BGPT: Paper Review: KRAS Binders Hidden in Nature

Fuel Your Discoveries

Quick Explanation Copied

What this paper actually does

Builds a natural-product-first discovery pipeline for KRAS pocket binders by combining virtual screening, deconstruction-based screening of a natural-products database, SAR-by-NMR, and structure-guided follow-up with X-ray crystallography ().
Biophysically validates weak KRAS G12D-active-state binding (millimolar KD range) for specific hits using 2D 1H/15N HSQC NMR and confirms binding poses with X-ray structures of KRAS G12D bound to those hits ().
Targets the SI/II pocket using a docking surrogate (GDP-KRAS co-crystal structure with indole 7; PDB 4EPV) to reduce the problem of scarce active-state structures ().

Long Explanation

Paper Review (visual-first): KRAS Binders Hidden in Nature

DOI: 10.1002/chem.201902810

One-line claim (as stated)

An integrated natural-product screening approach uncovered multiple new KRAS binders and validated binding to active-state KRAS G12D using NMR and X-ray crystallography, including a new crystal soaking system and a “snugness of fit” scoring function ().

1) Pipeline anatomy (what was chained to what)

Screening used a virtual natural-product library derived from CRC Dictionary of Natural Products (DNP) via deconstruction to make entries amenable to virtual screening ().
Docking was performed against the SI/II-pocket of a GDP-KRAS co-crystal surrogate (PDB 4EPV with indole 7) because active-form X-ray density was ambiguous at the time ().
Hits were filtered by docking-score heuristics (top 500 by MW-normalized Glide score) and structural hydrogen-bond logic (mimic indole occupation; H-bonds to D54) followed by chemical-stability pruning ().

2) Biophysical binding strength (validated hits only, as reported)

Evidence in the excerpt (HSQC 2D 1H/15N):

9b: cross-peak shifts with GCP-KRAS G12D; millimolar-range KD with no saturation up to 1.5 mM ().
15R: HSQC titration reported KD of ~1 mM and binding validated by crystal structure after soaking ().
15S: bound more weakly with a millimolar-range KD ().

3) Diversity of natural-product sources among virtual hits

Important skepticism: the excerpt lists many example scaffolds and hit classes, but does not provide a complete “8 scaffolds” breakdown table. The bar chart above reflects only hit classes explicitly enumerated in the provided text (e.g., β/γ-carbolines, bromopyrrole alkaloids, diffusarotenoids/vasicine/quinocarcinol, indolopyrrole alkaloid, and a natural-product-like corporate-library class) ().

4) Binding-mode logic: “SI/II pocket” + D54 hydrogen bonding

What is explicitly mechanistic in the excerpt

The authors emphasize a shallow polar SI/II-pocket that exists in an active, oncogenic form of KRAS, and they argue scaffolds that bind the active form are scarce ().
Selection criteria include mimicking indole 7 occupancy and forming hydrogen bonds with D54 in the SI/II pocket ().
Confirmed binding poses: 9b is reported to form two H-bonds to D54 in the crystal structure; 15R is reported to form two chelating H-bonds to D54 plus a weaker additional H-bond to the K9 sidechain via a carbonyl group ().

5) Claimed method innovations (with skeptical reading)

5A) “Snugness-of-fit” scoring function (grid-envelope overlap)

The excerpt defines the scoring concept as protein/ligand envelope overlaps across multiple atomic radii on 3D grids, interpreting small-radius overlaps as “snug” contacts and larger-radius overlaps as less snug ().
They apply this post-processing score to a “kinked” rotenoid-like library (1383 molecules) to select candidates for NMR testing, ultimately finding 15R/15S binding by NMR ().

Critical limitation (based on excerpt content only): the scoring function is described operationally, but the excerpt does not specify (i) training/validation against known binders, (ii) robustness to docking pose uncertainty, or (iii) how snugness improves enrichment relative to Glide alone. Those details are likely in the full paper/SI, but are not present in the provided text.

5B) First crystal soaking system for active KRAS G12D (as claimed)

The excerpt states they developed a robust crystal soaking system for weakly binding fragments (KD > 1 mM) using GDP-KRAS G12D with nucleotide exchange to a non-hydrolysable GTP analogue GMP-PCP (GCP) to yield a novel crystal form that mimics a β-sheet dimeric form of active KRAS G12D ().
They report a co-crystal structure for 9b with GCP-KRAS G12D at 1.2 Å, with binding mode confirming docking predictions (SI/II-pocket; two H-bonds to D54) ().

Critical limitation: the excerpt claims “first” and “mimics” but does not supply comparison benchmarks or alternative soaking controls; these would be essential to independently assess whether the system truly generalizes or merely works for the specific hits shown.

6) What was validated biophysically/structurally (explicitly in excerpt)

Compound	Source class (as stated)	Protein state used for validation	Binding readout	Reported affinity scale	Binding-mode confirmation
9b	Analogue of tricyclic indolopyrrole alkaloid (fungal context)	GCP-KRAS G12D	2D 1H/15N HSQC NMR cross-peak shifts + KD titration	Millimolar; no saturation up to 1.5 mM	Co-crystal structure at 1.2 Å; SI/II-pocket; two H-bonds to D54
15R	Rotenoid-like “kinked”/indoloisoquinolinone enantiomer	GCP-KRAS G12D	2D 1H/15N HSQC NMR	KD ~1 mM	Crystal structure after soaking; chelating H-bonds to D54 (two) + weaker H-bond to K9
15S	Rotenoid-like “kinked”/indoloisoquinolinone enantiomer	GCP-KRAS G12D	2D 1H/15N HSQC NMR	Millimolar range; weaker than 15R	Docking mode described; crystal validation is described for 15R in excerpt

Source for table entries is limited strictly to what is described in the provided excerpt ().

7) Skeptical critique (what could be misleading or incomplete)

Key blind spots visible from the excerpt alone

Surrogate model risk: docking started from GDP-KRAS (4EPV with indole 7) rather than a fully resolved active-state electron density map; mismatch between GDP-bound and GTP-bound conformational ensembles is a known failure mode in pocket-binding discovery, and the excerpt does not quantify docking error or pose ensemble uncertainty ().
Affinities are weak: validated KD values are millimolar-scale, which is compatible with crystallographic soaking of fragments but raises the question of whether these scaffolds translate into potent and selective inhibitors without major medicinal chemistry optimization ().
Scoring-function generalization unknown: the snugness-of-fit definition is clear, but the excerpt does not show how it was benchmarked (e.g., against known binders) or whether it robustly ranks true binders when docking poses are uncertain ().
Library composition and selection bias: vDNP deconstruction and subsequent filters can bias what survives to NMR testing; likewise, “kinked” library generation depends on their conformational hypothesis for rotenoids being flat and hence requiring a kinked alternative space. The excerpt states these steps, but does not quantify how much they reduce the search space versus increase enrichment ().

8) How the provided excerpt maps onto the paper’s figures

Figure-level anchors (from excerpt)

Fig. 1: examples of secondary metabolites with multiple fused rings ().
Fig. 2: selected virtual hits and indole 7 present in the GDP-KRAS co-crystal used for docking ().
Fig. 3: docking poses in SI/II-pocket for multiple hits including 7, 8, 9a, 10b, 11b, 12b, 13, 14b ().
Fig. 4 and Fig. 5: NMR titration and X-ray co-crystal structures for 9b and 15R, respectively ().

Author reviews to go deeper

Feedback:

Updated: March 22, 2026