease does not develop in these circumstances. When cytotoxic cells kill an infected target, they do so by inducing apoptosis. Therefore, the consequences of cell-mediated cytotoxicity will not be viewed by the immune system as a danger situation. Also, the activated antigen-presenting cells that simultaneously present microbial and self antigens in a danger situation will not persist indefinitely. After the offending foreign antigen has been cleared, self antigens will again be continuously presented in the absence of co-stimulation, thereby re-establishing normal self tolerance. Therefore, in a normal individual, autoimmune manifestations will be transient and self-resolving. This view is supported by observations reporting the ephemeral production of self-reactive antibodies during or following certain infections. As explained above, the mechanisms that explain the lack of an immune response to tumours are the same mechanisms that explain the lack of an autoimmune response against normal tissue. Therefore, the danger theory does not contradict the concept of tolerance to self, but proposes that we are tolerant to self antigens because they are usually viewed in a nondangerous situation and not because of unique properties in their expression, such as early in ontogeny.
Why, then, do some individuals eventually develop an autoimmune disease? The danger theory approaches autoimmunity from a different perspective and suggests that the primary cause of autoimmunity is in the presentation of antigen and in the activation of antigen-presenting cells. Polly Matzinger subdivides autoimmune diseases into five categories based upon their putative origin (Matzinger, 1998).
(1) There is an unrecognized, ongoing infection in the target organ. In this case, there is no real autoimmune disease: the immune system is simply trying to fight off a chronic infection. There have been numerous reports of associations between certain autoimmune diseases and certain infectious agents, but, to date, no rigorous correlation has been observed between chronic infection and autoimmunity.
(2) There is a cross-reaction with an environmental antigen. This mechanism, also known as molecular mimicry is not unique to the danger theory and has also been proposed in the context of classical self-nonself discrimination. The difference between the danger theory and conventional models is that the danger theory postulates that self-destruction should stop once the pathogen has been cleared, because self antigens, in the absence of danger signals, will act as tolerogens. An example of such a situation is represented by rheumatic fever, an inflammatory disease occurring after group A streptococcal infection. The most serious complication of this disease is represented by lesions of the heart valves that are caused by antibodies cross-reacting with heart membranes and streptococcal M proteins. In this case, no new damage will occur once the infection has been cleared, although the cardiac lesions can leave serious sequellae.
(3) Another situation would be represented by ‘bad death’: cellular death recognized as a danger signal by the immune system. As mentioned above, this is in contrast with apoptotic or programmed cell death which does not elicit danger signals and therefore should not induce an autoimmune syndrome. Autoimmunity may thus arise in a couple of circumstances corresponding to a ‘bad death’ situation. The first would be exposure of the immune system to an abnormally large amount of nonprogrammed cell death (i.e. necrotic) that would result in the triggering of an immune response towards cell components that are released during this process. Massive cellular death can occur during a variety of situations (certain infections, burn or crush injuries), but there is no evidence that extensive, uncontrolled release of cellular material can trigger a bona fide autoimmune disease in a normal individual. A second situation could occur if the body, because of an inherent defect, was unable to clear normal apoptotic material. There are several clinical and experimental situations that suggest that this type of mechanism may actually take place. For instance, individuals with inherited deficiencies of upstream components of the classical complement pathway (C1, C2 and C4) have a dramatically increased incidence of lupus. Patients with a congenital deficiency in C1q have an 80% probability of developing systemic lupus erythematosus (SLE) (Walport et al., 1998). C1q can bind to apoptotic cells and may facilitate their clearance (Walport et al., 1998). Therefore, absence of C1q may result in the persistent exposure of apoptotic material, which may induce an autoimmune response. Likewise, serum amyloid P component is a plasma protein that binds to DNA and chromatin, as well as to apoptotic cells under physiological conditions. Mice genetically deficient for serum amyloid P component develop antinuclear antibodies and severe glomerulonephritis, similar to human SLE, a serious autoimmune disease (Bickerstaff et al., 1999).
(4) Another possibility is that the autoimmune response is simply a consequence of the disease, rather than the cause. This has been suggested for several diseases, such as scleroderma, where autoantibodies are produced against various intracellular components such as the nucleolus. The clinical manifestations in these patients, however, appear to be unrelated to the autoimmune component. It has, therefore, been suggested that autoantibodies in this syndrome do not contribute directly to the lesions, but are produced in response to tissue damage during the course of the disease. If this is correct, then one would expect to observe sustained autoantibody responses in most situations where chronic tissue injury occurs. However, consistent autoimmunity is not a usual feature of such situations. On the other hand, lesions in scleroderma may have a unique character (reperfusion injury) that could elicit unique danger signals and lead to autoimmunity (Casciola-Rosen et al., 1997).
(5) The last category overlaps the previous four. In this situation, the immune response switches to a harmful class. The nature of the antigen that is targeted by the immune response is not the critical element. The immune response may directly target an infectious agent or a molecular mimic, or may occur because of ‘bad death’. What matters is that an immune response that is harmful for a particular organ is going to take place in the wrong anatomical location. The production of such cytokines at other sites may not exert harmful consequences but will do so in certain situations. For example, such a scenario could take place when proinflammatory cytokines such as TNFα or interferon γ are produced in locations such as the anterior chamber of the eye or the pancreatic islets of Langerhans. The question must then be ‘Why did the wrong class of response take place?’. This could be due to extrinsic factors, such as the nature of the pathogen itself which may drive an inflammatory T helper (TH) type 1 response. There is also a suggestion that commensal pathogens may help maintain an anti-inflammatory environment and their absence may create a situation prone to become inflammatory. A harmful response may also result from intrinsic factors, for instance an inability of the body to maintain immunosuppressive conditions in specific sites. This may be due to a decreased production of TH2 cytokines or of transforming growth factor β, or a diminished expression of apoptosisinducing factors such as Fas ligand (Matzinger, 1998).
If the donor and the recipient differ in their MHC antigens, most organ transplants will be rejected even though the immune system has clearly not evolved to prevent surgical transplantation. At first glance, the immune system should not view an organ graft as dangerous because, most of the time, the aim is to replace a diseased organ with a healthy one. However, it is inevitable that the surgical procedure will cause some level of tissue damage and ischaemia. These danger signals will induce the activation of endogenous antigen-presenting cells within the transplanted organ, which in turn will elicit an antigraft response from the host. This may explain why tissues that are naturally low in antigen-presenting cells, such as the cornea, are usually accepted by the host, even without MHC matching. Likewise, tissues that are experimentally depleted of antigenpresenting cells are less likely to be rejected. An interesting situation is represented by the liver, which is often successfully transplanted across MHC barriers, despite being a fully vascularized organ containing many antigen-presenting cells. Two properties of the liver make this organ unique: its size and its ability to regenerate. Initially after the transplantation, and as a consequence of the surgical trauma, the liver’s antigen-presenting cells will stimulate an immune response against the graft. However, because of its size, the liver is not entirely eliminated by the host immune response, allowing sufficient time for it to replace the destroyed hepatocytes. By then, the danger signals have subsided and the liver antigens are presented to the immune system with only signal 1, and tolerance is re-established (Matzinger, 1998).
Problems with and Objections to the Danger Theory
If the danger theory is correct, necrotic cells or their contents should activate the immune system whereas apoptosis will result in ignorance or tolerance (Steinman et al., 2000). At first glance, it may seem surprising that the immune system has been educated well enough (during evolution, not ontogeny) to differentiate between bad (necrotic) and good (programmed) cell death, as there are so many intracellular components that become accessible in both situations. The literature provides contradictory data on this topic. Gallucci and colleagues (1999) have reported that resting dendritic cells can be activated by coculture with necrotic (but not with apoptotic) syngeneic cells. Further, necrotic cell products can serve as adjuvants and increase the in vivo response to a foreign antigen, such as ovalbumin (Gallucci et al., 1999). Teleologically, it also makes sense that dendritic cells should be able to present some antigens from apoptotic cells, because certain viral infections can result in apoptotic death clearly a dangerous situation for the body. Indeed, dendritic cells can present viral antigens from apoptotic cells (Inaba et al., 1998), but it remains unresolved whether presentation in this context can be immunogenic or will always induce tolerance.
In contrast with the above-mentioned findings, some investigators have observed that apoptotic cells can also trigger dendritic cell maturation (Rovere et al., 1998). The differences among these observations may be the result of differences in the subtypes and in the level of differentiation of the dendritic cells, a notoriously complex issue (Shortman, 2000). The importance of the cell type in the response to necrotic or apoptotic products is reinforced by a recent report which suggests that co-injection of cellular extracts with an immunogen increases cytotoxic T-cell responses (Shi et al., 2000). This adjuvant activity is constitutively present in the cytoplasm of normal cells and may be released following injury. Interestingly, this activity increases about 10-fold in cells that have died via apoptosis. These data support the hypothesis that, under certain circumstances, some cell types may also view apoptosis as a potential danger situation. An additional complication to this already difficult question has been suggested by the results of Salioet al. (2000), who have observed that neither apoptotic nor necrotic cells can induce maturation of dendritic cells unless they are contaminated with Mycoplasma. Deliberate infection of cells with Mycoplasma resulted in an increase in co-stimulatory molecule expression and cytokine production by dendritic cells (Salio et al., 2000).
An experimental challenge to the danger theory comes from ingenious experiments from several groups (Bingaman et al., 2000; Anderson et al., 2001). In these studies, immunodeficient mice were grafted with skin or hearts from allogeneic animals. The levels of histoincompatibility between donor and recipient ranged from drastic MHC differences to lesser antigenic disparities, such as the presence of the H-Y antigen. The grafts were allowed to heal for several weeks or months, so that they became quiescent and would not represent a potential danger. The immunodeficient animals were then reconstituted with various cellular preparations (e.g. bone marrow, spleen) from immunocompetent animals. When the cells used for the reconstitution were mismatched with the graft, the transplant was rejected. These results suggest that peripheral expression of an antigen in the absence of danger is not sufficient to tolerize newly emerging T cells and that central tolerance, presumably based on self-nonself discrimination, is critical in this process. Interestingly, one of these studies was performed by Matzinger’s group (Bingaman et al., 2000). These investigators hypothesized that the results were due to the fact that, although the grafts appeared to have healed when the animals were reconstituted, differences in gene expression persisted in the transplant that may have been perceived as danger by the immune system. However, the identity of these long-lasting danger signals remains to be determined (Bingaman et al., 2000).
As mentioned above, antigen-presenting cells must provide a second signal to trigger an immune response, whereas, in contrast, antigen presentation in the absence of this second signal will be tolerogenic. An important issue, which is usually accepted both by proponents and adversaries of the danger theory, is that an antigen-presenting cell does not have an intrinsic ability to differentiate between self and nonself when deciding whether to provide this second signal (Vance, 2000). A clue to this question may be provided by recent improvements in our understanding of innate immunity. It has become increasingly clear that the initial stages of the innate immune response to an infectious agent are critical in determining which type of cognate immunity will be developed (Janeway, 1992; Medzhitov and Janeway, 2000).
Professional antigen-presenting cells (dendritic cells, macrophages and B cells) express receptors that recognize pattern-associated molecular patterns (PAMPs). These PAMPs are unique to microbes and are not found in their eukaryotic hosts. They are shared by large numbers of pathogens within a particular class and are usually required for their survival. Examples of PAMPs include bacterial DNA, double-stranded RNA, lipopolysaccharide, lipoteichoic acid, peptidoglycan and mannans. Pattern-recognition receptors can be secreted or present on the cell surface, where they can mediate endocytosis (followed by processing of the pathogen) or signalling functions. These signalling receptors are members of the toll family, and are known as toll-like receptors (TLRs). For instance, TLR-4 is involved in the recognition of lipopolysaccharide; TLR-2 is required for the response to bacterial lipoproteins and peptidoglycans (the latter in collaboration with TLR6); and CpG-containing bacterial DNA activates via TLR9 (Hemmi et al., 2000; Medzhitov and Janeway, 2000). Engagement of pattern-recognition TLR receptors by PAMPs may result in the upregulation of the B7-1 and B7-2 co-stimulatory molecules and may direct an antigen-presenting cell to produce inflammatory cytokines, such as IL-12 (Medzhitov and Janeway, 2000). Therefore, in this situation, a PAMP will unwittingly notify the immune system that it is in a dangerous situation. As a consequence, PAMPs are occasionally referred to as ‘stranger’ signals. Does this mean that the danger theory and the stranger model are mutually incompatible? The answer is ‘Not at all’, because both signals can mutually reinforce each other. For instance, a bacterial infection could result in the production of PAMPs and in the upregulation of heat shock proteins, which together can induce the activation and maturation of dendritic cells. Therefore, danger and stranger signals can cooperate in activating the innate immune system to direct an adaptive response.
The danger theory and the controversy surrounding it raise the issue of the usefulness of such theories. Proponents of the danger theory have pointed out flaws in the self-nonself model and, conversely, opponents of the danger model have provided experimental evidence against it. Biological processes, including the immune system, are endlessly intricate and it may be quixotic to attempt to reduce them to a limited set of rules as may be possible in the field of particle physics. The complexity of the immune system is such that exceptions to any novel paradigm must necessarily follow its proposition. In fact, the purpose of such models may not be to provide a ‘theory of everything’, but to contribute an intellectual framework that will elicit innovative experiments. By this meter, the danger theory has already been successful.