Comparing historical epitope groups to experimental immune epitope data

We asked whether there is experimental evidence that the human immune system displays a 5 bias towards particular regions of hemagglutinin; such a bias would traditionally be called an 6 immune epitope. We obtained all available human B cell epitopes for H3 hemagglutinin from the 7 Influenza Research Database (IRD). There were two broad categories of epitope data available 8 in the epitope database. One was non-linear epitopes and the other was linear epitopes. In 9 addition, the IRD splits epitopes into B and T cell epitopes and also into the various species 10 whose immune system was tested. Here, we were interested in antibody-driven immune escape, 11 and hence we focused on B cell epitopes. 12 We initially considered all available B cell epitope data in our analysis, inluding human and 13 non-human, because non-human data have traditionally been part of the data considered for 14 epitope definition in influenza (see e.g. [1]). In the database, there were two types of B cell 15 epitopes available for hemagglutinin: linear and non-linear. For humans, there were 31 separate 16 epitope entries consisting of 26 non-linear and 5 linear. For B cell epitopes, every linear human 17 epitope was also listed in the IRD as the linear epitope of some other species, most commonly 18 mouse but also, for example, ferret. For non-humans, there were 134 available epitope entries 19 consisting of 47 non-linear and 87 linear. The non-linear epitopes were provided as site numbers, 20 so that mapping them onto the protein structure was trivial. 21 For linear epitopes, we started with the short sequence fragments; each of the fragments 22 was between 5 and 40 amino acids in length. We tried initially to map the fragments onto the 23 emblematic A/Aichi/2/1968 sequence, but it turned out the epitope fragments were actually 24 generated from disparate strains along the H3N2 lineage. The differences between the original 25 founder strain and the fragments meant that we could not accurately map the vast majority 26 of short peptides onto a single sequence. Instead, we took the entire curated and pre-aligned 27 set of 3854 sequences that we used in the evolutionary rate calculations. We then aligned the 28 fragments with MAFFT using a very strong opening penalty of 10. We visually checked to be 29 sure all of the 87 fragments aligned reasonably well to the full H3 alignment. Then, as with the 30 non-linear epitopes, we counted the number of times each site was hit and the co-occurrences 31 of sites hit by the same antibody. Since the 5 linear human epitopes were a subset of the 87 32 non-human linear epitopes, we dropped the linear human epitopes from further analysis. 33