researchsquare.comSource publication:
Loading content…
Whether tumor clones can be told apart by their gene regulatory networks had remained largely unmapped, since clones are usually defined by copy-number profiles and rarely characterized by regulatory activity. Clone-specific networks were inferred from single-cell RNA-seq and compared with a Wasserstein-distance measure of regulatory divergence. Across 253 patients and 12 cancer types, clone-clone networks resembled each other more than clone-normal pairs, and the largest clone diverged from normal cells significantly more than smaller clones.
Distinct clones within a tumor carry distinguishable gene regulatory networks, and clone-clone networks resemble each other more than they resemble normal-cell networks. Intratumor heterogeneity had been characterized mainly at the genomic level, with clones defined by somatic variants or copy-number profiles, while the regulatory wiring linking clonal genotype to transcriptional state remained largely unmapped. Across 9 triple-negative breast cancer and 6 colorectal cancer datasets, clone-clone Jaccard similarity exceeded clone-normal similarity, significantly for TNBC (p=0.038) and as a non-significant trend for CRC (p=0.089); in the 253-patient, 12-cancer-type extended cohort, median clone-clone Jaccard similarity was about 0.64 versus about 0.59 for clone-normal pairs (p=1.2×10⁻²³).
Clone size tracks regulatory divergence: the largest clonal subpopulation departs further from normal cells than the smallest clone. Clonal fitness had been read mainly from clone frequency and genomic features, without a quantitative comparison of whether regulatory reprogramming accompanies clonal expansion. Across the 15 datasets, 1-Wasserstein divergence from normal was significantly greater for the largest than the smallest clone (binomial p=4.8×10⁻⁴; Wilcoxon p=6.1×10⁻⁵), replicated in the 253-patient cohort (binomial p=2.4×10⁻²³; Wilcoxon p=2.9×10⁻²⁶), though the pooled correlation between clone size and divergence was weak (Spearman ρ=0.20).
Treatment-associated regulatory rewiring can exceed the divergence between clones sampled at the same time, showing structural divergence with functional convergence. Clone-evolution analyses had generally attributed pre- to post-treatment change to clonal selection, without separately quantifying the rewiring induced by therapy itself. In three longitudinal samples from one stage IV breast cancer patient, only 10.7% and 15.07% of the pre-treatment network's edges were shared with post-treatment 1 and post-treatment 2; pre-versus-post Wasserstein divergences (7.39 and 5.20) exceeded the post 1 versus post 2 value (3.78) and the divergence among clones within a timepoint.
Regulon analysis ties clone-level differences to specific transcription-factor programs, sorted into clone-variable, sample-specific and broadly conserved classes. Differential expression reports gene-level shifts but does not directly say which transcription-factor programs differ across clones and which are shared across samples. In TNBC and glioblastoma samples, regulons including GRHL2, SIX1, HEY1 and HEY2 varied across clones in multiple samples, while BCL11A, TBX3, TRPS1 and POU2F3 were broadly active across clones in multiple samples; some regulon-associated genes were significantly associated with patient survival in TCGA cohorts.
The analysis covers pilot cohorts in triple-negative breast cancer, colorectal cancer and glioblastoma, a 253-patient extended cohort spanning 12 cancer types, and two longitudinal treatment cohorts of one stage IV breast cancer patient and 11 ovarian cancer patients; conclusions are limited to clone-resolved regulatory networks inferred from single-cell RNA-seq.
Clones were computationally inferred from copy-number profiles rather than lineage tracing, and for some datasets clonal structure came from single-cell RNA-seq alone, the same modality used to infer the networks; the ovarian cross-phase comparison rests on 7 within-phase and 7 cross-phase clone pairs from 5 patients and is treated as suggestive; network inference is sensitive to noise and modeling assumptions, and inferred interactions still require experimental validation.
