Ultraconserved elements as a new class of non-coding cancer drivers

This was one of my earliest journal clubs, and I chose it because it’s almost a thesis statement for my whole PhD: it takes a class of non-coding sequence nobody had systematically mutated in cancer, and shows, statistically and at the bench, that those sequences carry real driver mutations.

Why ultraconserved elements?

Ultraconserved elements (UCEs) are stretches >100 bp that are 100% identical across at least three of five placental mammals. That degree of conservation is extraordinary, evolution clearly won’t tolerate change there, yet most UCEs are non-coding (intronic, intergenic, regulatory) and their role in cancer was unknown. The hypothesis is simple and bold: if a sequence is that conserved, a somatic mutation in it should matter, and some of those mutations might be drivers.

A nice sanity check up front: across the 1000 Genomes Project, UCEs carry significantly lower germline SNP density than the genome or even the exome, they’re under strong purifying selection in the germline. Which makes their somatic behaviour in tumours all the more interesting.

The mutational landscape

Across 2,449 primary tumours (22 cancer types, PCAWG), they found 24,039 somatic mutations in ~73% of UCEs. Two numbers stood out to me:

• UCEs have a higher somatic mutation rate per Mb than coding regions (4.41 vs 2.93), and within UCEs the non-coding ones (ncUCEs) mutate more than coding UCEs.
• ncUCEs with deeper conservation (≥90% across 50 species) harbour more mutations, the most ancient regions are the most mutated in cancer. That’s a counter-intuitive, selection-flavoured signal worth sitting with.

Are they actually drivers?

This is the part that maps onto my Objective 2 (finding positively selected non-coding hotspots). They ran a mutation-burden test on individual UCEs plus a simulation test (UCEs matched for size, annotation and ±1 Mb local context, 1,000 permutations). Real ncUCE driver candidates ranked in the top 0.5% of the simulated distribution, i.e. excess burden beyond chance, the signature of positive selection. Six ncUCE driver candidates emerged (FDR < 0.05), including the 5′UTR of BCL6, the promoter of PAX5, and splice sites of FBXO11/ARHGAP15/EBF1, several already COSMIC cancer genes.

From statistics to function, the bench half

What elevates this paper is that they didn’t stop at a list. A multiplexed CRISPR–AsCpf1 screen of 2,247 recurrently mutated ncUCEs across three colorectal lines returned 35 hits that altered proliferation, and they dissected two mechanistically:

• UCE_11311 is an enhancer of ARID1B (a SWI/SNF tumour suppressor). It carries H3K27ac, and knocking it out lowers ARID1B and drives a pro-oncogenic phenotype, more colony formation, larger xenograft tumours, higher Ki-67.
• UCE_2272 is a silencer of RPS13. It carries H3K27me3, represses a reporter in luciferase assays, and its knockout de-represses RPS13 (an aggressiveness marker) and again grows bigger tumours.

They also showed MIR142 ncUCE mutations disrupt miRNA processing, mutations in the CNNC motif impair SRSF3-assisted Drosha processing, reducing mature miR-142. So ncUCE mutations act through at least three distinct mechanisms: enhancer loss, silencer loss, and miRNA-processing disruption.

Why it mattered for me

This paper is, in miniature, the full arc I want my PhD to run: define a non-coding sequence class, test it for positive selection with a properly matched null, then validate candidates functionally, discovery → selection → mechanism → bench. It’s also a reminder that conservation is a powerful prior for picking where in the non-coding genome to look. The main gap, from my angle, is that it’s pan-cancer and colorectal-focused; the breast-specific, regulatory-map-aware version of this analysis is part of what I’m building.