Evolution of metastasis revealed by mutational landscapes of chemically induced skin cancers

Abstract

Human tumors show a high level of genetic heterogeneity, but the processes that influence the timing and route of metastatic dissemination of the subclones are unknown. Here we have used whole-exome sequencing of 103 matched benign, malignant and metastatic skin tumors from genetically heterogeneous mice to demonstrate that most metastases disseminate synchronously from the primary tumor, supporting parallel rather than linear evolution as the predominant model of metastasis. Shared mutations between primary carcinomas and their matched metastases have the distinct A-to-T signature of the initiating carcinogen dimethylbenzanthracene, but non-shared mutations are primarily G-to-T, a signature associated with oxidative stress. The existence of carcinomas that either did or did not metastasize in the same host animal suggests that there are tumor-intrinsic factors that influence metastatic seeding. We also demonstrate the importance of germline polymorphisms in determining allele-specific mutations, and we identify somatic genetic alterations that are specifically related to initiation of carcinogenesis by Hras or Kras mutations. Mouse tumors that mimic the genetic heterogeneity of human cancers can aid our understanding of the clonal evolution of metastasis and provide a realistic model for the testing of novel therapies.

Figure 1

Chemically induced tumors carry a mutation signature of the carcinogen DMBA. (a) Scheme for the generation of genetically unique backcrossed (FVBBX) mice and of metastases. FVB/N Hras^−/−mice were crossed with SPRET/EiJ mice and the resulting offspring were crossed to inbred wild-type (WT), Hras^+/− or Hras^−/− FVB/N mice to generate a backcross population with all three Hras genotypes. Tumors were induced in FVBBX mice, carcinomas were resected upon reaching >1 cm in the longest diameter, and mice were allowed to progress to metastatic disease. (b) Frequencies of mutations observed in each of 96 possible trinucleotide mutation contexts for all of the mutations across all tumors. Trinucleotide contexts, arranged on the x axis, are grouped by the base pair change of the mutation. Peaks are observed at two specific contexts, CTG to CAG and GTG to GAG. The same pattern is also observed when only nonsynonymous mutations are considered (data not shown). (c) Frequencies of ‘early mutations’ (left graphs) or ‘late mutations’ (right graphs) in metastases, as classified by two strategies. Top, mutations were classified by whether they are shared with the ancestral primary tumor (early, upper left) or not shared (late, upper right). Bottom, mutations in tumors of all stages were classified on the basis of whether they were fully clonal (i.e., present in all tumor cells; early; lower left) or subclonal (i.e., present only in a fraction of tumor cells; late; lower right). Early mutations show an enrichment of A>T mutations (equivalent to T>A mutations on the opposite strand), and late mutations show a higher proportion of G>T mutations (equivalent to C>A mutations on the opposite strand).

Figure 2

Phylogenetic trees reveal evolutionary relationship between tumors. (a) In mouse 1664, carcinoma A had four metastases (met) that diverged early (sharing an average of 46% of mutations; 83 mutations), whereas carcinoma B did not form any detectable metastases. Distant metastases to the lung and thymus did not show evidence of disseminating from a lymph node, but rather appeared to diverge from the primary carcinoma at approximately the same time as the lymph node metastases. The asterisk (*) denotes normal tissue (root of tree). Pap 1 and Pap 2 are independently arising papillomas, and LR is a lower right lymph node metastasis (under the rear right limb). (b) In mouse 1383, carcinoma A had four metastases that diverged relatively late and approximately synchronously, and which shared an average of 88% of their mutations with the primary carcinoma. UL and LL designate lymph node metastases under the upper (front) left and lower (rear) left limbs, respectively. (c) In mouse 1407, two lymph node metastases departed synchronously from carcinoma A and a spleen metastasis diverged slightly earlier than the lymph node metastases. The primary carcinoma and its metastases shared 153 mutations, corresponding to an average of 69% of their total mutations. UR designates a lymph node metastasis at the upper (front) right limb. (d) In mouse 1984, a lymph node metastasis and a lung metastasis diverged from carcinoma B synchronously, sharing 160 mutations with the primary carcinoma. (e) In mouse 2104, a lymph node metastasis and a dorsal metastasis shared 118 mutations with carcinoma B, indicating that the metastases diverged synchronously. (f) In mouse 1949, two lymph node metastases and a lung metastasis shared 167 mutations with carcinoma A (average of 52% of total mutations) and an additional 90 mutations with one another, suggesting that one of the metastases may have given rise to the others. This mouse provides the only counter-example to parallel evolution that we observed. In all panels, scale bars represent 100 SNVs.

Figure 3

Histological analysis of primary tumors and metastases from mouse 1984. (a–c) H&E-stained samples of primary carcinoma B showing spindle tumor cells arranged in fascicles (a), lung metastasis arising from carcinoma B showing squamous histology (b) and lymph node metastasis arising from carcinoma B showing spindle histology (c). Images are shown at 200× magnification. Scale bars, 50 μm.

Figure 4

Mutated genes in carcinogen-induced tumors. Nonsynonymous (green bars) and stop-gain (black bars) mutations in genes observed to be mutated in human cancer. Tumors are arranged on the horizontal axis by mouse genotype (wild type or Hras^−/−). Independent mutations are defined as mutations in tumors that show no phylogenetic relationship (on the basis of shared mutations). Number of mutations observed in early papillomas is also shown.

Figure 5

Gene CNVs in mouse skin tumors. (a) CNVs are shown on each chromosome, with Ras genotype and tumor morphology shown on the y axis. Chromosome numbers are on x axis, and chromosomes are arranged with proximal end on the left and distal ends on the right. The red and blue scale represents copy number gains and losses, respectively. (b) Focal Cdkn2a deletions are visible in many tumors, frequently less than 1 Mb in size. The vertical dashed lines indicate the boundaries of the Cdkn2a gene, and the blue horizontal lines show the extent of the deletion. (c) Cdkn2a losses and Met amplifications in samples displayed by tumor type. Co-occurrence of these events increases substantially with tumor progression, from papillomas (7% co-occurrence) to SCCs (25% co-occurrence) to spindle tumors (59% co-occurrence). Each box represents one tumor, and tumors with shared CNVs are arranged vertically. Pap., papilloma; Spind., spindle morphology; M, mixed SCC and spindle morphology.