PD-L1 offers a few plausible binding faces on its IgV domain. Only the GFCC′ face — the one PD-1 itself docks against — was carried into stage D. Four reasons, roughly in order of how badly the campaign would fail if each one were ignored:
- It is the functional face. A binder anywhere else on PD-L1 can still be a real, measurable binder — good KD, clean SEC, good Tm — and still fail the point of the campaign, because it never touches the surface PD-1 needs. Nothing downstream catches that mistake early; not BLI, not SEC, not Tm, only the stage-L blockade assay does, months later. Getting the face right at stage C is the only place this gets caught cheaply.
- It is solvent-exposed and geometrically tractable. The GFCC′ face is a moderately convex β-sheet edge, not a deep pocket or a flat plateau. RFdiffusion and BindCraft both design well against convex-to-flat patches anchored on a small hotspot cluster; they design poorly into grooves (nothing to wrap around) and poorly against very flat, feature-free surfaces (nothing to key into). The alternative considered and dropped — PD-L1's flatter, more hydrophilic back face — falls into that second trap: mini-protein binders pack badly against it, and there is no antibody or biological precedent that binding there does anything.
- It is not masked by glycans. The target page's automatic check cross-references PD-L1's annotated N-glycosylation sequons against the IgV domain boundary; none of them fall on this face (see PD-L1 → Glycans).
- Contact residues are mostly conserved from human to mouse. That is what makes the campaign's mouse cross-reactivity goal plausible without a second, mouse-specific design cycle — "mostly," not "fully," so cross-reactivity still gets confirmed experimentally rather than assumed.
External benchmark. Atezolizumab and durvalumab, two clinically approved anti-PD-L1 antibodies, were independently mapped to an epitope overlapping this same face Reference/Lee 2017. That is clinical-grade evidence the face matters, not just a structural argument from one crystal structure.
Related: PD-L1 · PD-L1/Campaign/PDL1-R1