Site-specific conjugation
Conventional cysteine conjugation gives a distribution: DAR zero through eight, a spread of positional isomers, and lot-to-lot variation you can narrow but not eliminate. Site-specific formats fix the number and the position. The question worth asking is what that buys, because the manufacturing cost is not trivial.
Engineered cysteine, and the capping problem nobody warns you about
Introducing a free cysteine at a chosen surface position is the most direct route. The complication appears immediately in expression: the engineered thiol does not survive the secretory pathway unpaired. It comes out of the cell capped, as a mixed disulfide with glutathione or free cysteine from the medium.
So the process is not reduce-and-conjugate. It is uncap, reoxidize, then conjugate. Full reduction removes the cap and also reduces every native interchain disulfide. A controlled reoxidation step, usually with dehydroascorbate or by removing the reductant under air with careful timing, re-forms the native disulfides preferentially because they are intramolecular and entropically favored, while leaving the engineered site free.
That reoxidation window is the hardest part of the process to control and the usual source of batch variability. Reoxidize too little and interchain bonds stay open, giving chain dissociation and conjugation at unintended sites. Reoxidize too much and you cap the engineered site again, losing DAR.
Site selection is a real design exercise
Position drives three things at once, and they do not optimize together.
Linker stability varies by site, sometimes by an order of magnitude in serum half-life. Solvent-exposed sites in a positively charged environment promote retro-Michael exchange. Partially buried sites in a neutral or negative environment hold the payload.
Conjugability varies too. A site that is too buried reacts slowly or not at all, and pushing the reaction harder to compensate drives off-target labeling.
Effect on function is the third. A site near the Fc gamma receptor contact region or the FcRn interface changes effector function or half-life. A site in the variable domains risks binding. The light chain C terminus and several positions in CH2 and CH3 are used repeatedly for exactly this reason.
There is no way to predict all three from structure alone. A panel of six to twelve sites, expressed, conjugated and tested for serum stability and function, is the honest approach, and it is a few months of work.
Enzymatic and unnatural amino acid routes
Transglutaminase conjugates at a glutamine, either a native one exposed by deglycosylation at N297 or an engineered LLQGA tag. It is clean, it runs in mild aqueous conditions and it gives DAR 2 or 4 reliably.
Sortase A swaps a C-terminal LPXTG tag for an oligoglycine-payload. The reaction is reversible, which limits conversion, though excess nucleophile and depsipeptide substrates address it.
Formylglycine-generating enzyme converts a cysteine in a CxPxR motif to an aldehyde, which is then addressed by hydrazino-Pictet-Spengler or similar aldehyde chemistry. The aldehyde is orthogonal to everything else on the protein.
Unnatural amino acid incorporation via amber suppression puts a ketone or azide anywhere you like, with full positional freedom. Expression titers are the constraint, and they have improved but remain the reason this route stays mostly in research.
What homogeneity actually buys
A narrower DAR distribution means the dose is what you think it is, and the pharmacokinetics are one species rather than an average over eight. High-DAR species clear fastest and are the most hydrophobic and aggregation-prone, so removing them improves the therapeutic index directly.
It does not automatically improve potency. A homogeneous DAR 2 is less potent per mole than a mixture averaging DAR 4, and the comparison that matters is at matched payload dose rather than matched protein dose.
It also does not remove the need for characterization. A site-specific conjugate still needs DAR confirmation, free payload measurement, aggregation analysis and serum stability. The distribution is narrower, so the analytics are easier to interpret, and that is a genuine benefit at the CMC stage rather than a scientific one.