I was reading @nature today and saw this pair of papers addressing one of the hardest problems in immunology: how do you encourage B cells to make antibodies against a specific epitope while also making them cross-react across diverse viral variants *at that epitope*? This is far from trivial. A naive B cell with close to the right starting antibody must already exist. It must bind the first immunogen strongly enough to activate, then beat many easier off-target B cells in germinal centers, and then it must somatically mutate down a restrictive, yet stochastic pathway of mutations. In nature.com/articles/s41586-0… they target the HIV V3-glycan epitope recognized by BG18-class antibodies. They use an engineered prime plus seven sequential boosts to progressively restore the native epitope and present diverse HIV Env antigens. This generated broadly neutralizing lineages in at least half the animals they immunized, and isolated antibodies reaching up to 67% of BG18’s breadth. In nature.com/articles/s41586-0… they target a different broadly neutralizing antibody class against the V2 apex antigen. Their engineered antigen improved precursor priming 30-400x over WT and achieved neutralization breadth in >90% of animals, reaching 90% on a 21-virus panel. To me, this is an amazing demonstration that you can "program" an antibody response, but the program in this case is a sequence of engineered immunogens enabling the right B-cells to rise to the top of a molecular melee occurring inside germinal centers. One thing that strikes me is this is very complementary to antibody design, (something I think about a lot). With computational design, epitope specificity and crossreactivity can be explicit design objectives. But these different approaches could feed into each other! Germline targeting as done here starts from a rare broadly neutralizing antibody discovered in e.g. an infected human, then reconstructs its naive precursor, then works backwards to design a vaccination path. With design, you could rapidly discover novel antibodies capable of broad neutralization, perhaps even to epitopes not observed in the natural immune response. To be vaccine-relevant, these should be reachable from the naive human repertoire, with plausible VDJ gene usage, CDR3 features, and a realistic mutation path. All of these are sequence-level constraints that sampling algorithms used by antibody designers could directly encode (indeed, this is a similar problem to maximizing humanness or biasing decoding to avoid sequence liabilities while maintaining or improving binding). Once you find candidates that conform to these requirements, you could also use design on the prime side a prime that better activates the right set of precursor B cells and boosts that structurally select for the mutations needed for breadth. TL;DR: design could tell us what broadly neutralizing antibodies the immune system could produce, and germline targeting could design a curriculum for B cells to get there. Congrats to the authors (just tagging accounts I found) @profshanecrotty @WardLab1 @scrippsresearch @ljiresearch