Anyone who has watched a PROTAC look great in human lines and flat in a mouse efficacy study will recognize this. It is a reminder that degrader PK/PD in rodents can fail for E3 reasons rather than compound reasons, so a CrbnI391V or humanized-E3 model is worth setting up early. It also argues for profiling ternary complex formation across species before reading a negative in vivo result as a dead series.

Tubulin binders that dodge P-gp efflux are always of interest, and low efflux ratios plus retained activity in resistant lines is the right first filter. The zebrafish result is the honest sting in the tail: potent antimitotics often have a thin margin, so selectivity over fibroblasts in culture will not settle the safety question. I would look next at the binding site and how far the scaffold can be tuned toward a wider window.

STING inhibitors have mostly been tool compounds or covalent binders, so a non-covalent-looking, orally dosed chemotype that acts by blocking palmitoylation is a different handle on the target. I would want to know whether it truly binds the Cys91 pocket or acts upstream, because that decides selectivity versus other palmitoylated proteins. The Trex1 mouse data is the convincing part for an autoimmune indication.

This is a tidy example of phenotype-led off-target triage: odd toxicity in cell panels, a mechanistic guess, then SAR that separates the two activities. Endosomal TLR7/8 antagonists need to be both cell-permeable and cleanly selective, and BRD4 is an easy trap for flat heteroaromatics. I would check the pyridone's binding mode in the TLR7/8 dimer interface before borrowing the fix for other scaffolds.