Making a safe and effective vaccine isn't easy. Usually only a handful of protein snippets, or peptides, from a pathogen are able to spark a protective immune response. Now researchers have developed a computer program that homes in faster on that peptide needle in the protein haystack. Experts say the tool, described in the June Nature Biotechnology, could help rein in the skyrocketing costs of vaccine development.
To trigger a vigorous immune response, antigen-presenting cells (APCs) put on display chopped-up peptides from pathogens or tumor cells. The protein fragments draw the attention of T cells, one caste of the immune system's warriors, which secrete a variety of immune hormones and orchestrate the different players of the immune reaction. APCs keep a tight grip on the foreign peptides by sticking them in a cleft of what's called their major histocompatibility complex (MHC), a diverse set of surface proteins. To complicate matters for vaccine designers, humans have one of six different MHC genes, each coming in hundreds of slightly different forms, or alleles. Because MHC alleles differ in their binding affinity for various peptides, a vaccine based on just one peptide may not work for everyone.
About 8 years ago, Jürgen Hammer, an immunologist at Hoffman-La Roche Inc. in Nutley, New Jersey, set out to decipher the rules that govern peptide binding to MHC molecules. In a painstaking series of some 10,000 experiments, he determined affinity of 35 different binding sites within the MHC clefts for the 20 naturally occurring amino acids that make up peptides. This allowed him to estimate the peptide binding strength of the 51 most common MHC alleles that contain these binding sites. From this information Hammer devised a computer program that could calculate the binding affinity of the 51 MHC alleles for various pathogenic peptides. The more MHC alleles that bind to a particular peptide, the more likely this peptide is to induce an immune response in many people.
Putting the software to the test, Hammer's team analyzed some 200 peptides with known MHC-binding properties. The computer predicted correctly up to 80% of the peptides that bind to a given MHC, and it falsely picked less than 5% of the peptides that didn't bind.
The approach could "greatly accelerate vaccine development" by cutting down on the number of peptides that must be screened in the test tube, says vaccine researcher Anne DeGroot of Brown University in Providence, Rhode Island. But she cautions that the concept must be validated by pointing the way to a new vaccine. "You can do a lot of neat things on your computer," she says, "but ultimately you have to put [the predicted peptides] in vaccines and test whether they protect [from infection]."