Metabolic response to CDK4/6 inhibition in ER+ breast cancer creates a therapeutic vulnerability in drug-tolerant persister cells

Abstract

Although endocrine therapies prevent recurrence and progression of estrogen receptor alpha (ER)-positive breast cancer, approximately one-third of patients experience recurrent disease that is rarely cured in the advanced/metastatic setting. A subpopulation of endocrine-tolerant breast cancer cells persists as residual disease that confers risk for the eventual emergence of drug resistance. An analysis of persisters that continue to proliferate despite endocrine therapy revealed the activation of pathways related to metabolism and E2F transcription factor signaling. E2F signaling is driven by cyclin-dependent kinases 4 and 6 (CDK4/6), and CDK4/6 inhibitors (CDK4/6i) are used clinically to prevent and manage endocrine resistance. CDK4/6i slowed the cycling of endocrine-tolerant persisters. Analyzing metabolic alterations induced by CDK4/6i, we found that CDK4/6i-tolerant persisters had upregulation of mitochondrial content, mitochondrial membrane potential, respiration, and reactive oxygen species (ROS). Inhibition of mitochondrial complex I further increased ROS levels and enhanced growth inhibition in both endocrine-sensitive and -resistant cell lines and patient-derived xenografts. These findings collectively offer mitochondrial respiration as a therapeutic target in CDK4/6-tolerant persister breast cancer cells to help eradicate residual disease.

Introduction

Approximately two-thirds of breast tumors express estrogen receptor alpha (ER), which typically reflects a degree of dependence upon estrogens for growth without amplification/overexpression of the gene encoding HER2. Patients with early-stage ER+/HER2- breast cancer are commonly treated with surgery followed by adjuvant endocrine therapies that antagonize ER (e.g., tamoxifen) or aromatase inhibitors that suppress estrogen biosynthesis to inhibit ER transcriptional activity. Approximately 30% of these patients experience disease recurrence within 20 years of follow-up that is often metastatic (1). In the advanced/metastatic setting, disease is often managed systemically with drugs that include the ER-downregulating anti-estrogen fulvestrant (fulv), and inhibitors of cyclin-dependent kinases 4 and 6 (CDK4/6i) such as palbociclib, ribociclib, and abemaciclib. Abemaciclib and ribociclib are also approved for clinical use in combination with endocrine therapy in the adjuvant setting for patients at high risk of recurrence, and abemaciclib as monotherapy in the endocrine-resistant metastatic setting.

As the rate of recurrence of ER+ breast cancer remains nearly constant for ~20 years, a large proportion of recurrences occur after the completion of 5–10 years of adjuvant endocrine therapy. Recurrences are caused by cancer cells that persist despite adjuvant therapy. Even with the combination of abemaciclib and endocrine therapy in a high-risk population, risk of recurrence is 32.0% (95% CI: 22.8%-41.1%) (2). Therefore, understanding signaling mechanisms that drive tolerance of endocrine and CDK4/6i therapies is essential to develop therapeutic strategies to prevent and overcome persistence. One output of ER signaling is activation of CDK4/6 that phosphorylate retinoblastoma protein (Rb), which in turn activates E2F transcription factors that drive G1-to-S cell cycle progression (3,4). We and others showed that CDK4/6 activation, including through loss of the regulatory activity of Rb, can drive resistance to endocrine therapy through activation of E2F transcription factors (4–6), providing rationale for the therapeutic development of CDK4/6i in ER+ breast cancer. However, CDK4/6i-tolerant cells persist to often give rise to recurrent/progressive disease. We recently described the importance of oxidative phosphorylation (OXPHOS) in endocrine-tolerant persisters in ER+ breast cancer (7), and others have observed metabolic shifts in drug-tolerant persisters in other cancer types (8–12). Herein, we explored the role of OXPHOS in ER+ breast cancer cells that persist during CDK4/6i treatment, and the effects of Rb loss that confers resistance to CDK4/6i.

Materials and Methods

Cell culture and chemicals

MCF-7, T47D, HCC-1428, ZR75–1, and CAMA-1 cells were obtained from American Type Culture Collection (ATCC). Cells were maintained in DMEM supplemented with 10% FBS (HyClone Laboratories). A fulv-resistant (FR) derivative of MCF-7 cells (MCF-7/FR) was a gift from Matthew Ellis (Washington University). T47D/FR cells were generated through culture in 1 uM fulv for 4 months until a resistant line was established. FR cells were maintained in medium containing 1 uM fulv (Tocris Bioscience). Hormone-depleted medium (HD) was phenol red-free DMEM containing 10% dextran/charcoal-stripped FBS (DCC-FBS; Hyclone Laboratories) and 2 mM Glutamax (ThermoFisher Scientific). Rb-knockout cell lines (MCF-7/Rb-, T47D/Rb-, CAMA-1/Rb-) were generated by CRISPR/Cas9-mediated editing of RB1 (13) and provided as gifts from Erik Knudsen (Roswell Park Comprehensive Cancer Center). Cell lines were confirmed to be mycoplasma-free (Universal Mycoplasma Detection Kit; ATCC) and authenticated by STR genotyping (University of Vermont Cancer Center DNA Analysis Facility). IACS-010759, palbociclib, and abemaciclib were obtained from Selleck Chemicals. All other chemicals were obtained from Sigma unless otherwise stated.

Serial cell imaging and analysis

Cells were seeded in 12-well plates (10,000–50,000 cells/well). Plates were imaged every 3–4 d on a Cytation 5 (BioTek) with 4x magnification and phase contrast. Images were analyzed and cell number was quantified with Gen5 software (Biotek).

Real-time cellular metabolic analysis (Seahorse)

Cells were pretreated as indicated and then seeded in 96-well Seahorse XF96 plates (40,000–60,000 cells/well; Agilent) and allowed to adhere overnight. Oxygen consumption rate (OCR) was serially measured using the Seahorse XF96 Analyzer (Agilent). Mitochondrial respiration was inferred using OCR measurements using the Seahorse XF Cell Mito Stress Test Kit.

Flow cytometry

Cells were trypsinized and resuspended in serum-containing medium with either 100 nM MitoTracker Deep Red, 100 nM TMRE, 5 uM CellROX Deep Red (ThermoFisher Scientific #C10422), or 5 uM MitoSox (ThermoFisher Scientific #M36007). For cell cycle analysis, cells were fixed in 70% ethanol at −20°C overnight, washed with PBS, and stained with propidium iodide/RNase staining solution (ThermoFisher Scientific) for 30 min at room temperature in the dark. Samples were analyzed on a MACSQuant-10 (Miltenyi Biotec) or a ZE5 Cell Analyzer (Bio-Rad). Data were analyzed using FlowJo 10.8.2 software (BD Biosciences). Cell cycle analysis was performed using the Cell Cycle platform within FlowJo with the Watson Pragmatic algorithm.

Immunoblotting

Cells were lysed and frozen tumor fragments were homogenized in RIPA buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 1% NP-40, 10% glycerol, 1 mM EDTA, 1 mM EGTA, 5 mM NaPPi, 50 mM NaF, 10 mM Na b-glycerophosphate) with HALT protease inhibitor cocktail and 1 mM Na₃VO₄. Lysates were sonicated for 10 sec and centrifuged at 17,000 × g for 10 min at 4°C. Supernatant was collected, and protein concentration was measured by BCA assay (Pierce). Cell lysates were reduced and denatured using NuPAGE (ThermoFisher Scientific) plus 1.25% b-mercaptoethanol. Fifty ug of protein/sample was analyzed by SDS-PAGE. Protein was transferred onto nitrocellulose membrane and blocked with 5% BSA in TBS containing 0.1% Tween-20 (TBS-T) for 1 h. Membranes were incubated with primary antibody overnight at 4°C on a shaker in blocking solution. Primary antibodies included Rb (Cell Signaling 9313), pRb_S807/811 (Cell Signaling 8516), OXPHOS antibody cocktail (Abcam ab110411), b-actin (Cell Signaling Technology 3700), and vinculin (Cell Signaling Technology 13901). Membranes were then washed with TBS-T 3 times for 10 min, and incubated with HRP-conjugated secondary antibody in 5% milk in TBS-T for 1 h. Secondary antibodies included HRP-conjugated anti-mouse (Cytiva NA931V) and HRP-conjugated anti-rabbit (Cytiva NA9340V). After 3 washes with TBS-T for 10 min each, signal was developed using Pierce ECL Western Blotting Substrate or Supersignal West Pico PLUS substrate, and blots were imaged using a Chemidoc MP (Bio-Rad).

Animal studies

Animal studies were approved by the Dartmouth College IACUC (protocol 00002144). Female NOD/SCID/IL2Rγ^−/− (NSG) mice were obtained from the Dartmouth Cancer Center Mouse Modeling Shared Resource. The HCI-003 patient-derived xenograft (PDX) tumor model was obtained from University of Utah (14). PDX tumor fragments (~8-mm³) from a donor mouse were orthotopically implanted into experimental mice at 4 wk of age. Mice were supplemented with exogenous E2 (1 mg) via s.c. beeswax pellet (15). When tumors reached 200 mm³, mice were randomized to treatment with vehicle, IACS, abemaciclib, or the combination. IACS-010759 was dissolved in DMSO at 15 mg/mL, diluted in a 0.5% methylcellulose suspension, and administered by oral gavage in 100 uL at 2.5 mg/kg/d for 5 consecutive d/wk. Abemaciclib was dissolved in 1% hydroxyethyl cellulose in 20 mM phosphate buffer (pH 2.0), and administered by oral gavage in 100 uL at 50 mg/kg/d for 5 consecutive d/wk.

Watermelon reporter studies

Lentivirus was generated using LentiX cells transiently transfected with 3 plasmids (pMD2.G, psPAX2, and Watermelon backbone) using Lipofectamine 2000. pMD2.G and psPAX2 were gifts from Didier Trono (Addgene 12259 and 12260; RRID:Addgene_12259 and RRID: Addgene_12260.) Watermelon backbone (16) was a gift from Joan Brugge and Aviv Regev (Addgene 155258; RRID:Addgene_155258). Lentivirus was used to infect MCF-7 cells, which were then used for fluorescence-activated cell sorting (FACS) to collect mNeonGreen-positive cells to yield the stably transfected MCF-7/Watermelon cell line.

MCF-7/Watermelon cells were treated with 0.5 ug/mL doxycycline (dox) for 3 d to induce histone H2B-mCherry expression. Cells were then use for FACS to isolate mCherry+ cells that were reseeded in triplicate and treated as indicated. Cells were again sorted by FACS to isolate mCherry-high (slow-cycling or non-cycling) and mCherry-low (fast-cycling) persister subpopulations, and RNA was extracted using RNeasy Plus Mini Kit (Qiagen 1062832).

RNA quality was assessed on a fragment analyzer (Advanced Analytical Technologies, Agilent), and RNA was quantified by Qubit. In preparation for sequencing (RNA-seq), polyA libraries were prepared from 2.5 ug of total RNA using NEBNext Ultra II RNA Library Prep Kit for Illumina with Sample Purification Beads (New England Biolabs E7775) and TruSeq Stranded Total RNA (Illumina RS-122–2201) workflows according to manufacturer’s instructions. Each library was uniquely barcoded, quantified by qPCR (Kapa Biosystems KK4824), and pooled for sequencing on an Illumina NextSeq 500 (2 × 75-bp). Reads were checked for quality using fastqc (RRID:SCR_014583) and if necessary were trimmed using Trimmomatic (RRID:SCR_011848) to trim regions with phred Q>30 (RRID:SCR_001017). High-quality reads were then aligned to reference genome hg19 using STAR. Gene counts were normalized by frequency per kilobase million (17). Differential expression of genes was determined using the limma (RRID:SCR_010943) and DESeq2 (RRID:SCR_015687) packages in the R environment (RRID: SCR_001905), and multiple testing correction was performed using the FDR Benjamini–Hochberg method. To determine significant gene expression pathway enrichment between time points, we conducted gene set enrichment analysis (GSEA) for Hallmarks pathways in GenePattern (18) using default arguments.

Metabolite extraction and LC-MS/MS analysis

Analyses were performed in the MCW Cancer Center Mass Spectrometry Core (RRID: SCR_027908). Frozen tumor tissue fragments were weighed, transferred to Dounce glass homogenizers, and mechanically disrupted.. Ice-cold 80% methanol was then added to tissue lysate to normalize extraction volume to tissue weight (10 uL/mg tissue), followed by vortexing for 1 min and incubation on ice for 30 min. Mixtures were centrifuged at 16,000 × g for 15 min at 4°C. The upper clear layer (800 uL) was transferred to a new microcentrifuge tube, dried using a SpeedVac, and stored at −80°C. Dried extracts were reconstituted in 200 uL of methanol:water (1:1), vortexed for 2 min, filtered through a 0.2-um Nanosep MF filter, and subjected to LC–MS/MS analysis using an Orbitrap Exploris 240 mass spectrometer (Thermo Scientific) coupled to an ultra-high-performance liquid chromatography unit (Vanquish Flex, Thermo Scientific). Ions of analytes were generated using a heated electrospray ionization (HESI) source. Sample (2 uL) was injected by an autosampler, and metabolites were separated in a HILIC column (SeQuant ZIC-HILIC, 3.5 um, 100 mm × 2.1 mm, PEEK coated, Millipore Sigma) equipped with a guard column (SeQuant ZIC-HILIC, 20 mm × 2.1 mm, PEEK coated, Millipore Sigma). Throughout the analysis, the LC column and autosampler chamber were kept at 30°C and 10 °C, respectively. The mobile phase (A and B) was delivered at a flow rate of 0.2 mL/min for elution. Mobile phase A was prepared as follows: 20 mM ammonium acetate and 0.05% acetic acid in water. Mobile phase B consisted of acetonitrile. The LC gradient started at 90% mobile phase B and was held for 3 min, followed by a linear decrease to 40% B from 3 to 19 min, and held at 40% B for 2 min. The column returned to 90% B at 22 min, and re-equilibrated at 90% B until 32 min. A blank run (methanol-water, 1:1) was performed before each sample injection. The source settings were as follows: polarity, switching between positive and negative; ion transfer tube temperature, 325°C; vaporizer temperature, 350°C; spray voltage, 3.5 kV for positive and 3.0 kV for negative polarity; sheath gas flow, 40 (arb). auxiliary gas flow, 16 (arb); sweep gas flow, 1 (arb). Data were acquired in both full MS and data-dependent MS/MS (dd-MS2) scan modes using Xcalibur v3.0 Software (Thermo Scientific). The settings for full MS mode were as follows: mass resolving power, 120,000 (FWHM, at m/z 200); m/z range, 70–1050; RF lens, 50%; automatic gain control (AGC) target, 1 × 10⁶; maximum sample injection time (IT), 100 ms; and microscans, 1. The following conditions were used for dd-MS2 modes: mass resolving power, 30,000 (FWHM, at m/z 200); number of dependent scan (TopN), 5; isolation window, 1 m/z; first mass, m/z 50; AGC target, 5× 10⁵; maximum IT, 100 ms; collision energy, normalized (20%, 40%, and 60%); and microscans, 1. Intensity threshold was 1 × 10⁴. EASY-IC feature was enabled for lock mass correction. Dynamic exclusion properties were as follows: excluding after 1 time; exclusion duration, 5 s; mass tolerance, 5 ppm; and isotopes exclusion enabled. Peak selection, tentative identification, peak integration and alignment were performed using Compound Discoverer software (ver. 3.3.3.200; Thermo Scientific). The workflow used was “Untargeted metabolomics with statistics: detect unknowns with ID using online databases and mzLogic.” After data processing, metabolites were filtered applying a mass-error tolerance of ±5 ppm, RT (min) range of 3 to 22 and an mzCloud best-match score >70. In addition, chromatograms were manually inspected and excluded if they could not be distinguished from noise.

Statistical Testing

Flow cytometry data were analyzed by ANOVA followed by Bonferroni multiple comparison-adjusted posthoc testing between groups. Tumor growth data were analyzed using the following linear mixed model on the logarithmic scale: Log₁₀(tumor volume_it) = a_i + b * t + e_it, where i represents the i^th mouse, t represents time of tumor volume measurement, a_i represents the mouse-specific log tumor volume at t = 0, b represents the rate of tumor volume growth, and e_it represents deviation of measurements from the model over time (19,20). Mouse heterogeneity (in baseline tumor volume) is represented by variance of a_i, and b*log_e(10) * 100 indicates tumor volume increase (%) per week. Treatment groups were compared using a z-test for slopes with standard error derived from the output of the function ‘lme’ from the library nlme in R. Synergy was determined as described in ref. (21).

Data availability

RNA sequencing data are available at NCBI SRA under accession PRJNA1281637. All other raw data generated in this study are available upon request from the corresponding author.

Results

Fast-cycling endocrine-tolerant persister ER+ breast cancer cells exhibit metabolic adaptations

Much of the work to date on drug-tolerant persister cells makes the assumption that all persisters in a population exhibit the same biology. However, emergence from persistence is highly variable as evidenced by the recurrence of a limited number of tumors at metastatic sites and the limited clonal variability therein (22,23). We employed the Watermelon reporter system that was recently developed to enable tracking of proliferation kinetics. MCF-7 cells were stably transfected with Watermelon encoding constitutively expressed mNeonGreen and doxycycline (dox)-inducible histone H2B fused to mCherry (16). MCF-7/Watermelon cells were pulse-treated with dox for 3 d to induce labeling with H2B-mCherry, and mCherry+ cells were collected, reseeded (without dox), and treated for 7 d with either vehicle or 1 uM fulv in growth medium, or with hormone-depleted (HD) medium (Figure 1A). mCherry signal intensity is diluted with each round of cell division (16). We observed that the extent of mCherry dilution corresponded with MCF-7 growth kinetics: control-treated cells grew and diluted the mCherry label more than cells persisting during HD and fulv (Figure 1A).

Figure 1: Fast-cycling endocrine-tolerant persister ER+ breast cancer cells exhibit upregulation of metabolic gene expression programs.

(A) MCF-7/Watermelon cells were treated with dox for 3 d, and mCherry+ cells were collected by FACS and reseeded. Cells were then treated in triplicate for 7 d with vehicle control or 1 uM fulv in growth medium, or with HD medium. Cells were analyzed by FACS. Gates in representative plots reflect fast-cycling and slow-cycling persisters based on higher vs. lower mCherry signal intensity, respectively. (B-C) Fast- and slow-cycling FACS-collected persisters from HD- and fulv-treated cells in (A) were used for RNA-seq. Transcriptomes of fast- vs. slow-cycling cells were compared by Gene Set Enrichment Analysis (GSEA) for Hallmark gene sets. NES: normalized enrichment score.

Prior work with Watermelon in drug-tolerant persister lung cancer cells showed that a subpopulation of “fast-cycling persisters,” which continued proliferating despite drug treatment, exhibited metabolic differences compared to “slow-cycling (or non-cycling) persisters” (16). We therefore isolated subpopulations of fast- and slow-cycling persisters based on mCherry signal intensity (Figure 1A) and performed transcriptomic profiling. Gene set enrichment analysis (GSEA) of transcriptomes revealed enrichment for gene sets associated with the cell cycle, OXPHOS, and fatty acid metabolism in fast-cycling persisters during HD (Figure 1Band fulv (Figure 1C); evaluation of the genes underlying enrichment for the OXPHOS gene set revealed strong contributions from genes that encode components of the electron transport chain (Supplementary Tables S1 and S2). Therefore, fast-cycling persisters that are expected to be the most likely subpopulation to drive disease recurrence exhibit metabolic adaptations that may be therapeutically tractable.

CDK4/6 inhibition suppresses cell cycling in fast-cycling endocrine-tolerant persisters

CDK4 and CDK6 phosphorylate Rb family proteins, inhibiting Rb and enabling activation of E2F transcription factors. We confirmed the abilities of the CDK4/6 dual inhibitors (CDK4/6i) palbociclib and abemaciclib to suppress Rb phosphorylation and prevent progression to S-phase in parental ER+ breast cancer cells (Figure 2A–B). Evaluation of two sublines with acquired resistance to fulv revealed maintained and reduced sensitivity to CDK4/6i in MCF-7/FR and T47D/FR cells, respectively (Figure 2C). We then tested whether cycling of fulv-tolerant persisters can be arrested by CDK4/6i. MCF-7/Watermelon cells were pulse-labeled by dox treatment, sorted to collect mCherry+ cells, treated with fulv for 4 d, and then treated with vehicle or the CDK4/6i palbociclib or abemaciclib for 7 d. Flow cytometry analysis of fulv→CDK4/6i-tolerant persisters for mCherry signal showed that CDK4/6i decreased the proportions of fast-cycling and moderate-cycling persisters, leading to accumulation in G1 (Supplemental Figure 1). These observations support the use of CDK4/6i to target endocrine-tolerant cycling persisters to prevent recurrence/progression.

Figure 2: CDK4/6 inhibitors promote cell cycle arrest.

(A) Cells were treated +/− 500 nM palbociclib or abemaciclib for 7 d, and lysates were analyzed by immunoblot. Ratios of phospho-Rb:total Rb are noted below each lane. (B) Cells were treated as in (A) in triplicate for 7 d, fixed, stained with propidium iodide, and analyzed by flow cytometry to quantify proportions of cells in each phase of the cell cycle. (C) Fulv-resistant (FR) cells maintained in medium containing 1 uM fulv were treated and analyzed as in (B). Data are shown as mean +/− SD. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001 by Bonferroni-adjusted posthoc test of proportions of cells in S-phase. ns: not significant.

CDK4/6 inhibition increases mitochondrial content and membrane potential

Based on the inference of metabolic adaptations in fast-cycling endocrine-tolerant persisters (Figure 1B–C), and our prior reports describing increased mitochondrial content and dependence upon OXPHOS for ATP production in such persisters (7,24), we probed mitochondrial adaptations to CDK4/6i. Parental and FR cells treated with CDK4/6i showed significantly increased mitochondrial content and membrane potential (Figure 3A–B and Supplemental Figures 2A–B and 3A–B). In line with those observations, CDK4/6i consistently increased oxygen consumption rate (OCR) associated with basal respiration and spare capacity (Figure 3C and Supplemental Figure 2C). Oxygen consumption was nearly completely suppressed upon treatment with the mitochondrial complex I inhibitor IACS-010759 (Figure 3C and Supplemental Figure 2C–D), linking basal and CDK4/6i-induced OCR to OXPHOS.

Figure 3: CDK4/6 inhibition increases mitochondrial content, membrane potential, and respiration.

(A-B) Cells were treated in triplicate +/− 500 nM palbociclib, 500 nM abemaciclib, or 50 nM IACS for 7 d. Cells were then labeled with MitoTracker (A) or TMRE (B) for 30 min prior to flow cytometry analysis. Signal median fluorescence intensity (MFI) was calculated for each sample. (C) Cells were treated as in (A) for 4 d and then analyzed by Seahorse Mito Stress Test. Serially collected oxygen consumption rates (OCR) were used to calculate basal respiration, ATP consumption, and spare capacity. Data are shown as mean +/− SD. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001 by Bonferroni-adjusted posthoc test.

CDK4/6 inhibitor-induced metabolic adaptations are dependent upon Rb function

Loss of Rb function confers resistance to CDK4/6i through deregulated (CDK4/6-independent) activation of E2F transcription factors (25). We therefore evaluated the effects of Rb (RB1) knockout on metabolic response to CDK4/6i in ER+ breast cancer cells. Indeed, the cycling of Rb-deficient cells was only modestly affected by CDK4/6i (Supplemental Figure 4). Rb loss decreased mitochondrial content and membrane potential compared to Rb-proficient CAMA-1 and T47D cells, but not MCF-7 cells (Figure 4A–B). Palbociclib generally did not alter these mitochondrial phenotypes in Rb-deficient cells, with the exception of increased membrane potential in CAMA-1/Rb-knockout cells (Figure 4B). However, abemaciclib robustly increased mitochondrial content and membrane potential in Rb-knockout cells (Figure 4A–B), which may be attributable to the broader repertoire of kinases (e.g., CDK9, DYRK1A, HIP kinases) reportedly inhibitable by abemaciclib compared to palbociclib (26,27). Both drugs increased OCR in Rb-deficient T47D cells, but only abemaciclib elicited such effects in MCF-7 cells (Figure 4C).

Figure 4: Rb is required for mitochondrial response to CDK4/6 inhibition.

(A) RB1-null and parental control cells were treated +/− 500 nM palbociclib or abemaciclib for 7 d. Cells were then labeled with MitoTracker (A) or TMRE (B) for 30 min prior to flow cytometry analysis. Signal median fluorescence intensity (MFI) was calculated for each sample. (C) Cells were treated as in (A) for 4 d and then analyzed by Seahorse Mito Stress Test. Serially collected oxygen consumption rates (OCR) were used to calculate basal respiration, ATP consumption, and spare capacity. Data are shown as mean +/− SD. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001 by Bonferroni-adjusted posthoc test. ns: not significant.

CDK4/6 inhibition promotes oxidative stress that can be exacerbated by mitochondrial complex I inhibition

Given the effects of CDK4/6i on mitochondria and OCR, we postulated that CDK4/6i could increase the generation of reactive oxygen species (ROS). Indeed, CDK4/6i with palbociclib or abemaciclib increased the levels of mitochondrial and cellular ROS (Figure 5A–B and Supplemental Figures 3C–D and 5). Interestingly, single-agent IACS-010759 also increased ROS levels in multiple cell lines, suggesting that mechanisms independent of complex I can drive ROS production. Combined treatment with IACS-010759 and a CDK4/6i elicited variable effects on cellular ROS but consistently increased mitochondrial ROS more than single-agent treatments (Figure 5A). In parallel, combination treatment most effectively suppressed growth, but apoptotic effects varied between cell lines (Figure 5C–D and Supplemental Figure 6); however, co-treatment with the antioxidant N-acetylcysteine (NAC) did not prevent drug-induced apoptosis.

Figure 5: Inhibition of CDK4/6 and complex I increases oxidative stress in ER+ breast cancer cells.

(A-B) Cells were treated in triplicate +/− 500 nM palbociclib, 500 nM abemaciclib, or 50 nM IACS for 7 d. Cells were then labeled with MitoSox (A) or CellRox (B), and analyzed by flow cytometry. Signal median fluorescence intensity (MFI) was calculated for each sample. (C-D) Cells were treated as in (A), and growth was serially measured using Incucyte imaging. Areas under curves were used for statistical comparisons. Data are shown as mean +/− SD. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001 by Bonferroni-adjusted posthoc test. ns: not significant.

Given the promising effects of combined IACS-010759 and CDK4/6i treatment in vitro, we tested these agents against the HCI-003 PDX model of ER+ breast cancer that we previously showed to be endocrine-resistant (28). Single-agent abemaciclib or IACS-010759 each significantly slowed tumor growth compared to vehicle control (Figure 6A). The drug combination was significantly more effective than single-agent treatments, and only the combination provided tumor stasis (i.e., no significant change in volume over the course of the study; Supplemental Figure 7); however, the drugs did not display significant synergy (p=0.222). Immunoblot analysis of tumor tissues harvested after 4 wk of drug treatment confirmed downregulation of phospho-Rb and upregulation of mitochondrial OXPHOS proteins (Figure 6B) consistent with in vitro observations (Figure 3A); a caveat of this analysis time point is that surviving tumor cells may have adapted to drug treatments. The treatment-induced upregulation of OXPHOS proteins may be related to an upregulation of AMPK activity in response to an energy deficit (29), which in turn can drive the expression of genes encoding proteins to restore energy balance in a cell. Significant changes in tumor cell proliferation (Ki67) and apoptosis (cleaved caspase-3) were not detected. Since OXPHOS dysfunction can decrease HIF1-alpha levels (30), we assessed HIF1-alpha levels in tumors, but no significant changes were detected (Supplemental Figure 8). Since mitochondrial complex I is critical for aspartate synthesis (31), we measured aspartic acid levels in tumor lysates. Abemaciclib treatment significantly increased tumor aspartic acid levels that were suppressed by IACS (Figure 6C), demonstrating that IACS engaged complex I.

Figure 6: Combined abemaciclib and IACS-010759 treatment provides tumor stasis.

(A) NSG female mice bearing orthotopic HCI-003 tumors were randomized to treatments as indicated. Data are shown as mean + SD of tumor volume relative to baseline. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001 by mixed modeling. (B) Tumor tissues collected at endpoint were used for immunoblot analysis. Vertical black lines indicate that lanes were from the same blot (rearranged here for visualization). (C) Tumor tissues collected at endpoint were used for metabolite extraction and measurement of aspartic acid levels. *p≤0.05, **p≤0.01 by Bonferroni-adjusted posthoc test compared to vehicle unless otherwise indicated with brackets.

Discussion

Our findings suggest that therapeutic targeting of metabolic hubs such as mitochondrial complex I can overcome tolerance to CDK4/6i in ER+ breast cancer persister cells with enhanced metabolic pathways such as respiration. Analysis of endocrine-tolerant persister cell cycling phenotypes revealed a subpopulation of fast-cycling persisters with transcriptional enrichment for metabolic adaptations. We posit that persisters as a whole can survive treatment, and fast-cycling persisters are a subset of the persister population able to transiently bypass the growth-arresting nature of drug treatment and continue to proliferate. We recently showed that the persister state is reversible with a consequent reduction of mitochondrial content and recovered expression of glycolytic proteins (7). Such fast-cycling persisters remained sensitive to the growth-inhibitory effects of CDK4/6i, in support of the addition of abemaciclib or ribociclib to an adjuvant endocrine therapy backbone in patients at high risk of recurrence. CDK4/6i induced and potentiated mitochondrial phenotypes akin to those induced by endocrine therapy (7), including increased mitochondrial content, mitochondrial membrane potential (which may be related to mitochondrial content), and respiratory capacity. Such responses to CDK4/6i were dependent upon Rb function, affirming the centrality of CDK4/6-mediated Rb inactivation to CDK4/6i efficacy and its metabolic effects. Inhibition of CDK4/6 and complex I elicited therapeutic effects in both endocrine-sensitive/tolerant and endocrine-resistant models, offering applications for dual targeting of these pathways in both disease settings.

Oren et al. created the Watermelon reporter system that enables tracking of relative numbers of cell divisions. In line with their observations in PC9 lung cancer cells that persisted during treatment with the EGFR inhibitor osimertinib (16), we found that endocrine-tolerant ER+ breast cancer persisters exhibited heterogeneity in proliferation (Figure 1A). We showed that fast-cycling endocrine-tolerant persisters remained sensitive to CDK4/6i (Supplemental Figure 1). These findings align with our report that endocrine-resistant ER+ breast cancer cells remain dependent upon CDK4/6-E2F signaling (6), and support the clinical use of CDK4/6i to combat endocrine-resistant and endocrine-tolerant cycling persisters.

We observed that CDK4/6i-tolerant persisters showed upregulation of mitochondria and oxidative stress. Herrera-Abreu et al. also recently reported that CDK4/6i with palbociclib increases oxidative stress in MCF-7 cells (32). Oren et al. found increased ROS in osimertinib-tolerant PC9 persisters (16). Such metabolic responses have been observed in other cancer models and with other anti-cancer drug treatments (7–12), suggesting that metabolic reprogramming to enable a persister phenotype is a common theme across cancers that may also offer a common therapeutic vulnerability. Mitochondrial content can be upregulated due to increased biogenesis or decreased mitophagy; although the former has been shown to drive drug resistance, the latter can context-dependently promote drug resistance or sensitivity. Mitochondrial function can be affected by fusion, fission, trafficking, signaling, and structural changes, all of which have been implicated in drug resistance in cancer cells (33,34). In the case of ER+ breast cancer, the combination of a persister-inducing agent (such as a CDK4/6i) and a mitochondrial inhibitor (such as IACS-010759) was significantly more effective than either agent alone (Figures 5C–D and 6A, and Supplemental Figure 5B). These collective observations warrant the pursuit of treatment strategies addressing metabolic vulnerabilities that may find broad applicability across cancer types.