Ephraim Fuchs - 1:32pm May 12, 1997 (#5 of 47)  

The question being "What is the most fruitful model to explain immunologic tolerance?" the first thing that must be done is to define "tolerance." So let me state at the outset that I will proceed as if tolerance is either physical or functional inactivation of the responding lymphocyte population; that is, either clonal deletion or clonal anergy. This definition of tolerance is meant specifically to exclude certain phenomena, such as "immune deviation," which in a sense is just a measurement artifact. As an example, let's say that a normal, unprimed mouse responds to the placement of a skin graft from a nonidentical mouse by rejecting it. If we then do something to the animal so that it no longer rejects the graft but instead makes antibodies to the disparate histocompatibility molecules, I would not say that we have induced tolerance to the graft. This definition of tolerance is much more stringent because it requires the positive demonstration that the reactive cells have died or that the reactive cells do absolutely nothing in response to the antigen in question, not even make the as-yet-undiscovered interleukin-44 (obviously, this poses a problem for anybody who wants to invoke functional inactivation, or clonal anergy, as a mechanism of true tolerance). My definition of tolerance also specifically excludes the phenomenon of clonal ignorance, in which lymphocytes do not respond to an antigen because they have not seen it. This falls under the definition of "unresponsiveness." An animal that is tolerant of an antigen is unresponsive to the antigen, but not all animals that are unresponsive to an antigen are tolerant of the antigen. 

Having defined tolerance, it is now possible to summarize the decisions a lymphocyte must make. The two major decisions are (1) whether to be turned on or off in response to the antigen and (2) (contingent upon the cell having decided to be turned on) what effector class is the best at getting rid of the antigen. This being a discussion of tolerance mechanisms, I will confine myself strictly to addressing the on/off decision. There are, in fact, three possible outcomes of a T cell's encounter with an antigen presented by an antigen-presenting cell (APC): (1) The TCR for antigen is not sufficiently engaged and nothing happens, (2) the T cell is turned off (apoptosis or anergy, if it exists), or (3) the T cell is turned on and goes on to the effector-class decision. Now I am going to make the assumption that receptors function as analog-to-digital converters, in which TCR signaling yields an internal signal of "0" or "1" (this is not absolutely critical to the model). If receptors function as analog-to-digital converters, then two receptors will be needed to produce the three possible T cell-APC encounter outcomes as described. Thus we must bring in a second signal in addition to the specific TCR for antigen, and this will be called, in accordance with current terminology, "costimulation." Now it is very important to note here that costimulation may not be mediated by a single APC surface molecule, and in fact it is more biologically meaningful to talk about which cells are capable of delivering costimulation. 

Having established that a two-signal model is the simplest, most efficient means of achieving any of the three outcomes (on, off, or no change of state), I will now lay out a simple model to describe the T-cell decision process. For a detailed account, please refer to the 1994 article by Polly Matzinger. Here are the rules:  

(1) Antigen recognition in the absence of costimulation leads to tolerance, whereas activation results from antigen recognition in the presence of costimulation (Lafferty and Cunningham, 1975).  

(2) The decision of a T cell between activation and tolerance depends upon only three criteria: (a) the differentiation state of the T cell, (b) the type of cell that is presenting antigen, and (c) the activation state of the APC (to be defined later).  

(3) There are four relevant T-cell differentiation states: (a) immature thymocyte, (b) resting naive T cell, (c) resting memory T cell, and (d) activated effector T cell. 

(4) The requirements for signal[2] (costimulation) relax with progressive differentiation of the T cell: (a) Immature thymocytes are incapable of receiving signal[2] and so die when they recognize antigen, (b) resting naive T cells are activated only by appropriately activated dendritic cells, (c) resting memory T cells are activated only by dendritic cells, macrophages, and B cells, (d) activated effector cells are triggered to perform their effector function by receiving signal[1] alone, and (e) nonhemopoietic (tissue) cells cannot deliver signal[2] to resting T cells. 

(5) Activated effector cells perform their function for a small period of time, after which they die or revert back to the resting memory state.  

(6) Signal[2] (costimulation) differs for resting naive and resting memory T cells: (a) Signal[2] for naive T cells is induced on a dendritic cell when it binds something released from the cytoplasm of a stressed or dying tissue cell, and (b) signal[2] for memory T cells is induced on B cells (and possibly macrophages and dendritic cells) by resting memory CD4+ T-cell recognition of antigen.  

These rules make no reference whatsoever to the antigen, and this is the Achilles' heel of any theory that invokes the capacity of the immune system to make a self-nonself discrimination. Stated succinctly, the immune system does not discriminate self antigens from nonself antigens because antigen (signal[1]) is a necessary component of both the inductive (activation) and paralytic (tolerance) pathways. This is a point that I made in response to an article on the 20th anniversary of the Bretscher-Cohn two-signal model of lymphocyte activation (Fuchs, 1992; Fuchs, 1993). Now, if one wants to "save the appearances" of self-nonself discrimination, there are only two ways: (1) Propose the existence of a population of T cells that do not obey the strict two-signal rule or (2) propose that antigen regulates the delivery of signal[2]. The first way was proposed originally by Lederberg (1959), and, strangely enough, has been resurrected by Rod and Mel Cohn in their proposal of a class of inducer T cells that undergoes an antigen-independent differentiation from the i-state cell to the e-state (Langman and Cohn, 1996). That is a critical departure from a two-signal model. The two-signal model states that antigen (signal[1]) is absolutely required for induction of the T cell, either to death or to activation. I found Rod's proposal somewhat ironic in light of the fact that Peter Bretscher and Mel Cohn came up with the two-signal model (Bretscher and Cohn, 1970) to correct the flaws of the Lederberg model, and now Rod and Mel Cohn are back to invoking a Lederberg-like temporal model to correct a flaw in the two-signal model (the "primer" problem of who delivers the first signal[2]). The second way, proposing that antigen regulates the delivery of signal[2], was invoked in a way by Charlie Janeway (1989) when he postulated that costimulation is induced by microbial products such as lipopolysaccharide (LPS), which frequently accompanies new and dangerous antigens. This model has trouble with the immune response to viruses, which do not carry any molecular signatures of "nonselfness." It also runs into serious difficulties when trying to explain my finding that B cells activated by LPS still induce tolerance in naive T cells specific for an antigen presented by the B cells.  

The critical, defining feature of the danger model pertains to what turns on a dendritic cell to become an immunogenic (costimulatory) APC for a naive T cell. Polly Matzinger and I propose that danger signals are emitted by cells that are stressed or dying nonphysiologically. This model accounts for the requirement for adjuvants in generating immune responses to non-noxious antigens. Adjuvants provide the necessary tissue damage, thereby causing the liberation of danger signals. The model also accounts for the immune response to viruses, as lytic viruses will cause the release of danger signals that are normally kept away from dendritic cells. Now one might say that many viruses induce death by apoptosis. However, if the rate of apoptosis is high, the scavenging capacity of intraorgan phagocytes may be exceeded, and cells that are not scavenged promptly eventually rupture and liberate danger signals. Incidentally, this is the basis of the chromium-release assay of cellular cytotoxicity - even though targets are killed by apoptosis, there are no scavengers to gobble them up, and the killed cells eventually ooze their contents.  

Doug was concerned about a backup mechanism to provide for tolerance to tissue-specific antigens that are first presented to the immune system in a "dangerous" context (#2). The laws outlined above provide such a mechanism. Indeed, antiself effector cells would be activated, but they will eventually rest back down to the resting memory stage. At this point, if they go out into the tissue and encounter their antigen on a parenchymal cell, they will be turned off. I should note that it is important that CTLs kill cells by the induction of apoptosis. This way, they will not provide constant fuel to the fire of an inflammatory response. 

Rod Langman - 2:45pm May 13, 1997 (#6 of 47)  

Rhetoric and history aside, I did not follow Ephraim's argument (#5) for dismissing the AAR model as it stands, stated clearly, I hope, here in this forum. To help me understand, let me try and restate your point of view. Because antigen is needed to drive both tolerance and immunity, antigen alone cannot inform the immune system whether it was the product of a self or nonself genome. We would agree on this and say that something else is needed to sort antigens into those that should be eliminated and those that should not. What is the missing thing, and how is it detected by the immune system?  

The Coutinho-Stewart-Bandeira-et al. school would say that the missing thing is all a matter of timing, that it is not who you are but when you were there. The way to tell a self thingie from a nonself thingie is to know how long it has been around. If it was there before the immune system got started, and it never went away, then it would start and sustain a set of suppressive reactions, so that whenever a new cell arose with receptors able to see a self antigen, it would be suppressed and held harmless.  

The Bretscher-Cohn-Langman school would use the same criteria of antigens that are prior and persistent being treated differently from those that are subsequent and transient with respect to the immune system. We have our arguments over how history is converted into a regulatory mechanism.  

The remaining variants can be lumped into a general case in which nonself is nasty, pathogenic, inflammatory, disintegrative, etc., whereas self is safe, with none of the nasty properties of nonself. Depending on which form of nastiness is used, different mechanistic details emerge to describe how the immune system uses particular sorts of information.  

The historians would have to accept genetic nonself agents sneaking in early being treated as self (i.e., not eliminated) and self antigens popping up late being treated as nonself (i.e., eliminated) as well as any genetic self component that went missing for an extended period being treated as nonself upon its return. In other words, self and nonself have not been defined by genetics but by the immune system. 

The others - let me call them "alarmists" for want of a better cute name (apologies to Doug on cuteness) - would have to accept that genetic nonself agents that sound no alarm are treated as self and become obligate tolerogens (like Bill Weigle's deaggregated gamma globulins in the absence of LPS) and that any genetic self component in the near vicinity of an alarming attack will be treated as nonself and eliminated. Key for these alarmist elements is that they are seen the same way in all individuals because they are properties common to many (all) pathogens, etc., making their recognition germ-line encoded. In this way a self-nonself discrimination is made without reference to antigen. This avoids, as you so succinctly put it, the "Achilles' heel of any theory that invokes the capacity of the immune system to make a self-nonself discrimination."  

Let me stop at this point before getting too lost and see whether a midcourse correction is needed, or whether maybe I need to start again. 

William O. Weigle - 4:19pm May 13, 1997 (#7 of 47) 

It's nice to see a change in the subject. It gives us a chance to discuss some of the more practical aspects of self-nonself recognition, or whatever the panel wishes to call it. I also agree with Doug's suggestion that we ought to begin dealing with real life and remind Rod that there is real life in some parts of California. Before dealing with the more theoretical aspects of self-nonself recognition, tolerance and autoimmunity, we should first discuss the merits of models of immunologic tolerance and how they apply to at least one of the popular theoretical models. The in vivo behavior of antigens after their injection is seldom considered by tolerologists. It seems to be the contention of many that most, if not all, antigens serve as either tolerogens or immunogens depending upon the animal species, how they are injected, immune state of the host, etc. In fact, there are very few endogenous proteins or carbohydrates that can induce a solid unresponsive state similar to that we enjoy to our own body constituents.  

Antigens such as viruses, red blood cells, bacteria, tumor cells, etc., are very complex antigens that are aggressively handled by the first line of defenses when they enter the body and do not readily lend themselves to the induction of immunologic unresponsiveness. The suppressed responses following the administration of such antigens are most likely due to negative immune regulation. For example, after injection into mice, keyhole limpet hemocyanin is rapidly cleared by the reticuloendothelial system; after 24 hours, approximately 1% of this protein remains in the body. Likewise, because of its size, within minutes after egg albumin is injected, it is found in the glomeruli of the kidneys and subsequently eliminated intact in the urine. These proteins are difficult if not impossible to induce tolerance to, even in the neonate (Weigle, 1980). Thus, workers using such proteins as tolerogens have reported tolerance in one set of the CD4+ T cells and responsiveness in the other (Burstein et al., 1992). These studies and many that have followed in the last couple of years have nothing to do with tolerance but merely are the result of preferential usage of one T-cell subset and downregulation of the other. This interpretation has become clear in recent studies involving the response of CD4+ T-cell subsets to parasitic infections.  

However, there are antigens that can be used in tolerogenic forms that can give us a wealth of information concerning self-nonself recognition and insight into autoimmunity. With the exception of peptides, there is no other group of antigens that as readily lend themselves to tolerance (that mimic self tolerance) as serum protein antigens. Most typical of such antigens are monomeric forms of mammalian gamma globulin. These antigens and most likely other serum protein antigens readily lend themselves to understanding cellular and subcellular events involved in the induction of tolerance. Again, the best rationale to explain their ability to induce solid unresponsive states that are complete and lasting is that they mimic the physicochemical properties and in vivo behavior of self components. After monomeric human gamma globulin (HGG) is injected, it equilibrates between the intra- and extravascular fluid spaces, coming into contact with all the potential antigen-reactive cells, and it persists in these spaces with a half-life of approximately 7 days.  

Although different quantities of antigen are required, tolerance is induced in both T cells and B cells (Weigle, 1980). The tolerant state results in the failure to generate responses by either the Th1 or Th2 subsets, as evidenced by the failure to induce B-cell help and proliferation and release of cytokines upon immunization and subsequent challenge (Romball and Weigle, 1993). It should also be pointed out that monomeric HGG is not aggressively processed in vitro by APCs and thus does not cause release of the cytokines capable of causing deliverance of the second signal to T cells. On the other hand, aggregated HGG (immunogen) is rapidly degraded into peptides by APCs, causing the release of cytokines (Levich et al., 1987). This unresponsive state is of a peripheral nature in that it does not require the thymus (Gahring and Weigle, 1989). The above approach enables us to come to some conclusions concerning peripheral tolerance but does not address the tolerance induced in the thymus. Others using immunoglobulins to selectively target APCs have also demonstrated a solid unresponsive state in both CD4+ and B-cell subsets (Eynon and Parker, 1993; Finkelman et al., 1996).  

It is my belief, as well as that of others in the field, that tolerance with these models as well as that with peptide antigens results from bypassing activation of APCs. The ability to inhibit the induction of tolerance to monomeric HGG by cytokines and generators of cytokines suggests that the failure to activate APCs is the result of failure to activate the cytokine cascade (Weigle et al., 1987). Thus, the second signal in T-cell activation results from release of cytokine after activation of APCs, and bypassing this second signal allows for tolerance. The ability to induce tolerance by peptides, which also do not activate APCs and result in release of cytokines, supports this contention. Thus, it is unnecessary to invent an effector T cell to deliver a second signal and then offer yet another hypothesis as to how the first effector T cell arrives. I imagine that the postulated requirement for the antigen-specific effector T cell for the second signal suggests a requirement for specificity. Specificity of this second signal through APCs and cytokines is that the cytokine signal occurs in the immediate microenvironment of the interacting cells. After diffusion (even a few angstroms) from the microenvironment, cytokine concentration is not sufficient to activate receptor on cells even in neighboring microenvironments. Thus, mixtures of monomeric HGG with immunogens would not be expected to interfere with the tolerance induction to the former. On the other hand, injections of polyclonal cytokine releasers (endotoxin) or large amounts of cytokines (IL-1-alpha and -beta and TNF-alpha) result in sufficient concentrations of the cytokine in the body, including microenvironments of the interacting cells, to supply the second signal and convert a tolerogenic event to an immunogenic one. Thus, the Bretscher-Cohn theory as applied to T cells by Rod is not convincing. On the other hand, a danger signal is also an inadequate explanation for a second signal to T cells.  

The above points raise questions as to the nature of APCs in tolerance induction and the relative importance of tolerance induced through the thymus vs. peripheral tolerance as it applies to embryo (neonate), young adult, and aged adult.  

Ephraim Fuchs - 1:54am May 14, 1997 (#8 of 47) 

I agree totally with Rod (#6) that antigens that are prior and persistent are treated by the immune system as "self," in that they do not evoke destructive effector mechanisms, and that antigens that are late and transient are treated as "nonself." I am putting quotation marks around "self" and "nonself" because I really do not believe in such terms; I think it is better to refer to antigens as those that either turn off or turn on lymphocytes that respond to them. Thus "nonself" antigens are transient precisely because the immune system was turned on to produce a destructive mechanism.  

I also agree completely that the history of the individual is crucial in determining how the immune system will respond to a given antigen. Rod mentions that genetic nonself that sneaks in early may be treated as "self" by the immune system, and genetic self that pops up late may be treated as "nonself." But how are early and late defined? Lederberg defined these in terms of the age of the individual. More modern models define these in terms of the age of individual lymphocytes. But are we really talking chronological age? Are lymphocytes equipped with Timex watches (apologies to our European contributors; perhaps Rolex or Audemars Piguet)? I would state that the time or experience factor is defined by the differentiation state of the lymphocyte. If an immature thymocyte is exposed to its antigen, it is turned off. If the antigen was present early enough during the life of the animal such that there is no immature thymocyte that is not exposed to the antigen, then a state of deletional tolerance will result, and the antigen will be treated as self.  

Now we turn to those genetic self elements that arise after the immune system is mature. Rod states that such antigens may be eliminated (i.e., treated as "nonself"), but the very existence of tumors bearing tumor-specific antigens is evidence that not all genetic self is eliminated. What, then, is the deciding factor here? I contend it is primarily the cell presenting antigen. Polly Matzinger has argued at length that B cells, because they present their idiotypes (unique peptides usually belonging to the antigen-binding region on immunoglobulin), should not be allowed to activate naive T cells, and she showed this experimentally. Being a true believer in the two-signal model, I reasoned that because B cells are indeed capable of presenting antigen, then if B cells can't turn on naive T cells, they must be turning them off, and this appears to be the case. Polly and I have argued that naive T-cell activation is the responsibility of the "professional" APC, but even when a dendritic cell presents antigen to a naive T cell, activation is not guaranteed.  

Rod goes on to talk about nonself agents that sound no alarm. What is alarm? Is it possibly a danger signal? He then goes on to say, "Key for these alarmist elements is that they are seen the same way in all individuals because they are properties common to many (all) pathogens, etc., making their recognition germ-line encoded." This, I believe, is in line with Janeway's theory (1989) of discriminating noninfectious self from infectious nonself, in which some microbial pathogens carry signatures of nonselfness. Such signatures may include LPS, which may activate a dendritic cell to become an immunogenic (costimulatory) APC. But the real problem for Janeway's model is the immune response to viruses, which carry no signatures of nonself. Here we would propose that there must be an endogenous danger signal, one that is made by the host to indicate that a virus is causing real trouble. That is, the host must be able to sound the alarm in response to the virus. We don't know what this molecule is yet, perhaps heat-shock proteins that, under physiological circumstances, do not contact the surface of a dendritic cell. But if a virus, or a late-appearing genetic self antigen, does no harm to the vertebrate host, no alarm bell is sounded because no abnormal cell death is occurring. If a virus that did no harm could spread among cells without evoking an immune response, if such a virus encoded genes that were beneficial to the host, and if such genes could pass from the soma to the germ line, then maybe vertebrates could pass on acquired characteristics to their offspring. Maybe Lamarck was right after all (Lamarck, 1984)?