Moderator Kenneth Schaffner - 11:41am May 11, 1997 

What is (are) the most fruitful model(s) to explain immunologic tolerance? What, in your view, are the key recognition elements, cell types, interactions, time period(s), and site(s)? Can you relate the model(s) back to a general principle, such as self-nonself distinction, integrity preservation, danger, etc.? 

Rationale: 

There seems to be a consensus (of sorts) in this group that the self-nonself distinction has and can be useful but may be limited. It may be limited because of our language (and conceptual associations), because it is not fruitful in application to autoimmune diseases, transplantation biology, or cancer, or because it may not help in generating novel experiments.  

Even if self-nonself is a key general principle, all seem to agree that we need to move to more specific descriptions of what the immune system does in current vertebrates (as conceptually helpful as the discussion of "thingies" may be) and what turns it on and turns it off (forgive the first-approximation dichotomy). This move may or may not involve a unitary theory of tolerance. Perhaps, from this more specific level, we may be able to look back (up?) and see whether the self-nonself distinction guides us well, or whether some other general principle (integrity, danger, etc.) is better - or even if all such general principles are too abstract to be currently useful.

Rod Langman - 7:20am May 12, 1997 (#1 of 47)  

Everyone is looking to take a shot at self-nonself discrimination, so I'm going to give you a big sitting duck as a target for shooting practice. We can call it the webbed associative antigen recognition (AAR), or wAAR for short (as the mood strikes, you can shorten this to "war"!). 

In the beginning, there were defense mechanisms like the alternate complement pathway and perforins, which were coupled to germ-line-selected recognition devices that picked on one or another organization of biochemical building blocks to find a marker that would indicate whether the biodestructive response should be activated or not. Then, as the pathogens became more numerous and diverse, so the recognitive system that told the difference between host and pathogen needed greater and greater specificity, and with more specificity per recognition device, the more different specific devices were needed. These are the non-immune defense mechanisms of invertebrates. Then, when the germ-line selection pressure on any one specificity element was so rare that it was often lost by genetic drift, and when the specificity elements were faced with diverse host antigens, the solution needed to go to the next step in protection was a somatically selected specificity system that would allow individual cells to be kept or killed instead of whole animals being selected upon. The twin conditions of needing to recognize as many pathogens per receptor and as few self components per receptor were the driving force for immune specificity. If recognition of pathogens were all that was needed, a universal glue would have been perfect, but because the host could not be made to go the same way as the pathogen, an adequate degree of specificity became essential. 

As the specificity per receptor increased, so increased the number of different receptor specificities needed to cover the pathogenic universe. However, the bigger the mutational substrate, the more likely became germ-line mutations to antihost, and there was an impasse. The only way to increase the size of the repertoire, and hence the number of different pathogens that could be recognized without self recognition and destruction, was to somatically select on the specificities so that cells, not whole individuals, live and die.  

A somatically selected repertoire is generated in a state where each cell has open to it two choices, either to live or die, and, just as in the whole-animal case, self recognition led to cell death and failure to recognize self led to cell survival. This is the point at which the Lederberg model (1959) can be introduced because it dealt with how a cell would make a "once-in-a-lifetime" decision regarding self and nonself. Under an updated Lederberg model, all cells would begin with one choice - see antigen and die. Among the survivors of this carnage would be cells that failed to see self; these would then be allowed to differentiate into biodestructive effectors such that, upon encounter with antigen, they would reflexively execute their destructive effector reactions. Unfortunately, animals needed to live quite a bit longer than any given cell, and so the regenerating immune system had to repeat this self-nonself discrimination. However, in the adult, when a pathogen was present, there was no way to tell whose genome, the host's or the pathogen's, made any particular protein, and the Lederbergian cell that was born during an infection would treat the epitopes of the pathogen as self. 

What are some of the major difficulties with a Lederberg-style model? Perhaps the simplest place to start is at the point where nonself antigen initiates an immune response. Parenthetically, it seems safe to assume that a constitutive expression of immunity is not going to last long in evolution. Thus, a cell is hit by antigen and swings into an active protective mode, but when the antigen is gone, or at some fixed time, the cell has to revert to a quiescent state. This is a crucial decision because if the cell reverts only after antigen has been eliminated, one missed persistent self antigen would be enough to kill the host, or one mutated receptor that produces antiself would mean inevitable death. But, if the cell reverts after a decent time interval, then any antigen that can hang around for more than one tick on this reversion clock is going to be able to trick the immune system into thinking it is a self component. Neither of these choices looks promising for explaining the present-day immune system.  

"Wait," I hear, "how about introducing something new to take care of these problems, some kind of fail-safe for the mutants, or slow-growing bugs? Can we have a self-nonself discrimination for the fast-growing bugs and another for the slow growers?" I suspect not, because without some kind of historical measuring device the life span of a bug cannot be predicted, only measured after the fact. Once a historical learning process has to be invoked, it would be germ-line selectable to measure the life span of the host rather than that of all the different bugs. This is the next step in developing this game of wAAR.  

At issue is going to be whether history remembers or forgets self. The Coutinho-Stewart-Bandeira approach would favor remembering self, whereas the AAR approach is to forget self and remember nonself. How does this work?  

A cell is born and it has open to it two choices: live or die. If the cell sees an epitope (above some minimum threshold concentration, of course), then it is sent on the pathway to death. This is true at all times, whether the immune system is days or years old. If the cell fails to see an epitope in some magic time period (a few days would be reasonable), then it undergoes an antigen-independent differentiation to become an effector that is an inductive, upregulating cell (i.e., a helper T cell). This antigen-independent differentiation to the effector state is unique to regulatory cells of the helper class. We like to call the class of newborn cell an "i-state" cell, meaning that its future is indeterminate. This i-state cell advances to an "a-state," which can be thought of as anticipatory but undecided. In the absence of any further interactions, the a-state cell decays and dies. In the presence of "help," it advances to the effector, or "e-state," and expresses its biodestructive effector function. Like all e-state cells, the Th (helper T cell) slowly reverts to the i-state, and in the absence of antigen there is a steady-state level of eTh and iTh established by the antigen-independent differentiation of iTh to eTh and the antigen-independent reversion of eTh to iTh.  

There is another pathway of iTh to eTh that is antigen dependent and requires the presence of eTh. This second pathway is very fast and acts so that small, almost catalytic, numbers of eTh generated by the antigen-independent pathway are able to rapidly promote corresponding antigen-specific iTh to eTh status, as might be required during an acute infection. These eTh, whatever their origin, catalyze the conversion of antigen-specific a-state cells to the e-state when the eTh and the a-state cell can be linked by recognition of the same antigen (hence the name "associative antigen recognition").  

This system represents a mechanism whereby a self-nonself discrimination can be made by the cells of the immune system using criteria for self and nonself that are useful only to the immune system. Any antigen that the immune system could not respond to (e.g., it is not available or is at too low a concentration to signal the i-state cell) cannot cause deletion (forgetfulness), and this explains how some genetically defined self antigens fail to cause tolerance because they could not signal the i- to a-state conversion. Self is defined by the repertoire of antigens that the iT cells can be exposed to in some shortish period of time (a time less than the time taken to make the transition from i- to e-state by the antigen-independent pathway). The critical historical period occurs when the very first embryonic Th cell arises in the i-state, before any eTh could have been produced. All antigens present at this time would perforce cause cells to be eliminated. It is this particular set of antigens, present at early embryonic times, that becomes the immune system definition of "self"; all else is nonself. Nonself antigens are characterized as being recognizable by specific eTh at the time of entry into the immune system.  

As an aside, this schema attaches no special role to the thymus in determining when a cell can undergo the antigen-independent conversion from i-state to e-state. In short, central and peripheral tolerance are misleading boundaries.  

A critical question is whether no new self antigens are revealed to the immune system after the initial period of embryonic absence of eTh - that is, self antigens that would cause the death of the animal if subject to immune elimination. Only if such antigens can be directly demonstrated is there a need for some additional means of establishing tolerance to a self component in the face of a preexisting population of eTh. There may be excellent reasons for the immune system to develop a means of becoming unresponsive in the face of a preexisting eTh population. For example, if a continued immune response to a persistent infection is likely to overrun the limited resources of the immune system (e.g., devoting too many cells to the response) when the infection cannot be eliminated, maybe an immune response is not helpful. After all, if the infection has persisted for some time and the host has not died, the nonself agent cannot be too lethal, and it may even prove beneficial provided some kind of tolerance was developed. One other special form of unresponsiveness is the switch in the class of an immune response from, say, antibody to cell-mediated immunity. When only one class of immunity is measured, a switch in class could be interpreted as apparent tolerance. Finally, the immune responses that rid effete cells and proteins are directed at degraded self components, and because degradation is slow (often taking months), these antigens would be classified as nonself by the scheme described here. 

Obviously this is a large sitting duck, with many missing points and explanations. So, go at it, give it your best shot, and let's see what is left standing 24 hours later. 

Doug Green - 8:13am May 12, 1997 (#2 of 47)  

Well, now we're to the juicy bit - can we explain how the immune system does what it does? What I'm going to try to do is offer a first pass at such an explanation. Naturally it will be full of holes, and there will be lots of things that are left unexplained. I don't think that it contains anything that isn't within the domain of accepted wisdom, although I'm sure that many will prefer to change it here or there. My goal, though, is to show that not only is it possible to explain the function of the immune system (within a reasonable first approximation), but we can also do so without invoking any principles or processes that cannot be rigorously defined at a mechanistic level. I expect that most who read it will feel that sense of deja vu that says, "There's nothing new here, we know all this," and that's exactly the point. We just needed to put it all together to see where it gets us. I think it gets us pretty far. 

I'm going to try this with a minimum of technical stuff as well, but in the interest of space (and time), I might use a bit of immunologic shorthand (sorry in advance). So here it is. 

Our bodies are tubes. The boundaries of the tube, separating us from the "outside" world, are the skin and the mucosa (which includes the gut and the lungs). Anything that is going to infect us must do so by penetrating one of these barriers. The immune response begins when one of these barriers is broken. The damage itself triggers a local reaction, and it is this local reaction that controls any potential infection. Like politics, all immunity is local. 

Breaking the barriers by physical means actually disrupting cells, and the cells themselves respond to this by releasing factors (cytokines). In addition, mechanical impact can also directly affect cells that line all these surfaces, the mast cells, and trigger them to release granule contents in the immediate area of the impact. These contents, and the cytokines, act on the closest capillaries and induce contraction of the endothelial cells to allow fluid to enter the site. The fluid carries complement molecules. Within seconds of even a sterile insult, the site is washed in fluid from the blood. Meanwhile, the cytokines induce expression of adhesion molecules on the endothelium, which allows binding of passing neutrophils. Some of these will exit the blood at the gaps caused by the contracted endothelial cells. These early events can be greatly increased if complement is activated by the presence of bacteria in the site of damage. Similarly, bacterial products will also act on nearby cells to increase the release of cytokines that amplify these effects. (If we wish to consider a vaccination with only a protein, then adjuvants or even scarification will trigger the amplification at this site. If we wish to consider a viral infection, the response occurs after the virus has damaged cells as a consequence of its lifestyle.) The neutrophils recruited to the site will eat any bits of damaged tissue as well as any bacteria (especially those bound with complement). The cells eventually die and are cleared by other phagocytic cells that enter the site more slowly - the macrophages. 

This local inflammation isn't the end of it, of course, but it has produced a critically important change in some cells of the tissue, the dendritic cells (for review, see Watts, 1997). The immature dendritic cells constantly take up and cleave proteins, and these fragments (regardless of source) bind to class II MHC molecules present in the endosomes and are expressed on the surface. The MHC-peptide complexes are turned over at a fairly high rate. When the dendritic cells are exposed to inflammatory cytokines (IL-1, TNF) and/or bacterial products (e.g., endotoxin), they mature and change their lifestyle. First, they stop expressing so much new class II MHC (and it's no longer in the endosomes), and any complexes that have formed now persist for a much longer period on the surface (days rather than hours). The cells also now express surface molecules that are capable of providing a costimulatory signal to T cells (e.g., B7-2/CD86). These dendritic cells basically display a peptide "snapshot" of the proteins present in the site where the inflammatory response occurred. The dendritic cell then drifts with the lymph to the lymphatic drainage and toward the nearest lymph node, where millions of CD4 T cells are waiting. 

This large population of CD4 T cells contains a very large number of receptor specificities, but not every one that's possible. We've been over the idea of negative selection, but it's worth going over it again, because there is often some confusion about it. T cells develop in the thymus, and this goes on throughout our lives (not only while we ourselves are developing). As it develops, the T cell reaches a stage where it expresses a unique T-cell receptor (TCR). If that receptor contacts its ligand (MHC-peptide) on a dendritic cell in the thymus, the activation signal stimulates the T cell to kill itself. It doesn't matter whether the ligand is derived from our own proteins or proteins encoded by a potential pathogen; the cell will die. (TCRs preferentially recognize proteins presented on the platforms formed by MHC molecules. This is very useful, because proteins only come from living things, and if a protein is present, it's a pretty good guess that something living is or was present.) The thing is, most proteins from our environment are not always present in the body, and when they aren't, then T cells with TCRs capable of recognizing them will mature. So if my developing T cells that might be capable of recognizing a flu virus die in response to flu viral proteins present in my thymus because I have a flu, I will still be able to make an immune response to the virus, thanks to the T cells that matured last week when I didn't have the flu. It's for this reason that I like to say that the immune system responds to things that are only sometimes in the system and not to things that are always there. 

What about those molecules that are present in our bodies but not in the thymus? Many of these the immune system doesn't know about unless they appear for some reason (and it treats them as things that are only sometimes present). But it's also possible that some things are readily available for recognition in the periphery but not in the thymus (we don't know this, but it's possible). Perhaps the freshly generated T cells that leave the thymus retain the capacity to die for a day or two while they sail around the body checking things out. It's difficult (though not impossible) to test this, because we need a way to know which are the freshly generated T cells. I'm sure, though, that it can be done. 

Another point that is often raised here is that the proteins that we produce in our bodies are always changing, and therefore the immune system has to know which of these it can and cannot respond to (without causing disease). I previously mentioned the possibility that this simply might not be true; there is no evidence that I know of that new proteins arise during maturation that can elicit immune responses in less mature, syngeneic individuals. But let's say they do. Here there are still a couple of ways for negative selection to play a role. Certainly, if a new "self" protein is synthesized and readily available to the immune system, then any newly developing T cells that might recognize it will die, and these won't contribute to any potential problems. So we only have to concern ourselves with those T cells that were generated previously but that now come in contact with the new protein. Let's come back to those in a bit. 

Our dendritic cell has come to the lymph node, and if it contacts a T cell bearing a TCR that binds to its specific ligand on the dendritic cell (MHC plus some peptide that isn't normally there), then this generates a signal. In the presence of the costimulatory signal (generated by B7 interacting with CD28, for example), the cell will respond by producing cytokines and expressing cytokine receptors, and this will lead to (1) the proliferation of the cell bearing the specific receptor and (2) its loss of adhesion and entry into the lymphoid circulation, where it will eventually enter the blood and exit again at any site at which an inflammatory response is occurring (and thus adhesion molecules are expressed on the capillary endothelium). It thus can find itself back at the original site where the ligand was first generated (and where more class II MHC-expressing cells are displaying the same ligand). The resulting response amplifies those mechanisms present at the site, such as the activity and function of macrophages. 

So, coming back to our problem of the new "self" protein, this will be processed into peptides and presented in class II MHC molecules just like any other protein. However, in the absence of tissue damage or an inflammatory site, there won't be any expression of costimulatory molecules on the class II MHC-expressing cells, and those T cells that might bind to the new ligand will do so without a costimulatory signal. In this case, the signal generated produces a state of nonresponsiveness in the T cell (we don't know whether this eventually leads to the cell's demise or whether it allows the possibility that the cell will eventually function only to produce inhibitory cytokines in response to this ligand). In any case, the function of these cells will be eliminated. 

Well, then, what happens if the new "self" protein appears and there's an inflammatory response at the site where this new protein is expressed? First, it's possible that this is just a bad thing to happen, and we're going to see an example of autoimmune disease. However, I suspect that there are additional backups in place to help to limit this. Before the onset of reproductive maturity, there is often a tremendous drop in lymphocyte number owing to, for example, glucocorticoid production. Similarly, pregnancy is a state that seems to be not only "immunosuppressive" but also anti-inflammatory, and again, it's possible that additional safeguards are in place to limit responses to "new" proteins (whether truly foreign or just new). Many such mechanisms have been described, and I think that the reason why none of them has been shown to be essential is because they are backups - the major mechanisms may well be of the sort I've outlined above. 

This is a very cursory overview of a difficult subject, but as I said, it's only a first pass. But maybe the reader will notice a couple of things. The process I've described, which might provide a reasonable facsimile of self-nonself discrimination, doesn't involve anything other than demonstrable cellular and molecular mechanisms - there is no need for the system to do much more than respond to cellular injury (which is arguably a very ancient sort of thing to do), monitor the development of a lymphocyte (the cell has to respond differently at different stages in its maturation), and respond in fundamentally different ways to TCR ligation, depending upon whether a costimulatory signal is present. I didn't even mention the word "antigen," and I only said "self" (in quotation marks) because that's the nature of this discussion, not because the term is needed. 

Of course, I also didn't say anything that we don't all know about. This is the emerging view of how the immune system works, and we've known about this for a number of years. I don't really feel that it's a theory, or a model, just an amalgam of what a lot of us already know. I've just said it all in one place. 

Zlatko Dembic - 9:33am May 12, 1997 (#3 of 47)  

The integrity protection/restoration principle is basically an "analytic/ synthetic" principle and not a discrimination principle. It is a process through which a combination of signals selects the outcome. This process aims to "synthesize" or restore integrity, which, perhaps, might not be complete (it is not perfect). The outcome that can be expected by each population of specific immunocytes that receive signal[1], [2], or [3] is listed (Day 2, #26):  

(1) Signals[1M + 2 + 3] (the activation of the destructive immune response to "parasites," graft rejection);  

(2) Signal[1] (deletion tolerance);  

(3) Signals[1 + 2] (autoimmune disease);  

(4) Signals[1M + 2 + 3 (- [2])] (one kind of peripheral tolerance [anergy?], tumors, symbiosis, downregulation of the active-destructive immune response, appearance of memory and autoimmunity; depending on time); and 

(5) Signals[1M + 3 (- [3])] (certain type of tolerance, tumors, also time dependent). 

Specific immunodeficiency would be a state when no cell is capable of getting signal[1]. The final outcome to any antigenic complex load would depend on the selection of the winning subpopulation due to availability of particular signal. 

According to the self-nonself discrimination principle, a decision by some external feature on what happens with signal[1] and signal[2] is required, and this poses a great difficulty for me. That is why it is so hard to explain the occurrence of the first primed T cell by this principle, and in the AAR model it had to be doom for every T cell; in other words, given enough time, every T cell would become activated by default. It looks attractive on first glance, but is there any evidence?  

Zlatko Dembic - 1:00pm May 12, 1997 (#4 of 47)  

Just one small clarification: 

(1) Signal[1] = TCR - peptide-MHC interaction, for T cells; BCR - antigen, for B cells; M = modulated; 

(2) Signal[2] = costimulation (including, for example, for T cells: CD28-CD80/86 interaction and cross-talk; for example, downregulation of signal[2] can include CTLA4/CD28 switch and CTLA4-CD80/86 interaction); T-cell help (for naive B cells and precursor cytotoxic T lymphocytes); 

(3) Signal[3] = disruption of integrity (including, for example, for dendritic cells: upregulation of costimulatory and cross-talk capabilities).