Affinity maturation without breaking developability
Maturation selects for binding, not for behaviour. In our own published programme it did not merely fail to protect developability — it destroyed it, and we had to engineer it back.
Short answer: affinity maturation optimises one number and leaves every other property unprotected. Self-association, polyreactivity and expression can all get worse while affinity improves. The fix is counter-selection inside the campaign and orthogonal developability readouts on every variant — not a developability gate bolted on at the end.
Why a matured clone can be worse than its parent
Nothing in a standard panning scheme asks whether the winning clone is soluble, monomeric, non-sticky or expressible. Those properties are free to drift while the number being optimised improves.
Two mechanisms account for most of it. The parent’s liability can simply survive the campaign, if it is driven by framework or by CDR positions the libraries never touch. More awkwardly, selection can create the liability: the residue changes that raise affinity — added hydrophobicity in the paratope, patches of complementary charge — are the same physical features that drive non-specific surface interaction. A scheme that scores only antigen binding cannot tell a clone that binds the antigen tightly from one that binds everything slightly more tightly, including itself.
What this looked like on a real molecule
We ran our platform end to end on chimeric infliximab, a molecule with documented Fab–Fab self-association, and published the result including the part that did not work.
Humanization did what it should. Grafting the CDRs onto human germline frameworks removed the self-association entirely: AC-SINS Δλmax went from 19.33 nm for the chimeric parent to −4.00 nm for hInfliximab. It also cost potency, as grafts usually do — IC50 moved from 96.8 to 166.1 ng/mL.
STEM™ affinity maturation then recovered and exceeded parental potency, reaching 42.0 ng/mL. And it brought the self-association back. hInBG4, the matured clone, read +9.00 nm on AC-SINS — a molecule that had been clean before maturation was now a formulation risk. Its BVP ELISA ratio rose in parallel, from 1.1 to 2.3.
The liability was not inherited. It was created by the maturation step, because nothing in the affinity readout was scoring it.
| Clone | TNFα neutralization IC50 (ng/mL) | Self-association AC-SINS Δλmax (nm) | Non-specificity BVP ELISA ratio |
|---|---|---|---|
| Infliximab — chimeric parent | 96.8 | 19.33 | 1.3 |
| hInfliximab — humanized parent | 166.1 | −4.00 | 1.1 |
| hInBG4 — after STEM™ maturation | 42.0 | +9.00 | 2.3 |
| LW2Y — after targeted CDR substitution | 34.5 | −1.67 | 1.5 |
Data from Table 1 of our preprint, doi:10.64898/2026.08.18.745573 (bioRxiv). Preprint — not certified by peer review.
Fixing it took a separate, deliberate round. hInBG4 was treated as a new scaffold, a panel of single and combined CDR substitutions was built, and every variant was profiled on four orthogonal readouts. Three cleared all four. The final clone, LW2Y, is 2.8× more potent than the chimeric parent and self-associates less than adalimumab.
Screen inside the campaign, not after it
Running developability as a gate after maturation means discovering the problem once the options have already narrowed to whichever clones survived. Counter-selection during selection stops polyreactive and thermally fragile clones being enriched in the first place.
In our staged scheme every selection round is heat-challenged and subtracted against a polyspecificity panel — baculovirus particles, dsDNA, ovalbumin, HSP90-Fc and KLH — before panning on antigen. Libraries are designed at preselected CDR positions using the amino acid usage and positional frequencies observed in human antibodies, which biases the output toward a human-like sequence rather than away from one, and complexity is held within an electroporation capacity of roughly 1 × 1010 cfu so the library is genuinely sampled.
This reduces how often a second engineering round is needed. It does not guarantee it is never needed. On infliximab it was.
Four readouts, and what a bad number costs later
| Assay | What a bad number would cost you later |
|---|---|
| AC-SINS | Aggregation, high viscosity at concentration and poor colloidal stability — the difference between a subcutaneous product and an infusion |
| BVP ELISA | Polyspecificity against a heterogeneous membrane surface; high ratios track with fast clearance and poor PK |
| SE-HPLC | High-molecular-weight species and peak tailing — the aggregate burden a regulator will ask about |
| Transient expression | A cheap first read on whether the molecule will express at manufacturing scale |
Four orthogonal readouts are enough to rank a variant panel. One is not, because each misses a different failure mode.
If your lead has already failed developability
A matured lead carrying a self-association or polyreactivity flag is not dead. It is a scaffold with a localisable problem. The approach that worked on hInBG4 was targeted substitution rather than another round of random maturation: build a panel of single and combined CDR substitutions, profile all of them on orthogonal assays, and keep the variants that clear every readout without surrendering potency. LW2Y surrendered nothing — it gained potency while its AC-SINS moved from +9.00 to −1.67 nm.
Frequently asked questions
Does affinity maturation always worsen developability?
No. It removes the selection pressure that would have protected it. Whether a molecule gets worse depends on the parent’s liabilities and on whether the scheme includes counter-selection. Campaigns that screen polyspecificity and thermal stability during selection produce matured clones with developability profiles comparable to their parents far more often.
Can self-association be fixed without losing affinity?
Often, yes. In our published programme the final engineered clone was both more potent and less self-associating than the matured clone it came from. It required a second, measurement-guided engineering round rather than a repeat of maturation.
What AC-SINS value is acceptable?
There is no universal cutoff, and the number is assay- and format-dependent. The useful comparison is against clinical-stage benchmarks run in the same assay on the same day. In our panel adalimumab read −1.33 nm and chimeric infliximab 19.33 nm.
Should developability screening come before or after affinity maturation?
During. A gate after maturation tells you which clones failed; counter-selection during maturation stops them being enriched at all.
How long does an affinity maturation campaign take?
STEM™ campaigns run five months or longer depending on the target and the starting data. Scope and timeline are set on a feasibility call.
Bring us the antibody you cannot fix
If you have a lead that binds well but fails developability screening, the scoping call is with the PhD scientist who would run the work — not a salesperson.
Talk to a scientist about your lead →
Free 30-minute call · Initial feasibility assessment within two business days · NDA available before you disclose your target
Related: STEM™ affinity maturation · antibody humanization · the full infliximab case study