BGPT: Paper Review: Concurrent genetic and non-genetic resistance mechanisms to KRAS inhibition in CRC

Fuel Your Discoveries

Quick Explanation Copied

Key finding (skeptical read)

Across matched CRC biopsies and organoid models, the paper argues that resistance to KRAS G12C inhibition + EGFR inhibition is often mixed: genetic escape can be present (often subclonal) and simultaneously non-genetic adaptive transcriptional states (inflammation → EMT/YAP/fetal-like) appear, with TBK1 inhibition targeting an early inflammatory phase in a cancer-cell–autonomous manner.

Long Explanation

BGPT • Science Raw-Text Paper Review

Concurrent genetic and non-genetic resistance mechanisms to KRAS inhibition in CRC

Preprint DOI: 10.1101/2025.08.05.668666

1) What the study claims (and what’s actually measured)

Design: patient-matched CRC biopsies (pretreatment, on-treatment ~day 7–21, and post-progression) from KRASG12C + EGFR combination trials, analyzed by targeted exome sequencing (MSK-IMPACT) plus CosMx spatial transcriptomics; followed by organoid modeling and drug perturbations.
Genetic escape: acquired mutations were detected in 7/11 progression-available patients; most were subclonal and only a subset met a “clonal” threshold relative to preexisting clonal events; in 4/11 patients there were no identifiable acquired genetic events.
Non-genetic adaptation: on-treatment cancer-cell programs enriched inflammatory/cytokine and interferon signaling; progression samples enriched EMT, YAP, and fetal-like signatures; transcriptional reprogramming occurs in all progression biopsies analyzed.
Timing + autonomy: organoid experiments suggest early drug-induced inflammatory programs (and TBK1–IRF3 targets) appear before later regenerative/fetal-like resistance states, and these early inflammatory programs are at least partially cancer-cell autonomous.
Mechanistic lever: focused kinase perturbations identify TBK1 blockade (and momelotinib) as delaying outgrowth with KRAS inhibition; TBK1 shRNA phenocopies the combination effect; combined TBK1 + KRAS inhibition targets an early TBK1–IRF3 inflammatory phase.

2) Visual synthesis from the paper’s reported cohort numbers

2.1 Genetic escape detected at progression (MSK-IMPACT)

Based on the preprint’s stated counts, acquired genetic events are observed in most progression-sample patients, but not in all.

2.2 Clonality of de novo mutations in progression biopsies

The preprint reports 17 de novo mutations in progression biopsies with only 7 classified as “clonal” under their VAF-relative criterion.

2.3 Sample sizes and timepoints (as stated)

The preprint states 11 matched pre- and on-treatment biopsies, with on-treatment windows spanning day 7 to day 21; and 8 progression biopsies. (Exact per-timepoint distributions are not fully enumerated in the provided excerpt.)

Modality / stage	n (as stated)	Notes from paper text
Pretreatment biopsies (matched to on-treatment)	11	Matched pre- and on-treatment biopsies are collected prospectively.
On-treatment biopsies	11	Window: day 7 to day 21.
Post-progression biopsies	8	Used for progression analyses including targeted sequencing and spatial transcriptomics.
CosMx spatial transcriptomics (total cells QC-passed)	933,903 cells	Includes 409,858 cancer cells after QC across all patient samples.

3) Mechanistic model (paper’s narrative) as a knowledge graph

This directed graph visualizes the study’s proposed temporal logic: KRAS inhibition triggers an early inflammatory (TBK1–IRF3) cancer-cell program, which precedes (and may enable) later regenerative fetal-like / EMT / YAP states; genetic escape can coexist with non-genetic programs, with intratumoral spatial heterogeneity.

4) Evidence quality check (skeptical, mechanism-focused)

4.1 Strengths visible from the provided text

Multi-modal, longitudinal patient sampling: matching pretreatment → on-treatment → progression helps separate early vs late transcriptional adaptations, reducing “post hoc” ambiguity relative to single timepoint studies.
Spatial context: intratumoral heterogeneity is explicitly addressed (zones with different adaptive programs), which is critical for resistance biology.
Orthogonal perturbation logic for TBK1: (i) focused pharmacologic screen, (ii) concordant pathway activation (IRF3 targets), and (iii) genetic mimicry via two independent doxycycline-inducible shRNAs support TBK1 sufficiency within their experimental frame.

4.2 Red flags / uncertainties explicitly implied by the excerpt

Small cohort for genotype–state correlations: the preprint itself states cohort size limits ability to correlate specific genomic profiles with transcriptional adaptations.
Autonomy conclusion depends on ligand–receptor panel completeness: CellChat analysis found limited predicted signaling to cancer cells within the 1000-gene panel, but the preprint explicitly notes it cannot exclude microenvironmental ligands not present in the panel.
Multiple mechanisms may be co-existing, but disentangling causality is hard: the paper presents mixed genetic and non-genetic programs coexisting; however, the excerpt does not show (here) whether inflammatory programs are always upstream of fetal-like state emergence in genetically escaped zones or vice versa. This remains testable, not established beyond their models.
Therapeutic framing vs mechanistic specificity: TBK1/IKK inhibitor and momelotinib target related kinases; the authors use shRNA to argue TBK1 sufficiency, but off-target contributions outside the excerpt cannot be fully ruled out without supplemental specificity profiling.
Experimenter blinding not applied (as stated): the statistical methods mention experiments were not blinded to conditions, which can affect subjective scoring or imaging-based endpoints (if any are used).

5) What would most strongly falsify the TBK1 “early phase” story?

Temporal inversion: if early inflammatory / IRF3 target upregulation does not precede or does not track with later fetal-like/yap/EMT emergence in independent models, the proposed ordering weakens.
Genetic independence: if TBK1 knockdown (or pathway blockade at IRF3) fails to reproduce combination synergy across multiple organoid lines, TBK1’s causal role would be less supported.
Microenvironment dependence: if inflammatory programs require microenvironmental ligands outside the CosMx panel and disappear in cancer-cell–only settings where panel ligands are absent, autonomy would be overstated.

6) Conflicts of interest (from provided manuscript text)

Stated disclosures in the excerpt

LED is a consultant and has received research funding from Revolution Medicines; Weill Cornell licensed KRAS-related technologies to Boehringer Ingelheim, Amgen, and Revolution Medicines.
RY served as an advisor for Mirati Therapeutics.

This doesn’t negate the biology, but it does raise the bar for mechanism clarity and independent replication.

7) Data availability (as stated)

Deidentified CosMx data available at GEO: GSE293124.
Raw RNA-seq FASTQ files at SRA: PRJNA1244364.

Author-review deep dives (BGPT links)

This will iteratively re-check mechanistic claims against the deposited CosMx and RNA-seq accessions stated in the manuscript, and regenerate additional quantitative summaries from those raw resources.

Feedback:

Updated: March 28, 2026