When the transcriptome forgets but the epigenome remembers

Chronic inflammation is a well-established cancer risk factor, ulcerative colitis patients are 2–5× more likely to develop colorectal cancer, but the molecular why has stayed fuzzy. This paper gives a strikingly clean answer: the tissue keeps a memory of the inflammation written in chromatin, long after the inflammation itself is gone, and that memory primes it for tumours. I picked it because the methods and the molecular cast (AP-1, FOXA1, chromVAR) overlap almost exactly with my own breast-cancer atlas.

The central dissociation

Using a DSS colitis model with SHARE-seq (joint scATAC + scRNA from the same cell) across control, acute, chronic and recovery stages, they find the result the whole paper turns on:

97% of inflammation-activated genes return to baseline after recovery, but the chromatin does not. Recovered stem cells stay epigenomically distinct from controls.

The memory has a dominant signature: cumulative gain of AP-1 (FOS/JUN) motif accessibility. About 9.2% of recovered stem cells carry exceptionally high AP-1 accessibility (vs 1.6% in controls), and, the part I find wild, it persists >102 days, dozens of generations of epithelial turnover. And it’s independent of the protein: FOS protein peaks during injury then drops within 21 days, while the AP-1 chromatin accessibility lags and stays high. The memory is in the chromatin structure itself, reinforced by concordant DNA methylation changes (4,397 regions: accessibility up, methylation down).

It’s clonally inherited

They built SHARE-TRACE (SHARE-seq + lentiviral clonal barcoding) to ask whether these states are heritable through cell division. They are: cells within a clone resemble each other more than random cells, and AP-1 is the only TF with memory both in vivo and through clonal lineages. The in-vivo and ex-vivo distributions of high-AP-1 cells match (~9–12%), which means the heterogeneity is maintained cell-intrinsically, not by signalling from the niche.

FOX stabilises AP-1, and the tumour payoff

A deep-learning footprinting tool (seq2PRINT) discovered a composite AP-1/FOX motif de novo, and biochemistry confirmed FOX TFs cooperatively stabilise AP-1 binding at memory sites (FOXP1+AP-1 up to 8.3×), even via protein–protein interaction without the FOX DNA motif (AlphaFold3-backed).

Then the functional payoff: in an APC-loss adenoma model, colitis-recovered mice grow larger early tumours (not more tumours, bigger ones), spatial transcriptomics shows colitis tumours activate the AP-1 wound-healing program (P20), and, the rescue, AP-1 inhibition during initiation cuts tumour size ~40%, specifically in colitis-memory mice. Importantly, inhibiting AP-1 blocks the phenotype without erasing the memory (chromatin + methylation persist), so the memory is the durable substrate, and AP-1 activity is what reads it out.

My critical read

The strengths are obvious, an unusually complete validation chain (in vivo → organoid → clonal → biochemical → structural), extended timepoints, and a human IBD-organoid translation. The honest limits: it’s a colitis/colorectal model only, DSS is chemical injury rather than true autoimmune UC, the colitis→cancer link in humans is still inferred from mouse, and they lack WGS to fully exclude a somatic-mutation contribution to clonal fitness. That last one matters to me directly, it’s the exact place my non-coding-mutation work meets their epigenetic-memory work. Their question is “what does chronic injury write into chromatin?”; mine is “what do somatic mutations write into the same regulatory elements?” The most interesting version of breast-cancer biology is probably where those two answers meet, heritable epigenetic state and the mutations that lock it in.