Small-molecule targeting of brachyury transcription factor addiction in chordoma

Abstract

Chordoma is a primary bone cancer with no approved therapy¹. The identification of therapeutic targets in this disease has been challenging due to the infrequent occurrence of clinically actionable somatic mutations in chordoma tumors^2,3. Here we describe the discovery of therapeutically targetable chordoma dependencies via genome-scale CRISPR-Cas9 screening and focused small-molecule sensitivity profiling. These systematic approaches reveal that the developmental transcription factor T (brachyury; TBXT) is the top selectively essential gene in chordoma, and that transcriptional cyclin-dependent kinase (CDK) inhibitors targeting CDK7/12/13 and CDK9 potently suppress chordoma cell proliferation. In other cancer types, transcriptional CDK inhibitors have been observed to downregulate highly expressed, enhancer-associated oncogenic transcription factors^4,5. In chordoma, we find that T is associated with a 1.5-Mb region containing 'super-enhancers' and is the most highly expressed super-enhancer-associated transcription factor. Notably, transcriptional CDK inhibition leads to preferential and concentration-dependent downregulation of cellular brachyury protein levels in all models tested. In vivo, CDK7/12/13-inhibitor treatment substantially reduces tumor growth. Together, these data demonstrate small-molecule targeting of brachyury transcription factor addiction in chordoma, identify a mechanism of T gene regulation that underlies this therapeutic strategy, and provide a blueprint for applying systematic genetic and chemical screening approaches to discover vulnerabilities in genomically quiet cancers.

Extended Data Figure 1. Dependency scores for sgRNAs targeting commonly essential genes.

Viability after sgRNA treatment (represented by sgRNA-dependency scores; see Methods) corresponding to each of the primary screening sgRNAs targeting either RPS11 or RPS19 across 127 cancer cell lines (CCLs), including 2 chordoma CCLs (blue circles). For a given sgRNA, the median sgRNA-dependency score across 127 cell lines is indicated (gray line). Lower values indicate greater sgRNA depletion and thus essentiality of the target gene (shaded region).

Extended Data Figure 2. Sensitivity of chordoma cells to EGFR and/or ERBB2 inhibitors and of non-chordoma cells to CDK9 inhibitors.

a) Validation of primary screening hit compounds targeting EGFR and/or ERBB2. Four chordoma cell lines were treated with indicated concentrations of candidate antiproliferative compounds and assayed for cell viability after 6 d with CellTiter-Glo. Response data are represented by a fitted curve to the mean fractional viability at each concentration relative to vehicle-treated cells; error bars represent the SEM (n = 4 biological samples measured in parallel). b) Dinaciclib and alvocidib have antiproliferative effects across a wide range of cancer cell lines. Area-under-curve (AUC) values corresponding to cell lines in CTRP treated with either dinaciclib or alvocidib. Each point represents a cancer cell line in CTRP treated with the indicated compound. Boxplots depict the inner quartiles (boxes) and median value (horizontal line) with whiskers representing 1.5 × the interquartile range of 445 (dinaciclib-treated) or 440 (alvocidib-treated) cell lines. AUCs were computed as described in the Methods and at https://github.com/remontoire-pac/ctrp-reference/tree/master/auc.

Extended Data Figure 3. Chordoma cells are less sensitive to CDK4/6 and CDK8/19 inhibitors.

a) Response of chordoma cells to compounds targeting CDK4/6 and CDK8/19 proteins. Four chordoma cell lines were treated with indicated concentrations of compounds and assayed for cell viability after 6 d with CellTiter-Glo. Response data are represented by a fitted curve to the mean fractional viability at each concentration relative to vehicle-treated cells; error bars represent the SEM (n = 4 biological samples measured in parallel). b) Immunoblot analysis of UM-Chor1 cells treated with indicated concentrations of inhibitors or DMSO for 24 h. The experiment was performed twice for CCT251545 (one representative experiment displayed) and once for other compounds.

Extended Data Figure 4. T is super-enhancer-associated across chordoma cell lines.

a) Enhancers in 5 chordoma cell lines ranked by H3K27ac signal in each sample. Enhancers proximal (within 100 kb) to the T gene start site are annotated, as described in the figure. Super-enhancers (SEs) were determined by the inflection point of the plot. b) Table showing the ranks of top T-associated enhancers in each chordoma sample.

Extended Data Figure 5. Brachyury is highly expressed in chordoma cell lines.

a) Immunoblot analysis of chordoma and chondrosarcoma cell lines. Chordoma cell lines selectively express high levels of the brachyury protein. The experiment was performed once. b) Expression of T and MAX, as measured by RNA-sequencing, across 935 non-chordoma cancer cell lines derived from diverse tumor types. Data were generated as part of the Broad Institute Cancer Cell Line Encyclopedia (quantified data obtained from: https://ocg.cancer.gov/ctd2-data-project/translational-genomics-research-institute-quantified-cancer-cell-line-encyclopedia). Boxplots depict the inner quartiles (boxes) and median value (horizontal line) with whiskers representing 1.5 × the interquartile range. c) Gene expression levels of 115 super-enhancer (SE)-associated transcription factors in 5 chordoma cell lines (points), ranked by mean expression (horizontal ticks). d) T is amplified in the JHC7 chordoma cell line. Genomic copy-number alterations, inferred from whole-exome sequencing data, in five chordoma cell lines. A region of 2.06 Mb around the T locus on chromosome 6 shows 26-fold amplification in JHC7. This finding is consistent with the 2.6-Mb amplicon inferred from ChIP-seq whole-cell extract.

Extended Data Figure 6. T is super-enhancer-associated in patient-derived chordoma tumors.

a) Enhancers in chordoma tumors ranked by H3K27ac signal in each sample. Super-enhancers (SEs, red) and typical enhancers (black) proximal (within 100 kb) to the T gene start site are annotated. SEs were determined by the inflection point of the plot. b) Table showing the ranks of top T-associated SEs or typical enhancers in each chordoma sample.

Extended Data Figure 7. Patient-derived chordoma tumors express brachyury.

Immunohistochemical staining of patient-derived chordoma tumors for brachyury expression. H&E, hematoxylin and eosin. The experiment was performed once.

Extended Data Figure 8. THZ1 and actinomycin D reduce expression of T (brachyury) in a concentration- and time-dependent fashion.

a) Immunoblot analysis of UM-Chor1 cells treated with indicated concentrations of THZ1 or DMSO for 12, 24, 36, or 48 h. The experiment was performed once. b) UM-Chor1 cells were treated with indicated concentrations of compound or DMSO for 4, 8, or 12 h and subjected to RT-qPCR. Data are expressed as the log₂ fold-change of transcript levels relative to vehicle-treated cells, normalized to GAPDH levels, and represent the mean ± SD (n = 3 biological samples measured in parallel). Results of statistical analyses of RT-qPCR data, derived from a one-sided Welch’s t-test, are reported in Supplementary Table 9. c) Response of chordoma cells to treatment with actinomycin D. Four chordoma cell lines were treated with indicated concentrations of compound and assayed for cell viability after 6 d with CellTiter-Glo. Response data are represented by a fitted curve to the mean fractional viability at each concentration relative to vehicle-treated cells; error bars represent the SEM (n = 4 biological samples measured in parallel).

Extended Data Figure 9. Expression of ATP6V1B2, SAE1, SOX9, and TPX2 is downregulated following sgRNA-mediated T (brachyury) repression.

a) UM-Chor1 cells were transduced with sgRNAs targeting T or a non-targeting sgRNA control and subjected to RT-qPCR. Data are expressed as the fold-change of transcript levels relative to sgRNA control-treated cells, normalized to GAPDH levels, and represent the mean (n = 2 biological samples measured in parallel, represented by black points). * p < 0.05; ** p < 0.01; *** p < 0.001; p-values were derived from a one-sided Welch’s t-test. Exact p-values and effects sizes are reported in Supplementary Table 9. b) Immunoblot analysis of UM-Chor1 cells transduced with sgRNAs targeting T or a non-targeting sgRNA control. SgRNA treatment was performed once and immunoblotting was performed twice (one representative experiment displayed).

Extended Data Figure 10. THZ1 treatment can reduce brachyury expression in CH22 cells ex vivo and in vivo.

a) CH22 chordoma cells were treated with indicated concentrations of transcriptional CDK inhibitors and assayed for cell viability after 6 d with CellTiter-Glo. Response data are represented by a fitted curve to the mean fractional viability at each concentration relative to vehicle-treated cells; error bars represent the SEM (n = 4 biological samples measured in parallel). b) Immunoblot analysis of CH22 cells treated with indicated concentrations of inhibitors targeting CDK4/6 (palbociclib), CDK7/12/13 (THZ1), or CDK9 (dinaciclib, NVP-2, alvocidib) or DMSO for 48 h. The experiment was performed once. c) Weight change of mice treated with THZ1 or vehicle for the study depicted in Fig. 4h–i. d) THZ1 can downregulate brachyury expression in vivo. Immunoblot analysis of CH22 xenograft tumors following treatment with indicated doses of THZ1 or vehicle twice daily for 5 d. The experiment was performed once. e) Immunoblot analysis of CH22 xenograft tumors following treatment with THZ1 or vehicle twice daily for 3 d. Top and bottom panels represent two independent studies (bottom panel corresponds to the study depicted in Fig. 4h–i).

Figure 1. Genome-scale CRISPR-Cas9 screening identifies T (brachyury) as a selectively essential gene in chordoma cells.

a) Comparative analysis of genome-scale CRISPR-Cas9 screening results generated in UM-Chor1 and MUG-Chor1 chordoma cell lines versus 125 non-chordoma cancer cell lines. SgRNAs were ranked by and plotted against selectively lethal effects (RNMI scores; see Methods) in chordoma versus non-chordoma cell lines. The top three sgRNAs (red circles) all target T. b) Viability after sgRNA treatment (represented by sgRNA-dependency scores; see Methods) corresponding to each of the four primary screening sgRNAs targeting either T or RPL23 across 127 cancer cell lines (CCLs), including 2 chordoma CCLs (blue circles). For each sgRNA, the median sgRNA-dependency score across 127 cell lines is indicated (gray line). Lower values indicate greater sgRNA depletion and thus essentiality of the target gene (shaded region). c) (Left) Proliferation of chordoma cell lines transduced with sgRNAs targeting T or a non-targeting sgRNA control. Points represent the mean ± SD (n = 3 biological samples measured in parallel). **** p < 0.0001, derived from a two-way ANOVA (p-values for the test comparing sg-EGFP and sg-T #1 are displayed). Exact p-values and effect sizes are reported in Supplementary Table 9. (Right) Immunoblot analysis of transduced cells confirming sgRNA-mediated protein repression. Proliferation experiments were performed twice with UM-Chor1 and U-CH2 cells (one representative experiment displayed for each) and once with MUG-Chor1 cells; immunoblots were performed once. d) Relative gene expression of UM-Chor1 cells transduced with one of two sgRNAs targeting T versus a non-targeting sgRNA control. Gene expression was measured with RNA sequencing. Data represent two biological replicates per condition. e) Gene-set enrichment analysis results show significant downregulation of genes associated with cell cycle progression, such as drivers of G2/M checkpoint progression and E2F targets.

Figure 2. Small-molecule sensitivity profiling identifies inhibitors of CDK7/12/13 and CDK9 as potent antiproliferative agents in chordoma cells.

a) Profiling results for 459 small molecules tested in four chordoma cell lines. Data represent area-under-curve (AUC) values calculated from 8-point concentration-response curves generated in duplicate, and coloring in the heatmap is representative of smaller AUC values (blue) corresponding to more potent effects (AUC value of 1 = vehicle control; value of 0 = complete killing at all concentrations). Left heatmap, all compounds were ranked by average potency in chordoma cell lines; potent compounds (average AUC < 0.8, boxed) were then filtered to exclude those with cytotoxic effects non-selective for chordoma cell lines versus up to 891 non-chordoma cancer cell lines, when such data were available in CTRP. Compounds that were only tested in chordoma cell lines and were not available in CTRP were not filtered beyond the initial potency filter. Twenty-eight compounds (right heatmap) passed these criteria. b) Validation of primary screening hits and related compounds. Four chordoma cell lines were treated with indicated concentrations of candidate antiproliferative compounds and assayed for cell viability after 6 d with CellTiter-Glo. Response data are represented by a fitted curve to the mean fractional viability at each concentration relative to vehicle-treated cells; error bars represent the SEM (n = 4 biological samples measured in parallel). c) Rankings of sgRNA-level dependency scores (see Methods) for the indicated genes following genome-scale CRISPR-Cas9 screening (see Fig.1). Points represent each of four sgRNAs targeting a given gene that were present in the pooled CRISPR library used for screening. P-values were derived from a one-sided Mann-Whitney test. *, p < 0.05; **, p < 0.01; *** p < 0.001; n.s., not significant. The gray waterfall plot represents ranked dependency scores for all sgRNAs tested (median dependency score = −0.3257 for MUG-Chor1; median = −0.3283 for UM-Chor1). Median dependency scores for the four sgRNAs and exact Mann-Whitney p-values corresponding to each gene are reported in Supplementary Table 9. d) Immunoblot analysis of UM-Chor1 cells treated with indicated concentrations of THZ1 or DMSO for 24 h. Data are representative of two independent experiments. e) Caspase-3/7 activity (top) and cell viability (bottom) following THZ1 treatment of UM-Chor1 cells. Caspase-3/7 activity and cell viability were measured in parallel at the indicated time points using Caspase-Glo 3/7 and CellTiter-Glo reagents, respectively. Data are expressed as the fold-change of caspase-3/7 activity (top) or fraction of cell viability (bottom) relative to vehicle-treated cells and represent the mean ± SD (n = 4 biological samples measured in parallel).

Figure 3. T (brachyury) is super-enhancer-associated and highly active across chordoma cell lines and in patient-derived chordoma tumors.

a,b) Gene tracks of H3K27ac, brachyury, and ATAC-seq signal (units of reads per million per base pair) at the T and MAX gene loci. For datasets with multiple samples (H3K27ac and ATAC-seq), signals for samples are plotted as a translucent shape and darker regions indicate regions with signal in more samples. An opaque line is plotted and gives the average signal across all samples in the group. H3K27Ac ChIP-seq was performed on 5 chordoma cell lines (UM-Chor1, MUG-Chor1, U-CH2, U-CH1, JHC7); brachyury ChIP-seq was performed on 1 chordoma cell line (U-CH1) as previously reported; and ATAC-seq was performed on two chordoma cell lines (2 biological replicates for U-CH2 and 1 replicate for MUG-Chor1). c) Bar graphs of RNA-seq mRNA levels for T and MAX across 5 chordoma cell lines. Units are in transcripts per million (tpm). d) Gene expression levels of the top 30 (of 115) super-enhancer (SE)-associated transcription factors in 5 chordoma cell lines (points), ranked by mean expression (horizontal ticks). e,g) Enhancers in chordoma cell lines (UM-Chor1, MUG-Chor1, U-CH2, U-CH1, JHC7) or patient-derived chordoma tumor tissue ranked by average H3K27ac signal across samples. SEs and associated genes are annotated along the vertical axis and were determined by the inflection point of the plot. f,h) Gene tracks of H3K27ac signal at the T-amplified region in the JCH7 cell line and the corresponding region in the MGH_1 chordoma tumor. The amplicon and T-proximal SEs are shown (red boxes). i) Gene tracks of H3K27ac signal across eight patient-derived chordoma tumor samples and one matched normal adjacent muscle sample at the T locus. j) Clustergram of chordoma samples hierarchically clustered by similarity of H3K27ac signal at the union of all SE regions across samples.

Figure 4. Inhibitors of CDK7/12/13 and CDK9 downregulate T (brachyury) expression, and THZ1 treatment reduces chordoma tumor proliferation in vivo.

a) Immunoblot analysis of chordoma cells treated with indicated concentrations of antiproliferative compounds or DMSO for 48 h. The experiment was performed four times with UM-Chor1 cells (one representative experiment displayed) and once with MUG-Chor1 cells. b) Immunoblot analysis of chordoma cells treated with indicated concentrations of inhibitors targeting CDK4/6 (palbociclib), CDK7/12/13 (THZ1), or CDK9 (dinaciclib, NVP-2, alvocidib) or DMSO for 48 h. The experiment was performed twice with MUG-Chor1, UM-Chor1, and U-CH1 cells (one representative experiment displayed for each) and once for JHC7 cells. c) Relative gene expression of UM-Chor1 cells treated with THZ1 (100 nM or 500 nM) (top), or actinomycin D (10 ng/mL or 150 ng/mL) (bottom), versus vehicle-treated cells. Cells were treated with compound or vehicle (DMSO) for 4 h. Gene-expression profiling was performed with RNA-seq, and genes downregulated following treatment with both concentrations of a given compound are indicated (shaded region). Ranked by significance, T has a rank of 128 and 139 out of 37,043 genes for 100 nM and 500 nM THZ1, respectively; and a rank of 11,607 and 2,539 out of 37,043 genes for 10 ng/mL and 150 ng/mL actinomycin D, respectively. Data represent two biological replicates per condition, and false-discovery-rate (FDR) values were derived from a two-sided Wald test, with a Benjamini-Hochberg correction. d) Immunoblot analysis of UM-Chor1 cells transduced with V5-tagged ORFs encoding T or a LACZ control gene and then treated with indicated concentrations of THZ1. Unlike endogenous brachyury expression, ectopic brachyury expression was not greatly reduced with THZ1 treatment. The experiment was performed twice (one representative experiment displayed). e) Cell viability of UM-Chor1 cells transduced with an ORF encoding either LACZ or T and then treated with 500 nM THZ1. Data are expressed as the fraction of cell viability relative to vehicle-treated cells and represent the mean ± SD (n = 12 biological samples measured in parallel). **** p < 0.0001, derived from a two-tailed, unpaired t test. Exact p-values and effect sizes are reported in Supplementary Table 9. f) Integration of super-enhancer-associated and essential genes in UM-Chor1 and MUG-Chor1 cell lines. Essential genes were identified for each cell line by applying the STARS algorithm (see Methods) to genome-scale CRISPR-Cas9 screening data described in Fig. 1. Asterisk, genes downregulated following sg-T-mediated T repression as measured by RNA-seq (experiment described in Fig. 1d). Cross, gene loci bound by brachyury as measured by brachyury ChIP-seq (using a previously reported dataset; see Methods). g) Model of T (brachyury) gene control. h) Schematic of the dosing schedule for THZ1 treatment of a human xenograft mouse model of chordoma. i) Tumor proliferation in mice engrafted with CH22 cells and treated with vehicle or THZ1. Data represent the mean tumor volume ± SEM of 12 (vehicle-treated) or 14 (THZ1-treated) mouse tumors at the indicated time points. **** p < 0.0001, derived from a two-way ANOVA with repeated measures. Exact p-values and effect sizes are reported in Supplementary Table 9.