Combating Antibody Autoimmunity
Qingyu Cao, Ph.D., head of business development at AlgoNomics (www.algonomics.com), presented her company’s approach to dealing with immunogenicity of therapeutic proteins. As all therapeutic proteins demonstrate some level of immunogenicity, this is a common problem.
AlgoNomics’ Epibase® uses structural bioinformatics to identify potential epitopes on a protein based on the structural characteristics of HLA molecules and their binding property to epitopes. The whole sequence of target protein is scanned using the Epibase algorithm, and the binding affinity between epitope and HLA is computed based on the interaction energies, taking into account side-chain flexibilities. A risk profile of the particular protein molecule is thus provided. This method becomes increasingly more accurate as the amount of experimental data grows, correlating in silico modeling with results from the company’s T-cell activation assay.
While the in silico approach cannot replace thorough animal and clinical studies, it allows for a more rapid and accurate assessment of the immunogenicity risk and can provide substantial cost savings by elimination of blind alleys, Dr. Cao reported.
Dr. Cao’s contentions were buttressed by clinical data from a study on patients with rheumatoid arthritis who were receiving Adalimumab directed against TNF-a. The study, performed in collaboration with Sanquin and Genmab, identified a number of potential troublemaker epitopes.
Of 109 patients enrolled in the study, 19 showed antidrug antibodies and 17 of these possessed HLAs that were targeted by unruly epitopes. These HLAs are also highly associated with RA.
While high-throughput screening is now widely employed in the search for new pharmaceutically active molecules, there is little information available on the factors that affect the ability of positive hits to move to later stages of drug development. Novartis Pharma (www.novartis.com) has extensive experience in this area, according to Andreas Bender, Ph.D., post doctoral fellow at the Novartis Institutes for BioMedical Research.
Dr. Bender’s approach is to combine in silico target prediction, which gives mechanistic information about compound action, with cellular, image-based readouts profiling multiple parameters, yielding holistic information but lacking the mechanistic parameter. “We combine both processes to have a better of idea of the true character of a compound, in order to measure both the effect of a compound as well as to develop an hypothesis of its function,” he stated.
Dr. Bender and his associates compared the results of thousands of high-throughput screening runs using 15 different readout technologies and 18 target classes including protein kinases and GPCRs. The group collected data on which aspects of the screening process were empirically correlated with downstream success, defined as the fraction of HTS campaigns that advance into the later stages of drug discovery.
The Novartis high-content screening platform gives 36-dimensional readouts in three channels. HeLa cells are grown for 24 hours, stained with a laundry list of fluorescent dyes, then analyzed by automated microscopy following a factor analysis to reduce dimensionality of the readout. This approach yields a multidimensional readout that details the assembled complexity of the signaling network. “We can build statistical ligand-target relationships to predict the targets of molecules, based on their molecular structure,” Dr. Bender said.
When Dr. Bender and his colleagues analyzed the results of these large-scale screens of 6,500 compounds they found that “similar effects (phenotypes) can be obtained from dissimilar molecular structures. On the other hand, similar structures can cause dissimilar effects. Different information is obtained from both sources and employing both types of knowledge is crucial.”