Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Scand J Immunol. Author manuscript; available in PMC 2010 July 1.
Published in final edited form as:
PMCID: PMC2729654

On the critique by Colin Anderson of “A Reply to Dembic: On an end to the beginning of mis-understanding the immune response”

In order to organize this discussion, I presented a set of postulates that describe immune responsiveness [1]. Dembic argued that the role of regulatory T-cells, life long tolerance later in life, and effector cell-type switching, were unexplained [2]. I answered this critique [3], which drew a commentary by Colin Anderson [4] that I will address here. The thoughtful and probing criticism of my essay [1] by Colin Anderson [4] must be read before considering my reply here.

The broad framework

Let me begin with a general comment. I don’t think that I can answer Colin Anderson in a way that would satisfy him. The reason is that his arguments are derived from pure empiricism. In my mind, understanding the behavior of the immune system requires a synthesis (a conceptualization) from which observation is interpreted. Any theory that incorporates all of the varied mini-interpretations of observation would have to be wrong as not all of these mini-interpretations could possibly be correct. In addition, I appreciate his complaint of oversimplification, but I don’t understand what the problem is that he is trying to solve. I am aware that the immune response is complex so I am asking, how do you deal with it?

My answer was to “modularize” [1]. Each Module must have coherent rules of behavior that define it. Those rules can be used to guide the writing of a computer program. This latter will not only reveal any cryptic conclusions but more importantly would allow evaluation of the parameters of the Module (i.e., quantitative constraints). To this end I defined three Modules each with its own structure and logic.

  • Module 1: The structure and generation of the repertoire.
  • Module 2: The sorting of the repertoire.
  • Module 3: The coupling of the repertoire to effector class.

Under each of the three Modules I took what the state of our present knowledge appeared to be and tried to encapsulate it. If there is a missing component, my comment was, add it and we will discuss how to integrate the additions/corrections or reject them. We are trying to guide the construction of a comprehensive, predictive model.

The ARA Model under attack by Anderson deals with Module 2 only. It equates the sorting of the repertoire (a defined event) with the popular self-nonself discrimination. If now, we are all agreed that the repertoire must be sorted, then the Danger, Pathogenicity, Localization, Integrity, Morphostasis, etc. Models deal with Module 3, the choice and magnitude of the effector response. I did not imply their lack of value; in fact, I formalized the argument that they are key to analyzing Module 3 (Ref 1, page 377 Postulate 12). The constant request (or challenge) that I use the ARA model to explain Module 3, the determination of effector class, is misplaced.

I also pushed the role of Module 2 to its limit, namely that the sorting of the repertoire is both necessary and sufficient to explain the self-nonself discrimination, not the responsiveness-unresponsiveness discrimination when nonself is involved. This led to obscurantism so I defined tolerance and unresponsiveness as well as autoimmunity and immunopathology to clarify the interpretation of experiment (see later).

In his general commentary Anderson argues that Cohn “…….recognizes that evolutionary processes do not evolve to perfection, and yet seems to ignore the corollary that the mechanisms that have evolved act under quantitative constraints.

If I might detail my comment, evolutionary selection cannot operate to perfection in the eyes of the human who uses self-serving criteria of perfection (e.g., total absence of autoimmunity). Considered from an evolutionary point of view the selection has operated to an acceptable limit (i.e., evolutionary perfection) because that limit is no longer a significant factor in the procreation of the species. What we as self-centered humans are trying to do is manipulate the system to serve out goals (e.g. reduce autoimmunity and infectious disease to zero). I cannot see how “quantitative constraints” are relevant, or for that matter are a “corollary”. Further, I do not see how “quantitative issues” will “change the core of the ARA model” or make “integrity” a mechanism of “natural” tolerance. If my set of postulates limited to antigen-specific interactions [1] are “oversimplified” based on “quantitative constraints,” add a set that fills the gap.

The immune system responds to “thresholds.” The effector function of antibody is concentration-dependent, which implies that the response is T/B cell density-dependent. It must respond by reaching an effective threshold in a short enough time without debilitating the host either by autoimmunity or immunopathology. This gives us Protecton theory [5, 6] where an immune boundary and an autoimmune boundary are steady states defined by the aid of a computer program. The latter allowed the analysis of so-called “quantitative constraints.” They were not ignored, they were confronted for both B [5] and T [7, 8] cells. This having been said, I do not see how they impact on my set of postulates; nor is it clear how the postulates can be characterized as “all or none” unless it means a “threshold” for response.

Continuing the general commentary Anderson views my challenge as unjustified because neither Dembic or others have dismissed “the need for somatic selection in the self-nonself discrimination.”

When it is argued that the immune system is not interested in or concerned with a self-nonself discrimination, it means that the necessity to make a self-nonself discrimination was not a selection pressure in the evolution of immune behavior. We have argued that, to the contrary, this discrimination was a dominant selection pressure for specificity and repertoire size [5, 9, 10].

I have defined the “self-nonself discrimination” as the “sorting of the repertoire.” The sorting of the repertoire is necessitated because the effector output is biodestructive and ridding. The ARA model is designed to explain the sorting of the repertoire (Module 2). It makes no difference whether or not one gratuitously grafts Sorting onto Integrity or Danger, if the immune system is not concerned with the problem. This is why I put these formulations under Module 3 (see later).

A somatically generated random repertoire can only be sorted by a somatically learned or historical process. Somatic selection differs from germline-selection in that the output is not transmitted to the next generation. As I pointed out earlier, from an evolutionary perspective, the sorting of the repertoire is both necessary and sufficient to make a self-nonself discrimination [11, 12]. This argument is challenged by a large school of immunologists who operate in the danger [1315], cytopathicity [16], pathogenicity [17, 18], localization [19, 20], integrity [21], morphostasis [22], and context [23]/circumstances [24] framework. As a school, they contend either that the immune system is not interested in/concerned with a self-nonself discrimination (e.g., [21]) or that the solution lies in a sorting of the effector output not the repertoire [25]. As the sorting of the effector output is accomplished by a germline-selected mechanism, whereas the sorting of the repertoire must be accomplished by a somatically derived mechanism, I feel justified in my commentary that the “integrity model” has been placed in the wrong framework. As an example, to quote Dembic, “The immune system neither discriminates between ‘Self’ and ‘Nonself,’ nor it acts to confront ‘Danger,’; rather, it reacts to disruption of tissue integrity… …” [21].

That “it reacts to disruption of tissue integrity” may be true, but this is hardly a reason to reject that it discriminates Self from Nonself. If the mechanism of the self-nonself discrimination is claimed to be solved by a germline-selected process like “integrity” or “danger” and the sorting of the “adaptive” repertoire is only solvable by a somatically selected process, then my criticism that these are models of regulation of class (Module 3), not sorting of the repertoire (Module 2), remains valid. It is also appropriate for me when analyzing these models to consider their commonalities as a simplification not “oversimplification.”

If I interpret Anderson correctly, we now all agree that the sorting of the repertoire (Module 2) is obligatory, if autoimmunity is to be controlled at the evolutionarily selectable level. The question, then, is what model for the mechanism of sorting competes with the ARA model? The sorting of a somatically generated random repertoire requires a prior sorting of the antigenic universe into self and nonself. I have argued that this process can only be accomplished as a function of developmental time. There must be a window during ontogeny when the immune system arises in the presence of all self and no nonself. In order to maintain the state of “tolerance,” self must persist. This developmental time aspect of the ARA model is under constant challenge but no substantial argument against it has surfaced (see later).

As I have pointed out [2628], it is not unreasonable to argue that evolutionary selection operates at the level of the effector output which must be regulated to rid the target without debilitating the host. This, at the experimental level, means that one challenges a manipulated animal with an antigen (infectious agent, toxin, innocuous polymer, graft, etc.) and observes the output using a responsiveness-unresponsiveness assay (e.g., survival-death, acceptance-rejection of graft, serum antibody, etc.). This black box approach buries both the logic of the system as well as the mechanism because there are several levels at which the responsiveness-unresponsiveness, antigen-specific, discrimination can be made. I have tried to open the box and look inside because the extrapolation from an observation of unresponsiveness to a theory of “natural” tolerance can only be tested by opening the box. This divides the antigen-specific events into three Modules as discussed.

Lastly, whether an “algorithm is oversimplified” is determined by the accuracy and robustness of its output. That which appears to be missing can then be evaluated.

Specific comments

1 -“…….nowhere in Dembic’s article that he denies the need for a sorting of the repertoire.”

The logic of my comment was simple enough. If one proposes a germline-encoded mechanism for a self-nonself discrimination by assuming that all nonself is red, whereas all self is blue, then the adaptive repertoire need not be sorted. At the level of effector output, only a germline-selected device to detect red or blue is required.

Why didn’t evolution come up with that solution? Simply put, there is no physical or chemical property of antigens that the immune system can use to define them as self or nonself as classes. If, as I see it, the Integrity Model is concerned solely with regulation of effector class, not the self-nonself discrimination, then the comment of Anderson would have been valid. If not, then my extrapolation from Dembic’s Model based on logic stands. Sorting of effector function, not repertoire, based on the germline-encoded recognition of ‘color’ would have been adequate, were it possible. The immune system must discriminate between self and nonself, integrity, danger, pathogenicity, context, etc. notwithstanding. In the absence of a mechanism to sort each of the non-overlapping repertoires of Th, Tc and B, the individual would succumb to autoimmunity. The denial by Dembic [21] lies in the assumption that a germline-encoded recognitive system (i.e., Integrity) obviates the need for a self-nonself discrimination.

2 -Cohn’s “…….argument against Tregs role in the sorting is based on a qualitative ‘all or none’ argument…..”

Unresponsiveness is not tolerance [29], page 82, [30, 31].

“Unresponsiveness” is the observation that a manipulated animal will not respond to an antigenic challenge to which it would normally respond.

“Tolerance” is the theoretical extrapolation of the observation of unresponsivness to a mechanism for the sorting of the repertoire (i.e., the self-nonself discrimination).

In both cases there are many ways to render an adult antigen-specifically unresponsive but most of them are not extrapolatable to the mechanism of “natural” tolerance. My proposal is that Tregs (T-suppressors) act as a feedback mechanism that drives the system towards unresponsiveness at the level of Module 3. That feedback control is required, is known for many biological communication systems [32]. This postulated role avoids the indeterminacy raised by Anderson’s comment, “whether the small degree of unresponsiveness to the self-of-the-species is experimental unresponsiveness or true tolerance is an issue of debate.” T-suppressors (Ts) are Class II MHC restricted (oddly enough, no one has asked, why?) and express a random anti-peptide repertoire. If this repertoire is sorted to be anti-nonself then no question of a role in “natural” tolerance arises. If this repertoire is sorted to be anti-self then its role is by definition to establish and/or maintain “natural” tolerance. There cannot be, in this latter case, “a small degree of unresponsiveness” as the mechanism of tolerance; functional unresponsiveness to self (absence of autoimmunity) is required. If the Ts repertoire is anti-nonself then “a small degree of unresponsiveness” is meaningless unless it is translated into regulation of magnitude (i.e., a large degree of responsiveness). I might point out, as an aside, that a Ts anti-self population regulating “tolerance” would have to distinguish the unbound TCR/BCR antigen-receptor treated as a self-component from the same antigen-receptor bound to its nonself ligand. If it could not distinguish them (as is likely), the immune system would be rendered unresponsive to all nonself.

Are Ts sorted to be anti-self, anti-nonself or unsorted? That is the question to answer if one wishes to place them as a “failsafe” mechanism of “tolerogenesis.” Dembic assumes that the Ts have been sorted to be anti-nonself (specifically anti-commensals) (see later). We agree then that they cannot play a role in sorting the repertoire. Normally Ts-activity is regulated to allow a sufficient not excessive effector response, the mechanism of which is a valid question but not a test of the validity of the postulate (see refs. [12, 33].

3 -A new hypothesis: The need for Tregs (T-suppressors) is driven primarily by the ability of low avidity interactions with self-peptide MHC complexes to cause immunopathology.

While I have not seen the developed theory, it would have to deal with the following points. Consider the behavior of a given single TCR with high affinity for a nonself (ns) ligand (Pns) and a low affinity for a self (s) ligand (Ps). This is what I refer to as the Standard Model of the TCR.

—If the TCR is of too low affinity for Ps to be negatively selected (absence of Signal[1]) then it would have to be of too low affinity when induced by Pns to be effective in attacking self-cells expressing Ps (autoimmunity). After all, most of us deal with our antigen loads (Pns) without succumbing to autoimmunity (attacking our self, Ps). Presumably, the role of a TCR conformational change trigger (Signal[1]) is to keep a concordance of affinity between inducibility and ability to mediate effector function. If the response to Pns is spilling over to affect innocent bystanders (immunopathology) then my proposal for the role of Ts is the only extant solution.

—There is no way that the random repertoire of Tregs (Ts) can determine the affinity of the defensive T-effectors that it is postulated to regulate, particularly since antigen-specificity would require associative (linked) recognition by Ts of peptides derived from the given self antigen. Under the Standard Model, if the Ts, like all other T-cells are high affinity anti-Pns and low affinity anti-Ps, any antigen-specific regulation by Ts anti-Ps of low affinity defensive T-cells anti-Ps would be marginal, if existent. Antigen non-specific regulation would be counter productive in this context. My view is that there is no self-antigen triggered immunopathology without prior autoimmunity (breaking of tolerance) in which case any concomitant immunopathology would be a second order problem.

—In spite of its popularity, there is no reason to assume the Ps is acting as a specificity element during positive selection. More likely it is functioning as a structural element necessary for the stability and conformation of the presenting MHC-encoded restricting element. Only the allele-specific epitope need be recognized as a determinant of positive selection. The very assumption of a TCR of high affinity for Pns, positively selected by low affinity for Ps, needs rationalization if it is to be credible [34, 35].

—For those who insist on Tregs (Ts) playing any role in the self-nonself discrimination (i.e., protection against autoimmunity) it would be informative to have a prediction of what happens to the postulated Ts anti-self cell when one breaks ‘tolerance’ to a given self-component. The breaking of tolerance (autoimmunity) converts a self-antigen into a nonself-antigen by inducing eTh that recognize it. A Ts-repertoire that was sorted to be anti-nonself would be blind to a converted self-component and, therefore, would have no role in protecting against autoimmunity. If the Ts-repertoire were anti-self then breaking tolerance requires either specific deletion of Ts anti-the-given-self or overriding of its inhibitory activity. If one argues that the induced effector T-helper level when breaking tolerance overrides the inhibitory Ts-level, then the relationship between Ts and Th would be that which I have postulated for regulation of the response to nonself, applied, in this case, to self. As the Th-repertoire is known to be sorted to be anti-nonself, I leave it to the reader to define the steps involved in overriding, antigen-specifically, a Ts anti-self activity that maintains tolerance, by the Th-anti-self population extracted from the autoimmune boundary by induction when tolerance is broken. This is a food-for-thought exercise.

Anderson points out that there is “……no compelling theoretical or empirical argument to dismiss either of these hypotheses” (Ts regulate magnitude vs. Ts prevents rejection of beneficial commensals).

During our discussion, T-suppressors were postulated to play one or the other of five roles:

  • - purge anti-self from the repertoire (most of the immunological community)
  • - purge only low affinity anti-self from the repertoire ([36] Anderson and Thangavelu)
  • - determine the class of the humoral response (many idiotype network theorists)
  • - protect “useful” bacteria (commensals) [21]
  • - determine the magnitude of the response [12, 26, 31, 33]

My conclusion was made by subtraction. I ruled out a role of Ts in purging anti-self (low or high affinity) ([11, 12, 33], see above) and in directly determining the class of humoral response [28]. The proposed role of the immune system in protecting commensals was more of a question of definition, as among what might be viewed as commensals are opportunistic pathogens. There is no doubt that the immune system plays a role in keeping intestinal flora with the potential to be harmful at bay. However, it responds to both non-pathogen and pathogen. If a commensal is defined as harmless even in the absence of an adaptive immune system, then it probably did not act as a selective pressure characterizing that system. However, one must be careful. A RAG-minus mouse in an animal quarters and in the wild are under somewhat different pathogenic and commensal universes as selective agents.

While non-pathogenic flora can have an effect via the innate system on antigen-unspecific events [37], this is not what we are discussing. In any case, under the ARA model, the sorted random anti-nonself repertoire of Ts could not distinguish commensals from pathogens as classes; at best the class of the response to each might be different due to additional antigen-unspecific tissue signals as proposed by Dembic and others. Consequently, regulation of the magnitude of the response to nonself remains in my mind a default postulate for the role of T-suppression.

In sum, then to use Alexandre Corthay’s encapsulation of my position:

  1. The repertoire of Tregs (Ts) is sorted to be anti-nonself.
  2. The function of Tregs (Ts) is to down-regulate the magnitude of the effector response.
  3. Tregs (Ts) are not involved in the self-nonself discrimination.

4- “In terms of control of effector class and unresponsiveness to an antigen, Cohn states…. ‘The antigen is a nonself-component that under conditions of encounter induces a level of T-suppressors that inhibits an observable response.’”

The quote, as treated by Anderson, is out of context.

When a black box experimentalist fails to observe a response to an antigen, there are a priori several explanations, one of which was quoted by Anderson from my essay [1]. The quote is not in the context of evolutionary selection; it deals with the manipulation of the system by the experimenter. For example, a nonself graft might not be rejected if a Ts population is experimentally induced or introduced to a sufficient level, a point also made by Anderson [4] under his Section 3. This finding is not extrapolatable to ‘tolerance’ because the Ts population in this example must have been sorted to be anti-nonself. Its repertoire cannot regulate what it cannot see, namely self. Dominant “unresponsiveness” is an experimental observation (fact) important for the understanding of Module 3, regulation of magnitude of response. “Unresponsiveness,” not “tolerance,” can be infectious; Waldmann please note [38].

5-“Regarding the question of tolerance to antigens arising later in life, Cohn raises the criticism that examples of tolerance later in life…..are not the equivalent of naturally evolved self-tolerance but is, instead, experimental unresponsiveness.

So-called ‘tolerance’ to antigens arising late in life are experimentally contrived examples of unresponsiveness, not extrapolatable to the sorting of the repertoire. Once the system is responsive (the developmental window is closed), it cannot distinguish newly arising self from nonself. Anderson challenges this statement by asking me to explain every extant example interpreted by him to challenge this statement. If I explain them, then the explanation is ipso facto to be viewed as hiding behind complexity with an “implausible interpretation.” This is intimidating rhetoric; I know of no contradicting experiments that warrant abandoning the ARA model for the sorting of the repertoire (Module 2).

Consequently I would challenge Anderson to produce the “myriads of examples of late life ‘tolerance’” and of “fetal life ‘responsiveness’” that are not understandable under or incompatible with the ARA model. In essence, Anderson is arguing (as does a large segment of the immunological community) that the sorting of the ‘adaptive’ repertoire can be accomplished by a mechanism independent of developmental time. I view the role of developmental time as a default postulate.

We seem to agree that late life unresponsiveness to a nonself antigen due to suppression cannot be extrapolated to the mechanism of “natural’ tolerance (the sorting of the repertoire). Late life unresponsiveness due to negative selection by the nonself-epitope requires that the rate of inactivation of the antigen-responsive T-helper be more rapid than the rate at which it is induced to effectors. There are many ways to achieve this state.

Let us consider Anderson’s specific example because it illustrates the poverty of pure empiricism. “How does Cohn explain life long tolerance to male islet grafts (and subsequently male skin grafts to the same recipients) by ‘syngeneic’ female recipients given the grafts as adults?

While this is reasonably explained under the ARA model, it is difficult to explain under all other models.

The islets do not express Class II MHC and present H-Y peptide on Class I MHC only. H-Y when associated with islets transplanted under the renal capsule, is not available as a ligand for T-helpers and B-cells. As the islets are healthy/viable cells, H-Y is either poorly presented by APC or is presented as a “tolerogen only,” because there is an insufficiency of regulatory eTh anti-[PH-Y-Class II MHC] to activate an effective eTh response. The closer that a challenge antigen is to self (in this case differing from host by a single epitope) the higher the probability it will establish unresponsiveness. There is competition between inactivation and activation; closer to self favors inactivation, further from self favors activation. Even if defensive eTh anti-H-Y were induced they would not see the islets as a target because they lack Class II MHC. The islets could only be destroyed by them as innocent bystanders. More likely eTh anti-H-Y are not induced as the skin graft is accepted. Given an insufficiency of functioning regulatory effector T-helpers anti-H-Y, the naive cytotoxic T-cells, Class I MHC-restricted, are deleted (Signal[1]) upon interaction with islets. When a male skin graft is tested, under conditions such that its rejection is dependent on cytotoxic effector T-cells, the graft is accepted because the cytotoxic T-cells anti-H-Y are maintained at a steady state level below the autoimmune boundary by the tolerogenic-only islets. There is competition between the negatively selecting islets and the potentially immunogenic skin graft that prevents the breaking of unresponsiveness. This cited observation supports the postulates that peripheral negative selection is operative in adults (maintenance of tolerance), that effector T-helpers are required to activate cytotoxic T-cells, and that the acceptance of an H-Y skin graft is dependent on an insufficiency of effector cytotoxic T-cells anti-PH-Y, Class I MHC restricted. I would suggest that these findings should be viewed as supportive of the ARA model because, for the immune system, its stage of development is determined (or defined) by the insufficiency or sufficiency of effector T-helpers (eTh) not by the age of the animal. Converting an adult immune system into an embryonic immune system requires the purging of its effector T-helpers (eTh) and allowing the naive T-helpers to arise when there is an insufficiency of eTh.

There are other ways that an experimentalist can establish antigen-specific unresponsiveness in adults. Some may actually be in the literature (I haven’t looked). For example, if one fixes a cell with formalin, glutaraldehyde, or any other agent under properly controlled conditions, then it might become resistant to processing because crosslinking inhibits proteolysis. However, the fixed cell might still be able to inactivate B-cells that interact with conformationally unaffected cell-surface epitopes (Signal[1]) rendering the host immune system unresponsive to them in the humoral class.

Another example might be to engineer a cell that expresses Class II MHC covalently attached to dominant peptide from a given antigen. This cell when appropriately fixed would inactivate T-helpers rendering the immune system unresponsive in all classes to the antigen from which the peptide was derived.

While many more examples can be envisioned none are extrapolatable as being examples of late life “tolerance.” They are examples of late life “unresponsiveness.”

Anderson’s reference to the Lederberg Model as an alternative to the ARA Model is basically incorrect. The Lederberg Model is embedded in a developmental time assumption. Consider an inactivatable-only population arising in the presence of self and nonself (an unsorted antigenic universe). No response to either is possible. The Lederberg Model requires a period in the absence of nonself for activatable-only anti-nonself cells to accumulate. In such a system a role for eTh is obviated. Both activation and inactivation are mediated epitope-by-epitope. Further, all models for sorting the repertoire require a period of development when antigen-responsive cells arise under an inactivatable-only condition. This could be inherent to the cell (Lederberg Model) or to the system because of the insufficiency of eTh (ARA Model), or, less likely, to both.

Let us develop further the problem of delayed expression to self-antigens that is so troubling to Anderson. Under the ARA model no self-antigen can arise as a de novo ligand for the immune system after the developmental time window closes (sufficiency of eTh). Consider three experiments relevant to the delayed expression of self-antigens.

Adams et al. [39] established two transgenic murine lines that express the SV40 T-antigen uniquely in the β-cells of the pancreas. One line expresses the T-antigen early during embryonic life; the other expresses the T-antigen delayed after birth. The early expressor treats the T-antigen as self and there is no immune response to it; the delayed expressor treats the T-antigen as nonself and responds to it with the destruction of the β-cells resulting in diabetes, a predictable consequence under the ARA Model.

If the T-antigen provokes a response when expressed delayed in the periphery, it could not have been ectopically expressed earlier in thymus as an appropriate ligand acting either as a negative selector for T-helpers or as a positive selector for T-suppressors. Consequently, the unresponsive state of the early expressor can be reasonably postulated to be due to negative selection on T-helpers by the T-antigen processed in peripheral APC while the developmental time window was open. The responsiveness to the T-antigen seen in the late expressor seems to be initiated by the accumulation of eTh anti-T-antigen during the period of its absence (i.e., the primer nonself antigen-independent pathway [7]).

As an aside, in the experiment of Adams et al. [39] the transgenic T-antigen induced tumors in both the responder (delayed expressor) and nonresponder (early expressor) lines. This argues that any deleterious or traumatic effect attributable to the tumor does not impinge on the establishing of tolerance (Module 2), although it could well play a role in determining the class of the effector response (Module 3).

The experiments of Le Douarin et al. [4042] can be interpreted similarly. A quail limb bud was grafted onto a chicken embryo before the immune system appears. The chicken is hatched with a healthy functioning integrated quail limb that is rejected acutely at around one week after birth. One explanation would be that the quail limb expressed a delayed self-antigen. A delayed self-antigen cannot be distinguished from a nonself antigen and becomes a target of attack by the chicken immune system.

If this explanation has validity, one must ask why isn’t the quail limb rejected in quail. There are two possibilities, either the postulated delayed self-antigen is expressed early in quail or it is expressed delayed in both chicken and quail but in quail it is expressed ectopically in thymus as a negatively selecting ligand for T-helpers.

The graft of embryonic quail thymus epithelium onto a chick embryo results in a functioning chicken-quail thymus. In such an animal the quail limb is accepted into adult life favoring ectopic expression.

These experiments set the stage for our interpretation [7, 31] of the Aire controlled ectopic expression of peripheral self-antigens in thymus (and in APC in general), while the developmental time window is open. The Aire controlled family of ectopically expressed peripheral self-antigens must include all those that are expressed as delayed ligands for T-helpers. If this be the case then, there is no example of a peripheral self-antigen expressed as a ligand for the immune system after the developmental window closes, much less a “myriad of examples.” Further, I have had no need to make “multiple assumptions for the late life tolerance,” except for the one that it doesn’t exist. Only late life unresponsiveness exists.

Further, the ARA framework provides an explanation for the finding that well-healed H-Y skin grafts on a RAG-minus adult female mouse are rejected by a female fetal liver cell transplant [43]. In order to reject such a graft, effector T-helpers anti-H-Y must appear in the periphery at a rate that is faster than the rate that they are negatively selected by the graft. Predictably, if the size of the graft is increased, it will be accepted. This has been explained under the ARA Model by postulating a nonself antigen-independent pathway to effector T-helpers [28] that initiates their induction. H-Y is never a ligand for B-cells but would, in this situation, be a negative selector for cytotoxic T-cells anti-[PH-Y-Class I MHC]. However, their induction to effectors requires eTh anti-H-Y. What assumptions to explain this finding that are also testable must be envisioned under any of the other models?

The same holds true for “lack of tolerance in numerous studies where antigens are present during the putative tolerogenic window.” There are no examples of responsiveness to antigen during the “tolerogenic window” (insufficiency of effectors T-helpers, eTh). If the antigen is administered with a Signal[2] substitute like LPS or CpG (advertently or inadvertently) then all that was demonstrated is that the naive T/B cells (i-cells) are born tolerizable and inducible (an assumption of the ARA-Model). The eTh-independent antigens are ligands for the innate system for which a self-nonself discrimination is germline, not somatically selected. It is the sufficiency or insufficiency of eTh (Signal[2]) that determines the pathway to activation or inactivation. If the response to any challenge of an embryo with a foreign agent (protein, virus, etc.) is to be considered as a contradiction (disproof) of the ARA model, then it must be shown that it was expressed as a processed peptide-Class II MHC ligand during embryonic life (i.e., insufficiency of eTh) accessible to the newly arising iTh for negative selection [31]. Given this, what examples should we discuss?

At this point the question of an embryonic response to eTh-independent antigens always arises. These are ligands for the innate immune system for which a self-nonself discrimination is germline, not somatically selected. If the antigen is seen by the adaptive system, the epitope that is recognized must be absent in the host. All that such a response shows is that immune cells are born activatable and inactivatable at all stages in developmental time.

Lastly, let me comment on a philosophical point raised by Anderson. It is not a question of, “the ratio of contradicting to supporting experiments” that would make me change my mind, as that ratio is dependent on interpretation. I would change my mind if there existed a competing theory for the sorting of the repertoire (the self-nonself discrimination) that dealt with the data in as disprovable, comprehensive, predictive, accurate and logical a manner as the ARA theory. In this framework, it would take only one decisive experiment to disprove the model (i.e., to change my mind). Fortunately, a theory of how the effector class is determined and its magnitude regulated, is largely (not completely) independent of one’s conceptualization of the sorting process. The three modules have comfortable boundaries but this does not negate that they can be eventually melded to model an immune system.

6- “Cohn’s assertion that the immune system cannot distinguish commensals from parasites……

The B- and T-cell repertoires, after sorting, recognize both commensals and pathogens equally well as nonself. If the Ts repertoire is sorted to be anti-nonself then it cannot distinguish the commensal from the pathogen because it recognizes epitopes (peptide-Class II complexes) not antigens. To the extent that there is one, the distinction must come from the antigen-unspecific commensal-tissue vs pathogen-tissue interaction (Dembic’s Signal 3 or in my discussion Postulates 12 and 13 [1]).

I did not wish to imply that all commensals are potential pathogens. What is defined as commensal in the presence of an intact immune system might be a pathogen in its absence (opportunistic infections).

The goal of my essay [1]

It is becoming increasingly clear that computer modeling of immune behavior will be a major tool in dealing with its complexity. If we are interested in how the immune system does work, not in how it might work, then important constraints are put on the ground rules for modeling. Real world immune systems are bounded by evolutionary principles and testable predictions. Artificial immune systems are unbounded and deal with our ability to design protective mechanisms superior or equivalent to the ones selected by evolution.

The comment that I “oversimplified” is in essence a critique that I ignored or brushed aside complexity. As the real world immune system evolved as a collection of fortuities interactively selected upon for function, it is dauntingly complex and heavily booby-trapped for theorists. My approach to this problem was simple. Bite off as big a chunk of the subject as you can assimilate without choking. This, for me, divided the antigen-specific elements of immune responsiveness into three modules each with its own logical structure. The generalizing postulates within each module are based on that logic and can be the basis for an algorithm that permits one to explore the quantitative limits to the parameters and to reveal cryptic relationships between them. The algorithms describing each module can be linked to yield an output. If the output is ‘absurd,’ then it becomes possible to ask questions such as:

  • – Which of the postulates is at fault?
  • – Was that postulate arrived at due to a misinterpretation of the data?
  • – What postulate could be substituted or added that would give a satisfactory output?
  • – What are the predictions of the algorithm that are testable?

If a model is posited in the wrong framework, it is misleading even though its elements might be essential to another framework. I have not argued that models based on danger, pathogenicity, cytopathicity, localization, integrity, context, etc. are without merit. What I have argued is that they do not (as claimed by their proponents) obviate the requirement for a self-nonself discrimination and do not in any way explain it. These models do give us key insights into Module 3, the regulation of class [1].

Computer modeling must be guided by postulates based on evolutionary principles and as firmly supported by experiment as possible. Formulating these postulates should be the role of the immunologist who must articulate and clarify them for the computer modeler. This SJI Discussion Forum is a unique place to do this, as a default set can only be arrived at as a consensus. If my set is unsatisfactory, by all means produce a competing or supplemental set that we can discuss


There is no way that one can overestimate the value of conscientious reviewing of manuscripts, particularly when open refereeing is involved. To this end, I thank and appreciate Alexandre Corthay whose thoughtful suggestions were incorported into the body of this commentary.

This essay is also a testimonial to the insightful criticism of Colin Anderson. His caring about “truth” and accuracy of conceptualization, as well as fairness, make debating with him a privilege. I am deeply grateful for his effort to keep me honest.

This work was supported by a grant (RR07716) from the National Center For Research Resources (NCRR), a component of the National Institutes of Health (NIH) and its contents are solely the responsibility of the authors and do not represent the official view of NCRR or NIH.


Associative Recognition of antigen
B-cell antigen-receptor
T-cell antigen-receptor
Complex between self-peptide (Ps) and MHC-encoded Class II restricting element (RII)


1. Cohn M. A rationalized set of default postulates that permit a coherent description of the immune system amenable to computer modeling. Scan J Immunol. 2008;68:371–80. [PMC free article] [PubMed]
2. Dembic Z. Beginning of the end of (Understanding) the immune response. Scand J Immunol. 2008;68:381–2. [PubMed]
3. Cohn M. A reply to Dembic: On an End to the Beginning of Misunderstanding the immune Response. Scand J Immunol. 2009 In press. [PubMed]
4. Anderson CC. Placing regulatory T cells ito global theories of immunity: An analysis of Cohn’s challenge to integrity (Dembic) Scand J Immunol. 2009 In press. [PubMed]
5. Cohn M, Langman RE. The Protecton: the evolutionarily selected unit of humoral immunity. Immunol Reviews. 1990;115:1–131.
6. Langman RE, Cohn M. The E-T (elephant-tadpole) paradox necessitates the concept of a unit of B-cell function: The Protecton. Mol Immunol. 1987;24:675–97. [PubMed]
7. Langman RE, Mata JJ, Cohn M. A computerized model for the self-nonself discrimination at the level of the T-helper (Th genesis) II. The behavior of the system upon encounter with nonself antigens. Int Immun. 2003;15:593–609. [PMC free article] [PubMed]
8. Cohn M, Langman RE, Mata J. A computerized model for the self-nonself discrimination at the level of the T-helper (Th-genesis). I. The origin of “primer” effector T-helpers. Int’l Immunol. 2002;14:1105–12. [PMC free article] [PubMed]
9. Langman RE. The specificity of immunological reactions. Mol Immunol. 2000;37:555–61. [PubMed]
10. Cohn M. A new concept of immune specificity emerges from a consideration of the Self-Nonself discrimination. Cell Immunol. 1997;181:103–8. [PubMed]
11. Cohn M. On the opposing views of the self-nonself discrimination by the immune system. Immunology and Cell Biology. 2008 in press. [PMC free article] [PubMed]
12. Cohn M. What roles do regulatory T-cells play in the control of the adaptive immune response? Int Immunol. 2008;20:1107–18. [PMC free article] [PubMed]
13. Matzinger P. Tolerance, danger and the extended family. Annu Rev Immunol. 1994;12:991–1045. [PubMed]
14. Matzinger P. The Danger Model: A renewed sense of self. Science. 2002;296:301–4. [PubMed]
15. Matzinger P. The real function of the immune system or tolerance and the four D’s (danger, death, destruction, and distress) 2003
16. Zinkernagel RM. On ‘reactivity’ versus ‘tolerance’ Immunology and Cell Biology. 2004;82:343–52. [PubMed]
17. Janeway CA. The immune system evolved to discriminate infectious nonself from noninfectious self. Immunol Today. 1992;13:11–6. [PubMed]
18. Janeway CA, Goodnow CC, Medzhitov R. Immunological tolerance: Danger pathogen on the premises! Current Biology. 1996;6:519–22. [PubMed]
19. Zinkernagel RM. Uncertainties - Discrepancies in Immunology. Immunol Rev. 2002;185:103–25. [PubMed]
20. Zinkernagel RM. Immunology Taught by Viruses. Science. 1996;271:172–8. [PubMed]
21. Dembic Z. Immune system protects integrity of tissues. Mol Immunol. 2000;37:563–9. [PubMed]
22. Cunliffe J. Morphostasis: an evolving perspective. Medical Hypotheses. 1997;49:449–59. [PubMed]
23. Anderson CC. Time, space and contextual models of the immunity tolerance decision: Bridging the geographical divide of Zinkernagel and Hengartner’s ‘Credo 2004’ Scand J Immunol. 2006;63:249–56. [PubMed]
24. Miller J. Self-nonself discrimination by T lymphocytes. CR Biologies. 2004;327:399–408. [PubMed]
25. Matzinger P. Friendly and dangerous signals: is the tissue in control? Nature Immunol. 2007;8:11–3. [PubMed]
26. Cohn M. A biological context for the Self-Nonself discrimination and the regulation of effector class by the immune system. Immunol Res. 2005;31:133–50. [PubMed]
27. Cohn M. A Commentary on the Zinkernagel-Hengartner “Credo 2004” Scan J Immunol. 2005;61:477–84. [PMC free article] [PubMed]
28. Cohn M. On “Credo 2004” as viewed under the “Development-Context” model of Colin Anderson. Scan J Immunol. 2006;64:97–103. [PubMed]
29. Cohn M. Molecular Approaches to Immunology. New York: Academic Press; 1975. The logic of cell interactions in determining immune responsiveness; pp. 79–107.
30. Cohn M. The self-nonself discrimination: Reconstructing a cabbage from sauerkraut. Res Immunol. 1992;143:323–34. [PubMed]
31. Cohn M. Conceptualizing the Self-Nonself discrimination by the vertebrate immune system. In: Timmis J, Flower D, editors. In Silico Immunology. New York: Springer; 2007. pp. 375–98.
32. Segel LA, Bar-Or RL. On the role of feedback in promoting conflicting goals of the adaptive immune system. J Immunol. 1999;163:1342–9. [PubMed]
33. Cohn M. Whither T-suppressors: If they didn’t exist would we have to invent them? Cell Immunol. 2004;227:81–92. [PubMed]
34. Cohn M. An in depth analysis of the concept of “polyspecificity” assumed to characterize TCR/BCR recognition. Immunologic Research. 2008;40:128–47. [PMC free article] [PubMed]
35. Cohn M. Degeneracy, Mimicry and Crossreactivity in Immune Recognition. Molecular Immunology. 2005;42:651–5. [PubMed]
36. Allan-Baecher C, Viglietta V, Hafler DA. Inhibition of Human CD4+ CD25+high Regulatory T Cell Function. Journal of Immunology. 2002;169:6210–7. [PubMed]
37. Bouskra D, Brezillon C, Berard M, et al. Lymphoid tissue genesis induced by commensals through NOD1 regulates intestinal homeostasis. Nature. 2008;456:507–12. [PubMed]
38. Waldmann H. Tolerance can be infectious. Nature Immunol. 2008;9:1001–3. [PubMed]
39. Adams TE, Alpert S, Hanahan D. Non-tolerance and autoantibodies to a transgenic self antigen expressed in pancreatic βcells. Nature. 1987;325:223–8. [PubMed]
40. Ohki H, Martin C, Corbel C, Coltey, Le Douarin N. Tolerance induced by thymic epithelial grafts in birds. Science. 1987;237:1032–5. [PubMed]
41. Coutinho A, Salaun J, Corbel C, Bandeira A, Le Douarin N. The role of thymic epithelium in the establishment of transplantation tolerance. Immunol Rev. 1993;133:225–40. [PubMed]
42. Le Douarin N, Corbel C, Bandeira A, et al. Evidence for a Thymus-Dependent Form of Tolerance that is Not Based on Elimination or Anergy of Reactive T cells. Immunol Rev. 1996;149:35–53. [PubMed]
43. Anderson CC, Carroll JM, Gallucci S, Ridge JP, Scheever AW, Matzinger P. Testing Time-, Ignorance-, Danger-Based Models of Tolerance. J Immunol. 2001;166:3663–71. [PubMed]
44. Gray DHD, Gavanescu I, Benoist C, Mathis D. Danger-free autoimmune disease in Aire-deficient mice. PNAS. 2007;104:18193–8. [PubMed]
45. DeVoss J, Hou Y, Johannes K, et al. Spontaneous autoimmunity prevented by thymic expression of a single self-antigen. Journal of Experimental Medicine. 2006;203:2727–35. [PMC free article] [PubMed]