Proteasome Inhibition Sensitizes Liposarcoma to MDM2 Inhibition with Nutlin-3 by Activating the ATF4/CHOP Stress Response Pathway

Targeting the proteasome in combination with MDM2 inhibition activates the ATF4/CHOP stress response axis to induce apoptosis in liposarcoma, providing a potential therapeutic approach for the most common soft–tissue sarcoma.

Abstract

Liposarcoma is the most commonly occurring soft-tissue sarcoma and is frequently characterized by amplification of chromosome region 12q13–15 harboring the oncogenes MDM2 and CDK4. This unique genetic profile makes liposarcoma an attractive candidate for targeted therapeutics. While CDK4/6 inhibitors are currently employed for treatment of several cancers, MDM2 inhibitors have yet to attain clinical approval. Here, we report the molecular characterization of the response of liposarcoma to the MDM2 inhibitor nutlin-3. Treatment with nutlin-3 led to upregulation of two nodes of the proteostasis network: the ribosome and the proteasome. CRISPR/Cas9 was used to perform a genome-wide loss of function screen that identified PSMD9, which encodes a proteasome subunit, as a regulator of response to nutlin-3. Accordingly, pharmacologic studies with a panel of proteasome inhibitors revealed strong combinatorial induction of apoptosis with nutlin-3. Mechanistic studies identified activation of the ATF4/CHOP stress response axis as a potential node of interaction between nutlin-3 and the proteasome inhibitor carfilzomib. CRISPR/Cas9 gene editing experiments confirmed that ATF4, CHOP, and the BH3-only protein, NOXA, are all required for nutlin-3 and carfilzomib-induced apoptosis. Furthermore, activation of the unfolded protein response using tunicamycin and thapsigargin was sufficient to activate the ATF4/CHOP stress response axis and sensitize to nutlin-3. Finally, cell line and patient-derived xenograft models demonstrated combinatorial effects of treatment with idasanutlin and carfilzomib on liposarcoma growth in vivo. Together, these data indicate that targeting of the proteasome could improve the efficacy of MDM2 inhibitors in liposarcoma.

Significance:

Targeting the proteasome in combination with MDM2 inhibition activates the ATF4/CHOP stress response axis to induce apoptosis in liposarcoma, providing a potential therapeutic approach for the most common soft-tissue sarcoma.

Introduction

Liposarcomas are the most commonly occurring sarcoma type representing up to 20% of cases (1, 2). Clinically, liposarcomas are classified into four subtypes based on differentiation status and genetic characteristics: well-differentiated (WDLPS), dedifferentiated (DDLPS), myxoid/round cell, and pleomorphic liposarcomas (3, 4). Notably, WDLPS and DDLPS make up more than 50% of all liposarcomas and are the least responsive to currently available treatments. Surgical resection is the current standard of care for both WDLPS and DDLPS; however, recurrence is common (5). Therefore, the development of novel treatment strategies for liposarcomas could benefit many patients.

A hallmark of tumorigenesis across cancer types is the inactivation of the p53 pathway via numerous mechanisms, including mutations in TP53, MDM2 copy-number gain, and CDKN2A (ARF) mutations (6, 7). Primarily, WDLPS and DDLPS are characterized by the presence of large marker chromosomes containing amplification of chromosome region 12q13–15, resulting in massive MDM2 copy-number gain (8–10). Consistent with this mechanism of inactivation of the p53 pathway, liposarcoma tumors retain wild-type (WT) p53 activity in 95% of cases (9, 10). These characteristics make liposarcomas an attractive target for p53-based therapeutic approaches, such as MDM2 inhibitors. Extensive studies over the past ∼15 years have revealed that MDM2 inhibitors, such as nutlin-3, are highly effective at inducing the p53 transcriptional program, however, the ultimate cellular response elicited, such as cell-cycle arrest or apoptosis, is highly variable (11–14). Consistent with these findings, several clinical trials for single-agent MDM2 inhibitors in solid tumors have failed, with clinical cohorts showing limited response and undesirable side effects such as cytopenia (15). Furthermore, several small-scale trials for MDM2 inhibitors in liposarcomas have shown only modest efficacy, with little ability to induce disease regression and rapid emergence of resistance via selection for p53 mutations (15, 16). Taken together, these data indicate that the use of MDM2 inhibitors as single agent therapies, even for promising cancer types with favorable molecular profiles, is likely to be limited.

Identification of rational drug combinations represents one strategy to expand the clinical utility of numerous small molecules, including MDM2 inhibitors (17–19). The discovery of druggable targets that interact with the MDM2 inhibitor-induced p53 response to drive cell death could serve as the foundation for the development of combinatorial therapies in liposarcomas. With this in mind, we set out to identify pathways that sensitize liposarcomas cells to MDM2 inhibition which could reveal druggable targets for combinatorial treatments in liposarcomas.

Our initial characterization of the molecular response of WDLPS and DDPLS cell lines to the MDM2 inhibitor nutlin-3 revealed a previously unreported perturbation in the proteostasis network via wholesale upregulation of both ribosome and proteasome subunits at the RNA level. We next performed a genome-wide CRISPR-Cas9 screen for nutlin-3 sensitizing genes (NSG) which identified numerous promising candidates, including PSMD9 which encodes a component of the 26S proteasome. We validated PSMD9 specifically, and the proteasome more generally, as sensitizing to nutlin-3 using genetic and pharmacologic approaches. Mechanistic investigations revealed that the unfolded protein response (UPR), specifically the activation of the ATF4/CHOP signaling axis, was necessary for the interaction between nutlin-3 and proteasome inhibitors and the activation of the UPR was sufficient to sensitize liposarcoma cells to nutlin-3–induced cell death. Finally, using xenograft models of DDLPS, we demonstrated combinatorial efficacy of idasanutlin and carfilzomib in vivo.

Materials and Methods

Cell culture

The 94T778 cells were acquired from ATCC (catalog no. CRL-3044, RRID:CVCL_U613) and Lipo-246 cells were provided by Dr. Keila Torres and the University of Texas M.D. Anderson Cancer Center (RRID:CVCL_U613). All cells were cultured in DMEM/F12 medium (Thermo Fisher Scientific) supplemented in 10% EquaFETAL (Atlas Biologicals) and antibiotic and antimycotic (Anti-Anti, Thermo Fisher Scientific). HEK293FT cells were cultured in DMEM (Thermo Fisher Scientific, RRID:CVCL_6911) and were supplemented with 10% FBS (Peak Serum) and antibiotic and antimycotic (Anti-Anti, Thermo Fisher Scientific). Cells were maintained at 37°C in a humidified atmosphere with 5% CO₂. Cell lines were authenticated with STR profiling and routinely monitored for Mycoplasma via PCR (Supplementary File 1).

Cytogenetics

Liposarcoma cell cultures were treated with Colcemid for 2 hours and harvested the next day. Cells were detached using Trypsin-EDTA, incubated in 0.075 mol/L KCl for 2 hours, and fixed in 3:1 methanol:glacial acetic acid solution. Fixed cells were dropped onto precleaned glass slides. GTL banding was performed with standard GTL method including incubations at 85°C for 90 minutes, 25 seconds of trypsin digestion, and 2 minutes of Leishman's staining. Metaphase spreads were imaged and karyotyped using the BandView software (Applied Spectral Imaging Inc). Ploidy (chromosome counting) was investigated in at least 30 spreads per specimen.

For FISH, Leishman's stain on the G-banded slides was completely removed by washing in CitriSolv twice for 5 minutes each, followed by dehydration and fixation in 3:1 methanol:glacial acetic acid overnight. Cells were then hybridized with the CytoCell DNA probe set LPS016 which contains probes for MDM2 (mapped at 12q15, labeled in red) and centromeric control, D12Z1 (mapped at 12p11.1-q11.1, labeled in green). Metaphases previously karyotyped were examined for identification of chromosomes with homology to the FISH probes. Chromosome classification and nomenclature followed the ISCN (2016) guidelines.

Cell line mutational analysis

Targeted DNA library preparation was performed via the Illumina TruSight Tumor kit per the manufacturer's instructions (with minor modifications; Illumina). Libraries were sequenced on the Illumina MiSeq platform for a targeted depth of no less than 500x for any individual amplicon. A custom-built bioinformatics pipeline utilizing GSNAP for sequence alignment (20) and FreeBayes for variant calling (21) was employed for data analysis. All genomic regions were verified to be covered by at least 500 sequencing reads and identified variants were manually inspected using Integrative Genomics Viewer (Broad Institute).

Transcriptome library preparation and sequencing

RNA quality was assessed using an Agilent 2200 TapeStation and quantified by Qubit (Life Technologies). Poly(A)+ RNA enrichment, and strand-specific library preparation were carried out using a Universal Plus mRNA-Seq with NuQuant (Nugen/Tecan). Paired-end 150 bp sequencing was carried out on an Illumina NovaSeq 6000 instrument by the Genomics Shared Resource at the University of Colorado Anschutz Medical Campus.

Transcriptome analysis

Trimming and filtering of low-quality reads was performed using bbduk from BBTools (v37.99, RRID:SCR_016969; ref. 22) and fastq-mcf from ea-utils (v1.05, https://expressionanalysis.github.io/ea-utils/, RRID:SCR_005553). Alignment to the human reference genome (GRCh38) was carried out using HISAT2 (v2.1.0, RRID:SCR_015530; ref. 23) in paired, spliced-alignment mode with a GRCh38 index with a Gencode v33 annotation GTF, and alignments were sorted and filtered for mapping quality (MAPQ > 10) using Samtools (v1.5; ref. 24). Gene-level count data were quantified using HTSeq-count (v0.6.1, RRID:SCR_011867; ref. 25) with the following options (--stranded=reverse –minaqual=10 – type=exon --mode=intersection-nonempty) using a Gencode v33 GTF annotation file. Preprocessing, statistical analysis, and plot generation for all datasets was carried out using R (R 4.0.1, RRID:SCR_001905 / Rstudio 1.3.959, RRID:SCR_000432 / Bioconductor v 3.11, RRID:_006442; refs. 26–28). Reads were demultiplexed and converted to fastq format using bcl2fastq (bcl2fastq v2.20.0.422, RRID:SCR_015058). Data quality was assessed using FASTQC (v0.11.5; https://www.bioinformatics.babraham.ac.uk/projects/fastqc/, RRID:SCR_014583) and FastQ Screen (v0.11.0, https://www.bioinformatics.babraham.ac.uk/projects/fastq_screen/, RRID:SCR_000141). Differential gene expression was evaluated using DESeq2 (version 1.28.1, RRID:SCR_000154; ref. 29) in R, using q < 0.1 (10% FDR) as the threshold for differentially expressed genes (DEG).

CRISPR screen

Lentiviral particles for the Human GeCKOv2 A and B libraries (30) were produced by transfection of HEK293FT cells (12.5×10⁶ per library across 15 cm² plates) with 16.25 μg of pooled GeCKOv2 A and B plasmid libraries, 6.9 μg of pdelta8.9, 2.03 μg of pCMV-VSVG (RRID:Addgene_8454), and 80 μg polyethyleneimine (Polyscience). Following overnight incubation, the media was replaced, and HEK293FT cells were cultured for 48 hours. Lentivirus-containing supernatant was then collected, clarified at 300 g for 5 minutes, and sterilized by passing through a 0.45 μm cellulose acetate filter (VWR International).

The 94T778 cells (12.5×10⁶ cells each to maintain ∼100× library coverage) were transduced by adding lentiviral supernatant along with 8 μg/mL polybrene. Following overnight incubation, the media was replaced and 1 μg/mL puromycin was added to select for cells with stable integration of the GeCKO constructs. Cells were then cultured for 3 passages (∼10 days) to allow for clearance of cells with knockout of essential genes. The resulting cell population was plated at ∼100x coverage across 3× T225 flasks in DMEM/F12 supplemented with 0.05% DMSO (vehicle control) or 10 μmol/L nutlin-3R (Cayman Chemical #10004372, CAS Number 548472–68–0) for 72 hours. Each population was then allowed a recovery period of 3 to 4 population doublings (∼1 week). Genomic DNA was then harvested with a DNeasy Blood and Tissue kit (Qiagen).

Sequencing library preparation

A nested PCR strategy was used to prepare barcoded guide RNA (gRNA) cassette libraries suitable for Illumina sequencing. PCR1 used genomic DNA as a template, producing a 445 bp product that was used as the template for PCR2 which adds Illumina sequencing adaptors and barcode sequences, producing a 195 bp product. To maintain GeCKO library coverage without high genomic DNA concentrations inhibiting amplification, PCR1 was performed as eight individual 50 μL reactions: 500 ng genomic DNA, 10 μL 5x Phusion HF buffer, 2.5 μL PCR1-fwd primer (10 μmol/L), 2.5 μL PCR1-rev primer (10 μmol/L), 1 μL dNTPs (10 mmol/L each), 1 μL Phusion polymerase (Thermo Fisher Scientific), nuclease-free H₂O to 50 μL. PCR1 cycling conditions: 1 cycle of 98°C for 5 minutes, 19 cycles of 98°C for 30 seconds, 55°C for 30 seconds, 72°C for 30 seconds, and 1 cycle of 72°C for 7 minutes. The 8 reactions per sample were then pooled, cleaned up using a MinElute PCR Purification Kit (Qiagen), and quantified. PCR2 was performed as two individual 50 μL reactions/sample. Each PCR2 reaction contained 75 ng of PCR1 amplicon, 10 μL 5x Phusion HF buffer, 2.5 μL PCR2-fwd primer (5 μmol/L), 2.5 μL PCR2-rev1–6 barcoded primer (5 μmol/L), 1 μL dNTPs (10 mmol/L each), 1 μL Phusion polymerase (Thermo Fisher Scientific), nuclease-free H₂O to 50 μL. PCR2 cycling conditions: 1 cycle of 98°C for 5 minutes, 7 cycles of 98°C for 30 seconds, 62.9°C for 30 seconds, 72°C for 30 seconds, and 1 cycle of 72°C for 7 minutes. Duplicate PCR2 reactions were then pooled and resolved on a 1.5% TAE Agarose gel. The 195 bp PCR2 band was excised and purified using a QIAquick PCR Purification Kit (Qiagen) with a MinElute column and quantified. Library quality was assessed using a High-Sensitivity DNA Kit (Agilent) on a 2100 Bioanalyzer System (Agilent). Libraries were sequenced on an Illumina HiSeq 2500 system at the University of Colorado Cancer Center Genomics and Microarray Core facility. Primers used for PCR1 and PCR2 are listed in Supplementary File 1.

CRISPR screen bioinformatics

Trimming and filtering of low-quality reads was performed using fastx_trimmer and fastq_quality_filter from the FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/index.html, RRID:SCR_05534). Alignment to GeCKO reference library was carried out using Bowtie2 (RRID:SCR_016368) and alignments were sorted, filtered for mapping quality (MAPQ > 10), and counted using Samtools (v1.5, RRID:SCR_002105; ref. 24). Gene candidates were ranked by thresholding each individual gRNA by the criteria of > 8 CPM with a log₂[fold change (FC)] < 0.5 for NSG, or log₂FC > 3 for genes that confer resistance. For each gene, net depletion of at least three gRNAs was required to be considered NSG.

Ribonucleoprotein transfections and screening

1×10⁵ 94T778 cells were seeded into each well of a 6-well plate and allowed to attach overnight. Ribonucleoprotein (RNP) transfections were performed using the crRNA::tracrRNA duplex following the manufacturer's recommended protocol (IDT). In brief, individual crRNAs (Supplementary File 1) were annealed to tracrRNA in equal molar amounts (1 μmol/L each). Duplexed gRNAs were then complexed with Cas9 protein in OptiMEM (Thermo Fisher Scientific) and packaged in Lipofectamine3000 (Thermo Fisher Scientific). The transfection mix was added dropwise to overnight cell cultures and allowed to incubate for 16 hours. The next day, transfection media was aspirated and replaced with fresh media. Cells were expanded into 10 cm² plates prior to single cell sorting of individual clones into 96-well plates then expanded until wells were > 50% confluent. Then, one third of the cells were passaged (replicated) into a DNA screening plate and the remaining two thirds were frozen by addition of a 2x freezing media solution (complete culture media with 20% FBS and 20% DMSO). To replicate the 96-well plate into a DNA screening plate, wells are washed 1× with 150 μL PBS, then trypsinized with 25 μL of 0.25% Trypsin-EDTA. Trypsin was quenched by mixing 50 μL of complete media, then 1/3 passaged by taking 25 μL of cell suspension and adding to a new plate already containing 100 μL of complete media (125 μL final o/n culture). Next day, overnight culture/residual trypsin containing media is aspirated and replaced with 100 μL complete media.

When the wells within the 96-well DNA screening plate reached > 80% confluency the wells were aspirated and washed with 1× PBS. 50 μL of lysis buffer (10 mmol/L Tris-HCL pH 7.5, 10 mmol/L EDTA, 10 mmol/L sodium chloride, 0.5% Sarkosyl) + Proteinase K at 100 ng/mL was added to each well and incubated overnight at 37°C. A 96-well rubber lid seal was used for each plate to collect evaporation, which was centrifuged back into each corresponding well the following morning. Salt precipitation of DNA was then performed by adding 100 μL of 0.075 mmol/L sodium chloride in ice-cold 100% ethanol to each well and allowing to sit for 10 minutes at room temperature. Precipitated DNA was then pelleted by centrifugation at 2,000 g for 15 minutes at 4°C. Pellets were then washed twice with 150 μL of 70% ethanol, then allowed to air dry at room temperature for 15 minutes. DNA was then resuspended in 60 μL of water. For subsequent PCR screening, 2 μL of this DNA solution was used as template in a 50 μL PCR reaction.

Annexin V assay

75,000 cells were seeded into each well of a 6-well plate and allowed to attach overnight. Next day, media was aspirated and replaced with media containing DMSO, single-agent compounds or combinational treatments. Treatments were left undisturbed for 72 hours at 37°C in a humidified atmosphere with 5% CO₂. At 72 hours posttreatment, media and a PBS wash were collected and 250 μL of 0.25% Trypsin-EDTA was used to detached remaining cells. The media/PBS was used to quench and collect trypsinized cells. Cells were centrifuged at 300 × g for 3 minutes, then the media was aspirated. 1 mL PBS was used to resuspend each pellet, followed by centrifugation and aspiration. Propidium iodide and Annexin-FITC antibody (Thermo Fisher, RRID:AB_2935694) were diluted in binding buffer (10 mmol/L HEPES, 140 mmol/L sodium chloride, 2.5 mmol/L calcium chloride, pH 7.4) prior to use. Pellets were resuspended in 10 μL of assay buffer (for each reaction, 6.5 μL of binding buffer + 1.5 μL of 1 mg/mL propidium iodide + 2 μL Annexin-FITC) and incubated at room temperature in the dark for 20 minutes. 100 to 250 μL of binding buffer was then added to each treatment to dilute cells prior to measurement. The propidium iodide and Annexin-FITC staining was measured using Accuri C6 Flow cytometer. Both early and late Annexin-FITC positive populations were summed.

96-well combinatorial synergy assay

200 μL of PBS was added to each outer well of a 96-well plate to prevent evaporation. Then, 3.5×10³ 94T778 cells or 5.0×10³ Lipo-246 cells were seeded into each interior well and allowed to attach overnight. Next day, media was aspirated and replaced with treatment media that was prepared as follows: (i) Preparation of a nutlin-3 dilution series down the rows at twice the final concentrations, with the final row containing fresh media only. (ii) Preparation of a proteasome inhibitor dilution series across the columns at twice the final concentrations, with the final column containing fresh media only. (iii) Combining the two dilution series plates together at a 1:1 volume to yield the working concentrations of each compound. 100 μL of treatment media was then added to the experimental plate. Cells were cultured in treatment media for 72 hours followed by the SRB assay to measure total protein content within each well (both fixed cells and cell corpses; ref. 31). Synergy mapping was performed by importing the 96-well SRB absorbance reads in a matrix format to the SynergyFinder 2.0 (RRID:SCR_019318).

Clonogenic growth assay

A total of 5×10² 94T778 cells or 1×10³ Lipo-246 cells were seeded into four wells of a 6-well plate and allowed to attach overnight. The following day, treatments were added and remined for 72 hours. Media was then replaced every 72 hours for 16 days (94T778) or 25 days (Lipo-246). Cells were then washed once with 2 mL PBS, then stained by adding 750 μL of crystal violet (0.5% crystal violet in 20% methanol/H₂O) to each well for 30 minutes at room temperature. Each well was then washed five times with 2 mL PBS then once with DI H₂O. Images were taken with the ImageQuant800 digital camera system (Amersham).

siRNA knockdown

A total of 3.75×10⁴ 94T778 cells or 7.5×10⁴ Lipo-246 cells were seeded to each well of a 6-well plate. The next day 5 μL of 20 μmol/L stock siRNAs (Invitrogen) were complexed with 15 μL Lipofectamine 3000 in 600 μL OptiMEM (Life Technologies). The transfection mix was vortex and allowed to complex at room temperature for 15 minutes. Then, 120 μL was added to each well for a final concentration of 10 nmol/L. siRNAs used as follows; nontargeting medium GC content (catalog no. 462001), ATF4-targeting (catalog nos. s1703 and s1704), DDIT3-targeting (catalog nos. s3996 and s3996) TP53-targeting (catalog nos. VHS40366 and VHS40367). 24 hours following siRNA transfection, drug treatments were added without media change.

Western blotting

Sample preparation, quantitation, and Western blotting were carried out as previously described (32). Detection was completed by chemiluminescence using SuperSignal West Pico PLUS (Thermo Fisher Scientific), and images were captured with an ImageQuant800 digital camera system (Amersham). Antibodies used are listed in Supplementary File 1. Densitometry was performed with ImageJ (RRID:SCR_003070).

Animal work

All in vivo experiments were approved by the Institutional Animal Care and Use Committee at the University of Colorado Anschutz Medical Campus (IACUC protocol 00432).

Cell line–derived xenografts

A total of 5×10⁶ to 7×10⁶ Lipo-246 cells suspended in Hank's Balanced Salt Solution:Matrigel (1:1) were subcutaneously injected into the right flank of male and female NOD.Cg-Prkdc<scid>Il2rg<tm1Wjl>/SzJ mice purchased from The Jackson Laboratory (Strain ID# 005557, RRID:IMSR_JAX:005557). Mice were randomized into treatment groups. Idasanutlin, a nutlin-3 derivative optimized for in vivo use, was administered via oral gavage at a final dose of 37.5 mg/kg, was prepared by suspending the compound in a proprietary solution obtained from Roche at a final concentration 7.5 mg/mL. Idasanutlin treatment occurred on days 1 to 5 of each week for up to 3 weeks (21 days). Carfilzomib was initially dissolved in DMSO at 100 mg/mL and prepared for tail vein injection at 1 mg/mL by adding to a solution of 10% captisol and 10 mmol/L sodium citrate was administered via tail vein injection at a final dose of 3 mg/kg on days 1 and 4 of each week for up to 3 weeks. Combination of the two drugs were administered on the same schedules for the same duration. Idasanutlin-treated animals received tail-vein injections of the carfilzomib vehicle (10% captisol, 10 mmol/L sodium citrate, and 1% DMSO) on days 1 and 4 and carfilzomib-treated animals received idasanutlin vehicle (proprietary) via oral gavage on days 1 to 5. All animals were weighed on days 1 to 5 of each week prior to administration of drugs to determine gavage and tail vein injection volumes, and to check for humane end point weight status. Tumors were measured on days 1, 3, and 5 of each week. Animals that lost more than 15% of body weight or with tumors that exceeded 2,000 mm³ reached humane endpoints and were euthanized.

Patient-derived xenografts

A primary patient-derived xenograft (PDX; CUSARC27) was generated by implantation of freshly obtained surgical tissue from a patient with DDLPS into NOD/SCID gamma) mice. Written informed consent was obtained from the patient following an Institutional Review Board–approved tumor banking protocol, according to principles of the Belmont Report, and in accordance with the Common Rule, Institutional Review Boards (COMIRB 19–0708), and state and federal HIPAA regulations. The PDX was passaged twice in NOD/SCID gamma mice prior to use for these experiments and the F2 specimen was compared to the original human sample by bone and soft tissue pathologist to confirm etiology. Treatments were performed using the same dosing schedule as the cell line–derived xenograft (CDX) experiments.

Complete blood count collection and analysis

A cardiac puncture was performed to collect approximately 0.5 mL of blood from each mouse. Mice were placed in dorsal recumbence (in researcher's hand) with their dorsal skin pinched by the researcher. A 26G 3/8″ needle with a 1 mL syringe was inserted bevel up just below the ribs on the midline at a 20-degree angle relative to the researcher's hand. Once a flash of blood was witnessed the plunger is slowly drawn on the syringe. After blood collection the needle is withdrawn, and the blood is added to a lithium heparin tube (Sarstedt LH 41.1393.105) for complete blood count (CBC) analysis. CBC analysis was completed using the HemaTrue (Heska) according to the manufactures protocol.

Data availability

The CRISPR screen and transcriptome data generated by this study are publicly available in Gene Expression Omnibus (GEO) at GSE214891. All other raw data are available upon request from the corresponding author.

Results

Liposarcoma cells upregulate transcripts encoding ribosome and proteasome subunits in response to MDM2 inhibition

We first characterized the cytogenetics of two liposarcoma cell lines using standard methods. MDM2 amplification status was determined using FISH which confirmed widespread amplification of the 12q13–15 region in both the 94T778 (WDLPS) and Lipo-246 (DDLPS) cell lines (Fig. 1A). Quantification of the ratio of the MDM2-probe signal to the control-probe estimated MDM2 copy number to be ∼100. Furthermore, karyotyping of both WDLPS and DDLPS showed a high degree of aneuploidy with individual nuclei ranging from 60 to 80 chromosomes (Fig. 1A). Consistent with the low tumor mutational burden (TMB) of many liposarcoma tumors, sequencing analysis of select regions of 26 cancer-related genes with the Illumina TruSight Tumor kit revealed no mutations in any of the genes examined.

Figure 1.

MDM2 inhibition induces expression of ribosomal and proteasomal genes in liposarcoma. A, FISH of the 12q15 region (red) and Chr12 centromere (green). B, Western blot time course characterization of canonical p53 target genes in both 94T778 and Lipo-246 at 0, 24, 48, 72 hours (hr) posttreatment with 10 μmol/L nutlin-3. C, Cell-cycle profile of 10 μmol/L nutlin-3 treated 94T778 and Lipo-246 cell lines at 72 hours posttreatment. Bottom/dark gray, G1; middle/white, S; top/light gray, G₂–M. PI, propidium iodide. D, Annexin V assay following identical conditions from (C). Statistical analysis was completed using an unpaired two-tailed t test. **, P < 0.01; ****, P < 0.0001. E, Heat map of expression levels for Hallmark p53 pathway and E2F targets leading edge genes in both 94T778 and Lipo-246 from 16-hour treatment of 10 μmol/L nutlin-3, n = 2 replicates per cell line per condition. F, GSEA using KEGG Pathways. Ox. Phos, oxidative phosphorylation. G, Volcano plots of 94T778 transcripts highlighting directionality from KEGG Ribosome and KEGG Proteasome observed in (F). Statistical significance for RNAseq was calculated using DESeq2. All experiments were completed as biological triplicates unless otherwise noted. Nut, nutlin-3.

To establish conditions for studying MDM2 inhibition in liposarcomas, we next characterized the response of WDLPS and DDLPS cell lines to 10 μmol/L nutlin-3 treatment over a 72-hour time course (Fig. 1B). Nutlin-3 treatment effectively stabilized p53 by 24 hours of treatment while upregulating direct p53 targets MDM2, p21, and PUMA. Under these same conditions, most liposarcoma cells undergo cell-cycle arrest (Fig. 1C), with approximately 20% of cells undergoing apoptosis, a significant increase over DMSO-treated cells (Fig. 1D).

To investigate the liposarcoma transcriptional response to nutlin-3, we treated both WDLPS and DDLPS cells for 16 hours and performed transcriptome analysis. In both cases, we observed genome-wide transcriptional changes, with thousands of genes both up and down regulated in each cell line (Supplementary Fig. S1A–S1B; Supplementary File 2). Among the top upregulated genes in response to nutlin-3 treatment were the well-known direct p53 target genes MDM2 and CDKN1A (Fig. 1E; Supplementary Fig. S1A-S1B). Gene set enrichment analysis (GSEA; ref. 33) of the Hallmark gene set collection identified the p53 pathway as the most significantly enriched gene set in both WDLPS and DDLPS, and E2F targets as the most significantly depleted, consistent with cell-cycle arrest (Supplementary Fig. S1C and Supplementary File 3). Closer inspection of the leading edge genes enriched in the p53 pathway gene set revealed consistent upregulation of canonical p53 target genes involved in cell-cycle arrest (i.e., CDKN1A) and apoptosis (i.e., FDXR) as well as stress response genes such as SESN1 and ATF3 (Fig. 1E). Among the leading edge genes depleted among the E2F targets gene set were E2F8 and the DNA polymerase subunit POLD3.

To further explore these data, we repeated GSEA using Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. Interestingly, among the most strongly enriched gene sets were two pathways associated with proteostasis, Ribosome and Proteasome, in both liposarcoma cell lines (Fig. 1F and Supplementary File 3). Strikingly, nutlin-3 treatment resulted in significant upregulation of 80 transcripts in the KEGG Ribosome gene set for WDLPS (Fig. 1G and Supplementary File 3) and DDLPS (Supplementary Fig. S1D and Supplementary File 3), the vast majority of which encode for ribosomal proteins, including the direct p53 transcriptional target RPS27L. Similarly, we found wholesale upregulation of proteasome subunits in liposarcoma cell lines in response to nutlin-3 (Fig. 1G; Supplementary Fig. S1E). To define whether induction of proteasome and ribosome genes in response to MDM2 inhibition was p53 dependent, we knocked down p53 for 24 hours using siRNAs and treated with nutlin-3 for an additional 24 hours prior to harvesting RNA for qRT-PCR analysis. We found that in WDLPS cells, induction of three of four selected proteasome genes required p53, whereas only one of four ribosome genes did (Supplementary Fig. S1F–S1H). Taken together, these data indicate that both WDLPS and DDLPS induce a proteostatic stress response to MDM2 inhibition in addition to the canonical p53 response.

A genome-wide CRISPR screen identifies PSMD9 as a NSG

To identify putative drug targets to combine with MDM2 inhibition in liposarcomas, we performed a genome-wide NSG loss-of-function CRISPR screen. First, the WDLPS cell line was transduced with the GeCKO library and propagated for both clearance of gRNAs targeting essential genes and expansion of library-containing cells. Then, cells were treated with DMSO or nutlin-3 for 72 hours and allowed to recover for three population doublings prior to isolation of DNA, assembly of barcoded libraries, high throughput sequencing, and bioinformatic analysis (Fig. 2A; Supplementary File 4). Notably, comparison of gRNA counts from nutlin-3–treated cells versus DMSO-treated cells revealed a striking enrichment of p53 gRNAs, consistent with p53 as the primary downstream effector of MDM2 inhibition and providing confidence in the results (Fig. 2B; Supplementary Fig. S2A; Supplementary File 4). To identify NSG candidates, we filtered the complete list of gRNAs based on 1) abundance (>8 counts per million), 2) median fold-change across replicates (<0.5), and 3) number of gRNAs scoring in the same direction (at least 3 depleted). This high-stringency primary filtering resulted in a list of 64 candidate genes, which we first ranked on the basis of number of gRNAs scoring in the same direction and then based on the smallest median fold-change of scoring gRNAs (Table 1; Supplementary File 4). Our top 10 candidate NSGs represented a diverse group of cellular functions ranging from AMPA receptor signaling (CACNG2), to meiosis (CNTD1), to DNA double strand break repair (LIG4).

Figure 2.

A genome-wide NSG screen identifies the proteasome subunit PSMD9. A, Schematic representation of the NSG screen in 94T778. B, MA plot of all filtered gRNAs highlighting the thresholding criteria (dotted lines). C, Validation of individual PSMD9 RNP-edited clones. D, Validation of pooled PSMD9-RNP clones. E, Characterization of the pooled PSMD9-RNP clones over a 72-hour time course. All experiments were completed as biological triplicates. Statistical analysis was completed using an unpaired two-tailed t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant. Ctrl, control; Nut, nutlin-3.

Table 1.

NSGs ranked by number of gRNAs depleted and smallest median fold change.

Top ten ranked NSGs
Rank	Gene ID	# gRNAs depleted	Median FC
1	CACNG2	4	0.245
2	DNAJC22	4	0.262
3	THBS3	4	0.426
4	GRAP2	4	0.432
5	ACMSD	4	0.433
6	CNTD1	3	0.139
7	PSMD9	3	0.188
8	ZBTB7C	3	0.236
9	SRI	3	0.247
10	LIG4	3	0.25

Among the top 10 NSGs, PSMD9 stood out as the only candidate that is currently clinically actionable, as it is a subunit of the 26S proteasome, for which there are numerous FDA-approved small molecules (34). PSMD9 scored with three gRNAs and a median fold-change of 0.188, consistent with strong depletion of these gRNAs (Table 1). Therefore, we selected PSMD9 for validation using RNP-Cas9 editing to knockout PSMD9. We employed a two-hit approach using gRNAs targeting upstream of exon 1 and downstream of exon 2, respectively (35) (Supplementary Fig. S2B). We screened individual clones for the desired deletion using PCR and confirmed decreased PSMD9 levels by Western blot. Notably, PSMD9 is encoded on chr12q24 and potentially amplified in our liposarcoma lines leading to incomplete knockout of some clones. Next, we validated the response of five independent PSMD9-RNP clones to nutlin-3 treatment. We observed a dose dependent increase in apoptosis in PSMD9-RNP cell lines relative to Control-RNP clones as assessed by Annexin V staining (Fig. 2C). Four of the five clones showed statistically significant increases in Annexin V levels after 72 hours of treatment with 10 μmol/L nutlin-3, whereas all five clones exhibited increased apoptosis at 20 μmol/L (Fig. 2C). For further characterization of the response of PSMD9-RNP cells to MDM2 inhibition, we pooled three RNP-edited clones, as well as three Control-RNP clones for comparison purposes, and confirmed the increased apoptotic response (Fig. 2D). We performed a nutlin-3 time course on these pooled collections of cells for molecular assessment of their response. We observed lower levels of PSMD9 in the PSMD9-RNP pool as well as an intact p53 response, demonstrated by induction of p53 and MDM2 protein levels (Fig. 2E). Furthermore, these cells retain activation of both the cell-cycle arrest and apoptotic arms of the p53 response as measured by p21 and PUMA levels, respectively (Fig. 2E). In fact, the most striking difference between PSMD9-RNP cells and controls was at the level of cleaved caspase-3, which was induced earlier and more strongly in the PSMD9-RNP pool (Fig. 2E). Altogether, these data indicate that the proteasome subunit PSMD9 is a NSG, which prompted us to test the effects of combinatorial targeting of the proteasome and MDM2 in liposarcomas.

Proteasome inhibition sensitizes liposarcoma cells to MDM2 inhibition

Having genetically validated PSMD9 as an NSG, we next tested the efficacy of MDM2 inhibitors in combination with proteasome inhibitors. We performed combinatorial profiling through a two-dimensional dose response using the Sulforhodamine B (SRB) assay. SRB binds to total protein, detecting proliferating cells as well as cells which are arrested or apoptotic, allowing for the determination of both single-agent and combinatorial efficacy. To define drug interactions, we used the Highest Single Agent (HSA) method with the SynergyFinder 2.0 software (36). The first proteasome inhibitor we tested in combination with nutlin-3 was the tool compound MG132 which did indeed interact with the MDM2 inhibitor at a wide range of doses for both small molecules in WDLPS cells (Fig. 3A; Supplementary File 5). We next tested a panel of proteasome inhibitors approved by the FDA for blood cancer therapy: bortezomib, ixazomib, and carfilzomib (37). We observed increased effectiveness at various doses with each of the proteasome inhibitors, however, carfilzomib clearly had the strongest profile (Fig. 3A and B; Supplementary Fig. S3C; Supplementary File 5). In parallel, we characterized the combinatorial response of all four proteasome inhibitors to nutlin-3 in the DDLPS cell line and obtained similar results (Supplementary Fig. S3A–S3B; Supplementary File 5). One possible explanation for the differences in overall effectiveness among the proteasome inhibitors is the mechanisms of action for these drugs. For example, both bortezomib and ixazomib reversibly bind the chymotrypsin-like subunit (PSMB5) of the 20S core/26S proteasome, whereas carfilzomib binds irreversibly to the 26S proteasome, as does MG132 (37–39).

Figure 3.

Proteasome inhibitors increase the efficacy of nutlin-3 (Nut)–induced apoptosis in liposarcoma cells. A, 94T778 96-well synergy assay with the proteasome inhibitors MG132 (5 μmol/L-19.5 nmol/L), bortezomib (50 nmol/L-195 pmol/L), ixazomib (50 nmol/L-195 pmol/L), and carfilzomib (50 nmol/L-195 pmol/L) in combination with nutlin-3 (40 μmol/L-2.5 μmol/L). HSA synergy score was calculated using SynergyFinder 2.0. B, %SRB absorbance values representing the concentrations showing the strongest HSA synergy score (from panel A). %SRB values were obtained by normalizing all SRB absorbance values to the mean of the no treatment control values across three biological replicates. C, Annexin V staining of a 94T778 treated with 10 μmol/L nutlin-3 for 72 hours, three individual doses of carfilzomib ranging from 50 nmol/L to 12.5 nmol/L, and all three carfilzomib doses in combination with 10 μmol/L nutlin-3. D, Time course characterization of 94T778 treated with 10 μmol/L nutlin-3, 12.5 nmol/L carfilzomib, and in combination. E, Annexin V transfected with the indicated siRNA for 24 hours prior to 72-hour treatment with DMSO, 10 μmol/L nutlin-3, 12.5 nmol/L carfilzomib or the combination. F, Western blot characterization of samples from panel E. All experiments were completed as biological triplicates. Statistical analysis was completed using an unpaired two-tailed t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Closer inspection of the individual doses of proteasome inhibitors in combination with nutlin-3 that yielded the strongest HSA scores showed once again that MG132 and carfilzomib performed similarly, with high levels of combinatorial efficacy achieved even at proteasome inhibitor doses with limited efficacy as single agents (Fig. 3B and C; Supplementary Fig. S3C-S3D). Bortezomib and ixazomib, conversely showed a lower degree of combinatorial efficacy, even at doses with moderate single agent efficacy for both nutlin-3 and the proteasome inhibitor (Fig. 3B). On the basis of these findings, we proceeded with characterizing the cell response to nutlin-3 and carfilzomib treatment as single agents and in combination.

Next, we used Annexin V staining to quantify the percentage of the population undergoing apoptosis in the nutlin-3–carfilzomib combination. We tested three concentrations of carfilzomib both as single-agent and in combination with 10 μmol/L nutlin-3. Strikingly, several doses of carfilzomib that did not induce apoptosis as a single agent provided massive sensitization to apoptosis in combination with nutlin-3 (Fig. 3C). A clonogenic growth assay confirmed the strength of the interaction between nutlin-3 and carfilzomib and demonstrated the durability of the combinatorial effects. As shown in Supplementary Fig. S3E, 72 hour treatment of WDLPS and DDLPS cells with nutlin-3 alone suppressed clonal outgrowth compared to DMSO treatment as did carfilzomib, albeit to a lesser extent. The combination treatment, however, resulted in no detectable clonal outgrowth at the 16-day (WDLPS) or 25-day (DDLPS) time point at which DMSO-treated cells were confluent.

To examine the mechanism of increased apoptosis in the nutlin-3–carfilzomib combination, we used Western blotting on WDLPS cells treated with single agents and in combination (Fig. 3D). As expected, both nutlin-3 alone and the combination treatment resulted in substantial stabilization of p53, while carfilzomib alone had a more modest effect on p53 levels (Fig. 3D). Consistent with p53 levels, both MDM2 and p21 were strongly induced by nutlin-3 alone and in combination with carfilzomib. The most striking synergy we observed at the level of p53 target induction was for the pro-apoptotic BH3-only protein NOXA. Combinatorial treatment of the DDLPS cell line Lipo-246 resulted in a similar induction of p53 target genes (Supplementary Fig. S3F). Finally, to test the role of p53 in the nutlin-3–carfilzomib interaction, we knocked down p53 with siRNAs for 24 hours prior to drug treatment and performed Annexin V staining and Western blot analysis. We found that knockdown of p53 completely rescued the loss of viability associated with nutlin-3 alone or the combination treatment (Fig. 3E). At the protein level, loss of p53 resulted in lower levels of MDM2 and NOXA, consistent with a requirement for p53 in the induction of these genes (Fig. 3F). Taken together these data indicate that nutlin-3 and carfilzomib interact to kill liposarcoma cells in a p53-dependent manner without combinatorially upregulating the p53 pathway itself.

Proteasome inhibitors synergize with nutlin-3 through the ATF4/CHOP axis

Our assessment of the impact of simultaneous proteasome and MDM2 inhibition revealed limited cooperative impact on the canonical p53 response, with the notable exception of NOXA. To define alternative mechanisms of interaction, we performed transcriptome analysis of liposarcoma cells treated with DMSO, nutlin-3, carfilzomib, or the combination for 28 hours and analyzed the DEGs as for Fig. 1. Ingenuity Pathway Analysis (IPA) revealed that the pathway predicted to be most strongly activated in the nutlin-3–carfilzomib treatment was the UPR in both WDLPS and DDLPS, while additional pathways associated with proteostasis were also predicted to be activated by the drug combination including Nrf2-mediated Oxidative Stress and Autophagy (Fig. 4A; Supplementary Fig. S4A; Supplementary File 6). Inspection of the DEGs comprising the UPR signature, indicated transcriptional upregulation of the ATF4/CHOP (encoded by DDIT3) axis, including ATF4 and DDIT3 themselves, as well as several of their transcriptional targets such as HSPA5, PPP1R15A, and HSPA1A (Fig. 4B; Supplementary Fig. S4B; Supplementary File 6). IPA identified a modest p53 activation signature as well, including combinatorial induction of NOXA (encoded by PMAIP1) transcript levels, consistent with our Western blot results (Supplementary Fig. S4C and S4D; Supplementary File 6). Additionally, when we examined the expression patterns of the KEGG Proteasome and Ribosome gene sets, we found a strong combinatorial effect on the Proteasome, with the majority of DEGs being significantly upregulated in by nutlin-3–carfilzomib treatment relative to single agents, and mixed effects on the Ribosome DEGs (Supplementary Fig. S4E–S4H). We next examined the impact of nutlin-3 and carfilzomib on the activation status of ATF4/CHOP pathway at the protein level. Interestingly, we found that all three treatment conditions result in eIF2α phosphorylation and activation, albeit with different kinetics (Fig. 4C; Supplementary File 7). Furthermore, we found that nutlin-3 and carfilzomib single-agent treatments had little to no effect on the expression of ATF4 protein levels, whereas the combination resulted in a strong increase in ATF4 levels (Fig. 4C). Next, we interrogated the impact of our combinatorial treatment on several effectors downstream of ATF4, CHOP (a transcription factor that is a direct target of ATF4), and BiP (an ER stress sensor that is a direct target of CHOP). Strikingly, we found that the nutlin-3–carfilzomib combinatorial treatment had a profound impact on the expression of both proteins. Here again, we observed similar results for all proteins in the DDLPS cell line Lipo-246 (Supplementary Fig. S5A). Additionally, NOXA, which we found to be combinatorially upregulated in Figs. 3D; Supplementary Figs. S3F; and S4C-S4D, is known to be activated by ER stress as well as p53, consistent with ER stress induced activation of the ATF4/CHOP axis in response to nutlin-3–carfilzomib combinatorial treatment. Taken together, these data indicate carfilzomib interacts with nutlin-3 to trigger the ATF4/CHOP ER stress response axis.

Figure 4.

Nutlin-3 and carfilzomib act through the ATF4/CHOP axis to induce apoptosis. A, Top 10 Canonical Pathways from Ingenuity Pathway Analysis of DEGs (adjusted P < 0.0001) from Lipo-246 DDLPS cells treated with nutlin-3, carfilzomib, or the combination for 28 hours. Positive Z-score indicates predicted pathway activation. B, Heat map of expression levels for top 5 genes in the UPR from IPA for 94T778 WDLPS cells. C, Time course of 94T778 treated with 10 μmol/L nutlin-3, 12.5 nmol/L carfilzomib (carf), or with a combination of both. D, Annexin V staining of 94T778-RNP edited cell line pools treated for 72 hours with both 10 μmol/L nutlin-3 and the 10 μmol/L nutlin-3 – 12.5 nmol/L carfilzomib combination. E, Characterization of the ATF4/CHOP axis in the 94T778-RNP edited cell line pools treated for 24 hours with the combination of 10 μmol/L nutlin-3 and 12.5 nmol/L carfilzomib. F, Characterization of direct p53 targets and caspase-3 cleavage in the 94T778-RNP edited cell line pools treated for 24 hours with the combination of 10 μmol/L nutlin-3 and 12.5 nmol/L carfilzomib. G, Annexin V staining of 94T778 cells treated individually with 0.5% DMSO, 10 μmol/L nutlin-3, 2 μmol/L - 0.5 μmol/L tunicamycin (Tun) and in combination with 10 μmol/L nutlin-3. Thapsigargin (Thap) was used individually at 100 nmol/L - 25 nmol/L and in combination with 10 μmol/L nutlin-3. All experiments were completed as biological triplicates. Statistical analysis was completed using an unpaired two-tailed t test. *, P < 0.05; **, P < 0.01; ****, P < 0.0001. Ctrl, control.

To define the role of the ATF4/CHOP-mediated ER stress response in carfilzomib and nutlin-3–induced apoptosis, we used CRISPR-Cas9 gene editing to alter expression levels of the two transcriptional effectors identified in this cascade, ATF4 and CHOP, as well as NOXA. To achieve this, we again generated CRISPR RNP clones using the two-hit technique (Supplementary Fig. S5B; Supplementary File 8). Following PCR screening for the expected genomic edits, multiple RNP clones for each gene were pooled for subsequent experiments, including control clones that received RNPs, but were not edited. Notably, for ATF4 and CHOP, we were not able to isolate any complete knockout clones. This is consistent with data from the Cancer Dependency Map (40), which classifies ATF4 as a common essential gene and CHOP as strongly selective with liposarcomas as the second most significantly enriched lineage (Supplementary Fig. S5C and S5D, www.depmap.org). For NOXA, we were able to isolate both partial (het) and complete (null) knockout cell lines. We next assayed the RNP-edited pools of cells for sensitivity to nutlin-3 or the nutlin-3–carfilzomib combination. Interestingly, for the ATF4, CHOP, and NOXA-null RNP pools we observed a statistically significant reduction in apoptosis at 72 hours with nutlin-3 as a single agent, however, the NOXA-het pool failed to rescue this treatment (Fig. 4D). Importantly, the ATF4 RNP pools completely blocked the combinatorial apoptotic response observed in the 10 μmol/L and 20 μmol/L nutlin-3–carfilzomib combination (Fig. 4D; Supplementary Fig. S5E). At the increased 20 μmol/L nutlin-3 dosage, the CHOP-het and NOXA-null RNP pools were not sufficient to block apoptosis (Supplementary Fig. S5E). These data suggest that ATF4 is a key factor in driving the increased apoptotic response in the nutlin-3–carfilzomib combination.

To further characterize changes in p53 and UPR signaling associated with the ability of our RNP pools to rescue nutlin-3–carfilzomib-induced apoptosis, we next performed a series of Western blots that yielded several insights. First, ATF4, CHOP, and NOXA RNP pools showed lower levels of their edited targets in response to the combination treatment (Fig. 4E). Second, knockdown of ATF4 protein levels resulted in decreased levels of downstream factors, such as CHOP and BiP (Fig. 4E). Third, constitutive downregulation of ATF4 and CHOP both result in stabilization of NOXA in response to the combinatorial treatment (Fig. 4E). Of note, NOXA protein levels are elevated at baseline in the ATF4 and CHOP RNP pools, relative to the control RNP pool (Supplementary Fig. S5F). Furthermore, transient knockdown of ATF or CHOP with siRNA did not induce NOXA to a greater extent than a control siRNA either at baseline or in response to the nutlin-3–carfilzomib combination (Supplementary Fig. S5F). These data are consistent with chronic depletion of ATF4 and CHOP inducing NOXA protein expression at baseline, and subsequent combinatorial drug treatment leading to further upregulation. Fourth, along with ATF4, knockdown of either CHOP or NOXA blocks the full induction of BiP (Fig. 4E). Fifth, the ATF4, CHOP, and NOXA RNP cell lines all have modest, but detectable impacts on the levels of p53 and PUMA induction and a stronger effect on the levels of cleaved caspase-3 (Fig. 4F; Supplementary Fig. S5G). Taken together these data indicate that depletion of several factors in the ATF4/CHOP axis is sufficient to dampen the increased apoptotic effect of dual inhibition of MDM2 and the proteasome. Finally, the NOXA results indicate that this protein, a member of both signaling cascades, is necessary, but not sufficient for the induction of apoptosis by the nutlin-3–carfilzomib combination.

The observation that the nutlin-3–carfilzomib apoptotic phenotype can be rescued by reducing expression of genes involved in the UPR pathway led us to ask if ER stress alone is sufficient to synergize with nutlin-3. Toward this end, we tested two separate ER stress-inducing small molecules, tunicamycin and thapsigargin, alone and in combination with nutlin-3. Indeed, we found that, even at doses with weak single-agent effects, both tunicamycin and thapsigargin strongly sensitized liposarcoma cells to nutlin-3–induced cell death (Fig. 4G). Treatment of liposarcoma cells with these ER stressors had a modest effect on pEIF2a and ATF4 levels at 24 hours, possibly due to the kinetics of their activation, but resulted in clear upregulation of CHOP and BiP (Supplementary Fig. S5H). At these low doses, tunicamycin and thapsigargin alone did not result in p53 stabilization or caspase-3 cleavage, but in combination with nutlin-3 resulted in a marked increase in these apoptotic markers (Supplementary Fig. S5I). Taken together, these results suggest activation of p53 in parallel to ATF4/CHOP signaling is sufficient to increase the apoptotic potential of nutlin-3 in liposarcomas.

Combinatorial proteasome and MDM2 inhibition suppresses tumor growth in a dedifferentiated liposarcomas xenograft model

Next, we tested the more clinically relevant MDM2 inhibitor, idasanutlin, a nutlin-3 derivative which has been optimized for in vivo use and tested in numerous clinical trials (41), with carfilzomib, both as single agents and in combination, in vivo. First, we confirmed the idasanutlin-carfilzomib interaction in DDLPS cells in vitro (Supplementary Fig. S6A). Next, we engrafted the DDLPS cells onto a single flank of NOD scid gamma mice and let them grow until palpable, at which point tumor-bearing animals were randomized into four groups and treated with (i) Vehicle, (ii) idasanutlin, (iii) carfilzomib, or (iv) a combination of the two drugs for up to 21 days. Analysis of body weights for each group revealed a decrease in body weight for both the carfilzomib and combination treated groups after the first treatment, however, this weight loss quickly stabilized, and no animals lost more than 15% of their body weight in this study (Supplementary Fig. S6B). Under these dosing conditions, idasanutlin-treated animals experienced no weight loss, nor did vehicle-treated animals (Supplementary Fig. S6A). We next performed survival analysis and found that all tumors in the vehicle treatment group had reached the 2,000 mm³ humane end point by day 12 and those animals had to be removed from the study, while the final single-agent carfilzomib- and idasanutlin-treated animals had to be removed by days 15 and 19, respectively (Fig. 5A). Strikingly, only 2 of the 7 animals that received the combination of carfilzomib and idasanutlin had tumors that reached 2,000 mm³, and even then, not until days 15 and 17 (Fig. 5A). Notably, both carfilzomib and idasanutlin single-agent therapies significantly extended survival by slowing tumor growth (Fig. 5A and B; Supplementary Fig. S6C). Assessment of tumor volume at the final time point at which all animals remained in the study, day 8, demonstrates a modest but not significant difference in tumor volumes among vehicle and single-agent treatment groups (Fig. 5C). However, the combination treatment group showed significantly reduced tumor volume relative to vehicle treated animals and both single-agent groups (Fig. 5C). These data indicate that the combination of idasanutlin and carfilzomib significantly enhanced overall survival by reducing tumor growth.

Figure 5.

Carfilzomib increases efficacy of idasanutlin in liposarcoma in vivo. A, Kaplan–Meier curve from a Lipo-246 CDX mouse model treated with idasanutlin (idasa) on days 1 to 5, carfilzomib (carf) on days 1 and 4, or the combination of the two for up to 21 days. Statistical analysis was completed using a Mantel-Cox test. *, P < 0.05; **, P < 0.01; ***, P < 0.001. B, Tumor growth over time in the Lipo-246 CDX mouse model. Individual tumors volumes were normalized to the volume at day 1 of the treatment start date. Black arrow indicates the time point which the first control animal was removed and indicates the time point of all subsequent CDX endpoint analysis as seen in C. Error bars represent SEM. C, Day 8 analysis of the CDX normalized tumor volumes. Time point is indicated by the black arrow in B and is the last day all animals remained in the experiment. D, Tumor growth over time in the PDX mouse model treated as in A. Error bars represent SEM. E, Endpoint analysis of the PDX normalized tumor volumes. F, End point CBC analysis resulted in no significant changes in CBC populations including platelets, RBCs, lymphoblasts, and granulocytes. Ns for Lipo-246 CDX animal model are as follows: Veh (vehicle)/Veh n = 3, idasanutlin/Veh n = 6, Veh/carfilzomib n = 3, idasanutlin/carfilzomib n = 7. N's for PDX animal model are as follows: Veh/Veh n = 4, idasanutlin/Veh n = 4, Veh/carfilzomib n = 4, idasanutlin/carfilzomib n = 5. Unless otherwise indicated, statistical analysis was completed using an unpaired nonparametric Mann–Whitney U-test. *, P < 0.05; **, P < 0.01.

To extend the preclinical significance of these findings, we next tested our therapeutic combination in a newly established PDX model, CUSARC 27. Tumor samples were engrafted into animals and allowed to grow until palpable prior to randomization into groups as described for the DDLPS cell line. Once again, we saw that carfilzomib treatment alone or in combination with idasanutlin resulted in decreased body weight after the first two treatment points that stabilized throughout the duration of the experiment (Supplementary Fig. S6D). Notably, the PDX tumors grew at a markedly slower rate than did the cell line-derived tumors, and consequently, no animals had tumors that reached 2,000 mm³ humane end point during the 21 days of drug treatment (Supplementary Fig. S6E). Neither idasanutlin, nor carfilzomib alone, had any measurable impact on tumor growth, however, the combination of drugs had a similar effect, with three of five tumors in this group demonstrating only moderate growth and two of five undergoing a modest regression (Fig. 5D and E). Finally, because MDM2 inhibitors have been reported to cause thrombocytopenia and neutropenia, we performed CBC analysis on all animals from the PDX study at the time of sacrifice. Importantly, we found no impact of any treatment group on the number of platelets or red blood cells present, nor on the frequency of lymphocytes or granulocytes in the blood (Fig. 5F). Taken together, these data indicate that the combination of proteasome and MDM2 inhibitors may represent a promising combinatorial strategy for liposarcomas and that, with abbreviated treatment schedules, this strategy may have more modest side effects than prior treatment regimes.

Discussion

Liposarcomas are the most frequent class of soft-tissue sarcoma and are characterized by low tumor mutational burden and a high degree of aneuploidy, specifically amplifications of chromosome region 12q13–15. Despite their prevalence, treatment options have been slow to develop, with surgical resection remaining as the primary intervention. In the past two decades, genomics advances have identified CDK4 and MDM2 as central nodes in the 12q13–15 locus that are amplified in ∼95% of all liposarcoma cases. Parallel developments in targeted therapeutics have led to the development of CDK4/6 inhibitors, which have now been approved for several non-sarcoma cancer types, and MDM2 inhibitors, which have yet to realize their clinical potential. Here, we report that in response to one such MDM2 inhibitor, nutlin-3, both WDLPS and DDLPS upregulate numerous components of the proteostasis network. Using a genome-wide CRISPR-Cas9 screen, we identify the proteasome as putative combinatorial target for MDM2 inhibitors and demonstrate interactions between nutlin-3 and a panel of proteasome inhibitors. We show that the combinatorial efficacy of nutlin-3 and carfilzomib is largely driven by the ATF4/CHOP/Noxa UPR axis, and that pharmacologic induction of the UPR by ER stress is sufficient to sensitize liposarcoma cells to nutlin-3. Finally, using two different xenograft models we show that the combination of carfilzomib and the clinically optimized nutlin-3 derivative, idasanutlin, are effective at slowing tumor growth and promoting survival in vivo.

The finding that nutlin-3 treatment results in transcriptional upregulation of subunits the ribosome and proteasome, as well as activation of the UPR, contributes to a growing body of evidence that p53 interacts with the proteostasis network in a cell type- and stimulus-specific manner. Indeed, previous studies demonstrated that nucleolar stress or inhibition of ribosome biogenesis leads to stabilization and activation of p53 (42–44). Conversely, however, data showing that MDM2 inhibition leads to upregulation of ribosome or proteasome biogenesis are lacking. Our data indicate that, in the context of liposarcomas, nutlin-3 induces both phenomena, and the combination of nutlin-3 and carfilzomib further induces levels of proteasome genes. Notably, PSMD9, the proteasome subunit identified in our NSG screen, was recently implicated in ribosomal protein shuttling to the nucleolus and subsequent activation of p53 (45). Numerous reports have demonstrated interactions between p53 and the UPR, however, in most cases p53 is reported to suppress the UPR. For example, Namba and colleagues showed that p53 null cells upregulate both IRE1α and XBP1s in response to ER stress relative to their WT counterparts (46). This report demonstrated that mutant p53 cell lines have higher expression of IRE1α at baseline, consistent with the notion that WT p53 suppresses activation of the UPR under basal conditions (46). Using an in vivo model of malignant rhabdoid tumors, another group showed that p53 suppressed an ER stress response in part through induction of another proteostatic pathway, autophagy (47). Interestingly, in contrast with what we report here, p53 has been shown to induce apoptosis during ER stress by suppressing the chaperone BiP (48). Relatively few studies have taken aim at the impact of p53 activation on the UPR in isolation in a nongenotoxic environment. We show here that nutlin-3 treatment alone has a modest impact on the UPR, increasing phosphorylation of pEIF2α and expression of ATF4. Activation of the UPR by proteasome inhibitors is a well-defined phenomenon, wherein accumulation of misfolded proteins generates an ER stress response (49, 50). Our data indicate that combinatorial inhibition of MDM2 and the proteasome converge on the UPR—specifically the ATF/CHOP axis—to induce a strong apoptotic response, with little to no additional impact on the canonical p53 response.

Interestingly, liposarcomas may represent a particularly strong candidate for the combination of MDM2 and proteasome inhibitors given their genetic profile. As described above, the amplification of MDM2 makes MDM2 inhibitors a natural fit, but the highly unbalanced aneuploid nature of liposarcomas could make them a uniquely promising target for proteasome inhibitors, such as carfilzomib, due to increased protein expression and protein stoichiometry imbalances that increase demand on proteasome activity. Despite a lack of success, clinical trials with MDM2 inhibitors, including for treatment of liposarcomas, proceed both as single agents (www.clinicaltrials.gov, NCT05012397 and NCT05218499) and in combination with other drugs, such as the growth factor receptor inhibitor pazopanib (NCT05180695). Proteasome inhibitors have yet to enter clinical trials for liposarcomas, however, a recent drug screening study identified a proteasome inhibitor as a potential therapeutic strategy for liposarcomas (51). Our data suggests that the combination of MDM2 and proteasome inhibitors could hold promise for the treatment of liposarcomas and interactions between proteasome inhibitors and MDM2 inhibitors has been reported in other cancer types (52, 53). Indeed, one study demonstrated that the combination of nutlin-3 and bortezomib can act combinatorially even in the absence of WT p53 in vitro, suggesting that this strategy could be effective even against p53 mutant clones within a tumor potentially preventing the recurrence of tumor resistant to MDM2 inhibitors (52). Further studies will be required to demonstrate the impact of longer-term administration of idasanutlin and carfilzomib in animals on a host of clinically relevant metrics, including the development of resistance and hematopoietic stem cell viability, among others. Finally, one of the stumbling blocks for MDM2 inhibitors in the clinic has been cytopenia (15, 54, 55); it will be critical to assess the impact of any new combinatorial strategies involving MDM2 inhibitors on this toxicity.

The potentially broadly acting mechanism of interaction we report here could portend combinatorial efficacy not only for tumors with similar genomic profiles, such as neuroblastoma and acral melanoma (56, 57), but also for many more tumors that retain WT p53 activity.

Supplementary Material

Supplementary File 1

Reagents used in this study.

Supplementary File 2

Nutlin-3-treated LPS transcriptome analyses.

Supplementary File 3

GSEA of nutlin-3-treated LPS transcriptome data.

Supplementary File 4

Genome-wide CRISPR Nutlin-3 Sensitizing Gene screen data.

Supplementary File 5

Nutlin-3-proteasome inhibitor drug interaction SRB data.

Supplementary File 6

Nutlin-3-carfilzomib-treated LPS transcriptome analyses.

Supplementary File 7

Western blots.

Supplementary File 8

Agarose gels.

Authors' Disclosures

M.P. Ludwig reports grants from NIH and nonfinancial support from Roche during the conduct of the study. M.D. Galbraith reports grants from NIH during the conduct of the study. B.A. Wilky reports personal fees from Boehringer Ingelheim outside the submitted work. J.M. Espinosa reports other support from Eli Lilly, Gilead Sciences, and Perha Pharmaceuticals outside the submitted work; in addition, J.M. Espinosa has a patent for U.S. Provisional Patent Application No. 63/356,432 title “Dual Inhibition of MDM2 and eIF2alpha Induces Cell Death in Multiple Cancer Cell Types” issued. K.D. Sullivan reports grants from NIH, Ventus Therapeutics, nonfinancial support from Roche during the conduct of the study, and grants from NIH outside the submitted work. No disclosures were reported by the other authors.

Authors' Contributions

M.P. Ludwig: Conceptualization, resources, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.D. Galbraith: Data curation, formal analysis, writing–review and editing. N.P. Eduthan: Data curation, formal analysis. A.A. Hill: Resources, investigation. M.R. Clay: Resources, writing–review and editing. C. Moreno Tellez: Resources. B.A. Wilky: Resources, funding acquisition, writing–review and editing. A. Elias: Conceptualization, funding acquisition, writing–review and editing. J.M. Espinosa: Conceptualization, supervision, funding acquisition, writing–review and editing. K.D. Sullivan: Conceptualization, formal analysis, supervision, funding acquisition, visualization, methodology, writing–original draft, writing–review and editing.