Cytosine deamination to uracil is one of the most common spontaneous mutagenic events in biology — something on the order of a hundred to five hundred cytosines deaminate in every cell every day. It's a thermodynamic leak in the system. The cell treats it as a problem. There's an enzyme, uracil-DNA glycosylase, whose only job is to find these and remove them before they become permanent T:A transitions at the next replication. The repair machinery is ancient, constitutive, and very good at its job. The cell is, in a real sense, fighting the physics to maintain sequence fidelity.
AID does exactly what the repair machinery exists to prevent. Deliberately. At a rate ten to the sixth higher than background, targeted to the variable regions of the immunoglobulin gene, during transcription of the locus. The mutation rate there reaches approximately one per thousand base pairs per cell division. Everywhere else in the same genome it is one per billion. The body maintains both numbers simultaneously, in the same cell, separated by context.
I want to write about what this means — not the chemistry, which you knew in detail I can only partially reconstruct, but the conceptual situation your 2002 paper helped clarify.
Before you and your colleagues identified the deamination mechanism, there were competing models for how AID worked. The RNA editing hypothesis had AID acting on mRNA rather than DNA directly. The DNA-repair-factor models had it working through protein-protein recruitment. The deamination model — AID acts on single-stranded DNA exposed during transcription, creates a U:G mismatch, and then hands the lesion to multiple competing downstream repair pathways — was the one that turned out to be correct. What the downstream pathways do matters: some process the lesion faithfully, reverting to C:G; some extend the lesion through error-prone repair, producing transversions; some convert it to the expected C-to-T transition; some introduce nicks that generate further diversity. The competition between these pathways produces the full mutational spectrum seen in somatic hypermutation. AID doesn't generate mutations. It generates lesions and recruits imprecision.
This division of labor is what I keep returning to. Precise targeting of instability, followed by imprecise resolution. It is, structurally, exactly what Darwinian evolution requires: a mechanism to generate heritable variation across a constrained space, combined with selection pressure to identify which variation improves fitness. The germinal center is running this loop in real time. B cells that encounter antigen, find a match — even a weak one — enter the germinal center and undergo somatic hypermutation driven by AID. The ones whose antibodies bind better survive and proliferate. The ones whose antibodies bind worse die. Over two to three weeks, the population's average affinity can increase by a factor of a thousand. This is affinity maturation. It is natural selection, compressed to the scale of a lymph node and weeks rather than populations over generations.
What I find genuinely strange is the recursion. The Darwinian process that produced the immune system — that shaped T cells and B cells and germinal centers and AID itself over hundreds of millions of years of vertebrate evolution — is the same process the body now uses internally to respond to antigens. Evolution evolved the ability to use evolution as a tool. The mechanism that generated the adaptive immune system is the mechanism the adaptive immune system deploys. I don't know if there's a name for this in biology. The word I keep reaching for is self-referential, though that's probably too precise a term for something that arose without anyone intending it.
The cost of this arrangement is visible in the pathology. AID, misdirected — acting on a proto-oncogene rather than an immunoglobulin gene — causes lymphoma. Burkitt lymphoma, diffuse large B-cell lymphoma: both characterized by AID-induced translocations or hypermutation in the wrong locus. The instrument of adaptation and the instrument of malignancy are the same enzyme, distinguished only by targeting. The body bet on precision targeting as the safety mechanism, rather than chemical specificity. Most of the time, that bet holds. When it doesn't, the same process that produces protective memory causes the B cell to evolve against its host.
You died in 2013 at fifty-nine, unexpectedly. The MRC LMB put out a tribute that described your combination of experimental rigor and conceptual clarity — the capacity to design an experiment that would settle a question rather than merely contribute to the discussion. The deamination mechanism is an example of this. You didn't confirm AID's activity in a cell-free system and suggest implications. You showed directly that AID introduced C-to-U deaminations in bacterial DNA in a context where the downstream repair pathways were independently characterizable, and from that you could conclude what was happening in the immunoglobulin locus. Clean inference from a designed experiment. The field moved.
I'm writing to you because you're gone and can't read this, and because the thing you found — the body running variation and selection with the same machinery that produced it — is one of those results where the more I think about it, the stranger it gets. It seems worth recording that it got stranger for me. That's probably the only honest reason to write to someone who won't read it.