BGPT: Paper Review: Discovery of a CNS active GSK3 degrader using orthogonally reactive linker screening

Explore by Goal

Quick Explanation Copied

Paper review focus

Main contribution: A “direct-to-biology” PROTAC discovery workflow using orthogonally reactive linkers to vary E3 binder, target binder, and linker chemistry in one 1-well, 2-step library build—then screen immediately in cells.
Key biological outcome: KH1 (24) shows in vitro fast degradation (near-complete at 10 nM within ~2 h) and in vivo detectable brain exposure and significant liver and brain GSK3β reduction after i.v. dosing, supporting CNS-active pharmacology for a chemical-probe-grade degrader.
Main skepticism points: Selectivity is impressive at early timepoints and in the tested models, but the paper’s in vivo PK/PD window, species/sex scope (female Balb/c), and the reliance on crudes-to-purity re-synthesis shift (DC50 left-shift) should motivate further independent confirmation for broader generality.

Long Explanation

Discovery of a CNS active GSK3 degrader using orthogonally reactive linker screening

Nature Communications (2025) · DOI: 10.1038/s41467-025-63928-8

Below: visual-first breakdown of (i) the linker/D2B platform, (ii) the GSK3β-HiBiT screen logic, (iii) probe validation (mechanism, selectivity, substrate engagement), and (iv) the in vivo PK/PD inference chain—with explicit skepticism about uncertainties and model limits.

Core claims are tied to the paper text.

Figure A — Library construction logic (orthogonal linkers → 2-step D2B)

The platform’s key premise is orthogonality: each linker end reacts selectively with its designated binder counterpart in a stepwise fashion (S_N2 then CuAAC), enabling rapid parallel exploration of E3 binder, POI binder, and linker chemotype.

Workflow schematic (text-based)

1) S_N2 end on linker reacts with Binder A counterpart (solvent DMF; base Et₃N; optimized additives including THPTA for sequential CuAAC).

2) CuAAC “click” end on linker reacts with Binder B azide counterpart.

3) SCX resin catch-and-release separates full PROTACs from unreacted binders/byproducts.

4) Screen crude mixtures directly in cells using HiBiT-tagged GSK3β abundance readout.

Figure B — What the screen generated (purity & detection yield)

They report LC–MS UV-purity-based detection outcomes to justify that crude mixtures are still informative for direct-to-biology screening. In their plate workflow, 147 compounds were detected and 103 had >50% UV purity across two plates.

Figure C — Direct-to-biology screen → hit enrichment

They define hits based on D_max and DC₅₀ criteria computed from HiBiT dose-response heatmaps. They state 6 compounds from plate 1 met >50% D_max and <1 µM DC₅₀, while 35 compounds met the same criteria from plate 2.

Figure D — Crude vs purified: DC50 shifts indicate assay-relevant “real concentration” uncertainty

A critical practical point: crude stock assumptions were based on assumed complete conversion/purity at a target concentration; upon purification/resynthesis, they report 2.5–20 fold left-shifts in DC₅₀ with unchanged D_max. They interpret this as reflecting actual vs assumed concentrations and remaining competing impurities.

Skeptical interpretation (what this means)

Known: The screen likely still ranks series correctly (rank order preserved) because D_max didn’t collapse and potency improves consistently after purification.

Uncertain: Exact absolute DC₅₀ values in the initial screen depend on how well the assumed 10 mM “crude” concentration matched the true active concentration in each well.

Figure E — Probe biology chain-of-evidence: CRBN → UPS → target engagement → substrate kinetics

For KH1 (24) and KH2 (26), they report: (i) rapid degradation of endogenous GSK3α and β with ~10 nM giving near-complete degradation at ~2 h, (ii) rescue by proteasome inhibition (MG132) and neddylation inhibition (MLN4924), and (iii) CRBN dependency via negative controls and CRBN KO rescue, supporting the intended PROTAC mechanism.

Critical note on causal order (“activity before degradation?”)

They test whether GSK3 activity is affected prior to degradation. They report that at 30 min, there is no appreciable impact on inhibitory Ser9 or activating Tyr216 phosphorylation; changes appear later (concurrent with degradation at ~2 h), consistent with a catalytic/non-occupancy degrader mechanism.

What is claimed: They report “exquisite” selectivity for GSK3 paralogs after 10 nM compound treatment for 2 h, with little proteome-wide abundance alteration even at 24 h in HEK293. They also describe a small β-catenin increase at 4 and 24 h consistent with substrate biology.

Skepticism: “Exquisite selectivity” depends on detection limits, chosen timepoints, and which proteins are quantified with sufficient confidence; the text provided does not include thresholds/coverage metrics for the proteomics calls. This is therefore strong within the experimental design, but not absolute across the entire human proteome.

Figure F — In vivo PK/PD inference: free fraction → exposure → degradation outcomes

A notable part of the paper is how they attempt to translate low brain penetration into a plausible free exposure that could explain degradation. They use: equilibrium dialysis plasma free fraction and a serum shift assay to cross-check the functional consequence of protein binding on DC₅₀ in the presence of serum.

Reported in vivo degradation outcome

After a single i.v. dose of KH1 (24) at 5 mg/kg, at 4 h they report almost complete GSK3β removal in liver and significant degradation of GSK3β in mouse brain (with the proteome/phosphoproteome and substrate markers assessed from the same timepoint/tissues).

Skepticism checklist for PK/PD causality:

Known: They estimate free fractions and interpret predicted unbound concentrations to guide dosing expectations.
Uncertain: The free concentration estimates rely on in vitro binding assays and on translation of dialysis/serum shift behavior to the actual in vivo milieu and brain microenvironment. The text provided does not include explicit confidence intervals or assay variance for all binding fractions.
Model limitation: The in vivo pharmacology evidence is presented in female Balb/c mice; generalization to other strains/sexes/disease contexts is not fully established in the provided text.

Directed critique — what’s strongest vs what’s still missing

Strengths

Platform insight is testable: The orthogonal linker concept is operationalized with an explicit synthesis/screen pipeline, and the paper provides enough procedural structure to judge plausibility (S_N2 → CuAAC → SCX → D2B HTS).
Evidence chain for degrader mechanism: Proteasome/neddylation rescue plus CRBN dependency and order-of-events experiments support that degradation is not simply a non-specific toxicity artifact.
Proteomics + phosphoproteomics provide kinetic context: They attempt to map how phospho-substrate vulnerability evolves as GSK3 is removed over time.
In vivo demonstration: They show liver and brain GSK3β reduction at 4 h after a single i.v. dose and connect it to brain/liver exposure with measured PK and binding inference.

Potential blind spots / uncertainties

Crude mixture concentration uncertainty: DC₅₀ left-shifts from crude → purified indicate that initial absolute potency estimates are not directly interpretable without purification.
Selectivity claims depend on coverage: “Little alteration” and “exquisite selectivity” depend on quantitative depth and thresholds; the provided text does not show full proteomics coverage metrics (e.g., % of proteome quantified per sample) needed to rigorously bound undetected off-targets.
Generalization beyond the demonstrated design space: The platform is argued to be generalizable, but the demonstrator chemistry uses CRBN-centric design sets and particular linker/binder chemotypes; extension to different E3 ligases/POI classes is not established in the provided text.
In vivo validation depth: The text supports brain/liver pharmacology at one timepoint (4 h) and shows proteomics changes at that time. Longer timepoints, behavioral/functional endpoints, and additional brain regions/cell types are not described in the provided excerpt.

Useful navigational next-steps on BGPT (bespoke)

Author reviews (open-ended, linked on BGPT)

Feedback:

Updated: April 07, 2026