LEARN · DESIGN

Epitope selection

Most discovery effort goes into affinity, and most program failures trace back to the epitope. A tight binder on the wrong surface of the target blocks nothing, triggers nothing and delivers nothing. Choosing the epitope is a design decision, and it is worth making explicitly before any candidate exists.

Start from the mechanism

Write down what the molecule has to do, then ask which surface makes that possible.

To block a protein-protein interaction, the binder needs to cover or distort the interface. The interface, not the most immunogenic surface, is the target.
To deliver a payload inside a cell, the epitope has to support internalization, and epitopes on the same receptor differ in how well they do that.
To neutralize an enzyme, the site or a surface that controls access to it is the target, which often means a cleft where a small, convex paratope such as a VHH has an advantage.
To capture an aggregate specifically, the epitope has to be exposed on the aggregate and hidden on the monomer, which is a conformational requirement rather than a sequence one.
To be a detection reagent, the epitope needs to survive whatever sample handling the assay uses.

Then check the constraints

Conservation across species. If toxicology will be done in a species, the epitope has to exist there. Discovering after a year that the molecule does not bind the tox species is a common and expensive surprise.

Conservation across variants. For a target with allelic or disease-associated variation, an epitope that lands on a variable patch will work in some patients and not others.

Accessibility in context. A surface that is exposed on a recombinant ectodomain can be buried in the full-length protein on a cell, at a junction, or inside a complex with its partners.

Modification. Glycans, phosphorylation and proteolytic processing all change what is available, and they differ between recombinant material and tissue.

Steering a campaign toward an epitope

Several levers work, and they combine well.

Masking the surfaces you do not want by pre-blocking them with a known binder during selection. Using a truncated or stabilized antigen that only presents the region you want. Counter-selecting against a variant that lacks the epitope. Selecting on cells to keep only binders that see the native context. For computational design, specifying the target site directly is one of the things the current generation of methods does well.

Verify before committing

Whatever the route, confirm where the binders landed before the panel narrows. Competition against reference binders gives bins quickly. A mapping method gives residues. A predicted epitope from co-folding is a hypothesis, and the cost of being wrong about it grows with every month the program continues.