The art of deimmunizing therapeutic proteins
The consideration of potential immunogenicity is an essential component in the development work flow of any protein molecule destined for use as a therapeutic in a clinical setting. If a patient develops an immune response to the molecule, in the best case scenario, the patient's own antibodies can neutralize the drug, blunting or even completely ablating its therapeutic activity. In the worst case scenario, the immune response to the drug can endanger the health or even the life of the patient. Thanks to the incredible molecular diversity that can be achieved by VDJ recombination in antibody-producing lymphocytes (B-cells), the antibody repertoire of even a single individual is so vast (as many as 1011 distinct antibodies for a single individual) that it is difficult to imagine ever being able to design all potential antibody (or B-Cell) epitopes out of a protein while still preserving its structure and function. There is however a chink in the antibody defense's armor that can be successfully exploited to make therapeutic proteins less visible to the immune system - the presentation of antigens to T-cells by antigen-presenting cells (APCs), a critical first step in the development of an adaptive immune response to an antigen.
Protein antigens captured by antigen-presenting cells such as B-cells, are digested into peptide fragments that are subsequently presented on the cell surface as a complex of the peptide bound to a dual chain receptor coded for by the family of Major Histocompatibility Complex (MHC) Class II genes. If this peptide/MHC II complex is recognized by a T-cell antigen receptor of one of the population of circulating T- helper (Th) cells, the B-cell and its cognate Th cell will form a co-stimulatory complex that activates the B-cell, causing it to proliferate. Eventually, the continued presence of the B-cell antigen that was captured by the surface-bound antibody on the B-cell, will result not only in the proliferation of that particular B-cell clone, but also in the production of the free circulating form of the antibody (it should be noted that antibody responses to an antigen are typically polyclonal in nature, i.e. a family of cognate antibodies is generated against a specific antigen). It is through this stimulatory T-cell pathway that the initial detection of an antigen by the B-cell is escalated into a full antibody response to the antigen. Incidentally, one of the major mechanisms of self-tolerance by the immune system is also facilitated by this pathway via the suppression of T-cell clones that recognize self-antigens that are presented to the immune system during the course of its early development.
This T-Helper pathway is therefore a key process in mounting an antibody-based immune reponse to a protein antigen and while the repertoire of structural epitopes that can be recognized by B-cells is probably far too vast to practically design a viable therapeutic protein that is completely free of them, the repertoire of peptides that are recognized by the family of MHC Class II receptors and presented to T-cells (T-cell epitopes), while still considerable in scope, is orders of magnitude smaller than the set of potential B-cell epitopes.
So as designers of therapeutic proteins and antibodies, how can we take advantage of this immunological "short-cut", to make our molecules more "stealthy" with regard to our patient's immune system?
The solution lies in remodeling any non-self peptide sequences within our molecules, that are determined to have a significant binding affinity for the MHC Class II receptors. The two chains of an MHC Class II receptor form a binding cleft on the surface of an APC into which peptide sequences of approximately 9 amino acids can fit. The ends of the cleft are actually open, so longer peptides can be bound, but the binding cleft itself is only long enough to sample about 9 amino acid side chains. It is this cleft with the bound peptide that is presented on the surface of an APC for recognition by T-cells.
The genetic evolution of MHC Class II alleles in humans is such that there are about 50 very common alleles that account for more than 90% of all the MHC Class II receptors found in the human population. There are of course, many more alleles in the entire human population, but they become ever rarer as you go down the list from the 50 most common ones, with some of the rarer alleles being entirely confined to very specific populations and ethnicities. What this means for us as engineers of therapeutic proteins is that if we can predict potential T-cell epitopes for the 50 or so most common MHC Class II alleles, we can predict the likelihood of a given peptide sequence being immunogenic for the vast majority of the human population.
It actually turns out that some researchers have published experimental peptide binding data for the 50 most common MHC Class II alleles and their results are very encouraging for the would-be immuno-engineer. The peptide binding motif of the MHC II receptor essentially consists of 9 pockets, each of which has a variable binding affinity across the 20 amino acid side chains that is independent of the side chains bound in the other 8 pockets. This last property is of particular importance because it means that we can calculate the relative MHC II binding affinity for any particular 9-mer peptide by the simple summation of the discrete binding pocket/side chain affinities, rather than having to consider the vast combinatorial space of binding affinities that would be possible if the amino acid binding affinity of each pocket was dependent upon the side chains bound in the other 8 pockets.
This is the point at which a computer and some clever software can be enormously helpful. While I was employed at a major biotechnology company, I created software that could use a library of this kind of MHC II peptide affinity data, in order to scan the peptide sequences of protein drugs and antibodies that we were developing for the clinic. The software not only predicted the regions of the peptide sequence containing potential T-Cell epitopes, but it also used other structural and bioinformatics algorithms to help the scientist to successfully re-engineer the molecule to reduce its immunogenicity while preserving its structure and function.
This last phrase explains why I used the word "art" in the title of this article.
What we learned from experience was that while it is relatively easy to predict T-cell epitopes in a peptide sequence, reengineering the sequences while preserving the structure and function of the protein is the much greater challenge.
Based upon this experience, it was no surprise to me that the great majority of the thousands of lines of Java code that I wrote developing our deimmunization software, was dedicated to functionality that guided the scientist in selecting amino acid substitutions that would have the highest probability of preserving the structure and function of the protein. Even with this software however, the essential elements in this process were still the eyes and brain of the scientist, guided by training and experience in protein structure and biochemistry.
In other words, the art and craft of the experienced protein engineer.
Much like the old joke "My car is an automatic but I still have to be there" - the software could not substitute for the knowledge and experience of a skilled protein engineer, but it could make her life a lot easier by suggesting amino acid substitutions with a high probability of being structurally and functionally conservative; and by keeping track of all the changes and their impact upon the sequence and structure.
The software really showed its value in the improvement it brought to our success rate in converting our computational designs to successful molecules in the laboratory. For any given project with a new biologic, we would typically design a bunch of variants to be tested in the lab, of which one or two might have all the properties we were shooting for. Once we started using the software, there was a noticeable increase in the proportion of our designs that tested well in the lab, compared to previously. This was interesting to me insofar as it showed that while the software could not replace the scientist's knowledge and experience, it could certainly enhance and augment its application to the problem at hand - probably by keeping track of the many moving parts in the deimmunization process, so that the scientist is free to think more carefully about the actual science.
In spite of all this technological support however, a successful deimmunization depends heavily upon skill and experience in protein engineering, and there's arguably still as much art in successfully re-engineering T-cell epitopes as there is science in predicting them.
© The Digital Biologist | All Rights Reserved