TOLERANCE AND THE FOUR D's (danger, death, destruction and distress)
by Polly Matzinger

As a student, I was taught that the immune system discriminates between self and non-self; that it attacks foreign antigens and is tolerant (except in disease states) of self and that it defines "self" as the set of antigens present early in life, when the immune system is developing. After years of having trouble with this view, I have now abandoned it.

Although the old view explained most of the features of immunity, it also left some real gaps. For example, if "self" is learned during early fetal or neonatal life, what happens at puberty, or during lactation? Though individuals change greatly at puberty, most are not destroyed by their immune systems and the newly lactating breast, generating large quantities of (new) milk proteins, is not rejected. The changes that occur in 'self' are not the only problem with the old view. Why do normal individuals contain natural antibodies and/or T cells able to see stable self antigens, such as DNA, tubulin, or myelin basic protein, and, given that such autoantibodies and T cells exist, why are autoimmune diseases rare? How are T cells made tolerant of tissue-restricted MHC/peptide complexes that are not in the thymus? Why are livers more easily transplanted than skin? Why are tumours not rejected more often? As long as we continue in the old beliefs, there are no good answers to these questions. However, Ephraim Fuchs and I have built a model of the immune system which supplies these answers. In an earlier essay I described such an immune system in detail and laid down the rules by which it operates to maintain self tolerance. In this essay, I will describe it briefly and touch mainly on the last two questions listed above, namely tumors and transplantation.

Let's begin with the definition of "self" which, over the years, has sometimes been extended from 'everything encoded by the genome' to 'everything under the skin', in order to include commensal genomes. It has been narrowed to exclude the 'privileged' sites such as brain, cornea and testes. It has been modified for T and B cells. For T cells, Waldmann suggested that self consists only of MHC/peptide complexes, and Zinkernagel et. al. reduced it further, arguing that self consists only of peptides found in the thymus and that all other tissues are "ignored". For B cells, Cohn defined self as cell surface and soluble molecules, proposing that"housekeeping" antibodies to intracellular components might help clear cellular debris. Mitchison defined it as those bodily proteins present at a concentration above a certain threshold, Jerne excluded antibody idiotypes because their individual concentrations are too low, and Coutinho and his colleagues defined it in terms of an idiotype network, which they call a "positive definition of self".

"Non-self" is equally difficult to define. There are plenty of non-self structures that the immune system does not attack (e.g. silicone, bone, peptides {depending on MHC type}, solitary haptens, and food) because, as Dresser pointed out years ago, they lack a curious characteristic called "adjuvanticity". For example, though ovalbumin is certainly "foreign" to a mouse and MBP is "self", both proteins elicit responses if injected with adjuvant while neither does if injected intravenously. Clearly self vs non- self matters less here than adjuvanticity. Recently Janeway pointed out that few antigens are particularly immunogenic. He suggested that "adjuvanticity" is a property of evolutionarily distant antigens, like bacterial cell walls, to which the immune system is poised to respond. Ultimately all of the models boil down to the idea that the immune system regards a certain subset of the body as self and a particular fraction of the rest of the universe as foreign. In short, it doesn't really discriminate self from non-self but some self from some non-self. The immune system thus classifies antigens into four categories.

1) Visible Self = bodily structures to which the immune system is tolerant.

2) Invisible Self = bodily structures to which it is not tolerant but to which it does not normally respond.

3) Visible Non-self = structures to which the immune system normally responds.

4) Invisible Non-self = non-immunogenic structures to which it does not respond.

The problem with this kind of classification is that it is virtually impossible to uncover the rules by which the immune system could make the necessary distinctions. For example, Waldmann's idea that self, for T cells, consists of common peptide/MHC complexes, begs the question of what happens when, during the course of a normal life, the peptides change. Zinkernagel's idea, that T cells are tolerant only of antigens found in the thymus and "ignore" all other tissue antigens, also runs into trouble. Since T cells respond to short peptides and crossreact on non-identical but similar peptides, Zinkernagel's "ignorant" T cells should often be activated by cross-reactive environmental antigens and autoimmunity should be a frequent and irrevocable event, rather than a rare occurance. Janeway's idea, that the immune system responds to "infectious non-self" because it has a phylogenetic memory of infectious organisms, may be partially right but it does not cover such cases as graft rejection, or responses to many viruses. Mitchison's and Jerne's view, that the immune system is tolerant of "normal" concentrations of self antigens, cannot deal with proteins whose concentrations change. In fact, though change is one of the hallmarks of life itself, the self vs non-self models all suffer from the idea that the immune system learns (either during evolution or ontogeny) a static definition of self that it must then live with for the life of the individual.

If we forget about self for the moment and step sideways to look at the other side of the equations above, we find it possible to ask a different question, namely "how does the immune system decide whether tor espond or not?" I have conjectured that it might respond to danger,not to non-self. Clearly, distinguishing dangerous from harmless entities would be an efficient and evolutionarily sensible thing to do, and it turns out that the task is not as difficult as it first might seem. Among the many potential definitions for "dangerous", the one with the most explanatory force and which leads to the simplest model with the most predictive power is "anything that causes cell stress or lytic cell death". Cell death is an integral part of living systems. It occurs during ontogeny and in adult life. We find death in the thymus, death in the bone marrow and blood, death in the brain, skin, gut, liver. Literally, death everywhere. But this is normal, programmed cell death. It usually involves apoptosis and the dying cells are ordinarily shed to the outside environment or scavenged by specialized cells. This sort of death does not appear dangerous to the immune system. However, should a tissue be stressed or die abnormally, the immune system is alerted and responds. Let's take an example of a virus infection to see how it could work. part 1) Initiating the response.

Danger is sensed by tissues themselves, which signal the professional antigen presenting cells among them. Figure 1A shows what a typical tissue (say skin) might look like under normal, conditions. It consists of a mixture of actively growing (and dying) skin cells and local antigen-presenting dendritic cells, in this case Langerhans cells, which are essentially quiescent. When a lytic virus infects the skin, the infected cells die by non-apoptotic death, activating the Langerhans cells, which then capture the antigens in their neighborhood, up-regulate MHC molecules, lose Fc receptors, and travel to the local lymph node, where they present the captured antigens to passing T cells. The signals sent by stressed or dying cells are not known at the moment but they could be of several sorts. A dendritic cell might become activated if a cell to which it is connected suddenly dies, simply from the sudden loss of connection. It might have receptors for a heat shock protein elaborated by stressed cells or for a protein, normally found only on the inside of intact healthy cells, that leaks out only if the cell dies lytically. There are a myriad of potential ways in which a tissue could communicate danger to its local APCs and I have no reason to choose any particular ones. It suffices for the moment to postulate that such danger signals exist, that the constant normal (apoptotic) cell death found in many tissues does not elicit them, and that without them, the APCs remain quiescent and intiate no immune responses. part 2) training the T cells

Since APCs cannot distinguish self from non-self, they capture normal skin antigens as well as viral antigens and present both to passing T cells. In self-non-self models, T cells must be trained to tell the difference between these foreign and self antigens. In an immune system poised for danger, the T cells need only to distinguish APCs from everything else. In a nutshell, if any T cell specific foran APC antigen were deleted, and if APCs were the only cells able to activate T cells, then the only T cells that would be activated during an immune response would be those specific for new antigens presented by APCs. To build such a T cell population, we need three basic laws and one minor exception. the laws of lymphotics The First law, taken from the Bretscher-Cohn and Lafferty-Cunningham models, assumes that resting T cells need two signals to be activated; signal One from TCR binding to MHC/peptide and signal Two (co-stimulation) from an APC. It states that T cells die if they receive signal One without signal Two and become activated if they receive both.

The Second law states that resting T cells can only receive co- stimulatory signals from APCs . Interdigitating dendritic cells (and, perhaps, macrophages) can serve as APCs for both virgin and experienced T cells, and B cells can re-stimulate experienced but not virgin T cells. Though other tissues might express surface MHC/peptide complexes, they do not express the appropriate co-stimulatory signals needed by T cells.

The Third law states that the activated effector stage only lasts for a certain period of time. During this time, T cells do not require co- stimulatory signals and can be triggered to function (eg. help B cells or kill targets) by signal One alone. After a while, they either die or return to a resting state from which they can only be drawn again by the appropriate combination of signal One plus signal Two.

Together, the first and second laws compel any mature T cell that encounters a non-APC tissue (like skin) to die from the lack of a second signal. The third law ensures that the first two apply to experienced T cells. Add the idea that stressed or dying tissues signal their local APCs, and a picture emerges of an extended immune system in which every bodily tissue is deeply involved in its own protection. Each tissue essentially has three functions. First it does its normal job; e.g. it filters plasma or pumps blood. Second, by elaborating stress signals, it alerts the rest of the immune system to the presence of danger. Third, by displaying its own antigens in the absence of co-stimulation, it induces tolerance to itself. deleting T cells in the thymus:

There is a single necessary exception to the three laws. It occurs in the thymus and is designed to delete T cells able to recognize dendritic cells. Since thymic dendritic cells are perfectly capable of providing co- stimulatory signals, we borrow from Lederberg and propose that thymocytes pass through a developmental stage in which they are tolerizable but not yet activatable, a stage in which the pathways for receipt of second signals are not hooked up. Thus, as an exception to the second law, thymocytes at this stage would be unable to receive second signals from any cell, including professional APCs. Any thymocyte recognizing the normal surface MAP (MHC/antigen profile) of a dendritic cell would thus be eliminated by the receipt of signal One without signal Two.

Having finished maturation, the remaining T cells exit into the periphery and begin to circulate. Let's return to the hypothetical virus infection to see how the system works. activating and tolerizing T cells in the periphery

Among the virgin T cells circulating through the lymph node draining the infected site, some are specific for viral peptides (V) displayed by the activated Langerhans cell and others, having had no opportunity to become tolerant of non-thymic tissue antigens, will recognize peptides from the skin (Sk). Let's look first at the cells specific for V.

Stimulated by signal One and Two from the activated Langerhans cells, they multiply, become effector cells and circulate out of the node looking for virus infected targets. Each killer destroys a few infected cells and then reverts to a resting state and drains back into the local lymph node, where, if the infection is still going on, it will be re-activated. This cycle of activation into effector cell and reversion to resting cell should continue until the infection is cleared and there are no longer any activated APCs presenting the viral antigen. The experienced cells, or at least a selected few of them, will now recirculate as resting memory cells, waiting for the next encounter3. The autoreactive Sk specific T cells are also stimulated by the activated APCs, but because Sk is everywhere on normal skin, they are ina critically different situation from the killers specific for V. While the virus specific killers accumulate at the infected site and return to the original lymph node, where they are likely to meet a virus- presenting APC, the Sk specific killers will distribute all over the body. Each will kill a few skin cells and then drain to a local lymph node. Unless that node is draining an infected site, the Sk specific killers will not be reactivated. They will exit the node as resting memory cells and recirculate from blood to tissues. Encountering Sk again on normal skin cells, and thus receiving signal One without signal Two, they will die. Any Sk specific cells returning to the original infected area will also be deleted by normal skin cells after the infection is cleared.

A critical point to remember here is that killer cells induce apoptosis in their targets. Although the chromium release assay gives the impression that targets die by membrane disruption, DNA fragmentation actually occurs hours before the membranes disintegrate. In vivo, these apoptotic cells are probably scavenged long before their membranes fall apart. Consequently death induced by killer cells does not signal local APCs to perpetuate the immune response. Only real danger can do that.

Thus, although both autoreactive and virus specific killers follow the laws of lymphotics, their fates differ because their antigens are critically dissimilar. Autoantigens are continuously expressed by healthy cells incapable of delivering signal Two. By sheer numbers and persistence, they ensure that circulating autoreactive cells should regularly be deleted. Of course the efficiency of deletion will depend on the size of an organ and the rate at which lymphocytes circulate through. Large organs, and those that have minimal blood-tissue barriers, like the liver, should tolerize quite effectively whereas small and/or well barricaded organs (brain, pancreas?) should induce deletion at a much slower rate. part 4: lactation and puberty, tumors and transplants.

Life's changes: An organism in which tissues tolerize for themselves has no problem with change. Take puberty. Though many organs and their hormones are affected, there is no associated lytic cell death and hence no activation of local APCs. Therefore any T cells that recognise the changes will simply die because of the first law of lymphotics. The newly lactating breast making casein and perhaps expressing casein peptide/MHC complexes on its surface will not activate casein-specific T cells, but, by offering signal One without signal Two, it will delete them. These are examples of the constant conversation going on between lymphocytes and other tissues. The tissues themselves, offering their antigens in the absence of signal Two, need not be static entities, stuck with an immune system tolerant only of yesterday's 'self'. They can afford to change.

Tumors: A newly arising tumor cell may express antigens not expressed by its normal tissue mates, but this is not enough to alert the immune system. There is no intrinsic difference between a rapidly dividing tumor cell and a rapidly dividing hematopoietic cell, gut cell or thymocyte. If it dies, it dies by apoptosis. It does not normally produce stress proteins nor activate local APCs and there is no particular reason why the immune system should be able to distinguish it from any other rapidly dividing cell type unless it becomes infected, stressed or otherwise necrotic. Consequently, as it grows, any tumor unable to deliver signal Two should induce deletion of tumor specific T cells.

Sometimes, however, tumors spontaneously regress. The cause may be simple. Should a melanoma, for example, be traumatized by viral, bacterial or physical insult, the local APCs would become activated, capture the tumor antigens (as well as normal melanocyte antigens) and present them to passing T cells in the draining nodes. Any tumor specific T cells that had not yet been deleted would become activated and begin to destroy the tumor. If the melanoma were small or if the T cells were repeatedly activated (e.g. by a chronic, remittent or long lasting viral infection), the melanoma would be destroyed. In some cases, the T cells might also destroy normal melanocytes expressing similar antigens, leading to the vitiligo occasionally found with spontaneous melanoma regression. This view fits with the finding that T cell clones isolated from such patients can kill melanoma cells taken from HLA matched patients that have not rejected their tumors, showing again that it is not lack of antigen expression that prevents tumor rejection.

A similar scenario describes the finding that an injection of heavily irradiated tumor cells transfected with GMCSF can evoke protection against the untransfected tumor. This treatment has two important features. First, though radiation death is usually apoptotic, an injection of large numbers of dying cells may well overload local scavenging capacities, allowing some of the tumour cells to disintegrate before they can be cleared, and thus signalling local APCs. Second, GMCSF enhances the well being of dendritic cells in vitro and may do the same in vivo. The combination of dying cells and APC-enhancing cytokine would thus provoke activation of tumor specific T cells that could now kill a certain number of living, untransfected tumor cells. Soon, however, the killers would rest down and, in the absence of more danger signals, would not be re-activated, explaining why the protection generated by the transfected tumor can be overcome with a large enough challenge of normal tumor cells. A prediction would be that the larger challenge doses could be dealt with by repeated immunization with irradiated transfected tumor cells.

The same reasoning may also explain why tumor infiltrating lymphocytes (TILs) work only in some cases. If TILs are removed from the tumor, activated in vitro and re-injected along with a source of IL-2, they may well return to the tumor and engage in a round of destruction. If the tumor burden is small, this may be enough to destroy it. However, if there are too many tumor cells for the injected T cells to destroy in the first round, the T cells will go through their natural cycle of resting down and waitng to be re-stimulated. Without a source of activated APCs, they will remain in the resting state, now to be tolerized by recognition of tumor antigens in the absence of signal Two.

Direct trauma to the tumor itself is not the only way in which specific immunization might occur. Activation by any crossreactive antigen should be enough, and this suggests that getting rid of tumors might be a simple matter of immunizing repeatedly with appropriate antigens. The immunizations could merely be injections of disrupted tumor cells in adjuvant. The injections might consist of tumor cells that had been infected with a lytic virus or an inducible death gene, with or without the addition of APC-enhancing cytokines. One could paint visible surface tumors (repeatedly) with noxious substances, calculated to kill off enough tumor cells to activate local APCs. It might help to add carrier determinants to activate helper cells, or to inject professional APCs fed with the tumor cells (or their antigens, if known). In any case, repeated immunizations would be necessary. Though a single immunization would initiate immunity, the response would soon die down for lack of repeated stimulus. Even if some of the tumor were destroyed by killers activated during the first immunization, this apoptotic death would not maintain the response. The tumor meanwhile, like any other tissue expressing signal One without signal Two, would induce deletion of tumor-specific memory cells as they rested down. To be effective, therefore, immunizations should be repeated until the last vestiges of tumor are gone. In addition such immunizations should be done early, while the tumor is small and has not had time to delete a large number of tumor specific T killers, or alternatively they should be held off for a while after removal of the main tumor mass in order to give the thymus time to repopulate the periphery with new tumor specific T cell populations.

Transplantation: Graft rejection is usually considered a function of signal One; the grafted tissues are rejected because they express foreign antigens. However, a great deal of evidence suggests that signal Two is also critical here. For example, ovaries, kidneys, thyroid and pancreas can all be successfully transplanted if they are first carefully purged of passenger APCs. If fresh donor type APCs, are given at the time of transplantation or soon after, the grafts are rapidly rejected, showing that their failure to act as immunogens is not due to a lack of expressed antigen. More importantly, the APC depleted grafts gradually induce tolerance to themselves such that the recipient mice are eventually unable to reject their grafts even when immunized with fresh APCs. This tolerance is not a generalized suppression because the mice are still able to respond to donor type spleen cells, showing that the tolerance induced by thyroid or pancreas, for example, does not extend to cover the MAPs of other tissues such as spleen.

Livers seem to be especially extraordinary, since they are often accepted even without APC depletion. Though a rejection crisis is initiated, it wanes after a few weeks and eventually dies out altogether, leaving the recipients tolerant. This is tough to explain from a self-non-self viewpoint. Why should liver grafts, replete with APCs, be tolerogenic when skin or heart grafts are rejected? The reason is that livers are big, they do not have strong vascular barriers, and they regenerate; consequently they are a large and easily accessible source of signal One. The process begins when APCs in the liver become activated by the surgical trauma, home to the local (recipient) lymph node and activate donor specific T cells which then migrate to the organ and begin killing all donor cells, including hepatocytes, endothelial cells and APCs. After a while, all the bone marrow-derived APCs will have died, but the regenerating liver lasts longer, continues to offer signal One without signal Two and, rather quickly because of its size, deletes all the relevant T cells. Should any hematopoietic stem cells remain, they can now take up residence, creating the chimerism that is often (but not always) seen in recipients of liver grafts.

An apparant exception to the idea that tissues induce tolerance to themselves comes from transplant recipients who suddenly stop taking Cyclosporin A and reject their grafts, even though they may have had them for over 20 years! According to the model, these patients should be tolerant. The problem here is not the model but the drug. Since CsA blocks signal One not signal Two, the first law of lymphotics cannot operate and no deletion can occur. This is nicely illustrated by an experiment in which rats were given allogeneic livers under a two week course of CsA and, at various times later, were immunized with donor-type skin. Skin grafts given 0-4 weeks after CsA withdrawal were rapidly rejected and also stimulated rejection of the livers. By eight weeks, the livers had tolerized for themselves and, though the skin grafts were rejected, the livers were not6. The authors determined that there were three stages of reactivity. In the first, during CsA treatment, alloreactive T cells were unable either to respond or be tolerized. In the second "transitional" stage, when the drug was withdrawn, the T cells were slowly tolerized unless activated by a new source of APCs. In the third stage, the animal had become solidly tolerant of the graft. They wondered whether the transitional stage was delayed by the CsA treatment. Today, we would answer yes. Tissues cannot induce tolerance to themselves in the presence of a signal One blocker. This may also be why irradiated rodents whose immune systems are allowed to regenerate under the cover of CsA suffer autoimmune symptoms when the drug is discontinued.

The solution, of course is to find drugs that block signal Two without obstructing signal One, and some steps have already been made in this direction. For example, in two back to back studies, soluble CTLA4 evoked long lasting tolerance of xenogeneic islet grafts. It also inhibited antibody responses to KLH and SRBC though, to the perplexity of the authors, not permanently. I would suggest that the treatment had the same effect in both cases. It blocked signal Two and stopped the immune response. In the case of the islets, the grafted tissue stayed in place, continually tolerizing for itself by expressing its antigens in the absence of signal Two. However, neither KLH nor SRBC persist for long in mice and, in their case, a temporary ablation could not generate long lasting tolerance.

I am less certain of protocols in which antibodies to adhesion molecules are used. They run the risk of inhibiting signal One, in which case, like CsA, they may stop rejection but nevertheless not allow for the induction of tolerance. An optimal combination might be to give blockers of signal Two along with a source of stem cells in addition to the grafted organ. In this case, the organ itself should tolerize any mature T cells and the stem cells should generate the necessary chimerism to induce tolerance of newly maturing T cells in the thymus, a protocol occasionally mimicked by liver grafts, accounting for their success.

Although the danger model does eventually produce a rough definition of self, the difference from that made by self-non-self discrimination models is more than just semantics. An organism in which the availability of second signals governs immunity and tolerance needs no static definition of self. Its immune system is not a separate army protecting (and regulating) the rest of the organs of the body, but an extended, highly interactive network making its decisions on the basis of input from all bodily tissues. This is a flexible immune system that changes as the organism changes, that welcomes the presence of useful commensal organisms and allows the passage of harmless opportunistic ones. In short this is an immune system that exists in harmony with both its internal and its external environment.

Downloaded from: