Supplemental Data

Full Text
The Danger Model: A Renewed Sense of Self
Polly Matzinger

Supplementary Material

In the years since the Danger model was first published, certain questions have been asked repeatedly by both immunologists and non-immunologists, and were brought up once again by the referees of "Danger, a renewed sense of self". This supplement gives some brief answers, showing how the model can be used (see also refs [1, 2])

Tolerance to a changing self: Puberty, lactation, heart attacks and transplants between identical twins: One of the most common and important questions asked about the Danger model concerns self tolerance. Although the question takes many forms, its roots lie in a concern about tolerance in the presence of danger.

Why, if damage-induced alarm signals turn on the immune system, are transplants between identical twins not rejected? Or why doesn't an arm fall off after a cut or bruise? Shouldn't there be an autoimmune response?

Yes, there should be, and this important question lies at the heart of the difference between the SNS, INS, and Danger models. Although it has not been tested (to my knowledge) for minor cuts or bruises, such autoimmune responses have been documented for damaged hearts. Up to 50% of people suffering heart attacks transiently produce auto-anti-heart antibodies [3, 4]. The Danger model explains both the initiation and the transience of this antibody response, suggesting that it is initiated by APC-stimulating alarm signals sent by damaged heart cells, and ends when the heart heals and the alarm signals cease.

The assumption underlying this suggestion is that tolerance occurs under conditions of health (no costimulation) and activation occurs under conditions of danger (because the alarm signals upregulate the costimulatory capacity of APCs). Because normal heart antigens are always present, new heart-specific T cells should become tolerant as they and encounter the antigens. Some of the T cells may be tolerized in draining lymph nodes by unactivated APCs presenting the tissue antigens without sufficient costimulation, while others may become tolerized by direct encounter with the tissues themselves (which do not express costimulatory signals). Therefore, at the time of the heart attack, there will be very few heart-specific T cells able to respond to the newly activated APCs. Although these few will respond, their numbers are too low to do noticible damage.

Also, because activated APCs seem to live only for a few days, the response will cease when the damaged heart cells heal and stop sending alarm signals. Any remaining heart-specific T cells will then die or become resting memory cells. As they circulate and transit through the healthy heart, they will be tolerized by the lack of co-stimulatory signals. Thus, though a transient autoimmune response may occur after any damage or infection, the response stops and tolerance is slowly re-established after the tissue heals.

Tolerance in self-reactive cells looks like a self-non-self definition; consequently the difference between Danger and SNS has sometimes been dismissed as semantics. But semantics has to do with meaning, and the models do not give the same meanings to the words "self" and "foreign". In a SNS model, the word "foreign" means "that to which the immune system should respond" and "self" means "that to which the immune system should be tolerant". The Danger model uses the words "self" and "foreign" to distinguish things encoded by our own genes from those encoded by other organisms. We can therefore speak about "self-tolerance" and understand each other. But these meanings have no impact on the decision to respond, which is governed by the presence or absence of alarm signals. Because most self is harmless (and persists) we will be constantly inducing tolerance to it. But tolerance will also be induced to antigens expressed by commensal organisms, and by parasites that do no cellular damage, so that the set of things to which the immune system becomes tolerant is not the same as "self", but encompasses any persistent antigen that is not associated with danger.

I would predict that, if we looked closely, we would find transient autoimmune responses occurring whenever there is danger. Examples include trauma-induced-uveitis, where injury to one eye can sometimes lead to autoimmune damage to the other [5], and the transient T and B cell responses to the blood clotting factors FVIII and FIX found in healthy people [6], sometimes (Conti-Fine, personal communication) after accidental bruising.

Viruses that don't cause necrotic death:
Although some viruses cause necrotic deaths in the cells they infect, others induce apoptotic death. A referee asked "why is there an immune response to these viruses?"

This question arises from a misunderstanding of the basic principle of the Danger model. Although it is sometimes tempting to simplify it into the nice little sound bite "apoptosis vs. necrosis", this is a bit misleading. There are many forms of apoptosis and the Danger model does not use them as defining criteria. The model is based on the difference between healthy and stressed/injured cells. It suggests that cells do not release alarm signals when they die by normal physiological processes of cell death, and are scavenged by normal physiological processes of scavenging (perhaps we could call this entire process "natural death"); whereas cells do release alarm signals when they are stressed, injured, or die abnormal deaths, or are ineffectively scavenged. It is difficult to compress this full idea into a sound bite, but it is important to get it right because the model depends on it. Any departure from the normal healthy state of a cell may be enough to release alarm signals that activate local APCs.

Very few viruses can complete their life cycles without causing some form of cellular stress or damage. In fact, a prediction from the Danger model is that any virus that is truly benign will elicit no immune response. In addition, many virally infected cells produce and release interferon alpha (IFNGreek Letter Alpha), which acts as a warning signal to neighboring cells, signaling them to produce inhibitors of viral RNA transcription. Because of this, we wondered if IFNGreek Letter Alpha might also act as an alarm signal to warn local APCs of the infection. We found [7] that IFNGreek Letter Alpha is indeed a very potent activator of bone marrow derived dendritic cells.

Pregnancy:1) If fetuses are not rejected because they are healthy and do not send alarm signals, why does an Rh negative mother respond to her Rh positive fetus?

The SNS models assumed that the mother should reject her foreign fetus, and devised a number of explanations for why she doesn't: 1) there is a physical and immunological barrier between mother and fetus. 2) the fetus immunosuppresses its mother. 3) the placenta suppresses local immunity. None of these solutions are evolutionarily sound. Should a fetus become infected, a mother unable to deal with that infection would die. Immunosuppressing the individuals carrying the next generation seems a poor evolutionary strategy, and it amazes me that these "solutions" are taught. In fact, there is cell traffic in both directions across the placenta; [8, 9], and pregnant females are not immunosuppressed but can respond to dangerous organisms [10].

Instead, the Danger and INS models both suggest [11] that healthy fetuses are not rejected because they don't send exogenous or endogenous alarm signals to activate APCs. However, should the fetus become infected, the maternal immune system will be alerted and if, in clearing the infection she also rejects the fetus, she will at least live to have another.

Why then does Rh disease occur? The reaction of Rh negative mothers to their Rh positive fetuses is an interesting case, illustrating that it is critically important to consider the details when offering explanations for any particular phenomenon. The salient detail about the maternal reaction to Rh is that it is usually not a problem in the first pregnancy, but only in subsequent ones. Mothers are immunized to the Rh antigens of their children during parturition. Birth can be a dangerous process, with bleeding and rampant alarm signals. During the first pregnancy itself, while there were no alarm signals, mothers do not respond to the fetal Rh antigens. This may be why treatment with Coomb's factor (anti-Rh antibody) is not usually needed during the entire first pregnancy but only around birth.

2) Why do pregnant women (and their fetuses) in malaria-endemic areas do poorly compared to non-pregnant women? For some time this was offered as evidence that pregnancy is immunosuppressive; however recent data show that it is actually evidence for immunity to a well adapted parasite. A variant of malaria has been discovered that grows only in the placenta, binding specifically to placental glycosaminoglycans [10]. Primiparous women in endemic areas have protective memory responses to many other malaria variants, but are immunologically na�ve to this pregnancy-specific variant. During the first pregnancy their immune responses to the variant are slow and inefficient, but, by the second they do better, and they continue to improve. Men and nulliparous women are not immune to this variant; only older females are immune, and the variant only grows during pregnancy. Thus pregnant women are definitely not immunosuppressed and can mount protective immune responses, even to well adapted parasites [10].

Tumors: If tumors are not rejected because they are healthy cells that don't send alarm signals, what about solid tumors that have large amounts of necrotic death in their centers?

Although the interactions of tumors with their "host" are complex, both the Danger and INS models suggest that tumors are not rejected primarily because they are healthy un-infected cells that do not alert APCs. What then of necrotic tumors? At first glance, the Danger model suggests that these should be cleared. However, a more detailed look gives a different picture. The tumors in question are large, solid, poorly vascularized tumors that eventually begin to starve for lack of nutrients. Starvation, however, causes a form of apoptotic, rather than necrotic death (often dubbed "sterile necrosis" by onco-pathologists), and should elicit no immune response. Eventually, when the number of dying cells overloads the local scavenging capacity, dead cells will disintegrate, releasing APC-stimulating alarm signals [12]. If APCs reside in the tumor, an immune response should then ensue. By this time, however, the tumor has had a long time to tolerize any circulating tumor-specific T cells. Although a few remaining tumor-specific killers may become activated and infiltrate the tumor, each killer can only kill a limited number of tumor cells before it needs to be reactivated. Bt this time, the race is already lost and the tumor, already large, will continue to grow, a case of too little, too late.

Expanding tumor infiltrating lymphocytes (TILs) in vitro is a step towards effective treatment, but they will not stay active forever. This may be why TIL therapy has had some spectacular successes but also many failures. When TILs are isolated, expanded and re-injected into patients, tumors can become visibly smaller, and sometimes disappear. However, in most cases, the initial onslaught by activated TILs is not enough to clear a large tumor and, without re-stimulation, the response wanes, and the remaining tumor cells grow [13]. Because killers kill by inducing apoptosis [14], they cannot maintain their own response. It will be necessary to activate the patient's APCs, by using a vaccine, or by causing damage to the tumors, in order to maintain the response and clear the tumors.

I believe that TIL therapy, and anti-tumor vaccination, can cure tumors. But we will have to let go of the old fashioned idea that the immune system, once turned on, continues to fight until the antigen is gone. It won't. It will continue to fight only until the Danger is gone, and it does not recognize most tumors as dangerous because the cells do not send enough alarm signals. To make TILs and anti-tumor vaccines more effective, we should boost the patients again and again and again, until the last tumor cell is gone.

Autoimmunity [2]: For years, researchers studying autoimmunity have been asking the question "what breaks tolerance?" I think this is the wrong question. We are never completely tolerant. As long as the thymus and bone marrow are producing new B and T cells, we will have a few new circulating self-reactive lymphocytes. If instead we ask "What stimulates these self-reactive cells and what maintains the response?", a world of possibilities opens up. I think there are at least four different categories of stimulus and they result in different sorts of autoimmune diseases.

1) An unrecognized infection in the target tissue: These aren't true autoimmune diseases, as self antigens are not the primary target. The immune system is doing its job (clearing a pathogen) and damages the target tissue in the process. Lyme disease, for example, has symptoms very similar to juvenile rheumatoid arthritis [15].

2) Molecular mimicry by a pathogen that has some similarity to a self tissue: T cell recognition is far less specific than previously thought [16], and some pathogen-specific T cells will also see a self antigen. These will respond against both the pathogen and the self tissue while the infection persists and stop when the pathogen is cleared (and the damage stops), only to be re-activated if the infection soon re-occurs. This category differs from the first, as the disease has a truly self-reactive component and the infection need not therefore be in the targeted self tissue. Essentially similar to rheumatic fever, this may be the category for some "flaring" diseases.

3) Bad death: (This category is unique to the Danger model, as there are no infectious agents and no "foreign" components necessary.)

All over the body cells die every day. In some cases, such as red cells, thymocytes, skin and gut, death is fairly constant. In others, such as the ovaries and uterus, it is regular, but not constant. In yet others it is completely irregular, for example in germinal centers, where mutating B cells die in droves. Each of these deaths is controlled by a series of genes coding for different aspects of the death and clean up processes and, as with all genes, mutations happen. Any one of these mutants could result in the release of alarm signals that initiate immune responses. Environmental toxins that cause cell damage could also lead to the release of alarm signals. There need not be a foreign antigen involved, as the autoimmune response occurs because of persistent cellular distress. Some forms of lupus, for example, have been associated with deficiencies in death and/or scavenging genes [17], and some forms of autoimmune hemolytic anemia may relate to a somatic mutation that results in necrotic cell death [18].

Livia and Antony Rosen have suggested something similar for scleroderma. Their idea is that this is primarily not an autoimmune disease, but that the primary defect is one in the cell death process, and that the immune response occurs as a result of the bad death [19].

A question that often arises is "why are lupus and some other autoimmune diseases primarily seen in females?". The SNS and INS models have no answer to this, but the Danger model potentially offers a partial explanation based on the forms of cell death that occur in female but not male bodies. We pop an egg, and kill off, clean up and regenerate an entire uterine lining every month. That is a lot of death and cleaning up to do and leaves us more susceptible to mutations in death and scavenging genes. It might also explain why some autoimmune flares are less frequent in pregnant women (in whom these death and clean up processes are temporarily suspended), only to increase again after giving birth.

4) The wrong class: The antibodies and cytokines released during an immune response come in different effector classes (Th1, Th2, Th3 and their respective antibody classes etc). These are potent molecules that can wreak enormous havoc, and some tissues are more sensitive to certain effector molecules than others. Celiac disease, for example, seems to be caused by a Th1 response to a protein in wheat; the cells of the gut villi cannot deal with the cytokines made in a Th1 response, and the disease stops when celiac patients stop eating wheat [22]. In experimental situations, Diabetes and EAE, both characterized by TH1 responses, can be alleviated by a switch to a Th2 response [20, 21]. In these cases, the response continues, but no longer damages the target tissue. Such diseases show that the immune system can kill a tissue with the wrong class of immunity even if the response is not directed against that tissue.

Unlike the areas of transplantation and tumor treatment, the Danger model does not offer immediate prescriptions for treatments for autoimmune diseases. However, it does offer a new viewpoint and this can be as important. Immunologists have spent over half a century studying autoimmunity from the point of view that these diseases result from a defect in self tolerance, and we have not gotten very far in understanding them. Perhaps if we changed our viewpoint, we would see things that we have not noticed before.

If we accepted for a moment, the Danger model's point of view that immunity is controlled by the tissues that the immune system in meant to protect, we might ask questions we have not asked before, or follow leads we have not followed before, and we might discover features of immunity that have long lain hidden.

  1. Matzinger, P., Annu.Rev.Immunol., 1994. 12: p. 991-1045.
  2. Matzinger, P., Semin. Immunol., 1998. 10: p. 399-415.
  3. Elahi, A., Vijayakumar AN, Lichstein E, Mokhtarian F., J Lab Clin Med, 2001. 138(2): p. 112-8.
  4. Prohaszka Z, et al Eur J Clin Invest., 2001. 31(4): p. 283-4.
  5. Kilmartin, D., et al Br J Ophthalmol., 2001. 85: p281-6.
  6. Reding, M., et al Thromb Haemost, 2000. 84(4): p. 643-52.
  7. Gallucci, S., M. Lolkema, and P. Matzinger, Nat Med, 1999. 5: p. 1249-55.
  8. Bonney, E.A. and P. Matzinger, J.Immunol., 1997. 158: p. 40-47.
  9. Zhang, L. and R.G. Miller, J. Immunol., 1993. 56: p. 918-921.
  10. Beeson, J., Reeder JC, Rogerson SJ, Brown GVTrends Parasitol, 2001. 17: p. 331-7.
  11. Actually, Janeway has not discussed what predictions his PRR view would make, usually limiting discussion to the activation of the APC. I am inferring what he might say, in order to outline the difference and similarities between the two models.
  12. Shi, Y., W. Zheng, and K.L. Rock, Proc Natl Acad Sci U S A, 2000. 97(26): p. 14590-5.
  13. Rosenberg, S.A., et al., J.Natl.Cancer Inst., 1994. 86: p. 1159-1166.
  14. Duke, R.C., R. Chervenak, and J.J. Cohen, Proc.Natl.Acad.Sci.U.S.A., 1983. 80: p. 6361.
  15. Steere, A.C., Scand J Infect Dis Suppl, 1991. 77: p. 51-4.
  16. Garcia, K., et al 1998. 279(5354): p. 1166-72.
  17. Maekawa, Y.a.K.Y., Trends in Immunology, 2001. 22: p. 662-663.
  18. Hillmen P, R.S Br J Haematol., 2000. 108: p. 470-9.
  19. Rosen, A. and L. Casciola-Rosen, Cell Death and Differentation, 1999. 6: p. 6-12.
  20. Katz, J.D., C. Benoist, and D. Mathis, Science, 1995. 268: p. 1185-1188.
  21. Chen, Y., et al., Science 1994, 1994. 265: p. 1237-1240.
  22. Nilsen, E.M., et al., Gut., 1995. 37: p. 766-776.