DNA replication timing and long-range DNA interactions predict mutational landscapes of cancer genomes

Abstract

Somatic copy-number alterations (SCNA) are a hallmark of many cancer types, but the mechanistic basis underlying their genome-wide patterns remains incompletely understood. Here we integrate data on DNA replication timing, long-range interactions between genomic material, and 331,724 SCNAs from 2,792 cancer samples classified into 26 cancer types. We report that genomic regions of similar replication timing are clustered spatially in the nucleus, that the two boundaries of SCNAs tend to be found in such regions, and that regions replicated early and late display distinct patterns of frequencies of SCNA boundaries, SCNA size and a preference for deletions over insertions. We show that long-range interaction and replication timing data alone can identify a significant proportion of SCNAs in an independent test data set. We propose a model for the generation of SCNAs in cancer, suggesting that data on spatial proximity of regions replicating at the same time can be used to predict the mutational landscapes of cancer genomes.

Figure 1

Long-range DNA interactions and the distribution of SNCAs with regard to replication timing zones. (a) Organization of genomic DNA with long-range interactions between distant replication timing zones can increase the risk of interference between adjacent replication forks, leading to genomic alterations. (b–e) The two boundaries of SCNAs are significantly more likely to reside in genomic regions with the same replication timing than that expected by chance for all SCNAs (b), somatic copy number amplifications (SCNA-Amplifications) (c), somatic copy number deletions (SCNA-Deletions) (d) and SCNAs that overlap with known cancer genes listed in the Cancer Gene Census (SCNA_CGC) (e). Supplementary Module 3 provides separate analyses of SCNA amplifications and deletions as identified by GISTIC. The absolute number (n) of observed and expected cases is provided below each bar. *, P = 1.21 × 10^–5; **, P = 1.98 × 10^–5; ***, P = 3.76 × 10^–6.

Figure 2

SCNA frequencies vary between different replication timing zones. (a–c) Observed (black) and expected (gray) proportions of SCNA boundaries in early-, mid- and late-replication timing zones. The absolute number (n) of observed and expected cases is shown inside each bar. Similar graphs are shown for somatic copy number amplifications (SCNA-Amplifications) (b) and somatic copy number deletions (SCNA-Deletions) (c). The triangle reflects the direction of enrichment. (d) SNP-chip log₂ ratios indicating SCNAs with boundaries in genomic regions within early- and late-replication timing zones. The dashed line, drawn along the median of late-replication timing data points, serves to highlight the difference with early-replication timing data points. (e) SNP-chip log₂ ratios indicating SCNAs with boundaries in genomic regions within early- and late-replication timing zones, which overlap with known cancer genes listed in the Cancer Gene Census (SCNA_CGC). Supplementary Module 6 provides the contingency tables for SCNA amplifications and deletions as identified by GISTIC. The dashed line serves the same purpose as in d. (f) Distribution frequencies of SCNA boundaries near early-, mid- and late-replication transition zones, after dividing the transition zones into 10-kb nonoverlapping windows. (g) Distribution of frequencies of SCNA_CGC boundaries near early-, mid- and late-replication transition zones, after dividing the transition zones into 10-kb nonoverlapping windows. Replication transition is continuous. *, P < 2.21 × 10^–16; **, P = 1.59 × 10^–4; ***, P < 2.2 × 10^–16; ****, P = 1.42 × 10^–8.

Figure 3

Genome-wide distributions of long-range interactions and SCNAs. (a) Pairs of genomic regions that were at least 5 Mb apart on the same chromosome were classified into four categories depending on whether they share no HiC reads, ≥1, >5 or >10 HiC reads. We estimated the proportion of region pairs that also harbor the two boundaries of one or more SCNAs. (b) Same as a, but only considering those SCNAs that overlap with known cancer genes. Supplementary Module 7 provides the information for SCNA amplifications and deletions as identified by GISTIC. (c) We classified large SCNAs (>5 Mb, black), and those that overlap with known cancer genes (gray) into four groups depending on the number of HiC reads that link the two boundary regions of the SCNAs. (d) Distributions of SCNAs and long-range interactions around BCL6, a cancer gene on human chromosome 3. The HiC reads that link two boundary regions of SCNAs are shown in red. (e) Pie chart showing the proportion of ovarian cancer SCNAs whose boundaries were predicted using replication timing and HiC data alone.