Developability liabilities in CDRs
A developability liability is a sequence feature likely to cause a problem during manufacturing, storage or dosing. Screening for them is cheap, and the screen is only useful if the output is ranked by consequence rather than counted.
The motifs worth finding
Asparagine deamidation. Asparagine followed by glycine is the fast case, and asparagine followed by serine or threonine is slower. Deamidation converts the residue to aspartate or isoaspartate, changing charge and often binding. In a CDR it is the liability most likely to cost potency over shelf life.
Aspartate isomerization. Aspartate followed by glycine is the classic motif, and the reaction is favored in acidic conditions, which includes the low-pH hold after Protein A elution.
Methionine and tryptophan oxidation. Both oxidize under light, metal ions and peroxide. Exposed methionine in a CDR or at the Fc interface is the usual concern.
Unpaired cysteine. A free thiol in a variable domain promotes scrambling, dimerization and aggregation, and it complicates any conjugation chemistry later.
N-linked glycosylation sequon. Asparagine, any residue except proline, then serine or threonine, inside a variable domain leads to heterogeneous glycosylation of the paratope.
Hydrophobic patches and charge extremes. Large exposed hydrophobic surface on the CDRs drives aggregation and non-specific binding. Strongly positive patches correlate with poor clearance and with polyspecificity.
Sequence is a filter, not a verdict
A motif is a possibility, not a defect. Whether it matters depends on solvent exposure, local flexibility and the neighboring residues, which is why a structure, even a predicted one, adds so much to a sequence screen. An NG buried against the framework behaves very differently from an NG on the tip of CDR-H3.
So the workflow that holds up is: find the motifs from sequence, weight them by exposure from structure, then confirm the ones that matter experimentally with forced degradation, a thermal ramp, a non-specific binding assay and a short stability study.
The problem with editing CDRs
Most of the worst liabilities are in the CDRs, because that is where diversity lives. But the CDRs are the paratope, and editing them risks the binding you selected for. Framework edits are cheap; CDR edits require re-testing.
This is the tension every developability tool runs into. A score that says "fix the NG in CDR-H2" is only actionable if the fix preserves affinity, and that has to be measured. In practice the sequence is that you generate the conservative alternatives, express them alongside the parent, and measure both binding and stability, rather than choosing one from a score.
What to do at the design stage
Screening early is much cheaper than fixing late. Filtering a designed panel for the worst motifs before synthesis costs nothing and removes candidates that would fail after months of work. It does not replace measurement, and a panel filtered only by computed scores can lose the best binders, so keep some liability-bearing candidates in the panel when their binding is exceptional, and plan the engineering.