Positively selected enhancer elements endow osteosarcoma cells with metastatic competence

Abstract

Metastasis results from a complex set of traits acquired by tumor cells, distinct from those necessary for tumorigenesis. Here, we investigate the contribution of enhancer elements to the metastatic phenotype of osteosarcoma. Through epigenomic profiling, we identify substantial differences in enhancer activity between primary and metastatic human tumors and between near isogenic pairs of highly lung metastatic and nonmetastatic osteosarcoma cell lines. We term these regions metastatic variant enhancer loci (Met-VELs). Met-VELs drive coordinated waves of gene expression during metastatic colonization of the lung. Met-VELs cluster nonrandomly in the genome, indicating that activity of these enhancers and expression of their associated gene targets are positively selected. As evidence of this causal association, osteosarcoma lung metastasis is inhibited by global interruptions of Met-VEL-associated gene expression via pharmacologic BET inhibition, by knockdown of AP-1 transcription factors that occupy Met-VELs, and by knockdown or functional inhibition of individual genes activated by Met-VELs, such as that encoding coagulation factor III/tissue factor (F3). We further show that genetic deletion of a single Met-VEL at the F3 locus blocks metastatic cell outgrowth in the lung. These findings indicate that Met-VELs and the genes they regulate play a functional role in metastasis and may be suitable targets for antimetastatic therapies.

Extended Data Figure 1. Met-VEL profiles of osteosarcoma patient lung metastases and human osteosarcoma cell lines

a, Aggregate plots showing H3K4me1 ChIP-seq and H3K27ac ChIP-seq signal +/− 3Kb from midpoints of gained and lost Met-VELs in paired patient lung metastases and primary tumors. b, Aggregate plots showing H3K4me1 ChIP-seq, H3K27ac ChIP-seq and DNase-seq signal +/− 3Kb from mid-points of gained and lost Met-VELs in metastatic/parental human osteosarcoma cell lines.

Extended Data Figure 2. Met-VEL clusters occur across metastatic cancers

a. UCSC browser view of H3K4me1 profiles in MG63.3 (metastatic) and MG63 (parental) cell lines illustrating an example of a gained (left) and lost (right) Met-VEL cluster. Met-VELs identified by black bars. 200kb Met-VEL clusters highlighted in gray. b. Genome-wide lost Met-VEL landscape for MG63.3 cell line. Rows represent scaled chromosomal coordinates. Peaks represent maximum gained Met-VEL counts in 200kb sliding windows. Predicted target genes for selected peaks are labeled. c. Gained and lost Met-VEL cluster counts in patient lung metastases/primary tumors and metastatic/parental cell line pairs. d. Percentage of total gained (top) and lost (bottom) Met-VELs within and outside of clusters in patient lung metastases/primary tumors and metastatic/parental cell line pairs.

Extended Data Figure 3. Assessment of Met-VEL associated gene expression during metastatic colonization of the lung

a. Schematic of experimental design for assessment of Met-VEL gene expression in parental and metastatic cell lines in ex vivo lung metastasis model. Image adapted from . b. Log2 quantile-normalized FPKM values for gained (left) and lost (right) Met-VEL and Met-VEL cluster genes in HOS/MNNG (top) and HOS/143B (bottom) cell line pairs. Asterisks indicate significant differences in FPKM distributions between parental and metastatic cell lines (* P<0.05; ** P<1E-3; *** P<1E-4). P-values calculated by Mann-Whitney Test. c. Heatmap of up-regulated gained Met-VEL genes in HOS/MNNG (top) and HOS/143B (bottom) cell line pairs. d. Expressed transcription factors with enriched motifs in lost Met-VELs in three metastatic/parental cell line pairs and corresponding motif enrichment p-values. e. Gene Ontology (GO) terms for lost Met-VEL genes in three metastatic/parental cell line pairs and corresponding p-values.

Extended Data Figure 4. Assessment of Met-VEL associated gene expression in vitro and in ex vivo lung culture

a. Log2 fold-change in quantile-normalized FPKM values of gained Met-VEL genes in metastatic cell lines versus parental cell lines in various conditions. P-values calculated by Mann-Whitney Test. b. Log2 fold-change in quantile-normalized FPKM values of lost Met-VEL genes in metastatic cell lines versus parental cell lines in various conditions. P-values calculated by Mann-Whitney Test. c. Violin plots of log2(fold-change quantile-normalized FPKM values) distributions for gene sets in three metastatic cell lines relative to parental non-metastatic lines. Fold-change values for all expressed genes and gained Met-VEL gene sets represent comparisons of expression in metastatic lines at day 14 in ex vivo lung culture to non-metastatic lines at 24hrs. Fold-change values for lost Met-VEL gene set represent comparison of expression in metastatic lines at 24hrs in ex vivo lung culture to non-metastatic lines at 24hrs. d. Bar charts indicating overlap between gained (left) and lost (right) Met-VEL gene sets and top 1000 up- and down-regulated genes in corresponding conditions, respectively.

Extended Data Figure 5. Assessment of enriched functions of gained Met-VEL genes in osteosarcoma patient metastases

Enriched Map representation of all Gene Ontology (GO) terms for gained Met-VEL genes calculated by aggregating gene lists from two patient metastases (Lung Met 4 and Lung Met 5).

Extended Data Figure 6. Analysis of anti-metastatic and gene expression effects of BET inhibition with JQ1

a. Kinetics of metastatic outgrowth of MNNG (top) and 143B (bottom) metastatic cell lines in ex vivo lung culture with 250nM JQ1 or vehicle (DMSO) treatment. Metastatic burden measured as total GFP+ area per lung section normalized to GFP+ area on day 0. Values represent averages of 8 lung sections (4 sections per mouse x 2 mice) +/−SEM. P-value calculated by Mann-Whitney Test. Dashed line indicates time points chosen for RNA-seq studies. b. Representative 2.5x images of vehicle (left) and 250nM JQ1 treated (right) lung sections at day 15. Lung sections outlined with dashed white line. Scale bar = 500μm. c. 20X image of hematoxylin and eosin stained section of lung slice after 15 days in ex vivo culture treated with DMSO (top) and 250nM JQ1 (bottom) illustrating viable lung cells and architecture. Scale bar = 100μm. d. GSEA plots of 2-fold up-regulated gained Met-VEL gene sets in vehicle versus JQ1-treated MNNG (left) and 143B (right) cells isolated from ex vivo lung culture. Cells isolated at time points indicated by dashed lines in Supplemental Figure 2–5a. e. Log2 fold-change expression 2-fold up-regulated gained Met-VEL gene sets in vehicle versus JQ1-treated MNNG (left) and 143B (right) cells isolated from ex vivo lung culture at time points indicated by dashed lines in Extended Data Figure 7a relative to parental cell line. f. Percent reduction in gene expression with 250nM JQ1 treatment of all genes up-regulated 2-fold in metastatic cell lines relative to parental cell lines at sorting time points, lost Met-VEL genes, gained Met-VEL genes 2-fold up-regulated relative to parental cell lines at sorting time points, up-regulated gained Met-VEL genes without SE genes, and all SE genes in MNNG and 143B cells growing in ex vivo lung culture sorted at time points indicated in Extended Data Figure 7a.

Extended Data Figure 7. Assessment of LT3REPIR shRNA construct induction and leakiness

a. Schematic of doxycycline-inducible LT3REPIR shRNA construct. Modified from (Fellmann et al., 2013). b. Cytometric analysis of GFP and DsRed expression in GFP+ MG63.3 cells transduced with LT3REPIR construct at baseline (left) and 40hrs after exposure to 5ug/ml doxycycline (right).

Extended Data Figure 8. Assessment of Tissue Factor (F3) dysregulation in metastatic osteosarcoma

a. IGV browser view of H3K27ac, H3K4me1, and DNase profiles at F3 gained Met-VEL cluster locus in MG63.3 (metastatic) and MG63 (parental) cell lines. Top of figure shows local contact profile analysis of F3 locus in MG63.3. In the top panel (main trend), the contact intensity (black line) is calculated by using a running mean analysis of normalized read counts with a 1kb sliding window. The 20th and 80th percentile are visualized as a gray trend graph. In the bottom panel, contact intensities are computed using linearly increasing sliding windows (scaled 2–50 kb) and are displayed as a color-coded heatmap of positive 4C signals (maximum of interaction set to 1). Local color changes are log-scaled to indicate changes of statistical enrichment of captured sequences, corresponding to the enhancer-promoter interaction. Areas of significant contact highlighted. b. Fold-change quantile normalized F3 FPKM values in 3 metastatic cell lines at 24hrs and 14 days of metastatic outgrowth in ex vivo lung culture relative to parental line at 24hrs. c. Tissue Factor (F3) relative expression in human patient primary tumors and lung metastases normalized to expression in normal osteoblasts. d. IGV browser view of H3K27ac, H3K4me1, and DNase profiles at F3 gained Met-VEL cluster locus in MG63.3, MNNG, and 143B metastatic cell lines.

Extended Data Figure 9. F3 expression in human osteosarcoma tumors

Immunohistochemical staining of F3 in human osteosarcoma lung metastases and primary tumors. Scale bars = 200μm.

Extended Data Figure 10. Assessment of Tissue Factor (F3) Met-VELs in patient lung metastases

a. IGV browser views of the F3 locus showing H3K27ac ChIP-seq tracks of lung metastases from 10 osteosarcoma patients. Blue bars below each track correspond to super enhancers defined using the ROSE script. Hi-C defined topologically associating domains (TADs) are displayed below. b. Volcano plot of 5571 total super enhancers detected in all osteosarcoma patient samples and cell lines used in this study. Points marked in red denote super enhancers meeting the threshold of significance (P < 0.05) for being gained or lost in metastatic samples. Points in grey denote super enhancers below the significance threshold. The F3 super enhancer is indicated by the arrow.

Extended Data Figure 11. Assessment of F3 knockdown by RT-qPCR

a, Amplification plots and standard curves of GAPDH RT-qPCR using 0.2ng, 2ng, 20ng, and 200ng of template cDNA demonstrating efficiency value of 100% and R² value of 1.000. All Cq values used for quantification of GAPDH were confirmed to be within linear range of standards. b, Amplification plots and standard curves of F3 RT-qPCR using 0.2ng, 2ng, 20ng, and 200ng of template cDNA demonstrating efficiency value of 97.2% and R² value of 0.999. All Cq values used for quantification of F3 were confirmed to be within linear range of standards. c, Relative GAPDH expression in DsRed+ (induced) cells transduced with F3 shRNA constructs after 40hrs treatment with 5ug/ml doxycycline relative to uninduced controls (average of 3 replicates +/−SEM). d, Relative F3 expression normalized to GAPDH in DsRed+ (induced) cells transduced with F3 shRNA constructs after 40hrs treatment with 5ug/ml doxycycline relative to uninduced controls (average of 3 replicates +/−SEM).

Extended Data Figure 12. Assessment of F3 knockdown on metastatic osteosarcoma cell in vitro growth and lung colonization

a. Change in confluence relative to day 0 of MG63.3 (top) and MNNG (bottom) cells transduced with F3 shRNA constructs grown in standard culture conditions over 108hrs in the presence or absence of 5ug/ml doxycycline. Values represent averages from 6 plates +/− SD. b. Representative 2.5x images of from day 21 of ex vivo lung culture sections of GFP+ MG63.3 (left) and MNNG (right) cells transduced with F3 shRNA constructs untreated (top) or treated with 5ug/ml doxycycline (bottom). Lung sections outlined with dashed white line. Scale bar = 500μm. c. Quantification of metastatic burden at day 21 of ex vivo lung culture sections of GFP+ MG63.3 (left) and MNNG (right) cells transduced with F3 shRNA constructs untreated (red) or treated (blue) with 5ug/ml doxycycline. Values represent averages +/−SEM from 8 sections per condition (4 sections per mouse x 2 mice) normalized to the same section at day 0. P-Value calculated by Mann-Whitney Test. d. Representative 2.5x images of in vivo metastatic burden in untreated (left) or doxycycline-treated (right) mice receiving tail vein injection of 5×10⁵ GFP+ MG63.3 cells transduced with shF3B construct. Scale bar = 500μm. e. Quantification of in vivo metastatic burden in untreated (red) or doxycycline-treated (blue) mice receiving tail vein injection of 5×10⁵ GFP+ MG63.3 cells transduced with shF3B construct. Values represent log2 of total GFP+ area per 2.5x field, black bars represent average +/− SEM (N= 5 mice per condition, 5 images per mouse). P-Value calculated by Mann-Whitney Test.

Extended Data Figure 13. Assessment of lung metastatic burden at experimental end point of orthotopic spontaneous metastasis experiment with F3 knockdown

a. All images used for quantification of in vivo metastatic lesions in lungs 21 days after measureable tumor formation in untreated mice receiving orthotopic injection of 8×10⁵ MG63.3 cells transduced with shF3B construct (5 mice per condition, 5 images per mouse). Scale bar = 250μm. b. All images used for quantification of in vivo metastatic lesions in lungs 21 days after measureable tumor formation in doxycycline-treated mice receiving orthotopic injection of 8×10⁵ MG63.3 cells transduced with shF3B construct (5 mice per condition, 5 images per mouse). Scale bar = 250μm.

Figure 1. H3K4me1 ChIP-seq identifies metastatic variant enhancer loci (Met-VELs) and Met-VEL clusters

a, Schematic representation of human tumor and metastatic human osteosarcoma cell line cohort. b, UCSC browser views of H3K4me1 profiles from MG63.3 (metastatic) and MG63 (parental) cell lines illustrating an example of gained (top) and lost (bottom) Met-VEL(s). Met-VELs are boxed in red. c, Heatmap showing H3K4me1 ChIP-seq signal +/−5kb from H3K4me1 peak midpoints for all putative enhancers in MG63.3/MG63 pair sorted by differences in signal. Sub-panel shows heatmap for gained and lost Met-VELs alone. d, Aggregate plots showing H3K4me1 ChIP-seq and H3K27ac ChIP-seq signal +/− 3kb from mid-points of gained (left) and lost (right) Met-VELs for a representative matched primary/lung metastatic human tumor pair (top) and MG63.3/MG63 cell line pair (bottom). DNase-seq signal +/− 3kb from Met-VEL mid-points is also shown for the MG63.3 and MG63 cell lines. e. Percentage of enhancers gained and lost in metastatic samples relative to primary tumors or non-metastatic cell lines. f. Heatmap of fold change normalized RPKM in metastatic samples vs. primary tumor or non-metastatic cell lines for aggregated list of all gained and lost Met-VELs across all samples. H3K4me1 signal shown on the left, H3K27ac signal shown on the right. The samples from left to right are as follows M112, 143B, MNNG, MG63.3, LM7, Lung Met 1, Lung Met 2, Lung Met 3, Lung Met 4, Lung Met 5. Asterisks indicate 143B and MNNG samples. g. Genome-wide gained Met-VEL landscape for human osteosarcoma metastatic tumor (Lung Met 4) and MG63.3 cell line. Rows represent scaled chromosomal coordinates. Peaks represent maximum gained Met-VEL counts in 200kb sliding windows. Predicted target genes for selected peaks are labeled.

Figure 2. Met-VELs modulate gene expression during metastatic colonization of the lung

a. Schematic of experimental design for assessment of Met-VEL gene expression in parental and metastatic cell lines in ex vivo lung metastasis model. Image adapted from . b. Log2 quantile-normalized FPKM values for gained (left) and lost (right) Met-VEL and Met-VEL cluster genes in MG63/MG63.3 cell line pair. Asterisks indicate significant differences in FPKM distributions between parental and metastatic cell lines (* P<0.05; ** P<1E-3; *** P<1E-4). P-values calculated by Mann-Whitney Test. c. Heatmap of up-regulated gained Met-VEL genes in MG63/MG63.3 cell lines illustrating phasic expression pattern. d. Enriched Map representation of all Gene Ontology (GO) terms for three classes of gained Met-VEL responder genes calculated by aggregating gene lists from all three cell line pairs. e. GSEA plot of up-regulated gained Met-VEL gene set compiled from three metastatic cell lines in human patient lung metastases versus primary tumors. f. Expressed transcription factors with enriched motifs in gained Met-VELs in three metastatic/parental cell line pairs and corresponding motif enrichment p-values. g. Aggregate plot of H3K4me1, FOS, and FOSL1 ChiP-seq signal at all gained Met-VELs in MG63.3 cell line.

Figure 3. In Vivo high-throughput RNAi functional assay of candidate metastasis dependency genes

a. Schematic of doxycycline-inducible LT3REPIR shRNA construct. Modified from . b. Schematic of experimental design for in vivo high-throughput functional assay of candidate metastasis dependency genes. c. Volcano plot of relative abundance of shRNAs targeting 33 genes in GFP+/DsRed+ sorted osteosarcoma cells from doxycycline-treated mice (N=15 mice per replicate x 3 replicates) versus input cell population. 2^nd most depleted shRNAs for each gene are plotted as well as negative and positive shRNA controls. Negative controls contained groups of 2–4 shRNAs. d. Volcano plot of relative abundance of shRNAs targeting 13 genes meeting initial hit criteria (Fig. 3c) in GFP+/DsRed+ sorted osteosarcoma cells from doxycycline-treated mice (N=5 per replicate x 3 replicates) versus GFP+/DsRed+ sorted osteosarcoma cells treated with doxycycline in vitro. 2^nd most depleted shRNAs for each gene are plotted as well as negative controls groups of 2–4 shRNAs.

Figure 4. Tissue Factor (F3) mediates lung metastasis of osteosarcoma

a. In Vivo 2.5x images of GFP+ MG63 (parental) and MG63.3 (metastatic) cells in the lung 24hrs following tail-vein injection of 1×10⁶ cells. Sections stained for GFP, tissue factor (F3, red), and DAPI. Arrowheads indicate individual tumor cells within lung. Scale bars = 20μm. b. Quantification of mean red pixel intensity within GFP+ (tumor) area in MG63 (parental) and MG63.3 (metastatic) cells 24hrs after tail-vein injection (N=15 images per condition). c. Representative images of immunohistochemical staining of F3 in human osteosarcoma lung metastases (M), primary tumor (P), and omission control (OC). Tissue microarray contained 18 scoreable lung metastases of similar quality to those displayed. d. Percentage of lung metastases with various levels of F3 positivity. e. Kaplan-Meier plot of untreated (red) and doxycycline-treated (blue) mice tail-vein injected with 5×10⁴ GFP+ MG63.3 cells transduced with shF3B construct (N=5 mice per condition). P-value calculated by Gehan-Breslow-Wilcoxon Test. f. Primary tumor growth in untreated (red) and doxycycline-treated (blue) mice receiving orthotopic injection of 8×10⁵ GFP+ MG63.3 cells transduced with shF3B construct. Values represent averages +/−SEM (N=5 mice per condition). P-values calculated using student’s t-test. g. Representative 2.5x images of in vivo metastatic lesions in lungs 21 days after measureable tumor formation in untreated (left) and doxycycline-treated (right) mice receiving orthotopic injection of 8×10⁵ GFP+ MG63.3 cells transduced with shF3B construct. Arrowheads indicate individual tumor cells within lung. Scale bar = 500μm. h. Quantification of lung metastatic burden 21 days after measureable tumor formation in untreated (red) and doxycycline-treated (blue) mice receiving orthotopic injection of 8×10⁵ GFP+ MG63.3 cells transduced with shF3B construct (N=5 mice per condition, 5 images per mouse). P-values calculated using Mann-Whitney test. i. Amount of activated factor X (FXa) formed in in vitro assay by MG63 (left) and MG63.3 (right) cells treated with 25μg/mL IgG control, 10H10, or 5G9 antibodies for 20 minutes prior to the addition of FVIIa and FX to a final concentration of 100nM. FXa formation assessed 30 minutes after adding FX. P-values calculated using student’s t test with Welch’s correction. j. Representative 2.5x images of in vivo metastatic lesions in lungs of mice 14 days after tail vein injection of 5×10⁵ MG63.3 cells with 500μg of IgG, 5G9, or 10H10 antibodies. Scale bar = 500μm. k. Quantification of lung metastatic burden 14 days after tail vein injection of 5×10⁵ MG63.3 cells with 500μg of IgG, 5G9, or 10H10 antibodies (N= at least 5 mice per condition, 5 images per mouse). P-values calculated using Mann-Whitney test.

Figure 5. Deletion of single gained Met-VEL blunts F3 expression and mitigates lung metastasis of osteosarcoma cells

a. IGV browser view of region targeted for deletion with TALENs. Schematic shows strategy for 4,211bp deletion. Sanger sequencing shows resulting clonal homozygous deletion. b. Representative 40x images of WT and Met-VEL deleted MG63.3 cells in the lung 24hrs after initiation of an ex vivo lung metastasis experiment. Scale bar = 50μm. c. Quantification of mean red pixel intensity (F3 expression) within GFP+ (tumor) area for WT and Met-VEL deleted MG63.3 cells 24hrs after initiation of ex vivo lung metastasis assay (N= at least 20 images per condition, sections taken from N=3–5 mice per condition). P-value calculated by Mann-Whitney Test. d. Representative 2.5x images of WT (left) and Met-VEL deleted (right) lung sections at day 5. Lung sections outlined with dashed white line. Scale bar = 500μm. e. Quantification of metastatic burden at day 5 of ex vivo lung culture for GFP+ WT and Met-VEL deleted MG63.3 cells. Bars represent mean +/−SEM from at least 8 sections per condition (4 sections per mouse x 2–3 mice) normalized to the same section at day 0. P-Value calculated by Mann-Whitney Test.