Genetic alterations driving metastatic colony formation are acquired outside of the primary tumour in melanoma

Abstract

Mouse models indicate that metastatic dissemination occurs extremely early; however, the timing in human cancers is unknown. We therefore determined the time point of metastatic seeding relative to tumour thickness and genomic alterations in melanoma. Here, we find that lymphatic dissemination occurs shortly after dermal invasion of the primary lesion at a median thickness of ~0.5 mm and that typical driver changes, including BRAF mutation and gained or lost regions comprising genes like MET or CDKNA2, are acquired within the lymph node at the time of colony formation. These changes define a colonisation signature that was linked to xenograft formation in immunodeficient mice and death from melanoma. Thus, melanoma cells leave primary tumours early and evolve at different sites in parallel. We propose a model of metastatic melanoma dormancy, evolution and colonisation that will inform direct monitoring of adjuvant therapy targets.

Fig. 1

Dissemination of melanoma cells as function of tumour thickness. a Yellow function: estimated cumulative probability of dissemination as a function of tumour thickness (Turnbull) (n = 1027 patients). Blue line: Standard log-logistic distribution incorporating a fraction of patients without long-term dissemination (95% CI: blue dashed lines). Upper black dashed line: only 65.2% of melanomas disseminate lymphatically (95% CI: 60.4–70.0%). Fifty percent of this value (32.6%) provides the median thickness (0.5 mm, 95% CI: 0.3–0.7 mm) of disseminating melanomas (lower black dashed line). b Comparative analysis of histopathological and immuncytological SLN halves. Displayed are representative examples with immunocytological scores of DCCD ≤ 100, 100 < DCCD ≤ 1000 and DCCD > 1000. Samples LN 72 and LN 89 are stained against melan A LN 10, LN 135 and LN 168 against S100. LN 154 shows a highly pigmented melanoma in H&E staining. A close-up of the subcapsular region 3 is shown in the main figure; more central or core regions 1 and 2, see Supplementary Fig. 4a for higher magnification. c Evaluation of histopathological findings in corresponding sections of samples with a positive DCCD < 100. d Percentage of DCC-positive patients (n = 525) with colonisation (DCCD > 100) according to the Turnbull estimate (yellow). The percentage of colonisation (blue curve, 95% CI blue dashed curves) is described by a cumulative exponential distribution function (median 10.3 mm; 95% CI: 8.4–13.0 mm). e Hazard functions for dissemination (yellow line), and colonisation (blue line) describing the instantaneous risk per unit thickness for an event (dissemination, n = 1027; colonisation, n = 525) for those tumours, for which it has not yet occurred (de novo dissemination/ colonisation). f Survival analysis of melanoma patients (n = 1027) according to T stage (T1: ≤ 1 mm (n = 83); T2: 1.01–2.0 mm (n = 496); T3: 2.01–4.0 mm (n = 315); T4: >4 mm thickness (n = 133); log-rank test, p < 0.001)

Fig. 2

gp100-positive cells from SLNs and primary tumours display multiple copy number alterations (CNAs). a Ninety-one DCCs from our patients were selected according to QC criteria (see main text and Supplementary Figure 1) for CGH and mutation analysis. Histograms depict the genomic gains (yellow) or losses (blue) per cell. The identifiers indicate cell ID. b Cluster analysis of paired PTs (n = 23 samples) and DCCs (n = 24 samples) from 19 patients for chromosomal aberrations (gain = +1; loss = −1). Only the 10 most variable regions are included (see Supplementary Figure 5 for all 53 variable regions). Side-line identifiers indicate patient ID, sample type (PT and DCC) and sample index; bottom-line labels indicate chromosomal regions. Black (28T), white (30T) and dark grey (09T) filled squares indicate examples of PT-DCC pairs for which several areas of the PT were available. Examples of paired DCCs and PTs with varying PT thickness are indicated by coloured squares (green: 1 mm, grey: 1.4 mm, pink: 2 mm, blue: 7 mm PT thickness). c Comparison of paired PTs and DCCs for chromosomal aberrations. Displayed are the 18 chromosomal regions that differ significantly (FDR-adjusted Fisher’s exact test p value ≤0.05) between paired PTs (n = 23, 19 patients) and DCCs (n = 24, n = 19 patients) regarding aberration frequency. Gains and losses are given in percent. d Maximum number of DCC samples N_DCC (colour-coded) that cannot be excluded to derive from corresponding DCC-like clones within the PT plotted against the assumed DCC-like clone fraction in the PT and the successful dissemination fold change. If DCC-like and non-DCC-like PT clones disseminate equally (fold = 1), at most 11/24 DCC samples may result from DCC clones within the PT at the maximum indicated clone fraction of 0.40 (A); for twofold dissemination not more than 14/24 (B). For fold = 5, the same number of samples can be excluded up to a fraction of 0.124 (11/24) (C) and 0.21 (14/24) (D), respectively. These intermediate fractions correspond to a maximum of 5/24 (0.124) (E) and 7/24 (0.21) (F) DCC samples, respectively, if clones disseminate similarly (fold = 1). See 'Methods' section for details. Significance level α = 0.05 (Poisson-binomial distribution)

Fig. 3

Phylogenetic trees of paired PT and DCC samples. Representative examples for the three distinct groups observed in our data (for all 19 patients, see Supplementary Figures 6–8). The ordinate of the tree panels indicates the number of aberrations per CGH profile (square root scale). Heat maps show aberration profiles along paths from normal cells (N) to PT (P1–4) and DCC (D1, D3) samples, respectively. Profiles A1, A2 and A5 denote inferred common ancestors (intermediates) (see 'Methods' section)

Fig. 4

BRAF and NRAS mutations in PTs and DCCs. a Single cell WGA reliably captures wild-type and mutated alleles. Exon 15 mutation c1799T>A (BRAF) and Exon 2 mutation c181C>A (NRAS) were detected in all single cells (lanes 1–9 or 1–15) of cell lines with BRAF (cell lines 70–61 (n = 9) and MelHo (n = 9)) or NRAS (cell line 102–4; n = 15) mutation. The allelic ratio of wt vs. mut alleles of each cell line was determined using pooled DNA. Note that this ratio is preserved in most single cells. b Summary of results in a and detection of BRAF and NRAS mutation in single gp100+ cells isolated from (i) lymph nodes from healthy controls spiked with melanoma cell line cells, processed and stained as SLN of melanoma patients; (ii) an enzymatically digested DCC-xenograft derived from NRAS-mutated DCCs and (iii) primary melanomas with BRAF mutation. c Mutation analysis of BRAF and NRAS for paired PT-DCC samples (n = 32 patients). Different mutations (either NRAS or BRAF) are indicated by mut1 and mut2. Fisher’s exact test p values indicate differences in BRAF mutational status between PTs and DCCs. d Percentage of patients (n = see Table 1) with homogeneous (all cells harbouring the mutation) and heterogeneous BRAF/NRAS mutational status among DCCs. DCCs were detected using two markers, gp100 (n = 43 cells) or MCSP (n = 61 cells)

Fig. 5

Molecular analysis of DCCs at dissemination. a Left: Separation of primary melanomas according to median thickness into thin (<1.8 mm) and thick (≥1.8 mm) tumours, from which DCCs were analysed. Right: Number of chromosomal aberrations per DCC from thin (n = 17) and thick (n = 17) tumours . b Samples were selected for suggested early arrival in the SLN as indicated by small DCCD values up to 105 and then tested for genomic differences regarding their origin from thick and thin primaries. The y axis specifies the DCCD limit up to which samples were included in the analysis, with sample numbers ranging from 6 (DCCD = 1) over 31 (DCCD = 24) to 46 (DCCD = 105). The x axis reports the limiting thickness according to which samples were split regarding their association with thin or thick tumours (set points left of these values). Colour indicates the minimum FDR-adjusted p values (Fisher’s exact test) across loci/mutations assessing the largest genomic differences between these two groups. The smallest p value of FDR = 0.084 occurs at DCCD = 24 and thickness = 1.6 mm for locus 18q21–23 (black central dot). p values up to FDR = 0.2 are indicated by grey dots and may refer to the loci 18q21–23, 1q24–44 or 8q23–24.3. The colour key includes the non-log-transformed FDR values 0.05, 0.1 and 0.2 in brackets. No significant differences were observed

Fig. 6

DCCs acquire genetic alterations within and outside the PT. a, d, g, j Genome alterations are explained by four different risk scenarios. Panels a, d, g and j show different risk functions E (colour-coded) as a function of the number of cell divisions in the SLN (n_L) and the PT (N_T). The two arrows indicate a prototypic state trajectory: initially, cell divisions exclusively occur within the PT (N_T > 0, n_L = 0) until after n_T cell divisions a tumour cell disseminates to the SLN. Subsequently, PT and SLN cells grow simultaneously (N_T > 0, n_L > 0). The grey dashed horizontal lines at N_T = 14.4 and 26.6 correspond to the minimum (0.6 mm) and maximum (10 mm) experimental thickness values ('Methods' section). Growth rates are presumed equal in both environments implying (i) inaccessibility of the region below the diagonal (grey solid line), consistent with the empty lower right triangular area in c, f, i, l and (ii) N_T = n_T + n_L (after dissemination; 'Methods' section). The angle indicates the direction of the normal vector regarding equal risk lines. b, e, h, k Classification results for four prototypic CGH results (loci/mutations) using linear classifiers with 2D set points according to the indicated DCCD and PT thickness values and directions as pictured by the small arrows within each rectangle of the colour matrix ('Methods' section). Colour encodes the negative decadic logarithm of the FDR-corrected p values of each classification (Fisher’s exact test). Basically, the red/orange areas separate regions differing in their distribution of genomic alterations. Their orientation parallels the equal risk lines in a, d, g, j. DCCD and thickness values were chosen according to experiment while classifier set points were slightly displaced relative to these values (white points included for the 5% most significant FDR values). Minimum FDR values are indicated by black points. N = 87 DCCs, 57 patients, 82 loci/mutations. c, f, i, l Corresponding measurement results as well as the class assignments of the best (lowest FDR) classifiers indicated by the light (class1) and dark (class2) grey areas. The class boundary does not necessarily appear linear as samples are listed according to rank while linear classification was performed in log(DCCD)-log(thickness) space. The classifier angle and the distribution of deletions (−1), balances/wt (0) and amplifications/mutations (+1) in each class are given in the title

Fig. 7

DCC genetic alterations acquired in the SLN correspond to colonisation. a Distribution of best classifier angles for different loci/mutations. These can be grouped according to the prototypic categories in a, d, g and j. Vertical bars correspond to the range of angles associated with the same minimal FDR value, diamonds indicate angles that are closest to the ideal prototypes. Only the 62/82 loci/mutions with at least 10 samples per class were included. b DCCD-dependence of genomic changes derived using a vertical classifier (angle = 0°) positioned according to the experimentally measured values (set points left of these values). Only category 3 and 4 loci whose FDR values fall below 0.05 are shown. Numbers in brackets following the legend (e.g. BRAF mut (95)) indicate peak DCCD. c Proliferation of DCCs in sentinel nodes. Immunofluorescence of Melan A⁺ cells in G0-, G1-, G2-phase and mitosis (from left to right). Nucleus (blue), Melan A (red), Ki-67 (green; for n, see figure; Fisher’s exact test, p = 0.0005 for G0 vs. non-G0)

Fig. 8

Tumour-forming ability of DCCs before and after colonisation and patient survival. a Left to right: Isolated MCSP⁺ DCCs from a patient-SLN; DCC-derived sphere; H&E staining of a patient DCC-derived xenograft (DCC-PDX); DCC-PDX (7 s.c. injected DCCs). Pictures are representative for 15 transplanted patient samples. b Side-by-side transplantation of paired MCSP⁺ DCCs (n = 3 patients, 20 injection sites) and DCC-derived spheres (n = 3 patients, 15 injection sites) from the same patient into NSG mice. Left: Kaplan–Meier analysis of tumour-free mice (p < 0.0001, log-rank test). Right: number of injected MCSP⁺ DCCs and DCC-derived spheres per injection site. Black filled circles indicate tumour formation. c Number of MCSP⁺ DCCs (DCCD > 100: n = 25 injection sites) or DCC-derived spheres (DCCD > 100: n = 14 injection sites; DCCD ≤ 100: n = 14 injection sites) that were transplanted into NSG mice. Each circle represents one injection site. Black filled circles indicate tumour formation (engraftment). The p value (p = 0.03 Fisher’s exact test) indicates a significant difference in the engraftment rate for samples with DCCD > 100 (cells and spheres were pooled, n = 39, n = 10 patients) vs. DCCD ≤ 100 (spheres, n = 14, n = 5 patients). d Mutational status of patient-derived DCCs and their respective xenografts regarding the colonisation signature from Fig. 7a and Supplementary Table 3 (i.e., loss of genetic locus 9p21–24 and gain of 7q21–36, BRAFmut and NRASmut), n = 6 patients. Note that samples 125 and 277 add NRAS mutations to the colonisation signature. e Left: Kaplan–Meier survival analysis of patients with DCCs that display at least one of the final colonisation signature features (i.e., loss of genetic locus 9p21–24, gain of 7q21–36, BRAFmut or NRASmut, n = 39) or not (wt regarding colonisation signature, n = 22). Right: Kaplan–Meier survival analysis of patients with DCCs that display BRAF mutation (BRAFmut, n = 15) or wild type sequence (wt, n = 46)

Fig. 9

Model of melanoma progression from local to metastatic disease. Histologic appearance, patient-derived dissemination estimate, proliferation rate and BRAF mutational state are integrated into the scheme. Data are taken from this study and from references^{, ,}