news and views

were still restricted to zero-sum games, that

MARK DOBSON

is, where one player’s loss is exactly the other’s

gain. Non-zero-sum games have more gener-
al payoffs, allowing both players to win in
some outcomes, and thereby encouraging
some degree of cooperation. But the question
of whether such games also could be solved
eluded researchers for 20 more years.
Finally, Vieille has proved that they can
all be solved. It is an extremely long
proof, which spans a series of four difficult
8
papers5–8. It is not a constructive proof,
in the sense that in most cases one would
not be able to actually give the solution,
as I did in the preceding story: one would
just know that it exists. However, it is an
outstanding achievement, requiring ex-
tremely sophisticated tools from all sides
of mathematics, from ergodic theory to
algebraic geometry. It gives us a new class of
models with which to study conflict and
cooperation — models that will certainly
find their way into the study of social inter-
actions and economics.
Ivar Ekeland is at the University of Paris-Dauphine,
75775 Paris, Cedex 16, France.
e-mail: [Link]@[Link]
1. Gillette, D. Ann. Math. Stud. 39, 179–187 (1957).
shoes. So if, for instance, there were a way for back and throws dice to that effect. As for the 2. Blackwell, D. & Ferguson, T. S. Ann. Math. Stat. 39, 159–163
(1968).
the minister to reach a definite conclusion as minister, his best policy is to flip a fair coin
3. Mertens, J. F. & Neyman, A. Int. J. Game Theory 10, 53–66
to whether he should work or shirk today, the every morning and to work if it comes up (1981).
king, having precisely the same information tails. Note that, as long as A is close to B, the 4. Mertens, J. F. & Neyman, A. Proc. Natl Acad. Sci. USA 79,
(namely, what the minister did yesterday and probability of the king returning today is 2145–2146 (1982).
the days before), would reach the same con- close to the constant 1/N 2. 5. Vieille, N. Int. J. Game Theory 21, 395–404 (1993).
clusion, and take advantage of it — returning After this solution, things stalled for ten 6. Vieille, N. 2–Players Stochastic Games 1: A Reduction, preprint,
Univ. Paris-Dauphine (1997).
if the minister planned to shirk, and staying more years, until Mertens and Neyman3,4 7. Vieille, N. 2–Players Stochastic Games 2: The Case of Recursive
away if he planned to work. The way out is to extended the result to games with more than Games, preprint, Univ. Paris-Dauphine (1997).
resort to probabilistic strategies: as he wakes two possible outcomes, and where chance 8. Vieille, N. Large Deviations and Stochastic Games, preprint,
up, the minister decides on certain probabil- comes in more complicated ways. But they Univ. Paris-Dauphine (1997).
ities with which to work or shirk, and then
casts lots to find out what he will actually do. Immunology
Similarly, the king will throw dice every
morning to find out whether to return or to Licence to kill
stay away, only needing to decide the odds for
Antonio Lanzavecchia
returning or staying away.
This problem was first stated by Gillette1
in 1957. Its solution depends on the values hen the body’s own cells become The need for helpers and the nature of this
the minister and the king attach to the
different outcomes, and also on the way they
value the future, compared with the present.
W infected, the immune system can
take the extreme step of killing
them to preserve the rest of the organism. It is
help have intrigued immunologists for years.
The first model (Fig. 1b) suggested that
T-helper and T-killer cells recognize their
In 1968, Blackwell and Ferguson2 found the no surprise that this potentially dangerous specific antigens simultaneously on the same
solution to the particular case when the king mechanism has to be kept under strict con- APC, and that cytokines (such as interleukin-
and the minister have exactly opposite inter- trol, and on pages 4741, 4782 and 4803 of this 2) produced by the activated T-helper cell
ests (say, for instance, that if the minister is issue, three groups suggest a new mechanism would act on the T killer and facilitate its
caught, he can stay out of jail by paying every by which this control operates. response4,5. Although interleukin-2 can sub-
day a large amount of money M to the king, The immune system’s T-killer cells exist stitute for helper cells in some systems, there
and that if the king is caught out, he can keep as inactive precursors that must be activated are two main problems with this model. First,
his kingdom by paying the minister this same to develop killing capacity. Activation entails it requires that two rare antigen-specific cells
amount M every day). They proved that, if recognition of antigenic peptides bound to meet on the same antigen-bearing APC — an
everyone holds a long-range view, the best class I molecules of the major histocompati- event that has an extremely low probability.
strategy for the king is to choose at the outset bility complex (MHC) on the surface of anti- Second, and more importantly, some killer
some large number N, and then to compute gen-presenting cells (APCs). But this precur- responses can be elicited in the absence of
every morning the number D = N + A – B, sor–APC interaction is often not enough to helper cells6, implying that the need for help is
where A is the number of days the minister stimulate the T-killer cell (Fig. 1a, overleaf) conditional rather than absolute.
has worked since the game began and B the and, in these cases, a T-helper cell that can The alternative theory, proposed by
number of days he has shirked. The king recognize an antigen on the same APC is also Polly Matzinger on theoretical grounds, is a
then assigns a probability p = 1/D 2 to coming required4. ‘licensing’ model. Matzinger suggested that
NATURE | VOL 393 | 4 JUNE 1998 413
Nature © Macmillan Publishers Ltd 1998
news and views
by recognizing antigen on APCs, T-helper Figure 1 Antigen-presenting cells
cells deliver a signal that activates the APCs7. a b
Interleukin-2 (APCs) need a licence to help
The licensed APC can then directly stimulate T killer T helper
T-killer cells. a, T-killer
T killer
T-killer cells (Fig. 1c). The key to this model is precursors recognize antigen on
a specialized ‘professional’ APC, which has resting APCs, but do not receive
been identified as the dendritic cell. Dendritic enough signal to be activated.
cells sample antigens in all body tissues and Resting APC
APC b, According to the traditional
migrate to lymph nodes where they present
8
model, T helpers and T killers
antigens to T cells8. The molecules responsible recognize antigen on the same
for the interaction between the T-helper and c APC. The helper is activated and
dendritic cells are called CD40L and CD40. T helper
produces interleukin-2 which
CD40L is a membrane molecule expressed by contributes to activation of the
antigen-stimulated T-helper cells. It interacts CD40L T killer killer. c, In the new model,
CD40
with and triggers CD40, a surface receptor supported by the experiments of
that can activate dendritic cells9–11. APC Ridge et al.1, Bennett et al.2 and
The three new papers support this model Schoenberger et al.3, the APCs are
of T-killer-cell activation. Bennett et al.2 and Activated APC
licensed to activate T-killer cells
Schoenberger et al.3 show that mice lacking Tumour—necrosis factor—a by T helpers or by other stimuli.
T-helper cells cannot mount a killer response Lipo—
polysaccharide APC
when they are injected with APCs that
display an antigen recognized by T-killer Virus
cells. Remarkably, however, the authors
could induce a killer response by injecting optimally. In their resting state, they possess Thus, rather than being chaperones that
the mice with an antibody against CD40 — low levels of MHC and co-stimulatory molec- bring T helpers and T killers together, the
by binding CD40 and triggering the APCs, ules, so they are poor at stimulating T-killer dendritic cells have now become active pro-
the anti-CD40 antibody acts as a surrogate of cells. Resting dendritic cells can be activated tagonists that are licensed to activate killer
T-helper cells. Moreover, Schoenberger et al. by inflammatory cytokines and bacterial cells. Stimuli from pathogens, inflammatory
noticed that, in normal mice, they could products that upregulate MHC and the cytokines and T-helper cells are integrated in
block the induction of a helper-dependent co-stimulatory molecules, thereby increasing dendritic cells, which ultimately tune the
T-killer response using antibodies to CD40L. the ability to stimulate T cells9. Furthermore, capacity to stimulate T-cell responses. It
This inhibition was reversed by adding the activation of dendritic cells by CD40 makes sense to have different ways to activate
stimulatory anti-CD40 antibody. Bennett results in increased production of cytokines, dendritic cells so that an appropriate
and colleagues also showed that mice lacking especially interleukin-12 (refs 10, 11). The response to any offending agent can be
either CD40 or CD40L cannot mount latest findings1–3 can thus be explained by the mounted effectively.
helper-dependent T-killer responses. fact that co-stimulatory molecules and inter- The possibility of manipulating the
To define the stimuli that enable the APC leukin-12, which are well-known require- activation of dendritic cells will open new
to activate T-killer cells, Ridge et al.1 manipu- ments for activation of T-killer cells, need to avenues for vaccination and therapy. By
lated dendritic cells in vitro. They show be induced in dendritic cells by T-helper cells stimulating dendritic cells, the response to
that dendritic cells carrying specific antigen and inflammatory stimuli. weak antigens such as tumour-specific
cannot stimulate T-killer responses in vitro The licensing model reconciles, in a sim- antigens (which are normally encountered
or in vivo unless they are first stimulated by ple concept, many apparently contrasting outside an inflammatory context) may be
anti-CD40 antibodies or specific T-helper observations. For instance, it explains how boosted. This could be done using anti-
cells. The interaction between T-helper cells the response to the same antigen can be T- CD40 antibodies, as proposed by Ridge et
and APCs can be dissociated in time from the helper dependent or independent. In the al.1, Bennett et al.2 and Schoenberger et al.3.
interaction between APCs and T killers. So, presence of adjuvant (which Charlie Janeway Alternatively, as others have suggested11,
the rare antigen-specific T cells do not need calls “the little dirty secret of immunolo- dendritic cells could be pulsed with an anti-
to interact with APCs simultaneously — gists”12), the inflammatory cytokines gener- gen that is recognized by the T-helper mem-
they can do it sequentially. ated may be enough to activate dendritic ory cells, thereby causing them to be activat-
These results1–3 clearly illustrate that the cells to a state where they can autonomously ed in a timely fashion in lymphoid organs.
interaction between CD40L and CD40 can stimulate a T-killer response, even in the Antonio Lanzavecchia is at the Basel Institute for
be crucial in bringing the APCs to a state in absence of T-helper cells. On the other hand, Immunology, Grenzacherstrasse 487,
which they can autonomously trigger T- if antigen is presented by resting dendritic CH 4005 Basel, Switzerland.
killer responses. Interestingly, however, they cells, the T-helper-mediated CD40 trigger- e-mail: lanzavecchia@[Link]
also suggest that there may be alternative ing would be absolutely required. But this 1. Ridge, J. P., Di Rosa, F. & Matzinger, P. Nature 393, 474–478
(1998).
ways to activate APCs. Bennett et al. show model is still compatible with the possibility
2. Bennett, S. R. M. et al. Nature 393, 478–480 (1998).
that a T-killer response to the same antigen that particularly strong antigens might be 3. Schoenberger, S. P., Toes, R. E. M., van der Voort, E. I. H.,
may or may not depend on the T-helper able to stimulate T-killer responses, even in Offringa, R. & Melief, C. J. M. Nature 393, 480–483 (1998).
cells, according to whether the antigen is the absence of help or inflammation. 4. Keene, J. A. & Forman, J. J. Exp. Med. 155, 768–782 (1982).
5. Mitchison, N. A. & O’Malley, C. Eur. J. Immunol. 17, 1579–1583
administered on cells or in association with The licensing concept has further impli- (1987).
an adjuvant (a substance that induces in- cations for the generation of polarized 6. Buller, R. M., Holmes, K. L., Hugin, A., Frederickson, T. N. &
flammation). In addition, Ridge et al. show immune responses. Activation of the APCs Morse, H. C. Nature 328, 77–79 (1987).
7. Guerder, S. & Matzinger, P. J. J. Exp. Med. 176, 553–564 (1992).
that dendritic cells can also be activated after and production of interleukin-12 are 8. Banchereau, J. & Steinman, R. M. Nature 392, 245–252 (1998).
infection with influenza virus. required for the T-helper responses that pro- 9. Sallusto, F. & Lanzavecchia, A. J. Exp. Med. 179, 1109–1118
What happens to dendritic cells during tect from intracellular pathogens. Converse- (1994).
10. Caux, C. et al. J. Exp. Med. 180, 1263–1272 (1994).
this activation, and why do different pathways ly, suppressive T cells might inhibit or modu- 11. Cella, M. et al. J. Exp. Med. 184, 747–752 (1996).
lead to the same effect? It is well known that late the function of the dendritic cells, and in 12. Janeway, C. A. Jr Cold Spring Harb. Symp. Quant. Biol. 54,
dendritic cells need to be activated to perform this way interfere with the immune response. 1–13 (1989).

414 NATURE | VOL 393 | 4 JUNE 1998

Nature © Macmillan Publishers Ltd 1998