Using DNA methylation patterns to infer tumor ancestry

Abstract

Background: Exactly how human tumors grow is uncertain because serial observations are impractical. One approach to reconstruct the histories of individual human cancers is to analyze the current genomic variation between its cells. The greater the variations, on average, the greater the time since the last clonal evolution cycle ("a molecular clock hypothesis"). Here we analyze passenger DNA methylation patterns from opposite sides of 12 primary human colorectal cancers (CRCs) to evaluate whether the variation (pairwise distances between epialleles) is consistent with a single clonal expansion after transformation.

Methodology/principal findings: Data from 12 primary CRCs are compared to epigenomic data simulated under a single clonal expansion for a variety of possible growth scenarios. We find that for many different growth rates, a single clonal expansion can explain the population variation in 11 out of 12 CRCs. In eight CRCs, the cells from different glands are all equally distantly related, and cells sampled from the same tumor half appear no more closely related than cells sampled from opposite tumor halves. In these tumors, growth appears consistent with a single "symmetric" clonal expansion. In three CRCs, the variation in epigenetic distances was different between sides, but this asymmetry could be explained by a single clonal expansion with one region of a tumor having undergone more cell division than the other. The variation in one CRC was complex and inconsistent with a simple single clonal expansion.

Conclusions: Rather than a series of clonal expansion after transformation, these results suggest that the epigenetic variation of present-day cancer cells in primary CRCs can almost always be explained by a single clonal expansion.

Figure 1. A picture of a colorectal adenocarcinoma with a cartoon illustrating three possible growth models: rapid growth, gradual growth, asymmetric growth.

Figure 2. Tumor sampling scheme with bisulfite sequence data generated from one colorectal tumor.

Grey circles denote cancer glands. Six to seven glands are sampled from each tumor half and multiple clones are sequenced for each tumor gland (e.g. 7–8). Black circles denote methylated CpGs and white circles unmethylated CpGs. Three physical distances: within a cancer gland, between cancer glands from the same tumor half, and between cancer glands from opposite tumor halves.

Figure 3. The number of glands in a tumor as a function of the number of generations of cell division for Gompertzian growth models with inflection times of 49 and 320 generations and a maximum of 524,288 glands.

Figure 4. Average Hamming distance by tumor half for the 12 tumors published by Siegmund et al.

The x-axis represents the distance in the right tumor half and the y-axis the distance in the left tumor half. Within-gland distances are denoted by G and within-half by H. Lines connect the pairs (G,H) from the same tumor. The line type indicates differences in distance between-half and within-half (BH-WH). Solid lines denote small BH-WH differences (difference <0.1, 8 “flat” tumors) and broken lines denote larger differences (BH-WH difference >0.4, 4 tumors). The type of broken line indicates WH differences between left and right tumor half; dotted lines indicate tumors with large WH differences (>0.6, 3 tumors); the dashed line indicates one tumor with small WH difference (0.14).

Figure 5. The relationship between physical proximity in the tumor and epigenetic distance.

(a) The relationship between the fraction of tumor sampled and time to the most recent common ancestor. (b) Mean estimates of average Hamming distance for three physical distances: between tumor half (BH), within tumor half (WH) and within gland (WG) under different growth curve inflection times (inflection time is 49 for figure on left and 320 for figure on right) (N = 10,000 replicates).

Figure 6. Mean estimate of average Hamming distance between-half (BH), within-half (WH), and within gland (WG) from a simulation study (N = 10,000 replicates).

Growth curve inflection time is 49 generations of cell division for fast growing tumors and 320 generations for slow-growing tumors.

Figure 7. Schematic of sample genealogy.

(a) Example genealogy. Black circles denote cancer stem cells and grey circles denote cancer glands. (b) Two models for cell division. Black circles denote CSCs and white circles denote non-CSCs.