Colorectal cancer: what public data can and cannot tell us

This is not a treatment and not medical advice. Over one night we ran 10 rounds of analysis (16 completed analyses) on public cancer genomics: more than 3,700 tumours and more than 1,100 cancer cell lines (TCGA and other cohorts, plus the DepMap cell-line screens). The aim was to see where a treatment idea might have doors, and to say clearly which doors stayed shut.

Result: nine hypotheses ranked by strength of evidence, several honest negatives, and one correction. None is tested in a lab. None is a cure.

• Most colorectal cancers are microsatellite stable (MSS), roughly 80 to 85%, and these respond poorly to current immunotherapy. Colorectal Cancer Alliance; Postgraduate Medical Journal, 2023.
• In a phase 2 trial of 56 patients (UNICORN, Nature Medicine, 2 Nov 2026), a short course before surgery with botensilimab plus balstilimab gave a major pathological response in 36% of MSS-type (pMMR) colon cancers, and 0% with botensilimab alone. This is a small early trial in 14 patients per group. Nature Medicine.
• LY6G6D is a surface protein that is more common in MSS colorectal cancer and is being targeted by T-cell engager antibodies. A phase 1 study of one such drug (linclatamig) reported limited anti-tumour activity: 1 confirmed and 2 unconfirmed partial responses (AACR 2025 abstract). A second phase 1 study of CBI-1214, also an anti-LY6G6D/CD3 engager, is recruiting in MSS colorectal cancer (NCT07321106). These are early studies, not proven treatments.
• SHOC2 has been linked to resistance to MEK inhibitors in RAS-mutant cancers (Nature Communications, 2019), and a 2025 Nature paper describes targeting the SHOC2-RAS interaction (Nature, 2025).
• A 2025 study of 981 colorectal cancer genomes from 11 countries found that mutation signatures linked to the bacterial toxin colibactin were 2.5 and 4 times more common in cancers diagnosed before age 50. The authors suggest childhood exposure may contribute to early-onset disease. It is a possible driver, not a settled cause. Nature, 2025.

Source: published abstract, Nature Medicine, 2026 (doi link above). Small groups: 95% confidence intervals are wide (for example 12.8 to 64.9% for 36%). This chart is from the literature, not from our analysis.

These are results of our own analysis of public tumour data. In 1,038 tumours (374 right-sided, 664 left-sided), the mutation pattern differs sharply. KRAS is mutated in 59% of right-sided and 33% of left-sided tumours. TP53 goes the other way: 58% right, 81% left. BRAF (18% against 7%), RNF43 (16% against 3%) and PIK3CA (30% against 16%) are also more common on the right. All of these passed correction for many tests (q below 0.001).

Source: our analysis table of right versus left mutation frequencies, Fisher exact test with Benjamini-Hochberg correction. Observational data: it shows a difference, not a cause. Part of the right-sided difference may reflect MSI-high tumours, which sit mostly on the right.

Among 506 MSS tumours, KRAS-mutant ones carry a PIK3CA mutation in 36% of cases against 15% in KRAS-normal ones (odds ratio 3.2, q below 0.001). NRAS and BRAF mutations are rare when KRAS is mutated, as expected because they do the same job. The idea that follows is a hypothesis: a KRAS tumour may need a combined block of both routes.

Source: our analysis table of co-mutations in MSS tumours (196 KRAS-mutant, 310 KRAS-normal). Tumour sequencing data. Not tested in cells or patients.

In the DepMap cell-line screens, cells carrying a KRAS mutation depend more on SHOC2 (correlation 0.68) and RAF1 (0.67) than other cells, and a little more on YAP1 (0.40). All three passed correction for many tests. This matches the literature on SHOC2 and MEK resistance and ranks among our strongest hypotheses. It is a cell-line result. Some effects are not specific to the bowel: for SHOC2 the bowel lines were not different from the rest (p=0.59).

Cold versus hot tumours

MSI-high tumours carry far more immune signals (for example PD-L1, LAG3, PRF1, CXCL9) than MSS ones: 78 MSI-high against 504 MSS tumours, all with q below 1e-10. The immune gap is real and large.

WNT pathway

WNT and stemness signals are higher in MSS tumours. Cells with a high WNT score depend on their own pathway (beta-catenin, correlation -0.52), a sanity check that the score works. No new synthetic-lethal partner passed correction.

Surface targets

LY6G6D, CEACAM5, CDH17 and GPA33 are expressed on MSS tumours and are candidates for antibody-based drugs. LY6G6D is already in phase 1 testing.

• Our finding: 9 tumours carried a driver POLE mutation (about 1.6% of the MSS group), and 8 of the 9 sit inside the group labelled MSS. All 9 were extremely mutated (median 166 mutations per Mb). This is a computed subgroup of public data, not a measured share of patients who would respond.
• Literature: in a retrospective study of 27 patients with metastatic colorectal cancer and a proofreading-deficient POLE or POLD1 mutation, the response rate to immunotherapy was 89% against 54% in MSI-high cancers. It is a small, retrospective cohort, not a prospective trial. Annals of Oncology, 2024.
• Literature: a Japanese national registry of 127 patients with high mutation burden (10 or more mutations per Mb) who received pembrolizumab found much worse results in MSS than in MSI-high cancers: median time to treatment failure 2.0 against 10.6 months, median survival 4.5 against 33.6 months. The authors conclude that the 10 mutations per Mb cut-off does not reliably predict benefit in MSS cancers. ESMO Open, 2025.
• Regulatory context: in June 2020 the US FDA gave pembrolizumab accelerated approval, not full approval, for unresectable or metastatic solid tumours with 10 or more mutations per Mb that progressed after prior treatment and have no satisfactory alternative, using an FDA-approved test. The basis was a response rate of 29% across tumour types. It is a US decision and not an automatic route to immunotherapy for MSS colorectal cancer. FDA.

• Early-onset against later-onset colorectal cancer: after correction for many tests, no gene separated the two groups. Small raw differences exist (TP53 77% against 69%, BRAF 6.5% against 11.6%).
• Survival: no single mutation predicted survival in the TCGA data after correction. In stage IV (86 patients) APC-mutated tumours looked slightly better (14.9 against 13.1 months) but not after correction.
• The hottest tenth of MSS tumours (by immune activity) is not explained by high mutation burden or by POLE (p=1.0). What warms them is an open question.
• A search for a new synthetic-lethal weak spot in WNT-high tumours found nothing that passed correction. A metastatic survival dataset could not be reached, so that question stays open.

• Observational public data and cell lines. No patients and no laboratory test.
• Hypotheses are graded by evidence strength. None is a treatment recommendation.
• Right-versus-left and KRAS tables are descriptive. They do not show cause.