Mutational Biases Drive Elevated Rates of Substitution at Regulatory Sites across Cancer Types

Abstract

Disruption of gene regulation is known to play major roles in carcinogenesis and tumour progression. Here, we comprehensively characterize the mutational profiles of diverse transcription factor binding sites (TFBSs) across 1,574 completely sequenced cancer genomes encompassing 11 tumour types. We assess the relative rates and impact of the mutational burden at the binding sites of 81 transcription factors (TFs), by comparing the abundance and patterns of single base substitutions within putatively functional binding sites to control sites with matched sequence composition. There is a strong (1.43-fold) and significant excess of mutations at functional binding sites across TFs, and the mutations that accumulate in cancers are typically more disruptive than variants tolerated in extant human populations at the same sites. CTCF binding sites suffer an exceptionally high mutational load in cancer (3.31-fold excess) relative to control sites, and we demonstrate for the first time that this effect is seen in essentially all cancer types with sufficient data. The sub-set of CTCF sites involved in higher order chromatin structures has the highest mutational burden, suggesting a widespread breakdown of chromatin organization. However, we find no evidence for selection driving these distinctive patterns of mutation. The mutational load at CTCF-binding sites is substantially determined by replication timing and the mutational signature of the tumor in question, suggesting that selectively neutral processes underlie the unusual mutation patterns. Pervasive hyper-mutation within transcription factor binding sites rewires the regulatory landscape of the cancer genome, but it is dominated by mutational processes rather than selection.

Fig 1. Functional binding sites are enriched for somatic mutations that are deleterious to motif binding potential.

(A) In the 1KG dataset, high frequency polymorphisms (>5% minor allele frequency) are depleted at functional binding sites compared to control sites (Wilcoxon test; p-value = 0.003), whereas the opposite trend is observed for somatic mutations (Wilcoxon test; p-value = 6.522e-09). (B) There are more somatic substitutions in functional, relative to control sites in cancer compared to 1KG polymorphisms (Wilcoxon test; p-value = 3.652e-10). (C) The relative change of the PWM-score is lower at functional sites compared to control sites in 1KG (Wilcoxon test; p-value = 8.442e-05), whereas the PWM-score introduced by somatic mutations is indistinguishable between functional and control sites. Each of the 118 binding motifs contributes one data point to the plots in this Figure.

Fig 2. Mutation accumulation and TFBS motif disruption in cancer compared to control sites and polymorphism data.

The X-axis shows the ratio of the number of substitutions in functional, relative to control sites in cancer, divided by the corresponding numbers for 1KG polymorphisms, i.e. (cancer_functional/cancer_control)/ (1KG_functional/1KG_control). Values > 1 indicate an excess of mutations in functional binding sites in cancer, correcting for the amount of variability that is tolerated at these sites at the population level. The Y-axis shows the corresponding ratio for the reduction in PWM-score. Values < 1 indicate that the matrix score is reduced to a greater extent in functional, relative to control sites in cancer compared to 1KG polymorphisms. The color and shape of the data points indicate the significance of their departure from random expectation. Note that some motifs were excluded from this plot because the p-value for the difference in PWM-score reduction could not be calculated (full list in S1 Dataset).

Fig 3. Somatic mutation and polymorphism patterns within TF binding sites.

Substitution counts across all binding sites for each of three motifs, selected from the full list of 118 motifs (see S8 Fig for similar plots for all motifs tested). For comparison, substitution counts at control sites and 1KG high frequency polymorphism counts are shown in the panels below. (A) Substitution counts for CTCF: MA0139.1. (B) Substitution counts for USF1: MA0093.2. (C) Substitution counts for ZBTB33: MA0527.1.

Fig 4. CTCF-binding sites are often mutated when acting as modulators of chromatin structure.

CTCF-motifs are highly enriched inside loop anchor points and across domain boundaries (A, C). The number of substitutions in CTCF-motifs is increased for motifs that are located in loop anchor points (B); for domain boundaries, no significant increase in substitution rate was observed (D).

Fig 5

A) The number of mutations inside functional and control TFBSs, plotted for each tumour sample in this study. B) Cancer mutations in TF binding sites, stratified by tumour type. Heatmap of mutation counts in functional motifs, relative to control sites and 1KG polymorphisms. Only motifs that occur in at least 1,000 binding sites in the genome are shown, for the four tumour types with the highest total number of mutations. Increasing shades of red: ratio > 1, indicating an excess of functional mutations in cancer; grey: ratio <1.