BGPT: Paper Review: Proteogenomic Analysis of Human Colon Cancer Reveals New Therapeutic Opportunities

Explore by Goal

Quick Answer Copied

Paper in 1 line

A CPTAC proteogenomics study of 110 colon cancer patients (tumor + matched normal + blood) integrates WXS/RNA-seq with global and phospho TMT proteomics to link mutations → protein/PTM changes, define unified multi-omics subtypes (UMS: MSI/CIN/Mesenchymal), and generate therapeutic hypotheses including RB phosphorylation–CDK2 and MSI glycolysis–immune suppression.

Source:

To go deeper with an iterative “run tools/code” agent, use the button below.

Long Answer

Proteogenomic Analysis of Human Colon Cancer Reveals New Therapeutic Opportunities (Cell, 2019)

Critical, evidence-based review focused on what the paper does mechanistically, what it shows, and what it does not prove.

1) What the study claims (and why it matters)

Multi-omics integration across prospective colon tumors (tumor + matched normal adjacent tissue (NAT) + blood) to connect genotype (WXS/MSI/SCNA) with protein state (global proteome) and signaling/PTMs (TMT phosphoproteome).
Therapeutic hypothesis generation via three mechanistic threads explicitly highlighted in the provided full text: (i) RB phosphorylation as a driver tied to CDK2 activity, (ii) MSI subtype glycolysis with an association to reduced CD8 T-cell infiltration, and (iii) proteomics-supported tumor antigens (neoantigens and cancer-testis antigens).

2) Core quantitative outputs (sanity checks you can audit)

Values come directly from the paper’s reported tumor vs NAT differential protein analysis thresholds (adj. p < 0.01, paired Wilcoxon signed-rank; fold-change > 2 defines the highlighted “colon cancer-associated proteins”).

MSI-H vs MSS counts were explicitly stated (MSI-H: n=24; MSS: n=82) after WXS-based bimodal MS INDEL distribution and confirmation agreement with PCR when available.

3) Mechanistic focus: RB phosphorylation / CDK2 (what’s strong vs what’s inferential)

3.1 What they directly measured

RB1 amplification at the genomic level and increased total Rb protein in tumors vs NAT are reported.
Phospho-Rb sites (T373, S807, S811, T826) show higher tumor abundance, and the paper reports a tumor vs NAT fold change for phospho-Rb relative to total Rb.

3.2 What they infer (and a skeptical check)

E2F1 activity is “inferred” using tumor-normal protein level changes and the paper reports positive correlations between Rb change and inferred E2F1 activity.
CDK2 activity is “inferred” using tumor-normal phosphorylation levels; phospho-Rb change shows the strongest correlation with inferred CDK2 activity compared to other kinases.

Critical caveat: these “activities” are computational inferences. They are consistent with known RB/E2F biology, but they are not the same as direct kinase activity assays or perturbation experiments inside this colon cohort (the paper itself frames validation as beyond scope).

4) Mechanistic focus: MSI glycolysis–immune suppression (what’s explicitly shown)

MSI protein-level metabolic shift: broad increase in glycolytic enzymes and near-universal decrease in TCA cycle enzymes at the protein level (not at mRNA).
Association with CD8 infiltration: glycolytic activity (median protein abundance of measured glycolytic enzymes) is negatively correlated with CD8 infiltration in MSI (Spearman r = −0.61, p = 0.02).
Targeted SRM validation: SRM (and PRISM-SRM) shows correlations between SRM and TMT for CD8A and SLC2A3 and identifies PKM2 as the major isoform in the targeted SRM panels.

Critical caveats: correlations are not causality; infiltration is inferred from RNA signatures using computational methods (ESTIMATE/GSVA described in STAR Methods).

5) Antigen discovery: what is “proteomics-supported” and what remains uncertain

Neoantigens: they searched customized protein databases incorporating coding variations from matched exome sequencing and RNA-seq, evaluated 9–11 aa mutant peptides for HLA-I binding using netMHCpan, and define high-affinity peptides as putative neoantigens (IC50 < 150 nM).
Cancer-testis (CT) antigens: they identified 16 CT antigens in TMT global data, with IGF2BP3, SPAG1, ATAD2 elevated in >5% of tumor–NAT pairs (IGF2BP3 in 51% of pairs by their report).

Critical caveat: “putative neoantigens” rely on in silico HLA binding thresholds and proteomics identification; immunogenicity and presentation on MHC are not directly shown in this paper’s provided text.

6) Subtyping: unified multi-omics subtypes (UMS) and what they resolve

The paper integrates CMS (consensus molecular subtypes classifier), ProS proteomic subtypes, and MSI status into a network and defines three unified multi-omics subtypes: MSI, CIN, and Mesenchymal.
They then show orthogonal subtype-consistent signals (e.g., CIN subtype shows higher chromosome instability; mesenchymal subtype shows higher stromal infiltration; MSI subtype enriched with cytotoxic immune cells including NK and activated CD8 T cells).

7) Scientific quality and skepticism checklist (what I would scrutinize next)

Strengths

Prospective cohort + matched normals (tumor + NAT + blood) supports tumor-vs-normal proteomic contrasts and improves interpretability of expression differences.
Multiple proteomics modalities: label-free shotgun plus TMT global and TMT phosphoproteomics, plus targeted SRM/PRISM-SRM validation for selected proteins.

Red flags / limitations to audit

Inference vs causality: kinase “activity” and transcription factor “activity” are inferred computationally from phosphorylation/protein changes, not directly measured; therapeutic targeting remains hypothesis-level in this study’s scope.
Rectal cancer generalization: the paper focuses on colon tumors only and explicitly contrasts with TCGA being colon+rectal; it argues for cohort correlations but does not directly validate rectal.
Platform effects: while they report that TMT and label-free protein patterns correlate and batch effects were not observed between TMT plexes, residual measurement bias can still affect subtle PTM comparisons.

8) “What would disprove key conclusions?” (falsification targets)

If RB phosphorylation is not elevated in tumors or does not align with inferred E2F1/CDK2 activity across independent cohorts, the proposed RB phosphorylation–CDK2 axis would be weakened.
If the MSI glycolysis–CD8 infiltration relationship disappears when infiltration is measured differently (e.g., direct cell sorting) or after controlling for cell-composition/covariates, the association-driven immune suppression model would need revision.
If proteomics-supported neoantigen/CT antigen candidates do not show evidence of presentation and immunogenicity, their utility for vaccine prioritization would be reduced despite computational plausibility.

Go deeper: run an iterative “AI Scientist” agent on your behalf

Author reviews (bespoke BGPT links)

Feedback:

Updated: March 29, 2026