The Affirmation of Self: A New

Perspective on the Immune

System

John Stewart
CNRS
COSTECH
Centre Pierre Guillaumat
Universite de Compi`egne
BP 60649
60206 Compi`egne
France
John.Stewart@utc.fr

Antonio Coutinho
Abstract The fundamental concepts of autopoiesis, which
emphasize the circular organization underlying both living
organisms and cognition, have been criticized on the grounds
that since they are conceived as a tight logical chain of
definitions and implications, it is often not clear whether they
are indeed a scientific theory or rather just a potential
scientific vocabulary of doubtful utility to working scientists.
This article presents the deployment of the concepts of
autopoiesis in the field of immunology, a discipline where
working biologists themselves spontaneously have long had
recourse to cognitive metaphors: recognition; a
repertoire of recognized molecular shapes; learning and
memory; and, most striking of all, a self versus non-self
distinction. It is shown that in immunology, the concepts of
autopoiesis can be employed to generate clear novel
hypotheses, models demonstrating these ideas, testable
predictions, and novel therapeutic procedures.
Epistemologically, it is shown that the selfnon-self
distinction, while quite real, is misleadingly named. When a
real mechanism for generating this distinction is identified, it
appears that the actual operational distinction is between (a)
a sufficiently numerous set of initial antigens, present from
the start of ontogeny, in conditions that allow for their
participation in the construction of the systems organization
and operation, and (b) single antigens that are first presented
to the system after two successive phases of maturation. To
call this a selfnon-self distinction obscures the issue by
presupposing what it ought to be the job of scientific
investigation to explain.

Instituto Gulbenkian de Ciencia

PT-2781-901 OEIRAS
Portugal
coutinho@igc.gulbenkian.pt

Keywords
Autopoiesis, immune system, idiotypic network, self, non-self

1 Introduction
The concept of autopoiesis is one of the major contributions that Francisco Varela made
both to biology and to cognitive science.
Living organisms are characterized by their circular organization: they are metabolic
processes, pure fluxes of matter and energy, that have the very special property of producing themselves. They are dynamic dissipative structures; but unlike purely physical
and/or chemical structures, which are intrinsically ephemeral, living organisms function
in such a way that they can go on producing themselves indefinitely; in other words,
although they can of course die at any moment if an accident befalls them, they are
potentially immortal.
c 2004 Massachusetts Institute of Technology

Artificial Life 10: 261276 (2004)

J. Stewart and A. Coutinho

The Affirmation of Self

Cognition is also characterized by a circular organization: the objects of cognition

(for example, colors) are not given in advance, and do not even exist as such independently of the perceiving subject; these objects are brought forth together with the
subject, and are inseparable from the organization of that subject. Along with Humberto Maturana, Francisco Varela had the deep insight that these two forms of circular
organization are, at root, one and the same: this insight can be summed up by the
formula life=cognition=autopoiesis.
These were, and still are, heady ideas, with an extraordinary capacity to evoke
an immediate intuitive conviction that they have to be right. However, it is equally
obvious that in order to make a substantial contribution to science these ideas need to
be deployed quite concretely; they have to do some real work in relation to precise,
detailed experimental observations. In the 1980s it appeared to a number of people
Varela himself, the Brazilian immunologist Nelson Vaz, and the authors of this article
among othersthat the immune system offered a particularly promising field for putting
these ideas into practice. It is a historical fact that immunology is an area where working
biologists themselves spontaneously have long had recourse to cognitive metaphors:
recognition; a repertoire of recognized molecular shapes; learning and memory;
and of course, most striking of all, a self versus non-self distinction. And it did indeed
turn out that the application of the concept of autopoiesis to the immune system led
with extraordinary rapidity to a dramatic reversal in perspective. The aim of this article
is to explain what is involved in this potential paradigm shift.
2 Classical Immunology and a Potential Paradigm-Shift
In classical immunology, the immune system is conceived as a linear input-output system. The inputs are antigens: substances, generally foreign to the body of the animal,
which trigger the production as an output of antibodies, each of which specifically recognizes the antigen that evoked it. The specificity of this recognition is impressive: the
number of different antigens that the immune system of a mammal is able to recognize
has been estimated to be of the order of 1017 . Moreover, immunologists consider that
the repertoire of antibodies is complete: the mammalian immune system is capable
of recognizing the totality of all possible molecular shapes (of an appropriate size),
including molecules that have never before existed in the course of biological evolution because they were synthesized for the first time by human chemists. Classically
again, immunologists consider that the function of the immune system is to protect
the body against foreign antigens (typically, those that belong to pathological microorganisms); more precisely, it is considered that the function of an antibody is to trigger
the destruction of the antigens that it recognizes (Figure 1).
This classical scheme can be summed up by saying that the immune system can
potentially perceive everything, and that it triggers the destruction of everything that it
actually perceives. Now this bald schematic formulation raises a troublesome question:
what about the relation between the immune system and the body in which it is housed?
This body is composed of molecules, many of which are of the appropriate size for
being perceived by the immune system; the proof of this is that if such molecules are
injected into another animal, they do indeed provoke a destructive immune response.
It is for this reason that grafts of tissues or organs are practically impossible in mammals,
whereas they are easy in plants or invertebrate animals such as insects. But now, if
we apply the classical schema literally, we arrive at an astonishing conclusion: the
prediction is that the immune system should systematically destroy the body in which
it is housed.
It is clear that this is not what happens; or more precisely, when something of this sort
happens, one speaks of autoimmune disease, but autoimmune diseases are relatively
262

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

Figure 1. The linear input-output system of classical immunology. An antigen triggers the production of an antibody
that specifically recognizes and destroys the antigen.

rare, and when they do occur, they are far less catastrophic than the scheme predicts.
In fact, quite apart from empirical observations, it is obvious that this prediction cannot
be fulfilled, because systematic autoimmunity would be horribly dysteleological and
quite incompatible with the constraints of viability and natural selection. The classical
immunologists invented a term, horror autotoxicus, to indicate that this prediction of
their theory was not and indeed could not be systematically verified. Now, contrary to
the opinion of Karl Popper, a refutation does not necessarily induce scientists to abandon the theory in question. In classical immunology, the refutation was circumvented
by a straightforward adjustment to the theory: the immune system perceives everything except its own body. One suspects that this bald-faced adjustment, which is quite
shamelessly ad hoc in order to avoid what would otherwise be a straight refutation,
caused a certain unease, which may be subliminally expressed by the very term horror
autotoxicus, quite horrible indeed in its mixture of Greek and Latin roots. However
that may be, the conclusionthat the immune system perceives everything except its
own bodyis quite inescapable, given the premises of the argument; and it is for this
reason that the so-called self versus non-self distinction plays such a central role in
classical immunology.
What light does the perspective of autopoiesis shed on this rather murky situation?
According to the notion of autopoiesis, the objects of cognition are specified, constituted, by the organism itself. This can be summed up very neatly by a verse of the
Portuguese poet Pessoa:
What we see
Is not what we see
But what we are
The point is, in a way, a very simple one; but it runs so radically counter to ordinary
common sense that a few words of explanation may not be amiss. The usual point of
view is to consider that the objects of perception are ontologically primary: they exist,
and are what they are, quite independently of any perception that there may or may
not be concerning them. A perception by a cognitive subject then corresponds to an
internal representation of the referential object; ideally, the representation will tend
to be more or less isomorphic with the pre-existing referent. Technically, this point of
view is termed objectivist; we may note that it corresponds to the point of view of an
external observer who is presumed to be omniscient, and is thus able to examine both
the object and the representation and to check out on the degree of correspondence or
Artificial Life Volume 10, Number 3

263

J. Stewart and A. Coutinho

The Affirmation of Self

isomorphism between them. The perspective of autopoiesis is radically critical of the

objectivist position. The difficulty is that, from the point of view of a cognitive subject,
there is no way of getting out of its skin and perceiving the object in itself directly as
such. The only thing that an organism can know is the effect that its interaction with an
object has on its own functioning; having access to only one of the two terms, there is
just no way that an organism can judge whether the content of its percept is, or is not,
an adequate representation of an external object. Thus, the percept is not separable
from the cognitive subject. This is reminiscent of Berkeleys position, to be is to be
perceived, with the difference that an organism is subject to a viability constraint, so
that its percept cannot be an arbitrary hallucination but must bear significantly on the
organisms interactions with its environment.
In other words, in this new perspective, whatever one perceives is, by the very fact
of being perceived, the self; and whatever is not perceived is, ipso facto, non-self.
This amounts to an exact reversal of classical immunology, according to which the
immune system does not perceive, ignores, the self (otherwise it would destroy it),
and perceives only the non-self. Varela was the first to develop these considerations,
in close collaboration with Vaz; together, they proposed that in order to denote this
radical reversal in perspective, it might be better to speak of a distinction between self
and nonsense [16].
Before going further, it may be well to say a few words to try and dispel some confusion that has quite understandably arisen. It is abundantly clear that a major (if not necessarily the exclusive) function of the immune system is to protect the organism against
infectious diseases caused by pathological microorganisms. This is demonstrated quite
straightforwardly by the simple observation that severely immunodeficient miceand
humansdo indeed die of uncontrolled infectious disease. It is equally clear that in
order to do this, the immune system must make a distinction between pathological microorganisms and the body of the organism itself. The perspective of autopoiesis does
not gainsay any of this. What is at stake is the choice of an appropriate nomenclature
in order to designate the two terms of this distinction. The key here is a remark that
Humberto Maturana never tired of repeating: Everything said is said by an observer.
In particular, whenever a distinction is being made, we should always ask: Who is
making the distinction? In the case of classical immunology, the distinction is being
made by an external human observer: it is the immunologist who can see the difference between the organism on the one hand and pathological microorganisms on the
other; it is the immunologist who designates molecules from the body of the organism
as self, and molecules that he knows came from a microorganism (or another source
external to the organism) as non-self. By contrast, when we employ the conceptual
framework of autopoiesis, we are in a certain sense looking at things from the point
of view of the immune system itself.1 It is from this point of view that a selfnon-self
distinction is impossible, because, unlike a human immunologist, the immune system
itself has no means of knowing where the molecules came from. The immune system is composed of cells, the lymphocytes; at the level of a local interaction between
an individual lymphocyte and a molecule, there is nothing that distinguishes what the
immunologist calls a self molecule from a non-self molecule. The conceptual framework of autopoiesis is thus the more appropriate one if we are trying to understand the
mechanisms and mode of operation of the immune system itself. As we shall see, the
immune system as a whole, considered over the history of its development, is capable
of making a distinction that, to all practical intents and purposes, does roughly coincide
1 It is well to acknowledge that this is something of a rhetorical flourish: the point of view in question is not really that of the
immune system itself, but rather that of a human biologist acting as spokesman for the immune system. This being said, there
remains a valid and important distinction between the two points of viewthat of the classical immunologist, and that of the
spokesman.

264

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

with the immunologists self versus non-self distinction. However, when we identify
the mechanisms whereby the immune system is capable of making this distinction,
we shall appreciate that calling this a self versus non-self distinction is a misnomer.
We shall return to this point; but the time has come to present the understanding of
the actual functioning of the immune system that comes from adopting the conceptual
framework of autopoiesis.
3 A Mathematical Model of the Immune Network
A key element that made it possible to apply the framework of autopoiesis to the
workings of the immune system was the work of the great Danish immunologist, Niels
Jerne. The starting point of Jernes theory is this: if the repertoire of antibodies really
is complete (or open-ended, as he called it), then it is logically inescapable that the
antibodies themselves should be included in this repertoire. After all, antibodies are
protein molecules of the same size as many other antigens. In other words, there are
strong a priori grounds for supposing that the set of antibodies forms a connected
network, where each antibody is recognized by other antibodies in the system. We
have put the term antibody in scare quotes because it is clear that in this perspective
that recognition does not necessarily lead to total and immediate destruction. For this
reason, it is preferable to use the term immunoglobulin to designate the molecules
produced by the lymphocytes. In order to designate this sort of interaction between
immunoglobulins we employ the term idiotypic; consequently, the sort of network
predicted by Jerne is an idiotypic network.
When Varela came to Paris in 1985, he worked with Antonio Coutinho (himself a student of Jernes) to set down the basis of a mathematical model of idiotypic networks.
This is not the place to enter into technical details and the mathematical equations
[13, 14]. Qualitatively, in natural language, the basic idea was the following. The
survival of a lymphocyte, its proliferation, and its capacity to secrete immunoglobulins depended on the field that it received as a result of its idiotypic interactions with
other immunoglobulins. The dynamics of the network were highly nonlinear: if the
received field was below a lower threshold, the lymphocyte perished (Vaz expressed
this by saying that the lymphocyte died of loneliness); if the field was above an upper threshold, the lymphocyte died of suffocation; however, if the field was in the
favorable window between the two thresholds, the lymphocyte could survive and/or or
proliferate and produce immunoglobulins (Figure 2).2 Now, the immunoglobulins that
provide for the field itself, in the form of idiotypic interactions with the immunoglobulin receptors of the newly emerging lymphocytes, are produced as a consequence
of these very interactions. Hence the field determines itself and its maintenance, but
it is never the same, as the emerging lymphocytes never repeat the same repertoire
of diverse immunoglobulins. That is, the field is self-producing, and it selects newly
formed lymphocytes into its own operation, so that these dynamics are supplemented
by a process of meta-dynamics. We have already indicated that a lymphocyte could
disappear from the system (if its received field lay outside the window); but at each time
step, it was also possible to recruit new lymphocytes into the network from a sample of
random lymphocytes freshly produced by the bone marrow. Specifically, a candidate
lymphocyte would be recruited if, but only if, its received field was situated within the
window between the lower and upper thresholds. There was thus a circular relationship between the dynamics and the meta-dynamics (Figure 3). On the one hand, the
dynamics of the concentrations of the different lymphocytes determined the field at all
2 These rules for lymphocyte stability (survival within a favorable window of sensed field) are analogous to Conways well-known
game of life. They are also similar to the neural homeostasis that is currently being implemented in evolved robot controllers.

Artificial Life Volume 10, Number 3

265

J. Stewart and A. Coutinho

The Affirmation of Self

Figure 2. The bell-shaped activation curve for lymphocytes as a function of the received field. Below the lower
threshold LT, and above the upper threshold UT, the lymphocyte declines; between the two thresholds, there is a
window where the lymphocyte survives and/or proliferates and secretes immunoglobulins.

Figure 3. The circular relationship between dynamics and meta-dynamics. The dynamics presupposes a set list of
variables (the concentrations of the clones) and a fixed matrix of pairwise affinities between these clones. The
temporal variation in the concentration of each of the clones is then determined by a set of differential equations. At
any one point in time, the concentrations of the clones determine the field at all points in shape space. This in turn
determines the meta-dynamical processes of maintenance and elimination of existing clones, and the recruitment of
new clones: a clone will be maintained or recruited if the field that it receives lies within the window between lower
and upper threshold; it will be eliminated or not recruited otherwise. Thus, the meta-dynamics determines the list
of variables and the matrix of affinities; which brings us full circle.

266

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

points in shape space, and so the dynamics determined the meta-dynamics; conversely,
the meta-dynamics gave rise to the structure and connectivity of the network at each
instant, and so the meta-dynamics determined the dynamics.
This circularity between the dynamics and the meta-dynamics was quite deliberately
in the spirit of the circular organization so fundamental to the general theory of autopoiesis. Whether the circularity as expressed in this particular mathematical model is
really adequate to fully capture the circularity of autopoiesis is an open questionof
great importance, of course, but outside the scope of the present article. What we
may note, however, is that this model aimed at a good compromise between biological
realism on one hand, and maximal simplification on the other. All the components,
properties, and relations in the model were based on entities and processes that were
known to exist on grounds of actual biological observations; on the other hand, these
elements were represented in the model in the simplest possible form that would still
give rise to interesting emergent properties of the system as a whole.
4 Morphogenesis in Shape Space
In 1988, John Stewart joined the group that had formed around Varela and Coutinho at
the Pasteur Institute in Paris. His contribution was to run computer simulations, based
on the model that we have just described, in order to examine the emergent properties
of an idiotypic network. The immediate result of this was to reveal the necessity of
a method for specifying the structure of connectivity of an idiotypic networkmore
specifically, of defining the matrix of all possible pairwise affinities between a given set
of immunoglobulins. The experimental data on this point were (and still are) scanty
and quite insufficient; in addition, we needed a mode of representation that would
render the evolution of the connectivity structure as the system matured over time
graphically visible and comprehensible. We solved this problem by adopting a modified version of the shape space concept originally suggested by Perelson & Oster [9]
and developed by Segel & Perelson [11]. According to this concept, the universe of
stereochemical shapes that determine intermolecular affinities can be represented as
points in a multi-dimensional shape space. In our version, we used a two-dimensional
space (for obvious reasons of graphical visualization), and each point in shape space
was taken as representing a pair of perfectly complementary shapes with maximum
affinity (conventionally, the members of a pair are labeled black and white). This mode
of representation has the following advantages: firstly, relations of similarity in molecular shape (and hence affinity profiles) are immediately perceptible as the proximity of
corresponding points of the same color in the shape space; and secondly, relations of
complementarity (and hence high affinity) are also immediately perceptible in the form
of proximity between black and white points. The generation of random immunoglobulins as candidates for meta-dynamical recruitment was then quite straightforward: it
was effected by generating black or white shapes at random positions in the total
shape space. Mathematically, the affinity between two clones i and j , mij , is defined
as follows:


0 if i and j are the same color;
mij =
(1)
exp(dij2 ) if i and j are of opposite colors
where dij is the scaled distance between i and j in shape space.
The dynamics were deliberately reduced to their simplest expression: at each time
step in the simulation, a clone was maintained at unit concentration if its received field
lay within the window, and eliminated otherwise. At each time step, candidate clones
were generated randomly until one new clone was recruited.
Artificial Life Volume 10, Number 3

267

J. Stewart and A. Coutinho

The Affirmation of Self

Figure 4. Schematized results of a simulated morphogenesis in shape space. Lymphocyte clones are represented by
black and white circles. Clones of the same color have no affinity with each other. Clones of one color create a
field for clones of the opposite color that are close to them in shape space. (a) The self-organization of clones in
the absence of external antigens. The clones form chains of the same color, which face chains of the opposite color.
In the region between chains of opposite color, shaded with dots, the field is high; in the region enclosed by chains
of the same color, the field is low; the clones are all situated in the region where the field has an intermediate value
within the window of activation (see Figure 2). (b) The adjustment of the emergent pattern induced by the presence
of two antigens, represented by a white square and a black square. It can be seen that the patterns of Figure 4a are
adjusted so that the antigens are included in chains of the same color.

Using this procedure,3 we very rapidly obtained some promising results [12]. Firstly,
we showed that under these conditions, a self-sustaining idiotypic network could
arisewithout either collapsing or exploding. We quite deliberately started by studying
the behavior of the system in the absence of external antigens, in order to characterize
its eigen-behavior. The idiotypic network did indeed exhibit interesting properties of
self-organization: as can be seen in Figure 4a, the combined dynamic and meta-dynamic
process gives rise to clear patterns in shape space: there are chains of lymphocytes of
the same color, and the chains of complementary shapes mutually sustain each other.
We can consider that these patterns correspond to the identity of a molecular self as
defined and indeed constituted by the autonomous dynamics of the immune system
itself.
Secondly, we studied the modulation of this eigen-behavior when the system was
perturbed by the introduction of external antigens, modeled here as points in shape
space, which produced a field for the lymphocytes, but whose own concentration was
constant irrespective of the field they themselves received. Typical results are shown
in Figure 4b. What we see is that the idiotypic network adjusts smoothly so as to
integrate the antigens harmoniously into its own pattern of behavior; in other words,
the antigens are effectively assimilated as a part of the molecular self constituted by the
network. More precisely: We see in Figure 4b that all the antigens of a certain color
3 Technically, it may be noted that these results were obtained by computer simulation. It would be practically impossible to obtain
these results analytically, not only because the meta-dynamical process of recruitment and disappearance of clones changes the list
of variables, but because the process itself depends on relations between individual clones distributed in shape space and these
relations are historically contingent during the ontogeny of the system. In general, models that can only be resolved by simulation
are at a strong disadvantage because they are difficult to interpret, compared to mathematical models based on differential
equations; the latter, even if they cannot be solved analytically but only by numerical methods, can be described qualitatively in
terms of attractor dynamics. This latter resource is not available in the case of simulation models. However, in the present case
this was not an overwhelming difficulty: as shown by the subsequent discussion, it proved possible to interpret the emergent
patternsprobably thanks to the choice of an appropriate mode of graphical representation.

268

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

(black or white) are surrounded by lymphocytes of the same color. We know that all
the lymphocytes in a chain of a certain colour receive a field (from lymphocytes in
the facing chain of the opposite color) that is within the limits of the window (if this
were not the case, the lymphocytes would already have been eliminated). Since, under
these network conditions, the antigens receive the same field as the lymphocytes of
the same color that surround them, there is an important corollary: the field received
by the antigens remains within the limits of the window, that is, it is at most equal
to the upper threshold. Thus, if we suppose (as is reasonable) that the destruction of
antigens is only triggered by fields well above the upper threshold, in the presence of
an idiotypic network antigens will not be destroyed, but will be tolerated.
Thirdly, we can compare these results with what happens if the idiotypic network is
abolished. This would be difficult to realize experimentally in a real biological situation,
but in the model it can be achieved by a stroke of the penfor example, by recruiting
lymphocytes of only one color, which have zero affinities with each other. In this
case, the lymphocytes are activated only by the antigens; consequently, lymphocytes
complementary to the antigen are recruited without limit, and the field received by the
antigens increases indefinitely until it reaches levels that we may suppose do trigger
destruction of the antigens.
The results of these computer simulations contributed to a renewal of interest in
network ideas. When Jerne [6] first presented his concept of an idiotypic network in
1974, the idea received quite a favorable reception from the community of immunologists. However, over the years, the idea gradually fell into disrepute. What seems to
have happened is this. As we have seen, classical immunology is centered around the
phenomenon of strong, destructive immune responses to external antigens. Thus it was
quite natural that in the first-generation models of the immune network, the aim was to
make the network produce immune responses. However, the result of these attempts
was general failure: there seemed to be no way that an idiotypic network could be got
to produce a good immune response.4 Retrospectively, it seems clear that this failure
stemmed from the fact that the first-generation models were trying to make the network
do exactly the wrong thing. In order to produce classical immune responses, Burnets
mechanism consisting of the selection of unconnected clones is both straightforward
and perfectly adequate; at this level, a network organization is not only unnecessary
but actually counterproductive, because the network prevents the development of a
strong immune response. A much more appropriate role for the immune network, for
which its natural emergent properties are an advantage rather than a handicap, is to
promote tolerance by protecting the antigens of the body from attack by the immune
system. These considerations led Varela and Coutinho [15] to make a proposal for
second-generation immune networks, whose distinctive feature is that the immune
system is composed of two complementary compartments: the central immune system
and the peripheral immune system.
5 The Central Immune System and the Peripheral Immune System
It is important to note that these theoretical considerations were undertaken in close
conjunction with the ongoing experimental work in Coutinhos laboratory at the Pasteur
4 Theoretical immunology was, and still is, a flourishing field with a strong international community [10]. Around 1990 there was
a certain interest in idiotypic networks. However, as explained in the text, mainstream work in this area focused on immune
responses, immune memory, and vaccination. The work of Perelson and others showed that vaccination could only be obtained
by mini-networks composed of groups of two or three clones unconnected to each other. Global network connectivity gave rise
to results that could not be interpreted in this classical framework; this is poignantly expressed by the title of an article by de Boer
and Hogeweg [4]: Unreasonable implications of reasonable idiotypic network assumptions. On the other hand, when Perelson
and de Boer studied the emergent properties of an idiotypic network, without presupposing that the network should mediate
immune responses, they obtained results comparable to those presented here [5].

Artificial Life Volume 10, Number 3

269

J. Stewart and A. Coutinho

The Affirmation of Self

Institute. In particular, there was great interest in the so-called natural antibodies, the
circulating immunoglobulins that are found in the sera of all normal vertebrates even
when they are secluded from all antigenic contact with the environment. These natural
antibodies are produced in a seemingly spontaneous manner, and thus appear to be the
result of the autonomous internal activity of the immune system. Following previous
work of Avrameas at the Pasteur Institute [1], Coutinhos group quickly found that
these antibodies bind to self (i.e., antigens of the body) and are often multi-reactive.
Further work demonstrated that increases or decreases in the concentration of certain
specific natural antibodies had an influence on the natural antibody repertoire as a
whole. The natural antibodies are produced by naturally activated lymphocytes, which
represent about 10% of total lymphocyte numbers; the remaining 90% are resting cells,
which are mitotically inactive, do not secrete immunoglobulins, and are thus devoid of
effector functions. Fitting nicely to all these facts were the observations, independently
produced in John Kearneys [8] and Coutinhos groups, that spontaneously produced
antibodies often reacted with each other, while the connectivity of antibodies produced
upon activation of naturally resting lymphocytes was orders of magnitude lower.
The second-generation immune network model arose by putting together these empirical observations with the theoretical considerations outlined above. Every day, the
bone marrow produces a large number of new lymphocytes, each of which carries a
unique immunoglobulin receptor. The numbers produced are so large that the total
population of lymphocytes can be replenished in a few days; in addition, the repertoire
of these new lymphocytes is complete in the sense defined above, that is, it covers the
totality of all possible molecular shapes (of an appropriate size). If these new lymphocytes are not stimulated, they remain in a resting state and die after 2 or 3 days.
However, the rate of production is such that at any one time, these resting cells make
up 90% of total lymphocyte numbers. These resting cells constitute the peripheral
immune system (PIS); they have no functional idiotypic connections. The central immune system (CIS) is composed of the 10% of lymphocytes that are naturally activated;
according to the model, this activation is primarily the result of idiotypic interactions
between these lymphocytes, so that they form a connected network. As predicted by
the computer simulations, the repertoire of the CIS incorporates all the antigens of the
body of the organism that are permanently present. Also in line with the computer
simulations, it is the fact that body antigens are included in the repertoire of the CIS
with a network organization that protects them from immune attack and thus accounts
for the phenomenon of tolerance.
It is to be noted that according to this model, the two compartments, CIS and PIS,
are complementary. The CIS is composed of lymphocyte clones that have been rescued from death within 2 or 3 days (their fate if they had remained in the PIS) by their
meta-dynamical recruitment into the CIS. We may recall that the repertoire of lymphocytes freshly emerged from the bone marrow is complete. Hence, by construction, the
repertoire of the PIS is complete minus the repertoire of the CIS. Since the repertoire of
the CIS includes all the body antigens, it follows that to a first approximation (but we
shall have occasion to return to this point) the repertoire of the PIS is complete minus
body antigens. Since the lymphocytes in the PIS are isolated (unconnected by network
interactions either with each other or with the CIS), if they are stimulated by a novel
antigen (for example belonging to an invading microorganism), they will mount an
unfettered immune response. Thus, the PIS is ideally constituted both by its repertoire
and by its mode of functioning to the role of protecting the organism from pathological
microorganisms. The relationship between the CIS and the PIS is illustrated in Figure 5.
Conceptually, a possible extension of this notion is that the PIS includes lymphocytes that are rescued from death by minimal affinity to the idiotypic network, but
that affinity is too low to promote their activation to immunoglobulin secretion and
270

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

Figure 5. A schematic illustration of the relation between the central immune system (CIS) and the peripheral
immune system (PIS). The bone marrow continually exports large numbers of new lymphocytes. Over several days,
the repertoire of these lymphocytes is complete, that is, they recognize the totality of possible molecular shapes (of
an appropriate size). A small proportion of these new cells are recruited into the CIS, defined as a set of clones
with mutual affinities that sustain each other to form a connected network. The remainder of the new cells form a
residue of clones that are not stimulated and that will die after a few days; this residue constitutes the PIS. It follows
that the repertoire of the PIS is complete minus the CIS.

consequent participation in building the network itself. It follows that the CIS-PIS organization makes it possible to maintain a set of lymphocytes that survive as resting cells,
available to stimulation by microbial antigens (non-self), but capable of producing immunoglobulins that are potentially within the limits of network control. This explains
the fact that a number of conventional clonal immune responses may be submitted to
natural idiotypic regulation. In addition, this twist may well provide the mechanism for
selection of the long life spans of a small fraction of all lymphocytes that participated
in conventional responses, such that they are currently called memory cells. In other
words, memory of non-self is also a property of the network, which maintains alive
some specific cells even after the antigen has been eliminated from the body. However, although these extensions are attractive and plausible for experimentalists, they
were never implemented in the computer simulations; to that extent, their theoretical
validity remains uncertain.
We may now compare this second-generation network model with the scheme of
classical immunology. In a certain sense, the distinction between the CIS and the PIS
corresponds to the classical distinction between self and non-self. This is comforting,
and means that the new network view renders unto Caesar that which is Caesars. However, there remains a major difference from classical immunology, which is interesting.
The difference is that in classical immunology, tolerance to body antigens results from
eliminating all the lymphocytes that interact with themthe so-called clonal deletion
theory. According to the second-generation network model inspired by the concept of
autopoiesis, however, tolerance to body antigens is the result of a positive process: the
lymphocytes that interact with the body are not eliminated, but on the contrary they are
Artificial Life Volume 10, Number 3

271

J. Stewart and A. Coutinho

The Affirmation of Self

activated by being incorporated into the dynamics of an idiotypic network. This has
a number of consequencesconceptual, experimental and practicalwhich are worth
spelling out.
The conceptual difference is that the category of self is not defined by an external
human observer on the basis of knowledge that is intrinsically inaccessible to the immune system. In this new conception, self is first and foremost defined by the immune
system itself on the basis of its autonomous functioning as a self-sustaining idiotypic
network. It is only subsequently that the body antigens are incorporated into the repertoire of this network. The body antigens are not intrinsically self as such (and even less
because they are decreed to be self by an immunologist); they become self by virtue
of being assimilated into an immunological self that has already been constituted by
the autonomous operation of the immune system. This notion has a sound empirical
demonstration in a series of experiments carried out in the late 1980s by Le Douarin
and colleagues in Paris [3]. Concerned with analyzing cell fate in embryonic development, this group conducted tissue grafts between embryos of related, yet different,
species and strains, at a stage that preceded the existence of lymphocytes. Whereas
conventional theory would have predicted that such graftspresent in the body from
before the development of the immune systemwould be taken as part of self and thus
tolerated, Le Douarins group discovered that all embryonic grafts were rejected once
the animals reached immunological maturity. This observation was most surprising,
and it was at the origin of the current concepts of dominant tolerance (see below),
demonstrating that incorporation into the immunological self only occurs when the
grafted tissues share critical antigens with the cells that either make up the immune
system (bone marrow) or ensure its development and self-restricted operation (thymic
epithelial cells). We may also note here that even if we accept that the body antigens
are normally incorporated into the immunological self, it does not follow that the immunological self reduces to just the body antigens. As already mentioned, we shall
have occasion to return to this point.
The experimental difference is this. On the classical view, tolerance is due to the
elimination of lymphocyte clones that interact with the antigen in question. Thus, if
a hybrid immune system is produced experimentally, tolerance should be recessive
(i.e., a hybrid between a tolerant and a non-tolerant system should be non-tolerant).
On the new view, tolerance is due to the positive effects of a functional network;
thus, on condition that the hybridization is carried out in such a way that the network
is not disrupted, tolerance should be dominant (the hybrid, or chimeras containing
a mixture of tolerant and non-tolerant cells, should be tolerant). Without going into
details, we note that many experiments have been performed that amply demonstrate
the phenomenon of dominant tolerance. It suffices to say that, after a few decades
of great predominance of recessive-tolerance theories, it is now widely accepted that
natural tolerance is in fact dominant.
The practical, clinical difference is this. On the classical view, autoimmune disease
arises because the immune system is functioning overzealously; it is therefore quite
logical to treat autoimmune disease by immunosuppression. The results are generally
not very satisfactory: immunosuppressive treatments are, at best, symptomatic, and
may have serious side effects. To date, there are no observations indicating the cure
of autoimmune patients by such treatments. On the new view, autoimmunity arises
from a deficiency in the normal ongoing activity of the immune system (and, quite
apart from treatment, it is a widespread clinical and experimental observation that
immunodeficiency is indeed often associated with autoimmunity); the logical treatment
thus consists of an (appropriate) activation of the immune system. In line with this
prediction, the treatment of autoimmune disease by the injection of a balanced mixture
of normal serum immunoglobulins has had some very positive results [7].
272

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

6 Subsequent Developments
The second-generation network models described in the previous section had one
major gap. What our work showed was that if the immune system functioned in
a network mode, then the antigens that fall within its repertoire will be integrated
into the network dynamics and will hence be tolerated (this is the basis of the CIS);
whereas if the immune system functioned in a non-network mode without idiotypic
connections, then the antigens that fall within its repertoire will provoke an immune
response leading to their destruction (this is the basis of the PIS). However, the major
issue that remained quite unresolved was to specify how the distinction between the CIS
and the PIS actually came about. In our first simple models, illustrated in Figure 4, the
network spread over the whole available shape space (the self-organized patterns only
arose when the whole shape space was saturated), thus leaving no room for a residual
PIS. It is true that if we abolished the network interactions by simple fiata hand
of God, as it werethen the system would function in a PIS mode. However, this
sort of intervention (divine or otherwise), coming from outside the system, was quite
contrary to the spirit of autopoiesis, where the whole point is to explain phenomena
as resulting from the autonomous operation of the system itself. To be more precise,
what was missing was an account of how the CIS-PIS distinction (the successor to the
classical selfnon-self distinction) could arise through the autonomous ontogeny of the
system itself.
After 1993, this problem was tackled with great energy and imagination by Jorge
Carneiro, at that time a PhD student at the Pasteur Institute. Carneiro came to the
conclusion that this problem could only be solved by extending the model to include
not only the B lymphocytes, which produce immunoglobulins, but also macrophages,
which present digested fragments of antigens via the MHC molecules on their cell
surface, and the T lymphocytes, which recognize the MHC-presented antigens and
provide help to B lymphocytes. Without going into the details [2], we state that Carneiro
came up with an aesthetically beautiful model that exhibited the emergent properties we
were looking for. The first condition was that the ontogeny of the system should start
with a sufficiently numerous and diverse set of antigens (these plausibly correspond
to the body antigens already present when the immune system starts developing in
embryonic life, but we will refer to them more prudently here as initial antigens). On
this condition, the immune system went through several phases in its ontogeny. Firstly,
T cells were activated, and these in turn recruited B cells. On condition that the initial
set of antigens was sufficiently numerous, this process continued until the repertoire
of the B cells was complete, and so the B cells started to exert a regulatory feedback
influence on the T cells. At this stage, the second phase commenced: now that the
system was saturated and complete, competition set in amongst the B cells for the T-cell
help that they themselves were limiting. Under these conditions, the repertoire of the
B-cells shrank until only those with a B-cell receptor (BCR, an immunoglobulin) that
directly recognized a T-cell receptor (TCR) remained. It is aesthetically pleasing to note
here that since the TCRs are a mirror of the initial antigens, and these BCRs are a mirror
of the TCRs, the BCRs are a mirror of a mirrora sort of internal image of the initial
antigens. Thus, the B-cell repertoire in the mature CIS was far from complete, and in
fact was restricted to an internal image of the initial antigens. This set the stage for the
third phase in the ontogeny. If a single novel antigen, which the immune system had
not seen so far, was now presented to the system after the maturation of the first two
phases, then the global regulatory processes that occurred with the initial set of antigens
were not reproduced. The reason is that the B-cell repertoire was now tightly restricted
to an internal image of the initial antigens, and had virtually no chance of percolating
to include a BCR that would regulate the T cell activated by the novel antigen. The
Artificial Life Volume 10, Number 3

273

J. Stewart and A. Coutinho

The Affirmation of Self

result was that the immune system would mount an unfettered immune response to
the novel antigen.
Unfortunately, several straightforward predictions of this model are refuted by the
observations: for example, -knockout mice that have no B-cell receptors do not exhibit uncontrolled T-cell proliferation. Even more seriously (for after all, a refutation
can always be overcome by an appropriate adjustment and/or subsequent experimental findingsit should be said that very recent observations do show alterations in the
development of TCR repertoires in the immune system of mice that produce no immunoglobulins), this model has not given rise to experimental observations that would
validate it positively. The most serious problem is that it is not at all sure that idiotypic interactions between the highly variable BCRs and TCRs have any physiological
significance at all. Jernes initial argument in favor of idiotypic networks was purely
logical and qualitative: if the immunoglobulin repertoire is complete, then there must
logically exist interactions between immunoglobulins. Incontrovertible as far as it goes,
this is not enough to show that the concentrations and affinities involved are sufficient
for these interactions to have real physiological effects. Thus, for the moment at least,
idiotypic network models have fallen again into the shade.
7 Epilogue
Francisco Varela left immunology in 1993, to return in a more focused way to brain
biology, which he had never left either in the laboratory or in his mind. His feeling was that neuroscience, to which he brought the highly original contribution of
neurophenomenology, was more likely to provide the system best able to test and illustrate his general, fundamental ideas. Interestingly, one of his major contributions
to neurosciencethe demonstration that a neural network selects what it pays attention to, which in turn reverberates through a large part of the system and adds to the
memory of past experiences in selecting what we next pay attention tois closely akin
to the second-generation immune networks that he offered to immunologists. So, ten
years later, what has become of Franciscos contribution to immunology?
An interesting achievement of this work in immunology is to show that the concepts of autopoiesis can be employed to generate clear novel hypotheses and produce
models demonstrating these ideas. The concepts of autopoiesis have been criticized
on the grounds that since they are conceived as a tight logical chain of definitions and
implications, it is often not clear whether they are indeed a scientific theory or rather
just a potential scientific vocabulary of doubtful utility to working scientists. Here we
see that ideas inspired by autopoiesis can provide clear and testable predictions and
suggest novel therapeutic procedures.
More precisely, as regards the field of immunology itself, we have noted that idiotypic
network models are not currently in vogue. However, whatever its ultimate empirical
validity, the culminating network model of Carneiro provides a very good illustration
of an important epistemological point raised at the end of Section 2 above. This model
provides a mechanism, hypothetical certainly but quite precise and explicit, as to how
the demarcation between the CIS and the PIS could actually come about in the course
of an autonomous ontogeny, without the providential help of an omniscient external
agent. And to all practical intents and purposes, this demarcation does roughly coincide
with the classical immunologists selfnon-self distinction. However, if the mechanism is
indeed akin to that proposed by Carneiro, the actual operational distinction is between
(a) a sufficiently numerous set of initial antigens, present from the start of ontogeny, and
(b) single antigens that are first presented to the system after two successive phases of
maturation. To call this a selfnon-self distinction obscures the issue by presupposing
what it ought to be the job of scientific investigation to explainrather like Molieres
274

Artificial Life Volume 10, Number 3

J. Stewart and A. Coutinho

The Affirmation of Self

doctors, who grandiloquently claimed that the reason why opium makes you sleep is
that it contains a dormitive principle. If Carneiros model is not the right one, it will
have to be replaced by an alternative account with the same operational capacity to
explain how the CIS-PIS distinction does come about.
The focus of current research on the phenomena of tolerance to body antigens has
shifted to the so-called regulatory T cells. However, this recent work is complementary
rather than contradictory to the basic insights that came from the second-generation
network concepts. Firstly, the work on auto-reactive regulatory T cells has yet to address
the questions of repertoire formation and the 20-year-old observations on reciprocal
and recursive selection amongst naturally activated B and T cells. Secondly, and even
more important, the basic renewal of perspective that came from the deployment of
the concepts of autopoiesis in the field of immunology has become irreversible. It
is now clear that tolerance to body antigens is dominant, and so it cannot possibly
be explained on a one-by-one clonal basis and certainly not by the elimination of
lymphocytes that interact with the body. On the contrary, tolerance is primarily the
result of a positive, active process involving lymphocytes that interact productively
with the body; these lymphocytes probably include the newly fashionable regulatory
T cells but are not necessarily restricted to them. To be sure, the concept of dominant
tolerance as actually adopted by the immunological community is a rather watereddown version of what Varela proposed, and falls short of adopting the autopoietic
perspective as such. Thus the intellectual potential for a paradigm shift, clearly present
in Varelas ideas, has not fully materializedto date, at any rate. However, the future is
not closed. Francisco liked to say that the central, primordial function of the immune
system is that of self-assertion; and this insight, today more than ever, holds promise
for the future.

References
1. Avrameus, S. (1991). Natural autoantibodies: From horror autotoxicus to gnothi seaton.
Immunology Today, 12, 154159.
2. Carneiro, J., Coutinho, A., Faro, J., & Stewart, J. (1996). A model of the immune network
with B-T cell co-operation. I. Prototypical structures and dynamics. II. The simulation of
ontogenesis. Journal of Theoretical Biology, 182, 513530, 531548.
3. Coutinho, A., Salaun, J., Corbel, C., Bandeira, A., & Le Douarin, N. (1993). The role of
thymic epithelium in the establishment of transplantation tolerance. Immunological
Reviews, 133, 225240.
4. De Boer, R. J., & Hogeweg, P. (1989). Unreasonable implications of reasonable idiotypic
network assumptions. Bulletin of Mathematical Biology, 51, 381408.
5. De Boer, R. J., & Perelson, A. S. (1991). Size and connectivity as emergent properties of a
developing immune network. Journal of Theoretical Biology, 149, 381424.
6. Jerne, N. K. (1974). Towards a network theory of the immune system. Annales
dImmunologie (Institut Pasteur), 125C, 373389.
7. Kazatchkine, M. D., & Morell, A. (Eds.) (1997). Intravenous Immunoglobulin Research and
Therapy. CRC Press.
8. Kearney, J. F., Vakil, M., & Nicholson, N. (1987). Non-random V-gene expression and
idiotype-antiidiotype expression on early B cells. In G. Kelsoe and D. Schulze (Eds.),
Evolution and vertebrate immunity: The antigen receptor and MHC gene families (pp.
175190). Austin: Texas University Press.
9. Perelson, A. S., & Oster, G. (1979). Theoretical studies of clonal selection: Minimal antibody
repertoire size and reliability of self non-self discrimination. Journal of Theoretical Biology,
81, 645670.
Artificial Life Volume 10, Number 3

275

J. Stewart and A. Coutinho

The Affirmation of Self

10. Perelson, A. S. (Ed.) (1988). Theoretical Immunology, Vols. I, II. Reading, MA:
Addison-Wesley.
11. Segel, L. A., & Perelson, A. S. (1989). Shapespace: An approach to the evaluation of
cross-reactivity effects, stability and controllability in the immune system. Immunology
Letters, 22, 91100.
12. Stewart, J., & Varela, F. J. (1991). Morphogenesis in shape-space: Elementary metadynamics
in a model of the immune network. Journal of Theoretical Biology, 153, 477498.
13. Varela, F., Coutinho, A., Dupire, B., & Vaz, N. M. (1988). Cognitive networks: Immune,
neural and otherwise. In A. S. Perelson (Ed.), Theoretical Immunology, Vol. II (pp.
357374). New York: Addison-Wesley.
14. Varela, F., & Stewart, J. (1990). Dynamics of a class of immune networks. I. Global stability
of idiotypic interactions. II. Oscillatory activity of cellular and humoural components.
Journal of Theoretical Biology, 44, 93102, 103115.
15. Varela, F. J., & Coutinho, A. (1991). Second generation immune networks. Immunology
Today, 12, 159166.
16. Vaz, N. M., & Varela, F. (1978). Self and non-sense: An organism-centered approach to
immunology. Medical Hypothesis, 4, 231267.

276

Artificial Life Volume 10, Number 3