CD28 has been a feared target since 2006, when the anti-CD28 superagonist TGN1412 put six healthy volunteers in intensive care within hours of the first human dose. The receptor sits on T cells and tells them to scale up when the T-cell receptor sees antigen. Push it the wrong way and the result is a cytokine storm. Push it the right way, with a knob a prescriber can dial back, is what nobody has reliably done on this target since the TGN1412 disaster.
A group reporting May 30 in Advanced Science ↗ used an AI-guided design pipeline to find a cyclic peptide they called CIP-3 that binds the CD28 extracellular domain with nanomolar affinity and pulls the costimulatory signal down rather than turning it on. CIP-3 disrupts the CD28-ligand interaction directly, blocks T-cell activation in primary human cells, and shows no intrinsic agonist activity in the assays the authors ran. The TGN1412 mechanism is not on offer here. CIP-3 is an antagonist and only an antagonist.
The pharmacology angle is the part biologics do not have. Anti-CD28 antibodies sit on the receptor for days. Their effect on a patient is the integrated outcome of a slow-off-rate occupancy that the prescriber cannot pull back if something goes wrong. CIP-3 is reversible on a fast timescale that tracks drug exposure. When the peptide is around, CD28 signaling is muted. When the peptide clears, the receptor is back in business. Whether this matters in any given clinical setting depends on what is being treated, but it gives the prescriber a kinetic margin the antibody class does not.
The functional data are early but consistent. In a T-cell-transfer model of chronic mouse colitis, CIP-3 produced dose-dependent reductions in disease activity, and the same dose-dependence read out in systemic inflammatory cytokines. In primary peripheral blood mononuclear cells from healthy donors and from patients with active ulcerative colitis, the peptide suppressed cytokine production at levels the authors describe as comparable to a benchmark anti-CD28 biologic. That comparison is the practical question for translation. An oral or short-half-life subcutaneous CD28 modulator that matches a long-acting biologic on efficacy and adds a kinetic safety margin is a different product, not a worse one.
The cyclic peptide format is doing two specific jobs here. The cycle stabilizes a constrained backbone that has to dock into the CD28 protein-protein interface, a flat shallow interaction surface that conventional small molecules have generally failed at and that antibodies attack from a different angle. The same cycle limits the conformational entropy on the way to binding, which is what the AI pipeline was selecting for. CIP-3 is a synthetic modality designed for an interface that the field had given up on as undruggable by chemistry, not biology.
What the paper does not establish is whether CIP-3 can reach the doses needed for human chronic disease without losing the safety margin its kinetics suggest. Cyclic peptides as a class have made progress on oral and subcutaneous formulations but remain harder to dose than biologics or small molecules. The chronic colitis model is mouse. The patient PBMC data is ex vivo. The next data point that matters is a human PK/PD study showing that the exposure window the authors are after is actually achievable in patients.
CD28 itself is not on peptidemodel as a discrete card target. The receptor sits inside the broader immune ↗ panel. The design pattern here, an AI-picked cyclic peptide targeting a protein-protein interface that conventional medicinal chemistry walked away from, is the same one that produced several of the cyclic-peptide leads the platform hosts in unrelated indications. It is the format catching up to the receptors that need it.