Preclinical modeling of leiomyosarcoma identifies susceptibility to transcriptional CDK inhibitors through antagonism of E2F-driven oncogenic gene expression

Abstract

Purpose:

Leiomyosarcoma (LMS) is a neoplasm characterized by smooth muscle differentiation, complex copy-number alterations, tumor suppressor loss and the absence of recurrent driver mutations. Clinical management for advanced disease relies on the use of empiric cytotoxic chemotherapy with limited activity, and novel targeted therapies supported by preclinical research on LMS biology are urgently needed. A lack of fidelity of established LMS cell lines to their mesenchymal neoplasm of origin has limited translational understanding of this disease, and few other preclinical models have been established. Here, we characterize LMS patient derived xenograft (PDX) models of LMS, assessing fidelity to their tumors of origin and performing preclinical evaluation of candidate therapies.

Experimental Design:

We implanted 49 LMS surgical samples into immunocompromised mice. Engrafting tumors were characterized by histology, targeted next-generation sequencing, RNA-seq and ultra-low passage whole-genome sequencing. Candidate therapies were selected based on prior evidence of pathway activation or high-throughput dynamic BH3 profiling.

Results:

We show that LMS PDX maintain the histologic appearance, copy-number alterations and transcriptional program of their parental tumors across multiple xenograft passages. Transcriptionally, LMS PDX co-cluster with paired LMS patient-derived samples and differ primarily in host-related immunologic and microenvironment signatures. We identify susceptibility of LMS PDX to transcriptional CDK inhibition, which disrupts an E2F-driven oncogenic transcriptional program and inhibits tumor growth.

Conclusions:

Our results establish LMS PDX as valuable preclinical models and identify strategies to discover novel vulnerabilities in this disease. These data support the clinical assessment of transcriptional CDK inhibitors as a therapeutic strategy for LMS patients.

INTRODUCTION

Leiomyosarcoma (LMS) is a common type of soft-tissue sarcoma putatively derived from smooth muscle cells or their mesenchymal precursors. While LMS has no known recurrent single-nucleotide variants leading to oncogene activation, tumors characteristically display multiple copy-number alterations with common loss of tumor suppressor genes including RB1, TP53, ATRX and PTEN (1–3). Multiple large scale sequencing efforts have identified a recurrent gene expression program in LMS, including the description of three molecularly-defined subtypes within this disease (3–5). Based on genomic, transcriptional and other profiling data, proposed therapeutic targets in LMS have included inhibition of IGF1R signaling given its high expression in a subset of LMS and association with poor clinical outcomes (4,6), loss of argininosuccinate synthase 1 causing susceptibility to arginine deprivation (7), ATR inhibition owing to recurrent ATRX disruption and associated activation of the alternative lengthening of telomeres (ALT) pathway (1,8), and defects in homologous recombination and susceptibility to DNA repair pathway inhibition (1,9). We and others have shown the near-universal loss of RB1 in LMS, with comparative enrichment of E2F1 expression and activation of an E2F1-driven oncogenic gene expression program (2–4); as RB normally functions as a tumor suppressor by negatively regulating the cell cycle, including by directly binding to and disrupting transcriptional activation through E2F1 (10), antagonizing E2F1 represents a rational mechanism-based therapeutic approach in a majority of LMS cases.

Current clinical management of LMS relies on the use of empirically identified cytotoxic therapies that have modest clinical activity (11). Investigation of novel targeted therapies supported by mechanism-based hypotheses are needed to improve clinical outcomes in this disease. While several large sequencing efforts have broadly characterized LMS, mechanistic studies have been limited by lack of available preclinical models. Several cell line models of LMS have been described, though studies of available models suggest loss of fidelity to their tumor of origin (4,12), possibly arising from divergent in vitro growth patterns in the setting of tumor suppressor loss and lack of defined oncogenic pathways characteristic of this disease. Several patient derived xenograft (PDX) models of LMS have been described (9,13,14), though detailed characterization of such models is necessary to demonstrate their conformity to the human disease and value in preclinical testing.

To address the need for model development and validation in LMS and support translational research, we have generated and characterized a panel of LMS PDX models. All engrafted PDX lines exhibited alterations in TP53, RB1, PTEN and/or ATRX, among other tumor suppressor genes, with the exception of one PDX line that was driven by an ALK rearrangement. Evaluated LMS PDX retain the histopathologic features and genomic copy number alterations of their tumors of origin for up to 17 generations, and circulating tumor DNA (ctDNA) was found in the plasma of tumor-bearing mice. Transcriptional profiling showed co-clustering of PDX from several generations with their tumor of origin, with maintenance of LMS-subtype specific gene expression. In PDX with target-relevant genomic and transcriptional profiles, we found no benefit to monotherapy treatment with ATR or IGF1R inhibitors. Utilizing high-throughput dynamic BH3 profiling (HT-DBP), we identified susceptibility of LMS to transcriptional CDK inhibitors. Treatment of LMS PDX with broad and narrow spectrum transcriptional CDK inhibitors reduced tumor growth and antagonized the RB-deficient E2F-associated growth program. Our findings describe the development and detailed characterization of high-fidelity preclinical models of LMS, detail strategies to define potential therapeutic vulnerabilities, and identify transcriptional CDK inhibition as an effective means of antagonizing a principal oncogenic program in LMS.

MATERIALS AND METHODS

Xenograft Models.

Tumor samples isolated for xenografting experiments were obtained from patients undergoing clinically indicated surgery and following written informed consent to a Dana-Farber Cancer Institute (DFCI) and Brigham and Women’s Hospital (BWH) IRB-approved and U.S. Common Rule compliant research protocol. Fresh or cryopreserved tumors were implanted subcutaneously into ~6 week old female nude mice (NU/NU; Charles River Laboratories). Histologic review by a pathologist with expertise in sarcoma was performed to confirm the diagnosis of leiomyosarcoma, assess mitotic rate and grade, and compare parental tumors to PDX. Three attempted xenografts had no archival parental tumor available for pathology review, and mitotic rate and grade were obtained from the original clinical report. LMS7 and LMS29 were grown in the presence of subcutaneously implanted 17β-estradiol-sustained release pellets (Innovative Research of America, Cat# NE-121).

Drug Administration.

Engrafted mice were enrolled into treatment groups when tumors reached approximately 100–200 mm³ in size, as measured by calipers and determined by the tumor volume equation: volume = long diameter² × short diameter × 0.5. Mice were randomly assigned to treatment groups administered berzosertib (60 mg/kg solubilized in 5% DMSO/45% PEG300/50% H2O administered by oral gavage 5 days per week; Medchem Express Cat# HY-13902), linsitinib (50 mg/kg solubilized in 10% DMSO/40% PEG300/5% Tween 80/45% normal saline administered by oral gavage 5 days per week; LC Laboratories Cat# L5814), flavopiridol (50 mg/kg solubilized in 5% DMSO/30% PEG300/65% H2O administered i.p. 5 days per week; Selleck Chemicals Cat# S1230) or YKL-5–124 (2.5 mg/kg 10% DMSO/40% PEG300/5% Tween 80/45% normal saline administered by i.p., 5 days per week; Selleck Chemicals Cat#S8863). The berzosertib and flavopiridol treatments were performed simultaneously, and the same LMS33 control group was used for comparisons. The number of mice in each experimental group are indicated in the figure and associated legend. For drug efficacy studies, mice were treated for 21 days. For histologic evaluation of flavopiridol (Fig. 4D–E), drug was administered for 3 days. For transcriptional profiling of YKL-5–124 (Fig. 5C–H), drug was administered for 5 days. No statistical methods were used to predetermine sample size, and no animals died during drug treatment. All procedures were conducted under protocols approved by the Institutional Animal Care and Use Committee at DFCI. Tumors were dissected and flash frozen or fixed in 10% formalin for corollary studies including H&E staining, immunohistochemistry, in situ hybridization of sectioned tumors and sequencing-based studies.

Figure 4. HT-DBP of LMS patient samples and PDX identifies sensitivity to transcriptional CDK inhibitors.

A, Heatmap showing delta priming from HT-DBP of four LMS PDX and two patient samples (n = 2 per data point). Compounds used in the screen are organized into rows by drug class (Table S2). B, Cytochrome c release assay in LMS33 utilizing the indicated concentrations of CDK inhibitors, with DMSO, doxorubicin and navitoclax used as controls (n = 6). C, Tumor volume of LMS33 PDX in response to treatment with vehicle (n = 7) or flavopiridol (50 mg/kg i.p., 5 days per week; n = 6). Data were analyzed by two-way ANOVA, compared to vehicle; ***,P<0.001. D, Representative images of LMS33 treated with flavopiridol or vehicle for 3 days and stained for Ki-67 (scale bar = 50 μm) or CC3 (scale bar = 100 μm). E, Quantification of Ki-67-positive cells in LMS33 PDX treated with vehicle (n = 4) or flavopiridol (n = 3) for 3 days. Data were compared by two-tailed t-test; *,P<0.05.

Figure 5. Selective CDK7 inhibition decreases tumor growth and the E2F-driven oncogenic program in LMS.

A, Tumor volume of LMS4 PDX in response to treatment with vehicle (n = 5) or YKL-5–124 (2.5 mg/kg i.p., 5 days per week; n = 5). B, Tumor volume of LMS33 PDX in response to treatment with vehicle (n = 7) or YKL-5–124 (2.5 mg/kg i.p., 5 days per week; n = 6). Data were analyzed by two-way ANOVA, compared to vehicle; ***,P<0.001. C, GSEA butterfly plot of FDR and NES for RNA-seq data comparing LMS33 tumor-bearing mice treated with vehicle (n = 3) or YKL-5–124 (n = 3) for 5 days. The top three gene sets are labeled in each condition. D-F, GSEA plots of Hallmark gene sets for Epithelial Mesenchymal Transition (EMT), E2F Targets and G2M Checkpoint; the NES and FDR are shown. G, Expression in FPKM of exemplary genes downregulated by YKL-5–124 and involved in oncogenic gene transcription and cell division. H, Expression in FPKM of exemplary genes upregulated by YKL-5–124 and involved in gene transcription, cellular stress response and DNA repair. Data were compared by two-tailed t-test; *,P<0.05; **,P<0.01; ***,P<0.001.

Immunohistochemistry and In Situ Hybridization.

Immunohistochemical staining was performed on 4 μm sections prepared from formalin-fixed, paraffin-embedded tissue blocks after antigen retrieval using a citrated buffer pressure cooker protocol with following antibodies: SMA (1:400, Cell Signaling Technology Cat# 19245, RRID:AB_2734735), desmin (1:1,000, Thermo Fisher Scientific Cat# MA5–33065, RRID:AB_2810158), Ki-67 (1:400, Cell Signaling Technology Cat# 9027, RRID:AB_2636984) and cleaved caspase-3 (1:250, Cell Signaling Technology Cat# 9579, RRID:AB_10897512). A secondary antibody conjugated to HRP was used (Cell Signaling Technology Cat# 8114). Reactions were developed using DAB (Cell Signaling Technology Cat# 8059) or NovaRed (Vector Laboratories Cat# SK-4800) substrate kits per manufacturer recommendations. For Ki-67 quantification, the percent of Ki-67-positive cells in two separate areas of a stained section were divided into quadrants and individually counted by three blinded reviewers (8 regions per tumor). The resulting percent Ki-67 positive cells were averaged across quadrants and then the entire section to generate a single value for each PDX sample.

RNAscope hybridization and staining assays with IGF1R probe (Advanced Cell Diagnostics Cat# 415811) were performed on 4 μm thin sections prepared from formalin-fixed, paraffin-embedded tissue blocks per manufacturer recommendations (Advanced Cell Diagnostics Cat# 322300).

Ultra-Low Passage Whole Genome Sequencing.

Extracted DNA was quantified using Quant-iT PicoGreen dsDNA Assay Kit (Thermo Fisher Scientific Cat# P7589). For cell-free DNA, a QIAamp Circulating Nucleic Acid Kit (Qiagen Cat# 55114) was used to isolate cell-free DNA in frozen plasma from LMS33 and LMS19, though ctDNA could only be identified in LMS33. High molecular weight DNA contamination of the cell-free DNA was determined by Bioanalyzer (Agilent) and size selection was performed if necessary (AMPure XP beads, Beckman Coulter Cat# A63881). DNA extracted from FFPE or fresh frozen tumors was fragmented with sonication (Covaris) to approximately 250 bp and purified with AMPure XP beads. Up to 40 ng of cell-free DNA, 100 ng of DNA from fresh frozen tissue, and 200 ng of DNA from FFPE tissue were used for KAPA Hyper library preparation (Kapa Biosystems Cat# KK8500). Libraries were assessed for quality by Bioanalyzer followed by quantification using the MiSeq Nano flow cell (Illumina). Barcoded libraries were pooled and sequencing was performed on a HiSeq 2500 (Illumina). Sequencing results were demultiplexed, aligned, and processed using Picard, BWA alignment (15), and GATK tools (16,17). To assess for copy number segments in tumor and cell-free DNA, ichorCNA software was utilized (18). For copy neutral segments predicted by ichorCNA, the log2 ratio was set to zero.

RNA-seq.

Total RNA was isolated using the RNeasy Plus Kit (Qiagen Cat# 74136), and concentration measured by Nanodrop (Thermo Fisher Scientific) and quality by TapeStation 4200 (Agilent). Library preparation was performed using the NEBNext Ultra II DNA Library Prep Kit (New England Biolabs Cat# E7645S). Sequencing was performed on a NextSeq 500 or NovaSeq 6000 (Illumina). RNA-seq data were aligned to hg19 using STAR (19) with expression quantification using Cufflinks (20) to generate gene expression values in fragments per kilobase of transcript per million mapped reads (FPKM) units. Gene set enrichment analysis (GSEA, RRID:SCR_003199) (21) was performed using Hallmark or C7 gene lists in the Molecular Signatures Database (software.broadinstitute.org/gsea/).

High-Throughput Dynamic BH3 Profiling.

HT-DBP using chemical libraries or drug printed plates was performed as previously described (22). LMS tumors and PDX were dissociated using collagenase 4 (Life Technologies Cat# 17104019) and hyaluronidase (Worthington Biochemical Corporation Cat# LS002594) for 0.5–1 hour. Cells were plated in collagen coated 384 well plates (Corning Cat# 354667), drugged with indicated compounds, and were BH3-profiled 24 hours after drug treatment using the synthetic PUMA BH3 peptide. For chemical screens (Fig. 4A), compounds from the LINCS4 kinase inhibitor library at the Institute of Chemistry and Cell Biology (ICCB) at Harvard Medical School were pin transferred at 4 concentrations between 0.1 and 1.1 uM with n = 2 per concentration. For limited drug validation experiments in Fig. 4B, selected drugs were dispensed into 384 well plates using the HPD300 digital drug dispenser (Hewlett Packard and Tecan) with n = 6 per concentration. Upon drug treatment, cells were incubated at 37°C for 24 h. Subsequently, medium was washed from plates with PBS using the BioTek 406EL plate washer (BioTek). A 2x concentrated BH3 profiling buffer was added to cells with 0.002% digitonin and the highest PUMA BH3 peptide concentration which did not result in mitochondrial outer membrane permeabilization when pre-treated with DMSO. For peptide titrations, 2x BH3 profiling buffer was added manually. For BH3 profiling of drug-treated plates at a single peptide concentration, the 2x buffer was added using the Thermo Multidrop Combi (Thermo Fisher Scientific) and incubated for 1 hour at 27°C. Cells were fixed in 2% paraformaldehyde. Fixatives were neutralized using a tris/glycine buffer, and cells were stained overnight with Hoechst 33342 (Invitrogen Cat# H3570) or Cytochrome c–Alexa Fluor 647 antibody (BioLegend Cat# 612310). All imaging was performed on the IXM XLS high-content widefield microscope (Molecular Devices; at the ICCB at Harvard Medical School). A 10x objective was used to perform all imaging. Image analysis was performed in MetaMorph using the multiwavelength cell-scoring module and the adaptive background correction module to segment cells on the basis of an intensity above local background. This results in an approximate single-cell segmentation and the area of cytochrome c intensity. Cells were scored as being positive or negative on the basis of the area, and in each well the percentage of cells that are cytochrome c positive were quantified. To calculate delta priming for a single drug concentration, the percent positive cytochrome c cells per drug treated well was subtracted from the percent positive cytochrome c cells from DMSO treated wells. We determined a per drug delta priming by calculating the trapezoidal area under the curve of multiple drug concentrations.

Data Analysis and Availability.

Center values, error bars, P-value cutoffs, number of replicates and statistical tests are identified in the corresponding figure legends. Samples sizes were not predetermined. Data analysis not otherwise specified was performed in Excel or GraphPad Prism. RNA- and DNA-sequencing data is available through the dbGaP accession number phs002587.v1.p1. We endeavor to make these PDX models available to the research community to advance basic and translational investigation for this poorly understood disease.

RESULTS

Generation of LMS PDX and characterization of tumor histology, tumor suppressor loss and copy-number alterations.

Previously, we and others have used sequencing-based strategies to demonstrate the divergence of established LMS cell lines from LMS clinical specimens at epigenetic and transcriptional levels (4,12). These findings motivated the development of xenografts of LMS to establish valid preclinical models of this disease and enable translational research. We attempted to generate 49 LMS PDX lines from independent surgical samples (Table S1). Of these 49 samples, 17 (35%) were able to engraft and propagate at least three generations (F3) following implantation. The 32 tumors that failed to engraft and propagate to F3 were observed for growth for an average of 141 days (range 82–268 days) prior to termination. Among the entire cohort of implanted tumors, most samples were derived from female patients (43/49, 88%). Comparing engrafting and non-engrafting tumors, there were comparable rates of prior treatment with chemotherapy and/or radiation and similar rates of derivation from primary versus metastatic disease; while the frequency of extrauterine (ELMS) and uterine (ULMS) tumors were identical in non-engrafting samples, a higher percentage of the engrafting samples were derived from ULMS (11 of 17 tumors, 65%). As compared with non-engrafting tumors, engrafting tumors had a significantly higher average mitotic rate (Fig. 1A). Tumors derived from high-grade specimens were nearly twice as likely to engraft compared to low- or intermediate-grade specimens (Fig. 1B). Four tumors that were treated with pre-operative radiotherapy (i.e. immediately preceding sample collection for attempted engraftment) failed to engraft. Tumors from two patients with Li-Fraumeni syndrome and one with hereditary retinoblastoma were implanted, with successful engraftment of one Li-Fraumeni derived PDX (LMS37). Repeat derivation of PDX lines were attempted in three patients who underwent successive resections at least two years apart, and in each case neither the original nor the repeat tumor sample was able to engraft.

Figure 1. LMS PDX maintain histologic and copy number changes across generations.

A, Mitotoses per 10 high-power fields (HPF) comparing parental tumors of non-engrafting and engrafting PDX. Data were compared by two-tailed t-test; **P = 0.0011. B, Engraftment rates comparing parental tumors of high- or low/intermediate-grade. C-D, H&E, smooth muscle actin (SMA) and desmin (DES) staining of the parental tumor and serial passages of LMS7 (C, intermediate-grade) and LMS20 (D, high-grade); scale bar = 50 μm. Inset images in D indicate focal and weakly desmin-positive cells. E, Copy number plots generated from ULP-WGS in the LMS33 parental tumor and subsequent generations of PDX extending to F17. The x-axis indicates chromosome and y-axis copy number (log₂ ratio). F, Copy number plots generated from ULP-WGS of plasma derived from LMS33 to detect circulating tumor DNA (ctDNA).

Of the successfully propagating PDX, we evaluated histologic, genomic and transcriptomic features in a majority of lines (Table 1). The majority of these PDX were derived from ULMS and metastatic in origin, and 11 of 17 tumors had prior treatment with chemotherapy and/or remote radiation (i.e. not pre-operative radiation). Two ULMS tumors, LMS7 and LMS29, stained positive for estrogen receptor (ER) expression; LMS7 was able to propagate with prolonged passage times, while LMS29 could only propagate with simultaneous administration of estrogen through subcutaneously implanted 17β-estradiol sustained release pellets. We characterized parental tumors of successfully engrafted PDX with a targeted next-generation sequencing (NGS) assay probing for alterations in up to 447 cancer genes including surveying 60 genes for rearrangement detection (23). All parental tumors were found to have detectable alterations in at least two of the most commonly altered tumor suppressor genes in LMS (TP53, RB1, PTEN and ATRX), with the exception of LMS45. LMS45 had a lower tumor mutational burden (TMB) compared to most other samples, and NGS profiling identified an oncogenic ALK rearrangement. TMB varied between 1.521 and 9.885, and all samples were mismatch repair proficient by NGS testing.

Table 1. LMS PDX.

Columns indicate the engrafted PDX line, patient age at sample collection, patient gender, tissue of origin (ELMS, extrauterine LMS; ULMS, uterine LMS), whether tumors were collected from primary or metastatic samples (Status), treatment(s) prior to PDX collection (C, chemotherapy; RT, radiation therapy; -, none), approximate passage time in days from PDX implantation to growth ~1 cm in diameter, whether RNA-seq and ULP-WGS was performed, genomic alterations detected in tumor suppressor genes TP53, RB1, PTEN and ATRX, tumor mutational burden/megabase (TMB; n.d. = not determined), and notes on specific PDX features.

LMS PDX	Age	Gender	Origin	Status	Treatment	Passage (d)	RNAseq	ULP-WGS	TP53	RB1	PTEN	ATRX	TMB	Features
LMS4	52	F	ULMS	Metastatic	C	55	+	+	+	+	+	+	3.042
LMS7	67	F	ULMS	Metastatic	-	125	+	+		+	+		3.802
LMS16	66	F	ULMS	Metastatic	C,RT	90	+	+	+	+	+		6.083
LMS17	56	F	ELMS	Metastatic	C	83
LMS19	63	F	ULMS	Metastatic	C	70	+	+		+	+	+	9.885
LMS20	82	F	ELMS	Primary	C	40	+	+	+	+		+	5.232
LMS28	48	F	ULMS	Metastatic	C	90	+	+	+		+		1.521
LMS29	63	F	ULMS	Metastatic	C,RT	70			+	+		+	n.d.
LMS30	48	F	ELMS	Metastatic	-	70			+	+	+	+	3.802
LMS31	48	F	ELMS	Metastatic	-	60	+	+	+	+			2.281
LMS33	49	F	ULMS	Metastatic	C,RT	60	+	+	+	+	+	+	3.802
LMS36	50	F	ULMS	Metastatic	C	74
LMS37	52	F	ULMS	Metastatic	-	60			+	+	+	+	n.d.	Li-Fraumeni
LMS38	59	F	ULMS	Metastatic	-	70	+	+	+	+	+	+	8.365
LMS42	68	F	ELMS	Primary	-	114
LMS45	66	F	ULMS	Metastatic	C	85	+						1.521	ALK rearrangement
LMS48	28	F	ELMS	Metastatic	C,RT	90	+		+	+			3.802

Data Not Collected

No Alteration Detected by NGS

Compared to parental tumors, early and late passages of PDX retained the histologic features of their tumors of origin, specifically growth pattern, cytomorphology, cellularity and comparable mitotic rates. Further, LMS PDX retained staining patterns for smooth muscle actin (SMA) and desmin (DES, Fig. 1C–D, Fig. S1) identical to their tumors of origin. To evaluate for genomic copy number evolution over time in the xenografted tumors, we performed ultra-low passage whole-genome sequencing (ULP-WGS) on parental tumors and several generations of PDX. Compared to parental tumors, successive generations of PDX faithfully retained copy-number alterations across the genome (Fig. 1E, Fig. S2–S3). As has been previously noted in patient samples (3,24), we found genome-wide heterogeneity in gained and lost segments between different PDX lines. In one of two PDX lines assessed, we were able to detect shedding of tumor DNA into the plasma of xenografted mice (Fig. 1F), as we have previously shown for a subset of LMS patients with active disease (24). Taken together, these results illustrate the feasibility of generating a cohort of LMS PDX models and demonstrate histologic and genomic fidelity to their tumors of origin.

LMS PDX retain the transcriptional program of their parental tumors.

LMS has a recurrent gene expression program distinct from other sarcomas and related to its underlying smooth-muscle lineage (4). To evaluate PDX fidelity to this recurrent program, we performed RNA-seq on parental tumors, early passage PDX (F0-F3), and late passage PDX (>F3). Unsupervised hierarchical clustering of LMS tumors, PDX and LMS cell lines showed co-clustering of the majority (10 of 11) of PDX with their parental tumor, regardless of early or late passage status (Fig. 2A, Fig. S4A). PDX maintain the expression of commonly used LMS histologic markers (Fig. S4B), including universal expression of smooth muscle actin and variable expression of desmin, which corroborate histologic findings (Fig. S1, Fig. 1C–D). LMS cell lines clustered separately from LMS tumors and PDX, as we have previously seen in comparisons to larger tumor cohorts (4). Dimensionality reduction using principal component analysis (PCA) of RNA-seq data showed similar separation of LMS cell lines from tumors and PDX (Fig. 2B).

Figure 2. Transcriptional profiling of LMS tumors and preclinical models.

A, Unsupervised hierarchical clustering of RNA-seq data from the 10,000 highest expressed genes (rows) across LMS patient samples, early and late PDX passage and cell lines (columns). * indicates one PDX that failed to co-cluster with its parental tumor. Patient samples were included in this analysis which did not have an associated engrafted PDX. B, PCA of LMS patient samples (red), early and late PDX (blue and yellow, respectively) and cell lines (gray). C-E, Plot of Log₂ FPKM in primary tumors and early and late generations of PDX for genes associated with molecular subtypes of LMS including SYNM and ADIRF (cLMS, black), PDGFRA and DCN (iLMS, blue), and ESR1 and CHRDL2 (uLMS, red). F, GSEA butterfly plot of false discovery rate (FDR) and normalized enrichment score (NES) for the C7: Immunologic Signatures gene sets. The Hallmark (H) gene sets for Angiogenesis, Hypoxia, Inflammatory Response, Interferon Gamma Response, Interferon Alpha Response, Allograft Rejection, Complement, IL6-JAK-STAT3 Signaling, TNFA Signaling via NFKB and IL2-STAT5 Signaling are also shown. G-H, GSEA plots of Hallmark gene sets for Inflammatory Response and Hypoxia.

We and others have previously shown that LMS can be molecularly classified into three subtypes, which we designated conventional (cLMS), inflammatory (iLMS) and uterogenic (uLMS) based on gene sets enriched in each subtype (4). LMS PDX can similarly be stratified by LMS subtype-enriched genes, with transcripts within each group showing enrichment including SYNM and ADIRF in cLMS (Fig. 2C), PDGFRA and DCN in iLMS (Fig. 2D), and ESR1 and CHRLD2 in uLMS (Fig. 2E). To explore systematic transcriptional differences between parental LMS tumors and PDX, we performed gene set enrichment analysis (GSEA) comparing these two groups. Among Hallmark gene sets, those related to angiogenesis and immune function were significantly enriched in parental tumors (Fig. 2F–H). Similarly, evaluation of all Immunologic Signature (C7) gene sets showed recurrent enrichment in parental tumors. These data demonstrate the overall fidelity of LMS PDX to their parental tumors and that oncogenic transcriptional programs are fastidiously maintained in PDX following multiple passages. The predominant transcriptional differences in PDX as compared to parental tumors arises from endothelial and immune cell gene expression, which are anticipated in the context of focusing analysis on the human transcriptome and expected host microenvironment factors in these immunocompromised murine models.

ATR and IGF1R inhibitors fail to decrease growth of target-enriched xenografts.

Several targeted therapies have been proposed for LMS but preclinical evaluation has been limited due to the lack of available models. LMS is among the most common tumor subtypes to genetically delete or mutate ATRX, leading to activation of the ALT pathway that is dependent upon ATR (25,26). ATR inhibitors have been developed that can target tumors which rely on the ALT pathway, and this therapeutic strategy has already entered early clinical trials for ATRX-mutant LMS (e.g. NCT03718091). To establish whether ATR inhibition shows preclinical activity in an LMS PDX, we utilized LMS33, which has bi-allelic ATRX inactivation and lacks ATRX gene expression (Table 1, Fig. 3A). The potent and selective ATR inhibitor berzosertib was administered to LMS33 over a 21-day period at a dose of 60 mg/kg by oral gavage, which was previously shown to be efficacious in sensitive epithelial-derived xenograft models lacking ATRX alterations (27). However, LMS33 tumor growth was not significantly decreased over the 21-day treatment period, indicating the lack of activity of this drug as monotherapy in this model.

Figure 3. Failure of ATR and IGF1R inhibitors to perturb LMS PDX growth.

A, Expression in FPKM of ATRX across the indicated PDX lines. Data were analyzed by one-way ANOVA with Dunnet multiple comparisons test (compared to LMS33; ***,P<0.001). B, Tumor volume of LMS33 PDX in response to treatment with vehicle (n = 7) or ATR inhibitor berzosertib (60 mg/kg by oral gavage, 5 days per week; n = 5). C, Expression in FPKM of IGF1R across the indicated PDX lines. Data were analyzed by one-way ANOVA with Dunnet multiple comparisons test (compared to LMS20; ***,P<0.001). D, ISH of IGF1R in LMS19 and LMS20 parental tumors and PDX; scale bar = 5 μm. E, Tumor volume of LMS20 PDX in response to treatment with vehicle (n = 3) or IGF1R inhibitor linsitinib (50 mg/kg by oral gavage, 5 days per week; n = 4).

We have previously shown elevated IGF1R expression in a subset of cLMS tumors, which was predictive of worse clinical outcomes (4). IGF1R-directed combination therapy has previously been evaluated in soft tissue sarcoma including LMS, though the lack of robust IGF1R immunohistochemistry represents a challenge to deploying these therapies in a targeted manner (6,28). Among LMS PDX, LMS20 was found to have the highest IGF1R expression by RNA-seq (Fig. 3C). As an alternative to immunohistochemistry, we found that in situ hybridization (ISH) was successfully able to detect IGF1R expression in LMS20 compared to LMS19, with detection of IGF1R transcript in PDX as well as archival parental tumor samples (Fig. 3D). To assess the response of an LMS PDX to IGF1R inhibition, we treated LMS20 with linsitinib over a 21-day period at 50 mg/kg (29). However, linsitinib treatment failed to decrease LMS20 tumor growth (Fig. 3E). Taken together, these data show the feasibility of preclinical testing of molecularly targeted therapies in LMS, and that neither berzosertib nor linsitinib show significant preclinical activity as monotherapy in molecularly defined and target-enriched PDX. These results highlight the limitations of selecting therapeutic candidates arising from genomic or transcriptional profiling, and indicate that functional assessments of therapeutic activity in preclinical models may represent an optimal strategy to define lead clinical compounds.

High-throughput dynamic BH3 profiling identifies susceptibility to transcriptional CDK inhibitors in LMS.

High-throughput genetic and chemical screening platforms have significantly advanced the understanding of cancer biology and disease-specific vulnerabilities, though are commonly restricted to immortalized cell lines. To facilitate the identification of therapeutic vulnerabilities in LMS, we utilized high-throughput dynamic BH3 profiling (HT-DBP) in LMS PDX and patient-derived samples. HT-DBP tests whether a candidate molecule increases the “apoptotic priming” of short-term ex vivo cultures. Specifically, tumor cells are cultured for 4 to 24 hours with a drug of interest and are subsequently exposed to standardized concentrations of synthetic peptides derived from proapoptotic Bcl-2 family proteins and combined with a drug of interest to test the candidate molecule’s ability to promote apoptosis (22,30). We assayed 35 compounds with diverse targets and mechanisms of action (Table S2) for apoptotic priming in four LMS PDX and two patient-derived surgical samples. While no compound caused apoptotic priming in all PDX and patient-derived samples, among the library tested several transcriptional CDK inhibitors were found to increase priming in a majority of samples (Fig. 4A). We further evaluated the ability of CDK inhibitors of diverse selectivity to promote cytochrome c release in LMS33 ex vivo, with navitoclax and doxorubicin used as positive controls. LMS33 was chosen among sensitive PDX as it had the highest delta priming score. Compared to palbociclib, which is selective for cell cycle regulatory kinases CDK4 and CDK6, the pan-CDK inhibitor flavopiridol and transcriptional CDK inhibitor THZ1 caused cellular toxicity comparable to navitoclax or doxorubicin (Fig. 4B).

To confirm pan-CDK inhibitor activity in vivo, we treated LMS33 PDX bearing mice with the pan-CDK inhibitor flavopiridol at 7.5 mg/kg 5 days per week by intraperitoneal (i.p.) injection. Compared to vehicle-treated mice, flavopiridol significantly decreased tumor growth over a 21-day treatment period (Fig. 4C), without observed toxicity or weight loss (Fig. S5A). Histologic evaluation of LMS33 tumors treated for 3 days with flavopiridol or vehicle control showed a significant decrease in Ki-67-positive cells, with a corresponding increase in the number of cells staining positive for cleaved caspase-3 (CC3) as an indicator of apoptosis (Fig. 4D–E). Taken together, these results demonstrate a high-throughput approach to identifying therapeutic vulnerabilities in LMS and distinguish transcriptional CDKs as therapeutic targets in this disease.

CDK7 inhibition antagonizes the E2F-driven oncogenic program in LMS.

Genetic, transcriptional and histologic data suggest near universal loss of RB1 in LMS (2,3), which leads to dysregulation of E2F1 and activation of cancer-associated growth programs (10). We hypothesized that the observed toxicity of transcriptional CDK inhibitors in LMS may arise from antagonism of an E2F-driven program. To further explore the mechanisms of toxicity of transcriptional CDK inhibitors in LMS, we treated mice bearing two LMS PDX lines with YKL-5–124, a selective and covalent CDK7 inhibitor (31), at 2.5 mg/kg 5 days per week by i.p. injection. Compared to control mice, those treated with YKL-5–124 showed significant reductions in tumor growth in both LMS4 and LMS33, with regression of tumors from baseline in LMS33 (Fig. 5A–B). Treated mice showed no overt toxicity or weight loss (Fig. S5B–C).

We performed a 5-day treatment of LMS33 with YKL-5–124 or vehicle control, with tumors harvested 4 hours after the last dose and analyzed by RNA-seq, to study the acute transcriptional effects of CDK7 inhibition on LMS33. Compared to vehicle, YKL-5–124 treated tumors showed significant reductions in E2F-associated transcriptional activity and cell cycle progression using GSEA, with treated tumors showing upregulation of gene sets associated with DNA repair and reactive oxygen species (Fig. 5C–F). The Epithelial-Mesenchymal Transition gene set was also significantly downregulated by YKL-5–124 treatment (Fig. 5D). YKL-5–124 treatment significantly reduced expression of several key transcriptional and cell cycle regulatory targets, including E2F1 and its binding partner TFDP1, MYC, PLK1, AURKB and TOP2A (Fig. 5G). In contrast to E2F1 and in support of the induction of transcriptional reprogramming, YKL-5–124 treatment led to the significant upregulation of other transcriptional regulators, including FAM89B, HES6 and TAF6, with significant increases in genes associated with DNA repair and stress response (Fig. 5H). Taken together, these transcriptional changes demonstrate that CDK7 inhibition antagonizes the E2F pathway in LMS, ultimately disrupting cell cycle progression and tumor growth. The loss of the EMT signature further suggests a change in global transcriptional state following CDK7 inhibition, potentially reflecting evolving tumor fibrosis or cellular differentiation (32).

DISCUSSION

Despite multiple large-scale sequencing-based efforts to characterize LMS, deciphering active and targetable oncogenic programs driving tumor biology has remained challenging. Research efforts have been limited by the lack of available and well characterized preclinical models to functionally validate candidate therapies. Here, we describe the development of a cohort of LMS PDX models that faithfully recapitulate fundamental aspects of their tumors of origin. The engraftment rate of LMS PDX was 35% (17 of 49), with a higher percentage of engrafting tumors arising from ULMS. Successful engraftment was associated with higher mitotic rate and high-grade tumors. Engraftment was not predicted by prior oncologic therapy, except for pre-operative radiation which was associated with lack of engraftment. In contrast to pre-operative radiation, tumors that remotely received radiation (at least 22 months prior to attempted engraftment) and recurred within or metastasized from a prior radiation field were able to engraft. Alternative implantation sites, different murine host strains or use of hormonal supplementation may conceivably improve the engraftment rate for LMS. A subset of ULMS has high ER expression (33), and uniform use of estrogen pellets may improve xenograft growth, as has been observed in breast cancer PDX models (34). That our two engrafted LMS PDX with ER expression propagated slowly or required estrogen supplementation for growth may indicate the dependency of a subset of ULMS on hormonal signaling, as has been suggested for a subset of ULMS with high expression of ER treated with aromatase inhibitors in clinical studies (35).

LMS PDX were found to retain the histologic architecture of their primary tumors across successive passages. Global copy number alterations, which are characteristic of LMS and heterogeneous between different tumors, were similarly preserved across PDX passages, and ctDNA could be detected in the plasma of tumor bearing mice and may be a useful surrogate for treatment assessments in select PDX lines. Further endorsing the validity of LMS PDX models, the transcriptional landscape of successive generations of PDX co-clustered with parental tumors, with the major differences between parental tumors and PDX arising from the host microenvironment (e.g. endothelia and immune cells). Previously, PDX from multiple cancer subtypes were reported to develop early mouse-specific copy number alterations evolutionarily divergent from their tumors of origin (36). In contrast to these findings, we observed preservation of aneuploidy and transcriptional programs across successive generations of LMS PDX. Our data further reinforce the divergence of available LMS cell lines from patient tumors and PDX, highlighting the limitations of preclinical models in this disease.

Our efforts to generate LMS PDX were able to produce relevant models of disease subtypes that warrant additional investigation. Li-Fraumeni syndrome, characterized by germline mutation in TP53, bears an increased risk for LMS (37), and one such tumor was able to successfully engraft. When characterizing an engrafted tumor bearing a histologic diagnosis of LMS, genomic profiling identified the absence of characteristic tumor suppressor loss, low TMB and the presence of an ALK rearrangement. ALK rearrangements have been previously described as rarely occurring oncogenic events in LMS and other uterine mesenchymal tumors (38,39). Though ALK inhibition has shown activity in ALK-translocated sarcomas (40), the ALK translocation that gave rise to LMS45 was identified retrospectively only after the patient had succumbed to their disease. While genomic testing is not currently a standard of care in LMS, this finding suggests the clinical utility of such profiling in identifying rare and targetable alterations.

Based upon the recurrent loss of ATRX, disruption of the ALT pathway has been a proposed vulnerability in LMS. We tested this hypothesis in LMS33, which has bi-allelic ATRX inactivation and undetectable mRNA expression, using the ATR inhibitor berzosertib. While a 21-day treatment showed no significant change in tumor growth, berzosertib showed a trend towards reduced tumor volume at later timepoints. Additional evaluation of more potent ATR inhibitors, or the use of combination therapies (41), may be warranted to further explore the utility of ATR inhibition in LMS. IGF1R inhibition in the subset of LMS tumors that express this kinase has also been proposed as a therapeutic strategy (6), though we found no activity of the IGF1R inhibitor linsitinib as a single agent in an appropriately matched LMS PDX. These results demonstrate the utility of these models in the preclinical assessment of hypothesis-driven targeted therapies, and suggest that ATR and IGF1R inhibition with the evaluated compounds may not be clinically active as monotherapies in LMS.

To investigate LMS tumors and PDX for therapeutic vulnerabilities in an unbiased and high-throughput fashion, we utilized HT-DBP. HT-DBP requires only 4–24 hours of ex vivo culture, which minimizes the opportunity for the accumulation of molecular changes associated with prolonged in vitro propagation. Moreover, HT-DBP can identify compounds that prime cells for apoptosis, but do not induce frank cell death, enhancing the ability to identify therapeutic leads which may represent ideal candidates for combination therapy. Drugs that are able to increase the apoptotic priming of tumor cells are superior candidates for therapeutic interrogation in vivo (22,30). Through profiling LMS patient samples and PDX by HT-DBP, we identified transcriptional CDKs as a therapeutic target. Both the pan-CDK inhibitor flavopiridol and the selective CDK7 inhibitor YKL-5–124 decreased LMS tumor growth, and histologic evaluation of LMS PDX treated with flavopiridol showed decreased proliferation and increased apoptosis compared to control. Through transcriptional profiling of YKL-5–124-treated PDX, we determined that these drugs directly decreased levels of E2F1 and collaborative transcriptional and cell cycle regulators while globally decreasing E2F-driven transcription. YKL-5–124 treatment also produced changes in the EMT-related signature and activation of other transcriptional regulators, suggesting a global change in cellular transcriptional state. Loss of direct RB-mediated regulation has been shown to expand and diversify the genomic regions bound by E2F1, broadly modifying gene regulation and associated transcriptional output to permit oncogenesis (42). As RB1 is lost in a majority of LMS, targeting E2F1-driven gene expression with transcriptional CDK inhibitors may represent a therapeutic vulnerability in this disease. Selective transcriptional CDK inhibitors are now entering clinical trials (NCT04247126, NCT03263637) and may embody optimal therapeutic strategies for targeting RB1 loss and E2F1 dysregulation in LMS. Given the heterogeneity of LMS, additional screening efforts using HT-DBP and other methods in patient samples and PDX are needed to identify additional therapeutic vulnerabilities in LMS across the spectrum of disease. With the frequently modest benefit offered by current cytotoxic chemotherapy, these and related efforts are urgently needed to develop biologically based and hypothesis driven strategies for clinical translation in LMS.

STATEMENT OF TRANSLATIONAL RELEVANCE.

Leiomyosarcoma (LMS) is a mesenchymal malignancy harboring recurrent loss of tumor suppressor genes and a poorly defined oncogenic program. Empiric cytotoxic chemotherapy is the standard systemic treatment for this disease, and while more targeted and effective therapies are urgently needed insights into therapeutic vulnerabilities in LMS have been limited by available preclinical models. We generated a panel of LMS patient derived xenografts which preserve the histology, genomic copy number alterations and transcriptional landscape of their parental tumors. We utilized high-throughput dynamic BH3 profiling to screen tumor and xenograft samples ex vivo, identifying therapeutic activity of transcriptional CDK inhibitors in LMS. These drugs inhibit tumor growth and antagonize an oncogenic E2F-driven transcriptional program commonly present in RB1-deficient LMS. These findings describe a strategy to develop, validate and discover novel therapies in rare cancer subtypes and justify clinical evaluation of transcriptional CDK inhibitors in LMS patients.