ACSS2 Suppresses Ferroptosis to Drive Breast Cancer Brain Metastasis

Abstract

Brain metastasis in breast cancer patients represents a terminal disease stage, with median survival typically measured in months. Tumors that colonize the brain must adapt to its unique microenvironment, such as high acetate levels. Primary brain tumor cells enhance acetate conversion to acetyl-CoA through phosphorylation of acetyl-CoA synthetase 2 (ACSS2) by cyclin-dependent kinase 5 (CDK5), a process regulated by the nutrient sensor O-GlcNAc transferase (OGT). In this study, we showed that brain-metastatic breast cancer cells exhibited elevated O-GlcNAc, OGT, and phosphorylated ACSS2 (Ser267) compared to their parental counterparts. Both OGT and CDK5 were essential for in vivo tumor growth in the brain, and ACSS2 and a phospho-mimetic S267D mutant drove progression of brain metastatic breast cancer. Mechanistically, ACSS2 supported tumor cell survival by suppressing ferroptosis through E2F1-dependent transcription of the anti-ferroptotic protein SLC7A11. Treatment with brain-penetrant ACSS2 inhibitor AD-5584 induced ferroptosis and significantly suppressed breast cancer brain metastatic growth ex vivo and in vivo. Together, these findings identify ACSS2 as a key metabolic regulator of brain-metastatic breast cancer survival and a promising target for ferroptosis-inducing therapies.

Introduction

Breast cancer brain metastases (BCBM) remain a devastating consequence of breast cancer progression, contributing to reduced overall survival post-diagnosis. Breast cancer is the second most common cancer to metastasize to the brain, with the triple-negative breast cancer (TNBC) and HER2 subtype exhibiting the highest incidence rate compared to other molecular subtypes (1). TNBC brain metastases have the lowest overall survival (OS) of 4.9 months (2), highlighting an urgent need to understand the molecular mechanisms of growth and survival in the brain microenvironment to aid in developing targeted therapies.

The brain provides a unique metabolic microenvironment. Glucose is the primary oxidative fuel for the healthy brain (3), while other metabolites such as glutamine and acetate are recognized as preferred fuels for astrocytes (4). The blood-brain barrier (BBB) limits nutrient availability to the brain, creating an environment that is scarce of metabolites, growth factors, proteins, and amino acids (5). Importantly, the cerebrospinal fluid and brain tissue contain low levels of lipids (6) and triacylglycerides (7) compared to other tissues, imposing unique metabolic constraints that necessitate cellular adaptation to this microenvironment. In situ ¹³C-glucose infusion studies revealed that although tumors in patients with glioblastoma (GBM), lung, and breast cancer brain metastases oxidize glucose in the citric acid cycle, a significant portion of the acetyl-CoA pool was derived from other metabolic sources, including acetate (8,9). Outside the gut, the brain has the highest levels of acetate (10) and acetate has been shown to support the growth of tumors in the brain (9). These observations were accompanied by evidence of upregulation of the enzyme that catalyzes the conversion of acetate to acetyl-CoA, acetyl-CoA synthetase 2 (ACSS2), in these tumors, which has been shown to play a key role in GBM (9,11,12). Acetate uptake and upregulation of ACSS2 increases the acetyl-CoA pool, which can serve as a substrate for de novo lipid biosynthesis enzymes and histone acetylation, ultimately regulating gene expression that supports growth and survival in cancer cells (13). However, whether ACSS2 regulates growth in breast cancer brain metastatic cells is not known.

The hexosamine biosynthetic pathway generates UDP-N-acetylglucosamine (UDP-GlcNAc), which utilizes major metabolites (14). UDP-GlcNAc is a substrate for N- and O-Glycosylation, regulating cellular behaviors in response to nutrient availability (15). It also acts as a substrate for O-linked ß-N-acetylglucosamine (O-GlcNAc) transferase (OGT), which catalyzes the addition of O-GlcNAc moieties onto serine and threonine residues of nuclear and cytoplasmic proteins (16). O-GlcNAcylation is a post-translational modification that alters protein functions, protein-protein interactions, and phosphorylation states (17). OGT and O-GlcNAcylation are elevated in most cancers (18), including breast cancer and glioblastoma (GBM), and targeting this modification inhibits tumor cell growth in vitro and in vivo (19,20). We have previously shown that OGT is upregulated and required for glioblastoma growth (12). Elevated expression of OGT promotes phosphorylation of ACSS2 at Ser267 (p-Ser267-ACSS2) via a non-canonical cyclin-dependent kinase 5 (CDK5). This post-translational modification leads to increased stability of the ACSS2 protein and promotes increased acetyl-CoA levels (12). Given the metabolic similarities between GBM and brain metastases (9), we hypothesized that this pathway may also be enriched in BCBM and that ACSS2 can regulate tumor growth in the brain parenchyma.

Ferroptosis is a form of non-apoptotic, iron-dependent cell death driven by excess lipid peroxides on cellular membranes, leading to membrane damage, rupture and regulated cell death (21). Accumulation of lipid peroxides and lethal reactive oxygen species (ROS) are controlled by integrated oxidant and antioxidant systems (22). Cancer cells, particularly metastasizing ones, undergo high levels of oxidative stress, thus imposing a requirement on their antioxidant system to also be elevated to maintain the redox balance. Glutathione (GSH) plays an important role in regulating ferroptosis as it can reduce toxic phospholipid hydroperoxides to nontoxic lipid alcohols, thus acting as a critical repressor of ferroptosis in cancer cells (22). The amino acid cysteine is the main rate-limiting factor in the de novo synthesis of glutathione (GSH). In the highly oxidative tumor microenvironment of cancer, de novo synthesis of cysteine represents a metabolic demand that cannot be sustained, thus leading to its requirement to be met via transmembrane nutrient transporters. In the highly oxidative extracellular space, cysteine exists mainly in its oxidized dimer form, cystine, and it is imported intracellularly through the cystine transporter solute carrier family 7 member 11 (SLC7A11; also, xCT), the functional light-chain subunit of the system x_c⁻, a cystine/glutamate antiporter. SLC7A11 exchanges cystine and glutamate across the plasma membrane, and cystine is then rapidly reduced into cysteine intracellularly and used to generate GSH (22). Inhibition of SLC7A11 by drugs like erastin can trigger ferroptosis in cancer cells (21). Whether BCBM cells are susceptible to ferroptosis is not known.

Here, we present evidence that levels of OGT, CDK5, and ACSS2 are elevated in breast cancer brain metastatic trophic cells compared to primary tumor parental breast cancer cells, and their overexpression provides an advantageous growth of primary tumor breast cancer cells in the brain. Importantly, ACSS2 and the phosphomimetic version of ACSS2 (ACSS2-S267D) are required for breast cancer growth preferentially in the brain parenchyma, contributing to cell survival by protecting against ferroptosis in BCBM cells in an E2F1-SLC7A11 dependent manner. In addition, p-Ser267-ACSS2 is highly expressed in patients with brain metastases, providing further evidence for its potential role in the adaptation of breast cancer cells to the brain environment. Importantly, we show that a novel brain-permeable ACSS2 inhibitor can induce ferroptosis and block tumor growth in BCBM cells ex vivo and in vivo.

Material and methods

Cell Culture.

Triple negative breast cancer MDA-MB-231 (RRID:CVCL_0062), 4T1 (RRID:CVCL_0125), E0771 (RRID:CVCL_GR23) cells and HEK293T (RRID:CVCL_0063) cell lines were obtained from ATCC (Manassas, VA, USA) with short tandem repeat (STR) profiling and cultured according to ATCC protocol supplemented with 10% fetal bovine serum (FBS), 5% 10,000 Units/mL Penicillin-10,000 μg/mL Streptomycin, and 5% 200 mM L-Glutamine. Triple-negative brain trophic cells MDA-MB-231-BR and 4T1-BR are a kind gift from Dr. Patricia Steeg (Center for Cancer Research, National Cancer Institute) (23) and E0771-BR are a kind gift from Qing Chen (Wistar Institute), generated as previously described (24) and were grown under the same conditions. For crystal violet staining, 5 × 10⁴ cells were plated and subjected to the treatments as described in the individual figures and then stained with 0.5% crystal violet prepared in a 1:1 methanol-water solution followed by PBS washes. Cells stably overexpressing OGT (GeneCopoeia, EX-Z3428-M13–10- OGT plasmid, RRID:Addgene_29760), CDK5 (GeneCopoeia, EX-A0348-Lv120, RRID: Addgene_20449), HA-ACSS2-WT, HA-ACSS2-S267D (GeneCopoeia, pReceiver, RRID:Addgene_228594), pWPI-SLC7A11-HA (Addgene, plasmid #201643, RRID:Addgene_201643), and HA-E2F-1-pRcCMV (Addgene plasmid # 21667, RRID:Addgene_21667) were generated through production of lentivirus. All cell lines were quarterly tested for mycoplasma (abmGood PCR Mycoplasma Detection Kit). Following thawing, cells were only used for experiments for up to ten passages.

Production of Lentivirus and Viral Transduction.

HEK-293T cells were grown to ~90% confluency and transfected. Prior to transfection, 10 μg of shRNA or overexpression plasmid DNA, 10 μg VSVG, 5 μg RSV, and 5 μg RRE were gently mixed and incubated in 1.5 mL of Optimem for 5 minutes. Concurrently, 90 μL of PEI was added dropwise to 1.5 mL of Optimem and incubated for 5 minutes. Following the 5 minute incubation, the PEI solution was added dropwise to the DNA solution and incubated for at least 30 minutes. The PEI-DNA solution was then added dropwise to the HEK-293T cells already plated with 5 mL of Optimem, and the cells were incubated overnight in the transfection media. Approximately 16–18 hours later, the transfection media was replaced with normal growth media. Viral supernatants were collected at 24 and 48 hours following the media change. These supernatants were passed through a 0.45 μm filter and portioned into 1 mL aliquots to be stored at −80°C if not for immediate use. 1 mL aliquots from each collection time point where mixed with 2mL of growth media and 1:500 8mg/mL polybrene and added to 75% confluent cell line of interest for 6 hours. Media was replaced with 10mL of fresh growth media followed by the appropriate antibiotic selection.

RNA interference.

Control shRNA was acquired from Addgene (plasmid 1864, RRID:Addgene_1864). Control-scrambled shRNA sequence used was: CCTAAGGTTAAGTCGCCCTCGCTCTAGCGAGGGCGACTTAACCTT. All other shRNA constructs were acquired from Sigma and shRNA sequence used: for OGT-1, GCCCTAAGTTTGAGTCCAAATCTCGAGATTTGGACTCAAACTTAGGGC, and for OGT-2, GCTGAGCAGTATTCCGAGAAACTCGAGTTTCTCGGAATACTGCTCAGC. Cdk5 shRNA sequence used: for Cdk5–1, CCGGGTGAACGTCGTGCCCAAACTCCTCGAGGAGTTTGGGCACGACGTTCACTTTTTTG, and for Cdk5–2, CCGGCAGAACCTTCTGAAGTGTAACCTCGAGGTTACACTTCAGAAGGTTCTGTTTTTTG.

ACSS2 shRNA sequence used: CCGGGCTTCTGTTCTGGGTCTGAATCTCGAGATTCAGACCCAGAACAGAAGCTTTTTG.

ACSS2–3’-UTR-shRNA (Sigma): GCTTCTGTTCTGGGTCTGAAT.

Human E2F1 shRNA (Sigma): TAACTGCACTTTCGGCCCTTT. Mouse E2F1 shRNA (Sigma): CCATACCCTCTGTCCCAATAG.

Animal models of cancer.

For the intracranial model, Nu/Nu athymic 4–6 week old mice from Charles River Laboratories (Wilmington, MA, USA, RRID:IMSR_CRL:490) were immobilized using the Just for Mice ^™ Stereotaxic Frame (Harvard Apparatus, Holliston, MA, USA). Cell concentration was adjusted to 20,000 cells/uL in dPBS and kept on ice. The head was cleaned with 70% ethanol pads (Fisher Healthcare) and Povidone-Iodine Swabsticks (PDI, Woodcliff Lake, NJ) and a longitudinal incision of the skin was made between the occiput and forehead. A micromotor high speed drill (BrainTree Scientific Inc, BrainTree, MA) was used to make a hole above the right frontal cortex (2 mm frontal, 1.5 mm lateral from bregma). 100,000 cells in 5 μL dPBS were injected intracranially of MDA-MB-231 or MDA-MB-231-BR cells stably expressing luciferase and pReceiver-WT-ACSS2, pReceiver-S267D-ACSS2, N-Flag-OGT, HA-CDK5 overexpression or containing either Control, ACSS2, CDK5, OGT shRNA. Mice were injected intraperitoneally with 100 μL of D-Luciferin solution (30 mg/mL; Caliper Life Sciences, Hopkinton, MA) and tumor growth was monitored via bioluminescence imaging on the IVIS 200 system 19 (Perkin Elmer) and results analyzed using Living Image software (Caliper Life Sciences, Waltham, MA, 20 USA). For the drug trial, human ACSS2 inhibitor AD-5584 was prepared in 10 mg/mL saline solution. Nu/Nu mice were weighed and injected intraperitoneally daily for 10 days with 50 mg/kg of AD-5584/saline solution. For mammary fat pad (MFP) tumors, female athymic nude Nu/Nu mice (Charles River, Wilmington, MA, USA) (4–6 weeks old) were anesthetized with 4% isoflurane and inoculated with 1.5 × 10⁶ MDA-MB-231 cells stably expressing luciferase and either pReceiver-WT-ACSS2 or pReceiver-S267D-ACSS2, containing Control or ACSS2 shRNA. Mice were injected intraperitoneally with 200 μL of D-Luciferin solution (30 mg/mL) and results analyzed using Living Image software. For intracranial injections, mice were euthanized 3 weeks after injection. Brains were dissected, fixed in 4% formalin and prepared for processing/sectioning/embedding/blocking to generate paraffin-embedded slides. For MFP injections, tumors were measured weekly along and perpendicular to the longest dimension using digital calipers (Fowler Co., Inc., Newton, MA, USA). Tumor volumes were calculated as V=(length)x(width)2 ×0.52. After 8 weeks, tumors were excised, weighed, and photographed. All protocols involving the use of animals were approved by the Institutional Animal Care and Use Committee at Drexel University College of Medicine.

Human samples.

Paraffin embedded brain metastatic carcinoma tissue microarrays (GL861a) containing 43 cases of de-identified biosamples of metastatic carcinoma from different primary tumors were purchased from Biomax. Slides were subjected to immunohistochemistry using Thomas Jefferson University Department of Pathology and Genomic Medicine and stained for O-GlcNAc and pACSS2. Each slide was given a score from 1 (low) to 3 (high) based on the observed intensity of each staining. Final pACSS2 expression was determined as a ratio of pACSS2 to its correspondent ACSS2 staining score.

Reagents.

Anti-actin (Santa Cruz Biotechnology; Dallas, TX, USA, RRID:AB_626632), anti-OGT (RRID:AB_2716710), anti-ACSS2 (RRID:AB_2222710), anti-Cdk5 (RRID:AB_2773717), (Cell Signaling Technology; Danvers, MA, USA), anti-O-GlcNAc (Sigma Aldrich; St. Louis, MO, USA, RRID:AB_1079524) Anti-E2F1, Anti-SLC7A11 (Abcam). Puromycin, Polybrene, crystal violet, Thiamet-G (Sigma Aldrich, St. Louis, MO, USA). Ferrostatin-1, AD-5584, Liproxstatin-1 (Selleckchem, Houston, TX). Rabbit polyclonal antibody recognizing phosphorylated ACSS2 S267 was generated as previous described (12). D-luciferin potassium salt (Perkin Elmer, Waltham, MA, USA). Plasmids used were pReceiver N-Flag OGT, pReceiver HA-ACSS2, HA-ACSS2-S267A, and HA-ACSS2-S267D (Genecopoeia; Rockville, MD, USA). Cdk5-HA was a gift from Sander van den Heuvel (Addgene plasmid # 1872). D-luciferin potassium salt (Perkin Elmer, Waltham, MA, USA), and crystal violet (Sigma Aldrich, St. Louis, MO, USA).

mRNA expression.

RNA was extracted using Trizol (Invitrogen) as per manufacturer’s instruction from MDA-MB-231BR cells. Taqman gene expression assay primer probes for ACSS2 (Hs01122829_m1), GAPDH (Hs02758991_g1) E2F1 (Hs00153451_m1) SLC7A11 (Hs00921938_m1) were purchased from Thermofisher Scientific. qRT-PCR was performed using qScript XLT One-Step RT-qPCR Tough Mix (QuantaBio, Beverly, MA, USA) and analysis was performed using Applied Biosystems 7500 machine. Briefly, isolated RNA concentrations were determined by measuring them on a NanoDrop, and all samples were diluted to the lowest concentration among samples (in this case 40ng/uL). A mastermix was then generated for the individual primers using 5 μL dH₂O, 10 μL Brilliant PCR master mix and 1 μL primer per sample. A total reaction volume of 20 μL was used by gently mixing 4 μL of sample RNA and 16 μL of the mastermix into the reaction tubes, carefully avoiding bubbles. The samples were then loaded into the Applied Biosystems 7500 machine and the experimental setup was as follows: acquire experiment as “Quantitation ΔΔCT”, targets and samples were assigned, the initial holding stage adjusted to 50°C for 10 minutes, followed by a secondary holding stage at 95°C for 1 minute, then 45 cycles consisting of 95°C for 15 seconds and cooling to 60°C for 1 minute were completed. Data were exported from the Applied Biosystems software and expression levels were analyzed using Data Assist v2.0 (Life Technologies, Grand Island, NY, USA).

Immunoblotting.

Immunoblotting protocols have been previously described (12). Briefly, cell lysates from 1–5 × 10⁶ cells were prepared in radioimmune precipitation assay lysis buffer (150 mM NaCl, 1% NP40, 0.5% DOC, 50 mM Tris HCL at pH 8, 0.1% SDS, 10% glycerol, 5 mM EDTA, 20 mM NaF, and 1 mM Na₃VO₄) supplemented with 1 μg/mL each of pepstatin, leupeptin, aprotinin, and 200 μg/mL PMSF. Lysates were cleared by centrifugation at 16,000 × g for 15 minutes at 4 °C and protein concentration was determined with Bradford assay. Sample separation by SDS-PAGE, transferred to PDVF membrane and visualized with chemiluminescence. Proteins were analyzed by immunoblotting using primary antibodies indicated above.

RNA-Seq.

RNA-Seq was performed as previously described by your group (25). RNA was extracted using Trizol (Invitrogen) as per manufacturer’s instruction from MDA-MB-231BR cells stably expressing shSCR or shACSS2 (3 biological replicates per condition). RNA-Seq and analysis was performed by BGI Tech.

Immunohistochemical staining.

The slides containing brain metastatic tumors were deparaffinized by Xylene and subsequent rehydration was done by decreasing concentrations of ethanol-water mixture. Antigen retrieval was done by citrate buffer immersion and steaming the slides for 45 minutes. Tissue was treated with 200–400 μL 1% BSA+5% serum PBS solution for 1hr followed by primary incubation at 4° C overnight using OGT, O-GlcNAc RL2 (Santa Cruz 1:50), p-S267 ACSS2 (1:100), ACSS2 (Cell Signaling Technology 1:100), GFAP (Life Technologies 2.2B10 1:100) antibodies and secondary antibody incubation was done for an hour at RT with respective antibodies at 1: 100 dilution. For IHC the stain was developed using the DAB kit by Vectastain (Vector Labs, Burlingame, CA, USA). Finally, slides were mounted and imaged using EVOS FL (Life Technologies). The staining patterns of tumor tissues were assessed by a board-certified pathologist. For immunofluorescence, Alexa Fluor conjugated secondary antibodies were used at 1:100 dilution in the dark, slides were mounted and imaged under a fluorescent microscope. H&E and Ki-67 staining were performed at TJU, Department of Pathology, Philadelphia, PA.

Clonogenic Survival Assay.

Cells were cultured and detached in their respective growth media to assess clonogenic survival, then counted using a hemocytometer. 1,000 cells were plated in triplicate in 6- well plates in their respective growth media and allowed to form colonies. Cells were cultured for 14 days and were stained with crystal violet. Colony numbers were quantified (1 colony >40 cells).

BODIPY Staining of Cells.

Cells were stained with 5 μM BOPIDY C11 (Invitrogen) in PBS for 25 min at 37C in the dark and washed in 1xPBS prior to imaging on EVOS FL (Life Technologies) using GFP.

Lipid Peroxidation (MDA) Assay

The concentration of MDA in cells or tissues was determined using an MDA assay kit (Abcam, ab233471). The kit provided MDA standard and all necessary buffers. The assay involved the following steps: serial dilution of MDA standards and test samples; addition of 10 μl of MDA color reagent stock solution to each well of MDA standard; incubation at room temperature for 10–30 minutes; addition of 40 μl of reaction solution and further incubation at room temperature for 30–60 minutes; measurement of the end-point absorbance at OD 695–700 nm using an absorbance plate reader with path-check correction.

Chromatin Immunoprecipitation (ChIP).

Chromatin Immunoprecipitation was conducted following the Cell Signaling Technology protocol, SimpleChIP Enzymatic Chromatin IP Kit (Magnetic Beads) #9003. MDA-MB-231BR cells treated at 70% confluence with DMSO or AD-5584 (50 μM) for 6 hours on two 10 cm dishes. Cells are then fixed, washed, collected for preparation of nuclei and the digestion of chromatin utilizing Micrococcal Nuclease. Digested cross-linked chromatin samples were then incubated with Normal Rabbit IgG (Cell Signaling, 2729P) or E2F-1 Rabbit Ab (Cell Signaling, 3742S) overnight and purified utilizing ChIP- Grade Protein G Magnetic Beads #9006. Chromatin samples were then eluted and DNA purified for quantification using Real-Time Quantitative Polymerase Chain Reaction (RT-qPCR) method. Primer sequences for GAPDH (Forward Sequence: GTGGATATTGTTGCCATCAATGACC; Reverse Sequence: GCCCCAGCCTTCATGGTGGT) and SLC7A11 promoter (Forward Sequence: CAAGCCCATGAGGAAGCTGA as previously described (26); Reverse Sequence: TTCTCCACCTCCTCGTTCCA) from Integrated DNA Technologies. Results were analyzed following a delta delta Cq method.

EdU Staining.

Cell proliferation was assessed using the Click-iT^™ EdU Imaging Kit with Alexa Fluor^™ 488 (Invitrogen) according to the manufacturer’s instructions. Briefly, MDA-MB-231 or MDA-MB-231BR cells were grown on 8-well chamber slides and treated with DMSO or AD-5584 for 24h. After treatment, cells were incubated with 10 μM EdU in culture medium at 37°C for 1 h. Following incubation, cells were washed with PBS and fixed with 4% PFA in PBS for 30 mins at RT. After fixing, cells were washed twice with 3% BSA in PBS and permeabilized with 0.5% Triton X-100 in PBS for 20 mins at RT. After additionally washes with 3% BSA in PBS, incorporated EdU was detected using the Click-iT reaction cocktail for 30 minutes in the dark. Nuclei were counterstained with Hoechst. Images were taken on an EVOS FL Auto Imaging System (Thermo Fisher Scientific, RRID:SCR_026039). The percentage of EdU-positive nuclei was quantified using ImageJ (RRID:SCR_003070).

PDxO-BCBM Organoids.

Patient-derived xenograft organoid (PDxO) line HCI-015, derived from brain metastasis of TNBC/basal breast cancer patient, was received from the Huntsman Cancer Institute Patient Derived Xenograft Repository and culturing of PDxOs was performed as previously described (27). Briefly, organoids were cultured in PDxO complete media for triple negative subtype and embedded in Growth Factor Reduced Matrigel Matrix (Corning, Glendale AZ). Mature organoids were passaged by digestion of matrix by adding Dispase II (Sigma Aldrich) and dissociated into single cells by addition of Gibco TrypLE Express (ThermoFisher Scientific, Waltham, Massachusetts, USA). Single cells were seeded in 10 μL domes on a 96 well plate at a density of 5,000 cells with 200 μL of PDxO conditioned media containing DMSO or AD-5584 (50 μM). Cells are given 6 days to form organoids. To quantify, images are taken at 20X magnification and the total cross-sectional area of 30 counted organoids of a minimum width size of 50 μm from each treatment were calculated. Data analyzed using ImageJ.

Flow Cytometry.

Cells were prepared according to manufacturer protocol (BD Pharmingen, Propidium Iodide) for cell death and BODIPY C11 (Invitrogen) for lipid peroxidation. Briefly, cells were trypsinized (0.25% Trypsin), counted, washed twice with 1XPBS, and resuspended in 100uL 1X binding buffer incubated with 5 μL Propidium Iodide (PI) staining or 5μM BODIPY solution for 15 minutes in the dark at room temperature. Following incubation, the volume was brought up to 500μL of 1X binding buffer. Tubes were then analyzed using a Guava easyCyte flow cytometer. All data were collected and analyzed using a Guava EasyCyte Plus system and CytoSoft (version 5.3) software (Millipore). Data are gated and expressed relative to the appropriate unstained and single-stained controls.

Ex vivo brain slice model.

Organotypic hippocampal cultures were prepared as described previously (28). Briefly, adult mice (6–8 week) or mice after 12 days following intracranial injection were decapitated and their brains rapidly removed into ice-cold (4°C) sucrose-aCSF composed of the following (in mM): 280 sucrose, 5 KCl, 2 MgCl₂, 1 CaCl₂, 20 glucose, 10 HEPES, 5 Na⁺-ascorbate, 3 thiourea, 2 Na⁺-pyruvate; pH=7.3. Brains were blocked with a sharp scalpel and sliced into 250 μm sections using a McIlwain-type tissue chopper (Vibrotome inc). Four to six slices were placed onto each 0.4 μm Millicell tissue culture insert (Millipore) in six-well plates, 1 ml of medium containing the following: Neurobasal medium A (Gibco), 2% Gem21-Neuroplex supplement (Gemini), 1% N2 supplement (Gibco), 1% glutamine (Invitrogen), 0.5% glucose, 10 U/mL penicillin, and 100 μg/mL streptomycin (Invitrogen), placed underneath each insert. One-third to one-half of the media was changed every 2 days. Tumor growth was monitored via bioluminescence imaging on the IVIS 200 system (Perkin Elmer, RRID:SCR_018621) and results were analyzed using Living Image software (Caliper Life Sciences, Waltham, MA, USA, RRID:SCR_014247). For MTS assay, individual brain slices were transferred to a 96-well plate and subjected to Promega CellTiter 96^® Aqueous One Solution (Cat: G3582) mixed in a 1:5 ratio with culture media and treated as previously described. Tissues were incubated at 37°C, 5% CO₂ for 4 hours and absorbance at 490nm was measured with Tecan Spark Microplate reader.

Statistics and reproducibility.

All results shown are results of at least three independent experiments and shown as averages and presented as mean ± SEM. P<0.05 was considered statistically significant. The exact value of sample size (n) is given in the figure legends or represented as single dots in the figures. Data was analyzed using GraphPad Prism 10.4.1 (GraphPad Software, RRID:SCR_002798). P-values were calculated using a Student’s two-tailed test unless otherwise stated (* represents p-value ≤ 0.05, **p-value ≤ 0.001, ***p-value≤0.001 or as marked in figure legend). Statistical analysis of growth rate of mice was performed using ANOVA. *p-value < 0.05. Survival curves were described using Kaplan-Meier plots and compared using the log-rank test. All cells and the animals were randomly allocated to experimental groups.

Data availability

The RNA-seq data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE326480. Publicly available data generated by others were used by the authors. The brain metastatic transcriptome in breast cancer data analyzed in this study was obtained from Gene Expression Omnibus public dataset GSE184869 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE184869). All other raw data generated in this study are available upon request from the corresponding author.

Results

OGT/CDK5/ACSS2 levels are elevated in breast cancer brain metastatic cells and are required for tumor growth in vivo.

Given the metabolic similarities in acetate dependency between glioblastoma and brain metastases (9), we hypothesized that our previously identified pathway in GBM (12), where OGT governs the regulation of acetate metabolism and growth via CDK5 phosphorylation of ACSS2, may also be contributing to BCBM growth. To test this hypothesis, we examined this pathway utilizing human and mammalian parental and brain trophic (BR) TNBC cells selected to preferentially metastasize to the brain (29). Initially, to test that these brain tropic cells preferentially grow and have adapted to the brain microenvironment, we intracranially injected parental MDA-MB-231 TNBC cells or the brain tropic breast cancer MDA-MB-231-BR cells enriched to metastasize to the brain in the parenchyma of immunocompromised Nu/Nu mice (30). The brain trophic cells formed large and highly invasive tumors compared to parental MDA-MB-231 cells which exhibited smaller, localized focal tumors, indicating a clear adaptation of the BR cells to the brain environment (Fig. 1A). Consistent with our hypothesis, we found that the brain trophic cells contained elevated levels of O-GlcNAc and ACSS2-S267 phosphorylation as determined by immunohistochemical (IHC) analysis compared to parental cells (Fig. 1A) Using immunoblot analysis we found significant increase in total levels of O-GlcNAc, OGT, pACSS2-S267, ACSS2 and CDK5 (Fig. 1B, Supplementary Fig. 1A) when compared to parental breast cancer cells. We saw similar results in two additional pairs of parental and brain trophic mouse TNBC cells: 4T1-BR and E0771-BR cells compared to parental cells (Fig. 1B, Supplementary Fig. 1B). Importantly, using a tissue microarray, we found that 63% and 94% of human brain cancer metastases rom 86 different primary tumors including breast, lung, thyroid and others contained medium-high levels of O-GlcNAc and ACSS2-S267 phosphorylation, respectively, as determined by immunohistochemical (IHC) analysis (Fig. 1C). These findings suggest a possible role for OGT and pS267-ACSS2 in brain metastatic growth.

Figure 1. OGT/CDK5/ACSS2 axis is elevated in TNBC BCBM.

(A) Representative images of immunohistochemical staining for H&E, O-GlcNAc and p-S267-ACSS2 on coronal sections from mice inoculated with MDA-MB-231 (top) or MDA-MB-231-BR (bottom) cells. Scale bar: 200 μM. (B) Cell lysate from parental and brain trophic MDA-MB-231, 4T1 and E0771 cells were collected for immunoblot analysis with the indicated antibodies. (C) Immunohistochemical staining for O-GlcNAc and p-S267-ACSS2 on a brain metastatic carcinoma tissue microarray (n=86) (4X scale bar 1000 μm) and representative images of pACSS2 scoring as low, medium or high based on a 1, 2 and 3 score respectively for pACSS2 and O-GlcNAc staining. Scale bar: 200 μM.

To test the role of OGT and CDK5-ACSS2 axis in regulating BCBM macrometastatic growth in the brain, we targeted OGT and CDK5 using stable RNAi knockdown in MDA-MB-231-BR cells injected intracranially. We utilized intracranial injections of breast cancer brain-trophic cells as this is regarded as a suitable model for breast cancer brain macrometastases (31). Targeting OGT levels in MDA-MB-231-BR cells reduced total O-GlcNAc levels, ACSS2-S267 phosphorylation and total ACSS2 levels compared to control cells (Fig. 2A). Intracranial injection of BR cells stably expressing OGT RNAi significantly reduced growth in the brain compared to control cells (Fig. 2B). Targeting CDK5 with RNAi in MDA-MB-231-BR cells reduced ACSS2-S267 phosphorylation and total ACSS2 levels compared to control cells (Fig. 2C). Similar to the results with reduced OGT expression, intracranial injection of cells stably expressing CDK5 RNAi significantly reduced growth in the brain compared to control cells (Fig. 2D). To test whether OGT or CDK5 expression is sufficient to provide signals that allow poorly growing parental cells to adapt to the brain microenvironment and increase growth in the brain parenchyma, we tested the effects of stably overexpressing either OGT or CDK5 in parental MD-MBA-231 cells. OGT overexpressing cells increased O-GlcNAc, CDK5 levels, ACSS2-S267 phosphorylation and total ACSS2 levels compared to controls (Supplementary Fig. 2A) and allowed poorly growing parental breast cancer cells to now form significantly larger tumors in the brain parenchyma compared to control parental MDA-MB-231 cells (Supplementary Fig. 2B). Similar results were obtained in MDA-MB-231 cells overexpressing CDK5 (Supplementary Fig. 2C, 2D). These data suggest that OGT and CDK5-mediated phosphorylation of ACSS2 may regulate breast cancer cells metabolic adaptability and growth in the brain microenvironment.

Figure 2. OGT, CDK5, and ACSS2 are required for BCBM cell growth in vivo.

(A) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or OGT were collected for immunoblot analysis with the indicated antibodies. (B) Representative images of bioluminescent (top) detection of tumors from mice injected with MDA-MB-231-BR cells stably expressing shRNA against Scramble or OGT 14 days post-injection. Representative images of H&E analysis (bottom) on coronal sections from mice harboring cells with shRNA against Scramble or OGT tumors at Day 14. Graph of bioluminescence signal from mice injected with MDA-MB-231-BR cells stably expressing shRNA against Scramble (n=4) or OGT (n=4) (right). (C) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or CDK5 were collected for immunoblot analysis with the indicated antibodies. (D) Representative images of bioluminescent (top) detection of tumors from mice injected with Scramble or shCDK5 MDA-MB-231-BR cells 14 days post-injection. Representative images of H&E analysis (bottom) on coronal sections from mice harboring Scramble or shCDK5 tumors at Day 14. Graph of bioluminescence signal from mice injected with MDA-MB-231-BR cells stably expressing shRNA against Scramble (n=3) or CDK5 (n=3) (right). (E) Representative images of bioluminescent (top) detection of tumors from mice injected with shSCR and shACSS2 MDA-MB-231-BR cells 14 days post-injection. Representative images of H&E analysis (middle) on coronal sections from mice harboring shRNA against Scramble or ACSS2 tumors at Day 14. Data are quantified and presented as average relative bioluminescence signal from mice injected with MDA-MB-231-BR cells expressing shRNA against Scramble (n=3) or ACSS2 mice (n=3) (bottom). Student’s t-test reported as mean ± SEM;. (F) Representative images of bioluminescent (top) detection of tumors from mice injected with shSCR and shACSS2 4T1-BR cells 14 days post-injection. Representative images of H&E analysis (middle) on coronal sections from mice harboring shRNA against Scramble or ACSS2 tumors at Day 14. Data are quantified and presented as average relative bioluminescence signal from mice injected with 4T1-BR cells expressing shRNA against Scramble (n=3) or ACSS2 mice (n=3) (bottom). (G) Representative images of bioluminescent (top) detection of tumors from mice injected with shSCR and shACSS2 E0771-BR cells 14 days post-injection. Representative images of H&E analysis (middle) on coronal sections from mice harboring shRNA against Scramble or ACSS2 tumors at Day 14. Data are quantified and presented as average relative bioluminescence signal from mice injected with E0771-BR cells expressing shRNA against Scramble (n=3) or ACSS2 mice (n=3) (bottom). Student’s t-test reported as mean ± SEM; ****p<0.0001, ***p<0.001. Scale bar: 200 μM.

ACSS2 and ACSS2-pSer267 provide a specific advantage for breast cancer brain metastatic growth.

It was previously shown that ACSS2 expression is increased in brain metastases, where similarly to GBM, acetate is a major metabolite to be oxidized in patient tumors (9,32,33). Considering this evidence, we examined whether ACSS2 and ACSS2-pSer267 are critical for allowing breast cancer cells to adapt to the brain microenvironment. We implanted control and ACSS2 stable RNAi-expressing TNBC cells into the brain or mammary fat pad and assessed tumor growth. Brain tropic MDA-MB-231-BR cells stably expressing ACSS2 RNAi (Supplementary Fig. 2E) exhibited a significantly impaired tumor growth in the brain compared to control cells (Fig. 2E). A similar inhibition of tumor growth was seen in brain tropic mouse 4T1-BR (Fig. 2F) and E0771-BR (Fig. 2G) cells with stable ACSS2 knockdown (Supplementary Fig. 2F, Supplementary Fig. 2G) compared to controls and using a second shRNA targeting ACSS2 3’UTR in 4T1-BR cells (Supplementary Fig. 2H, 2I). However, targeting ACSS2 in parental breast cancer MDA-MB-231 cells had minimal effect on tumor growth compared to control when implanted in the mammary fat pad (Supplementary Fig. 2J, 2K). These data suggest a critical dependency on ACSS2 for breast cancer growth in the brain. To directly test the role of ACSS2-S267 phosphorylation on breast cancer growth in the brain, we targeted endogenous ACSS2 using a 3’UTR specific ACSS2 RNAi in parental MDA-MB-231 cells. We exogenously expressed either wildtype or ACSS2 phospho-mimetic S267D mutant (Supplementary Fig. 3A). Importantly, growth of tumors derived from ACSS2-S267D expressing cells implanted in the brain was significantly enhanced compared to wildtype ACSS2 expressing cells (Supplementary Fig. 3B). In contrast, no significant difference in tumor growth was measured between these same cells when implanted into the mammary fat pad (Supplementary Fig. 3C). These results suggest that ACSS2 and ACSS2-S267 phosphorylation play a unique and critical role in breast cancer cell growth specifically in the brain microenvironment.

We also tested whether the role of CDK5 in regulating BCBM growth required ACSS2-S267 phosphorylation. To test whether CDK5-mediated regulation of cell growth was dependent on ACSS2-S267 phosphorylation, we overexpressed ACSS2-S267D mutant in the context of CDK5 knockdown (Supplementary Fig. 3D) and noticed a partial growth restoration in MDA-MB-231-BR cells in vitro (Supplementary Fig. 3E). In addition, MDA-MB-231-BR cells expressing the phospho-mimetic mutant ACSS2-S267D were able to partly rescue the growth effects conferred by CDK5 depletion in vivo (Supplementary Fig. 3F). Thus, CDK5 regulation of BCBM growth is partly dependent on ACSS2-S267 phosphorylation.

ACSS2 regulates breast cancer brain metastatic cell survival via ferroptosis suppression.

Using an ex vivo brain slice tumor model (28), we treated brain slices containing MDA-MB-231-BR preformed tumors with a quinoxaline-based ACSS2 inhibitor VY-3–249 (32) (ACSS2i) (Fig. 3A) and found that targeting ACSS2 causes a complete regression of tumors in the brain parenchyma (Fig. 3B) compared to control treatment, suggesting that targeting ACSS2 induced cell death in BCBM cells. To exclude toxic traits of the VY-3–249 ACSS2 inhibitor in ex vivo brain slices, we examined brain slice viability in tumor-free slices. We found no difference with vehicle treated samples, indicating that the effect of VY-3–249 is specific to the cancer cells (Supplementary Fig. 4A). A four-fold increase in propidium iodide (PI) in MDA-MB-231-BR cells treated with VY-3–249 confirmed induction of cell death (Fig. 3C) in vitro. To determine the specific cell death mechanism mediated by targeting ACSS2, we tested several cell death-inducing inhibitors including autophagy (3-methyladenine: 3-MA), necroptosis (necrosulfonamide: NSA), apoptosis (caspase inhibitor: zVAD) and ferroptosis (ferrostatin-1: Fer-1). Only ferrostatin-1 (Fer-1), a specific inhibitor of ferroptosis (34), blocked VY-3–249 induced cell death (Fig. 3C). Ferroptosis is an iron-dependent cell death mechanism characterized by the accumulation of lipid-peroxides that leads to plasma membrane rupture (35). We detected a near seven-fold increase in lipid peroxides in MDA-MB-231-BR cells treated with VY-3–249, as measured by C11 BODIPY 581/591 staining (Fig. 3D). This effect was comparable to treating cells with erastin (Fig. 3D), an inhibitor of system x_c⁻ antiporter (xCT) and a well-characterized drug that triggers ferroptosis (36). We detected a similar increase in cell death (Supplementary Fig. 4B) and lipid peroxidation (Supplementary Fig. 4C) in 4T1-BR cells treated with VY-3–249, as well as in E0771-BR cells (Supplementary Fig. 4D, 4E). In all three cell lines, these effects were reversed by treatment with Fer-1. Consistent with these data, targeting ACSS2 with two different RNAi in MDA-MB-231-BR (Fig. 3E, Supplementary Fig. 5A) led to significant increase in cell death (Fig. 3F, Supplementary Fig. 5B) and lipid peroxides (Fig. 3G, Supplementary Fig. 5C) which were reversed by treatment with Fer-1 (Fig. 3F, 3G, Supplementary Fig. 5B, 5C) and liproxstatin-1 (Lip-1), another ferroptosis inhibitor (37). (Supplementary Fig. 5B, 5C). Similar results were seen in 4T1-BR cells (Supplementary Fig. 5D–5I).

Figure 3. ACSS2 regulates breast cancer brain metastatic cell survival via ferroptosis suppression.

(A) Representative images depicting tumor growth in organotypic brain slices derived from mice intracranially injected with MDA-MB-231-BR-luc cells detected via bioluminescence. Brain slices containing tumors are treated with Control (DMSO), ACSS2 inhibitor VY-3–249 (ACSS2i) 50 μM or ACSS2i and Ferrostatin-1 (Fer-1) 5 μM for indicated days. (B) Quantification of tumor Bioluminescence at indicated day (Control: DMSO n=3, ACSS2i n=3, ACSS2i +Fer-1 n=3). (C) Quantification of propidium iodide staining flow cytometry of MDA-MB-231-BR cells treated with ACSS2 inhibitor VY-3–249 (ACSS2i) (n=3), ACSS2i + 3-Methyladenine (3-MA) (n=3), ACSS2i + Necrosulfonamide (NSA) (n=3), ACSS2i + Caspase Inhibitor Z-VAD-FMK (ZVAD) (n=3), or ACSS2i + Ferrostatin-1 (Fer-1) for 48 hrs. (D) Representative images of MDA-MB-231-BR cells treated with Control (DMSO), ACSS2i (50 μM), or ACSS2i + Fer-1 (50 μM) and stained with Bodipy C11 (image magnification 20x, scale bar: 200 μM) (left). Quantification of lipid peroxides with Bodipy C11 and flow cytometry of MDA-MB-231-BR cells (right). (E) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or ACSS2 were collected for immunoblot analysis with the indicated antibodies. (F) Quantification of propidium iodide staining flow cytometry of MDA-MB-231-BR cells expressing shRNA against Scramble, ACSS2, and shRNA against ACSS2 and Fer-1. (G) Representative images of MDA-MB-231-BR cells stably expressing shRNA against Scramble or ACSS2 and treated with + Fer-1 and stained with Bodipy C11 (image magnification 20x, scale bar: 200 μM) (left). Quantification of lipid peroxides with Bodipy C11 and flow cytometry of MDA-MB-231-BR cells expressing shRNA against Scramble, ACSS2, and shRNA against ACSS2 and Fer-1 (right). Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM *p-value < 0.05, **p<0.01, ***p<0.001, ***p<0.0001.

We next examined whether Fer-1 treatment could also block tumor regression following VY-3–249 treatment in the ex vivo tumor brain slice model. First, we show that preformed tumors exhibited significantly reduced tumor growth when treated with the ferroptosis inducer erastin, while co-treatment with Fer-1 blocked this effect (Supplementary Fig. 6A), indicating that ex vivo brain slices can undergo ferroptotic regulation. Consistent with the in vitro results (Fig. 3C), VY-3–249 mediated effects on tumor growth ex vivo were reversed by Fer-1 co-treatment (Fig. 3A, 3B). Similar results were seen in 4T1-BR (Supplementary Fig. 6B) and E0771-BR cells using the ex vivo model (Supplementary Fig. 6C). Thus, targeting ACSS2 genetically and pharmacologically induces ferroptosis in BCBM cells in vitro and ex vivo.

ACSS2 regulates ferroptosis in an E2F1-dependent manner in BCBM cells.

To explore downstream mechanisms by which ACSS2 regulates ferroptosis in BCBM cells, we performed bulk RNA-sequencing (RNA-seq) analysis to identify transcriptional changes in gene expression patterns in MDA-MB-231-BR cells containing downregulated levels of ACSS2 by RNAi. Transcriptome analysis of MDA-MB-231-BR cells comparing stable expression of ACSS2 RNAi to non-gene targeting RNAi identified 422 differentially expressed genes consisting of 168 upregulated and 254 downregulated genes using a cut-off value of fold change >2 (and p<0.05); (Supplementary Fig. 7A, Supplementary Table S1). Our RNAseq analysis did not contain changes in major ferroptosis regulators. However, expression of early region 2 binding transcription factor (E2F1)-regulated genes was significantly decreased upon ACSS2 knockdown (Supplementary Fig. 7B). E2F1 is a central player involved in cell cycle and cell death and has recently been implicated in the regulation of ferroptosis (38–40). Consistent with the RNA-seq analysis, we found that the knockdown of ACSS2 leads to the downregulation of E2F1 protein (Fig. 4A, Supplementary Fig. 5A) and RNA levels (Fig. 4B) in MDA-MB-231BR cells and E2F1 protein levels in 4T1-BR cells (Supplementary Fig. 2H, 5D). Consistent with our hypothesis that ACSS2 regulation of E2F1 may be linked to ferroptosis, we found that targeting E2F1 with RNAi in MDA-MB-231-BR cells (Fig. 4C) significantly increases cell death (Fig. 4D) and lipid peroxide levels (Fig. 4E, 4F) that were reversed by Fer-1 co-treatment (Fig. 4D, 4E, 4F). Conversely, we show that overexpression of E2F1 in cells containing RNAi-depleted levels of ACSS2 (Fig. 4G) can reverse ACSS2-mediated induction of cell death (Fig. 4H) and fully rescue lipid peroxide levels (Fig. 4I) compared to controls. In addition, we show that E2F1 overexpression in BCBM cells (Supplementary Fig. 7C) can block cell death-mediated effects of treatment with ACSS2 inhibitor (Supplementary Fig. 7D), as well as block cell death (Supplementary Fig. 7E) and lipid peroxidation (Supplementary Fig. 7F) induced by erastin treatment. Importantly, we show that overexpression of E2F1 in cells containing RNAi-targeted ACSS2 levels can partly reverse tumor growth in vivo (Fig. 4J). Brain trophic cells that contain elevated ACSS2 levels (Fig. 1B) also contain increased E2F1 protein levels in MDA-MB-231-BR (Supplementary Fig. 7G), 4T1-BR (Supplementary Fig. 7H), and E0771-BR cells (Supplementary Fig. 7I) compared to corresponding parental cells, which further supports ACSS2 dependent regulation of E2F1 expression. Therefore, these data suggest that targeting ACSS2 genetically or pharmacologically induces ferroptotic cell death in an E2F1-dependent manner.

Figure 4. ACSS2 regulates ferroptosis in an E2F1-dependent manner.

(A) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or ACSS2 were collected for immunoblot analysis with the indicated antibodies. (B) Quantification of qRT-PCR of total RNA extracts collected from MDA-MB-231-BR cells stably expressing shRNA against Scramble or ACSS2 analyzing ACSS2 and E2F1 RNA levels normalized to housekeeping genes. Two-way ANOVA reported as mean ± SEM. (C) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or E2F1 collected for immunoblot analysis with the indicated antibodies. (D) Quantification of propidium iodide staining flow cytometry of MDA-MB-231-BR cells expressing shRNA against Scramble or E2F1, treated with and without Fer-1. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (E) Representative images of MDA-MB-231-BR cells expressing shRNA against Scramble or E2F1, and Fer-1 Bodipy C11 (image magnification 20x, scale bar: 200 μM). (F) Quantification of lipid peroxides with Bodipy C11 and flow cytometry of MDA-MB-231-BR cells expressing shRNA against Scramble or shRNA against E2F1 and Fer-1. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (G) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or shRNA against ACSS2 and/or overexpression of HA-tagged E2F1 collected for immunoblot analysis with the indicated antibodies. (H) Quantification of propidium iodide staining flow cytometry of MDA-MB-231-BR cells expressing shRNA against Scramble, ACSS2 and overexpressing HA-tagged E2F1. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (I) Quantification of lipid peroxides with Bodipy C11 and flow cytometry of MDA-MB-231-BR cells expressing shRNA against Scramble, ACSS2, or shRNA against ACSS2 and HA-Tagged E2F1 RNA. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (J) Representative images of bioluminescent (top) detection of tumors from mice injected with shSCR and shACSS2, shACSS2 HA-E2F1 MDA-MB-231-BR cells 14 days post-injection. Representative images of H&E analysis (bottom) on coronal sections from mice as above. Bioluminescence signal was quantified from mice injected with MDA-MB-231-BR cells stably expressing shSCR (n=4) and shACSS2 (n=4), shACSS2 HA-E2F1 (n=4) (right). Student’s t-test reported as mean ± SEM; *p<0.05, **p<0.01, ***p-value < 0.001, ****p<0.0001.

ACSS2 regulates ferroptosis in an SLC7A11-dependent manner in BCBM cells.

E2F1 was recently shown to regulate anti-ferroptotic protein SLC7A11 in colon cancer cells (41). SLC7A11 has been found to be elevated in breast cancer patients, including TNBC patients (42). SLC7A11 is an amino acid transporter that facilitates cysteine uptake in exchange for glutamate which controls glutathione metabolism and reduces levels of lipid peroxides (22). We investigated publicly available database and analyzed RNA-seq data from cohort of 32 patients with matched primary basal and HER2+ breast and brain metastatic tumor tissue (GSE184869) and examined levels of SLC7A11. We found that SLC7A11 was significantly upregulated in brain metastatic tissue when compared to matched primary tumors (Fig. 5A). Thus, we tested whether E2F1 can regulate SLC7A11 in BCBM cells. We found that genetically targeting E2F1 reduces the protein (Fig. 5B) and RNA (Fig. 5C) levels of SLC7A11 in MDA-MB-231-BR cells as well as protein levels in 4T1-BR cells (Fig. 5D). Conversely, we detected an increase in SLC7A11 protein levels in MDA-MB-231 cells overexpressing E2F1 (Supplementary Fig. 7C). Consistent with this data, using chromatin immunoprecipitation (ChIP) assay, we show that E2F1 binding is highly enriched on the SLC7A11 promoter in MDA-MB-231-BR cells (Supplementary Fig. 8A). Similarly, stable RNAi knockdown of ACSS2 reduces RNA (Supplementary Fig. 8B) and protein levels of SLC7A11 in MDA-MB-231-BR (Fig. 5E), and protein levels in 4T1-BR (Fig. 5F) and E0771-BR (Fig. 5G) cells. In addition, the reduction of SLC7A11 levels in MDA-MB-231-BR in ACSS2 RNAi targeted cells was reversed by overexpression of E2F1 (Fig. 4G). Further, pharmacologically targeting ACSS2 with VY-3–249 treatment reduced the expression of E2F1, and SLC7A11 in MDA-MB-231-BR (Supplementary Fig. 8C) and 4T1-BR (Supplementary Fig. 8D) cells. All these results suggest that ACSS2 regulates the anti-ferroptotic factor SLC7A11 via E2F1.

Figure 5. ACSS2/E2F1 regulates ferroptosis in a SLC7A11-dependent manner.

(A) Comparison of SLC7A11 mRNA expression between matched breast cancer (BC) and brain metastatic (BCBM) samples using online available Gene Expression Omnibus public datasets (GSE184869). Each dot represents a patient (Basal/HER2+ tumors, n=32). Student’s t-test reported as mean ± SEM. (B) Cell lysates from MDA-MB-231-BR cells stably expressing shRNA against Scramble or E2F1 collected for immunoblot analysis with the indicated antibodies. (C) Total RNA collected from MDA-MB-231-BR cells stably expressing shRNA against Scramble or E2F1. Quantification of qRT-PCR performed on RNA extracts analyzing E2F1 and SLC7A11 gene expression normalized to housekeeping genes. Two-way ANOVA reported as mean ± SEM. (D) Cell lysates from 4T1-BR cells stably expressing shRNA against Scramble or E2F1 collected for immunoblot analysis with the indicated antibodies. (E-G) Cell lysates from MDA-MB-231-BR, 4T1-BR, and E0771-BR cells stably expressing shRNA against Scramble or ACSS2 collected for immunoblot analysis with the indicated antibodies. (H) Cell lysates from MDA-MB-231BR cells stably expressing shRNA against Scramble or ACSS2 and overexpressing Control or SLC7A11 were collected for immunoblot analysis with the indicated antibodies. (I) Quantification of propidium iodide staining flow cytometry of MDA-MB-231-BR cells stably shRNA against Scramble or ACSS2 and overexpressing SLC7A11. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (J) Quantification of lipid peroxides with Bodipy-C11 flow cytometry of MDA-MB-231-BR cells stably expressing shRNA against Scramble or ACSS2 and overexpressing SLC7A11. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (K) Cell lysates from MDA-MB-231-BR cells stably overexpressing Control or SLC7A11 collected for immunoblot analysis with the indicated antibodies. (L) Quantification of propidium iodide staining flow cytometry of MDA-MB-231-BR cells stably overexpressing Control or SLC7A11 treated with DMSO or ACSS2 inhibitor VY-2–249 (ACSS2i) 50 μM. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (M) Quantification of lipid peroxides with Bodipy C11 and flow cytometry of MDA-MB-231BR cells stably overexpressing Control or SLC7A211 treated with DMSO or ACSS2 inhibitor (ACSS2i) 50 μM. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM; * p-value < 0.05,**p-value < 0.01, ***p-value < 0.001, ****p<0.0001.

To link these findings to the previously established pathway driven by OGT/CDK5/p-Ser267-ACSS2 upregulation in BCBM that drives large tumor formation in the brain parenchyma, we show that MDA-MB-231 parental cells overexpressing OGT (Supplementary Fig. 8E), CDK5 (Supplementary Fig. 8F) or ACSS2-S267D mutant (Supplementary Fig. 8G) contained increased protein levels of E2F1, and SLC7A11, suggesting an ACSS2 dependent regulation of E2F1, which in turn drives the regulation of anti-ferroptotic protein SLC7A11. We also show that treating BCBM cells with OGA inhibitor, thiamet-G, which increases total O-GlcNAcylation, leads to increase protein levels of ACSS2, E2F1 and SLC7A11 (Supplementary Fig. 8H). In addition, we show that overexpression of ACSS2-S267D mutant in BCBM cells (Supplementary Fig. 8I) blocks erastin-mediated induction of cell death (Supplementary Fig. 8J) and lipid peroxidation (Supplementary Fig. 8K).

We tested the role of SLC7A11 on cell death and ferroptosis regulation by inhibiting ACSS2. MDA-MB-231-BR cells stably overexpressing SLC7A11 (Fig. 5H) were able to block cell death (Fig. 5I) and lipid peroxidation (Fig. 5J) induced by ACSS2 knockdown as well as treatment with ACSS2 inhibitor VY-3–249 compared to control cells (Fig. 5K, 5L, 5M). Thus, ACSS2-mediated regulation of cell death and ferroptosis is dependent on the regulation of E2F1-SLC7A11 levels.

Novel brain permeable ACSS2 inhibitor induces ferroptosis and block BCBM growth ex vivo and in vivo

We have recently identified novel brain-permeable ACSS2 inhibitors, including compound AD-5584, that are predicted to be more stable VY-3–249 as well as derivative VY-3–135 (43) and can penetrate the blood-brain barrier (44). Consistent with the data above, treatment of BCBM cells with AD-5584 reduces E2F1 and SLC7A11 protein levels (Fig. 6A) and significantly reduced E2F1 binding to SLC7A11 promoter (Supplementary Fig. 8A). In addition, we found that treatment with AD-5584 increases cell death (Fig. 6B) and lipid peroxides (Fig. 6C) in MDA-MB-231-BR cells compared to control-treated cells, which were reversed by treatment with ferroptosis inhibitors Fer-1 and Lip-1. Similar results are shown for 4T1-BR cells treated with AD-5584 in vitro (Supplementary Fig. 9A, 9B, 9C). We also show that treatment of BCBM cells with AD-5584 significantly increases levels of malondialdehyde (MDA) (Supplementary Fig. 9D, 9E), another marker of ferroptosis (37). Since metastatic cells are often more sensitive to ferroptosis inducers (22), we tested whether BCBM cells are more sensitive than parental cells to AD-5584 or erastin treatment. We show that treatment with AD-5584 significantly increase cell death (Supplementary Fig. 9F) and lipid peroxidation (Supplementary Fig. 9G) in MDA-MB-231-BR compared to parental cells and found that erastin treatment also significantly increases cell death in BCBM cells compared to parental cells (Supplementary Fig. 9H). Treatment of AD-5584 also significantly reduced proliferation in both parental and BR cells as measured by EdU staining; however, there was no difference in sensitivity between parental and BR cells (Supplementary Fig. 9I, 9J). Thus, BCBM cells are specifically more sensitive to cell death caused by ferroptosis inducers including erastin and ACSS2 inhibitors compared to primary breast cancer cells.

Figure 6. Novel Brain Permeable ACSS2 inhibitor induces Ferroptosis in an E2F1-dependent pathway.

(A) Cell lysates from MDA-MB-231-BR cells treated with control (DMSO) or ACSS2 inhibitor AD-5584 (50 μM) for 48 hrs were collected for immunoblot analysis with the indicated antibodies. (B-C) Quantification of propidium iodide staining flow cytometry (B) and lipid peroxides with Bodipy C11 (C) of MDA-MB-231-BR cells treated with DMSO, AD-5584 (50 μM), or AD-5584 + Fer-1 (5 μM) or Lip-1 (5 μM). Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (D-E) Quantification of propidium iodide staining for cell death flow cytometry (D) and lipid peroxides with Bodipy C11 (E) of MDA-MB-231-BR cells stably overexpressing control or HA-E2F1 treated with DMSO, or AD-5584 (50 μM) for 48 hrs. Data reported as One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (F) Representative images depicting tumor growth in organotypic brain slices derived from mice intracranially injected with MDA-MB-231-BR-luc cells detected via bioluminescence. Brain slices containing tumors are treated with Control (DMSO), AD-5584 (50 μM) or AD-5584 and Fer-1 (5 μM) for indicated days (left). Quantification of tumor Bioluminescence at indicated day (right) (Control: DMSO n=3, AD-5584 n=3, AD-5584 +Fer-1 n=3). One-way ANOVA with Tukey’s multiple comparisons test reported as mean ± SEM. (G) Representative images of bioluminescent detection of tumors from Balb/C mice injected with luciferase-tagged 4T1-BR cells at Day 0 (prior to drug treatment) and at 10 days post-drug treatment (50 mg/kg) (left). Data are quantified and presented as average Relative Bioluminescence signal from mice injected with 4T1-BR cells treated with Vehicle (n=3) or AD-5584 treated mice (n=3) (middle). Student’s t-test reported as mean ± SEM. Representative images of brain sections stained for H&E 10-days-post treatment (right). (H) Kaplan-Meyer survival analysis quantifying survival of mice injected with 4T1-BR cells and treated with vehicle (n=5) or AD-5584 (n=5). (I) Working model schematic depicting OGT regulates ACSS2 via CDK5 phosphorylation at Serine-267 residue. BCBM cells upregulate ACSS2 to convert acetate to acetyl-CoA. Acetyl-CoA promotes the transcription of E2F1, leading to the transcription of anti-ferroptotic E2F1 target genes, SLC7A11. Expression of SLC7A11 protects BCBM cells from ferroptosis (Created in BioRender. Reginato, M. (2026) https://BioRender.com/hreyurg). *p-value < 0.05, **p<0.01, ***p-value < 0.001, ****p<0.0001.

Consistent with the model that ACSS2-mediates ferroptotic resistance via regulation of E2F1, we show that overexpression of E2F1 in MDA-MB-231-BR cells protects against AD-5584-mediated cell death (Fig. 6D) and lipid peroxidation (Fig. 6E). We also found that treatment with novel ACSS2 inhibitor AD-5584 was able to cause tumor regression in preformed tumors ex vivo, which Fer-1 co-treatment reversed in MDA-MB-231-BR (Fig. 6F) and in 4T1-BR tumors (Supplementary Fig. 10A). Importantly, we show that treating mice with a brain permeable ACSS2 inhibitor AD-5584 can significantly block breast cancer brain metastatic growth of 4T1-BR (Fig. 6G) and MDA-MB-231-BR (Supplementary Fig. 10B) in vivo and significantly extend survival of mice compared to control treatment (Fig. 6H). This inhibitor did not have any observable toxicity ex vivo, as treatment of tumor-free brain slices with AD-5584 did not alter brain slice viability (44). In addition, animals treated with AD-5584 did not lose weight compared to controls (Supplementary Fig. 10C, 10D), nor did they exhibit any observable change in behavior. Lastly, we show that treatment of Patient-derived xenograft organoid (PDxO) cell line HCI-015, derived from brain metastasis of TNBC/basal breast cancer patient (27), with AD-5584 significantly reduced organoid growth compared to control (Supplementary Fig. 10E). Thus, we show that targeting BCBM tumors with a novel brain-permeable ACSS2 inhibitor induces ferroptosis in vitro and blocks tumor growth ex vivo and in vivo.

Discussion

Breast cancer brain metastasis is a serious complication with multiple treatment challenges including limited therapeutic options, the BBB hindering delivery of many cancer treatments and little understanding of BCBM-specific signaling that allows for cell survival in the brain environment. Here, we report findings that address some of these major challenges.

First, we identify three potentially actionable targets, OGT, CDK5 and ACSS2 that are critical for growth of BCBM cells in the brain parenchyma. OGT is overexpressed in many cancers (18) including glioblastoma (12,45) and can regulate CDK5 and ACSS2 that results in increase of acetate conversion to acetyl-CoA and contributes to GBM growth in vivo (12). OGT inhibitors can block GBM cell growth in vitro and ex vivo (12). However, their clinical use has not yet been evaluated, and since inducible Ogt deletion in adult mice results in a lethal phenotype (46), toxicity is likely to be a concern with these compounds. Consistent with our results, CDK5 has recently been implicated as a key regulator of BCBM growth as CDK5 levels were found elevated in human BCBM samples compared to primary breast tumors and high CDK5 expression correlated with poor patient survival, and genetic inhibition of CDK5 suppresses breast cancer brain metastasis outgrowth in brain-trophic cell lines (47). Moreover, treatment with pan-CDK inhibitor Roscovitine delayed BCBM growth in vivo (47). Roscovitine has been used in many cancer models and shown in preclinical studies to have efficacy (48). However, in clinical trials, Roscovitine yielded mixed results, with evidence of tumor stabilization but no measurable objective responses (49). Nevertheless, Roscovitine may be able to enhance response to immune checkpoint therapy as it was recently shown in BCBM preclinical models (47). ACSS2 has been implicated as a critical regulator of tumors in the brain specifically in GBM (12). In this study, we show that targeting ACSS2 genetically and pharmacologically blocks BCBM growth in vivo. Small molecules targeting ACSS2 are in clinical trials for the treatment of advanced tumors and have been found to be well tolerated in Phase 1 clinical trials (50) (NCT04990739), consistent with the finding that Acss2-knockout mice are viable (51).

Second, we recently identified two novel ACSS2 inhibitors that penetrate the blood-brain barrier that show efficacy in pre-clinical models of BCBM (44) and show here that ACSS2 inhibitor AD-5584 induces cell death via ferroptosis and shrinks preformed BCBM tumors ex vivo and in vivo. Using an ex vivo BCBM tumor model, we previously showed that targeting ACSS2 can also synergize with radiation treatment (44). Since ferroptosis is a form of inflammatory cell death (52) that can increase the efficacy of immune therapy in breast cancer (53), targeting ACSS2 may alter immune cells in the brain tumor microenvironment and sensitize them to immune checkpoint inhibitors. Indeed, recent studies have shown that targeting ACSS2 may also regulate the host immune response to cancer (54). Further studies are needed to test if targeting ACSS2 can alter the brain immune microenvironment and sensitize to radiation and immune therapy.

Third, we identified that the ACSS2/E2F1/SLC7A11 signaling axis plays a key role in cell survival via inhibition of ferroptosis in BCBM cells. Interestingly, ACSS2 was critical for growth in breast cancer cells in the brain but not mammary fat pad in vivo. The latter aligns with recent findings showing that ACSS2 knockdown has no effect on the mammary fat pad of immune-deficient mice (54). Thus, ACSS2 represents a critical regulator of breast tumors in the brain parenchyma. We have previously demonstrated that CDK5 directly phosphorylates ACSS2 at Ser267, and that OGT regulates the protein levels of both CDK5 and pACSS2 in glioblastoma cells (12)—findings consistent with our current observations in BCBM cells. Our prior work demonstrated that ACSS2 phosphorylation on S267 blocks its degradation and enhances acetyl-CoA production in glioblastoma cells (12), and that pharmacologic inhibition of ACSS2 in BCBM cells reduces acetate conversion to acetyl-CoA (44). In glioblastoma cells, ACSS2 can also be phosphorylated at S659 by AMPK, which promotes its nuclear translocation (13). Future studies will examine whether other phosphorylation events occur on ACSS2 in BCBM cells.

Moreover, we showed that ACSS2 protects against ferroptosis in BCBM cells in a E2F1-SLC7A11-dependent manner. Our model suggests that ACSS2 may likely regulate acetylation on key transcriptional regulators on the E2F1 promoter, and E2F1, in turn, can regulate gene expression of SLC7A11 to protect against ferroptosis (Fig. 6I). ACSS2 regulates gene expression in neurons through direct binding to chromatin to promote histone acetylation (55). In addition, motif analysis of ACSS2-targeted, acetate-induced genes implicates the regulation of transcription factors, including E2F3 (56). Thus, ACSS2 in breast cancer brain metastatic cells may serve to protect from ferroptosis via regulation of gene expression.

ACSS2 has emerged as a critical regulator of tumor growth in a number of cancers by affecting cancer lipid metabolism, and acetylation (13). Our study is the first to link ACSS2 in regulating ferroptosis in brain tropic breast cancer cells. We found that targeting ACSS2 genetically or pharmacologically increased lipid peroxidation in multiple BCBM cells. Additionally, we show that ACSS2 regulates E2F1 proteo-transcriptomic levels and E2F1-regulated genes in BCBM trophic cells and that ACSS2 regulation of ferroptosis is E2F1-dependent. E2F1 alone was able to regulate ferroptosis, and targeting E2F1 induces cell death and lipid peroxidation in BCBM cells, which were reversed by ferrostatin-1 treatment. These data are consistent with recent studies showing E2F1 as a ferroptosis regulator in cancer cells. In prostate cancer, RB1 loss/E2F activation sensitized cancer cells to ferroptosis (39), while in fibrosarcoma cells, E2F1 overexpression protected cells from ferroptotic inducers, including erastin, and this protection was associated with E2F1 regulation of SLC7A11 expression (40). Interestingly, we found brain tropic cells containing elevated levels of E2F1 compared to parental breast cancer cells suggesting that E2F1-mediated protection from ferroptosis may be a mechanism for breast cancer cells to adapt to the unique brain microenvironment. An expression analysis comparison between primary human breast cancers with BCBM found E2F1 to be enriched in BCBM samples compared to primary breast cancers (57). Since E2F1 is a well-known regulator of proliferation, it will be of interest to understand whether E2F1-mediated regulation of proliferation can contribute to BCBM growth. Taken together, our data suggest that ACSS2/E2F1/SLC7A11 axis can regulate ferroptosis in breast cancer brain tropic cells.

Since the brain microenvironment is deficient in several nutrients that cancer cells require, including free fatty acids, tumors in the brain, including BCBM, must increase tumor fatty acid synthesis in order to proliferate (6). ACSS2 can contribute to de novo fatty acid synthesis by converting acetate to acetyl-CoA, which generates fatty acids (51). Ferroptosis highly depends on lipid metabolism as lipids are peroxidated and cause membrane disruption, leading to outer membrane rupture (58). Since BCBM cells contain elevated fatty acid synthesis, this may sensitize them to ferroptosis leading to a requirement for mechanisms to protect them from this form of cell death in the brain microenvironment. Our data suggests that BCBM cells increase ACSS2 levels, leading to protection from ferroptosis via regulation of E2F1-mediated gene expression of anti-ferroptotic protein SLC7A11. Human genomic analysis found that ferroptotic markers are associated with poor clinical outcomes in BCBM patients (59). Whether ACSS2 regulation of lipid synthesis can also contribute to the regulation of ferroptosis in BCBM cells needs further examination.

In summary, we have shown that targeting ACSS2 in breast cancer brain metastatic cells may have clinical utility in inducing cell death via ferroptosis without the toxicity that is seen with agents such as erastin (60). These new ACSS2 inhibitors may also have potential to synergize with current therapies including radiotherapy or immunotherapy. Our study provides a framework for targeting BCBM tumors with brain-penetrant ACSS2 inhibitors to induce ferroptotic cell death, exploit these tumors’ metabolic vulnerability and improve outcomes of BCBM patients.

Significance:

Targeting ACSS2 with brain penetrant inhibitors induces ferroptosis in brain metastatic breast cancer by perturbing an E2F1/SLC7A11 axis, providing an efficacious strategy for treating brain metastatic tumors.