Sustained Oncogenic Signaling in the Cytostatic State Enables Targeting of Nonproliferating Persistent Cancer Cells

Molecularly targeted therapy has emerged as a standard treatment approach for many cancer types including breast and lung cancers. Targeted agents that interfere with aberrant growth signaling and cell-cycle controls (1, 2) in cancer cells are extensively used and have shown profound efficacy in managing cancers as stable diseases (3). Unlike traditional chemotherapies that are predominately cytotoxic, many targeted therapeutic agents possess cytostatic properties that suppress cancer cell growth without directly inducing cell death. This effect confers a cytostatic treatment condition in which subpopulations of the cancer cells are suppressed and sustained in a largely proliferative-dormant (slowly- or nonproliferative) or quiescent state but remain viable and can persist under continual therapy (4). These treatment-mediated cytostatic cells are distinct from resistant cells that continue to grow even under drug treatments (4). Importantly, such persistent treatment-mediated cytostatic cancer cells require continual maintenance and constitute a reservoir from which drug resistance clones and recurrent growth can emerge, posing a fundamental challenge for archiving durable remission (3–5).

While cytostatic response of cancer cells is favorable and a goal for many targeted therapies, depleting the persistent cytostatic cancer cells could have important implications in improving therapeutic outcome. However, current management approaches remain mostly maintenance and monitoring as targeting cancer cells in a cytostatic state has been elusive. The nonproliferating nature renders cytostatic cancer cells significantly less sensitive to standard anticancer treatments, which often directly target proliferation mechanisms or are optimized for rapidly growing cells (6–8). Moreover, although the therapeutic mechanisms of growth suppression are well established, the properties of cancer cells while under cytostatic drug response are poorly understood. Whether specific mechanisms support the maintenance of cancer cells under continuous cytostatic treatments and the potentials of their exploitation for targeting are unclear.

Cancer genomics studies have begun to unveil genetic alterations that are associated with cancer recurrence and relapse (9–15). The PI3K–AKT pathway mediates many fundamental cellular processes in cell growth, metabolism, and biosynthesis and is one of the most frequently altered pathways in cancers. Recent studies have also implicated the PI3K–AKT pathway in cancer recurrence and relapse after prolonged remission (10, 12, 16–19). Persistent cancer cells harboring such genetic alterations are implicated to predispose to recurrent growth, making their quantitative depletion a therapeutic priority. However, prospective for targeting genetically predisposed persistent cancer cells while remaining in a cytostatic condition is largely obscure. Most oncogenic alterations associated with recurrent growth have supportive roles for cancer growth and progression. Their effects or roles under nonproliferating or cytostatic conditions are less clear.

In this study, we leveraged targeted therapy paradigms and engineered cytostatic states to explore therapeutic opportunity and mechanisms for depleting persistent cancer cells attributed to cytostatic treatment conditions. We identified an increased proteasome dependency driven by aberrant AKT signaling in cytostatic cell states that is therapeutically exploitable to deplete nonproliferating persistent cancer cells with PI3K pathway mutations. Mechanistically, this proteasome vulnerability is driven by sustained oncogenic AKT signaling under cytostatic conditions, which deregulates the homeostatic controls and drives an oxidative and proteotoxic environment that imposes an increased dependency on proteasome to maintain cell viability. Exploiting this proteasome vulnerability showed efficacy in depleting persistent treatment-mediated cytostatic cancer cells independent of breast cancer subtypes, epithelial tissue origin, and cytostatic agents. Moreover, transiently targeting treatment-mediated cytostatic cancer cells is sufficient to reduce tumor regrowth in spheroid and mouse models. Our work highlights that sustained oncogenic signaling in nonproliferating states as a driver of vulnerability that is therapeutically exploitable for targeting treatment-mediated cytostatic conditions in cancer therapies, which could offer new opportunities for depleting genetically predisposed persistent cancer populations and reduce cancer recurrence.

MCF10A (RRID:CVCL_0598), T47D (RRID:CVCL_0553), NCI-H1975 (RRID:CVCL_1511), SKOV-3 (RRID:CVCL_0532), and HEK293T (RRID:CVCL_0063) cells were obtained from ATCC. ZR-75–1 (RRID:CVCL_0588), HCC38 (RRID:CVCL_1267), HCC1954 (RRID:CVCL_1259), and SUM225 (RRID:CVCL_5593) cells were gifts from Dr. Deepali Sachdev, Dr. Douglas Yee, Dr. Julie Ostrander (University of Minnesota Medical School, Minneapolis, MN), and Dr. Fariba Behbod (University of Kansas Medical Center, Kansas City, KN). All cells were used within 15 passages from frozen stocks that have been authenticated by short tandem repeat (ATCC) and Mycoplasma tested by qPCR. Cells obtained directly from ATCC were expanded and frozen down as stocks within three passages and were not reauthenticated or tested for Mycoplasma separately. Detailed culture media conditions for each cell lines are described in the Supplementary Methods.

Three-dimensional cultures for MCF10A were performed similarly as previously reported (20). Detailed culture conditions are described in Supplementary Methods. Doxycycline induction (1 μg/mL) were performed on Day 16 unless indicated otherwise. Drug treatments were performed on Day 18. p70S6K inhibitors were added on the day of doxycycline induction and refreshed on Day 18. For tumor spheroid cultures, the cancer cells were trypsinized and seeded on Matrigel (GFR, growth factor reduced, Corning) bed as single-cells at 5,000 cells per well in the corresponding cell culture media with 5% FBS and 2% Matrigel (Day 0). Cytostatic drugs were added 1 day after seeding (Day 1) and refreshed with media daily for additional 5 days (Day 2–Day 6) to establish the cytostatic condition. Treatments with inhibitors were performed on Day 7 along with the cytostatic drug for 24 hours. Experiments with AKT and p70S6K inhibitors were pretreated from Day 5 unless indicated otherwise. Immunofluorescence staining for three-dimensional cultures were performed similarly as previously reported (21). Detailed protocols are described in Supplementary Methods.

pBABE-puro (RRID:Addgene_1764) and pBABE-hygro (RRID:Addgene_1765) were obtained from Addgene. The lentivector pLT-iGSP (IRES-GFP-SV40-puro) was generated by subcloning a SV40-puro cassette from the pBABE-puro plasmid and inserted downstream of the IRES-GFP cassette in the pLT-iG (IRES-GFP) lentivector. pLT-iG lentivector containing the tetracycline response element from pTre-Tight (Clontech #631059), multiple cloning sites, and a downstream IRES-GFP cassette from pIRES2-EGFP (enhanced GFP; Clontech #6029–1) were described previously (20). pLT-myrAKT1-iGSP was generated by subcloning the myrAKT1 fragment from pLNCX myrAKT1 (RRID:Addgene_17245) into the pLT-iGSP vector. pBABE-puro-H2B-GFP (Histone H2B fused to 5′ of EGFP in pBABE-puro) and pBABE-hygro-rtTA (reverse tetracycline-controlled transactivator in pBABE-hygro) were previously reported (20).

Viruses were packaged with psPAX2 (RRID:Addgene_12260) and pMD2.G (RRID:Addgene_12259; used to package pLT lentiviral vectors) or pCL-Ampho (Imagenex; used to package pBABE retroviral vectors) with TurboFect (Thermo Fisher Scientific). Viruses were collected on Day 3 and 4 following transfection, filtered through 0.45-μm filter, and stored at −80°C. Stable cell lines were generated by viral vector transduction and selection with 2 μg/mL puromycin.

Detailed protocols and instrument settings can be found in Supplementary Methods. Cells were trypsinized to single cells and seeded at 5,000 per wells in 0.5 mL of assay media supplemented with 4% Matrigel in 24-well ultralow attachment plates. One milliliter of assay medium supplemented with 1% Matrigel was added to each well 1 day after seeding. Half of the media were replaced every 4 days with fresh media supplemented with 1% Matrigel (no Matrigel on Day 12 and Day 16). On Day 16, cells were induced with 1 μg/mL doxycycline for 48 hours. Cells were collected after solubilizing the Matrigel in Cell Recovery Solution (Corning), rinsed with PBS, lysed and sonicated in extraction buffer [7 mol/L urea, 2 mol/L thiourea, 0.4 mol/L Tris pH 7.5, 20% acetonitrile, and 4 mmol/L tris(2-carboxyethyl)phosphine (TCEP), 5 mmol/L EDTA, 10 mmol/L sodium fluoride, 1 mmol/L sodium orthovanadate]. In-solution proteolytic digestion and TMT10plex isobaric labeling were performed, and samples were fractionated before running on the Thermo Fisher Orbitrap Fusion. Scaffold Q+ version 4.7 (Proteome Software Inc., RRID:SCR_014345) was used for TMT-based protein relative quantification for proteins detected in the database search. Setting the control MCF10A cell samples as a reference for the AKT^hyper cell samples, differentially expressed proteins were determined by Permutation Test with an unadjusted significance level P < 0.05 corrected by Benjamini–Hochberg. Canonical pathway analysis was performed using the Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems, RRID:SCR_008653).

Glucose-Glo and Lactate-Glo (Promega) were used. Control and AKT^hyper cells were induced with 1 μg/mL doxycycline on Day 16 for 48 hours. Media were refreshed on Day 18 with doxycycline and collected after 24 hours for analysis. The media were diluted with PBS for the glucose (400×) and lactate (100×) assays, and analyses were performed according to the vendor's protocols. Luminescence was measured using Tecan Infinite M1000 Pro Microplate Reader. Statistical significance was determined by two-tailed Student t tests.

Three-dimensional cell cultures were set up on coverglass-bottom 8-well chamber slides (Lab-Tek). Forty-eight hours after doxycycline induction, media were replaced with fresh media containing 5 μmol/L CellROX Deep Red Reagent (Invitrogen) and 5 μg/mL Hoechst 33342 and incubated in the cell culture incubator for 30 minutes. Cells were rinsed twice with PBS, fixed with 4% PFA at room temperature for 20 minutes, and replaced with PBS before analyzing using a Nikon A1Rsi Confocal with SIM Super Resolution microscope with a 20X objective or a Nikon A1R confocal microscope with a 25X objective.

Click-iT Plus OPP Alexa Fluor 594 Protein Synthesis Assay Kit (Invitrogen) was used. Forty-eight hours after doxycycline induction, media were replaced with fresh media containing 20 μmol/L Click-iT OPP (O-propargyl-puromycin) reagent and incubated in the cell culture incubator for 30 minutes. Cells were rinsed once with PBS, fixed with 4% PFA, and processed following the immunofluorescence staining protocol above. After TritonX-100, glycine and immunofluorescence buffer (0.1% BSA, 0.2% Triton X-100, 0.04% Tween 20, 7.7 mmol/L NaN₃ in PBS) washes, Click-iT reaction was performed according to the vendor's protocols. Samples were counterstained with 1X HCS NuclearMask Blue Stain and mounted with Prolong Diamond Antifade reagent. Images were acquired using a Nikon A1RMP confocal microscope with a 25X CFI Apo LWD Objective.

RNA was extracted with TRIzol Reagent (Invitrogen), and purified with RNeasy (Qiagen) according to the vendor's protocols. Total RNA was treated with DNase I and processed for cDNA preparation (SuperScript III First-Strand Synthesis System, Invitrogen). qPCR was performed using SYBR Green reagents and StepOnePlus Real-Time PCR System (Thermo Fisher). qPCR was performed with three individual biological samples, each averaged from technical duplicate runs. The qPCR primers are listed in Supplementary Methods. ΔΔC_t was calculated by normalizing to RPLR0, and means and SDs were determined. Statistically significant differences were determined by two-tailed Student t test.

Click-iT Plus TUNEL Alexa 594 In Situ Apoptosis Detection Kit (Invitrogen) was used. Twenty-four hours after bortezomib treatment, cells were rinsed once with PBS, fixed with 4% PFA, and processed following the three-dimensional immunofluorescence staining protocol. After TritonX-100, glycine and IF buffer washes, cells were washed twice in deionized water and incubated with the TdT Reaction buffer for 10 minutes at room temperature. Samples were then incubated with the TdT Reaction mixture overnight in a humidified chamber at room temperature. Cells were washed in IF buffer three times for 20 minutes each, once with PBS and processed with the Click-iT Plus TUNEL reaction cocktails for 30 minutes at room temperature in the dark. Samples were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) and mounted with Prolong Diamond Antifade reagent. Samples were analyzed using a Leica DMI 3000 fluorescent microscope with a 20X objective. Images were acquired using a Nikon A1RMP confocal microscope with a 25X CFI Apo LWD Objective. Statistical significance was determined by two-way ANOVA.

Click-iT Plus EdU Alexa Fluor 555 Imaging Kit (Invitrogen) was used. Six days after vehicle (water or DMSO) or a cytostatic drug (palbociclib or lapatinib) treatment, EdU reagent was added together with vehicle or the cytostatic drug for 24 hours. Cells were then fixed with 4% PFA and processed following the three-dimensional immunofluorescence staining protocol. After TritonX-100, glycine and IF buffer washes, Click-iT reaction was performed according to vendor's protocol. Samples were counterstained with DAPI, mounted with Prolong Diamond Antifade reagent, and analyzed using Leica DMI 3000 fluorescent microscope with 20X objective. Images were acquired using a Nikon A1RMP confocal microscope with a 20X objective. Statistical significance was determined by two-tailed Student t tests.

Cells were lysed in RIPA buffer with Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific) on ice. Lysates were homogenized by passing through a 27-gauge needle, centrifuged at 14,000 × g for 10 minutes at 4°C, and supernatants were collected. Protein concentrations were determined using the bicinchoninic acid assay. Samples were run on denaturing conditions with 4% to 20% Tris-Glycine gels and transferred to 0.45-μm polyvinylidene difluoride membranes. Primary antibodies diluted in blocking buffer were incubated overnight at 4°C. Horseradish peroxidase (HRP)–conjugated secondary antibodies were incubated for 1 hour at room temperature. Membranes were developed using Luminata Crescendo Western HRP substrate (EMD Millipore) and imaged with the myECL Imager (Thermo Fisher Scientific).

Chloroquine (20 μmol/L), buthionine sulfoximine (BSO, 50 μmol/L), arsenic trioxide (As₂O₃, 1 μmol/L), N-acetyl-L-cysteine (NAC, 2 mmol/L), palbociclib (1 μmol/L), and DAPI (1 μg/mL) were from Sigma-Aldrich. Lapatinib (1 μmol/L), bortezomib (50 nmol/L), PF-4708671 (10 μmol/L), LY2584702 (1 μmol/L), and MK-2206 (1 μmol/L) were from Selleckchem. Trolox (50 μmol/L, EMD Millipore), MG132 (1 μmol/L, Tocris), carfilzomib (50 nmol/L, APExBIO), selonsertib (10 μmol/L), and staurosporine (10 μmol/L) were from Adooq. Doxycycline (1 μg/mL) was from Research Products International. Hoechst 33342 (5 μg/mL) was from Thermo Fisher Scientific.

Phospho-AKT (Ser473; RRID:AB_2315049), cleaved caspase-3 (RRID:AB_2341188), phospho-ribosomal protein S6 (Ser235/236) (RRID:AB_916156), ribosomal protein S6 (RRID:AB_2238583), Ki67 (RRID:AB_2797703; for mouse tissue immunofluorescence), HRP-linked anti-rabbit IgG (RRID:AB_2099233), and HRP-linked anti-mouse IgG (RRID:AB_330924) were from Cell Signaling Technology. β-Actin (RRID:AB_476692) was from Sigma-Aldrich. Ki67 (RRID:AB_443209) was from Abcam. Goat anti-rabbit Alexa Fluor 488 (RRID:AB_2633280) and goat anti-rabbit Alexa Fluor 546 (RRID:AB_143051) were from Invitrogen. DyLight 549 Streptavidin (RRID:AB_2336408) was from Vector Laboratories.

T47D cells expressing H2B-GFP were resuspended in phenol red–free assay media supplemented with 2% Matrigel and seeded at 5,000 per well on Matrigel bed preformed on coverglass-bottom 8-well chamber slides. One day after seeding, media with palbociclib were added to a final concentration of 1 μmol/L. Afterward, the media were refreshed daily with palbociclib for 5 days. On Day 7, cells were treated with either vehicle or 50 nmol/L bortezomib together with palbociclib. Cells were imaged using a Leica SPE inverted confocal microscope with a 20X objective. Image stacks were taken every 20 minutes with 2-μm step-size for 25 hours under an environmental-controlled chamber (Humidified, 5% CO₂ and 37°C; Tokai Hit). Three-dimensional reconstruction and time-lapse videos were generated using Imaris software (RRID:SCR_007370) and ImageJ (RRID:SCR_003070).

Three thousand cells per well were seeded as in the three-dimensional cell culture protocol above. On Day 8, 24 hours after a proteasome inhibitor treatment, cells were rinsed once with PBS, and replaced with fresh cell culture media (10% FBS). The media were refreshed on Day 9 and then every 3 days. On Day 16, cells were counterstained with Hoechst 33342 and fixed with PFA or stayed in fresh cell culture media. Images were taken using a Leica DMI 3000 fluorescent microscope with a 10X objective. For each condition, images were taken and quantified from four biological replicates. The total number of nuclei of each spheroid was counted. On Day 16 (end of the experiments), spheroids with a greater number of nuclei than that of the average plus SD in the cytostatic state (Day 7) were scored as regrowth. Statistical significance was determined by two-tailed Student t tests.

Six- to seven-week-old female NU/J mice (RRID:IMSR_JAX:002019) were used. One million T47D or NCI-H1975 cells in 50% Matrigel (1:1 PBS-Matrigel) were injected subcutaneously on both sides of the flank. For the T47D model, animals were implanted with a 17β-estradiol pellet (0.36 mg/pellet, 60 days release; Innovative Research of America) subcutaneously before xenograft. For growth-arrest analysis under cytostatic conditions, mice were treated with vehicle (water) or 100 mg/kg palbociclib daily via oral gavage from one day after tumor cell injection for 6 days. Tumors were then harvested, fixed, and paraffin-embedded. For cell death analysis of cytostatic tumors after proteasome inhibition, mice were treated with 0.8 mg/kg bortezomib or vehicle (0.42% DMSO) intratumorally (22) after 6 days of daily palbociclib administration. Tumors were harvested 24 hours after bortezomib treatments, fixed, and paraffin-embedded. For regrowth analysis of cytostatic tumors after treatment discontinuation, mice were treated with bortezomib every 3 days for four times. Mice were treated with 100 mg/kg palbociclib daily for 2 additional days after the last dose of bortezomib. Tumor sizes were measured every 3 to 4 days with a caliper from the first day of the initial Bortezomib treatment dose until 25 days after discontinuation of all treatments (75% of the 17β-estradiol pellet release capacity for the T47D models) to monitor regrowth. Tumor volume was calculated by ellipsoid volume formula (π/6 × length × width²). Relative tumor size was calculated by normalizing each tumor to its initial volume before bortezomib or vehicle treatment on Day 7. Statistical significance was determined by Student t tests. Animal procedures were approved by the Institutional Animal Care and Use Committee of the University of Minnesota Medical School.

Tissue sections (7 μm thick) were deparaffinized and processed for antigen retrieval by heating in Antigen Unmasking Solution (Vector Laboratories) for 30 minutes. Tissue sections were stained using Mouse on Mouse Basic Kit (Vector Laboratories) according to the vendor's protocol. Primary antibodies against Ki67 and cleaved caspase-3 were used. Alexa Fluor Plus 488 and DyLight 549 Streptavidin secondary antibodies were used. Slides were counterstained with DAPI and mounted with Prolong Diamond Antifade reagent. Images were taken using Leica DMI 3000 fluorescent microscope with a 20X objective. Four to six different tumors for each condition were analyzed using CellProfiler 3.1.8 (RRID:SCR_007358). Statistical significance was determined by two-tailed Student t tests.

Statistical significance was determined by two-tailed Student t tests or two-way ANOVA followed by Sidak multiple comparisons test when interaction is significant (P < 0.05), using GraphPad Prism version 8.3.0 (RRID:SCR_002798). Specific statistical details of experiments can be found in the corresponding Figure Legends.

The data generated in this study are available within the article and its Supplementary Data files.

We initiated our study with a targeted therapy paradigm using palbociclib, a clinical-grade cyclin-dependent kinase 4/6 (CDK4/6) inhibitor with predominately cytostatic effects that is currently used for treating breast cancers and is under investigation for other epithelial cancers (1, 2, 23, 24), and T47D cells, a widely used breast cancer cell model that is highly sensitive to palbociclib based on the Genomics of Drug Sensitivity in Cancer (RRID:SCR_011956) database (25). To recapitulate a cytostatic treatment condition, we grew T47D into tumor spheroids on reconstituted basement membrane materials and administrated palbociclib daily. We chose a palbociclib dose (1 μmol/L) that induces growth arrest in majority of the cell population without significant toxicity (26, 27). Palbociclib treatment of T47D spheroids for 6 days significantly suppressed the cancer cell growth (Fig. 1A–C). EdU labeling showed that cell proliferation was significantly reduced (Fig. 1B and C). Immunofluorescent staining for Ki-67 further indicated that most of the persistent cells entered quiescence (Ki67-negative; Fig. 1B and C), a reversible growth-arrest state characterized by prolonged cell-cycle exit (28). These data indicated that continual palbociclib treatment largely suppresses and sustains T47D spheroids in a cytostatic state.

Growth inhibition and transition from active cell cycle to quiescence are often associated with downregulation of AKT signaling (29, 30). Interestingly, we found that aberrant AKT signaling was maintained in the cytostatic T47D cells despite prolonged cytostatic treatment (Fig. 1D). Notably, inhibiting AKT did not induce cell death, indicating that the sustained AKT signaling was dispensable for cell viability (Fig. 1E and F). Interestingly, even a high dose of the apoptosis inducer staurosporine (10 μmol/L; ref. 31) only yielded a subtle (1.8 ± 0.4%) apoptotic response (Fig. 1E and F), further underscoring the challenge of targeting persistent cancer cells under cytostatic treatment conditions. While nonessential, the predominant roles of AKT in diverse cellular processes and its sustained aberrant activation despite under cytostatic condition prompted us for further investigation.

PI3K pathway mutations have been associated with cancer recurrences and relapses (10, 12, 16–19). Persistent treatment-mediated cytostatic cancer cells that harbor PI3K pathway mutations thus represent a genetically predisposed populations with clinical importance. T47D cells harbor a PIK3CA mutation that aberrantly activates AKT signaling. We speculate that PI3K-pathway mutations cause sustained AKT signaling despite under cytostatic conditions. We next examined other epithelial cancer cells that harbor different PI3K–AKT pathway mutations and aberrant AKT activation (HCC38 and ZR-75–1, breast; NCI-H1975, lung; SKOV-3, ovarian). These cancer cell lines also sustained aberrant AKT signaling under palbociclib-mediated cytostatic conditions (Fig. 1G and H; Supplementary Fig. S1A–S1E). We also examined another cytostatic treatment with a different growth suppression mechanism. Lapatinib, a dual EGFR and HER2 inhibitor, arrests cancer cells by blocking growth factor signaling and exerts both cytotoxic and cytostatic effects (32). Treatment with a cytostatic dose of lapatinib that causes less than 5% cell death significantly induced cytostatic responses in two HER2-positive breast cancer cell lines with PI3K–AKT pathway mutation (HCC1954, PIK3CA^H1047R mutant; SUM225, PTEN-loss; Supplementary Fig. S2A–S2C). Sustained aberrant AKT signaling was also observed under the lapatinib-mediated cytostatic condition in HCC1954 cells (Fig. 1I). These data suggested that sustained AKT signaling under cytostatic treatments is common in cancer cells harboring PI3K–AKT pathway mutations, independent of cancer types and treatment agents.

Given the key roles of AKT in many fundamental cellular processes and intrigued by the sustained AKT signaling in diverse cytostatic treatment conditions (Fig. 1D and G–I), we undertook a bottom-up approach to investigate the effects of sustained aberrant AKT activation on the general cytostatic state. We leveraged an organotypic model of general cytostatic state based on the non-transformed human mammary epithelial cells, MCF10A, that undergo morphogenesis and develop into quiescent acinar structures (21). We previously adapted this platform to allow inducible activation of oncogenic signal in the acinar cells after establishing the quiescent condition (20). MCF10A cells engineered with a doxycycline-inducible expression cassette consisting of a myristoylated AKT1 (myrAKT1, a constitutively active AKT1 variant) and an Internal Ribosomal Entry Site (IRES)-driven GFP reporter (Fig. 2A) were first cultured without doxycycline for 16 days on reconstituted basement membrane materials to form growth-arrested, quiescent organotypic acini (Fig. 2B–D). These quiescent acini were then induced with doxycycline to express myrAKT1 (Fig. 2A and B). The mutant acinar cells showed AKT hyperactivation (AKT^hyper), indicated by the elevated levels of membrane-associated phospho-AKT (pAKT; Fig. 2E). The AKT^hyper acinar cells also showed an increase in glucose metabolism compared with the GFP controls (Fig. 2F). Importantly, these acinar cells remained quiescent (Fig. 2G and H), recapitulating a cytostatic state with sustained aberrant AKT signaling.

To gain insights into the effects of aberrant AKT signaling on cytostatic state properties, we performed quantitative mass spectrometry and compared the signaling controls between AKT^hyper and normal quiescent acinar cells by IPA. Seven of the ten most significantly altered canonical pathways are related to protein and redox homeostasis (Fig. 2I; Supplementary Tables S1 and S2). The Protein Ubiquitination Pathway was dysregulated and was the most significantly altered pathway in the analysis (Fig. 2I), implicating a defective control in protein turnover. Two pathways that mediate protein translation, Regulation of eIF4 and p70S6K Signaling and eIF2 Signaling Pathways, were upregulated, indicating an increase in protein synthesis. The analysis also showed a dysregulation in the Mitochondrial Dysfunction Pathway (Fig. 2I) and an upregulation in the NRF2-mediated Oxidative Stress Response (Fig. 2I), suggesting an imbalance in redox control. Increased levels of protein synthesis and reactive oxygen species in the AKT^hyper quiescent acini were validated using CellROX and OPP assays respectively (Fig. 2J and K). The dysregulated protein and redox homeostasis prompted us to further examine protein stress response in the cells. We examined the levels of ATF, CHOP, and spliced XBP1 mRNA to probe for the major signaling of unfolded protein response (UPR) and endoplasmic reticulum stress (33). While no change in ATF4 and CHOP expression were detected, qPCR analysis showed an increased level of XBP1 mRNA splicing (Fig. 2L), suggesting a sign of UPR and endoplasmic reticulum stress (34). Together, these data suggested that sustained aberrant AKT signaling dysregulates the cytostatic state homeostasis and promotes an oxidative and proteotoxic environment.

Elevated proteotoxic stress in the AKT^hyper quiescent cells led us to focus on the protein homeostasis pathways and their roles in the maintenance of the cytostatic mutant cells. We investigated the effects of perturbing the two major protein turnover pathways, the ubiquitin-proteasome system and autophagy. Inhibiting proteasome or autophagy by MG132 or chloroquine, respectively, induced apoptosis in the AKT^hyper quiescent populations compared with vehicle controls. However, only inhibiting proteasome, but not autophagy, showed selectivity over the normal quiescent cells (Fig. 3A). These results were further confirmed with different apoptotic assays (Supplementary Fig. S3A and S3B) and two clinical-grade proteasome inhibitors, bortezomib and carfilzomib (35), that have higher specificity than MG132 (Fig. 3B and C). Notably, these proteasome inhibitors exhibited both lower toxicity in the control quiescent cells and increased selectivity to the AKT^hyper quiescent populations (Fig. 3A–C). These results revealed a proteasome vulnerability of the quiescent cells with aberrant AKT activation, which enables selective targeting of the mutant from normal populations under cytostatic conditions.

We used bortezomib to gain further insights into the mechanisms underlying this AKT-driven selective proteasome vulnerability. The oxidative and proteotoxic environment in the AKT^hyper quiescent cells (Fig. 2I–L) led us to hypothesize that an elevated demand for protein turnover confers the increased proteasome dependency. To test this, we reduced the demand for protein turnover by blocking the AKT-driven protein synthesis or alleviating the damage-prone oxidative environment. AKT^hyper quiescent cells showed elevated levels of ribosomal protein S6 phosphorylation, a signal that promotes protein synthesis (Fig. 3D and E). Inhibiting p70S6K, the AKT downstream effector kinase of S6, with two separate inhibitors reduced the bortezomib-induced apoptotic response consistently in the AKT^hyper, but not the normal, quiescent cells (Fig. 3F). Similar results were observed after antioxidant treatments (Trolox or NAC) using previously established conditions for MCF10A organotypic cultures (Fig. 3G; Supplementary Fig. S4; refs. 36–39), indicating that the redox environment also contributes to the proteasome vulnerability. Notably, treatment with two oxidizing agents, BSO or As₂O₃, did not promote apoptosis in the AKT^hyper or normal quiescent cells (Fig. 3H), suggesting that exacerbating the oxidative environment alone without perturbing proteasome activities is not sufficient to trigger apoptosis. Together, these data are consistent with a role for an increase of proteostatic demand in contributing to the AKT-driven proteasome vulnerability in this cytostatic mammary cell model.

We next examined this proteasome vulnerability in the treatment-mediated cytostatic cancer cell models above (Fig. 1). Bortezomib treatment induced apoptosis in two (T47D and HCC38) of the three breast cancer cell lines tested. The two responsive lines showed a significant level of apoptosis (T47D, 62.8 ± 7.1% and HCC38, 31.3 ± 8.0%; Fig. 4A and B). Importantly, these apoptotic levels were 20.5 ± 3.7 (T47D) and 10.2 ± 3.0 (HCC38) folds higher compared with the control non-transformed mammary cells (MCF10A) under the same treatment conditions, indicating a high selectivity between malignant and normal cytostatic cells (Fig. 4C). Cell death without undergoing division was further confirmed by real-time live-imaging of the palbociclib-mediated cytostatic T47D cells showing apoptotic features of DNA fragmentation and chromatin condensation upon bortezomib treatments (Fig. 4D and E; Supplementary Video 1A, B). Notably, these cell lines were derived from different breast cancer subtypes and harbor different PI3K–AKT pathway alterations that cause aberrant AKT activation (T47D, luminal A, PIK3CA^H1047R; HCC38, triple-negative, PTEN-loss).

Bortezomib treatment of palbociclib-mediated cytostatic lung and ovarian cancer cells (NCI-H1975 and SKOV-3) also induced significant levels of apoptosis (Fig. 4F). Moreover, bortezomib treatment induced apoptosis in more than 55% of the lapatinib-mediated cytostatic HCC1954 cells (Fig. 4G), while SUM225 cells that did not maintain aberrant AKT activation under lapatinib-mediated cytostatic condition (Fig. 1I) were not sensitive to bortezomib treatment (Fig. 4G).

We further explored the signaling mechanisms mediating this proteasome vulnerability. On the basis of the increased XBP1 mRNA splicing, a downstream effect of IRE1α signaling in UPR and endoplasmic reticulum stress, observed in our engineered AKT^hyper cytostatic state model (Fig. 2L), we investigated the apoptotic signaling arm of IRE1α (40). Blocking ASK1 (also known as MAP3K5) significantly attenuated the apoptotic response (Fig. 4H). Importantly, this effect was observed in different breast cancer subtypes (T47D, HCC38, and HCC1954), both breast and lung cancer cell models, and different cytostatic treatment conditions (palbociclib and lapatinib; Fig. 4H).

Our results showed a strong correlation between sustained aberrant AKT activation and proteasome vulnerability in the treatment-mediated cytostatic cancer cells. To directly examine their causal relationship, we examined the effects of inhibiting AKT signaling on the sensitivity to proteasome inhibitor treatment. Interestingly, we found that a subset of the cytostatic cancer cell models was sensitive to AKT inhibition alone (Supplementary Fig. S5A and S5B), suggesting that AKT signaling also has a supportive role for maintenance under cytostatic treatments and that targeting AKT could be another candidate approach for depleting those persistent populations. Importantly, in the subset of cytostatic cancer cell models in which sustained aberrant AKT activation was dispensable (Supplementary Fig. S5A and S5B), AKT inhibition resulted in a significant reduction in the sensitivity to bortezomib-induced apoptosis (Fig. 4I). Consistent with the AKT^hyper mutant quiescent mammary cells (Fig. 3F), inhibiting p70S6K also reduced the levels of apoptotic response in these cytostatic cancer cells (Fig. 4I), suggesting a similar mechanism underlying the proteasome vulnerability.

Altogether, these data highlight that the proteasome vulnerability driven by sustained AKT signaling in cytostatic state is conserved in malignant cells and demonstrated a broad efficacy of its exploitation in targeting treatment-mediated cytostatic conditions independent of breast cancer subtypes, epithelial tissue origins, and cytostatic agents.

We further investigated the therapeutic implications of targeting persistent treatment-mediated cytostatic cancer cells using spheroid and mouse models of cancer recurrence after cytostatic treatment discontinuation. T47D spheroids were subjected to continuous palbociclib treatment to mimic cytostatic therapy conditions and a transient bortezomib treatment to induce apoptosis in the cytostatic cancer cells (Fig. 5A), similar to the procedures above. All drugs were then washed off and replaced with normal growth media (Fig. 5A). Most spheroids treated with vehicle regrew in 8 days (Fig. 5B), indicating the cytostatic effect is reversible. Importantly, transient bortezomib treatment significantly reduced the number of spheroids that regrew (Fig. 5B and C). Furthermore, the small number of spheroids that regrew were notably smaller in size (Fig. 5B). Similar results were also observed in the triple-negative breast cancer cell line HCC38 and lung cancer cell line NCI-H1975 (Fig. 5C).

Next, we examined the effects of targeting cytostatic treatment conditions on recurrent tumor growth using a mouse xenograft model of breast tumors (Fig. 5D–I). Mice with T47D xenografts were treated daily with palbociclib for 6 days to induce and maintain the cancer cells in cytostatic conditions (Supplementary Fig. S6A–S6C). Transient treatment with bortezomib (Fig. 5D), but not vehicle, significantly induced apoptosis in the cytostatic cancer cells (Fig. 5E and F), consistent with our in vitro data (Fig. 4A and B). Bortezomib treatment (Fig. 5G) also led to tumor shrinkage while the size of the tumors remained unchanged under vehicle treatment (Fig. 5H, statistics in red). Upon discontinuation of the cytostatic drug treatment, control tumors began to grow and significantly increased in size over three weeks compared with the bortezomib-treated tumors (Fig. 5H, statistics in black). Notably, the bortezomib-treated tumors remained stable at the reduced size over the same period (Fig. 5H, statistics in red). We tested the same treatment paradigm with a lung tumor model (NCI-H1975). The NCI-H1975 tumor xenografts also showed a statistically significant tumor shrinkage after bortezomib treatment (Fig. 5I, statistics in red). Although the bortezomib-treated tumors eventually regrew after drug discontinuation, the rate and size of the regrowth were significantly smaller compared with the control tumors (Fig. 5I, statistics in black). Together, these data strongly support an efficacy of targeting the persistent cancer cells while under cytostatic treatments in delaying or reducing cancer recurrence after therapy discontinuation.

Using targeted therapy paradigms and engineered cytostatic state models, we explored cancer cell controls under treatment-mediated cytostatic conditions and identified an AKT-driven proteasome-vulnerability that enables depletion of nonproliferating persistent cancer cells harboring PTEN–PI3K pathway mutations. Importantly, targeting this proteasome vulnerability showed efficacy in diverse treatment-mediated cytostatic conditions in vitro and in vivo. We also provided data supporting that transiently targeting the persistent cancer cells during cytostatic treatments could reduce cancer recurrence in spheroid and mouse models. These findings highlight the sustained oncogenic signaling in cytostatic state as a driver of vulnerability that could provide a therapeutic opportunity for depleting genetically predisposed persistent cancer cells while under cytostatic treatment conditions to delay or prevent cancer recurrence.

Led by an initial observation of sustained aberrant AKT signaling despite prolonged treatment-mediated cytostatic conditions, we undertook a bottom-up approach to model and investigate the effects of aberrant AKT signaling on general cytostatic state and found that sustained aberrant AKT signaling confers a selective sensitivity to proteasome inhibition. Mechanistically, aberrant AKT signaling alters the cytostatic state controls and promotes an oxidative and proteotoxic environment, which imposes an increased demand for proteasome function to maintain homeostasis and cell viability. Inhibiting proteasome induces apoptosis and enables selective depletion of the mutant cytostatic populations among normal nonproliferating epithelial cells. These findings highlight that sustained oncogenic signaling in cytostatic states drives vulnerabilities that can be exploited for targeting. Importantly, this targeting strategy showed broad efficacy in cytostatic cancer cells in vitro and in vivo, independent of breast cancer subtypes, epithelial tissue origins, and cytostatic agents.

Our study focused on sustained AKT signaling in cancer cells that persist under cytostatic drug treatments. Besides AKT, other oncogenic drivers, such as Myc and Notch, have also been implicated in persisting cancer cells and cancer recurrence (10, 15, 20, 41, 42). Sustained oncogenic signaling under cytostatic treatment conditions is likely a recurring theme in molecularly targeted cancer therapies, which intervene with specific molecular targets rather than suppressing global cellular activities. Oncogenic signaling pathways that are not directly suppressed by the treatments could sustain activities under drug response. Previous studies have also shown that upregulation of oncogenic signaling, such as AKT, underlies an adaptive response in persistent cancer cell populations under targeted drug treatments (43). Importantly, our data suggest that the oncogenic drivers that have major roles in supporting cancer cell growth may not be essential while under cytostatic conditions, rendering their direct targeting not efficacious (Supplementary Fig. S5A and S5B). Rather, the sustained oncogenic signals could elicit vulnerabilities in cancer cells under cytostatic drug response, enabling selective targeting over normal cells in the nonproliferating condition. Leveraging genetic alterations that could drive oncogenic signals associated with cancer recurrence or relapses (10, 15, 20, 41, 42), our study highlights a viable strategy for depleting genetic predisposed persistent cancer cell populations while they remain responsive to cytostatic treatments before drug resistance emerges.

We undertook a bottom-up approach using an organotypic model of quiescent non-transformed mammary cells to explore the effects of sustained oncogenic signaling on general cytostatic state and identified a vulnerability driven by sustained AKT signaling that showed broad efficacy for targeting in diverse cytostatic conditions. Sustained AKT signaling under cytostatic conditions strongly predict the responsiveness to proteasome inhibition (Fig. 4F and G). Interestingly, not all cancer cell models are sensitive. The observations suggest that other genetic or epigenetic factors may modulate the response and warrant future investigation to determine the underlying mechanism and identify additional biomarkers for responsiveness. The key roles of AKT signaling in many fundamental cellular processes likely confer a predominant effect on cytostatic state controls and underlie the broad efficacy of our targeting approach. It will be of great interest for future studies to examine the efficacy of targeting other cytostatic conditions in cancer progression such as early dissemination and metastatic dormancy (44–48). We speculate that other oncogenic drivers, such as Myc, that mediate fundamental cellular processes will also have predominant effects on the properties of cytostatic states and drive common vulnerabilities that could provide a basis for targeting cytostatic populations harboring those oncogenic changes. Our bottom-up approach should provide an amenable methodology for investigating those predominant effects.

Cancer cells that remain viable under therapies contribute to persistent conditions that underlie cancer recurrence and relapses. Previous studies on persistent conditions are mostly focused on the cancer cell populations that evaded cytotoxic drug treatment (49–54), and have identified a role of altered chromatin states and elevated GPX4 level in supporting drug tolerance and survival (49, 50). Interrupting these drug-tolerant mechanisms has shown efficacy in restoring drug sensitivity and targeting such persistent populations (49, 50). Our work focused on the persistent cancer cell populations attributed to the cytostatic drug effects that might not directly exert selection pressure. We demonstrated a distinct targeting strategy by exploiting vulnerabilities elicited by sustained oncogenic signaling in cytostatic state. Many advanced cancer therapeutics exert both cytotoxic and cytostatic effects (3, 5). Our work highlights different strategies for targeting persistent populations attributed to cytotoxic and cytostatic responses that might be required together to combat persistent conditions in cancer therapies.

Cancer cell populations in treatment-mediated cytostatic conditions underscore an extended remission period with minimal expansion and evolution that should confer a venue more vulnerable to targeting, but current management approaches remain mostly maintenance and monitoring (3, 4). Our study demonstrated a viable strategy to leverage this unique venue for targeting and showed an efficacy in depleting persistent cancer cells and reducing cancer recurrence. Our work also supports future studies to identify and investigate oncogenic signaling in cytostatic conditions, which could provide insights into developing such strategies for targeting during cytostatic conditions in cancer treatments or in other steps of cancer progression.

C.T. Leung reports grants from NIH/NCI, American Cancer Society Institutional Research Grant, and grants from University of Minnesota Foundation during the conduct of the study. No disclosures were reported by the other authors.

L.M. Kim: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. P.Y. Kim: Data curation, formal analysis, validation, visualization, methodology, writing–review and editing. Y.K. Gebreyohannes: Data curation, writing–review and editing. C.T. Leung: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing.

The authors thank Drs. Kaylee Schwertfeger, Douglas Yee, Jon Coloff, and the entire Leung lab for their helpful discussions and Dr. Thomas Cattabiani at the Stevens Institute of Technology for editing the manuscript. They thank the staff at the University of Minnesota Genomics Center, the University Imaging Centers, and the Center for Mass spectrometry and Proteomics at the University of Minnesota. The work is supported by grant awards from University of Minnesota Foundation, Lisa Rosenthal Fund for Metastatic Breast Cancer Research, American Cancer Society Institutional Research Grant (#124166-IRG-58–001–55-IRG36), and NCI (#R01CA200652).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.