GPNMB overexpression- a marker of resistance to CDK4/6 inhibitors

Breast cancer, resistance to targeted therapies, CDK4/6 inhibitors, DTPs, GPNMB (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 2 Funding: This study was supported partly by the Susan G. Komen for the Cure to Sunil Badve and Winship Invest$$ pilot grant to Yesim G ökmen-Polar. In addition, s tartup funds f rom Emory University were also utilized. Sunil Badve is a Komen scholar. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 3

Resistance to cyclin -dependent kinase 4/6 inhibitors remains a major clinical challenge in treating estrogen receptor -positive breast cancer, with no reliable predictive biomarkers currently available for patient selection. To investigate resistance mechanisms, we generated drug-tolerant persisters (DTPs) to abemaciclib and palbociclib in a panel of estrogen receptor- positive breast cancer cell lines. Functional analyses revealed that DTPs showed resistance to CDK4/6 inhibition, maintained G1 arrest, and exhibited increased senescence phenotype. To identify clinically relevant markers of resistance, we compared transcriptomic profiles from DTPs with publicly available gene‑expression data from the phase III PEARL trial. Glycoprotein non‑metastatic B (GPNMB) emerged as one of the most strongly upregulated transcripts in DTPs, and also was amongst the genes associated with resistance in the PEARL dataset. We further verified that GPNMB overexpression (GPNMB- OE) in sensitive cells conferred resistance to CDK4/6 inhibition, and enhanced migratory capacity. Overexpression of GPNMB drove substantially faster tumor progression and eliminated the growth‑inhibitory effect of abemaciclib, which remained highly effective in control tumors. Across all treatment arms, GPNMB‑ OE tumors failed to respond to CDK4/6 blockade, highlighting a strong resistance phenotype. These results identify GPNMB as a potent promoter of tumor progression and a key mediator of resistance to abemaciclib. Our findings position GPNMB as a potential biomarker and therapeutic target that may help identify patients unlikely to benefit from CDK4/6 inhibition. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 4

Cyclin-dependent kinase (CDK) 4 and 6 inhibitors (CDK4/6i) have been one of the most significant practice -changing advancements in the treatment of estrogen receptor positive (ER+) breast cancer in the recent decade (1-3). CDK4/6i such as abemaciclib, palbociclib, and ribociclib have been approved for the treatment of metastatic ER+ breast cancer. In addition, abemaciclib and ribociclib are also US Food and Drug Administration ( FDA)- and European Medicines Agency (EMA)-approved for treating early-stage breast cancer. CDK4/6i work by blocking the complex formation between CDK4/6 and cyclin D, which prevents the phosphorylation of the retinoblastoma protein (R B). This results in cell cycle arrest and prevents the uncontrolled proliferation (4). Although many patients respond to CDK4/6i, some patients have intrinsic resistance to these agents while o thers may acquire resistance to therapy over time. Despite the initial response to CDK4/6i, resistance and recurrence are frequent contributing to breast cancer mortality. Most clinical trials with CDK4/6i in metastatic ER+ breast cancer s do not require any biomarker analysis as entry criteria. T he monarchE clinical trial in high-risk early breast cancer required Ki67 greater than 20% as a risk defining factor s, which ultimately resulted in FDA-approval of Ki67 as a selection marker. However, responses to abemaciclib was observed in even in low-Ki67 patients, resulting in the removal of Ki67 as an FDA-approved biomarker for abemaciclib (5). The NATALEE clinical trial showed that adding ribociclib to endocrine therapy produced a meaningful and durable reduction in invasive disease‑free recurrence across a broad group of intermediate‑ and high‑risk HR+/HER2‑ early breast cancer (6, 7). This benefit was consistent across all clinicopathologic subgroups. However, no predictive biomarker has been identified to guide selection. Several studies have analyzed samples from patients with metastatic breast cancer treated with CDK4/6i to identify the cause(s) of resistance. Genomic profiling has identified (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 5 mutations in cell cycle-associated genes such retinoblastoma (RB) tumor suppressor in samples from patients treated with CDK4/6i (8). However, these were found to be rare (<5%) in patients enrolled in the PALOMA-3 clinical trial (9). Analysis of the monarchE and NATALEE studies did not find any alterations to be significantly associated with outcomes. Recent work by Kudo et al. has provided important genomic insights into resistance to first‑line CDK4/6 inhibition in metastatic HR+/HER2 – breast cancer (10). By analyzing MSKCC‑IMPACT sequencing data from patients treated with CDK4/6i plus endocrine therapy, a striking enrichment of TP53 and MDM2 alterations was present in roughly 30% of early progressors with markedly shorter progression‑free survival compared with long‑term responders. These findings were subsequently validated in a large , pooled dataset of 4,457 patients from the MONALEESA and m onarchE clinical trials, emphasizing the robustness of these genomic predictors. Complementary functional studies demonstrated that dual inhibition of CDK4/6 and CDK2 can mitigate the effects of p53 pathway disruption and restore durable therapeutic responses in ER+ breast cancer models, highlighting the biological relevance of p53 alterations in shaping CDK4/6i sensitivity and supporting rational combination strategies to overcome resistance. Stu dies involving mRNA analysis of patient samples treated with CDK4/6i have largely failed to identify definitive markers associated with outcomes (11, 12). An exception to this was the analysis of the PEARL trial which identified the expression of a number of genes (41 refractory and 10 sensitive) including Cyclin E1 ( CCNE1) and Polo-like kinase 1 ( PLK1) as important determinants of resistance (13). Kong et al also developed a 33 -gene signature that predicts neoadjuvant Ki67 response to anastrozole/palbociclib in a metastatic phase 2 trial (14). Other postulated mechanisms of CDK4/6i resistance in preclinical models include elevated CDK6 activity, cyclin E , Fibroblast Growth Factor Receptor ( FGFR) pathway activation , pyruvate dehydrogenase kinase 1 (PDK1) and protein Arginine Methyltransferase 5 (PRMT5) (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 6 (15-19). None of these molecular biomarkers are currently used for selecting patients for CDK4/6 inhibitors in clinical trials (20-23). Close inspection of clinical trial data suggests two important facets associated with resistance to CDK4/6i: a) resistance to the drug can appear early with up to 25% of patients exhibiting disease progression within the first few months, b) Data from the MAINTAIN and postMONARCH clinical trials in metastatic BC showed that patients who had tumors that progressed on primarily palbociclib could be successfully treated with ribociclib or abemaciclib, respectively (24, 25). The transient nature of CDK4/6i resistance has been also observed in cell line-based studies (26). These studies suggest that mechanisms of resistance may be transient and perhaps the “drug-tolerant persisters” (DTPs) might provide a good model to study resistance. These cells are believed to be a subpopulation of slowly replicating cells that can tolerate the drug treatment and repopulate the tumor following the termination of therapeutics (27-29). DTPs have been implicated as a bridge for an early stage of drug resistance mechanism allowing the survival of cells and adaptation to secondary resistance phenotypes (reviewed by Pu et al. (30) ). Although first described in lung cancer (31) , DTP cells has been reported in multiple cancer types, including breast cancer (32-34). In this study, we have developed abemaciclib and palbociclib resistant DTP models to study resistance to CDK4/6i. By comparing the DTP transcriptomic profiles with data from the PEARL clinical trial, we identified GPNM B, a transmembrane glycoprotein, as consistently overexpressed, supporting its role in resistance to CDK4/6i. Functional studies demonstrated that GPNMB promotes c ell growth and contributes to CDK4/6i resistance. We further demonstrated its importance in tumor progression and resistance to abemaciclib. Together, these findings, suggest that GPNMB may serve as a promising biomarker for identifying patients unlikely to benefit from CDK4/6 inhibition. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 7

Breast cancer cell lines and generation of DTPs Human LCC2 (tamoxifen-resistant) and LCC9 (fulvestrant (ICI 182,780) and tamoxifen cross- resistant) cell lines were kind gifts from Dr. R. Clarke (Georgetown University Medical School, Washington DC)(35, 36). MCF7 (RRID: CVCL_0031), and T47D (RRID: CVCL_0553) cell lines were purchased from American Type Culture Collection (ATCC, Manassas, VA , USA). Cell lines were carefully maintained in a humidified tissue culture incubator at 37°C in a 5% CO2:95% air atmosphere, and stocks of the earliest passage cells were stored. Parental cells (LCC2, LCC9, MCF7 and T47D) that are sensitive to the relevant drugs (abemaciclib or palbociclib) were treated at concentrations exceeding 100 times the established IC 50 values as described previously by Sharma et al. (31). DTPs appeared at 9 days later for the experimental analyses as indicated by Sharma et al. (31). In cell culture experiments, vehicle controls were used according to the solvents in which the drugs were prepared. Abemaciclib was dissolved in ethanol, whereas palbociclib was dissolved in dimethyl sulfoxide (DMSO). Accordingly, ethanol and DMSO were used as the respective vehicle controls at the same final concentrations as in the corresponding drug-treated groups (≤ 0.1% v/v). Cell proliferation Cell viability was performed using the CyQuant Cell Proliferation Assay (#C35011) according to the manufacturer’s instruction s (ThermoFisher Scientific , Waltham, MA, USA). Abemaciclib and palbociclib w ere purchased from Selleck Chemicals LLC (Houston, TX , USA). Briefly, cells were plated in 96 -well flat-bottom plates at a density of 2,000 cells per well and allowed to attach overnight. The following day, cells were exposed to increasing concentrations (100–2000 nM) of abemaciclib or palbociclib for 72 hours. After treatmen t, CyQUANT reagent containing the fluorescent dye and membrane -permeabilizing agent was (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 8 added according to the manufacturer’s instructions. Fluorescence was measured using a BioTek Synergy H1 plate reader (Agilent, Santa Clara, CA, USA) with excitation 480 nm and emission 53 5 nm. The r elative fluorescence units (RFU) were used as an index of cell proliferation, and IC₅₀ values, the concentrations of inhibitors that are necessary to kill 50% of the cells, were calculated using GraphPad Prism 10.3.1. Cell cycle analysis by flow cytometry Drug-sensitive cells and DTPs (2 x 10 5) were plated on a 60 -mm plate, harvested by trypsinization, pelleted, and resuspended in 1 mL of PBS. Cell cycle analysis was done as per manufacturer’s instruction (Propidium Iodide Flow Cytometry Kit, #ab139418, Abcam, Waltham, MA, USA) using a Becton Dickinson FACScan flow cytometer (Bedford, MA, USA). Data were analyzed with FlowJo™ v10. Senescence analysis Senescence-associated β- galactosidase (SA -β-gal) expression was evaluated by using a Senescence β-Galactosidase Staining Kit (#9860, Cell Signaling Technology, Danvers, MA) and a flow cytometric detection of cellular senescence via β-galactosidase hydrolysis as per the manufacturer’s instructions (CellEvent Senescence Green Flow Cytometry Assay, #C10840, Thermofisher Scientific, Waltham, MA, USA). Flow cytometry analysis was performed using BD FACSuite software and Data were analyzed with FlowJo™ v10. DTPs or parental cells are quantified by the extent of senescent cells (green) to determine the senescent cells induced by abemaciclib or palbociclib. ALDEFLUOR assay ALDEFLUOR assay was performed using the ALDEFLUOR Kit (StemCell Technologies, (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 9 #01700) according to the manufacturer’s instructions. Briefly, cells were resuspended in ALDEFLUOR assay buffer containing the ALDH substrate (BAAA) and incubated at 37 °C for the recommended time, with a parallel sample treated with the specific ALDH inhibitor DEAB as a negative control. After incubation, cells were washed, kept on ice, and immediately analyzed by flow cytometry. ALDH-positive cells were identified by comparing fluorescence intensity of DTP and their sensitive counterparts to their DEAB-treated controls, and data were analyzed using BD FACSuite software and FlowJo™ v10. RNA isolation and sample quality assessment Total RNA from DTPs (9 days treatment) and its parental drug- sensitive counterparts was extracted using the RNeasy isolation kit according to the manufacturer's instructions (Qiagen, Germantown, M D, USA). RNA was treated with Turbo DNase (ThermoFisher Scientific, Waltham, MA, USA ). The quality of RNA was assessed using the Nanodrop® ND -1000 spectrophotometer (ThermoFisher Scientific) and the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Clariom D Pico Human Transcriptome array and functional enrichment analysis Total RNAs from DTPs and their parental drug-sensitive cell lines were sent to the Applied Biosystems/Thermo Fisher Scientific Service laboratory (Santa Clara, CA , USA). Quality Control and Clariom D Pico Human Transcriptome Array were performed according to Applied Biosystems/Thermo Fisher Scientific’s instructions as previously described (37). Probe cell intensity (CEL) files generated from Clario m D arrays were analyzed using Transcriptome Analysis Console (TAC) software version 4.0 (Applied Biosystems/ThermoFisher). Differential gene expression analysis was performed using a limma-based linear modeling framework with empirical Bayes –moderated statistics. P values (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 10 and log₂-based fold changes were calculated for each comparison, and statistically significant genes were identified based on predefined thresholds. Differentially regulated genes (DEGs) were categorized as upregulated and downregulated sets in DTPs versus parental sensitive counterparts for each cell line. Venn diagrams were generated using genes that met the significance criteria of |Fold Change| ≥ 2 and P < 0.05, using the Venn diagram tool from Ghent University. of Ghent https://bioinformatics.psb.ugent.be/webtools/Venn/. P athway enrichment analysis was conducted using g:Profiler ( https://biit.cs.ut.ee/gprofiler/). E nriched pathways were identified across Gene Ontology (GO Biological Process). Q uantitative reverse transcription -polymerase chain reaction (RT -qPCR) of breast cancer cell lines RT-qPCR from DTPs and their parental drug -sensitive counterparts was performed as described previously (37). The mRNA l evels of GPNMB (Hs01095669_m1) were analyzed using TaqMan gene expression assays on an ABI Prism 7900 platform according to the manufacturer’s instructions (Applied Biosystems/ThermoFisher Scientific , Carlsbad, CA , USA). Actin (ACTB; Hs00357333_g1) was used as endogenous control. All experiments were performed as three independent sets. The relative quantification of the gene expression changes was analyzed according to the ∆∆Ct method using the Applied Biosystems DataAssistTM Software v3.0. All graphs were generated using GraphPad Prism 10.3.1 software. The error bars were calculated and represented in terms of the mean ± SD. The results presented are the combination of three independent assays, and unpaired t-test with Welch’s correction analyses were performed using GraphPad Prism 10.3.1 software (P < 0.05, statistically significant). Cell Migration Assay (Wound Healing Assay) To assess the effects of GPNMB, 24 well plates were coated wit h either 0.1% bovine serum (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 11 albumin (BSA) or recombinant human GPNMB ( osteoactivin (rhOA; 10 μg/mL , Acro Biosystems, # catalog GPB-H5229). Briefly, 250 μL of freshly prepared coating solution was added per well and incubated at 37°C for 2 hours or at 4°C overnight. Wells were then aspirated, washed once with 1× PBS, and used immediately or within one week. Wound healing assay has been performed to measure two-dimensional migration (34). Briefly, cells were seeded at a density of 2 × 10^5 cells/mL in coated 24 -well plates and cultured for 72 hours to a llow monolayer formation. A linear scratch was generated using a p200 pipette tip, and detached cells were removed by gentle washing with PBS. Images were acquired at the indicated time point using a Nikon Eclipse TS100 microscope with NIS-Elements imaging software (version 5.41.02; Nikon Corporation). Wound area was quantified, and percent wound closure was calculated using the formula: [(wound area at time 0 − wou nd area at time x) / wound area at time 0] × 100. Data were analyzed and graphed using GraphPad Prism v10.3.1 software .

represent three independent experiments performed in triplicate. Statistical analyses were conducted using an unpaired two- tailed Student t test with Welch’s correction. Differences were considered statistically significant at P < 0.05. Western Blot The protein lysates from DTPs and their parental drug- sensitive counterparts were prepared, and equal amounts of protein were subjected to SDS –PAGE and Western blot analysis as described previously (38). The Bio-Rad DC‐Protein assay kit (Bio -Rad, Hercules, CA, USA) was used to determine protein concentrations. Blots were incubated with the primary antibody against GPNMB (Cat# 38313 Cell Signaling Technology, Danvers, MA). Antibody against β- actin (ACTB, Sigma, St. Louis, MO , USA, dilution 1:5000) or GAPDH ( Cat# GTX627408, GeneTex, Inc., Irvine, CA , USA) was used as the loading control. Protein bands were visualized by SuperSignal West Pico PLUS Chemiluminescent Substrate kit (Amersham, (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 12 Piscataway, NJ, USA) and Amersham Imager 600 GE Healthcare Life Sciences (GE Healthcare Bio Sciences, Pittsburgh, PA, USA). The data are representative of three individual sets. Spheroid Immunofluorescence Spheroid development and immunofluorescence staining were performed as described previously (39). Briefly, primary a ntibody staining was performed using GPNMB (Cat# sc- 271415, Santa Cruz, CA, USA, dilution 1:50). Secondary antibody F(ab ’)2-Goat anti-Mouse IgG (H+L) Cross-Absorbed, Alexa Fluor Plus 555 was purchased from Invitrogen, cat#A48287, dilution 1:1000). Spheroids were imaged with the Leica TCS SP8 inverted confocal microscope (X 20). DAPI for nuclear staining is purchased by Cell Signaling (Cat#4083). Generation of stable full-length GPNMB overexpression in breast cancer cells Lentiviral particles encoding the human GPNMB ORF (NM_001005340, GPNMB Human Tagged Lenti ORF clone , (OriGene cat# RC207615L4V, pLenti -C-mGFP-P2A-Puro) were obtained from OriGEne Technologies (Rockville, MD, USA). The pLenti‑C‑mGFP‑P2A‑Puro vector (OriGene cat# PS100093V) served as the lentiviral control for all experiments and labeled as empty vector (EV) . LCC2, MCF7 and T47D c ells were transduced with either GPNMB ORF (GPNMB Overexpression, GPNMB-OE) or control lentivirus (Empty Vector - EV) according to the manufacturer’s instructions. After transduction, cells were selected with puromycin to establish stable cell lines expressing GPNMB or vector control. Individual puromycin‑resistant clones were isolated, expanded, and screened for GPNMB expression by Western blotting and GFP fluorescence. The clone demonstrating the highest GPNMB overexpression (GPNMB OE clone 1) from each cell line model was selected for all subsequent assays. mRNA and protein expres sion levels of GPNMB were further validated using RT - (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 13 qPCR and immunofluorescence, respectively. Statistical analysis Unpaired t-test with Welch’s correction analyses were performed using GraphPad Prism 10.3.1 software as indicated. All results are representative of three replicates and expressed as mean values SD. In all cases, differences were considered statistically si gnificant at *P < 0.05, **P<0.01, ***P<0.001, ****P<0.0001. In vivo orthotopic xenograft model All animal experiments were conducted under a protocol approved by the Emory University Institutional Animal Care and Use Committee (IACUC). Six-to eight-week-old ovariectomized athymic nude mice (Jackson Lab, Bar Harbor, ME, USA) were acclimatized for 3–7 days. All animals were housed in a specific pathogen- free (SPF) facility at Emory University. A controlled-release E2 pellet (0.72 mg E2, 60-day formulation; Innovative Research of America, Sarasota, FL, USA) was injected subcutaneously (s.c.) via a sterile 14-gauge trocar 24 h before tumor implantation. LCC2 cells stably expressing either empty vector ( 2EV) or GPNMB (2GPNMB OE C1) were injected into the mammary fat pad at a density of 5 × 10⁶ cells per mouse. Tumor growth was monitored by caliper measurements, and tumor volume was calculated using the formula: (length × width²)/2. When tumors reached approximately 100– 150 mm³ (day 24 post-injection), mice were randomized into four treatment groups: (1) LCC2- EV (2EV) treated with vehicle (n = 9), (2) LCC2 -GPNMB (2GPNMB OE C1 ) treated with vehicle (n = 9), (3) LCC2-EV (2EV) treated with abemaciclib (n = 10), and (4) LCC2-GPNMB (2GPNMB OE C1) treated with abemaciclib (n = 10). Abemaciclib (Selleckchem, Cat # S5716) was formulated as a suspension in 0.5% methylcellulose for delivery via oral gavage (PO) at a dose of 50 mg/kg/ once daily for 3 5 consecutive days. Vehicle only (0.5% methylcellulose) - (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 14 treated mice received an equivalent volume by oral gavage . Animals were sacrificed after 35 days of treatment when tumors of 2GPNMB OE C1 (the group with the highest tumor growth) reached a volume of 1,000 mm3. Statistical analysis in vivo : T umor growth data are presented as mean ± SEM. Statistical significance between treatment groups was determined using two-way ANOVA with repeated measures, followed by appropriate post hoc multiple -comparison tests. A n adjusted (adj.) p value < 0.05 was considered statistically significant. All statistical analyses were performed using GraphPad Prism v10.3.1 software. Resu lts Drug -Tolerant Persisters (DTPs) to CDK4/6i display distinct functional phenotypes compared to their drug naive sensitive counterparts To investigate CDK4/6i resistance, we developed DTP cell models resistant to CDK4/6 i (abemaciclib or palbociclib) in a panel of ER+ cells (LCC2 control , LCC9 control, MCF7 control and T47D control) treated as 100x IC50 doses. This resulted in a subpopulation of cells resistant to CDK4/6i (i.e., DTPs) after 9 days of treatment (Fig. 1A). The DTPs failed to reach 50% inhibition at the highest concentrations tested (2000nM), indicating that the cells became resistant to both abemaciclib and palbociclib. LCC2 control, sensitive cells to abemaciclib and palbociclib, exhibited a potent dose -response with an IC₅₀ of 47.32 nM and 49.96 nM, respectively, as described previously (40). A similar pattern was observed in the other DTP sublines confirming that this is general proc ess ((40) and Supplementary Fig. S1 A). The Palbociclib-DTPs similarly failed to reach 50% growth inhibition across the tested concentration range, further confirming resistance ((40) and Supplementary Fig. S1A). To determine whether the DTP phenotype alte red the canonical G1‑arrest response associated with CDK4/6 inhibition, we next evaluated cell‑cycle distribution following (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 15 treatment with abemaciclib and palbociclib. As expected, LCC2, LCC9, MCF‑7, and T47D DTP sublines showed G1‑phase growth arrest upon exposure to both inhibitors ( Fig. 1B, Supplementary Fig. S1 B). Because prolonged or repeated G1 arrest induced by CDK4/6 inhibition can drive cells toward a senescent state, we next asked whether DTPs exhibit enhanced senescence compared with their sensitive parental counterparts. We previously demonstrated that acquired resistant cells to abemaciclib or palbociclib developed a senescent phenotype (40) . To determine whether drug- tolerant persisters (DTPs) similarly display ed increased senescence, we measured senescence-associated β-galactosidase (SA-β-Gal) activity using histochemical staining and flow cytometry, a widely accepted surrogate marker of cellular senescence. In LCC2 DTP sublines, we observed marked blue staining, indicative of elevated SA -β-Gal activity and enhanced senescence ( Fig. 1C ). Quantification of mean fluorescence intensities further confirmed these findings: LCC2 DTPs generated with abemaciclib showed a 77% increase in SA-β-Gal–positive cells, while LCC2 DTPs generated with palbociclib exhibited 88% positivity, compared with only 3% and 28% in the respective parental controls (p < 0.0001) (Fig. 1D)). Similar results were observed in additional DTP models (LCC9, MCF7, and T47D), emphasizing that senescence is a prominent characteristic of CDK4/6 inhibitor-induced drug tolerance (Supplementary Fig. S1C-D). It has been reported that ALDH activity, particularly ALDH1, is essential for maintaining a subpopulation of DTP cells that survive prolonged exposure to targeted therapies (41). We next examined whether ALDH1 contributes to the drug‑tolerant persister (DTP) phenotype to abemaciclib and palbociclib. Using the ALDEFLUOR assay, we did not detect any measurable ALDH1 activity in DTP cells ( Supplementary Fig. S2). This absence of ALDH1 activity argues against a stem‑cell–like mechanism driving persistence in our model. Moreover, DTPs fully reverted to the parental drug‑sensitive phenotype after 20 days in drug‑free media, consistent with the transient and reversible nature of persister states reported (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 16 in clinical settings (Supplementary Fig. S3). Transcriptomic analysis in DTPs We next identified differentially expressed genes (DEGs) in four DTP cell lines treated with abemaciclib compared to their drug naïve sensitive counterparts using the Clariom D transcriptomics arrays and TAC 4.0 platform (Set 1-LCC2-DTPs versus LCC2 control, set 2- LCC9 DTPs versus LCC9 control, set 3-MCF7 DTPs versus MCF7 control and s et 4- T47D DTPs versus T47D control). This analysis revealed the upregulation of 154 genes common in all DTP models, while 274 common downregulated genes were identified (Fig. 2A). To gain deeper insight into the molecular programs altered in DTPs, we performed pathway enrichment analysis using g:Profiler with GO Biological Process annotations. Consistent with the known mechanism of CDK4/6 inhibition, the most significantly downregulated genes were predominantly enriched in pathways related to cell cycle progression and cell division, confirming the suppression of proliferative programs. In contrast, the top upregulated genes were significantly associated with biological processes including response to growth factor, cell migration, and integrin -mediated signaling pathways ( Fig. 2B). Together, these findings indicate that while CDK4/6i primarily suppresses proliferative programs, DTPs simultaneously activate signaling networks that may facilitate persistence and cellular remodeling. GPNMB Emerges as a Top Differentially Expressed Gene Associated with CDK4/6 Resistance in ER+ Breast Cancer To identify a clinically relevant dataset for validating the 154 genes upregulated in DTP s, we systematically searched for clinical trials that included both gene‑expression profiling and treatment with CDK4/6 inhibitors. This search identified the phase III PEARL trial, which randomized participants to palbociclib plus endocrine therapy versus capecitabine, as the only (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 17 publicly available cohort that combined large‑scale mRNA expression data with detailed clinical outcomes from patients with HR+/HER2‑ metastatic breast cancer treated with palbociclib plus endocrine therapy (13). Among the forty -one refractory genes previously linked to resistance to palbociclib plus endocrine therapy (13), glycoprotein non-metastatic B (GPNMB) emerged as one of the most strongly upregulated transcripts in our DTP models [FDR at 5% with the q -value method (SAM)] , further supporting its role as a clinically validated marker of CDK4/6 inhibitor resistance. GP NMB is significantly elevated in CDK4/6i- DTPs and overexpression of GPNMB in vitro confers resistance High expression of GPNMB mRNA was verified in LCC2 -derived DTP sublines to abemaciclib (100 -fold; P<0.0038) and palbociclib (33 -fold; P<0.0042) compared to their LCC2 control cells (Fig. 3A). This significant upregulation was further verified in additional ER‑positive, DTP sublines (LCC9 DTPs, MCF7 DTPs and T47D DTPs) in both abemaciclib and palbociclib, demonstrating that GPNMB upregulation is not restricted to a single cell‑line model (Supplementary Fig. S4 A-B). Flow cytometry analysis also verified that DTPs generated in response to abemaciclib or palbociclib exhibited a dramatic elevation of GPNMB at the cell surface, indicating strong upregulation of this protein during CDK4/6 inhibition (Fig. 3B; Supplementary Fig. 5 ). Using spheroid immunofluorescence, we further confirmed the upregulation of GPNMB at the protein level in all DTP sublines (Fig. 3C; Supplementary Fig. 6). Overexpression of GPNMB in parental sensitive cells mimic the DTP phenotype To further investigate the role of GPNMB in conferring resistance to CDK4/6 inhibitors, we generated stable GPNMB‑overexpressing LCC2 parental breast cancer cells using the OriGene human GPNMB ORF construct, followed by antibiotic selection and screening o f individual (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 18 clones. The clone with the highest GPNMB expression in LCC2 (named GPNMB clone 1 ; 2GPNMB-OE-C1) was selected for further functional studies (Supplementary Fig. S7A). We confirmed the dramatic increase of GPNMB expression at mRNA levels (P=0.0002; Fig. 4A). We also confirmed the elevated increase of GPNMB at the protein levels (Fig 4B); We next determined the sensitivity of GPNMB -overexpressing cells to CDK4/6i (abemaciclib ). 2EV control cells displayed an IC₅₀ of approximately 50 nM, whereas 2GPNMB‑OE‑C1 cells exhibited a dramatically elevated IC₅₀ of 1047 nM (Fig. 4C). This shift represents nearly a 20‑fold increase in resistance relative to parental cells, indicating that GPNMB overexpression substantially diminishes the efficacy of CDK4/6 inhibition. Consistent with its impact on cell‑cycle regulation, 2GPNMB‑OE‑C1 resulted in an elevated accumulation of cells in the G1 phase ( Supplementary Fig. S7B ). This G1 arrest was acco mpanied by a corresponding increase in senescence (Supplementary Fig. S7C), further supporting the role of GPNMB in modulating proliferative capacity and cell‑cycle progression. Together, these findings demonstrate that GPNMB overexpression not only alters cell‑cycle dynamics but also confers resistance to abemaciclib, highlighting a potential mechanism by which tumor cells may evade CDK4/6‑targeted therapy. We next confirmed that GPNMB conferred resistant to abemaciclib by culturing sensitive parental LCC2 cells on recombinant GPNMB protein. While cells coated with GPNMB protein did not reach IC50 values , sensitive control LCC2 cells coated with 0.1 % BSA reached IC50 values of (439.2 nM , Fig. 4D). We next demonstrated its role in cell migration using a wound‑healing assay performed on surfaces coated with either BSA or recombinant human GPNMB (GPNMB‑ECD). LCC2 cells cultured on GPNMB‑ECD–coated plates exhibited significantly enh anced wound closure compared with cells plated on BSA controls, indicating that extracellular GPNMB directly promotes migratory capacity (Fig. 4E; P=0.0003). Consistent with this observation, the addition of soluble recombinant (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 19 GPNMB‑ECD to parental LCC2 cells similarly accelerated wound closure, further supporting a functional role for the extracellular domain in driving a pro‑migratory phenotype. These findings align with our resistance model, in which GPNMB upregulation is associated with reduced sensitivity to CDK4/6 inhibition. Together, these results demonstrate that the GPNMB extracellular domain is sufficient to enhance cell migration and contributes to the invasive and drug‑resistant behavior observed in GPNMB‑high breast cancer cells. Overexpression of GPNMB Promotes Tumor Growth and Abemaciclib Resistance in ERpositive breast cancer xenografts. To determine how GPNMB overexpression influences tumor growth and therapeutic response to abemaciclib, we evaluated four xenograft conditions: These included LCC2 empty‑vector tumors treated with vehicle (2EV vehicle) and with abemaciclib (2EV abemaciclib), as well as LCC2 tumors overexpressing GPNMB ( GPNMB-OE-C1) treated with either vehicle or abemaciclib. This design allowed us to directly compare the impact of GPNMB overexpression under both untreated and drug‑treated conditions. Tumor growth in 2EV group was suppressed significantly in response to abemaciclib, confirming the sensitivity of 2EV group to abemaciclib (adjusted P<0.0001). On the other hand, GPNMB overexpression alone markedly enhanced tumor growth relative to 2EV tumors, demonstrating a significant oncogenic role for GPNMB in vivo (adjusted P<0.0356). Tumors expressing GPNMB ( GPNMB-OE-C1) continued to grow and remained unresponsive to abemaciclib compared with the 2EV group, demonstrating that GPNMB overexpression confer resistance to abemaciclib. Direct comparisons further highlighted the resistance phenotype driven by GPNMB. 2EV tumors treated with abemaciclib were significantly smaller than GPNMB -OE-C1 tumors receiving either vehicle or abemaciclib, demonstrating that tumors with GPNMB overexpression acquire a phenotype that confers resistance to abemaciclib blockade (adjusted P <0.0001). When (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 20 comparing GPNMB-OE-C1 tumors treated with abemaciclib to those receiving vehicle, tumor growth remained essentially unchanged, and the difference was not statistically significant, emphasizing the diminished therapeutic effect of abemaciclib in the context of GPNMB overexpression. These findings suggests that GPNMB overexpression not only drives aggressive tumor progression, but also confers resistance to abemaciclib, positioning GPNMB as a critical determinant of reduced responsiveness to abemaciclib.

The transie nt nature of CDK4/6i resistance observed in the MAINTAIN (24) and postMONARCH (25) clinical trials supports the application of DTPs -based models for investigation of CDK4/6i resistance. DTPs have been shown to repopulate drug- sensitive cellular progenies following a ‘drug holiday’ without harboring any known resistance - associated secondary mutations (42). We used high-dose treatment of human BC cell lines with CDK4/6i (abemaciclib and palbociclib) to develop an effective model of DTPs -model to explore the molecular mechanisms associated with resistance. CDK4/6i -DTPs exhibited slow growth, senescence, and resistant to CDK4/6i. We investigated the characteristics of CKD4/6i -DTPs to furthe r understand these clinical observations . Transcriptomics analysis identified 154 upregulated transcripts ( 138 protein coding) to be differentially expressed in CKD4/6i-DTPs as compared to parental cells across the four cell lines analyzed. To assess the clinical relevance of these genes, we compared this data with genomics analysis of 2,597 genes in the phase III PEARL study (13), wherein patients were treated with palbociclib plus ET. GPNMB was the only gene identified among the 41 top differentially expressed genes ( FDR at 5% with the q -value method (SAM) ) associated with resistance to palbociclib and endocrine therapy. This provided clinical evidence supporting the importance of GPNMB in palbociclib resistance. In our models, we observed (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 21 that most of the genes related to cell cycle and proliferation were dow nregulated in DTPs including CCNE1 and PLK1. GPNMB is expressed by senescent cells as has been referred to as a seno -antigen. Cellular sen escence is one of new hallmarks of cancer mediating tumor development and malignant progression. (43) Therapy- induced senescence, a cellular state triggered by therapeutics, has been implicated as a mechanism of resistance (44). In particular, reversible senescent cells such as in DTPs can escape fro m their non-proliferative condition and resume cell proliferation and rise to therapy-resistance and tumor progression. (45). In spite of this, the cells can show (near -)complete recovery to parental phenotype suggesting that they do not undergo “irreversible” senescence. Kudo et al have suggested that p53 deficiency s uppresses the DREAM complex in breast cancer cells, which enables cell -cycle re -entry (46). We recently have shown that palbociclib - and a bemaciclib-acquired r esistant cells induce senescence and exhibit m arkers of therapy-induced s enescence (40). However, we d id not observe any change in expression of TP53 mRNA in both acquired resistance and DTP models (Supplementary Fig. S8). and CDK2 mRNA levels were down in DTPs, but not significant, suggesting that therapy-induced senescence and these processes are likely independent of TP53 signaling. Overexpression of GPNMB, a transmembrane protein, has been associated with therapeutic resistance and poor overall survival in several tumors , including breast and head and neck cancers (47-49). A role of GPNMB has been reported in the development of resistance to chemotherapy and PD -L1-directed immunotherapy (50, 51) . Our study has identified that GPNMB-overexpression in breast cancer cells was associated with CDK4/6i resistance. Our in vivo studies in LCC2 xenograft model demonstrate that GPNMB overexpression profoundly alters both tumor growth dynamics and therapeutic responsiveness to the CDK4/6 inhibitor abemaciclib. In line with expectatio ns, control (2EV) tumors exhibited a robust reduction in tumor growth following treatment with abemaciclib, confirming the inherent sensitivity of this (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 22 model to CDK4/6 blockade. This finding reinforces the established role of abemaciclib as an effective inhibitor of cell‑cycle progression in hormone‑resistant breast cancer models. In contrast, GPNMB overexpression significantly accelerated tumor growth even in the absence of treatment, emphasizing the potent oncogenic capacity of GPNMB in vivo . In addition, GPNMB overexpression not only promoted tumor growth but also markedly diminished the therapeutic efficacy of abemaciclib. Tumors with GPNMB overexpression were entirely unresponsive to abemaciclib, maintaining growth rates indistinguishable fr om their vehicle‑treated counterpart. These results position GPNMB as a critical determinant of therapeutic resistance to CDK4/6 inhibition. The dual impact of GPNMB as enhancing tumor promotion while simultaneously reducing drug responsiveness highlights its relevance as both a prognostic marker and a potential therapeutic target. Strategies aimed at targeting GPNMB may restore sensitivity when combined with CDK4/6 inhibitors and improve outcomes in tumors with high GPNMB expression. Therapeutic targeting of GPNMB has been attempted to counter aging and to autoimmune diseases (52, 53) in addition to cancer. Elimination of G PNMB-positive cells by senolytic GPNMB vaccination reversed pathological aging in aged mice and prolonged the lifespan of mice with premature aging. GPNMB has been well documented to be an important target in metastatic Triple Negative Breast Cancer (TNBC)(47). In TNBC, an antibody-drug conjugate, Glembatumumab vedotin (GV) has been studied (49) . The phase II METRIC clinical showed that although GV had an acceptable safety profile, it was not associated with improvement in PFS as compared to capecitabine. The lack of clinical benefit in this trial has been attributed, at least in part, to GPNMB being predominantly in the cytosol in TNBC (49, 54) in contrast to our findings of cell surface GPNMB in CDK4/6i- DTPs and GPNMB -OE cells. In summary, our study identifies GPNMB as a previously unrecognized driver of (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 23 resistance to CDK4/6 inhibition. We demonstrate for the first time that GPNMB is markedly overexpressed at both the mRNA and protein levels in abemaciclib and palbociclib DTP models, and that its elevated expression is closely associated with an enhanced senescent phenotype as well as confer ring resistance to CDK4/6i. These findings position GPNMB not only as a functional contributor to CDK4/6 inhibitor resistance , but also as a potential biomarker for stratifying patients who are more likely to experience reduced progression free survival on CDK4/6i therapy. Together, our results highlight GPNMB as a promising therapeutic and prognostic target that warrants further investigation in the context of overcoming or preventing CDK4/6 inhibitor resistance. Data Availability Statement The CEL files for Clariom D Human Array will be deposited at GEO when the manuscript is accepted. Competing Interests The authors declare no conflict of interest. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 24

1. Grinshpun A, Tolaney SM, Burstein HJ, Jeselsohn R, Mayer EL. The dilemma of selecting a first line CDK4/6 inhibitor for hormone receptor -positive/HER2-negative metastatic breast cancer. NPJ Breast Cancer. 2023;9(1):15. 2. George MA, Qureshi S, Omene C, Toppmeyer DL, Ganesan S. Clinical and Pharmacologic Differences of CDK4/6 Inhibitors in Breast Cancer. Front Oncol. 2021;11:693104. 3. Sakach E, Keskinkilic M, Wood S, Canning M, Kalinsky K. CDK4/6 Inhibition in the Metastatic Setting: Where Are We Headed? Curr Treat Options Oncol. 2023;24(9):1103 - 19. 4. Glaviano A, Wander SA, Baird RD, Yap KC, Lam HY, Toi M, et al. Mechanisms of sensitivity and resistance to CDK4/CDK6 inhibitors in hormone receptor-positive breast cancer treatment. Drug Resist Updat. 2024;76:101103. 5. Finkelman BS, Zhang H, Hicks DG, Turner BM. The Evolution of Ki- 67 and Breast Carcinoma: Past Observations, Present Directions, and Future Considerations. Cancers (Basel). 2023;15(3). 6. Hortobagyi GN, Lacko A, Sohn J, Cruz F, Ruiz Borrego M, Manikhas A, et al. A phase III trial of adjuvant ribociclib plus endocrine therapy versus endocrine therapy alone in patients with HR-positive/HER2-negative early breast cancer: final invasive disease-free survival results from the NATALEE trial. Ann Oncol. 2025;36(2):149-57. 7. Fasching PA, Stroyakovskiy D, Yardley DA, Huang CS, Crown J, Bardia A, et al. Ribociclib Plus Endocrine Therapy in Hormone Receptor-Positive/ERBB2-Negative Early Breast Cancer: 4 -Year Outcomes From the NATALEE Randomized Clinical Trial. JAMA Oncol. 2025;11(11):1364-72. 8. Gomatou G, Trontzas I, Ioannou S, Drizou M, Syrigos N, Kotteas E. Mechanisms of resistance to cyclin-dependent kinase 4/6 inhibitors. Mol Biol Rep. 2021;48(1):915-25. 9. O'Leary B, Cutts RJ, Liu Y, Hrebien S, Huang X, Fenwick K, et al. The Genetic Landscape and Clonal Evolution of Breast Cancer Resistance to Palbociclib plus Fulvestrant in the PALOMA-3 Trial. Cancer Discov. 2018;8(11):1390-403. 10. Kudo R, Safonov A, Jones C, Moiso E, Dry JR, Shao H, et al. Long-term breast cancer response to CDK4/6 inhibition defined by TP53 -mediated geroconversion. Cancer Cell. 2024;42(11):1983. 11. Chandarlapaty S, Razavi P. Cyclin E m RNA: Assessing Cyclin- Dependent Kinase (CDK) Activation State to Elucidate Breast Cancer Resistance to CDK4/6 Inhibitors. J Clin Oncol. 2019;37(14):1148-50. 12. Konecny GE, Winterhoff B, Kolarova T, Qi J, Manivong K, Dering J, et al. Expression of p16 and retinoblastoma determines response to CDK4/6 inhibition in ovarian cancer. Clin Cancer Res. 2011;17(6):1591-602. 13. Guerrero-Zotano A, Belli S, Zielinski C, Gil-Gil M, Fernandez-Serra A, Ruiz-Borrego M, et al. CCNE1 and PLK1 Mediate Resistance to Palbocic lib in HR+/HER2 - Metastatic Breast Cancer. Clin Cancer Res. 2023;29(8):1557-68. 14. Kong T, Mabry A, Highkin M, Wang AZ, Hoog J, Guo Z, et al. Biomarkers of response to neoadjuvant palbociclib plus anastrozole in endocrine- resistant estrogen receptor - positive/HER2-negative breast cancer: a phase 2 trial. Nat Commun. 2026;17(1):949. 15. Knudsen ES, Shapiro GI, Keyomarsi K. Selective CDK4/6 Inhibitors: Biologic Outcomes, Determinants of Sensitivity, Mechanisms of Resistance, Combinatorial (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 25 Approaches, and Phar macodynamic Biomarkers. Am Soc Clin Oncol Educ Book. 2020;40:115-26. 16. Alvarez-Fernandez M, Malumbres M. Mechanisms of Sensitivity and Resistance to CDK4/6 Inhibition. Cancer Cell. 2020;37(4):514-29. 17. Formisano L, Lu Y, Servetto A, Hanker AB, Jansen VM, Bauer JA, et al. Aberrant FGFR signaling mediates resistance to CDK4/6 inhibitors in ER+ breast cancer. Nat Commun. 2019;10(1):1373. 18. Jansen VM, Bhola NE, Bauer JA, Formisano L, Lee KM, Hutchinson KE, et al. Kinome-Wide RNA Interference Screen Reveals a Role for PDK1 in Acquired Resistance to CDK4/6 Inhibition in ER-Positive Breast Cancer. Cancer Res. 2017;77(9):2488-99. 19. Lin CC, Chang TC, Wang Y, Guo L, Gao Y, Bikorimana E, et al. PRMT5 is an actionable therapeutic target in CDK4/6 inhibitor -resistant ER+/RB -deficient breast cancer. Nat Commun. 2024;15(1):2287. 20. Bui TBV, Burgering BMT, Goga A, Rugo HS, van 't Veer LJ. Biomarker s for Cyclin- Dependent Kinase 4/6 Inhibitors in the Treatment of Hormone Receptor - Positive/Human Epidermal Growth Factor Receptor 2 -Negative Advanced/Metastatic Breast Cancer: Translation to Clinical Practice. JCO Precis Oncol. 2022;6:e2100473. 21. Finn RS, Liu Y, Zhu Z, Martin M, Rugo HS, Dieras V, et al. Biomarker Analyses of Response to Cyclin-Dependent Kinase 4/6 Inhibition and Endocrine Therapy in Women with Treatment-Naive Metastatic Breast Cancer. Clin Cancer Res. 2020;26(1):110-21. 22. Stanciu IM, Parosanu AI, Orlov -Slavu C, Iaciu IC, Popa AM, Olaru CM, et al. Mechanisms of Resistance to CDK4/6 Inhibitors and Predictive Biomarkers of Response in HR+/HER2-Metastatic Breast Cancer-A Review of the Literature. Diagnostics (Basel). 2023;13(5). 23. Watt AC, Goel S. Cellular mechanisms underlying response and resistance to CDK4/6 inhibitors in the treatment of hormone receptor -positive breast cancer. Breast Cancer Res. 2022;24(1):17. 24. Kalinsky K, Accordino MK, Chiuzan C, Mundi PS, Sakach E, Sathe C, et al . Randomized Phase II Trial of Endocrine Therapy With or Without Ribociclib After Progression on Cyclin -Dependent Kinase 4/6 Inhibition in Hormone Receptor -Positive, Human Epidermal Growth Factor Receptor 2 -Negative Metastatic Breast Cancer: MAINTAIN Trial. J Clin Oncol. 2023;41(24):4004-13. 25. Kalinsky K, Layman RM, Kaufman PA, Graff SL, Bianchini G, Martin M, et al. Abemaciclib plus fulvestrant vs fulvestrant alone for HR+, HER2- advanced breast cancer following progression on a prior CDK4/6 inhibitor plus endocrine therapy: Primary outcome of the phase 3 postMONARCH trial. Journal of Clinical Oncology. 2024;42. 26. Navarro-Yepes J, Kettner NM, Rao X, Bishop CS, Bui TN, Wingate HF, et al. Abemaciclib is effective in palbociclib -resistant hormone receptor -positive metastatic breast cancers. Cancer Res. 2023. 27. Chisholm RH, Lorenzi T, Lorz A, Larsen AK, de Almeida LN, Escargueil A, et al. Emergence of drug tolerance in cancer cell populations: an evolutionary outcome of selection, nongenetic instability, and stress- induced adaptation. Cancer Res. 2015;75(6):930-9. 28. De Conti G, Dias MH, Bernards R. Fighting Drug Resistance through the Targeting of Drug-Tolerant Persister Cells. Cancers (Basel). 2021;13(5). 29. Mikubo M, Inoue Y, Liu G, Tsao MS. Mechanism of Drug Tolerant Persister Cancer Cells: The Landscape and Clinical Implication for Therapy. J Thorac Oncol. 2021;16(11):1798-809. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 26 30. Pu Y, Li L, Peng H, Liu L, Heymann D, Robert C, et al. Drug -tolerant persister cells in cancer: the cutting edges and future directions. Nat Rev Clin Oncol. 2023. 31. Sharma SV, Lee DY, Li B, Quinlan MP, Takahashi F, Maheswaran S, et al. A chromatin-mediated reversible drug -tolerant state in cancer cell subpopulations. Cell. 2010;141(1):69-80. 32. Marsolier J, Prompsy P, Durand A, Lyne AM, Landragin C, Trouchet A, et al. H3K27me3 conditions chemotolerance in triple- negative breast cancer. Nat Genet. 2022;54(4):459-68. 33. Echeverria GV, Ge Z, Seth S, Zhang X, Jeter -Jones S, Zhou X, et al. Resistance to neoadjuvant chemotherap y in triple -negative breast cancer mediated by a reversible drug-tolerant state. Sci Transl Med. 2019;11(488). 34. Risom T, Langer EM, Chapman MP, Rantala J, Fields AJ, Boniface C, et al. Differentiation-state plasticity is a targetable resistance mechanism in basal -like breast cancer. Nat Commun. 2018;9(1):3815. 35. Brunner N, Frandsen TL, Holst- Hansen C, Bei M, Thompson EW, Wakeling AE, et al. MCF7/LCC2: a 4-hydroxytamoxifen resistant human breast cancer variant that retains sensitivity to the steroidal antiestrogen ICI 182,780. Cancer Res. 1993;53(14):3229-32. 36. Brunner N, Boysen B, Jirus S, Skaar TC, Holst -Hansen C, Lippman J, et al. MCF7/LCC9: an antiestrogen-resistant MCF-7 variant in which acquired resistance to the steroidal antiestrogen ICI 182,78 0 confers an early cross -resistance to the nonsteroidal antiestrogen tamoxifen. Cancer Res. 1997;57(16):3486-93. 37. Gokmen-Polar Y, Neelamraju Y, Goswami CP, Gu Y, Gu X, Nallamothu G, et al. Splicing factor ESRP1 controls ER-positive breast cancer by altering metabolic pathways. EMBO Rep. 2019;20(2). 38. Gokmen-Polar Y, Mehta R, Tuzmen S, Mousses S, Thorat MA, Sanders KL, et al. Differential subcellular expression of protein kinase C betaII in breast cancer: correlation with breast cancer subtypes. Breast Cancer Res Treat. 2010;124(2):327-35. 39. Commander R, Wei C, Sharma A, Mouw JK, Burton LJ, Summerbell E, et al. Subpopulation targeting of pyruvate dehydrogenase and GLUT1 decouples metabolic heterogeneity during collective cancer cell invasion. Nat Commun. 2020;11(1):1533. 40. Mammadova A, Gu Y, Ruan L, Badve SS, Gokmen- Polar Y. Non -Canonical Senescence Phenotype in Resistance to CDK4/6 Inhibitors in ER -Positive Breast Cancer. Biomolecules. 2026;16(1). 41. Raha D, Wilson TR, Peng J, Peterson D, Yue P, Evangelista M, et al. The cancer stem cell marker aldehyde dehydrogenase is required to maintain a drug -tolerant tumor cell subpopulation. Cancer Res. 2014;74(13):3579-90. 42. Ramirez M, Rajaram S, Steininger RJ, Osipchuk D, Roth MA, Morinishi LS, et al. Diverse drug -resistance mechanisms can emerge from drug -tolerant cancer persister cells. Nat Commun. 2016;7:10690. 43. Hanahan D. Hallmarks of Cancer: New Dimensions. Cancer Discov. 2022;12(1):31- 46. 44. Chambers CR, Ritchie S, Pereira BA, Timpson P. Overcoming the senescence- associated secretory phenotype (SASP): a complex mechanism of resistance in the treatment of cancer. Mol Oncol. 2021;15(12):3242-55. 45. De Blander H, Morel AP, Senaratne AP, Ouzounova M, Puisieux A. Cellular Plasticity: A Route to Senescence Exit and Tumorigenesis. Cancers (Basel). 2021;13(18). 46. Kudo R, Safonov A, Jones C, Moiso E, Dry JR, Shao H, et al. Long-term breast cancer response to CDK4/6 inhibition def ined by TP53 -mediated geroconversion. Cancer Cell. 2024. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 27 47. Huang YH, Chu PY, Chen JL, Huang CT, Huang CC, Tsai YF, et al. Expression pattern and prognostic impact of glycoprotein non- metastatic B (GPNMB) in triple- negative breast cancer. Sci Rep. 2021;11(1):12171. 48. Rose AA, Annis MG, Dong Z, Pepin F, Hallett M, Park M, et al. ADAM10 releases a soluble form of the GPNMB/Osteoactivin extracellular domain with angiogenic properties. PloS one. 2010;5(8):e12093. 49. Rose AAN, Biondini M, Curiel R, Siegel PM. Targeting GPNMB with glembatumumab vedotin: Current developments and future opportunities for the treatment of cancer. Pharmacol Ther. 2017;179:127-41. 50. Xu X, Xie K, Li B, Xu L, Huang L, Feng Y, et al. Adaptive resistance in tumors to anti- PD-1 therapy through re-immunosuppression by upregulation of GPNMB expression. Int Immunopharmacol. 2021;101(Pt B):108199. 51. Chung JS, Ramani V, Kobayashi M, Fattah F, Popat V, Zhang S, et al. DC-HIL/Gpnmb Is a Negative Regulator of Tumor Response to Immune Checkpoint Inhibitors. Clin Cancer Res. 2020;26(6):1449-59. 52. Tsou PS, Sawalha AH. Glycoprotein nonmetastatic melanoma protein B: A key mediator and an emerging therapeutic target in autoimmune diseases. FASEB J. 2020;34(7):8810-23. 53. Suda M, Shimizu I, Katsu umi G, Yoshida Y, Hayashi Y, Ikegami R, et al. Senolytic vaccination improves normal and pathological age- related phenotypes and increases lifespan in progeroid mice. Nat Aging. 2021;1(12):1117-26. 54. Biondini M, Kiepas A, El -Houjeiri L, Annis MG, Hsu BE, Fortier AM, et al. HSP90 inhibitors induce GPNMB cell -surface expression by modulating lysosomal positioning and sensitize breast cancer cells to glembatumumab vedotin. Oncogene. 2022;41(12):1701-17.

This work was supported partly by the Susan G. Komen for the Cure to Sunil Badve (SAC220219) and Winship Invest$$ pilot grant to Yesim Gokmen- Polar. In addition, Startup funds from Emory University were also utilized. Sunil Badve is a Komen scholar. Research reported in this publication was supported in part by the Pediatrics/Winship Flow Cytometry Core of Winship Cancer Institute of Emory University, Children's Healthcare of Atlanta (NIH/NCI) and the Cancer Animal Models shared resource of Winship Cancer Institute of Emory University under award number P30CA138292) , and Emory University Integrated Cellular Imaging Core and Children’s Healthcare of Atlanta, (RRID:SCR_023534). The content is solely the responsibility of the authors and does not necessarily reflect (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 28 the official views of the National Institute of Health. E thics Approval / Consent to Participate All experimental procedures were approved by the Emory University Institutional Biosafety Committees. All protocols related to patient samples were reviewed and approved by the Institutional Review Board (IRB) of Emory University. Samples and clinical records were de-identified prior to access by the authors and linked with a numerical identifier. The IRB waived requirement of informed consent. Author Contributions YG -P, and SSB developed the study concept and design, obtained funding, and performed the interpretation, manuscript writing and execution of the entire project; YG and LR performed the experiments and statistical analyses. YH and MG-R carried out in vivo experiments and analyses, TB guided the entire project, KMK advised the conduct of the study and interpretation of the data. All authors have read and approved the manuscript. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 29 Figure legends Fig 1. Phenotype of CDK4/6i -treated drug tolerant persister cells (LCC2 DTPs) versus sensitive (LCC2 control) cells. A Representative sensitivity of LCC2 control treated with vehicle and LCC2 DTPs treated with abemaciclib (top) and palbociclib (bottom) with the indicated concentrations for 72hr using CyQUANT Direct Cell Proliferation Assay. IC50 Data (n = 3) were calculated using GraphPad Prism. B Representative cell cycle distribution of the LCC2 control and LCC2-DTPs developed to abemaciclib (top) and palbociclib (bottom). Cells at G1, S, and G2-M phases for each condition are shown in percentage (mean ±SD, n = 3). C Representative senescence-associated β-galactosidase (SA-β-gal) expression using Senescence β-Galactosidase Staining Kit and D Representative FACS plots showing SA-β-gal expression in LCC2 control and LCC2 DTPs developed to abemaciclib (top) and palbociclib (bottom) using CellEvent Senescence Green probe specific to β-gal. Numbers indicate the percentage of senescent cells. Fig. 2. Transcriptomic analysis of DTPs to CDK4/6i reveals enrichment of ECM pathways in comparison to their sensitive counterparts. A Venn diagram showing the differential expression analysis of DTPs compared with cells with their drug sensitive counterparts calculated by Transcriptome Analysis Console (TAC 4.0) software (Top: downregulated genes in DTPs, 2 2 cutoff (P < 0.05); bottom: upregulated genes in DTPs): Set 1- LCC2 DTPs versus LCC2 control, set 2- LCC9 DTPs versus LCC9 control, set 3- MCF7 DTPs versus MCF7 control and set 4- T47D DTPs versus T47D control. B g:profiler pathway analysis for down-regulated genes (top) and up regulated genes (bottom) in DTPs compared their sensitive counterparts. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 30 Fig. 3. GPNMB is elevated in CDK4/6i- DTP cells at both mRNA and protein levels . A Fold change (mean ± SD, n = 3; **P < 0.01) of GPNMB mRNA expression levels were assessed by RT -qPCR in CDK4/6i-tolerant persisters ( LCC2-DTPs) compared with LCC2 control cells. Unpaired t-test with Welch’s correction analyses were performed using GraphPad Prism 10.3.1 software as indicated B FACS-based analysis of cell-surface GPNMB expression in LCC2-DTPs compared to LCC2 control cells. High GPNMB cell surface expression in DTPs with the GPNMB antibody. C Fluorescent imaging of G PNMB protein in LCC2 DTPs and LCC2 control. GPNMB (Cat# sc -271415, Santa Cruz, CA, USA, dilution 1:50). Secondary antibody F(ab’)2-Goat anti-Mouse IgG (H+L) Cross-Absorbed, Alexa Fluor Plus 555) Fig. 4. Expression analysis of GPNMB overexpression in LCC2 cells. LCC2 cells were transduced either with empty vector (2EV) or GPNMB construct ( 2GPNMB-OE-C1) using lentiviral based transduction system. A GPNMB mRNA verification in 2GPNMB- OE-C1 compared to 2EV using RT -qPCR. U npaired t -test with Welch’s correction analyses were performed using GraphPad Prism 10.3.1 software (*** P<0.001). B GPNMB protein expression using spheroid immunofluorescent assay. C Representative sensitivity of 2EV and 2GPNMB-OE-C1 cells treated with the indicated concentrations of abemaciclib for 72hr using CyQUANT Direct Cell Proliferation Assay (n = 3). D Representative sensitivity to abemaciclib and E percentage of wound closure coverage healing were performed using LCC2 cells plated on rhGPNMB or 0.1 % BSA coated dishes; (n = 3). For statistical significance, unpaired t-test with Welch’s correction analyses were performed using GraphPad Prism 10.3.1. Fig. 5. Overexpression of GPNMB results in significant tumor growth and confers resistance to abemaciclib. Tumor growth and response to abemaciclib were assessed in mice bearing empty vector (2EV) or GPNMB overexpression (2GPNMB OE C1) in LCC2 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint 31 mammary xenografts and treated with vehicle as a suspension in 0.5% methylcellulose or abemaciclib at a dose of 50 mg/kg/once daily for 35 consecutive days via oral gavage (PO). Four experimental groups were analyzed: (1) 2EV Vehicle (red symbol), (2) 2EV Abemaciclib (blue symbol), (3) 2GPNMB OE C1 Vehicle (black symbol), and (4) 2GPNMB OE C1 Abemaciclib (green symbol). Tumor growth data are presented as mean ± SEM. Statistical significance between treatment groups was determined using two-way ANOVA with repeated measures, followed by appropriate post hoc multiple-comparison tests. An adjusted (adj.) p value < 0.05 was considered statistically significant. All statistical analyses were performed using GraphPad Prism v10.3.1 software. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint Abemaciclib (nM) Cell Viability (%) Palbociclib (nM) Cell Viability (%) A B LCC2 DTPsLCC2 control Count LCC2 control LCC2 DTPs Count Figure 1 Abemaciclib Palbociclib (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint LCC2 control LCC2-DTPs C D LCC2 control 3 % LCC2-DTPs 77% LCC2 control 28% LCC2-DTPs 88% LCC2 control LCC2-DTPs Figure 1 AbemaciclibPalbociclib Abemaciclib Palbociclib (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint Set 1 Set 2 Set 3 Set 4 B Figure 2 A Up in DTPs Set 1 Set 3 Set 4 Set 2 Down in DTPs (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint Figure 3 CA B GPNMB LCC2 control LCC2-DTPs LCC2 control LCC2-DTPs GPNMB Abemaciclib GPNMB 99.3% GPNMB 0.09% LCC2 control LCC2 DTPs LCC2 control LCC2-DTPs GPNMB 0.45% GPNMB 95.5% Abemaciclib Palbociclib Palbociclib (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint Cell Viability (%) Cell Viability (%) Abemaciclib (nM) Abemaciclib (nM) Figure 4 C D A B 2EV 2GPNMB-OE-C1 E Wound Closure (%) GPNMB (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint # Group Comparisons Significance Adj. P value 1 2EV Vehicle vs 2EV Abemaciclib Yes <0.0001 2 2EV Vehicle vs 2GPNMB OE C1 Vehicle Yes 0.0356 3 2EV Vehicle vs 2GPNMB OE C1 Abemaciclib No 0.4548 4 2EV Abemaciclib vs 2GPNMB OE C1 Vehicle Yes <0.0001 5 2EV Abemaciclib vs 2GPNMB OE C1 Abemaciclib Yes <0.0001 6 2GPNMB OE C1 Vehicle vs 2GPNMB OE C1 Abemaciclib No 0.5456 Figure 5 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 7, 2026. ; https://doi.org/10.64898/2026.03.04.709413doi: bioRxiv preprint