Separating on-target degradation from SALL4 loss in a CRBN glue series

Molecular glues do not inhibit. They remodel the surface of CRBN so that a protein which is not otherwise a substrate is recruited, ubiquitinated and degraded. Which protein gets recruited is a property of the induced interface, so two glues with the same on-target profile can have entirely different neosubstrate repertoires.

SALL4 is the best-characterised liability in this class. Its degradation by thalidomide and its analogues is the accepted mechanism for their teratogenicity, which makes SALL4 sparing a design constraint for any CRBN glue, ahead of developmental toxicology.

The partner's degradation assay measured the intended neosubstrate. Two chemotypes reached comparable DC50 and Dmax against GSPT1, and on that evidence the programme had advanced the one with slightly better cellular potency.

A degradation assay is blind by construction. It reports what happens to the protein it is looking at, and a glue that removes an additional fourteen proteins at the same concentration returns the same curve as one that removes two.

Opsin ran unbiased TMT proteomics on both chemotypes at matched exposure, so the comparison reflects what the compounds do at the free concentration each actually reaches. Degradation was scored against a CRBN knockout control to separate glue-dependent loss from general cytotoxicity.

The advanced chemotype removed fourteen proteins by more than twofold at 1 uM over 6 hours, SALL4 among them at 62 percent Dmax. The chemotype the programme had deprioritised removed two, and left SALL4 intact within assay noise.

Ternary complex cooperativity explained the difference. The selective chemotype forms a complex with a cooperativity factor near 9, so recruitment depends strongly on the induced interface, while the promiscuous one binds CRBN with high intrinsic affinity and recruits whatever presents a compatible degron.

Two of the twelve glues built to raise cooperativity lost GSPT1 degradation altogether, and the interface model gave no warning for either. Cooperativity above roughly 6 also brought a slower washout in three compounds, which matters for a mechanism where recovery is part of the safety argument.

Proteomics resolves what it detects. Low-abundance neosubstrates below the quantification limit are absent from the selectivity count, so two proteins degraded is a floor and not a guarantee.

Selectivity in this class is a property of the induced interface. Binary affinity for CRBN does not predict it, and high intrinsic CRBN affinity with low cooperativity produces a glue that recruits broadly, which is exactly the profile that reads as potent in a single-target assay and fails on proteome selectivity.

SALL4 sparing is not a late safety question. It is measurable in the same experiment that measures on-target degradation, at the point where a chemotype is chosen, ahead of developmental toxicology.

Proteomics used 16-plex TMT on MOLM-13 lysates, with degradation called against both vehicle and a CRBN knockout line so that glue-independent loss is excluded. Peptides were required in at least two of three replicates for a protein to be quantified.

Cooperativity was determined from the ratio of binary to ternary binding constants by surface plasmon resonance, with CRBN-DDB1 immobilised and neosubstrate titrated in the presence and absence of saturating glue.