Status of Non-Clinical Methods for Autoimmunity-Presenter:Paul-Henri Lambert, M.D., Chief, Vaccine Research & Development, WHO
Workshop on Non-Clinical Safety Evaluation of Preventive Vaccines: Recent Advances and Regulatory Considerations
The Society of Toxicology Contemporary Concepts in Toxicology Section
U.S. Department of Health and Human Services
Food and Drug Administration
Office of Women's Health
Center for Biologics Evaluation and Research
Monday, December 2, 2002
DR. LAMBERT: Thank you very much.
First, I would like to tell you that my voice remains somewhere between Paris and Washington, and I apologize for adding this poor voice to my poor Belgian accent.
I promise you that I will not give you the answer to the question of whether vaccine can cause autoimmune disease.
What is the issue? First, we know that we have examples of confirmed vaccine-associated autoimmune diseases. I have listed here three of these examples. One is encephalitis associated with a rabies vaccine, this is the old rabies vaccine, the sheep brain; thrombocytopenia associated with MMR and measles vaccination; and the Guillain-Barre associated with the swine influenza vaccination.
And I attract your attention to the fact that the incidence of these complications is still low. Although it reached a level of 30 per 100,000 for rabies encephalitis, it goes down to 0.8 per 100,000 when we speak about swine influenza.
And this is probably one of the key issues when we speak about autoimmunity and vaccination: that we deal with low incidence of complications, which are extremely difficult to pick up in the clinical trials.
The second point is that there is an increasing incidence of some autoimmune diseases, and this is resulting with an increasing risk of coincidence with vaccination events.
Here I'll just take one example, which is Type I diabetes. And all over the world the incidence is increasing. Here we just have the European picture, where we see that there is an annual increase of incidence of about 6 percent in the group of children between zero and four years of age.
And in this picture of increasing incidence, we know that Type I diabetes is occurring at an age which starts from six months to 15 years of age, which is when we have all our vaccines given. So it's practically unavoidable that any case of Type I diabetes will occur some time after one vaccination event. And it's always very difficult to disprove an association between the two.
The third point is that several autoimmune diseases appear to be caused or exacerbated by infection. And just a few examples: We know that a number of bacterial infections can be associated with rheumatic heart disease, reactive arthritis, Guillain-Barre syndrome, chronic arthritis, also with a number of viral infections. We have association with ITP, idiopathic thrombocytopenia. And even diabetes has been associated with various infections, but an association which is not very good.
So the question is: What is the potential mechanism by which vaccines might induce autoimmunity and autoimmune disease? And what can we do, in terms of non-clinical assessment, in relation to this mechanism?
First, the question of molecular mimicry. We know that molecular mimicry means that there is a similar B- or T-cell epitope on the vaccine and on host antigen. If we speak first about the B-cell epitope mimicry, we have to consider first oligosaccharide, oligosaccharide epitopes. And here we have a good example, which is the Guillain-Barre syndrome which can occur after campularvector [ph] [inaudible] infection, and is associated with the development of antigangliocyte [ph] antibodies. This is clearly associated with the mimicry between some epitopes on the LPS of campularvector and neurogangliocytes with the [inaudible] the association of GT-1 with one of the LPS type, in green GM-1, which is the same type, and in rat another LPS with another type of gangliocyte. So this is clearly shown, this correlation between this mimicry and the development, which is frequent, of Guillain-Barre after campularvector infection.
It is clear that in this situation it would be probably extremely risky to base vaccine against campularvector on this type of molecule. And for this reason, we could say that this type of homology involving oligosaccharide epitopes can be sufficient to select out the vaccine antigen.
In this case we have to recognize that the knowledge of an association of infection with autoimmunity is of critical importance to take our decision. And we don't have that for everything.
If we go to the Group B meninges capsular polysaccharide vaccine, again we have an antigen which is very similar to capsular polysaccharide which is expressed in humans developing neural tissue. And this is this poly-alpha neurominocasein [ph]. In this case we have no known association of Men-B antibodies with any autoimmune manifestation. What do we do? It is a question which is still open today. And obviously, it's a situation where we may like to move to animal models.
And what can we get out of animal models? Well, the animal model in this kind of case can be used if the cross-reacting epitopes are conserved. It can be used to test the possibility to induce cross-reacting responses in animals, keeping in mind that we do not always see the antibodies because they can be absorbed on the host tissue. They can be masked in some way. We can also assess the pathogenicity, but we know that in this particular situation it has been extremely difficult to show anything.
If we move to protein epitopes, there again we know that on a microbial antigen, like on any protein, most B-cell epitopes are exposed on the surface, they are conformational, and they are discontinuous.
If we look at autoantigens and autoantibodies, it's exactly the same thing. B-cell epitopes are seen by autoantibodies, and they are seen as conformational; they are surface exposed; they are discontinuous. And this makes the problem for identification.
Here we have the example of the GAT-65 islet cell antigen of islet cells in the pancreas mapped. And that's here, this area. It represents the B-cell epitopes as they are identified with autoantibodies coming from Type I diabetes patients. So this is clearly not a question of simple sequence. These are antigens which are highly complex, and where the confirmation is essential.
So the question: Can one predict the risk of autoimmune disease comparing these two things? The study can be based first on "in silico" studies, computer studies; in vitro analysis; and then finally in vivo analysis.
So the first animal experiment I would like to speak about is the experiment with the computer mouse, in silico prediction. Well, here is the first thing which is usually done, is to search for sequence homologies. I would say that the search for extensive sequence homologies is still of importance, because it means something. But short peptide homologies have no significance, or little significance, when we speak about B-cell epitopes, in view of this importance of confirmation.
Then one can move to the identification of B-cell epitopes. And we have a number of algorithms which have been developed for that, looking at areas of low hydrophobicity, high hydrophilicity, high flexibility, sophistication, antigenicity. And these B-cell epitopes which are identified can be compared between the vaccine antigen and the human protein epitopes.
Well, let's be clear. This is feasible, but it's really feasible when we know which are the target proteins. And if you don't know what to look for, this is practically a nightmare.
If we move to in vitro analysis, here we can search for cross-reactive antigens on tissue secreted protein with specific antibodies which are induced with the vaccine antigen in animals or in humans in initial trials.
We can also move to in vivo studies. And there I would think that the important point is to look in vivo for the possible binding of these antibodies, for their pathogenicity during active or passive immunization experiments. And this obviously can be done if the cross-reacting epitope is present in the animal.
So B-cell epitope mimicry I would say that this is more a concern for oligosaccharide than for protein. And when we speak about protein, this is of particular importance, maybe of importance, for vaccines which are against diseases which are known to be naturally associated with antibody-mediated autoimmune manifestations.
We have also to keep in mind that autoantibodies do not mean autoimmune disease; and that to be pathogenic, autoantibodies must have access to target antigen, they must have functional or cytopathic effects, have a sufficient ability, or be able to form pathogenic immune complexes.
If we move to T-cell epitopes, here we know that T-cell epitopes are small, linear epitopes--small, linear peptides. The size is different if we speak about CD-4 epitopes or CD-8 epitopes, from 11 to 20 amino acids, to eight to ten. We know that some core amino acids are important and can be recognized and used to identify these epitopes.
We know also that some infection induced T-cells can be associated with an autoimmune disease. And here I list a few examples, which are rheumatic heart disease, chronic Lyme arthritis, or reactive arthritis where T-cells and corresponding epitopes have been identified.
So can one predict the risk of autoimmune response if there is a mimicking T-cell epitope on a vaccine? First, again, we go to the computer and search for sequence homologies with the human protein data bank, looking for small peptides which are homologous, six to nine [inaudible].
And there usually you get in despair. Because what you find is a large number of homologies. Here you just have an example from a study which was done by Joel Tonnard [ph], who gave me this data, where the frequencing of sequence similarities has been studied between tetanus toxin and 15 human proteins. And you can see that at the six [inaudible] level more than 200 human proteins have peptide similarities with tetanus toxin and tetanus toxoid. Even if you go to the eight [inaudible] level with one mismatch, you still have 95 proteins which have similarities. So if this would be important, no one could be immunized today with tetanus toxoid.
Then, the next step is to search for common T-cell epitopes, using all kinds of algorithms for epitope prediction--which we call classical. And the questions which I ask are, first, are these mimicking peptides likely to be appropriately processed by the antigen presenting set? And if we speak about CD-8 epitopes it will be the question of processing at the [inaudible] level. And can this processed peptide bind to the various HLA molecules; particularly, entering and being captured in the groove of this molecule.
Well, this has limitations. We can again get a number of results, but we know that advanced HLA binding predictions are now visible only for a few HLA alleles. So we are limited in what we do.
In addition, even as this is being done, again, we can find quite a number of similar peptides on unrelated protein, predicted to bind to the same HLA allele. And taking just an example, again, tetanus toxoid, if we look for one TT DRB-1 binding epitope, this can be found on 12 unrelated human proteins. Again, tetanus toxoid would appear very dangerous.
And if we go one step further, then we can search for common T-cell epitopes using epitope prediction based on structural modeling. This is quite fancy in this modeling approach. The question which is being asked is: Is the mimicking peptide likely to be presented with a similar HLA peptide complex structure as a cell peptide?
And here we have an example where this has been done, comparing the binding in the HLA groove of an APC expressing the B-cell 27 class I antigen. On the one hand, in yellow, cell peptide of B-27, which is considered as a target in reactive arthritis. And here, the chlamydia DNA primae peptide which is shown, in green, to bind very similarly, to have the same structure as the cell peptide.
This is very fascinating. But as you can imagine, it can be done if you know the target protein. And it's taking so much time, in fact, that it's practically hopeless.
So the conclusion regarding this "in silico" prediction of T-cell epitope is that little useful information is likely to come out of random search approaches. And the search is more relevant when we deal with vaccines for infections which are known to be associated with autoimmune manifestation, and particularly if the target antigen, or one target antigen, is being suspected.
We have also the possibility for T-cell to look for in vitro and in vivo approaches. Animal models are not as good. But if the vaccine, again, is for an infection associated with autoimmunity, two questions can be asked. One is, can the vaccine antigen induce self-reacting T-cells in an appropriate model? And here I take the example of injecting chlamydia into a mouse which is transgenic for HLA B-27, and expresses a nice groove that we have seen before.
The second question is, can identified common epitopes be recognized by the patient T-cell? And again, this has been done in some studies, taking the patient T-cells from the articular fluid and showing that this can react with the same epitope which is being identified.
When moving in that direction, I would say that in this study of chlamydia, for example, starting with more than 80,000 putative potential mimicking epitopes, going down to the stage of reactivity in this in vivo model, this is allowing to restrict to eight or nine epitopes of potential significance.
One point which I think is essential is that the stringency of these different factors involved in T-cell stimulation and their potential role are very different according to what we look for. We know, for example, that in terms of binding to MHC there is a certain stringency, but not very high. Many things can bind.
In terms of recognition by T-cell, recognition by autoreactive T-cell, again, surprisingly, this is very degenerated. And many T-cells of low affinity can bind to many peptides. This is not selective.
So the real point: The selection, in terms of what comes out of this mimicry, depends on other things. It depends on the presence of co-stimulatory signals, which are provided either by an infectious agent--possibly by a very strong adjuvant. It has also to escape regulatory mechanisms, such as CD-25, CD-4 cells, which appear to be quite efficient normally. And probably, it needs as well a local inflammation in a target organ to get this really to lead to a pathogenic response and recognition.
So all this is so stringent at the end that it is very rare to get this complication. That's probably why every time we are infected we are not developing an autoimmune disease.
If we look at the other mechanisms, bistandard activation, this is different. The question is the relation to the fact that some infections have been shown not to induce autoimmune disease, but to trigger an underlying silent autoimmune disease. And the question is: Can vaccine do the same?
I just have here an example. We know that infection with the influenza virus in man has been shown to induce exacerbation of relapsing multiple sclerosis in one-third of the patients within the following six weeks. This is quite impressive. Fortunately, if these people are vaccinated, they do not develop this manifestation.
And we understand that now, in the following way; that some viral infections, particularly with IL-12 inducing viruses, such as the influenza virus, or exported to a number of microbial products, activate dendritic cells. And this activation can be strong enough, through [inaudible] receptors, to induce a release of a high level of pro-inflammatory cytokines, such as IL-1, IL-6, and IL-12. This can then lead to what we call a bistandard activation of other T-cells which are primed, which recognize different epitopes.
And the question obviously is raised: Can new vaccines with new adjuvants, such as MPL, the LT toxin, [inaudible], QS-21, CPG, or DNA vaccines, or some live attenuated viruses or viral vectors, particularly if they induce IL-12--Can they do the same thing? Can they induce all this mechanism and really create the same risk? This I think would be much more significant than any mimicry in the world.
And the question is: Can non-antigen-specific effects trigger an underlying silent autoimmune disease? How can we look at that? What kind of non-clinical assessment do we have?
Here, in vitro methods, we don't have much. We could imagine that we could compare the level of induction of cytokines, particularly IL-12, using different adjuvant formulation and using human PBMC or human purified dendritic cells, and look at this data. But the significance I think is still very difficult to define.
Other approaches which might be more relevant, in vivo, is to compare different vaccines, different adjuvants, in animal models of autoimmunity. And here we can look, for example, at the enhancement of murine lupus, tracking of EAE. So diabetic mice are not very good for that. And I am sure that this could help for the clinical trial planning. I don't think that the decision could be taken from animal models, but I am sure that we could better know what kind of monitoring we have to include in the initial clinical trial.
I just have here an example of spontaneous lupus, a model which is used in our center, and using New Zealand and B and W mice. In such a model, it's a model of systemic antibody-mediated autoimmune disease, although it's very much influenced by T-cells.
And we can test effects on anti-DNA; on the production and level of antiretroviral antigen, DP-17. We can assess the clinical expression. And here we see on the left the cumulative incidence of proteinuria in control New Zealand mice, as compared to mice which received adjuvants which are derivatives of LPS, which clearly do not accelerate the appearance of proteinuria; may have even some protection effect. And on the right, we see the same curve for cumulative incidence of mortality. And again, we see that the two adjuvants do not increase mortality.
So we can monitor survival. And we have to look at the histopathology to see to what extent the picture can be changed by what is given to these mice.
Another model which I think is very relevant--although like all models we don't know what to do with the reserves at the end--is that this is the model of silent priming for autoimmune experimental encephalitis. There are different possibilities, but the principal is to have mice which are primed.
Again, myelin is in the top part with infection with a thallus virus, or immunization with myelin-based protein in complete foreign adjuvant, or to use genetically predisposed mice which are transgenic for antimyelin T-cell receptor. These mice do not develop any clinical disease unless they are exposed to strong adjuvants, such as complete foreign adjuvant; or to IL-12 inducing viruses. In that case, the murine CMV. And this then leads very rapidly to delineating disease.
This is a kind of model where different adjuvants can be compared, different vaccine formulations, and see to what extent this kind of bistandard effect leads to real pathology.
One thing that I want to say is that I believe that non-antigen-specific effects of live or adjuvanted vaccine, at least the ones we know with the existing vaccine, appear to be time limited. And they are often localized to the regional lymph nodes. They are also very likely influenced negatively by regulatory mechanisms--the CD-4, CD-25 T-cells--and therefore, they are very unlikely to lead to these kind of dramatic results as we see in these models.
And I just take here the example of BCG. BCG is really considered as the strongest TH1 vaccine that we can give today. Well, in the studies in which we have collaborated with a group in The Gambia, we looked at the effect of BCG given to young children at birth, together with other vaccine, to see to what extent it is influencing the response to the other vaccine.
What has been seen is that if BCG is given at the same time--that means at birth or at two months--as the hepatitis-B vaccine, in the same arm, there is a very significant increase of the Interferon-gamma response to the unrelated vaccine, to the hepatitis-B. However, if BCG is given two months before or two months after hepatitis-B vaccine, there is no effect at all. This is really transient. It's like an adjuvant. If BCG is there at the time you get your vaccine, you get the effect.
And something which I found myself most surprising is that BCG has been used by an Italian group as an immunomodulator for the treatment of patients with multiple sclerosis. And they found no major adverse effects. And in fact, they even claim there is a beneficial effect of BCG. That means that BCG does not change completely the individual into a super TH1 person.
So in conclusion, I think that we can say that the potential risk of vaccine-associated autoimmune response is generally very low; often difficult to predict on a purely theoretical basis, such as mimicry; that for vaccines against infectious diseases known to be associated with an autoimmune pathology, we have to consider the theoretical risk of autoimmune response. And I think that a case-by-case approach has to be selected to see if this is a reality or not.
And finally, the potential risk of triggering an underlying autoimmune disease through non-specific bistandard effects, adjuvants, some live vaccine, is probably low, and even very low. But I think it would be a mistake to ignore it completely at this stage. Thank you.
DR. LAMBERT: Okay. So we have questions?
PARTICIPANT [In Audience]: I have two questions. The first one is the "in silico" analysis. Even if you do it on a longer stretch of amino acids or a longer peptide, you're going to come up with matches. So for example, you're developing a vaccine, and you come up with a match. What do you recommend on the next steps? You have no idea what this could be related to. Do you have any recommendations?
DR. LAMBERT: I think that, basically, if we suspect the potential B-cell epitope, we have to look for it, but we look for the cross-reactivity. And this may mean more.
At the T-cell level, I think that with the present studies which have been done, and if we have no indication at all that such an epitope, such an antigen, might be involved in an autoimmune manifestation, I think we might as well forget it. I think it's falsely giving the idea that this may be important, and I think that it's not.
PARTICIPANT [In Audience]: Okay.
DR. LAMBERT: You know, the real problem is that when we speak about vaccine and infection, we are used to speak about T-cells which have a relatively high avidity because they are directed against foreign antigens.
When we speak about the self-reacting peptide, and if they are seen by T-cells as they are seen by T-cells which have not been tolerized normally, and therefore they are the left-overs, they are low-avidity T-cells, low-avidity T-cell receptors, and they can bind as well these peptides to [inaudible] identified, but many other thousands, as was even without any amino acid in common with the first one--So it means nothing.
PARTICIPANT [In Audience]: Okay. So that leads actually into my second question, where you had those beautiful animal models of diseases that could be induced. I just wanted to make sure you're not recommending that we actually start using these animals indiscriminately; that you only use them if you think that there is a suspicion that whatever you found could cause or could induce some sort of autoimmune disease.
DR. LAMBERT: No, my feeling is that this type of model now has only a usefulness to compare different adjuvants, different formulations. For example, if you had the same adjuvant with some variance, or you derived different molecules from the same one, you might compare and see if, with the same level of activity, one is more likely for the question of binding capacity, diffusion, the kind of cell that would be seen--is more likely to induce this kind of bistandard activation than another one.
I think that by itself I would never recommend to do this kind of test just to see if you get the positive results, you know. By itself alone, it does not mean much.
PARTICIPANT [In Audience]: That's interesting. I'm trying to put that into perspective. I agree with you about the diabetic mouse not being a very good model. I'd like to hear why you think that; sort of what the criteria are for what you think are or would make good models; and at the end of the day, how you use the information you get from some of these things.
DR. LAMBERT: From which model do you speak of?
PARTICIPANT [In Audience]: The NOD mouse.
DR. LAMBERT: Yes. Again, I think I would put on the same level all of these kinds of models, spontaneous or induced models of autoimmunity. It's not giving you directly an answer, or the capacity to induce or reduce the disease which is present.
I think if we would have an adjuvant--For example, if in a model of NZB mouse we would inject LPS or [inaudible], we know that this will have a tremendous effect on the disease, will accelerate considerably the disease.
In this situation, this would tell you that with this kind of molecule, maybe you have to watch out. And when you move to your clinical trial, I would not stop the clinical trial for that. I would move into it, if you can, and then have maybe a special monitoring during your clinical trial for some of the indicators, some of the markers of systemic autoimmunity.
PARTICIPANT [In Audience]: Okay.