Chromobox 2 inhibition: A novel activity of alisertib, an Aurora A kinase inhibitor

Abstract

Chromobox 2 (CBX2), a subunit of Polycomb Repressor Complex 1 (PRC1), is expressed in high-grade serous carcinoma. CBX2 inhibitory peptide (CBX2i) has demonstrated efficacy in a syngeneic mouse model, but has limitations. We sought to identify an alternative approach to CBX2 inhibition.

A computational-based molecular docking screen was performed using the SelleckChem Bioactive library to identify inhibitors of CBX2. A similarity screen of top hits against the bound conformation of CBX2i pharmacophore model was performed in parallel. A series of in vitro validation studies evaluated the effect of alisertib on proliferation, a CBX2 target gene, and stemness. CBX2 knockdown cell lines and a syngeneic murine model were utilized to evaluate alisertib response in the context of CBX2 loss. Cell target engagement assay was performed. PRC1-activity was measured by H2AK119ub levels. Immune profiling of treated tumors defined the immune microenvironment.

The computational-based screen identified 10 candidate compounds. In vitro validation narrowed compounds of interest to raltitrexed, alisertib, GTX-007, LY315920, and PD0325901. Ultra-low dilution assay demonstrated dramatic decrease in spheroid formation with alisertib, an aurora A kinase (AURKA) inhibitor. Good structural overlap was observed between CBX2i and alisertib. Cell target engagement assay confirmed alisertib selectivity for both aurora A kinase and CBX2. Loss of CBX2 attenuated alisertib efficacy in vitro and in vivo. Treatment with alisertib leads to decrease in H2AK119ub and shift in the immune tumor microenvironment.

Alisertib efficacy in HGSC is dependent on functional CBX2 and cell target engagement confirms selectivity for CBX2, supporting that alisertib activity involves CBX2 inhibition.

INTRODUCTION

High-grade serous carcinoma (HGSC) is an aggressive malignancy of the ovary, fallopian tube and peritoneum. HGSC is usually diagnosed at an advanced stage and treated with a combination of cytoreductive surgery and platinum-based chemotherapy¹. In over 80% of cases, HGSC will recur and develop resistance to chemotherapy. These patients will eventually succumb to their disease, often within five years. Given the lethal nature of HGSC, therapeutic development and novel therapeutic combinations are critical for improved patient outcomes.

Chromobox 2 (CBX2) is an epigentic reader and subunit of the polycomb repressor complex 1 (PRC1). PRC1 functions to catalyze ubiquitin to Histone H2A Lysine 119 (H2AK119ub) to promote the recruitment of other chromatin regulators, such as polycomb repressor complex 2. CBX2 is directly involved in chromatin regulation and transcription, while PRC1 coordinates stemness, immune evasion, and the initiation of metastases ^2,3. CBX2 is expressed in >75% of HGSC and increased expression is associated with disease progression and poor prognosis ⁴. Increased expression of CBX2 has similarly been found to promote disease progression and dissemination in other cancer types, including prostate, breast, and colorectal cancer ^5-8.

Similarly, in multiple cancer types, including cervical, colorectal, breast, prostate, and HGSC, inhibition, or loss of, CBX2 leads to differential chemoresponse and attenuated tumor growth, shifts in the tumor microenvironment, and decreased stemness ^3,4,8-11. Given these findings, a peptide inhibitor, CBX2i, was developed to target the unique chromodomain and A/T hook region of CBX2 ¹². The peptide was evaluated and found to effectively decrease proliferation and stemness markers in vitro and decrease tumor burden in a mouse xenograft model with a human HGSC cell line ¹². However, developing novel therapeutics is resource-intensive, and for the peptide-based CBX2i, in vivo stability was a limitation. Given these concerns, we explored alternative options to therapeutically target CBX2.

Using the inhibitory peptide CBX2i as a target pharmacophore template, we performed a dual computational approach of molecular docking of a compound library and a biosimilarity screen to identify existing therapeutic agents with similar structural conformation and interactions with the CBX2 model in order to accelerate the clinical application of CBX2 as a potential therapeutic target. Top hits were identified and narrowed to the most effective agents. This work will describe the identification of candidate CBX2 inhibitors from the computational approaches, as well as the validation and in vitro and in vivo testing of the top hit, the aurora A kinase inhibitor, alisertib.

MATERIALS AND METHODS

Cell culture

As described in our previous work ¹², human HGSC cell lines (PEO1 RRID: CVCL_2686 [authenticated 4/15/2022], OVCAR4 RRID: CVCL_1627 [authenticated 1/31/2023], COV504 RRID: CVCL_2424 [authenticated 10/18/2023]) were obtained from the Gynecologic Tissue and Fluid Bank (GTFB) in 2016. GTFB banked the cells and authenticated via STR profiling at the University of Arizona Arizona Genetics Core (AZGC) (RRID:SCR_012429). A syngeneic (murine) cell line (ID8 Trp53^−/− Brca2^−/− RRID: CVCL_IU14 [authenticated 12/4/2023]) was generously provided by Ian McNeish¹³. STR profiling authenticated the ID8 lines at The American Type Culture Collection (ATCC). OVCAR4, COV504, and PEO1 cells were cultured in RPMI1640 supplemented with 10% Heat-inactivated FBS, 100 U/ml Penicillin/Streptomycin. ID8 cells were cultured in DMEM supplemented with 4% Heat-inactivated fetal bovine serum (FBS), 100 U/ml penicillin/streptomycin, 5 μg/ml insulin, 5 μg/ml transferrin and 5 ng/ml sodium selenite (Thermo Fisher, Massachusetts, USA). All cells were cultured at 37°C in 5% CO₂. Cells are only passaged up to 20 passages. Cells were routinely tested for Mycoplasma using LookOUT (Millipore-Sigma, Darmstadt Germany) or Mycostrip (InvivoGen, California, USA), most recently on November 6, 2024. Biological sex of all cells is female.

Compounds

All compounds were purchased from SelleckChem unless otherwise noted, and citations provide chemical structure. GTX-007 ¹⁴ (MedChem, Cat. #HY-12023), LY-315920¹⁵ (Cat. #S1110), Raltitrexed¹⁶ (Cat. #S1192), PD0325901¹⁷ (Cat. #S1036), BMS-582664¹⁸ (Cat. #S1138), Cabozantinib¹⁹ (Cat. #S1119), Dasatinib²⁰ (Cat. #S1021), SRT1720²¹ (Cat. #S1129), BMS-387032²² (Cat. #S1145), Alisertib²³ (Cat. #S1133), and LY3295668 ²⁴ (MedChem, Cat. #HY-114258)

Proliferation Assays

All drugs were suspended in DMSO at the concentration of 50 mM and aliquoted and stored in −80°C. 1000 cells were seeded in 96 well plate a day prior. Drugs were diluted with 10-fold dilutions in the culture media supplemented with FBS and penicillin/Streptomycin was added, and cells were cultured until the cells in the no-drug added wells became confluent (5-8 days). Cells were washed with PBS twice and fixed with fixation buffer (10% Methanol and 10% Acetic acid in PBS) followed by staining with crystal violet staining solution (0.4% Crystal Violet in 20% Ethanol) ²⁵. After the crystal violet in the wells was washed away with distilled water, plates were dried overnight, and crystal violet on the cells was dissolved in fixation buffer and 570 nm absorbance was quantified by a plate reader (OD570).

Small hairpin (shRNA) knockdown

As described previously ⁴, CBX2 specific shRNAs were obtained from the University of Colorado Functional Genomics Facility (human CBX2 #1: TRCN 0000020327 and human CBX2 #2: TRCN 0000232722). All CBX and AURKA shRNAs are included in Table S1. For Mouse shCbx2, shRNA sequence of the TRCN0000334429 was cloned into pLKO.1-blast vector (RRID:Addgene_26655) using EcoRI and AgeI site. An empty pLKO.1-puro (RRID:Addgene_10878) and pLKO.1-blast were utilized as shControl (shCTRL). Plasmid isolation was performed using a Plasmid Midi-Prep Kit (Qiagen Hilden, Germany). Following the manufacturer's instruction, HEK293T cells were transfected with lentivirus construct with packaging plasmids with Lipofectamine 2000 (Thermo Fisher Scientific, Waltham, MA). Five to six hours after transfection, cells were incubated overnight and transitioned to Dulbecco’s Modified Eagle Media (DMEM) supplemented with 10% FBS and 100 U/ml Penicillin/Streptomycin. Viral supernatant was collected 72 hours post-transfection, and applied to OVCAR4 or ID8 Trp53^−/− Brca1^−/− for 24 hours with Polybrene. Infected cells were selected using 1.0 μg/ml puromycin or 10 μg/ml blastcidin.

CRISPR knock out of CBX2

CBX2 CRISPR/Cas9 knocked out OVCAR4 cells were created by the University of Colorado Anschutz Medical Campus Cancer Center Functional Genomics Core Facility (RRID:SCR_021987) using the IDT Alt-R RNP system. ALT-R crRNA and ATL-R tracrRNA were suspended at 100 μmol/L in nuclease-free IDTE pH7.5. The same volumes of ALT-R crRNA and ATL-R tracrRNA were mixed to prepare the gRNA complex at 50 μmol/L, heated at 95°C for 5 minutes, and then cooled to room temperature. The RNP complex was prepared with 150 pmol of gRNA and 125 pmol of Alt-R Cas9-NLS in final volume of 5 μL in PBS. Cells were harvested in Necleofector solution SF with supplement (Lonza, Basel, Switzerland) at a concentration of 1x10⁶ cells per 100 μL. The transfection mix was made with 100 μL of cell suspension, 2.5 μL each of RNP complex, and 0.6 μL of 100 μL of Alt-R Cas9 Electroporation enhancer. For OVCAR4, program FE-132 was used. After pulsing, prewarmed medium was added into the Nucleocuvette vessel, and cells were plated for further culturing and isolating clonal populations. The following guide RNAs were used: Hs.Cas9.CBX2.1.AC:CCGAGTGCATCCTGAGCAAG and Hs.Cas9.CBX2.1.AA:GAGTACCTGGTCAAGTGGCG. PCR primers to screen for deletion included CBX2-F1: AGCATGGAGGAGCTGAGCA, CBX2-R2: GGTTACAGCGGGGAGAATCTG, and CBX2-R3: GGAGAATCTGGCCAAGAGGAG.

Computational-based Screening

Generation of the CBX2-CBX2i model.

The YASARA software platform (version 14.12.2) was used to construct a CBX2 homology model from crystal structure of the chromodomain of CBX2 (PDB: 5EPK)²⁶ and a CBX4 homology model from the crystal structure of the CD of CBX4 (PDB: 5EPL)²⁶. This homology model of CBX2 was then used to dock several peptides, including CBX2i, into a cleft spanning the chromodomain and the A/T hook. Molecular dynamics simulations revealed that the CBX2i peptide was the preferred inhibitor of CBX2 of the peptides examined^12,26.

Virtual library and similarity screening.

We used the docked conformation of CBX2i as a reference ligand to define the CBX2 binding cavity and docked the SelleckChem Bioactive Compound Library containing 8,270 compounds (downloaded from www.selleckcem.com in 2022) and CBX2i (as a control) using the GLIDE program within the Schrödinger Suite (release 2022-4, Schrödinger, LLC, New York, NY). The protein was prepared using the Protein Preparation Workflow and the compound library was prepared using LigandPrep to generate protonated species at physiological pH and retain defined stereochemistry. The docking Grid was defined using default parameters using CBX2i as reference and the compounds were docked using GLIDE with XP settings^27,28. A total of 200 compounds were identified as hits from this virtual screen and the top 10 compounds were ranked based on their Xtra Precision Glide (XPG) Score (Fig. 1A-B). As an additional approach we used the docked conformation of CBX2i in the binding cleft of CBX2 to generate a pharmacophore model using the Phase module within the Schrödinger Suite (Fig. 1C)²⁹. This pharmacophore model was used as a template to perform a Shape similarity screen against the 10 top-ranked compounds from the virtual screen using Phase feature definitions²⁹. Volume scoring was set to the typed pharmacophore model and ranking of compounds was performed using the Phase Screen Score. This similarity screen was performed to support the identification of compounds that could adopt a similar binding conformation as CBXi and retain a similar interaction fingerprint with amino acid residues of CBX2.

Figure 1. Computational-based docking simulation and similarity screen.

A) Molecular docking of the SelleckChem Bioactive Library into CBX2 using CBXi as a reference ligand in the CBX2-CBXi complex determined from Yasara molecular dynamic simulations. The top 10 ranked compounds by XPG Score from the docking are shown. B) XPG Scores were calculated for the top 10 compounds from the computational-based molecular docking. The more negative the score, the greater the potential for the compound to bind CBX2. C) The CBX2-bound conformation of CBX2i displays the pharmacophore model used in the similarity screen against the top 10 hits from the molecular docking screen; compounds are ranked by their shape similarity.

Immunoblotting

As described in McMellen et al, 2023,²⁵ protein was extracted with radioimmunoprecipitation assay buffer (150 mM NaCl, 1% TritionX-100, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate [SDS], 50 mM Tris pH 8.0) supplemented with Complete EDTA-free protease inhibitors (Roche,), 5mM NaF, and 1mM Na3VO4. Protein was separated on an SDS polyacrylamide gel electrophoresis and transferred to polyvinylidene fluoride membrane. Primary antibodies anti-CBX2 (Proteintech, Cat #15579-1-AP, RRID:AB_2737362) and anti-β-actin (Abcam Cat# ab6276, RRID:AB_2223210), anti-H2AK119Ub (Cell signaling Cat#8240) incubation was performed overnight at 4°C. Secondary goat anti-rabbit (IRDye 680RD or IRDye 800CW, LI-COR, Cat. #92568071; RRID: AB_2721181 or Cat. #926-32211; RRID: AB_621842; 1:20,000) and goat anti-mouse (IRDye 680RD or IRDye 800CW, LI-COR, Cat. # 926-68070; RRID: AB_10956588 or Cat# 925-32210; RRID: AB_2687825; 1:20,000) antibodies were applied for 1 hour at room temperature. Blots were visualized using the Licor Odyssey Imaging System and ImageStudio software (V4). Densitometry performed using ImageJ.

ALDEFLUOR Assay

ALDEFLUOR assay was performed as previously and according to the manufacturer's protocol (STEMCELL Technologies, Cat. #01700) ⁴. Briefly, 100,000 cells were trypsinized and resuspended in ALDEFLUOR buffer or buffer with diethylaminobenzaldehyde (DEAB) as a negative control. ALDEFLUOR reagent was added and incubated at 37°C for 45 minutes. Cells were washed with cold HBSS buffer (2% FBS/HBSS). Cells were stained with Zombie Red (BioLegend, Cat.# 423109) for viability. ALDH activity was analyzed using Novocyte Penteon (Agilent). Cytometry data were analyzed using FlowJo software (Tree Star).

Ultra-Low Dilution Analysis (ULDA) and MTT assay

Cells were trypsinized and passed through a 40-μm filter to generate a single-cell suspension. Cells were seeded into 96-well ultra-low attachment plates (Corning, Cat. #3474, flat-bottom; Corning, Cat. #7007, U-bottom for COV504) at 200 μL per well in CSC medium (DMEM/F12 supplemented with 1× B27, 4% FBS, 100 U/mL penicillin/streptomycin, 20 ng/mL human EGF, and 20 ng/mL human FGF). After 3 days, 100 μL of medium was carefully removed and drugs were added to achieve the final IC₅₀ concentration. Drug-containing medium was refreshed at half-volume every 4–5 days, and cultures were maintained for 3 weeks. At the end of the 3-week culture period, MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide] was added to a final concentration of 500 μg/mL. Plates were incubated at 37 °C for 4 hours, after which spheroids were examined microscopically for viability. The presence of purple spheroids indicated viable cells and was scored as 1; absence of spheroids or lack of purple coloration was scored as 0. These data were plotted, and stem cell frequency was calculated using ultra-low limiting dilution analysis - https://bioinf.wehi.edu.au/software/elda ³⁰.

Quantitative reverse-transcriptase PCR (qPCR)

RNA was isolated using RNAeasy Plus Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. NanoDrop spectrophotometry was used to measure the concentration of extracted RNA. Luna Universal One-step RT qPCR kit (New England BioLabs, Ipswich, MA) was used on BioRad CFX96 (Bio-Rad, Hercules, CA). TMEFF1 Fwd – GCATGCCAATTTCAGTGCCATA, TMEFF1 Rev – GTGCTTACAAGCAGCCCTTC; CBX2 Fwd – CGGCTGGTCCTCCAAACATAA, CBX2 Rev – TTGCCTCTCTTCCGGTTCTG; CBX4 Fwd – TGGGAACCGGAGGAGAACAT, CBX4 Rev – AAAGGTAGGCACCTGCACCA; CBX6 Fwd – GTGGGCGATCAAGTACAGCA, CBX6 Rev – CATACAGCTCACGCTCCCTC; CBX7 Fwd – GATGGCCCCCAAAGTACAGCA, CBX7 Rev- CTCCTTGCCCTTGGCCTTGT; CBX8 Fwd- CAACATGGAGCTTTCAGCGG, CBX8 Rev- ATTCCATGCGTCCTTTCCGT; HPRT Fwd – CCTGGCGTCGTGATTAGTGA, HPRT Rev – CGAGCAAGACGTTCAGTCCT; murine GAPDH Fwd – TGCACCACCAACTGCTTAG, murine GAPDH Rev – TGCACCACCAACTGCTTAG; murine CBX2 Fwd – AAGCTGGAGTACCTGGTCAAG, murine CBX2 Rev – ACCTCCTTCTCATGTTCCTTCTTC.

Size-exclusion-chromatography fractionation (fast-protein liquid chromatography, FPLC)

OVCAR4 cells were cultured and treated with DMSO or 100 nM Alisertib for 72 hrs, harvested, washed in PBS, and collected by scraping. Cells were suspended in 2 mL of Buffer A (10 mM HEPES pH7.9, 10 mM KCl, 1.5 mM MgCl2, 0.34 M sucrose, 10% glycerol, 1 mM DTT, 1 x complete protease inhibitor EDTA free) and incubated at 4°C for 5 minutes. Nuclear (pellets) and cytosol (supernatant) fractions were separated by centrifugation at 1000 x g for 5 minutes at 4°C. Nuclear pellets were resuspended in 1 mL of Buffer A and homogenized with 11 strokes of Dounce homogenizer (A). Suspension was layered onto 300 μl of Buffer A sucrose cushion containing 30% sucrose weight per volume, then centrifuged at 2400 x g for 1 hr at 4°C. Nuclear pellets (bottom of the sucrose cushion) were resuspended in 800 μl of high salt buffer (HSB, containing 50mM Tris HCl pH7.5, 300 mM KCl, 1mM MgCl2, 1mM EDTA, 1mM DTT, 1% TritonX-100, 1 x complete protease inhibitor), then rotated for 1 hour at 4°C, and centrifuged and 17,000 x g for 20 minutes at 4°C. Supernatant was flash frozen and stored at −80°C until gel filtration.

Gel filtration chromatography was performed using a Superose 6 Increase 10/300 GL column (23.5 mL, Cytiva) pre-equilibrated in PBS running buffer on an ÄKTA FPLC system (Cytiva). After receiving the prepared samples, 500 μL of each was loaded onto the column, and elution was performed at a flow rate of 0.5 mL/min. Fractions were collected at 0.5 mL per tube and stored at 4°C for downstream analysis. 40μl of each fraction was for immunoblotting (see above). Primary antibodies used for western blotting are CBX2 (Proteintech Cat# 15579-1-AP, RRID:AB_2737362, 1:1000), CBX4 (Cell Signaling Technology Cat# 30559, RRID:AB_2798991, 1:1000), PHC3 (Bethyl Cat# A301-569A, RRID:AB_1040014, 1:1000), β-Actin (Abcam Cat# ab6276, RRID:AB_222321,1:5000).

Cell Target Engagement Assay

Cell culture and transfection.

HEK293 cells were cultured in DMEM supplemented with 10% FBS and antibiotics (100 U/mL penicillin, 100 μg/mL streptomycin) at 37°C and 5% CO₂. Cells were transiently transfected with plasmids encoding S-tag-fused Aurora A kinase, chromobox protein homolog 2 (CBX2), or chromobox protein homolog 4 (CBX4) (TWIST Biosciences, San Francisco, CA). A background control (empty vector) was included in all experiments. Transfections were performed using Lipofectamine-mediated protocols, and cells were incubated for 48 h prior to assay.

S-tag-based enzyme complementation assay.

Target engagement was quantified using the Stag-based enzyme complementation system as previously described (³¹, [bioRxiv 2025.09.04.674281v1]). Following transfection, cells were perforated with 0.1% Triton X100 and incubated with S-protein reagent and fluorogenic substrate. Complementation activity was recorded as raw fluorescence units (RFU) using a multimode plate reader.

Thermal challenge series.

Cells expressing aurora A kinase, CBX2, or CBX4 were subjected to thermal challenge by heating between 37–68°C for 3 min, followed by recovery at 37°C for 30 min. Real-time fluorescence data was recorded using a multimode plate reader. Complementation activity was normalized to 37°C controls. Melting profiles were fit to a sigmoidal equation (nonlinear regression, four-parameter logistic, GraphPad Prism v10.1.1) to determine thermal aggregation, T-agg50, points.

Western blotting for cellular target engagement.

Immunoblotting was performed with HEK293 cells transfected with plasmid expressing CBX2-Stag. Cells were cultured in DMEM medium supplemented with 10% FBS. Fresh cell lysate prepared in non-denaturing buffer was dispensed into 96-well PCR plate (approx. 6000 cells/well/50 μl), then treated at various doses (20, 6.7, 2.2, 0.74 and 0.25 μM) of Alisertib together with DMSO control, for 1 hour. Samples were then subjected to a heat challenge at Tagg(50 °C) for 10 min, and unstable aggregated protein was removed by a centrifugation step (14,000 rpm). Supernatant was collected and SDS-PAGE gel was run, and immuno-detection was performed for CBX2 and Gapdh using corresponding anti-Stag and anti-Gapdh antibodies, respectively. Band intensity was quantified on LI-COR C-Digit Blot Scanner. Bands of stable target proteins were quantified, normalized to loading control. EC50 value of Alisertib with CBX2 was calculated.

Alisertib treatment and slope analysis.

Cells were treated with the alisertib across seven serial dilutions (2×10⁻⁵ to 2.7×10⁻⁸ M) plus vehicle (DMSO). For each target, compound responses were assessed at 40°C, 45°C, and 50°C, identified as common thermally sensitive points. Data were analyzed using slope-based integration as described in our referenced manuscript. EC₅₀ and Hill slope values were derived from total dose-series responses across all temperatures.

In vivo experiments

All mouse experiments were approved by the University of Colorado Anschutz Medical Campus Institutional Animal Care and Use Committee (IACUC protocol #569). Six- to eight-week-old C57BL/6J mice were purchased from The Jackson Laboratory (strain #000664), and each mouse was intraperitoneally (i.p.) injected with 5×10⁶ ID8 Trp53^−/− Brca2^−/− cells, followed by a seven-day tumor establishment period prior to the initiation of drug treatment. During drug treatment, body weights were recorded and served as a surrogate for general health and potential drug toxicity. ID8 cells preferentially target the mouse omentum within hours of cell injection³², which recapitulates characteristics of HGSC in humans ^33,34. As such, tumors herein were considered primary, and all others were considered secondary or disseminated tumors. Mice were i.p. injected with ID8 Trp53−/− Brca2−/− cells that were either CBX2-intact (shControl) or CBX2-depleted (shCbx2), as described above and in Iwanaga et al., 2024 ³. Seven days after cell injection, mice were treated via oral gavage (PO) with either alisertib (30 mg/kg) or vehicle control (5% DMSO, 35% PEG300, 5% Tween80 in sterile H₂O) five days per week for 28 days (n = 8). Mice were euthanized on the 29^th day and tumor burden was assessed through omentum weight, number of disseminated tumor sites, and total disseminated tumor weight. These tissues were also collected and processed for further analysis, as described below.

Tissue dissociation

Disseminated mesenteric tumors were dissected using micro-scissors, taking care not to gather extra mesenteric tissue. Harvested tumors from four mice per treatment group (described above) were placed into 2ml of Click’s media and minced with scalpels before dissociation using the Miltenyi Mouse Tissue Dissociation Kit (Miltenyi Biotec, Cat #130-096-730). Briefly, minced tumors were transferred to gentleMACs C Tubes (Miltenyi Biotec, Cat #130-093-237) containing RPMI with Miltenyi enzymes A, R, and D. GentleMACs tubes with tumors and enzyme mixture were then moved to gentleMACs Octo Dissociator and incubated at 37°C for 40 minutes with continuous rotation. After dissociation, tissue suspensions were passed through a 70μm filter and washed with 10ml of RPMI 1640 to gain single cell suspensions, which were then centrifuged at 300g for 5 minutes at 4°C. If the resulting cell pellet possessed significant red blood cells (by eye), then cell pellets were resuspended in ACK lysis buffer and incubated at room temperature for 10 minutes and centrifuged again at 300g for 5 minutes at 4°C. Cell pellets were then resuspended in Flow Isolation Buffer (1x HBSS with 1.6% 30%w/v BSA and 0.4% 500mM EDTA) and transferred to 96 well U bottom plates for flow cytometry staining.

Flow cytometry

Cells obtained from tissue dissociation (four mice per treatment group, described above) were counted and 1-3x10⁶ cells were plated into 96 well U bottom plates for staining. All cells were stained in 100 μl total volume consisting of 10ul Brilliant Stain Buffer (Invitrogen; Cat# 00-4409-750), 10 μl anti-Fc receptor blocking antibody (2.4G2 anti-CD16/CD32; acquired from Ross Kedl at University of Colorado, Anschutz Medical Campus, Denver, CO), sum of staining antibodies, and brought to 100 μl total volume with FACS buffer (1x PBS with 2% FBS). Samples were stained for 30 minutes at 4°C in the dark, centrifuged at 300g for 5 minutes at 4°C, and washed twice with 100 μl FACS buffer. Following all staining and washing, cells were filtered through 30 μm filters to prevent clogging and analyzed on the ZE5-YETI cytometer (UCCC Flow Cytometry Shared Resource, RRID:SCR_022035). Single stain (SS) controls and fluorescence minus one (FMO) controls were performed for each experiment. Cell gates were assigned based on the positive and negative thresholds for each fluorophore using SS and FMO controls for each experiment. All data were analyzed using FlowJo analysis software.

Cells were stained with Live/Dead (Zombie UV; BioLegend; Cat#423107), CD45 (1:300; BUV395; BD Biosciences Cat# 564279, RRID:AB_2651134), Ly6G (1:100; BV510; BioLegend Cat# 127633, RRID:AB_2562937), CD11c (1:200; BV786; BioLegend Cat# 117336, RRID:AB_2565268), CD11b (1:200; AF594; BioLegend Cat# 101254, RRID:AB_2563231), MHCII (1:200; Pacific Blue; BioLegend Cat# 107620, RRID:AB_493527), F4/80 (1:200; PerCP; (BioLegend Cat# 123126, RRID:AB_893483), CD206 (1:200; APC; BioLegend Cat# 141707, RRID:AB_10896057), PD-L1 (1:100; BV711; (BioLegend Cat# 124319, RRID:AB_2563619).

Immunohistochemistry on mouse omental tumors

As described previously,⁴ mouse omental tumors collected at the time of necropsy were formalin-fixed, paraffin-embedded, and five-micron sections were mounted onto microscope slides (n = 8). These samples were stained with antibodies specific for CBX2 (Thermo Fisher Scientific Cat# PA5-30996, RRID:AB_2548470, 1:50), H2AK119ub (Cell Signaling Technology Cat# 8240, RRID:AB_10891618, 1:5000), Ki67 (ThermoFisher Cat# RM-9106-S, RRID: AB_2341197, 1:250) and cleaved caspase 3 (CC3) (Cell Signaling Technology Cat# 9661, RRID:AB_2341188, 1:500) resuspended in 1% BSA/TBS and incubated overnight at 4°C. An isotype control (Rabbit IgG) was incubated in parallel. The secondary antibody, anti-rabbit Vector Laboratories ImmPRESS polymer Kit (Vector Laboratories MP-7451), was applied and incubated for 60 minutes. Slides were washed with TBS and developed with Liquid 3,3′-diaminobenzidine tetrahydrochloride (DAB) + Substrate Chromagen System (Agilent, Santa Clara, CA; Cat# K3468). Slides were counterstained with hematoxylin, and images were captured on an Olympus inverted microscope. Images were deidentified, and cell segmentation and DAB intensity per cell were calculated in quPATH v0.5.0 ³⁵.

Publicly Available Dataset

CBX and AURKA mRNA expression in ovarian lineage cell lines was correlated to alisertib sensitivity (area under the curve, AUC). Data was collected from the Dependency Mapping Portal ³⁶ in December 2024 and July 2025.

Statistical analysis

All statistical analysis was conducted in Prism GraphPad v10. Survival comparison used Kaplan-Meier with Logrank, pairwise comparison used t-test, multicomparison used Analysis of Variance with multiple test correction, pairwise comparison over time was analyzed using a mixed model effect. A p<0.05 was considered significant and when required False Discovery Rate multicomparison correction (q<0.05) was made. Error bars are shown as standard error mean (SEM). All in vitro experiments were performed in triplicate, and the in vivo experiment was run in duplicate.

Data Availability

All data is presented within the figures and supplemental information. Please contact the corresponding author for any additional information.

RESULTS

Computational screen to identify a known small molecule therapeutic to target CBX2

We previously developed and validated a CBX2 inhibitory peptide, CBX2i, which was confirmed to have efficacy in vivo ¹². Although the N-terminus of the CBX2i peptidomimetic was acetylated and the C-terminus aminated to enhance stability and improve bioavailability properties, extensive refinement of this compound would be needed for use as a therapeutic agent. Therefore, as an alternative approach to identify a small molecule that could target the same binding site within CBX2, we conducted a computational-based docking screen against the SelleckChem Bioactive Compound Library. This compound library was selected as it contains compounds that are structurally diverse, cell permeable, and have been examined in preclinical studies that would provide known bioactivity and acceptable dose range for our studies. For these docking simulations, we used CBX2i as a reference ligand to define the CBX2 binding site. From molecular docking of the SelleckChem Bioactive Compound Library to the defined binding site of CBX2, we identified 200 compounds that had the potential to bind the defined site within CBX2. We ranked the compounds by their XPG Score, a scoring function within Schrödinger Suite that represents the predicted binding energy between the ligand and the protein. CBX2i had the most favorable binding energy of −11.9 kcal/mol, which served as a positive docking control. The 10 top-ranked hits from the Bioactive Library were structurally diverse, and all had favorable predicted binding energies to support CBX2 binding (Fig. 1A-B). As an additional approach, we performed a shape similarity screen using the docked conformation of CBX2i from the docked CBX2-CBX2i complex ¹², to generate a pharmacophore model for CBX2i using Phase within the Schrödinger Suite.²⁹ The purpose of the pharmacophore model was to map the hydrogen-bond acceptor/donor groups and hydrophobic regions of CBX2i that were likely involved in the binding to CBX2. This pharmacophore model was then incorporated into the shape-based similarity screen, acting as a “fingerprint” to identify compounds that have the potential for similar interactions. The shape-based similarity screen is a widely used virtual screening method to identify compounds with similar 3D conformations and electronic properties to a known active compound, even from distinct chemical structures ^29,37-41.

Similarity or shape screening are often used in conjunction with other virtual screening methods, such as docking and molecular dynamics, to further refine the hit list ⁴¹. Given that the CBX2 structure was a computationally generated model and could contain artifacts, we used the shape-based similarity screen to examine the similarity of the 10 top-ranked compounds from the molecular docking to CBX2 to the CBX2i structure. The top 10 hits from the Bioactive Library molecular docking screen were applied to the shape-based similarity screen, which ranked the compounds based on their ability to adopt a similar conformational shape as CBX2i docked to CBX2 and retain interactions defined in the pharmacophore model (Fig. 1C). The rankings of compounds in the molecular docking differed from that of the shape-based similarity screen and the docked conformations of compounds in CBX2 differed from the aligned conformations with CBX2i. This was expected given the independent computational approaches of favoring docking into CBX2 or alignment with the CBX2i.

Evaluation of proliferation and CBX2 target genes

The top hits from the biosimilarity screen (Fig. 1A) were then evaluated in HGSC models. Dose-response assays of all 10 inhibitors were conducted in OVCAR4, COV504, and PEO1 HGSC cell lines to determine half inhibitory concentration (IC₅₀) (Fig. 2A, S1A-B). Understanding that many of these agents are known to be anti-proliferative, we sought to better understand the potential specific involvement of CBX2 in anti-tumor effect. We explored each of the top hits’ ability to downregulate a CBX2 target gene, a transmembrane protein with EGF-like and two follistatine-like domains 1 (TMEFF1) (Fig. 2B). TMEFF1 is demonstrated to play a role in tissue development and tumor promotion ^42,43 and in HGSC cell lines (OVCAR4 and COV504) and human HGSC tumors TMEFF1 was identified to be directly correlated to CBX2 expression (Fig. S1C-E). We found four inhibitors, raltitrexed, GTX-007, LY315920, and alisertib, significantly inhibited TMEFF1 expression as measured by qPCR after treatment in OVCAR4 cells (Fig. 2B).

Figure 2. Impact of treatment with IC50 of top compounds on markers of CBX2 and PRC1.

A) Dose response assays of 10 inhibitors were conducted to determine 50% inhibitory concentration (IC50) for OVCAR4 cells. B) qPCR of TMEFF1 after treatment of OVCAR4 cells with IC50 of each of the top compounds. Internal control, HPRT. C) Immunoblot for H2AK119ub after treatment of OVCAR4 cells with IC50 of each top compound. Error bars, SEM. Statistical test, non-linear regression with sigmoidal dose-response analysis, 95% CI included (A), and one-way ANOVA with Tukey multicomparison correction (B).

CBX2 is a known subunit of the polycomb repressor complex 1 (PRC1), which is critical for normal development and functions as a transcriptional repressor ². PRC1 is understood to have histone ubiquitin ligase activity, specifically targeting Histone H2A Lys119 ubiquitination (H2AK119ub). To evaluate the impact of treatment with alisertib on the broader PRC1 complex, H2AK119ub was used as a surrogate marker of PRC1 activity. We treated OVCAR4 cells with IC₅₀ doses of each of the top compounds, and observed that H2AK119ub was downregulated compared to control (Fig. 2C). Notably, alisertib reduced H2AK119ub levels by over 7-fold (0.14 vs. 1.0, Fig. 2C). These data demonstrate that a majority of the biosimilarity screen hits inhibited PRC1 activity and alisertib, an aurora A kinase (AURKA) inhibitor, was particularly effective in reducing H2AK119ub.

Stemness as a functional component of CBX2 inhibition

Based on our prior work and understanding that CBX2 drives a stem-like phenotype, stemness assays were utilized to explore the functional implications of treatment with each of the top candidate compounds ⁴. We sought to evaluate the impact of each of the top four inhibitors including PD325901 on stemness with an ultra-low dilution assay (ULDA) utilizing the HGSC cell lines OVCAR4 and COV504 (Fig. 3A). Single cell suspension was performed and cells were cultured in ultra low attachment plates, then treated with the inhibitors. The presence of spheroids was determined following treatment. Increased capacity to form spheroids from fewer cells implies more stem-like cells present in the population. Compared to DMSO-treated cells, alisertib and PD035901 reduced the spheroid-forming capacity and stem cell frequency (Fig. 3B-C, Fig. S2A). We also examined ALDH activity as a surrogate for stemness via ALDEFLUOR assay. Cells were treated with the IC₅₀ of each of the top five candidate molecules to identify the percentage of ALDH-positive cells, compared to DEAB-added cells as negative control (Fig. 3D). All five drugs reduced ALDH-positive cell frequency; however, alisertib reduced it to the greatest extent (Fig. S2B). Thus, taking all of the in vitro assays into account, we identified alisertib as the molecule most functionally similar to our CBX2 inhibitor. We next sought to demonstrate the relationship between alisertib and CBX2 activity in HGSC.

Figure 3. Treatment with alisertib leads to decreased stem cell frequency.

A) Protocol prototype for ultra-low dilution assay (ULDA) using OVCAR4 cells, including definition of spheroid positive vs negative. This was completed for the top 5 compounds: alisertib, raltitrexed, LY315920, GTX-007, and PD035901. B) Plot of percentage of wells without spheroid colonies against number of cells seeded initially and treated with the top five compounds. The greater the drop with treatment with a given agent, the greater decrease in stemness. C) Graph of percentage stem cell frequency after treatment with each of the primary compounds. Colored dots indicate matched experiments. D) ALDEFLUOR assay, measuring ALDH positivity in OVCAR4 cells treated with each of the primary compounds. Error bars, SEM. Statistical test, one way ANOVA with Dunnett multicomparison (adjusted p-values shown).

Confirmation of interaction and dependence between alisertib and CBX2 and polycomb repressor complex 1

Alisertib was developed to inhibit aurora A kinase (AURKA) ²³. The Cancer Dependency Map (https://depmap.org/portal/) was utilized to correlate alisertib sensitivity in ovarian cancer cell lines in relation to the mRNA expression of AURKA and all PRC1-associated CBX subunits (CBX2, CBX4, CBX6, CBX7, and CBX8) . The rho correlation values revealed that the efficacy of alisertib was significantly correlated with CBX2 expression (r = −0.52, p = 0.038), but not with AURKA expression or any other CBX subunit expression (Fig. 4A-B, Fig. S3A, Table 1). To experimentally confirm the dependency of alisertib effect on CBX proteins, we evaluated alisertib treatment response with modulation of each of the PRC1-associated CBX subunits. Using shRNA, we knocked down each of the CBX subunits, as well as AURKA. We observed that only knockdown of CBX2 impacted treatment response to alisertib (IC₅₀ shControl 73.93 nM vs shCBX2 >3000 nM) (Fig. S3B-C).

Figure 4. Alisterib activity is dependent on CBX2.

A) Dependency mapping of ovarian cancer cell line sensitivity to alisertib (y-axis, Area Under the Curve [AUC]) compared to AURKA mRNA expression. B) Dependency mapping of ovarian cancer cell line sensitivity to alisertib (y-axis, AUC) compared to CBX2 mRNA expression. C) FPLC fractionation of the chromatin compartment of OVCAR4 cells treated with alisertib or control. Protein used for immunoblotting against CBX2, CBX4, PHC3. Loading control, Actin. D) CBX2 knockout OVCAR4 cells were generated via CRISPR/Cas9, and knockout confirmed via immunoblot. Loading control, Actin. E) Alisertib dose response assay in OVCAR4 CBX2 Wildtype (CBX2WT) and OVCAR4 CBX2 knockout (CBX2KO). F) UItra-low dilution assay using alisertib in OVCAR4 CBX2WT and CBX2KO cells. Calculated stem-cell frequency graphed. G) OVCAR4 cells treated with increasing dose of alisertib. Protein lysates immunoblotted for CBX2 and H2AK119Ub. Loading control, Actin. H) OVCAR4 cells treated with two different aurora A kinase inhibitors, MNL8054 and VX680 . Protein blotted for H2AK119ub. Loading control, Actin. I) Chemical structure of alisertib. J) Overlap of CBX2 inhibitory peptide and alisertib. Spheres indicate essential residues. K) CBX2 inhibitory peptide and alisertib docked with the CBX2 protein. Error bars, SEM. Scale bar, 50 micron. Statistical test, Pearson Correlation (A,B), non-linear regression with sigmoidal dose-response analysis, 95% CI included (E), and unpaired t-test (F). L) Cell target engagement enzyme complementation signals for aurora A kinase, CBX2, and CBX4 constructs compared to background control at 37 °C. M) Potency analysis of aurora A kinase, CBX2 and CBX4 stabilization by alisertib. Nonlinear slope-based integration yielded an EC₅₀ of 1.6 μM and Hill slope of 0.83 for aurora A kinase, EC₅₀ of 2.2 μM and a steep Hill slope of 13.1 for CBX2, and EC₅₀ of 23.2 μM with a steep Hill slope of 15.9 for CBX4.

Table 1. Alisertib drug sensitivity correlated with the expression of CBX subunits and AURKA mRNA expression. Data obtained from Dep Map Sanger GDSC2 on July 7^th, 2025.

Gene	Pearson r (Alisertib Drug Sensitivity AUC vs. 17 ovarian cancer cell lines)	p-value
CBX2	−0.52	0.0328
CBX4	0.07	0.780
CBX6	0.10	0.718
CBX7	0.36	0.153
CBX8	−0.36	0.158
AURKA	−0.03	0.96

Chromobox proteins are understood to form complexes both within and outside of the polycomb repressor complex 1 (PRC1). We sought to assess the composition of CBX2-associated protein complexes following alisertib treatment. Using the chromatin isolate from OVCAR4 cells treated with alisertib or control, we performed Fast-Protein Liquid Chromatography (FPLC) fractionation of the native protein complexes. We observed that alisertib treatment led to a significant shift in the fractionation of CBX2, this was in contrast to two other PRC1 subunits, PHC3 and CBX4, where there was no shift in fractionation (Fig. 4C). These data suggest that treatment with alisertib leads to a shift specifically in CBX2-associated complexes.

To further define alisertib’s dependence on CBX2, we knocked out CBX2 using CRISPR/Cas9, and generated dose-response curves to alisertib (Fig. 4D-E). Notably, the knockdown and knockout of CBX2 resulted in slower growth of OVCAR4 cells, but in COV504 cells, CBX2 knockdown led to a modest increase in proliferation, and CBX2 knockout resulted in a minor loss of proliferation (Fig. S4A). We observed that the IC₅₀ of alisertib was ten times higher for CBX2 knockout (KO) cells compared to the CBX2 intact parental cells (CBX2 wildtype IC₅₀=77.7 nM vs. CBX2 KO IC₅₀=803.7 nM, Fig. 4E). In COV504, knockdown of CBX2 led to a 5.8-fold increase in the alisertib IC₅₀ (shControl IC₅₀=318 nM vs. shCBX2 IC₅₀=1832 nM) (Fig. S4B). We similarly examined a structurally distinct AURKA inhibitor, LY3295668, to understand if CBX2 inhibition was inherently linked to AURKA inhibition, however observed that efficacy of LY3295668 was independent of CBX2 (Fig. S4C-E).

We subsequently evaluated for stemness through ULDA and found that knocking out CBX2 attenuated the impact of alisertib on stemness (Fig. 4F). In OVCAR4 and COV504 CBX2 clonal knock out cells, we confirmed that knock out of CBX2 modestly reduced H2AK119ub, and did not change the expression of Aurora A kinase or PHC3, a PRC1 subunit (Fig. S4F-G). This highlights that reduced alisertib sensitivity was not due to CBX2-mediated changes in aurora A expression. These findings suggest that the functional impact of alisertib in high-grade serous cell lines is dependent on intact CBX2.

We then sought to determine whether alisertib inhibits the activity of the CBX2-associated PRC1. We used H2AK119ub to evaluate the impact of alisertib treatment on the activity of the broader PRC1 complex ⁴⁴. OVCAR4 and COV504 cells were treated with increasing doses of alisertib, and immunoblots were then performed to assess the subsequent levels of H2AK119ub. While H2AK119ub was inhibited with alisertib in a dose-dependent manner (Fig. 4G, Fig. S4H), other AURKA inhibitors did not inhibit H2AK119ub at the IC₅₀ of alisertib or their own IC₅₀ concentration (Fig. 4H).

Using computational-based molecular docking, we aimed to gain a deeper understanding of the structural interactions between alisertib and CBX2. We compared the docked conformations of alisertib (Fig. 4I) and CBX2i ¹², as well as the relationship between the CBX2-specific A/T hook region and chromodomain (CD) and alisertib (Fig. 4J-K). Interestingly, while alisertib did align to the CBX2i inhibitory peptide (Fig. 4J), it was ranked fifth in the molecular docking but seventh in the shape-based similarity screen (Fig. 1). The carboxylic acid group of the 2-methoxybenzoic acid of alisertib mimics the acetylated glutamate residue of CBX2i, making a hydrogen bond interaction with Lys61 of CBX2, whereas the terminal glutamate of CBX2i makes a hydrogen bond with the adjacent Lys94 of CBX2. The 2-methoxybenzoic acid and chlorophenyl rings of alisertib exhibit strong pi-cation interactions with Arg92, with the chlorophenyl ring forming an additional pi-cation interaction with Arg53 of CBX2. Although CBX2i displays many hydrogen bond interactions with CBX2 (including the glutamate of CBX2i with Arg92 around the CD-A/T Hook region) due to its amino acid backbone, CBX2i does not display these pi-cation interactions. In addition, the fluoromethoxyphenyl ring of alisertib occupies the area between Arg 92 and Arg34, around the CD-A/T Hook region, which is spatially occupied by the glutamate side chain of CBX2i.

Furthermore, upon examination of the docked compounds, a common feature observed with CBX2i and alisertib (Fig. 4K) is that their sidechains (the carboxylic acid group of Glu4 in CBX2i and 2-fluoro-6-methoxyphenyl in alisertib) occupy a cleft formed by Arg34 and Arg92 in CBX2 and display anchoring hydrogen bond interactions around this cleft that could be important for the further development of CBX2 inhibitors. This structural exploration demonstrates the potential interaction between CBX2 and alisertib including the potential for alisertib to occupy the specific A/T Hook region of CBX2.

Finally, a cell target engagement assay was performed to confirm selective binding. Aurora A kinase and CBX2 were evaluated with CBX4 as a control. This engagement assay found that alisertib selectively binds to both aurora A kinase and CBX2. All three S-tagged constructs resulted in similar expression levels (Fig. 4L). Thermal challenge yielded reproducible sigmoidal melting profiles for each target (Fig. S4I). Aurora A kinase exhibited the highest stability with a T-agg50 of 52°C, while CBX2 and CBX4 were less stable with T-agg50 values of 46°C and 45°C, respectively (Fig. S4I). From these profiles, 40°C, 45°C, and 50°C were identified as shared thermally sensitive points suitable for compound testing. Aurora A kinase displayed the strongest and most canonical stabilization in the presence of alisertib. Complementation data revealed clear dose-dependent engagement (Fig. 4M). Nonlinear slope analysis produced an aurora A kinase EC₅₀ of 1.6 μM and a Hill slope of 0.83 (Fig. 4M, S4J). These values align with previously reported ranges for aurora A kinase inhibition and confirm aurora A kinase as the primary intracellular target of alisertib under these conditions. CBX2 showed intermediate responsiveness to alisertib. Stabilization was evident in the complementation data, but the fitted potency (CBX2 EC₅₀ of 2.2 μM) was slightly weaker than that observed for aurora A kinase (Fig. 4M and Fig. S4K). Notably, the steep Hill slope of 13.1 (Fig. S4K-L) suggested a cooperative or threshold-like response, distinguishing its engagement profile from aurora A kinase. And, as expected, CBX4 displayed minimal engagement by alisertib, with weak stabilization signals evident only at the highest concentrations (Fig. 4M). The derived CBX4 EC₅₀ of 23.2 μM was an order of magnitude weaker than aurora A kinase or CBX2, and the Hill slope of 15.9 indicated a highly steep and atypical dose–response relationship (Fig. S4M). In summary, these data confirm that alisertib, while primarily an inhibitor of aurora A kinase, also inhibits CBX2 and its in vitro efficacy is dependent on the expression of CBX2.

Alisertib inhibits tumor growth in a CBX2-dependent manner in vivo

To further elucidate the dependence of alisertib on CBX2, an in vivo study was performed using a syngeneic mouse model with Cbx2 intact (shControl, n=16) compared to Cbx2 knock down (shCbx2, n=16) in mice treated with either alisertib or vehicle with eight mice per arm. ID8 (Trp53^−/− Brca2^−/−) cells were transduced with lentivirus containing shRNA of control (pLKO blastcidin resistant) or Cbx2, and transduced cells were selected with blastcidin and amplified to be injected 5 million cells/mice. Knockdown of Cbx2 was confirmed by qPCR before injection (Fig. 5A). shControl and shCbx2-tumor bearing mice were treated with either vehicle control or alisertib for 28 days, after which mice were euthanized and necropsy was performed (Fig. 5B). Treatment was well tolerated based on mouse weight and a comprehensive hematologic evaluation (Fig. S5A-B). Primary (omental) and secondary (disseminated) tumors were quantified. Further, tumor specimens were collected and immune cell composition was determined by flow cytometry (Fig. S6).

Figure 5. Alisertib anti-tumor activity is dependent on CBX2.

A) ID8 p53−/−, Brca2−/− cells were transduced with shControl (n=8) or shCbx2 (n=8). Cbx2 expression was confirmed via qPCR. Internal control, HPRT. B) ID8 shControl and ID8 shCbx2 were used for an in vivo study. C) At the time of necropsy, omental weight, D) number of dissemination sites, and E) disseminated tumor weight were measured. F) Tumors from the in vivo study were used for immunohistochemistry (IHC) against Cbx2. G) quPATH analysis of the Cbx2 intensity per tumor cell. H) Tumors from the in vivo study were used for IHC against H2AK119Ub. I) quPATH analysis of the H2AK119Ub signal per tumor cell. J) quPATH analysis of the Ki67 positive signal per cell. Average shown for each tumor. K) Tumors from the in vivo study were used for IHC against Ki67. L) quPATH analysis of the cleaved caspase 3 (CC3) positive signal per cell. Average shown for each tumor. M) Tumors from the in vivo study were used for IHC against (CC3). Error bars, SEM. Statistical Test, unpaired t-test (A) and two-way ANOVA (n = 8; C, D, E, G).

As previously reported ³, Cbx2 knockdown reduced omentum (Fig. 5C) and disseminated tumor weights (Fig. 5E) as well as the number of individual metastases, or dissemination sites (Fig. 5D) (Two-way ANOVA, n=8; p = 0.0078, 0.0043, and 0.0072, respectively). We also found that alisertib treatment reduced these measures of tumor burden (Two way ANOVA, n=8; all p < 0.0001). Still, this effect depended on the presence of Cbx2 (Two way ANOVA interaction, n=8; all p < 0.0001), as tumor weights and number of dissemination sites were largely equivalent in vehicle- and alisertib-treated shCbx2 mice. Disease proliferation was also similar if mice received vehicle control or the tumor had Cbx2 knockdown and received treatment with alisertib, and the anticipated treatment effect of alisertib was only seen in mice with intact Cbx2 (Fig. 5C-E). Furthermore, immunohistochemistry (IHC) studies for CBX2 (Fig. 5F) and H2AK119ub (Fig. 5H) were performed to evaluate intensity in the setting of shCbx2 and after treatment with alisertib. CBX2 IHC confirmed that shCbx2 tumors and alisertib-treated tumors, compared to shControl tumors, retained their knockdown and had reduced Cbx2 expression, respectively. In comparison to shControl tumors, there was a 3.8-fold reduction in H2AK119ub in tumors treated with alisertib and a 1.9-fold decrease in shCbx2 tumors (Fig. 5F-I). However, in comparing shCbx2 tumors treated with control versus alisertib, there is no additional loss of the H2AK119ub signal (Fig. 5F-I). Next, we assessed alisertib- and Cbx2-dependent proliferation and apoptosis. Based on Ki67 signal, treatment with alisertib and loss of Cbx2 led to a similar decrease in signal (Fig. 5J-K). Further, the combination of alisertib and Cbx2 loss did not result in any additional loss of Ki67 signal. In assessing apoptosis, the shCtrl alisertib-treated tumors displayed a significant increase in cleaved caspase 3 signal (Fig. 5L-M). However, the loss of Cbx2 expression attenuated the alisertib-dependent increase in cleaved caspase 3. Collectively, these data provide further evidence that alisertib effectively reduces HGSC development in a CBX2-dependent manner.

Treatment with alisertib leads to a shift in the tumor myeloid compartment

Modulation of CBX2 was recently observed to regulate the myeloid compartment of ovarian cancer tumors³. Thus, using flow cytometry analysis of the tumor, we sought to determine the impact of alisterib-mediated CBX2 inhibition on the tumor immune microenvironment. Using shControl and shCbx2 tumors treated with vehicle or alisertib (4 samples per treatment group), we focused on the myeloid compartment (Fig. 6A, gating strategy Fig. S6). We did not observe significant shifts in the lymphocyte compartment across the different genetic and treatment group (Fig. 6B). However, in the shControl tumor, alisertib led to a significant reduction of myeloid-derived cells compared to the vehicle-treated tumors (Fig. 6C). The reduction in the myeloid cells was also observed in the Cbx2 knockdown condition, but alisertib did not further reduce myeloid cells (Fig. 6C). Within the myeloid compartment, the treatment with alisertib and Cbx2 knockdown often resulted in similar shifts (Fig. 6D-H). Specifically, alisertib and/or Cbx2 knockdown did not shift the M2-like macrophage population but did result in reduced M1-like macrophages (Fig. 6F-G). Also, alisertib and/or Cbx2 knockdown similarly led to a reduction in immune suppressive PD-L1-expressing macrophages (Fig. 6H). These data recapitulate prior observations of CBX2-mediated myeloid remodeling. Notably, immune remodeling in the context of loss of CBX2 and alisertib treatment may be attributed to AURKA targeting. Overall, we show that CBX2 loss and alisertib treatment result in similar levels of remodeling.

Figure 6. Alisertib-mediated myeloid remodeling is dependent on CBX2.

A) ID8 shControl and ID8 shCbx2 tumors were used for immune profiling of the myeloid compartment. B) % Lymphocytes of total immune cells. C) % Myeloid cell of total immune cells. D) % Monocytes of myeloid cells. E) % Macrophages of myeloid cells. F) % of M1 of total macrophages. G) % of M2 of total macrophages. H) % of PD-L1+ macrophages of total myeloid. Error bars, SD. Statistical test, one-way ANOVA with Tukey multicomparison test (n = 4; C-F, H).

DISCUSSION

Endogenous CBX2 expression is limited to specific tissues, and the fact that tumor cells show high expression of CBX2 leads us to consider CBX2 as an promising and potentially ideal target for therapeutic development in HGSC, as well as other solid malignancies, including breast, colorectal, and prostate cancers. While we have demonstrated in vivo efficacy of an inhibitory peptide targeting CBX2, we sought to accelerate translation of this work using existing small molecule compounds. We conducted a tandem virtual and biosimilarity screen using the CBX2 inhibitory peptide as a template to explore existing therapeutic compounds. We identified several potential small molecule CBX2 inhibitors. Validation studies then narrowed the top hits by their inhibitory concentration and inhibitory effectiveness on downstream target. Across the stem cell assays we confirmed that treatment with alisertib had the most significant impact on reducing stemness and H2AK119ub, which closely aligns with known PRC1 and CBX2 activity .

To confirm the inhibitory effect of alisertib was dependent on CBX2, we knocked down CBX2 by shRNA or knocked out by CRISPR in the HGSC cell lines, and found that CBX2 loss attenuates alisertib treatment response. Alisertib showed a clear rank-order selectivity profile with aurora A kinase as primary target, CBX2 as secondary, and no selectivity for CBX4 was noted. This was explored in vivo utilizing a syngeneic murine model of HGSC (ID8 Trp53^−/− Brca2^−/−), and it was confirmed that knockdown of CBX2 attenuates alisertib’s anti-tumor efficacy in mice. Flow cytometry on tumor specimens demonstrated a shift in the tumor immune microenvironment, with a myeloid predominance. This is consistent with the previously described CBX2-mediated remodeling of the myeloid compartment. Moreover, treatment with alisertib was found to not only decrease expression of CBX2 but also decrease ubiquitination of HistoneH2A Lys119, a marker of PRC1 activity. The FPLC fractionation of the chromatin fractionation supports the hypothesis that alisertib modulates CBX2-associated protein complexes. Taken together, the data suggest that alisertib impacts the ability of CBX2 to read, thus destabilizing the PRC1 complex.

Targeting PRC1 in cancer is not a novel concept. PTC596, a BMI-1 inhibitor, as been trialed in many solid tumors (NCT02404480)^45-47 and most recently has been under investigation in HGSC in combination with neoadjuvant chemotherapy and to follow as maintenance (NCT03206645), but results have not yet been reported. BMI-1, while dramatically associated with poor outcomes^48,49, also tends to be more widely expressed among fallopian tube cells, and therefore, the benefit of targeting CBX2 may be that it is far more selective than BMI-1. For example, in a single cell analysis of benign fallopian tube cells, CBX2 expression is restricted to secretory epithelial cells, the cell of origin for HGSC, whereas BMI-1 is expressed in all cells ⁵⁰, suggesting that CBX2 may be a preferable target.

Alisertib (MLN8237) is an aurora A kinase inhibitor. Inhibition of aurora A kinase leads to disruption of mitotic spindle development and thus inhibits cell proliferation ⁵¹. Overexpression of aurora A kinase has been associated with tumor progression and poor prognosis in various cancers, including ovarian cancer ^52-56. Alisertib is orally available, well tolerated, and FDA approved with orphan drug status for small cell lung cancer ⁵⁷. It has been explored in triple negative breast cancer and head and neck cancer ⁵⁷, as well as recurrent ovarian cancer. In high-grade serous carcinoma, a phase I/II study of alisertib with weekly paclitaxel demonstrated moderate improvement in progression-free survival with addition of alisertib to paclitaxel over paclitaxel alone ⁵², however this raises the question of whether there are patients, based on biomarker status, who may benefit from alisertib.

CBX2 expression is quantifiable based on IHC, and the majority of HGSC cases exhibit high expression of CBX2. We are actively working to optimize CBX2 IHC for its utilization as a biomarker. Based on the data we have shared, a tumor with high CBX2 expression may be more likely to respond to treatment with alisertib. However, additional molecular aberrations, including homologous recombination status⁴⁷ or cyclin E amplification^58,59, may need to be considered in parallel to CBX2, given their impact on treatment outcome and prognosis. The establishment of CBX2 as a biomarker may allow for optimal utilization of alisertib in HGSC. Given the role of CBX2 inhibition in decreasing stemness and escape from anoikis, the best strategy for alisertib may be in the maintenance setting following chemotherapy. Understanding the critical need for a maintenance strategy in homologous recombination proficient (HRP) tumors, our next phase of work will focus on HRP and chemoresistent models, both in vitro and in vivo, to prepare for clinical trial development.

Ultimately, our biosimilarity screen has identified alisertib as an inhibitor of CBX2, and our experimentation confirmed that alisertib efficacy is dependent on intact CBX2 expression. Given the known tolerability of alisertib, as well as the fact that it is orally available and already FDA approved for small cell lung cancer, alisertib may be an optimal alternative approach to CBX2 inhibition, which will allow for accelerated translation and evaluation via clinical trial.

Translational Relevance:

Chromobox 2 (CBX2) drives progression and poor outcomes in high-grade serous carcinoma (HGSC) and CBX2 inhibition has been demonstracted to decrease tumor dissemination in a HGSC syngeneic murine model. To accelerate translational potential of these findings, this biosimilarity screen of the CBX2 inhibitor was performed, identifying aurora A kinase inhibitor, alisertib, as an existing compound that inhibits CBX2.

Alisertib is a known therapeutic agent that is orally available, well-tolerated, and holds FDA approval for small cell lung cancer. Alisertib has been previously evaluated in platinum resistant ovarian cancer with modest activity as a single agent and improvement in progression free survival in combination with paclitaxel, compared to paclitaxel alone. Better understanding the mechanism of alisertib is an opportunity to optimize a known therapeutic with a novel approach in HGSC. This work lays the foundation for CBX2 as a biomarker and development of mechanism-driven clinical trials to exploit potential vulnerabilities in HGSC.