Dual CDK4/6–PI3K/mTOR inhibition reinforces cytostatic programs and tumor control in preclinical models of primary and metastatic osteosarcoma

Highlights

• •Dual CDK4/6–PI3K/mTOR inhibition reinforces cytostatic programs in osteosarcoma.
• •Palbociclib induces senescence, while PI3K/mTOR blockade enhances autophagy.
• •Combination therapy suppresses compensatory PI3K signaling in osteosarcoma PDXs.
• •Dual inhibition improves tumor control and survival in metastatic osteosarcoma models.
• •CDK4/6 inhibition suppresses osteosarcoma lung colonization in vivo.

Abstract

Osteosarcoma (OS) in pediatric, adolescent, and young adult (AYA) patients is an aggressive bone cancer with limited treatment options. Dysregulation of the CDK4/6–cyclin D axis and the PI3K/mTOR pathway contributes to OS pathogenesis, providing a biological rationale for co-targeting these signaling nodes. However, pharmacologic CDK4/6 inhibition can trigger compensatory activation of the PI3K/mTOR pathway, restoring D-type cyclin expression and partially reactivating CDK4/6 signaling. Thus, dual inhibition of the CDK4/6 and PI3K/mTOR pathways not only addresses two parallel oncogenic drivers but may also prevent potential CDK4/6 inhibitor resistance mediated by feedback activation of PI3K/mTOR. In this study, we tested the hypothesis that coordinated targeting of these pathways would improve tumor control in preclinical OS models. In vitro sensitivity analyses using palbociclib and voxtalisib demonstrated additive to synergistic OS growth suppression, with palbociclib inducing G1 arrest and senescence, and the combination enhancing autophagy. Furthermore, the efficacy, tolerability, and mechanisms of palbociclib and voxtalisib, alone or in combination, were evaluated in molecularly defined primary treatment-naïve, and relapsed/metastatic OS models. In the relapsed/metastatic PDX77-TT2 model, short-term palbociclib exposure activated PI3K/mTOR signaling, whereas the combination of palbociclib and voxtalisib in long-term studies produced marked tumor suppression and extended survival. In the primary treatment-naïve PDX96 model, long-term palbociclib exposure generated a robust CDK4/6 pharmacodynamic response. The addition of voxtalisib reinforced autophagy, sustained CDK pathway inhibition, and improved overall tumor control. In an OS lung-colonization model, CDK4/6 inhibition alone markedly reduced OS lung nodules, with combination therapy providing comparable suppression. Dual CDK4/6–PI3K/mTOR inhibition achieves tumor control across various OS models, supporting the use of genomically guided, pathway-targeted strategies for pediatric and AYA OS.

Graphical abstract

Introduction

Osteosarcoma (OS) is the most frequent type of primary bone cancer among adolescents, young adults (AYA), and children [1]. The five-year survival rate for patients with metastatic disease remains below 20 %, compared to approximately 60 % for those with localized tumors [1]. At diagnosis, 25 %−50 % of OS patients present with metastatic disease, predominantly in the lungs, even while on frontline chemotherapy [1]. While the National Comprehensive Cancer Network (NCCN) provides recommended second-line therapeutic options for aggressive OS, no standardized or reliably curative therapies exist [1,2]. Thus, there is a critical need for better therapeutic options for treatment-naïve and relapsed/metastatic OS patients. Precision medicine approaches have the potential to identify actionable molecular targets in pediatric and AYA OS, thereby improving patient stratification for better therapeutic efficacy. For example, dysregulation of the Cyclin D–CDK4/6 signaling is a potential therapeutic vulnerability that has been identified in a subset of pediatric and AYA OS [3,4]. The Cyclin D-CDK4/6 complex regulates interactions between the E2F transcription factors and the retinoblastoma (RB1) protein [4]. Hypophosphorylated RB1 binds E2F transcription factors, repressing E2F-driven gene expression and thereby restraining G1 progression and preventing entry into S phase [4]. Following Cyclin D-CDK4/6-mediated RB1 phosphorylation, RB1 affinity for E2F is reduced, promoting the transcription of genes such as Cyclin E2 [4]. Hyperphosphorylation of RB1 by Cyclin E2–CDK2 frees E2F to drive expression of Cyclin A2, Cyclin B1, and other S/G2-phase genes [5]. Notably, cancer cells can bypass the RB1-dependent restriction point during the G1 phase of the cell cycle, typically through alterations in cell cycle machinery genes that activate Cyclin D–CDK4/6 constitutively or through the loss of RB1 [4]. Cyclin D–CDK4/6 axis dysregulation is also often caused by the loss of the endogenous CDK4/6 inhibitor, p16^INK4A, which serves as a biomarker of CDK4/6 activation in cancer cells [4].

CDK4/6 inhibitors (CDK4/6i) have emerged as key therapeutic agents across multiple cancer types due to their central role in regulating cell-cycle progression [4]. As in many other cancers, preclinical evaluation of CDK4/6i (palbociclib) as a monotherapy has not been as effective as hoped in OS [6]. The lack of therapeutic efficacy of CDK4/6i is likely due to its cytostatic mechanism of action [4] and activation of growth and survival pathways. Resistance to CDK4/6 blockade has also been reported in various cancers and is often driven by activation of compensatory signaling pathways, such as PI3K/mTOR and MAPK, especially in breast cancer [7]. Additionally, the PI3K/ mTOR pathway is already frequently hyperactivated in OS, contributing to its pathogenesis, making it an attractive target for combination therapy. Therefore, in pediatric and AYA OS with hyperactive CDK4/6 signaling or related actionable alterations, therapeutic strategies that concurrently inhibit CDK4/6 and PI3K/mTOR may enhance treatment efficacy and overcome adaptive resistance.

As such, Oshiro et al. investigated the efficacy of palbociclib in combination with the mTOR inhibitor everolimus in a primary OS orthotopic patient-derived xenograft (PDX) model, highlighting the potential for pathway co-targeting and laying the groundwork for broader investigation [6]. In this study, while the palbociclib+everolimus combination modestly decreased the tumor-volume ratio compared to the untreated control or doxorubicin, statistically significant differences between single-agent everolimus versus the palbociclib+everolimus combination were not demonstrated [6]. PI3K or mTOR inhibition alone may be insufficient, as blocking one node can activate compensatory signaling through the other or shift pathway dependency [8,9]. This supports using small-molecule inhibitors that target both PI3K and mTOR simultaneously. Dual targeting also suppresses AKT activation driven by the mTORC1–S6K–IRS1 negative-feedback loop [8,9]. Consistent with this rationale, a combination of CDK4/6i plus PI3K/mTORi, is in clinical testing, especially in the context of advanced/metastatic breast cancer being treated with hormone therapy (NCT02684032, NCT06757634). We prioritized voxtalisib due to its potent PI3K/mTOR inhibition in our in vitro functional screens as well as favorable tolerability in other cancers [9,10].

In our study, dual inhibition of CDK4/6 and PI3K/mTOR by palbociclib + voxtalisib was evaluated in a panel of human OS cell lines, and in PDX with molecular signatures indicative of CDK4/6i sensitivity (RB1+, CDKN2A null, CCND3 amplified) [11]. In OS lines, palbociclib + voxtalisib produced additive-to-synergistic cell growth inhibition. Combination palbociclib + voxtalisib was well tolerated, demonstrated antitumor efficacy, and enhanced the therapeutic activity of palbociclib in PDX models established from OS specimens from pretreated relapsed/metastatic and primary treatment-naïve OS patients. Kinome profiling analysis of a primary treatment-naïve PDX demonstrated that, compared to single agents, palbociclib + voxtalisib combination therapy significantly decreased PI3K pathway activity. Further, palbociclib monotherapy primarily induced senescence with modest autophagy, whereas in combination with voxtalisib, autophagy was enhanced without significantly altering senescence. In a metastatic OS lung-colonization model, palbociclib, either as monotherapy or in combination with voxtalisib, was the principal driver of reduced metastatic foci, further underscoring the therapeutic potential of CDK4/6 inhibition in aggressive OS. Taken together, these data suggest that while palbociclib remains the dominant driver of growth and metastatic suppression, co-targeting PI3K/mTOR with voxtalisib enhances these outcomes by more effectively suppressing PI3K signaling and reinforcing cytostatic responses through enhanced autophagy, supporting this combination as a rational therapeutic approach in OS.

Materials and methods

Cell lines

Human osteoblasts-femoral were purchased from (HO-F; CP4610 ScienCell Research Laboratories, Carlsbad, CA, USA) cultured in Osteoblast Growth Medium, according to the manufacturer’s instructions. We acquired the following established pediatric OS cell lines from American Type Culture Collection (ATCC; Manassas, VA) in 2018: G292, MG63, U2OS, 143B, and Saos-2. The lung OS metastatic cell line, Saos-LM7 [12], was kindly provided by Dr. Eugenie S. Kleinerman (MD Anderson, Houston, TX, USA). We gratefully acknowledge Dr. Ed Greenfield from IUSM for generously providing us with the lung OS metastatic cell line MG63.3 [13]. Notably, previously developed OS xenoline referred to as TT2 xenoline [14,15], was also used in this study. Summary of cell-line and donor demographics are highlighted in Supplementary Table 1. DNA fingerprinting analysis using a nine-marker short-tandem repeat (STR) analysis (IDEXX BioResearch, Columbia, MO, USA) was used to authenticate for their identity, as previously described [14,15]. They were reported to be 100 % human. STR confirmed identity compared with known STR patterns for the cell lines. All cell lines were mycoplasma-free (MycoAlert Kit; Lonza, Morristown, NJ, USA) and cryopreserved. All cell lines were cultured in DMEM supplemented with 10 % FBS and 1 % HEPES at 37 °C in a humidified atmosphere containing 5 % CO₂.

Compounds

Palbociclib (CDK4/6i), voxtalisib (PI3K/mTOR inhibitor; PI3K/mTORi), (ChemieTek, Indianapolis, IN, USA), were dissolved in 100 % dimethyl sulfoxide (DMSO) for in vitro studies. ARV825 was purchased from ChemieTek, CT-ARV825 (Indianapolis, IN, USA). For all drugs in all cell cultures, the final DMSO concentration was 0.1 % or less. Palbociclib and voxtalisib were dissolved in 10 mM HCL for the in vivo studies.

Cell proliferation assay

In 96-well plates, OS cell lines were seeded overnight at 2500-3500 cells/well except for MG63.3 cell line which was seeded at 250 cells/well. The next day, cells were treated at various concentrations of palbociclib, and voxtalisib, as single agents for five days (120 h). As previously described, methylene blue staining was used to determine cell growth/proliferation after the five-day treatment [16]. Three replicates of each experiment were conducted in sextuplicate. IC₅₀ values were calculated by fitting dose-response curves using nonlinear regression analysis to determine the concentration that inhibits 50 % of cell viability.

Western blotting

Lysates from cells and tumors were prepared in RIPA buffer supplemented with EDTA-free Complete Protease Inhibitor Cocktail (Roche, Basel, Switzerland) and 1 % Phosphatase Inhibitor 3 (Millipore/Sigma), which inhibits protein phosphatase 2A, alkaline phosphatases, and protein phosphatases 1 and 2A. Some lysates were prepared in Urea buffer (8 M urea, 50 mM Tris, pH7.5 and 5 mM DTT) [15]. Protein concentration was quantified using the DC Protein Assay (Bio-Rad, Hercules, CA, USA) for cell lysates and RCDC Protein Assay (Bio-Rad) for tumor lysates, and measured with a BioTek Synergy H4 plate reader. Proteins were separated on TGX Stain-Free gels (Bio-Rad) and transferred to LF PVDF membranes using the Trans-Blot Turbo Transfer System (Bio-Rad). Membranes were blocked with 5 % non-fat dry milk in TBS-T for 1 h. After antibody incubation, membranes were washed three times in TBS-T. Molecular weights were confirmed using Precision Plus All Blue Standards (Bio-Rad). Antibodies were diluted in 5 % non-fat dry milk or 5 % BSA in TBS-T and primary antibodies used for detection listed in Supplementary Table 2. The appropriate horseradish peroxidase-conjugated secondary antibody was then diluted 1:5000 in dry milk (5 % non-fat) at room temperature, for one hour (Promega, Madison, WI, USA; Anti-mouse IgG HRP-conjugate, cat#: W4021; Anti-rabbit IgG HRP-conjugate, cat#: W4011). After antibody incubation with TBS-T, membranes were washed three times for 10 min each. Membranes were again washed following secondary antibody incubation with TBS-T three times for 10 min each. The SuperSignal Western Chemiluminescent Substrate (Thermo Scientific, Grand Island, NY, USA) was used for protein detection and the Imaging System (Bio-Rad ChemiDoc) for image evaluation (Bio-Rad). The Image Lab software (Bio-Rad) was used to quantify protein expression, and proteins of interest were normalized to total protein levels, as quantified from the corresponding blot, and expressed in relation to the control cells (HO-F), media or cells treated with vehicle, or non-targeting control siRNA. Phosphorylation levels of prioritized proteins were also quantitated and normalized to their total protein levels.

Transient knockdown of RB1 with silencing RNA (siRNA) transfection

ON-TARGET plus non-targeting control and SMARTpool ON-TARGET plus RB1 siRNAs were purchased from Horizon Discovery (Waterbeach, UK). Non-targeting control siRNA sequences were D-001810-10-05: UGGUUUACAUGUCGACUAA, UGGUUUACAUGUUGUGUGA, UGGUUUACAUGUUUUCUGA, and UGGUUUACAUGUUUUCCUA. RB1 siRNA sequences from SMARTpool were Pool 1, J-003296-23: GAACAGGAGUGCACGGAUA; Pool 2, J-003296-24, GGUUCAACUACGCGUGUAA Pool 3, J-003296-25, CAUUAAUGGUUCACCUCGA; Pool 4, CAACCCAGCAGUUCGAUAU. TT2 and G292 cells were seeded overnight in 10 cm Petri dishes, and the following day transfected with either 100 nM non-targeting control siRNA or 100 nM RB1 siRNA. RB1 protein levels were evaluated by western blot at 48 and 120 h post-transfection to confirm knockdown. For cell growth assays, transfected TT2 and G292 cells were plated in 96-well plates and one day after transfection, cells were treated with either vehicle or palbociclib for five days to assess RB1 knockdown effects on palbociclib-mediated cell growth inhibition.

Cell cycle analysis by flow cytometry

TT2 and G292 cells were treated with a vehicle (DMSO) or 0.15 µM palbociclib and 0.5 µM voxtalisib, as single agents or in combination. After different time points including, 24 and 48 h, cells were collected and fixed in cold 70 % ethanol and stored at 4 °C for 30 min. Cells were then washed in cold PBS and were incubated with DNase-free RNaseA (100μg/mL) (Thermo Fisher Scientific #P3566), and Propidium Iodide (PI) (20 μg/mL) (Thermo Fisher Scientific, Waltham, MA, USA, cat#: AM2270). Cellular DNA content was analyzed on the BD Fortessa (BD Bioscience, Franklin Lakes, NJ, USA) flow cytometer analyzed with FlowJo software.

Apoptosis analysis

G292 cells were treated with a vehicle (DMSO) or 0.15 µM palbociclib and 0.5 µM voxtalisib, as single agents or in combination. TT2 cells were treated with vehicle, 0.2 µM palbociclib and 1 µM voxtalisib, as single agents or in combination. After various time points (24-96 h) cells and media were collected and washed twice in cold PBS, and 200ul of 1:10 diluted Annexin V Binding Buffer was added to the cells. In the dark, each cell was stained for 15 min with 5 uL FITC Annexin V (BD Pharmingen, Franklin, NJ, USA, cat#: 556420) plus PI staining Solution (Invitrogen, Grand Island, NY, USA, cat#: 00699050;). After 400ul of 1:10 diluted Annexin V Binding Buffer (BD Pharmingen, cat#: 556454) was added to each tube, flow cytometry was performed on the BD Fortessa (BD Bioscience) analyzed with FlowJo software.

Senescence flow cytometry

TT2 and G292 cells were treated with a vehicle (DMSO) or 0.15 µM palbociclib and 0.5 µM voxtalisib, as single agents or in combination. After five-days (120 h) of treatment, cells were washed with culture media and treated with 1:1000 diluted Bafilomycin A1 (FastCellular™ Senescence Detection Kit, MP Biomedicals, Irvine, CA, USA) in culture media at 37 °C plus 5 % CO₂ for an hour to inhibit endogenous β-galactosidase activity. Cells were then stained with 1:1000 diluted SPiDER-βGal (MP Biomedicals, FastCellular™ Senescence Detection Kit, cat#: 76337-152) and incubated at 37 °C plus 5 % CO₂ for 30 min. Subsequently, cells were collected and washed twice with cold PBS, resuspended in 250 ul PBS, and flow cytometry was conducted on the BD Fortessa (BD Bioscience) analyzed with FlowJo software.

Autophagy analysis via flow cytometry

TT2 and G292 cells were treated with a vehicle (DMSO) or 0.15 µM palbociclib and 0.5 µM voxtalisib, as single agents or in combination for five days (120 h). Cells were then washed and resuspended in fresh culture medium, stained with 50x Autophagy Probe Red according to the manufacturer’s instruction (Bio-Rad, Autophagy Assay Kit, Red, cat#: APO010B) at 37 °C protected from light for 30 min. Afterward, cells were washed in PBS containing 5 % BSA three times, resuspended in 500 ul of PBS containing 5 % BSA, and on the BD Fortessa (BD Bioscience), flow cytometry was performed and analyzed with the FlowJo program subsequently.

NOD.Cg-Prkdc Scid Il2rgtm1Wjl/SzJ (NSG) mice

NSG mice were obtained from the on-site breeding colony maintained by the Preclinical Modeling and Therapeutics Core (PMTC) at the IUSCCC. All procedures were approved by the Institutional Animal Care and Use Committee (IACUC) (protocols: 22028, 25041). Animals were maintained under pathogen-free conditions and maintained on a Teklad Lab Animal Diet (TD 2018SX, Inotiv, Indianapolis, IN, USA) with ad libitum access to reverse osmosis (RO) water under a 12 h light-dark cycle at 22–24 °C.

Development of PDXs from pediatric patients with aggressive and primary OS

PDX77-TT2 and PDX96 were established from OS samples derived from male pediatric patients under IRB # 1501467439 [14,15]. PDX77-TT2 is a pretreated OS model developed from a metastatic site in the pelvis of a relapsed/metastatic 18-year-old male patient [15]. Notably, TT2 xenoline mentioned above in the OS cell lines subsection was derived from this PDX. PDX96 is from a treatment-naïve OS model developed from a 9-year-old male OS patient with primary tumors in the right proximal femur [15]. Whole genome sequencing (WGS) was performed on the original patient tumors as well as on the serially passaged PDXs [15]. Notably, the models were confirmed to recapitulate the molecular landscape of the original tumor [15]. PDX77-TT2 and PDX96 were expanded into larger cohorts for archiving and in vivo studies. Both PDXs were authenticated by STR analysis (IDEXX). For in vivo studies, once tumor volumes reached 1000 mm³, tumor fragments (2 mm × 2 mm) were prepared, placed in 1 % FBS/DMEM, and implanted almost immediately into the subcutaneous flanks of male NSG mice without Matrigel or gelatin.Tumor growth was monitored with an electronic caliper, and mice with consistent tumor growth kinetics were randomized when tumors reached 100–200 mm³ in control and treated groups. Mice were treated with five-day cycles of palbociclib, voxtalisib, or combination therapy by oral dosing (PO), once-a-day (SID). See figure legends for specific details on each in vivo experiment.

CDKN2A/B loci characterization and copy number profiling from WGS

As previously reported, WGS of PDX77-TT2 and PDX96 revealed loss of the CDKN2A/B locus in both models [15]. Because CDKN2A/B encodes key cell-cycle regulators (p16^INK4A and p14^ARF) and p16^INK4A is a biomarker of CDK4/6i sensitivity, we performed high-depth genomic profiling of this region to precisely define the extent and structure of the deletions at this locus. FastQC was utilized to check the sequencing quality of PDX77-TT2 and PDX96. Reads were aligned to human genome hg38 by BWA-MEM2 and subsequently sorted with Samtools. Duplicate reads were marked with the Picard MarkDuplicates tool in GATK, and base quality scores were recalibrated using the BaseRecalibrator and Apply Recalibration tools in GATK. Copy number variation analysis was performed with CNVkit v0.9.12. Briefly, the “target” command of CNVkit was used to partition the accessible region of hg38 into 1-kb windows, followed by the “coverage” command summarizing read coverage within each window. A reference coverage profile for these windows was generated using the hg38 fasta file. The “fix” command was then applied to the coverage files of PDX77-TT2 and PDX96, respectively, to calculate copy number ratios (CNRs) relative to the reference genome. Finally, CNRs spanning CDKN2A and CDKN2B loci were extracted to examine copy number changes across their exons in both PDX77-TT2 and PDX96.

Development of pre-established metastatic OS lung foci

A human OS lung colonization model was established using MG63.3 cells. Six- to eight-week-old male NSG mice were injected with 2 × 10⁵ cells via the lateral tail vein (n = 7 mice/treatment group). After one week, mice were treated with single-agent palbociclib, voxtalisib, or combination therapy PO, SID. See figure legend for details.

Body weight was measured twice weekly, and predeath endpoint defined based on a clinical body-scoring system that monitors animal well-being, including assessments of activity, grooming, posture, respiration, prolapse, and skin condition. For metastatic foci analysis, lungs were collected when the first mouse reached the predefined endpoint. This occurred in most vehicle-treated mice at 31 days post-injection of the MG63.3 cells. One lung lobe from each mouse was flash-frozen and analyzed by paired quantitative real-time PCR using human- and mouse-specific primers, as previously described [17]. Additional lobes were formalin-fixed and paraffin-embedded (5 μm sections) for histopathological and immunohistochemical evaluation. Sections were stained with hematoxylin and eosin (H&E) or immunostained for human HLA to detect metastatic OS cells.

For HLA class I (ABC) immunohistochemistry (IHC) analysis, sections were quantified using HALO’s cytonuclear mask algorithm to determine the percentage of HLA-positive cells, scored as negative (blue), weak (yellow, 1+), moderate (orange, 2+), or strong (red, 3+). Histopathological assessment of H&E-stained sections was performed using HALO AI DenseNet, classifying metastatic lesions (red) versus normal lung tissue (blue), and calculating the percentage of metastatic tissue relative to total lung area. Statistical comparisons were performed using one-way ANOVA with Tukey’s multiple-comparisons test, and error bars represent the standard error of the mean (SEM).

DNA extraction

Genomic DNA was extracted from flash-frozen lung tissue obtained from NSG lung-colonized mice using either the QIAamp DNA Mini Kit (Qiagen, Germantown, MD, USA) or the DNeasy Blood and Tissue Kit (Qiagen, Germantown, MD, USA, cat#: 69504), following the manufacturer's instructions. The DNA quantity was then evaluated using nanodrop, QUBIT Fluorometer, and Agilent 2100 Bioanalyzer.

Quantitative PCR (qPCR)

qPCR was performed on 50 ng template DNA using PerfeCTa SYBR Green SuperMix (QuantaBio, Beverly, MA, USA, cat#: 84020,) according to manufacturer’s instructions and run on a CFX96 Real-Time System (Bio-Rad). PCR program conditions were denaturation at 95 °C for 10 min, followed by 40 cycles, 95 °C for 1 min, 56 °C for 45 sec, and 72 °C for 45 se. Human specific primers used: Forward 5′-CTGTTTTGTGGCTTGTTCAG-3′, Reverse 5′-AGGAAACCTTCCCTCCTCTA-3′. Mouse specific primers used: Forward 5′-TTGGTTGAGAAGCAGAAACA-3′, Reverse 5′-CACACAGTCAAGTTCCCAAA-3′. Data was analyzed to determine percentage of DNA template from human metastases versus mouse lung described previously [17].

Kinome profiling analysis

Custom-synthesized kinase inhibitors (CTx-0294885, VI-16832, PP58, Purvalanol-B, UNC-21474, and UNC-8088A) were linked to sepharose beads. These multiplexed inhibitor beads (MIBs) are a blend of Type I kinase inhibitors that target kinases in the "Asp-Phe-Gly (DFG)-in" conformation, capturing at least 90 % of the kinome [18,19]. MIB binding is used as a read-out for kinase activity, expression/abundance, and affinity [18]. PDXs were treated with 50 mg/kg palbociclib and 50 mg/kg voxtalisib, alone or in combination, or with DMSO for specific durations. After lysis, total protein lysates (2 mg) were flowed over the inhibitor-linked beads, followed by high and low salt washes. Eluted proteins were trypsinized and analyzed by LC-MS using MaxQuant software, searching against the Uniprot/Swiss-Prot database. Data were processed as fold-change relative to DMSO controls, with label-free quantification (LFQ). Principal component analysis was performed using Perseus software. For drug response LFQ, missing values were imputed, and two-sample tests were performed with permutation-based false discover rate (FDR) in Perseus. Volcano plots were generated using GraphPad Prism software [19]. Significance is established for any kinases with a -log10 p-adjusted value of 1.3 or above which indicates p < 0.05.

Proteomics (total and phosphoproteome) analysis

Sample preparation, mass spectrometry analysis, bioinformatics, and data evaluation were performed in collaboration with the Center for Proteome Analysis at IUSM. Methods described below are adaptations from literature reports [20] and vendor-provided protocols.

Sample Preparation/Peptide preparation: Flash frozen tumors of metastatic OS PDX model TT2-77 (0.4-0.5 mg each) treated with vehicle (n = 4) or 50 mg/kg palbociclib (n = 4) were submitted to the Center for Proteome Analysis. The samples were ground in a mortar and pestle under liquid nitrogen and denatured in 8 M urea, 50 mM Tris-HCl, pH 8.5 (Sigma-Aldrich cat#: 10812846001) with sonication using a Bioruptor® sonication system (Diagenode Inc. NJ, USA, North America, cat# B01020001) with 30 sec/30 sec on/off cycles for 15 min in a water bath at 4 °C. After subsequent centrifugation at 14,000 rcf for 20 min, protein concentrations were determined by Bradford protein assay (Bio-Rad cat#: 5000006). Approximately 1-2 mg equivalent of protein from each sample were then reduced with 5 mM tris (2-carboxyethyl)phosphine hydrochloride (Sigma-Aldrich, St. Louis, MO, USA, TCEP, S cat#: C4706) for 30 min at room temperature and alkylated with 10 mM chloroacetamide (CAA, Sigma Aldrich cat#: C0267) for 30 min at room temperature in the dark. Samples were diluted with 50 mM Tris.HCl, pH 8.5 to a final urea concentration of 2 M for Trypsin/Lys-C based overnight protein digestion at 37 °C (1:70 protease:substrate ratio, Mass Spectrometry grade, Promega Corporation, cat#: V5072). Digestions were acidified with trifluoroacetic acid (TFA, 0.5 % v/v) and desalted on Sep-Pak® Vac cartridges (WatersTM cat#: WAT054955) with a wash of 1 mL 0.1 % TFA followed by elution in 70 % acetonitrile 0.1 % formic acid (FA). Peptide concentrations were checked by Pierce Quantitative colorimetric assay (cat#: 23275) and confirmed to be consistent. Phosphopeptide Enrichment: For phosphoproteomics, each ∼ 1 mg peptide sample was applied to a Pierce High-Select™ TiO2 Phosphopeptide Enrichment Kits (Thermo Fisher Scientific, cat#: A32993). After preparing spin tips as per manufacturer’s instructions, each sample was applied to an individual enrichment tip, washed and eluted. The phosphopeptide elution was immediately dried and labeled as described below.

TMTpro labeling: Global and phosphopeptides were each labeled with Tandem Mass Tag (TMTpro) reagent (manufactures instructions, 0.5 mg per sample Thermo Fisher Scientific, TMTpro™ Isobaric Label Reagent Set; cat#: A44520 Lot VE299609) for two hours at room temperature, quenched with a final concentration v/v of 0.3 % hydroxylamine at room temperature for 15 min. Labeled peptides were then mixed and dried by speed vacuum.

High pH Basic Fractionation: For high pH basic fractionation, peptides were reconstituted in 0.1 % trifluoroacetic acid and fractionated on Sep-Pak® Vac cartridges using methodology and reagents from Pierce™ High pH reversed-phase peptide fractionation kit (Thermo Fisher Scientific cat#: 84868).

Nano–liquid chromatography/ mass spectrometry (Nano-LC-MS): Samples were run (1/8th of each global fraction and 1/5th followed by a technical replicate of 1/3rd of each phosphopeptide fraction) on an EASY-nLC 1200 HPLC system (SCR: 014993, Thermo Fisher Scientific) coupled to Eclipse Orbitrap™ mass spectrometer (Thermo Fisher Scientific). Peptides were separated on a 25 cm EasySpray™ C18 column (2 μm, 100 Å, 75 μm x 25 cm, Thermo Scientific cat#: ES902A) at 400 nL/min with a gradient of 4-30 % with mobile phase B (Mobile phases A: 0.1 % FA, water; B: 0.1 % FA, 80 % Acetonitrile (Thermo Fisher Scientific cat#: LS122500) over 160 min, 30-80 % B over 10 mins; and dropping from 80 to 10 % B over the final 10 min. The mass spectrometer was operated in positive ion mode with a 4 sec cycle time data-dependent acquisition method with advanced peak determination and Easy-IC (internal calibrant) on. Precursor scans (m/z 400-1750) were done with an orbitrap resolution of 120000, RF lens% 30, maximum inject time 50 ms, AGC target of 100 % (4e5), MS2 intensity threshold of 2.5e4, MIPS mode, including charges of 2 to 6 for fragmentation with a 60 sec dynamic exclusion. MS2 scans were performed with a quadrupole isolation window of 0.7 m/z, 34 % HCD collision energy, 50000 resolution, 20 % normalized AGC target (1e5), maximum IT of 120 ms, fixed first mass of 100 m/z.

SPS-MS3 phosphopeptide analyses: For this analysis, the gradient was shorted to 115 min total. A 4 sec cycle time was used (top speed). MS1 settings were Orbitrap resolution 120,000, scan range 400-1400 m/z, AGC target 100 %, RF lens 30 %, MIPS mode set to peptide, charge states 2-7, dynamic exclusion of 90 sec, minimum threshold of 5e3, and precursor fit window 0.7. MS2 level settings were quadrupole isolation of 0.7, fixed collision induced dissociation energy of 35 %, activation time of 10 ms, multistage activation on with neutral loss mass of 97.9673, detector ion trap in rapid mode with max IT of 60 ms, normalized AGC of 100 %. Precursor Ion exclusion of low 18 m/z and high 5 m/z with isobaric tag exclusion set to TMTpro. SPS-MS3 set to detector orbitrap with 50,000 resolution, scan range of 110-500, MS2 isolation window of 2 m/z, SPS to 10, fixed higher energy collision dissociation set to 55 %, max IT of 120 ms, normalized AGC target of 200 %.

Data analysis: Raw files were analyzed in Proteome Discover™ 2.4 (Thermo Fisher Scientific) with a database containing the UniProt reviewed Homo sapiens proteome plus common contaminants (Uniprot accessed on 09/29/2019). Control samples were channels TMTpro 130C, 131 N, 131C and 132 N; Palbociclib treated samples were channels 132C, 133 N, 133C, 134 N. Global and phosphoproteomics SEQUEST HT searches were conducted with a maximum number of 2 missed cleavages; precursor mass tolerance of 10 ppm; and a fragment mass tolerance of 0.02 Da. MS3 settings were 20 ppm precursor tolerance and 0.5 Da fragment tolerance. Static modifications used for the search were, 1) carbamidomethylation on cysteine (C) residues; 2) TMTpro label on lysine (K) residues and TMTpro on peptide N-terminus. Dynamic modifications used for the search were oxidation of methionines, phosphorylation on serine, threonine or tyrosine, and acetylation on protein N-termini. Percolator False Discovery Rate was set to a strict setting of 0.01 and a relaxed setting of 0.05. IMP-ptm-RS node was used for all modification site localization scores. Values from both unique and razor peptides were used for quantification. In the consensus workflows, peptides were normalized by total peptide amount with no scaling. Quantification methods utilized isotopic impurity levels available from Thermo Fisher Scientific. Reporter ion quantification was allowed with S/N threshold of 10, co-isolation threshold of 50 %, and minimum 37 % SPS mass matches threshold. Resulting grouped abundance values for each sample type, abundance ratio values; and respective p-values (ANOVA) from Proteome Discover™ were exported to Microsoft Excel.

Hematoxylin and eosin (H&E) staining and organ integrity analysis

Organ tissues were fixed in 10 % neutral-buffered formalin for 24-72 h, processed, and embedded in paraffin from 2 to 3 mice from each cohort. Sections (5 µm) were cut, stained with H&E, and imaged using the Aperio ScanScope CS system (Leica Biosystems, Deer Park, IL, USA). H&E stains were analyzed by a blinded, board-certified pathologist.

Analysis of in vivo senescence by confocal imaging

For the PDX96 tumor, 5-µm-thick cryosections were treated with 1:10 diluted Bafilomycin A1 (FastCellular™ Senescence Detection Kit, MP Biomedicals) to inhibit β-galactosidase activity, then stained with 1:10 diluted SPiDER-βGal (MP Biomedicals, FastCellular™ Senescence Detection Kit, cat# 76337-152) for 30 min at 37 °C with 5 % CO₂. Sections were also stained with Hoechst (1:1000, Thermo Fisher Scientific). Confocal imaging was performed using a Leica TCS SP8 confocal/2P microscope with a 20x/0.75IMM objective lens. Sequential laser illumination was performed for PMT1/DAPI (405 nm excitation, 415-483 nm emission) and PMT2/SPiDER-βGal (488 nm excitation, 525-570 nm emission). Images were 12-bit, 512 × 512 pixels, 400 Hz scan speed, with bidirectional X scan. Stitching was done using Leica LASX software v.3.5.7. Image segmentation was done with Imaris v. 9.82, using the ‘SPOT’ module for intensity and nucleus count, and ‘Surfaces’ for SPiDER-βGal staining intensity. Sum intensity was normalized per nucleus and graphed using One-way ANOVA followed by Holm–Sidak post-hoc test.

Statistical analyses

For in vitro drug combination studies, we calculated concentration required to inhibition 50 % cell growth (IC50) values and combination indices using CalcuSyn v2 (BioSoft), as outlined in our previous description [16]. The impact of combinations was assessed on an efficacy scale using the Bliss independence model [21]. The Bliss expected value was computed using the equation (A + B) – (A × B), where A and B represent the percentage of growth inhibition caused by agents A and B at specific dose combinations. Differences close to 0 % between the Bliss expected growth inhibition and the observed growth inhibition indicate additivity, values >0 % suggest synergy, and values <0 % suggest antagonism. Statistical analyses were performed using methods appropriate to each experimental design, including unpaired t-tests with Welch’s correction, one-way ANOVA followed by Tukey’s post-hoc test, and two-way or repeated-measures two-way ANOVA with Holm–Šidák post-hoc correction. All data are reported as mean ± SEM. Data were considered significant at p < 0.05, and additional statistically significant designations included within the respective figure legends. Survival in the PDX77-TT2 was evaluated using Kaplan–Meier plots and analyzed through the log-rank test. GraphPad Prism Software (GraphPad Inc., San Diego, CA, USA) was used for all data analysis and visualization.

Results

Dual inhibition of CDK4/6 and PI3K/mTOR decreases OS cell growth in vitro

In a panel of pediatric and AYA OS cell lines (TT2, G292, U2OS, MG63, Saos-2, Saos-LM7, and 143B, Supplementary Table 1) activation of CDK4/6–Cyclin D–RB1 and PI3K/mTOR signaling was verified by western blot (Supplementary Figure 1). Presence of CDK4/6, CCND3, and phosphorylation of RB1 was evident in RB proficient (RB+) OS cell lines. Expression of the CDKN2A gene product, p16^INK4A, was variable across the OS lines. In the TT2 xenoline, which harbors a CDKN2A/B deletion p16^INK4A, was not detected. In Saos-2 and Saos-LM7, both of which carry an RB1 splice-site mutation, RB1 protein was absent (Supplementary Figure 1). Similarly, PI3K/mTOR pathway components were broadly detected—including p110α, AKT, p-AKT, mTOR, and p-mTOR—supporting activation of this signaling axis in these models (Supplementary Figure 1). Given the activation of the CDK4/6 and PI3K/mTOR pathways in OS, functional drug sensitivity screens were conducted to evaluate potency and drug-interaction effects using Chou–Talalay (Fig. 1; Supplementary Table 3) and Bliss Independence analyses (Supplementary Table 4).

Fig. 1.

Palbociclib + voxtalisib combination produce additive to synergistic growth suppression across OS cell lines. TT2 xenoline and other established OS cell lines were treated with the CDK4/6i (palbociclib) and/or the PI3K/mTORi (voxtalisib) for five days. (A) Bar graphs show IC₅₀ values for single-agent palbociclib and voxtalisib across OS cell lines. Data represent mean ± SEM of three independent experiments, each performed in triplicate. (B-H) IC₅₀ isobolograms illustrate the effects of palbociclib (palbo) and voxtalisib (vox) alone and in combination. Isobole points falling below the diagonal line of additivity indicate synergy, points on the line indicate additivity, and those above the line indicate antagonism. Each point represents the mean of three experiments conducted in triplicate. Vertical and horizontal bars denote ± 1 SEM and are omitted when smaller than the symbol size.

To provide clinical context, the maximum plasma concentration (C_max) of CDK4/6i (palbociclib) in humans is approximately 0.2 μM at the maximum tolerated dose (MTD) of 100 mg once daily [22]. For PI3K/mTOR inhibitor (PI3K/mTORi; voxtalisib), the C_max reaches approximately 1.1 μM at MTDs of 50 mg twice daily or 90 mg once daily [23]. Dose–response analyses demonstrated that RB1⁺ OS cell lines (TT2, G292, U2OS, MG63, and 143B) were more sensitive to palbociclib, exhibiting lower IC₅₀ values than RB1-deficient lines (Saos-2 and Saos-LM7), consistent with RB1 proficiency being a key, though not exclusive, determinant of CDK4/6i sensitivity (Fig. 1A). In line with the role of RB1 in mediating CDK4/6i response, RB1 knockdown significantly reduced palbociclib sensitivity in TT2 and G292 cells, although the magnitude of the effect was modest (Supplementary Figure 2). In the IC₅₀–isobologram analyses (Chou–Talalay), the isoboles for palbociclib + voxtalisib combinations fell below the line of additivity, indicating synergy in all cell lines, with greater synergy observed in RB⁺ models (Fig. 1B–H). Correspondingly, combination index (CI) values were predominantly additive to strongly synergistic (Supplementary Table 3). Bliss independence analysis, which evaluates combination efficacy across a dose matrix, showed that Bliss scores for palbociclib + voxtalisib were uniformly within the additive range at clinically relevant concentrations in both RB⁻ and RB⁺ OS cell lines (Supplementary Table 4). These findings are consistent with clinical experience demonstrating that additive efficacy can yield meaningful therapeutic benefit in advanced cancers. Notably, Hwangbo et al. [24] reported that strong synergy is neither required nor commonly observed among clinically effective combination regimens.

Palbociclib induces cell-cycle arrest and senescence, while co-treatment with voxtalisib enhances autophagy in RB1⁺ OS cells

To explore the biological mechanisms contributing to the growth suppression observed with dual CDK4/6 and PI3K/mTOR inhibition, we analyzed cell-cycle distribution and molecular responses related to apoptosis, senescence, and autophagy in OS cells. Drug concentrations were selected based on Bliss Independence analyses to represent clinically relevant exposures within the range that produced reproducible growth inhibition in OS cells. (Supplementary Table 4). Palbociclib, alone or in combination with voxtalisib, induced a marked G1 arrest at 24 and 48 h in TT2 cells compared with vehicle or voxtalisib alone (Fig. 2A,B). In G292 cells, palbociclib alone or combined with voxtalisib did not alter cell-cycle distribution at 24 h (Fig. 2C) but produced significant G1 arrest by 48 h (Fig. 2D). There was no evidence of pronounced apoptotic cell death with single-agent or dual CDK4/6–PI3K/mTOR inhibition in these cell lines (Supplementary Table 5). Because cytotoxicity was not observed, we next assessed treatment-induced senescence using senescence-associated β-galactosidase (SA-β-Gal) staining and evaluated autophagy using the Autophagy Probe Red assay. Palbociclib alone induced senescence and measurable autophagy in TT2 (Fig. 2E,F) and G292 (Fig. 2G,H) cells. Voxtalisib monotherapy did not induce senescence in either line (Fig. 2E,G) and caused only minimal autophagy in G292 cells (Fig. 2H). Dual CDK4/6–PI3K/mTOR inhibition preserved palbociclib-induced senescence while further increasing autophagy in both TT2 (Fig. 2F) and G292 (Fig. 2H) cells. The increased autophagy is consistent with voxtalisib-mediated inhibition of mTOR, which relieves mTOR-mediated suppression of the autophagic machinery [10].

Fig. 2.

Palbociclib triggers G1 cell cycle arrest and senescence, while voxtalisib co-treatment enhances autophagy in RB1⁺ OS cells. Cell-cycle distribution was analyzed by flow cytometry following propidium-iodide (PI) staining in TT2 cells (A, B) and G292 cells (C, D) treated with single agents [0.15 µM palbociclib (palbo) or 0.5 µM voxtalisib (vox)] or their combination for 24 h (A, C) and 48 h (B, D). Bar graphs show the percentage of cells in G₁, S, and G₂/M phases. Senescence and autophagy were evaluated by flow cytometry using SA-β-Gal staining (E, G) and Autophagy Probe Red (F, H) in TT2 and G292 cells treated with palbociclib, voxtalisib, or their combination for five days. Data are representative of three independent experiments. Statistical analysis was performed using two-way ANOVA followed by Holm–Sidak post-hoc pairwise multiple comparisons; n = 3/group; *p < 0.05, **p < 0.01, *** p < 0.001, ****p < 0.0001. Note: 24 h = 24 h, 48 h = 48 h.

Molecular signatures of OS PDX models align with CDK4/6 and PI3K/AKT/mTOR pathway activation

Two OS PDX models, PDX77-TT2 and PDX96, were evaluated for their response to palbociclib + voxtalisib combination therapy versus single-agent treatment. Both models harbor a CDK4/6-hyperactivated molecular profile—RB1 monoallelic loss but detectable RB protein, CDKN2A homozygous deletion, and CCND3 amplification—and display protein signatures consistent with activation of the CDK4/6 and PI3K/mTOR pathways (Supplementary Figure 1). As previously reported, PDX77-TT2 is a relapsed/metastatic OS model that was established from a metastatic pelvic lesion [15]. The treatment-naïve PDX96 model was generated from the diagnostic biopsy of a patient who subsequently developed progressive OS [15]. Notably, both PDXs harbor extensive CDKN2A/CDKN2B deletions, which normally encode the p16^INK4A and p14^ARF proteins, which act as tumor suppressors and regulate cell proliferation through distinct mechanisms. Role for p14^ARF involves stabilization of p53 through sequestration of MDM2 so that p53 can function to regulate appropriate cues for apoptosis, DNA repair, etc. [25]. Notably, p16^INK4A is a major endogenous inhibitor of the Cyclin D–CDK4/6 complex, with loss of this checkpoint strongly promoting CDK4/6 activity [25]. To define the structural integrity of this locus, we used CNVkit to map copy number alterations across CDKN2A and CDKN2B and determined which exons encoding p16^INK4A and p14^ARF were deleted. In PDX77-TT2, a partial deletion spanning CDKN2B and the first exon of CDKN2A, together with complete loss of the last three CDKN2A exons encoding p16^INK4A and p14^ARF isoforms, was identified (Supplementary Figure 3). In contrast, PDX96 exhibited deletion of the entire CDKN2A/CDKN2B locus (chr9:21967446–22010446) (Supplementary Figure 3). Collectively, these findings demonstrate that both PDX models lack functional CDKN2A/B-encoded checkpoint control, supporting their suitability for evaluating the therapeutic impact of CDK4/6 inhibition.

CDK4/6 inhibition via palbociclib induces PI3K pathway activation and downregulates cell-cycle proteins in an OS PDX77-TT2

While combination palbociclib + voxtalisib showed in vitro OS growth inhibition, we next evaluated therapeutic efficacy in vivo. To determine whether CDK4/6 blockade induces compensatory signaling that could limit its efficacy, we analyzed early changes in the kinase network in the PDX77-TT2 model following short-term palbociclib exposure. As described in the Methods, MIB-MS kinome profiling measures the fraction of kinases in an inhibitor-binding–competent state, which reflects a combination of kinase abundance, conformational accessibility of the ATP-binding pocket, and affinity for immobilized pan-kinase inhibitors rather than catalytic activity alone [19]. Kinome profiling of PDX77-TT2 tumor lysates from mice treated with palbociclib (50 mg/kg, five days) revealed increased MIB binding to multiple growth factor receptors, including PDGFRB, TGFBR1, and ACVR1 which are all upstream activators of the PI3K/mTOR signaling cascade (Fig. 3A; Supplementary Table 6A) [[26], [27], [28]]. Consistent with the in vitro autophagy data in the TT2 xenoline (Fig. 2F), MIB binding to PIK3C3 (Vps34), a kinase required for autophagosome formation, was elevated following palbociclib treatment in PDX77-TT2-bearing mice (Fig. 3A; Supplementary Table 6A). Conversely, in tumor lysates from palbociclib-treated tumors, MIB binding to CDK1 was reduced. To extend these observations and capture the downstream consequences of these signaling changes, we next performed global and phosphoproteome analyses on the same tumor lysates. Global/total proteomic analysis demonstrated coordinated downregulation of proteins involved in cell-cycle regulation (CDK1), DNA replication (PCNA, MCM family members), microtubule dynamics (STMN1), and chromatin assembly (NASP) (Fig. 3B; Supplementary Table 6B). Phosphoproteome profiling further revealed reduced phosphorylation of key regulators of proliferation, including CDK2, RB1, MKI67, LMNA, and VIM (Fig. 3C; Supplementary Table 6C). Western blot confirmed increased PI3K (p110α; Fig. 3D; Supplementary Figure 4) and phosphorylation of AKT (Fig. 3D; Supplementary Figure 4), together with decreased phosphorylation of CDK1, and RB1 (Fig. 3D; Supplementary Figure 4) when normalized to total protein content; this was accompanied by a reduction in total RB1 protein (Fig. 3D; Supplementary Figure 4A), consistent with RB1 degradation and CDK4/6 inhibition (Fig. 3D; Supplementary Figure 4). Collectively, these data indicate that palbociclib blocks phosphorylation of downstream target RB1, triggers compensatory activation of PI3K/mTOR signaling, and downregulates CDK1/2-dependent cell-cycle proteins, providing a mechanistic rationale for combining CDK4/6i and PI3K/mTORi in OS.

Fig. 3.

Short-term CDK4/6 inhibition activates PI3K signaling and downregulates cell-cycle proteins in an OS PDX derived from a metastatic site in a pretreated/relapsed patient. Kinome, proteomic, and phosphoproteomic profiling were performed on PDX77-TT2 tumors treated with palbociclib (50 mg/kg) or vehicle for five days. (A) MIB-MS kinome analysis showing altered kinase binding, including increased receptor tyrosine kinase activity and decreased CDK1 binding following palbociclib treatment. (B) Global proteomic profiling demonstrating reduced expression of cell-cycle, replication, and chromatin-assembly proteins. (C) Phosphoproteomic profiling indicating decreased phosphorylation of key regulators of proliferation and cytoskeletal dynamics. (D) Western blot validation confirming activation of the PI3K/mTOR pathway and downregulation of CDK and RB1 signaling components [RB1/p-RB1, CDK1/p-CDK1, CDK2/p-CDK2, PI3K (p110α), AKT/p-AKT]. Western blot analysis was conducted once. Volcano plots lots show log₂ fold change vs –log₁₀ (p-adjusted value); horizontal line = 1.3 (*p < 0.05); n = 4/group; The mouse IDs shown in Fig. 3D correspond to the same animals used for the kinome and proteomic analyses.

Dual CDK4/6 and PI3K/mTOR inhibition reduces tumor growth and prolongs survival in the PDX77-TT2 model

We next evaluated whether dual pathway inhibition could improve therapeutic efficacy in vivo. PDX77-TT2-bearing mice were treated with palbociclib (50 mg/kg), voxtalisib (50 mg/kg), or their combination for eight weeks, with tumor growth monitored during and after cessation of therapy. Prior to reaching the predeath endpoint (tumor volume of 1500 mm³), both the single agents and palbociclib + voxtalisib combination significantly reduced tumor growth compared with the vehicle for days 41-57 of the dosing window. By day 57 of dosing, the palbociclib + voxtalisib-mediated tumor growth suppression was significantly decreased compared to palbociclib or voxtalisib alone (Fig. 4A). Combination therapy also extended median survival (121 days) relative to vehicle (76 days), palbociclib (111 days), and voxtalisib (99 days) (Fig. 4B). Body weights remained stable across all treatment groups, indicating good tolerability throughout the study (Fig. 4C). Together, these results demonstrate that concurrent targeting of CDK4/6 and PI3K/mTOR signaling produces superior and well-tolerated antitumor activity in vivo, reinforcing the rationale for dual-pathway inhibition in advanced or treatment-refractory OS.

Fig. 4.

Palbociclib + voxtalisib combination prolongs survival in the PDX77-TT2 OS model. PDX77-TT2 was established from a metastatic OS lesion in a relapsed patient. PDX77-TT2–bearing mice were treated with 50 mg/kg palbociclib, 50 mg/kg voxtalisib, or combination (PO, SID) for five consecutive days for eight weeks. (A) Tumor growth curves showing mean ± SEM tumor volumes. Table: Repeated measures two-way ANOVA with Holm–Sidak post-hoc pairwise multiple comparisons; n = 9/group. (B) Kaplan–Meier survival analysis demonstrating improved survival with single-agent and combination therapy compared to vehicle. Combination treatment further prolonged survival relative to either monotherapy. Table: Log-rank (Mantel–Cox) test; n = 9/group. (C) Body weights remained within the normal range across all treatment groups throughout dosing (measured twice weekly).

Dual inhibition of CDK4/6 and PI3K/mTOR suppresses tumor growth and modulates kinase activity in treatment-naïve OS PDX96

Building on findings from the pretreated PDX77-TT2 model, we next examined the impact of dual CDK4/6–PI3K/mTOR inhibition in the treatment-naïve PDX96 model, which shares similar high-risk oncogenic features to PDX77-TT2. This analysis focused on defining pharmacodynamic responses, antitumor efficacy, and acute treatment tolerability at the end of the dosing period. Mice bearing PDX96 tumors were treated with palbociclib (50 mg/kg), voxtalisib (50 mg/kg), or the combination for six weeks. Combination therapy produced the most significant suppression of tumor growth over time compared with vehicle or single agent (Fig. 5A). Endpoint tumor weights showed a significant decrease in single-agent and combination groups compared to vehicle, with a trend toward further reduction with the combination (Fig. 5B). Body weights remained stable across all treatment groups, indicating good tolerability (Fig. 5C), and histopathologic evaluation confirmed preserved organ architecture in all cohorts (Supplementary Figure 5).

Fig. 5.

Palbociclib + voxtalisib combination suppresses tumor growth, modulates cell-cycle and PI3K/mTOR signaling in the PDX96 OS model. PDX96 was derived from an OS tumor biopsy from a treatment-naïve patient. PDX96-bearing mice were treated with 50 mg/kg palbociclib, 50 mg/kg voxtalisib, or their combination (PO, SID) for six weeks. (A) Tumor growth curves showing mean ± SEM volumes. Repeated measures two-way ANOVA with Holm–Sidak with post-hoc pairwise multiple comparisons post-hoc comparisons: *p < 0.05 single agents or combo vs vehicle (days 56–79); #p < 0.05 single-agent palbo (red) or single-agent vox (blue) vs combination (days 68-79). (B) Final tumor weights (mean ± SEM). One-way ANOVA with Tukey’s test for post-hoc pairwise multiple comparisons: **p < 0.01 vs vehicle, ****p < 0.0001 vs vehicle; n = 8 mice/group. (C) Body weights remained stable and within the normal range during treatment (measured twice weekly). (D–H) Volcano plots from MIB-MS kinome profiling comparing combination, single-agent, and vehicle groups. Plots show log₂ fold change vs –log₁₀ (p-adjusted value); horizontal line = 1.3 (*p < 0.05); n = 4/group. (I) Western blot validation of selected targets [RB1/p-RB1, CDK1/p-CDK1, CDK2/p-CDK2, CDK4 CDK6, PI3K (p110α), AKT/p-AKT, mTOR/p-mTOR, Beclin-1, LC3BII/LCBI, BCL2]. GAPDH served as loading control. Western blot done once. (J) Confocal microscopy of SA-β-Gal staining in PDX96 tumors across treatment groups (scale bar = 500 µm). Bar graph shows sum intensity of SA-β-Gal/nuclei. One-way ANOVA with Tukey’s test for post-hoc pairwise multiple comparisons; n = 3/group; *p < 0.05, palbociclib vs. vehicle; other comparisons not significant.

To investigate how combination therapy blocks tumor growth, kinome profiling was performed on the PDX96 tumor lysates from each treatment group. We first examined how the combination compared with each single agent across the kinome. In tumor lysates prepared from combination palbociclib + voxtalisib versus vehicle or single agents, MIB binding to several tyrosine kinases was reduced, including PI3K pathway components (PIK3CA, mTOR) and PI3K/mTOR-activating receptor kinases like PDGFRA, compared to mice treated with vehicle or single agents (Fig. 5D,E,F; Supplementary Table 7A,B,C). In addition, the components of the JNK signaling pathway, one of three different MAPK pathways, including MAPK8 and MAPK9 [29], exhibited decreased MIB binding in tumor lysates from PDX96-bearing mice treated with the combination therapy compared to vehicle and single agents (Fig. 5D,E,F; Supplementary Table 7A,B,C). Additionally, a trend towards reduced pro-survival protein BCL2 following combination treatment was observed (Fig. 5I; Supplementary Figure 6M), consistent with prior reports linking BCL2 downregulation to decreased PI3K/mTOR pathway activity [30]. While pro-autophagy regulator, Beclin-1, remained unchanged (Fig. 5I), decreased mTOR phosphorylation (Fig. 5I; Supplementary Figure 6 J,K,R) in the combination group coincided with elevated LC3BII/LC3BI ratios (Supplementary Figure 6L), supporting a relationship between reduced mTOR activity and autophagy induction [10].

We next compared MIB binding in PDX96 lysates prepared from single-agent versus vehicle-treated PDX96-bearing mice. In PDX96 lysates prepared from palbociclib- versus vehicle-treated mice, there was no increase in MIB binding to the components of the JNK pathway or to MEKK2 (MAP3K2) [29], which activates MAPK8/MAPK9 (Fig. 5G; Supplementary Table 7D). In PDX96 lysates prepared from voxtalisib- versus vehicle-treated PDX96-bearing mice, increased MAPK8/9 and mTOR MIB binding were evident (Fig. 5H; Supplementary Table 7E). However, western blot indicated that total mTOR protein levels remained unchanged, and phosphorylation of mTOR at Ser2448, an AKT phosphorylation site widely used as a marker of PI3K/mTOR activation, was decreased in tumor lysates from voxtalisib-treated versus vehicle-treated mice (Fig. 5I; Supplementary Figure 6R). Given that MIB-MS reports inhibitor-binding–competent kinase rather than net abundance or phosphorylation, we interpret this as a change in mTOR’s conformational or complex-binding state, rather than increased mTOR activity, consistent with feedback signaling within the PI3K/mTOR pathway.

We next evaluated downstream CDK targets across all treatment groups. Kinome profiling also demonstrated that the palbociclib + voxtalisib combination of PDX96-bearing mice reduced MIB binding in the tumor lysates prepared from these PDX96 tumors across multiple cyclin-dependent kinases (CDK1, CDK2, CDK9, and CDK16), with a particularly strong decrease in CDK1 and CDK9 compared with either single agent (Fig. 5D–F). Palbociclib treatment of PDX96-bearing mice selectively reduced CDK1 MIB binding in corresponding PDX96 lysates, consistent with its primary role in CDK1 downregulation (Fig. 5G). In contrast, voxtalisib treatment of PDX96-bearing mice increased MIB binding for CDKs, including CDK4, CDK5, and CDK16 in PDX96 lysates, consistent with compensatory activation of cell-cycle–associated kinases following PI3K/mTOR inhibition (Fig. 5H). Western blot analysis corroborated these findings, showing reduced total and phosphorylated CDK1 and CDK2 protein levels following combination treatment (Fig. 5I; Supplementary Figure 6C-F, O,P). The palbociclib + voxtalisib combination also led to a greater decrease in total and phosphorylated RB1 than either single agent (Fig. 5I; Supplementary Figure 6A,B,N), confirming effective suppression of CDK4/6 downstream targets in vivo. Consistent with in vitro observations (Fig. 2), confocal imaging indicated that palbociclib alone induced senescence in PDX96 tumors. In contrast, the addition of voxtalisib did not significantly alter senescence levels relative to palbociclib alone (Fig. 5J). Collectively, these results indicate that palbociclib is the principal driver of senescence induction, while PI3K/mTOR co-inhibition enhances CDK suppression, blocks PI3K downstream signaling, and enhances autophagy, without further increasing senescence.

Palbociclib, in the presence or absence of voxtalisib, suppresses OS lung nodule outgrowth in the MG63.3 OS colonization model

Because pulmonary metastases are a major cause of mortality in OS, we next evaluated the impact of palbociclib + voxtalisib on metastatic outgrowth in the lung. In vitro IC₅₀ isobologram analysis of MG63.3 cells showed that isoboles for multiple dose ratios of palbociclib and voxtalisib fell below the line of additivity, indicating synergy between the two inhibitors (Fig. 6A). To assess treatment effects on established metastases in vivo, NSG mice were injected intravenously with MG63.3 cells and treated for three weeks with vehicle, palbociclib (50 mg/kg), voxtalisib (50 mg/kg), or the combination until the first vehicle-treated mouse reached the predefined humane endpoint. Human-specific qPCR of lung DNA demonstrated that both palbociclib monotherapy and the combination significantly reduced metastatic burden compared with vehicle controls (Fig. 6B). Histological analyses corroborated these findings, showing marked reductions in metastatic foci by H&E and HLA class I (ABC) staining (Fig. 6C,D). Although the reduction in metastatic burden with combination therapy did not reach statistical significance relative to palbociclib alone, all three independent measures—qPCR, H&E quantification, and HLA staining—showed concordant decreases in the combination group, suggesting a modest but consistent trend toward further suppression. Body weights remained stable across all treatment groups, apart from expected weight loss in vehicle-treated mice near endpoint (Supplementary Figure 7). Together, these findings demonstrate that CDK4/6 inhibition alone is sufficient to suppress the outgrowth of established pulmonary metastases in the MG63.3 model, with the addition of PI3K/mTOR inhibition showing a reproducible, though not statistically significant, trend toward further reduction in metastatic burden.

Fig. 6.

Palbociclib + voxtalisib induces synergistic growth inhibition in MG63.3 Cells and palbociclib-drives reduction of MG63.3 lung metastatic burden. (A) IC₅₀ isobologram showing in vitro synergistic growth suppression in MG63.3 cells across multiple palbociclib and voxtalisib dose ratios. Isobole points below, on, or above the line of additivity indicate synergy, additivity, or antagonism, respectively. Each point represents the mean ± SEM of three independent experiments performed in triplicate. (B) NSG mice were injected via the tail vein with 2 × 10⁵ MG63.3 cells and, one week later, treated with palbociclib (50 mg/kg), voxtalisib (50 mg/kg), or their combination (PO, SID) for three weeks. Human-specific qPCR analysis of lung tissue revealed significantly decreased metastatic burden in palbociclib- and combination-treated mice compared with vehicle controls (n = 6-7/group). (C, D) Representative photomicrographs of lung sections stained with H&E (C) and HLA class I (ABC) (D) demonstrating reduced metastatic lesions following treatment. Scale bars = 5 mm. HALO AI DenseNet segmentation masks show metastatic foci (red) and normal lung tissue (blue). Representative images from n = 3/group. (E) Quantification of metastatic area as a percentage of total classified lung tissue. (F) Quantification of HLA class I (ABC)–positive tumor cells as a percentage of total cells. Data represent mean ± SEM. One-way ANOVA with Tukey’s test for post-hoc pairwise multiple comparisons multiple comparisons test; n = 6-7/group; ****p ≤ 0.0001.

Discussion

Pediatric and AYA osteosarcoma (OS) remains difficult to treat due to its aggressive biology, the toxicity and limited long-term benefit of conventional chemotherapy, and the poor efficacy of current second-line therapies. Consequently, there is a critical need for more effective and durable therapeutic strategies. Increasing evidence highlights dysregulation of the Cyclin D–CDK4/6–RB1 and PI3K/mTOR pathways as recurrent and actionable vulnerabilities in OS, providing a strong rationale for combinatorial targeting approaches [4,31,32].

CDK4/6 represents a well-established therapeutic target across multiple tumor types [7,33]. Although CDK4/6 pathway hyperactivation occurs in a subset of OS, CDK4/6i rarely generate durable responses, consistent with limited long-term efficacy in other malignancies [6,[34], [35], [36]]. This reflects both their fundamentally cytostatic nature and the emergence of compensatory PI3K/mTOR signaling. CDK4/6 inhibition causes RB hypophosphorylation and disrupts mTORC2, activating AKT and downstream effectors [37,38]. AKT-mediated inhibition of GSK3β prevents Cyclin D degradation [37], and chronic PI3K/mTOR activation can upregulate Cyclin D, enabling CDK2-driven cell-cycle re-entry even in the absence of CDK4/6 activity [38]. These adaptive mechanisms highlight the need for coordinated inhibition of CDK4/6 and PI3K/mTOR to sustain pathway suppression.

In this study, dual CDK4/6–PI3K/mTOR inhibition produced additive to synergistic growth suppression across OS cell lines in vitro and achieved more durable tumor control in vivo. Given the breadth of kinome, proteome, and phosphoproteome alterations observed, we focused our mechanistic analyses on pathways with established relevance to CDK4/6 signaling, PI3K/mTOR activity, and cell-cycle and replication control. The full multi-omic datasets—including all significantly altered kinases and proteins—are provided in the Supplementary Tables. In vivo, kinome and proteomic profiling in PDX77-TT2 showed that short-term palbociclib modulated multiple networks, including expected downregulation of CDK1/2 activity. Increased CDK2 levels and noncanonical Cyclin D1–CDK2 activation are known mechanisms of escape from CDK4/6 inhibition [39]. PI3K/mTOR inhibition counters this bypass, preventing S-phase entry when CDK4/6 is inhibited [38]. Consistent with this, palbociclib-treated PDX77-TT2 tumors exhibited increased PI3K/mTOR activity, supporting dual inhibition with palbociclib and voxtalisib. Combination therapy in pretreated metastatic PDX77-TT2 improved survival and reduced tumor growth compared with single agents.

While the short-term (five-day) palbociclib exposure in the pretreated PDX77-TT2 elicited rapid PI3K/mTOR activation, prolonged palbociclib treatment (six weeks) in the treatment-naïve PDX96 did not show signs of increased PI3K/mTOR activation above baseline. These differences in PI3K/ mTOR responses to CDK4/6 inhibition as a monotherapy likely reflect a combination of (i) transient adaptive feedback activation of receptor tyrosine kinases (RTK) immediately after CDK4/6 inhibition, (ii) tumor-context dependence (prior therapy may ‘prime’ pretreated tumors for rapid compensatory RTK signaling), and (iii) longer-term homeostatic changes and clonal selection that reduce pathway throughput (e.g., receptor downregulation, induction of phosphatases). Importantly, these dynamics underscore that short-term signaling surges do not necessarily predict long-term pathway activity or therapeutic outcome; rather, they point to a window during which concurrent PI3K/mTOR blockade may prevent early compensatory signaling and thereby deepen and sustain cytostatic control.

Furthermore, while in vitro pharmacodynamic studies typically show rapid PI3K/mTOR pathway activation following palbociclib treatment [[40], [41], [42]], the temporal kinetics of this compensatory response in vivo remain poorly characterized and understudied [42]. Our findings help address this knowledge gap by providing in vivo time-course data defining the dynamics of PI3K/mTOR activation under CDK4/6 inhibition. Irrespective of these distinct temporal PI3K/mTOR pathway responses to CDK4/6 blockade as a monotherapy, dual inhibition suppressed tumor growth and altered kinase activity in treatment-naïve primary PDX96, possibly due to the endogenous hyperactivation of PI3K and CDK4/6 pathways already present in the model. Moreover, in PDX96, palbociclib monotherapy did not increase CDK1/2 activity, whereas the combination significantly reduced CDK1/2 activity, indicating more effective cell-cycle blockade. Kinome profiling also showed reduced binding to multiple CDKs (CDK1, CDK2, CDK9), along with decreased total protein and phosphorylation levels of CDK1 and CDK2 in tumor lysates from treated tumor-bearing mice. Combination therapy further reduced RB1 expression and phosphorylation, demonstrating more profound suppression of CDK4/6 signaling.

Mechanism-of-action studies showed that CDK4/6 inhibition resulted in G1 arrest and senescence, and the palbociclib + voxtalisib enhanced autophagy. These observations, consistent with findings in other solid tumors [38,43,43], can be extended to studies presented here. In PDX96, the palbociblib + voxtalisib combination induced greater autophagy than palbociclib alone, while senescence remained dominant with palbociclib monotherapy. The induction of both senescence and autophagy suggests that these cytostatic programs may co-occur within the same OS cells or may arise in distinct subpopulations. However, the precise cellular distribution of these responses remains to be determined. Palbociclib primarily induced senescence, consistent with other cancers [7,44,45], suggesting that pairing CDK4/6 inhibition with senolytics could be another avenue to improve efficacy further [44]. CDK4/6 inhibition can also trigger autophagy [45], while PI3K/mTOR inhibition promotes autophagy via mTORC1 suppression [10,38,43]. In line with these mechanisms, the combination increased autophagy more than monotherapy, supported by increased PIK3C3 activity and elevated LC3BII/LC3BI ratios. Long-term palbociclib did not increase PIK3C3, consistent with autophagy being strongest early in treatment [38]. Reduced BCL-2 expression with combination therapy may further promote autophagy by relieving Beclin-1 inhibition [46].

Therapy-induced senescence and autophagy are interrelated cytostatic programs that stabilize disease in the absence of cytotoxicity [45]. CDK4/6 inhibition induces RB1-dependent senescence and remodeling of the senescence-associated secretory phenotype, often resulting in tumor stasis rather than regression. PI3K/mTOR inhibition enhances autophagic flux, a survival mechanism that can transition to autophagy-associated death when prolonged [10,43]. The palbociclib + voxtalisib combination reinforced both programs, producing stable tumor control without overt cytotoxicity. Ongoing studies will test whether prolonged exposure can shift these cytostatic responses toward irreversible tumor suppression.

Because ∼25 % - 50 % of OS patients present with lung metastases at diagnosis [47], we also evaluated palbociclib + voxtalisib combination in an MG63.3 lung-colonization model. Palbociclib alone markedly suppressed metastatic outgrowth, whereas the greatest benefit in subcutaneous PDX models occurred with the palbociclib + voxtalisib combination. These differences may reflect biological distinctions between the different OS models as well as the differences in the tumor microenvironment. Although not statistically significant, several types of analyses suggested a trend toward reduced metastatic burden with the combination. Extended treatment duration may be required to fully realize the therapeutic potential of this combination in metastatic disease. Interestingly, palbociclib-mediated suppression of lung metastasis has been reported in nasopharyngeal carcinoma but has not previously been described in OS [41].

Limitations and Future Directions

This study evaluates dual CDK4/6–PI3K/mTOR targeting in OS, but several limitations remain. First, the observed reduction in OS lung nodules with palbociclib was seen in only one colonization model, requiring validation in additional metastatic systems. Second, while immunodeficient PDX models are valuable for mechanistic interrogation and therapeutic screening in pediatric cancers [15], they cannot evaluate immune-mediated effects which is important because CDK4/6i has been reported to modulate antitumor immunity [11]. Relatedly, a limitation of these immunodeficient PDX studies is that peripheral blood indices were not assessed; however, given the markedly altered baseline hematologic profiles of these strains, such measures can be difficult to interpret, and tolerability was therefore evaluated using longitudinal clinical monitoring, including body weight assessment and preserved multi-organ and splenic architecture. Comprehensive hematologic evaluation is being pursued in ongoing immunocompetent OS models, where these assessments are more robust and informative. Third, treatment windows in this study ranged from six to eight weeks; longer-term studies are needed to fully define the durability of pathway suppression, tolerability, and the potential transition from cytostatic to cytotoxic outcomes. Ongoing efforts include testing additional molecularly distinct PDX models, spontaneous metastatic OS models, immunocompetent approaches, and extended dosing regimens to determine whether sustained dual targeting enhances efficacy and limits adaptive resistance.

Conclusion

Dual CDK4/6–PI3K/mTOR inhibition is effective across treatment-naïve and relapsed/metastatic OS models, yielding durable and well-tolerated tumor control by blocking compensatory signaling and enhancing senescence and autophagy. Both palbociclib and voxtalisib inhibit tumor growth in OS PDX models, with the combination providing superior cytostatic effects, while palbociclib alone primarily suppresses metastatic OS lung colonization. Together, these data support a genomically informed therapeutic strategy with clinical potential for stabilizing high-risk pediatric and AYA OS.

Data availability/sharing statement

The raw and processed global and phosphoproteomics mass spectrometry data has been uploaded through ProteomeXchange partner MassIVE with accession MSV000099280 (PXD068837). Username: reviewer_ MSV000099280; Password: osteosarcoma.

The Kinome profiling of PDX77-TT2 treated with palbociclib has been successfully submitted to ProteomeXchange via the PRIDE database. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [48] partner repository with the dataset identifier PXD069093.

Funding

This research was funded by the Department of Defense (DOD CA230160, E.A.D., M.R.S., S.P.A., P.H.P., and K.E.P.); the NCI/NIH Cancer Center Support Grant to the Indiana University Simon Comprehensive Cancer Center for use of Shared Resource Facilities—Preclinical Modeling and Therapeutics, Flow Cytometry, Proteomics, and the Center for Medical Genomics (P30CA082709; M.A.T., K.C., F.M.K., A.L.S., and K.E.P.); the NICHD/NIH Specialized Centers in Research in Pediatric Developmental Pharmacology (P50HD090215; M.R.S., K.B.-V., B.J.B., P.H.P., and K.E.P.); the Sarcoma Foundation of America (P.H.P. and K.E.P.); the American Cancer Society Institutional Research Grant (ACS-IRG Grant Mechanism, Grant No. 16-192-31; P.H.P.); the IUSCCC American Cancer Society Post-Baccalaureate in Cancer Research Education Program (R.J.); the NCI/NIH T32 Pediatric and Adult Translational Cancer Drug Discovery and Development Training Program (PACT-D3; T32CA272370; R.M., E.E.S., and J.K.); the Caroline Symmes Children’s Cancer Endowment (M.R.S., B.J.B., E.A.D., P.H.P., and K.E.P.); the Emmie Joy Brooks Pediatric Cancer Research Fund (M.R.S., E.A.D., K.K., E.A.D., P.H.P., and K.E.P.); the Tyler Trent Cancer Research Endowment for Riley Hospital for Children at IU Health (M.R.S., P.H.P., and K.E.P.); the Danaher Foundation (M.R.S., P.H.P., and K.E.P.); the Indiana University Grand Challenge Precision Health Initiative—Pre-Sarcoma/Sarcoma Pillar (M.R.S., B.J.B., E.A.D., C.Y., F.B., R.M., P.H.P., and K.E.P.); and the Riley Children’s Foundation (M.R.S., P.H.P., and K.E.P.).

Ethics approval and consent to participate

Not applicable.

Institutional Review Board Statement

This study was conducted in accordance with the Declaration of Helsinki and approved by the IACUC (protocols: 22028, 25041). All patient samples from which the PDX were established were consented and collected under IRB protocol #1501467439 and have been described elsewhere.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

Portions of the text were edited for clarity and grammar using AI-based language models (ChatGPT and Grammarly). All scientific content, data analysis, interpretation, and conclusions were generated entirely by the authors, who take full responsibility for the accuracy and integrity of the manuscript.

CRediT authorship contribution statement

Farinaz Barghi: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. M. Reza Saadatzadeh: Writing – review & editing, Writing – original draft, Visualization, Supervision, Resources, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Erika A. Dobrota: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Methodology, Formal analysis, Data curation. Harlan E. Shannon: Writing – review & editing, Writing – original draft, Visualization, Supervision, Software, Methodology, Formal analysis, Data curation. Barbara J. Bailey: Writing – review & editing, Visualization, Validation, Supervision, Software, Methodology, Formal analysis, Data curation. Courtney Young: Writing – review & editing, Visualization, Validation, Formal analysis, Data curation. Rada Malko: Writing – review & editing, Visualization, Software, Methodology, Conceptualization. Ryli Justice: Writing – review & editing, Visualization, Validation, Software, Investigation, Formal analysis, Data curation. Niknam Riyahi: Writing – review & editing, Writing – original draft, Visualization. Christopher Davis: Writing – review & editing, Writing – original draft, Visualization, Software, Methodology, Investigation, Formal analysis, Data curation. Khadijeh Bijangi-Vishehsaraei: Writing – review & editing, Writing – original draft, Visualization, Resources, Methodology, Funding acquisition, Data curation. Keiko Kreklau: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation. Lauren K. Stevens: Writing – review & editing, Writing – original draft, Visualization. Jenna Koenig: Writing – review & editing, Visualization, Conceptualization. Shirzat Sulayman: Writing – review & editing, Writing – original draft, Visualization, Software, Methodology, Formal analysis, Data curation, Conceptualization. Sheng Liu: Writing – review & editing, Writing – original draft, Visualization, Software, Methodology, Formal analysis, Data curation, Conceptualization. Jun Wan: Writing – review & editing, Writing – original draft, Visualization, Supervision, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Melissa A. Trowbridge: Writing – review & editing, Visualization, Supervision, Software, Methodology, Formal analysis, Data curation, Conceptualization. Kathy Coy: Writing – review & editing, Visualization, Supervision, Software, Methodology, Investigation, Formal analysis, Data curation. Felicia M. Kennedy: Writing – review & editing, Visualization, Software, Methodology, Formal analysis, Data curation. Anthony L. Sinn: Writing – review & editing, Writing – original draft, Visualization, Supervision, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation. Marissa Just: Writing – review & editing, Visualization, Resources, Data curation. Kyle W. Jackson: Writing – review & editing, Visualization, Resources, Data curation. George Sandusky: Writing – review & editing, Visualization, Supervision, Software, Methodology, Formal analysis, Data curation. L. Daniel Wurtz: Writing – review & editing, Supervision, Resources. Christopher D. Collier: Writing – review & editing, Resources, Data curation. Dana Mitchell: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Emily E. Seiden: Writing – review & editing, Visualization, Software, Resources, Methodology, Formal analysis, Data curation, Conceptualization. Edward M. Greenfield: Writing – review & editing, Visualization, Supervision, Software, Resources, Methodology, Formal analysis, Data curation. Emma Doud: Writing – review & editing, Writing – original draft, Visualization, Software, Resources, Methodology, Formal analysis, Data curation. Amber Mosley: Writing – review & editing, Visualization, Software, Resources, Methodology, Formal analysis, Data curation. Steven P. Angus: Writing – review & editing, Writing – original draft, Visualization, Supervision, Software, Resources, Methodology, Formal analysis, Data curation, Conceptualization. Michael J. Ferguson: Writing – review & editing, Writing – original draft, Visualization, Resources, Methodology, Conceptualization. Pankita H. Pandya: Writing – review & editing, Writing – original draft, Visualization, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization. Karen E. Pollok: Writing – review & editing, Writing – original draft, Visualization, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Harlan E. Shannon is a retiree and stockholder of Eli Lilly.

Appendix. Supplementary Materials

The following supporting information can be downloaded at: www.mdpi.com/xxx/s1,

Supplementary Table 1: Summary of OS cell-lines and donor demographics.

Supplementary Table 2: Antibody list.

Supplementary Table 3. Combination index (CI) values for palbociclib + voxtalisib combination in OS cell lines.

Supplementary Table 4. Bliss analysis shows additive inhibitory effect of palbociclib + voxtalisib combination in OS cell growth.

Supplementary Table 5. Palbociclib and voxtalisib in combination or as single agent does not induce apoptosis in OS cell lines.

Supplementary Table 6A: Kinome profiling of PDX77-TT2 treated with 50mg/kg palbociclib for five-days (Figure 3A).

Supplementary Table 6B: Total proteome profiling of PDX77-TT2 treated with 50mg/kg palbociclib for five-days (Figure 3B).

Supplementary Table 6C: Phosphoproteome profiling of PDX77-TT2 treated with 50mg/kg palbociclib for five-days (Figure 3C).

Supplementary Table 7A: Kinome profiling of PDX96 treated with combination therapy vs vehicle (Figure 5D).

Supplementary Table 7B: Kinome profiling of PDX96 treated with combination therapy vs palbociclib (Figure 5E).

Supplementary Table 7C: Kinome profiling of PDX96 treated with combination therapy vs voxtalisib (Figure 5F).

Supplementary Table 7D: Kinome profiling of PDX96 treated with palbociclib vs vehicle (Figure 5 G).

Supplementary Table 7E: Kinome profiling of PDX96 treated with voxtalisib vs vehicle (Figure 5H).

Supplementary Figure 1: Baseline expression of CDK4/6- and PI3K/mTOR-pathway proteins across OS models.

Supplementary Figure 2. RB1 knockdown modestly decreases palbociclib-mediated growth sensitivity in OS cell lines.

Supplementary Figure 3: CDKN2A/B copy number alterations in OS PDX models.

Supplementary Figure 4: Protein quantification of short-term palbociclib-treated metastatic, pretreated/relapsed OS PDXs depicts that CDK4/6 inhibition activates PI3K signaling and downregulates cell-cycle proteins.

Supplementary Figure 5: Histopathological evaluation of organs from treated PDX96-bearing mice.

Supplementary Figure 6. Protein quantification of western blots confirms modulation of cell cycle and PI3K/mTOR, autophagy, and pro-survival pathways in PDX96 samples treated with palbociclib + voxtalisib combination.

Supplementary Figure 7. Body weights in treated mice in the MG63.3 lung colonization model.