Technologies

The EpiMatrix System

The original EpiMatrix algorithms and coefficient sets were developed at the TB/HIV Research lab at Brown University. In 1999, the EpiMatrix tool set was exclusively licensed to EpiVax. Since then, the predictive algorithms and coefficient sets that make up the EpiMatrix tool set have been significantly revised and extended. At present, we can predict against over 100 different MHC Class I and Class II alleles.

Scalable Performance

Since the beginning, the EpiMatrix System was designed as a commercial system. The core of the EpiMatrix System is an Oracle database used to store input protein sequences and analytical outputs. By contrast, many of the web-based predictive services that are available simply deliver result sets to a web browser, providing no data management capabilities. By designing the EpiMatrix applications around a shared core database, we have developed a high-throughput environment where the results of preliminary analyses can easily be passed as input to subsequent analytical and reporting applications.

Searching for Putative T cell Epitopes: EpiMatrix

In a typical EpiMatrix analysis, the target protein sequence is parsed into overlapping 9-mer frames where each frame overlaps the last by eight amino acids. Each of the derived 9-mer frames is then screened for predicted affinity against a panel of MHC Class I and/or Class II alleles. Raw scores are normalized before being reported. The resulting Z-scores fall on a common scale that can be directly compared across HLA alleles. In our experience Z-scores above 1.64 (approximately the top 5% of all 9-mers derived from any given protein) have a significant chance of binding to MHC molecules and scores above 2.32 (approximately the top 1% of all 9-mers derived from any given protein) are highly likely to bind to MHC molecules. The ability to rate putative epitopes on a common scale, an exclusive feature of the EpiMatrix System, greatly simplifies the process of selecting putative epitopes for in-vitro testing.

Finding Epitope Clusters: ClustiMer

We have observed that MHC Class II restricted T cell epitopes tend to co-locate in short well-defined regions within protein sequences. The ClustiMer algorithm reads EpiMatrix results sets and identifies regions (typically 15 to 25 amino acids in length) that contain significantly more predicted T cell epitopes than we would expect to find by chance alone. We refer to these regions as T cell epitope “clusters.” In our experience, these clustered regions are highly likely to contain promiscuous T cell epitopes (i.e. epitopes that can bind to more than one HLA allele). Because they can interact with multiple HLA alleles T cell epitope clusters are important drivers of adaptive immune response. In a vaccine context, these short amino acid sequences can be used as either priming antigens or as boosting adjuvants. In a deimmunization context, T cell epitope clusters are high-value targets, areas where a small number of amino acids substitutions can have a large impact on immunogenicity.

Homology Searching: BlastiMer

We have also developed automated Blast tools, capable of Blasting protein sequences, cluster sequences, or even individual 9-mer peptides against either the non-redundant protein database at Genbank, the patent database at Genbank, or our own proprietary database of known ligands and T cell epitopes. Stored Blast results can be displayed as alignments or in a summarized form. Identifying homologies can help to focus your design efforts. For example, human-like peptides may make poor vaccine components since matching T cells are likely to have been either deleted or anergized. In the deimmunization context, identifying natural variation may help to identify substitutions that are well-tolerated.

Protein Deimmunization: OptiMatrix

Once a T cell epitope cluster has been identified, it is necessary to devise a strategy for deimmunization. We start by identifying those individual amino acids that contribute the most to binding affinity across peptide frames and HLA alleles. We believe changes in these “sensitive” amino acids can have a disproportional impact on the immunogenicity of the underlying sequence. Once we have identified a set of target amino acids, we develop a set of viable replacement amino acids. We review and consider many inputs when compiling this list. We may look at the Blast Summary report (described above) to identify changes tolerated in other species or variants of the target protein. We may develop a 3-D model of the target protein and screen a set of deimmunizing changes for low impact alternatives. We also carefully weigh any input our clients can provide. With a list of targeted amino acids and viable alternatives in hand, we can run our protein deimmunization algorithm, PickaMer. The PickaMer algorithm will try every possible alternative sequence and list the best single amino acid changes, the best double changes, the best triple changes, and if necessary, even more complex changes. The deimmunized sequences suggested by PickaMer can then be validated in-vitro (see below) before being integrated into the target protein and tested for functionality.

Finding Conserved Epitopes: Conservatrix

In developing vaccines against highly variable targets such as HIV or HPV, it is important to identify T cell epitopes that are conserved across multiple stains of the target pathogen. High affinity epitopes that occur in just one or a few circulating stains make poor vaccine components. Highly conserved T cell epitopes, on the other hand are the Achilles heel of the target pathogen, they can expose the target pathogen to a vaccine induced immune response.

VaccineCAD

VaccineCAD (Vaccine Computer-Assisted Design) is a recently-developed algorithm that permits in silico vaccine design. The alignment of epitopes in a vaccine construct may result in the development of “nonsenses” epitopes at the junctions between epitopes, or pseudoepitopes. One means of reducing the potential for junctional immunogenicity is to order epitopes so as to diminish the likelihood that an MHC binder will be created from the tail of one epitope and the beginning of another. Another means of reducing junctional immunogenicity is to insert spacer sequences between the epitopes that also reduce the likelihood that a pseudoepitope will be created. VaccineCAD iteratively reorders epitopes so as to maximally reduce junctional immunogenicity and also introduces spacers where necessary.

HLA Transgenic Mice

Basic animal models such as in-bred mice, guinea pigs, and even primate models fall short when used to assess the potential of protein based therapeutics to induce immune responses in human beings. The proteins that make up MHC molecules, the dominant mediators of adaptive immune response are among the most highly variable in the human genome. That variability is common to all mammals. Animal models simply do not present the same T cell epitope repertoire as humans. HLA mice, mice engineered to present fully human MHC haplotypes, are the only effective pre-clinical model of human immune response. At EpiVax, we have assembled a panel of HLA knock out/knock in transgenic mice and a set of robust protocols for assessing immune response in those models. Mouse strains containing the human HLA molecules DRB1*0301 and DRB1*0401 are currently available. Several other strains of mice will soon be in production. We believe these mice represent the best available proxy for studying the immunogenic potential of your molecule in humans.