Adipose triglyceride lipase is a therapeutic target in advanced prostate cancer that promotes metabolic plasticity

Abstract

Lipid metabolism plays a central role in prostate cancer. To date, the major focus has centered on de novo lipogenesis and lipid uptake in prostate cancer, but inhibitors of these processes have not benefited patients. Better understanding of how cancer cells access lipids once they are created or taken up and stored could uncover more effective strategies to perturb lipid metabolism and treat patients. Here, we identified that expression of adipose triglyceride lipase (ATGL), an enzyme that controls lipid droplet homeostasis and a previously suspected tumor suppressor, correlates with worse overall survival in men with advanced, castration-resistant prostate cancer (CRPC). Molecular, genetic, or pharmacological inhibition of ATGL impaired human and murine prostate cancer growth in vivo and in cell culture or organoids under conditions mimicking the tumor microenvironment. Mass spectrometry imaging demonstrated ATGL profoundly regulates lipid metabolism in vivo, remodeling membrane composition. ATGL inhibition induced metabolic plasticity, causing a glycolytic shift that could be exploited therapeutically by co-targeting both metabolic pathways. Patient-derived phosphoproteomics identified ATGL serine 404 as a target of CAMKK2-AMPK signaling in CRPC cells. Mutation of serine 404 did not alter the lipolytic activity of ATGL but did decrease CRPC growth, migration, and invasion, indicating that non-canonical ATGL activity also contributes to disease progression. Unbiased immunoprecipitation/mass spectrometry suggested that mutation of serine 404 not only disrupts existing ATGL protein interactions but also leads to new protein-protein interactions. Together, these data nominate ATGL as a therapeutic target for CRPC and provide insights for future drug development and combination therapies.

Introduction

Otto Warburg received the Noble Prize in 1931 for his seminal work demonstrating that cancer cells can shift their metabolism to meet the increased energy and biosynthetic demands of a rapidly growing tumor. To that end, altered metabolism is now an accepted hallmark of cancer (1). While most of the research in the field has focused on aerobic glycolysis, it is now clear that many cancer types can also depend on other metabolic pathways. For example, prostate cancer exhibits high lipid metabolism. Accordingly, inhibitors of de novo lipogenesis like TVB-2640 (a fatty acid synthase (FASN) inhibitor) or fatty acid uptake have been developed and are currently being tested for the treatment of advanced solid tumors including prostate cancer (2,3). Despite the central importance of lipid metabolism in the disease, these inhibitors have not yet demonstrated any benefit to prostate cancer patients. Moreover, emerging data indicate functional redundancy that can bypass the efficacy of inhibitors of either de novo lipogenesis or lipid uptake (3,4). Thus, new approaches are needed to effectively target this oncogenic pathway in advanced prostate cancer.

Given the major role of lipid metabolism in prostate cancer but concerns with regards to compensation between de novo lipogenesis and lipid uptake, we sought to explore whether targeting downstream lipid droplet homeostasis could be efficacious in the treatment of prostate cancer. We reasoned that regardless of how lipids such as triglycerides (TGs) are accumulated in prostate cancer cells, the available pool of lipids that could be used by rapidly growing cells for energy, building blocks, or to buffer against lipid toxicity, would still be regulated at the level of the intracellular lipid droplets. By assessing genetic and transcriptional changes in men with metastatic, castration-resistant prostate cancer (mCRPC), we identified adipose triglyceride lipase (ATGL, encoded by the gene PNPLA2), the initial and rate-limiting step in the breakdown of TGs (5), but not other enzymes involved in lipid droplet TG synthesis or breakdown, as having the genomic and transcriptomic traits of an oncogene (described below). Notably, ATGL was described as a tumor suppressor in lung cancer (6,7), while in prostate cancer conflicting reports have been published (8,9). Here, using a variety of state-of-the-art in vivo and ex vivo approaches, we provide strong evidence that ATGL plays a pivotal functional role in advanced prostate cancer and represents a novel therapeutic target. Moreover, our mechanistic studies inform on strategies to target ATGL for cancer treatment.

Materials and Methods

Institutional Review Board Statement:

The animal study protocol was approved by the Institutional Review Board of the University of Texas MD Anderson Cancer Center (IACUC protocol#: 00001738-RN01, approved 10/4/2021).

Bioinformatics

Phosphoproteomic data (10) was mined using an updated AMPK substrate motif and PWM (11). AMPK targets were ranked by PWM score, P value, and change in phosphorylation. A complete list can be found in Supplemental File 1.

Cell lines

RWPE-1 (9RRID:CVCL_3791), LNCaP (RRID:CVCL_0395), C4–2 (RRID:CVCL_4782), PC-3 (RRID:CVCL_0035), 22Rv1(RRID:CVCL_1045), HEK293 (RRID:CVCL_0063) and NCI-H660 (RRID:CVCL_1576) were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA) and cultured according to ATCC protocols. C4–2B (RRID:CVCL_4784) and C4–2B-LT (C4–2B cells stably expressing firefly luciferase and tdTomato, gifts from Dr. Sue-Hwa Lin (UT MD Anderson Cancer Center)), MDA-PCa-144–13 (gift from Dr. Nora M. Navone (UT MD Anderson Cancer Center)), and RM-9 (RRID:CVCL_B461) cells (gift from Dr. Timothy Thompson (UT MD Anderson Cancer Center)) were cultured as previously described (12,13). Cell lines used for this study were authenticated prior to any experiments using short tandem repeat analyses via the MD Anderson Cytogenetics and Cell Authentication Core (CCAC) and cells were routinely tested for mycoplasma before cryopreservation and during experiments (every 6 weeks). If new cell lines were created such as knockout or addbacks, cells were tested for mycoplasma prior to cryopreservation. All wildtype cell lines were used for 10 passages or less. Knockout cell lines were obtained by single colony expansion. Once the cells were expanded to create cryo stocks, the cells were only used up to passage 8 before being replaced with fresh cells from cryo vials. The addbacks were created from the single colony expansion knockouts (PN+1) and were used for less than 8 passages before switching to new vial from cryo storage. Cell lines were maintained at all times without antibiotics.

CRISPR knockout and addback strategies

Single guide RNAs (sgRNA) were designed using the Dharmacon CRISPR design tool (2018) and tested for off-target effects using ChopChop (https://chopchop.cbu.uib.no/ 2019). Two sgRNAs targeting Exon2 of PNPLA2 (sgRNA1: 5’-GATGTTCCACGTCTTCTCGC-3’ (−)-strand, sgRNA2: 5’-CGCCGACGTAGTAGACGCCG-3’ (−) strand) and Exon 3 (sgRNA3 5’-TTGAGGTATCTAAAGAGGCC-3’) and a scramble sgRNA (https://doi.org/10.1038/nature20565) were ordered via Sigma as oligonucleotides. SgRNAs were then cloned into the pLentiCRISPRv2 (RRID:Addgene_52961). The different pLentiCRISPRv2 constructs containing each sgRNA were virus packaged using HEK293 (RRID:CVCL_0063) (Lentviral packaging, 3^rd generation) and then transfected into different prostate cancer cell lines and selected using puromycin. Prostate cancer cell lines containing the scramble sgRNA were used as a pool. For each of the three different sgRNAs, several single colonies were confirmed via sequencing, immunoblot, and then tested for proliferation, migration and TG accumulation using Oil Red O staining. A description of the knockout cell lines used in this study is described in Supplementary Table 1.

Addback constructs were created by first mutating the PAM sequence used by the sgRNA for the ATGL knockout. This was achieved by site direct mutagenesis in the pDONR223-PNPLA2 plasmid using primers containing a mismatch, resulting in the mutation of nucleotide 51 (C to T) while maintaining the amino acid residue (F). In addition to the PAM mutation, we also created a S404A (TCG to GCG) mutation and a S47A (TCG to GCG) mutation using site-directed mutagenesis. Constructs were then inserted into the pLenti-PGK-Neo-DEST backbone via an LR reaction (Gateway^™ LR Clonase^™ II Enzyme mix, Cat#11791020, Invitrogen, Waltham, MA, USA) virus packaged in HEK293T cells and then stably integrated into the ATGL KO single colonies via G418 selection.

shRNA cell lines

pGIPZ-shRNA constructs were purchased from the MDACC Functional Genomics Core and cloned via XhoI (Cat#R0146, NEB, Ipswich, MA, USA) and MluI (Cat# R0198, NEB, Ipswich, MA, USA) into pINDUCER10 (RRID:Addgene_44011; gift from Thomas Westbrook). Cells were transfected viral lentiviral packaging and selection via puromycin (2 μg/ml). For human prostate cancer cell lines seven different shRNAs and for murine prostate cancer cell lines three different shRNAs targeting ATGL were tested. The shRNA construct that showed the best knock down efficiency via immunoblot was selected for further experiments. A list of the shRNAs used in this study is provided in Supplementary Table 2.

siRNA treatment

Chemical siRNAs (Silencer select) were obtained via Invitrogen. For immunoprecipitation, cells were plated at 60% confluency in RPMI 1640 (no glucose) supplemented with 5 mM glucose and 5% dialyzed FBS. Cells were transfected with silencer RNAs for 72 h before cells were collected as described in the immunoblot section. A list of the siRNAs used in this study is provided in Supplementary Table 3.

Colony formation assays

5000 cells were plated in RPMI 1640 (no glucose) supplemented with 5% (LNCaP, C4–2, C4–2B-LT) or 0.5% (PC-3, RM-9, 22Rv1) dialyzed FBS and glucose adjusted as noted in figures. To test the combination of atglistatin and AZ-PFKFB3–26 using RM-9 cells, 15,000 cells/well were initially plated. Cells were fed every 48 h with media or retreated with inhibitors as noted in figures. Assays were run for various lengths of time (ex. low glucose conditions took longer to form colonies) but data were always normalized to controls on the same plate series. Cells were fixed using 10% formalin (Cat# HT501128, Sigma, St. Louis, MO, USA) and then washed with PBS. After drying overnight, fixed cells were stained with crystal violet (0.0125% dissolved in water) and washed with water until excessive crystal violet solution was removed. The plate was then dried and scanned at a resolution of 600 dpi. The colony formation assay was quantified using ImageJ (RRID:SCR_003070) by converting the image into grayscale, processing to binary, and then measuring the area % per well that is occupied by the colonies.

Wound healing (scratch test) assays

Cells were plated in 6-well plates at 80–90% confluency in RPMI 1640 (no glucose) supplemented with 5% (LNCaP, C4–2, C4-2B-LT) or 0.5% (PC-3, RM-9, 22Rv1) dialyzed FBS, as well as 5 mM glucose, unless otherwise indicated. If supplemented with fatty acids, cells were incubated 48h before scratch and then pretreated the day of scratch. Doxycycline-inducible shRNA-containing cell lines were pretreated in flasks with 600 ng/ml doxycycline for at least 7 days prior to the assay. RM-9 or RM-9 shRNA cell lines were treated with atglistatin (Cat#5301510001, Calbiochem, Sigma, St. Louis, MO, USA) immediately after the scratch (t=0). Scratch was performed using a 200 μl pipette tip. Images were taken with a Cytation 5 (BioTEK, Serial # 1705171D, 4xPL FL, Phase contrast) at timepoint 0 h and 24 h. Gaps from scratch were quantified via ImageJ (RRID:SCR_003070) using the MRI wound healing feature. Area of gap was recorded and then the % area wound healing was calculated using the formula $(\frac{area\left(t=0h\right)–area\left(t=24h\right)}{area\left(t=0h\right)})\mathrm{*}100$.

Cell viability assays

NCI-H660 and MDA-PCa-144–13 cells were plated in RPMI-1640 media containing 5 μg/ml insulin, 1 μg/ml transferrin, 30 nM sodium selenite, 10 nM hydrocortisone, 10 nM beta-estradiol, 4 mM L-glutamine, 10 mM HEPES, 1 mM sodium pyruvate and 5% FBS at a density of 5×10³ in 96-well plates. After 48 h, cells were treated with STO-609 (Cat # 1551, Tocris, Bristol, UK) or SGC-CAMKK2–1 (Cat#SML2834, Sigma, St. Louis, MO, USA) and incubated for 3 or 7 days. For siRNAs experiments, NCI-H660 cells were transfected with 100 nM final concentration siControl #1, siCAMKK2 #1 and siCAMKK2 #2 for 3 or 7 days. Then, a resazurin reduction assay, a fluorometric method, was performed according to the manufacturer’s protocols (CellTiter-Blue assay; Cat#G8080, Promega, Madison, WI, USA) to determine the presence of metabolically active cells. Fluorescent signal was measured at a wavelength of 560Ex/590Em using a Synergy H4 microplate reader (BioTek Instruments, Synergy H4, serial number 140417F).

Proliferation assays (Hoechst)

Media was removed and cell-containing 96-well plates were frozen at −80°C. Plates were then thawed and 100 μl of MilliQ water was added to each well. After an incubation at 37°C for 1 h, plates were frozen at −80°C. At the day of the assay, cell plates were thawed and 100 ul Hoechst stain working solution (Hoechst 33342, Cat#H3570, Invitrogen, Waltham, MA, USA; 2 μl diluted in 10 ml TNE buffer, pH 7.4) was added to each well. Plates were measured on a plate reader (Synergy H4, serial number 140417F) at the following settings: excitation: 360/20.0, emission: 460/20.0, optics: top, gain: 70, light source: xenon flash, read speed: normal, delay: 100 msec, measurements/data point: 10, read height: 8 mm.

Organoid models

Hi-MYC mice (strain number 01XK8) and Transgenic Adenocarcinoma of the Mouse Prostate (TRAMP) mice (strain number 003135) were obtained from the National Cancer Institute Mouse Repository at Frederick National Laboratory for Cancer Research. Five Hi-MYC male mice were sacrificed at 6 months of age for prostate harvesting, and the line was bred on the same genetic background FVB/N. For TRAMP, five male mice at 8 months of age were used, and the line was bred on a C57BL/6 background.

Single cell suspension (based on (14)):

Mouse prostate tissues were digested in Advanced DMEM/F12/Collagenase II (1.5mg/ml)/Hyaluronidase (1000 u/ml) (Life Technologies, Carlsbad, CA, USA) plus 10 μM Y-27632 (Tocris, Bristol, UK) for 1 h at 37°C with 1500 rpm mixing, continuously agitated. For prostates from TRAMP mice, the digestion time was extended to 75 min. Subsequently, after centrifuging at 150 g for 5 min at 4°C, digested cells were suspended in 1 ml TrypLE with 10 μM Y-27632 and digested for 15 min at 37°C and neutralized in DMEM/F12/FBS (0.05%). Dissociated cells were subsequently passed through 70 μm and 40 μm cell strainers (BD Biosciences, San Jose, CA, USA) to obtain a single cells suspension. Samples were resuspended in 1x PBS and sorted by Flow Cytometry (Becton-Dickinson Aria II and/or Becton-Dickinson Influx) for 4′,6-diamidino-2-phenylindole (DAPI) to enrich living cells.

3D culture (based on (15,16)):

Next, Hi-MYC DAPI negative, prostate epithelial cells were cultured in ADMEM/F12 supplemented with B27 (Life Technologies, Carlsbad, CA, USA), 10 mM HEPES, Glutamax^™ (Life Technologies, Carlsbad, CA, USA) and penicillin/streptavidin and contained following growth factors: EGF 5–50 ng/ml (Peprotech, Cranbury, NJ, USA), R-spondin1 conditioned medium or 500 ng/ml recombinant R-spondin1 (Peprotech), 100 ng/ml recombinant Noggin (Peprotech, Cranbury, NJ, USA) and the TGF-β/Alk inhibitor A83–01 (Tocris, Bristol, UK). Dihydrotestosterone (Sigma, St. Louis, MO, USA) was added at 1 nM final concentration. Murine prostate organoids were passaged via trituration with trypsinization with TrypLE for 15 min at 37°C. Passaging was performed every week with a 1:5–1:10 ratio.

Organoid treatment:

Following dissociation with TryplE for 15 min at 37°C, Hi-MYC or TRAMP organoids were seeded at 50,000 cells embedded in 20 μL of growth factor-reduced Matrigel^® per well in a 12-well plate. Organoids were cultured with aDMEM/F12 for high-glucose medium (17 mM glucose) or DMEM for low-glucose (5 mM glucose), in the presence of atglistatin or DMSO (vehicle). Organoids were grown for 14 days, and medium was refreshed every 3 days. At the end of treatment, organoids were imaged with a Leica Thunder microscope, and their diameter manually measured with ImageJ (RRID:SCR_003070). Organoids were then dissociated with TryplE and viable cells determined with Vi Cell BLU analyzer (Beckman Coulter, Brea, CA, USA) for cell growth analysis.

Organoid lipidomics:

Lipid species were extracted and analyzed by Lipometrix, at KU Leuven, Belgium. Briefly, lipid extraction was performed with 1 N HCl:CH3OH 1:8 (v/v), 900 μl CHCl₃ and 200 μg/ml of the antioxidant 2,6-di-tert-butyl-4-methylphenol (BHT; Sigma-Aldrich). A mixture of deuterium-labeled lipids SPLASH^® LIPIDOMIX^® Mass Spec Standard (#330707, Avanti Polar Lipids, Alabaster, AL, USA) was spiked into the extract mix. The organic fraction was evaporated using a Savant Speedvac spd111v (Thermo Fisher Scientific) at room temperature and the remaining lipid pellet was stored at −20°C under argon. Just before mass spectrometry analysis, lipid pellets were reconstituted in 100% ethanol. Lipid species were analyzed by liquid chromatography electrospray ionization tandem mass spectrometry (LC-ESI/MS/MS) on a Nexera X2 UHPLC system (Shimadzu) coupled with hybrid triple quadrupole/linear ion trap mass spectrometer (6500+ QTRAP system; AB SCIEX). Chromatographic separation was performed on a XBridge amide column (150 mm × 4.6 mm, 3.5 μm; Waters) maintained at 35°C using mobile phase A [1 mM ammonium acetate in water-acetonitrile 5:95 (v/v)] and mobile phase B [1 mM ammonium acetate in water-acetonitrile 50:50 (v/v)] in the following gradient: (0–6 min: 0% B > 6% B; 6–10 min: 6% B > 25% B; 10–11 min: 25% B > 98% B; 11–13 min: 98% B > 100% B; 13–19 min: 100% B; 19–24 min: 0% B) at a flow rate of 0.7 mL/min which was increased to 1.5 mL/min from 13 minutes onwards. TGs and DGs were measured in positive ion mode with a neutral loss scan for one of the fatty acyl moieties. Lipid quantification was performed by scheduled multiple reactions monitoring, the transitions being based on the neutral losses or the typical product ions as described above. The instrument parameters were as follows: Curtain Gas = 35 psi; Collision Gas = 8 a.u. (medium); IonSpray Voltage = 5500 V and −4,500 V; Temperature = 550°C; Ion Source Gas 1 = 50 psi; Ion Source Gas 2 = 60 psi; Declustering Potential = 60 V and −80 V; Entrance Potential = 10 V and −10 V; Collision Cell Exit Potential = 15 V and −15 V. The following fatty acyl moieties were taken into account for the lipidomic analysis: 14:0, 14:1, 16:0, 16:1, 16:2, 18:0, 18:1, 18:2, 18:3, 20:0, 20:1, 20:2, 20:3, 20:4, 20:5, 22:0, 22:1, 22:2, 22:4, 22:5 and 22:6 except for TGs which considered: 16:0, 16:1, 18:0, 18:1, 18:2, 18:3, 20:3, 20:4, 20:5, 22:2, 22:3, 22:4, 22:5, 22:6.

Peak integration was performed with the MultiQuantTM software version 3.0.3. Lipid species signals were corrected for isotopic contributions (calculated with Python Molmass 2019.1.1) and were quantified based on internal standard signals and adheres to the guidelines of the Lipidomics Standards Initiative (LSI) (level 2 type quantification as defined by the LSI).

Xenograft mouse models

This animal study was approved by and conducted under the Institutional Animal Care and Use Committee at the University of Texas MD Anderson Cancer Center (MDACC). NSG mice were obtained at the age of 6 weeks, castrated at week 7, and at 8 weeks 0.5 × 10⁶ cells in 50 μl PBS mixed with 50 μl Matrigel^® were injected subcutaneously into the flank. Tumor growth was monitored using caliper measurement and tumor volume was calculated using the formula $\frac{(length\times widt{h}^2)}{2}$. When tumors reached a length of 1.5 cm, tumors were dissected and cut into three pieces to 1) fix in formalin for IHC, 2) embed in OCT for DESI/MS, and 3) cryo-smash in liquid nitrogen for RNA/WB analysis.

Histology and immunostaining

Tumor tissue processing, embedding, H&E and immune staining was conducted by the MD Anderson Cancer Center Science Park Research Histology, Pathology, and Imaging Core (RHPI) (Smithville, TX, USA). The following antibodies were used: cleaved caspase-3 (Cat# 9661, Cell Signaling Technology, RRID:AB_2341188), p-HH3 (Ser10) (Cat# 06–570, Millipore, RRID:AB_310177). All image analyses were performed on the entire tumor section stained per slide. Image analysis to calculate % necrosis was conducted as before (17) using QuPath v0.3.0 (RRID:SCR_018257, University of Edinburgh, Edinburgh, UK).

Oil Red O staining

The Oil Red O staining protocol was modified from (18). For the panel of prostate cancer cell lines treated with siPNPLA2, cells were seeded in 24-well plates to 60% confluency in RPMI 1640 supplemented with 5% charcoal-stripped FBS and then treated for 72 hours with chemical siRNAs and/or R1881. For TG assessment of ATGL KO and addback cells, cells were seeded in 24-well plates (80,000 cells/well) in RPMI 1640 no glucose supplemented with 5% or 0.5% dialyzed FBS and 5 mM glucose. After 24 h treatment with BSA-coupled oleate or BSA, cells were fixed and stained with Oil Red O as described below. RM-9 (RM-9, RM-9-shPNPLA2, RM-9 shControl) cells were plated in RPMI 1640 no glucose supplemented with 0.5% dialyzed FBS and 5 mM glucose and treated atglistatin for 24 h hours and/or 600 ng/ml doxycycline, fixed with 10% formalin and stained with Oil Red O. In brief, the 6 parts of the Oil Red O stock solution (2.5 g/l Oil Red O, Cat#O0625, Sigma, St. Louis, MO, USA, diluted in isopropyl alcohol) was diluted the day of the assay with 4 parts of MilliQ water and filtered. Cells that were treated with BSA or different amounts of BSA-coupled oleate for 24 h, were fixed with 10% buffered formalin (Cat #HT501128, Sigma, St. Louis, MO, USA) for 30 min. Cells were washed with 60% isopropyl alcohol and incubated with the Oil Red O working solution for 20 min. After washing with 60% isopropyl alcohol, cells were kept in PBS and brightfield images at 20x were taken using the BioTek Cytation 5. To quantify the Oil Red O stains, images were converted to grayscale and the area of stain for each picture was determined using ImageJ (RRID:SCR_003070) and normalized to cell number per picture (manually counted).

Immunoblot analyses

Cells were washed twice with ice cold PBS and then lysed in RIPA buffer (50 mmol/L Tris (pH 8.0), 200 mmol/L NaCl, 1.5 mmol/L MgCl2, 1% NP-40, 1 mmol/L EGTA, 10% glycerol, 50 mmol/L NaF, 2 mmol/L Na₃VO₄, and protease inhibitors). After 0.5 – 1 h rotation at 4°C, the cell lysate was spun down and the supernatant was used for further analysis. Protein lysates were quantified via Bradford (Protein Assay Dye Reagent Concentrate Cat#5000006, Bio-Rad, Hercules, CA, USA) and 30 μg protein mixed with Laemmli buffer (62.5 mmol/L Tris-HCl (pH 6.8), 2% SDS, 25% glycerol, 5% β-mercaptoethanol, 0.01% bromophenol blue) were loaded per well on the SDS-PAGE (self-poured and adjusted to protein size (7.5–12.5%) or purchased premade (Criteron TGX precast gels, Cat#5671095, Bio-Rad, Hercules, CA, USA). The SDS-PAGE was separated at 100V or 150V (premade gels) in TRIS-Glycin buffer with SDS. Gels were transferred over night at 30V onto PVDF membranes (Bio-Rad Immun-Blot^® PVDF Membrane 0.2 μM) at 4°C using a TRIS-Glycine/methanol buffer system. Membranes were then blocked at room temperature with Tris buffered saline (TBS) containing Tween 20 (TBST: 200 mM Tris. 1.5 mM NaCl) containing 5% (w/v) milk powder for one hour. Following the blocking the membranes were incubated with primary antibody in TBST-containing 0.5% milk powder for up to 48 h. After three, 5-minute washes with TBST, membranes were incubated with the secondary HRP conjugated antibody according to primary antibody species (Bio-Rad, Hercules, CA, USA) for up to 90 min at room temperature. After three washes with TBST, blots were developed using the Biosystems C-600 Imager (Azure, Dublin, CA, USA).

Cell cycle analysis

Isogenic C4–2 PNPLA2 (ATGL) KO and add-back cell lines (WT, S47A, or S404A) were seeded at 200,000 cells/well in 6-well plates with RPMI 1640, 5% dialyzed FBS, and 5 mM glucose-containing media for two days. Cells were then trypsinized and treated with Hoechst 33342 (H3570, Invitrogen^™, 1:2000) in PBS for 30 min at 37°C. After spinning at 1000 RPM for 5 min, cells were washed twice in PBS and analyzed with flow cytometry using Cytek Northern Lights^™ (Cytek, Fremont, CA). FACS results were analyzed using Cytobank platform (https://cytobank.org).

Annexin V/propidium iodide (PI) labeling (apoptosis analysis)

The effects on cell viability, apoptosis, and necrosis caused by ATGL deletion were investigated using Annexin V-FITC/PI staining. Isogenic C4–2 PNPLA2 (ATGL) knockout (KO) and add-back cell lines (WT, S47A, or S404A) were seeded at 200,000 cells/well in 6-well plates with RPMI 1640, 5% dialyzed FBS, and 5 mM glucose media for 2 days. Cells were then trypsinized, washed with cold PBS, and resuspended in 1X annexin-binding buffer (Cat# V13242, Invitrogen^™). Cell density was accounted for before treatment with 5 μL of FITC Annexin V (Cat# V13242, Invitrogen^™) and 1 μL of PI working solution (Cat# V13242, Invitrogen^™, 100 μg/mL) per 100 μL of cell suspension for 15 min at room temperature. After the incubation period is finished, 300 μL of 1X annexin-binding buffer was further added. Cells were subsequently analyzed with flow cytometry, using Cytek Northern Lights^™ (Cytek, Fremont, CA). FACS results were analyzed using Cytobank platform (https://cytobank.org).

De novo lipogenesis assay

C4–2 scramble, ATGL KO, WT addback, S404A and S47A cell lines were cultured in RPMI 1640 with 2.5 mM glucose and 5% dialyzed FBS. The day prior to the lipogenesis measurement, cells were washed with PBS and changed to 1 ml serum starvation medium (RPMI 1640 with 2.5 mM glucose, 2 mM sodium pyruvate and 100 nM insulin). The day of the assay, the starvation media was replaced with lipogenesis medium [serum starvation medium + 10 μM sodium acetate and 0.5 μCi ³H-acetate (Cat# ART 202, American Radiolabeled Chemicals, Inc., St. Louis, MO, USA)] and cells were incubated for 2 h at 37°C. After the incubation, cells were washed twice with PBS (two times with 500 μl) and lysed with 120 μl of 0.1 N HCl. Lipids were extracted from the lysate using the Folch extraction method and part of the cell lysate (10 μl) was collected for protein quantification using the Pierce^™ BCA protein assay (Cat# 23225, Thermo Scientific, Rockford, IL, USA). The lipid-extracted phase (chloroform phase) was transferred to 4 ml scintillation vials, and to those, 3 ml of liquid scintillation fluid was added to each vial. Radioactivity trace in the chloroform phase was counted for 5 min using a liquid scintillation counter (PerkinElmer). Protein concentration of cell lysate quantified by Pierce^™ BCA protein assay was used to normalize the radioactivity count and expressed the data as CPM per mg protein.

Immunoprecipitations

Stable overexpression cell lines were created by lentiviral packaging plx304-ATGL WT or plx304-ATGL S404A in HEK293T cells and then transfecting C4–2 parental cells and selected using 10 μg/ml blasticidin. Cells were plated in 10 cm cell culture dishes at 60–70% confluency and the following day treated with R1881 or transfected with chemical siRNAs targeting both alpha subunits of AMPK or CAMKK2 (see siRNA section for sequences). After 48 h, cells were washed with ice cold PBS and lysed using RIPA lysis buffer (described above). Protein levels were detected via Bradford Assay (Bio-Rad) and 2 mg of each cell lysate was incubated with magnetic Dynabeads Protein G (Cat# 10003D, Invitrogen, Waltham, MA, USA) conjugated to either V5 (Cat# R960–25, Thermo Fisher Scientific, RRID:AB_2556564) antibody or mouse control IgG (Cat# sc-2025, Santa Cruz Biotechnology, RRID:AB_737182) for 2 h at room temperature. After three washes with PBS + Tween 0.02% using the Bio-Rad DynaMag2 magnet, samples were moved to a new tube, the supernatant removed via the magnet and 40 μl of Laemmli buffer (2x) was added. Samples were then boiled at 95°C for 6 min and stored at −20°C. Changes in ATGL phosphorylation at S404 were detected using the newly created pS404 ATGL antibody (Abclonal, Woburn, MA) as a primary antibody and HRP-conjugated secondary antibody. Membranes were stripped and then total levels of the protein were detected using the V5 antibody.

PNPLA2/ATGL immunoprecipitation/mass spectrometry

The immunoprecipitation (IP) and mass spectrometry (MS) were carried out as described previously (19) with slight modifications. The protein lysate was incubated with 5 μg of anti V5 antibody (Abcam ab15828) followed by incubation with 20 μl Dynabeads Protein A beads (Invitrogen 10002D) for 1 h at 4°C. The protein complex was eluted from the beads using NuPAGE^™ LDS Sample Buffer (Invitrogen NP0007) and heated at 95°C for 10 min. In-gel digestion was performed using trypsin protease and the entire lane was processed to generate two pools of peptides. The LC-MS/MS analysis was carried out on a nanoLC1200 system coupled to Orbitrap Fusion Lumos ETD (Thermo Fisher Scientific, San Jose, CA) mass spectrometer. The peptide elution was done using a 110 min discontinuous gradient of 90% acetonitrile buffer (B) in 0.1% formic acid at 200 nl/min (2–30% B: 86 min, 30–60% B: 7 min, 60–90% B: 7 min, 90–50% B: 10 min). The peptides were ionized at positive spray voltage- 2.4kV and ion transfer tube temperature was 320°C. The mass spectrometer was operated in the data-dependent acquisition mode with full MS scan (precursor) in Orbitrap (120,000 resolution, 300–1500 m/z, AGC target 3E5, 50 ms injection time). For each MS/MS scan (fragment), the top 30 most intense precursor ions were analyzed in IonTrap (HCD 32%, AGC 2E4, 30 ms ion injection) with 15 sec dynamic exclusion time. The MS raw files were searched using Proteome Discoverer 2.0 software (Thermo Fisher) with Mascot algorithm (Mascot 2.5, Matrix Science). The data was searched against the homo sapiens protein database from NCBI refseq [updated 2020_03_24]. Dynamic modification of Oxidation, protein N-terminal Acetylation, Deamidation on asparagine and glutamine was allowed. The precursor mass tolerance was confined within 20 ppm with fragment mass tolerance of 0.5 dalton and a maximum of two missed cleavage with trypsin enzyme was allowed. The peptides identified from mascot result file were validated with 5% false discover rate (FDR). The gene product inference and quantification were done with label-free iBAQ approach using ‘gpGrouper’ algorithm. The differentially expressed proteins were calculated using the moderated t-test to calculate p-values and log2 fold changes in the R package limma. For statistical assessment, missing value imputation was employed through sampling a normal distribution N(μ−1.8 σ, 0.8 σ), where μ, σ are the mean and standard deviation of the quantified values. The list of enriched proteins plotted on the heat map were generated using following criteria: protein should be missing in ‘parental’ and only present in ‘WT’ or ‘S404A’ genotype with at least 1 unique peptide or enriched by at least 100-fold over parental (log2 fold change >6.64). A summary of the complete normalized IP/MS data can be found in Supplemental File 2.

3D invasion assays

Cells were trypsinized and seeded at a density of 100,000 (C4–2) or 50,000 (RM-9) cells per well in the 5D spherical plate (Kugelmeiers, Zurich, Switzerland) and allowed to cluster in RPMI 1640 (C4–2) (Cat# 22400105, Gibco, Waltham, MA) with 10% (v/v) FBS, 1x penicillin/streptomycin or DMEM + glutamine (RM-9) in a 37°C in a humidified 5% (v/v) CO₂ atmosphere. After 24 h, spheroids were transferred to 1.75 mL centrifuge tubes and allowed to settle to the bottom of the tube via gravity. Media was removed and cells were resuspended in Glycosil (Advanced Biomatrix, Carlsbad, CA, USA; Cat# GS222) reconstituted in degassed water per the manufacturer’s instructions to 10 mg/mL. Peptides KGGGPQG;IWGQGK (PQ peptide) and GRGDS (RGD peptide) were purchased from Genescript USA Inc. (Piscataway, NJ, USA) and reacted as previously described(20). RGD at 73.7 mg/mL was added to the Glycosil spheroid suspension and then mixed into solution with a pipette with a cut off end. The combined Glycosil-RGD mixture was adjusted to pH 7.8 with 1M NaOH. PQ at 11.243 mg/mL was added and then mixed as previously described (20,21). The solution was allowed to begin crosslinking for 5 min prior pipetting in 50 μL pucks cast in PDMS molds and allowed to crosslink in the incubator. After 1 h, hydrogels were removed from the PDMS molds and cultured in 1 mL of media (C4–2: RPMI (no glucose) + 5 mM glucose + 5% FBS, RM-9: RPMI (no glucose) + 5 mM glucose + 0.5% FBS). RM-9 cells were treated with either atglistatin (40 μM) or DMSO (control) in assay media, replenished on days 3 and 5. C4–2 cells were assayed after 14 days and RM-9 cells were assayed after 5 days using live/dead assay reagents (4 μM EthD-1, 2 μM calcein AM, and 1 μg/mL Hoechst 33342 in PBS). Hydrogels laden with clusters were incubated in the live/dead reagents for 15 min prior to imaging on the Nikon A1R MP+. Images were acquired with .825 μm steps through the z-stack at 20x. The resonant scanner was used with 2x line average/integrate count and a pinhole size of 24.27 μm. Bitplane IMARIS 9.2.1 was used to analyze acquired z-stacks. C4–2 images were quantified using the spots feature (number of Hoechst positive nuclei) and surfaces (total encompassing cluster volume of calcein AM) to calculate number of cells per cluster (number of Hoechst-positive nuclei within the encompassing calcein AM volume). Invadopodia were manually counted, and final numbers were reported as invadopodia/cell. “Invadopodia” were defined as thin cellular processes extending outward from a cell cluster that were clear enough to be easily identified by eye. RM-9 z-stacks were acquired using the above acquisition settings then uploaded to IMARIS and counted manually. Cell counts include only those cells that contain a blue nucleus and green cytoplasm and were reported as the number of escaped cells per cluster. Both experiments are reported with three clusters per hydrogel done in three experimental replicates, n=9 clusters.

Shotgun lipidomics

C4–2 ATGL KO and addback cell lines were plated to 70% confluency in 10 cm cell culture plates and cells were harvested by washing twice with ice cold PBS, cell number and viability was measured and then diluted in PBS before snap freezing in liquid nitrogen. The frozen cells were then shipped to be analyzed by Lipotype GMBH, Dresden, Germany. Samples for shotgun lipidomics were isolated, injected, and high-resolution Orbitrap mass analysis including internal lipid-class specific standards were performed. Lipid classes were identified via the LipotypeXplorer (Lipotype GMBH, Dresden, Germany) software. Data is presented in mol%. The nomenclature was as following: glycerophosphoplids subclass, the total length of fatty acids:number of saturated fatty acids:number of hydroxyl groups (eg PC 36:2:0). If no hydroxyl groups were detected for the subclass, we labeled the lipid with the fatty acid length and number of saturated fatty acids (ex. PC 36:2). A list of the analyzed lipid classes is provided in Supplementary Table 4. A list of the equipment and software used for shotgun lipidomics is provided in Supplementary Table 5.

DESI-MS imaging

Xenograft tumor tissues were stored at −80°C prior to analysis. Frozen tissues were sectioned at 12 μm thickness, thaw-mounted onto glass slides, and immediately analyzed using a Q Exactive HF Orbitrap mass spectrometer (Thermo Fisher Scientific) coupled to a 2D OmniSpray stage (Prosolia Inc.) and equipped with a home-built DESI sprayer. DESI-MS imaging was performed on consecutive sections from each sample in both the positive and negative ionization modes using a mass resolving power of 70,000 and a spatial resolution of 200 μm. A histologically compatible solvent system comprised of 98% methanol was used as the DESI spray solvent at a flow rate of 5.0 μL/min. Ion images were assembled and visualized using Firefly (Prosolia, Inc.) and BioMap (Novartis) software. For ion identification, tandem mass spectrometry experiments were performed using HCD and high mass accuracy measurements. After DESI-MS imaging, the analyzed tissue sections were stained with H&E and annotated to delineate histopathological regions of interest. Mass spectral data was extracted from pixels corresponding spatially to tumor regions within the DESI-MS images using MSiReader software. Per-pixel ion intensities for selected metabolite and lipid species were normalized by the total ion count and averaged across each sample for subsequent statistical analyses.

[11C]-choline synthesis and uptake assay

[11C]-choline was synthesized under GMP-compliance, pyrogen-free and sterile conditions with a radiochemical purity of >97% by the Cyclotron Radiochemistry Facility (CRF) at MD Anderson Cancer Center (MDACC). Cellular uptake and cold blocking of [11C]-choline were studied in C4–2 scramble, ATGL KO, WT addback, S404A and S47A cell lines. For cell uptake, 5 × 10⁴ cells were plated using RPMI-1640 with 2.5 mM glucose and 5% dialyzed FBS in 24-well plates. Cell media was replaced with media without FBS and then [11C]-choline (1 μCi) was added to each well and incubated at 37°C for 1 h. Experiments that abrogated metabolism were performed at 4°C, a temperature at which cellular membranes become less fluid and permeability is low. After the incubation, an aliquot (200 μl) of media was transferred to a tube for gamma counting and the excess media was discarded. Cells were then washed with PBS and lysed with 1% SDS, 10 mM sodium borate for 30 min at room temperature. Radioactivity in cell lysates and media were measured using a WIZARD2 2480 automatic gamma counter (PerkinElmer) and corrected for the half-life of tracer. Protein concentration of cell lysates were measured using the Pierce^™ BCA protein assay kit according to the manufacture’s protocol using bovine serum albumin (BSA) as the protein standard. Cell uptake of [11C]-choline was normalized to protein content and expressed as a tracer ratio ((cpm/mg protein)/(cpm/mL)).

Seahorse analysis

Agilent Seahorse 96-well cell assay plates were coated with Poly-L-lysine (0.1% in H₂O, Cat# P8920, Sigma, St. Louis, MO, USA) for 5 min, followed by two washes with PBS and dried. After sterilization via UV, cells were seeded (C4–2: 20,000 cells/well, RM-9: 10,000 cells/well) in RPMI 1640 no glucose media supplemented with dialyzed FBS (C4–2: 5%, RM-9: 0.5%) and glucose (RM-9, C4–2: 5 mM). 24 h after seeding, cells were treated as indicated and after 24 h of treatment Seahorse analysis were performed using the Xfe96 analyzer. The Glycolysis Stress assay was performed according to manufacturer’s manual using the following final concentrations per well: Glycolysis Stress assay RM-9 (glucose: 10 mM, oligomycin: 2 μM, 2-deoxyglucose (2DG): 50 mM), Glycolysis Stress assay C4–2 (glucose: 10 mM, oligomycin: 1 μM, 2DG: 50 mM). The Mito Stress assay, using C4–2 cell line derivates, was performed according to manufacturer’s manual using the following final concentrations per well: oligomycin 1.5 μM, rotenone + antimycin A 0.5 μM. Optimal FCCP concentration was determined for each cell line derivate (Scramble: 1 μM, KO 0.5 μM, KO + WT 0.5 μM). Glycolysis was quantified as the difference between basal ECAR measurement before glucose injection (non-glycolytic acidification) and maximal ECAR measurement after glucose injection but before oligomycin injection. Glycolytic capacity was quantified as the difference between basal ECAR measurement before glucose injection (non-glycolytic acidification) and maximum ECAR measurement after glucose and oligomycin injection. In the Mitostress assay, ATP production was quantified as the difference between the last OCR measurement before oligomycin injection and the lowest measurement after oligomycin injection. Spare capacity was calculated as the difference of basal respiration (OCR) and maximal respiration after addition of FCCP uncoupling agent. All Seahorse results were analyzed using WAVE 2.6.3 and ECAR/OCR was normalized to relative cell number using Hoechst stain as described in proliferation section. Results were graphed using Graph Pad 9.3.1.

RNA sequencing

Isogenic C4–2 PNPLA2 (ATGL) KO and add-back cell lines, including the ATGL wildtype (WT), the lipolytic-inactive S47A mutant, and the S404A mutant, were used for RNA sequencing (RNA-Seq). Total RNA was extracted to construct a cDNA library for RNA transcriptome sequencing using NEBNext^® Ultra^™ II Directional RNA Library Prep Kit for Illumina^® (Massachusetts, USA). RNA sequencing was performed by Admera Health (New Jersey, USA). The top 500 most variant genes in all samples were identified in the heatmaps. Differential expression analysis was conducted between conditions using edgeR (22). To assess the pattern of expression of biological activities and signaling pathways among the differentially expressed genes, pathway analysis was performed for select pathways including GSEA HALLMARK, Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, using clusterProfiler (23). Gene sets that were significantly enriched were determined by using adjusted p-value ≤0.05 as a cutoff.

Immunofluorescence microscopy

Isogenic C4–2 PNPLA2 (ATGL) KO and add-back cell lines (WT, S47A, or S404A) were seeded at 25,000 cells/well in 8-well chamber slides coated with poly-L-lysine in 500 μl RPMI 1640, 5% dialyzed FBS, and 5 mM glucose media mixed with 75 μM of BSA-coupled oleate for two days prior to immunofluorescence microscopy. Cells were then washed with PBS for 5 min. Each following step was accompanied by three 5-min washes. The cells were fixed in 4% paraformaldehyde at room temperature for 30 min and then permeabilized with 0.5% w/v saponin reconstituted in PBS at room temperature for 1 h. After blocking on shaker for 1 h in PBS with 10% goat serum, ATGL primary antibody (Cat#2138S, Cell Signaling Technology, 1:500) diluted in 200 μl PBS with 10% goat serum was applied to cells overnight at 4°C. Following this, cells were washed using PBS and incubated with a goat anti-rabbit IgG (H+L) cross-adsorbed secondary antibody, Alexa Fluor^™ 488 (#A-11008, Invitrogen^™, 1:200) on a shaker for 1 h at room temperature. After five 3-min washes, lipid droplets were stained using LipidTOX^™ Deep Red (Cat# H34477, Invitrogen^™, 1:200) in PBS for 1 h. Nuclei were simultaneously stained using Hoechst 33342 (Cat# H3570, Invitrogen^™, 1:2000) for 10 min. 1.5 mm coverslips were lowered onto the slides, mixed with drops of ProLong^™ Gold Antifade mounting media (Cat# P10144, Invitrogen^™). Cell-containing slides were left to dry overnight in the dark at room temperature and then examined using a Zeiss LSM880 laser-scanning confocal microscope. Colocalization between ATGL and lipid droplet signals was determined using IMARIS 8 (v8.1, Bitplane, Belfast, Northern Ireland, UK, RRID:SCR_007370).

Antibodies

A list of antibodies used in this study is provided in Supplementary Table 6.

Primers

A list of primers used in this study is provided in Supplementary Table 7.

Plasmids

pLenti PGK Neo DEST (w531–1) was a gift from Eric Campeau & Paul Kaufman (plasmid # 19067, Addgene, Watertown, MA, USA, RRID:Addgene_19067). pLX304 was a gift from David Root (plasmid # 25890, Addgene, Watertown, MA, USA, RRID:Addgene_25890). lentiCRISPRv2 (RRID:Addgene_52961) was a gift from Junjie Chen (UT MD Anderson Cancer Center). A list of plasmids used and/or created for this study is provided in Supplementary Table 8. All plasmids used in this study are available from the Frigo lab upon request.

Statistical analysis

Statistical analyses were performed using Prism Graph Pad Version 9.3.1. One-way ANOVAs with Dunnett or Tukey (DESI-MS only) post-hoc as well as t tests (unpaired, two tailed) were performed as indicated for each figure. For colocalization studies, both Pearson correlation coefficient and the Mander’s overlap coefficient were used to quantify the degree of colocalization between ATGL and lipids.

Synergy model calculations

CDI was calculated by the following formula: CDI = AB (AxB), where AB is the combination of drug treatment, A and B each of the drugs by itself (24). The CDI of x<1 means synergetic, x=1 means additive and x>1 means antagonistic. For the synergy model using cell viability, cells were plated in 96-well plates (C4–2: 5,000 cells/well, RM-9: 500 cells/well) in RPMI 1640 no glucose media supplemented with dialyzed FBS (C4–2: 5%, RM-9: 0.5%) and glucose (RM-9, C4–2: 5 mM). RM-9 cells were treated 24 h after seeding with atglistatin or AZ-PFKFB3–26 in various combination treatments. C4–2 cells were treated 24 h after seeding with NG-497 (Cat#10–4626, Focus Biomolecules) or AZ-PFKFB3–26 (Cat# 5675, R&D Systems) in various combination treatments. Cells were retreated every 48 h and cells were harvested after five days of treatment and cell viabilities were determined via CellTiter-Blue^® assays. Synergy was determined using SynergyFinder 2.0 (25). Settings were applied as follows. Readout: viability. Detect outliers: Yes, Baseline correction: Yes, Curve fitting: LL4, Calculate synergy: HSA.

Data availability

The data analyzed in this study described in Figure 1A and Supplementary Figure S1 were obtained from cBio Portal at https://www.cbioportal.org/. The immune-precipitation/mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier “PXD045042” and a summary of all hits is provided in Supplementary File 2. The raw RNA-Seq data files are available from GEO using the identifier “GSE243159”. The remaining data generated in this study are available upon request from the corresponding author.

Figure 1: ATGL mediates prostate cancer growth in vitro and in vivo.

(A) SU2C patient data demonstrates that the increased expression of PNPLA2, genetic amplification, or copy number gain correlates with decreased overall survival compared to unaltered or decreased expression, homozygous deletion, or copy number loss of PNPLA2 (<2). n = 444; results are shown using tumor samples that were subjected to probe capture or poly(A)+ selection RNA analysis. Logrank test. *P < 0.05. (B) Knockout of PNPLA2 (ATGL KO) in C4–2 affected proliferation under physiological conditions (2.5 mM glucose, 0.5% FBS). (C) Colony formation (CFA) of AR+ (C4–2, C4–2B-LT) and AR− (PC-3) CRPC models. Scramble, ATGL KO and addbacks of ATGL wildtype (WT) were grown in the presence of 2.5 mM glucose and quantified for % area of well occupied by formed colonies at endpoint. Images (left) and quantification (right) are representative results (n≥3). (D) Mice were castrated and subcutaneously injected with scramble control (n=5), ATGL KO (n=4), or ATGL KO + WT C4–2 addback cell line derivates (n=4). (E) Individual tumor volumes until day 57 of experiment (when first mouse was sacrificed due to tumor size) and (F) survival curve. (G-I) Representative images of tumor tissue. (G) H&E stain (H&E), (H) cleaved caspase 3 (CC3), (I) p-Histone H3 (pHH3). Kaplan-Meier curves analyzed by Logrank. Other data analyzed using one-way ANOVAs. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = no significance. Data are mean ± SEM (n=3 or as indicated).

Results

ATGL mediates prostate cancer growth in vitro and in vivo

To examine whether altered lipid droplet TG homeostasis could be contributing to prostate cancer progression, we first assessed whether genes involved in lipid droplet TG metabolism exhibited genetic or transcriptional traits that correlated with patient outcomes in a large cohort of men with mCRPC (Figure 1A; Supplementary Figure S1; n = 444; divided into two RNA-Seq experimental methods; probe-capture and poly(A)+ selection). Of the genes examined, only the levels of PNPLA2, encoding ATGL, correlated with worse overall survival. This was surprising because prior reports demonstrated that ABHD5 (also called CGI-158), a co-activator of ATGL in murine adipose tissue, can function as a tumor suppressor in prostate cancer (9), suggesting that perhaps ATGL could also function as a tumor suppressor. However, humans and mice with ABHD5 and PNPLA2 (encoding ATGL) genetic alterations present different phenotypes suggesting that ABHD5 and ATGL may possess divergent functions (26,27). Accordingly, PNPLA2 exhibited the genomic and transcriptomic traits of an oncogene, while ABHD5 levels were unaltered, in mCRPC tumors. Beyond PNPLA2, only two other genes, DGAT2 and MOGAT2 (both encoding enzymes involved in TG synthesis and located next to each other on chromosome 11q13.5) correlated with survival, with both exhibiting genomic and transcriptomic traits of tumor suppressors (Supplementary Figure S1). Given the small number of mCRPCs with low DGAT2 or MOGAT2, challenges with targeting tumor suppressors, and ATGL’s druggability (28–31), we focused on ATGL, the initial rate-limiting step of TG breakdown (Supplementary Figure S2A). Functionally, KO of PNPLA2 had minimal effects on CRPC cell growth when cells were grown in the presence of standard cell culture media conditions (11.1 mM glucose) (Figure 1B; Supplementary Figure S2B). However, when glucose and serum concentrations were lowered to mimic physiological levels found in the tumor microenvironment, ATGL KO impaired cell growth (Figure 1B). When we tested additional isogenic CRPC cell lines in colony formation assays, which introduces density-dependent metabolic stress (32), we observed that ATGL KO decreased colony formation in all cell models tested (C4–2, C4–2B-LT, PC-3) (Figure 1C; Supplementary Figure S2B). The impaired colony formation was rescued by ATGL WT addback (Figure 1C; Supplementary Figure S2B). Similar results were observed when we added exogenous oleate, a common monounsaturated fatty acid, to the media, indicating that the growth inhibition caused by ATGL KO could not be rescued by the simple addition of any neutral lipid (Supplementary Figure S3). Fluorescence-activated cell sorting (FACS) analysis of control, ATGL KO, and ATGL KO + WT addback cells indicated that in cell culture, ATGL KO decreased prostate cancer cell numbers primarily through increasing cell death/apoptosis (Supplementary Figures S4A–B).

To test ATGL’s role in tumor growth in vivo, we used a human xenograft mouse model of CRPC where we injected C4–2 scramble control, C4–2 ATGL KO and C4–2 ATGL KO + WT addback cells into castrated 7-week-old NOD scid gamma (NSG) mice (Figure 1D) and monitored tumor growth. ATGL KO decreased tumor growth and subsequently prolonged survival, which was reversed by the re-expression of ATGL WT (Figures 1E,F). IHC analyses of tumors revealed that ATGL KO increased apoptosis as detected via cleaved caspase-3 (CC3) and had a non-significant trend towards decreased proliferation as assessed by phosphorylated histone H3 (Ser10) (pHH3) levels (Figures 1G–I), effects that were reversed by the re-expression of ATGL WT. ATGL KO tumors were also noted to have large regions of necrosis that were not observed in control or ATGL WT addback tumors (Figure 1G; Supplementary Figure S4C).

ATGL increases the levels of malignancy-associated glycerophospholipids and promotes choline uptake

Our functional data confirmed a role for ATGL in prostate cancer in vitro and in vivo. We next evaluated ATGL’s role in prostate cancer lipid metabolism. We first tested the effect of ATGL knockdown on neutral lipid levels (assessed by Oil Red O staining) using chemical siRNAs targeting PNPLA2 in a variety of prostate cancer cell models (LNCaP (AR+ hormone-sensitive), C4–2 (AR+ CRPC derivative of LNCaP), C4–2B (more aggressive variant of AR+ C4–2), 22Rv1 (AR+ CRPC), PC-3 (AR− CRPC)), and non-transformed prostate epithelial cells (RWPE1) (Supplementary Figure S5A–B). We observed an increase in neutral lipid accumulation following ATGL knockdown in multiple prostate cancer cell models, consistent with a block in the breakdown of TGs. However, ATGL knockdown did not change neutral lipid levels in non-transformed RWPE-1 prostate epithelial cells. As expected, hormone-sensitive LNCaP cells also demonstrated an increase in neutral lipids upon androgen treatment, consistent with AR’s known role in de novo lipogenesis, which was further increased following ATGL knockdown (Supplementary Figure S5A–B).

To determine if the ATGL-mediated changes in metabolism observed in cell culture were maintained in vivo and better understand ATGL’s mechanism of action, we determined if ATGL impacts membrane composition in tumors, since the released fatty acids from lipid droplets can serve as precursors for glycerophospholipids, a major component of cell membranes (33). For this, we used mass spectrometry imaging (desorption electrospray ionization-mass spectrometry, DESI-MS) to spatially evaluate ATGL-mediated metabolic changes in xenograft tumors derived from the experiment described in Figures 1D–I (Figure 2A–B; Supplementary Figures S6–7). This approach provides in situ spatial information on tumor metabolism that can be precisely correlated with tissue histopathology. Hence, DESI-MS imaging enables correlation between lipid distribution in specific tissue regions composed of cancer cells – apart from stroma and other neighboring cell compartments that may confound data analysis. As expected, DESI-MS revealed a clear increase in the relative abundance of structurally diverse TG species following ATGL KO, an effect that was reversed by the re-expression of ATGL (Figure 2A–B; Supplementary Figures S6–7). Conversely, diglycerides (DGs) were generally decreased or unchanged by ATGL KO. We observed clear decreases of several enriched glycerophospholipids in the ATGL KO tumors including phosphatidylcholines (PC), phosphatidylserines (PS), phosphatidylglycerols (PG), and phosphatidylinositols (PI) (Figure 2A–B; Supplementary Figures S6–7). This included several lipid subspecies that were also enriched in patients in the membranes of malignant prostate tumors compared to benign tissue (34). One of the most prominent glycerophospholipid classes associated with prostate cancer is phosphatidylcholine. The most commonly found phosphatidylcholines associated with prostate cancer malignancy in patient samples (PC 32:1, PC 34:1, PC 36:2, PC 38:2) (34) were clearly regulated by ATGL in our isogenic preclinical models (Figure 2A–B). This correlated with ATGL-mediated changes in choline and acetylcholine levels (Supplementary Figures S6–7). Additional changes were observed following ATGL knockout in PI 38:2, PS 36:1, PG 34:1 and PE 34:1 (Figure 2A–B), which have all been associated with prostate cancer (34). DESI-MS was also able to detect ATGL-dependent changes in the relative abundances of signaling lipids derived from glycerophospholipids, including lysophosphatidylcholines, lysophosphatidylethanolamines, and lysophosphatidylglycerols (Supplementary Figure S7). Interestingly, the relative abundance of longer-chain monoglycerides and fatty acids were increased in scramble and ATGL WT tumors relative to ATGL KO tumors, whereas short-chain species were less abundant or unchanged (Supplementary Figure S7).

Figure 2: ATGL regulates glycerophospholipid levels in human CRPC xenografts.

(A) Representative DESI-MS images from human CRPC xenografts described in Figure 1. (B) Normalized ion intensities for malignancy-associated phosphatidylcholine (PC), phosphatidylinositol (PI), phosphatidylserine (PS), phosphatidylglycerol (PG), and phosphatidylethanolamine (PE) lipid subclasses detected in the human CRPC xenografts. Scramble (n = 5), KO (n = 4), KO + WT (n=4). One-way ANOVA and Tukey. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = no significance.

Our DESI-MS data indicates that ATGL regulates intracellular choline levels (Supplementary Figures S6–7). But most choline comes from the diet. As such, we tested whether ATGL knockout impacted the uptake of ¹¹C-choline, an FDA-approved positron emission tomography (PET) tracer used for the detection of recurrent prostate cancer. Interestingly, ATGL KO decreased choline uptake, an effect that could be reversed by ATGL re-expression, but not expression of a lipolytic-dead mutant of ATGL (Supplementary Figure S8). These data indicate that ATGL promotes the uptake of extracellular choline. Hence, ¹¹C-choline could have utility in being repurposed as a pharmacodynamic biomarker of ATGL activity.

Pharmacological or molecular targeting of ATGL inhibits growth in diverse preclinical models of prostate cancer

Our genetic data indicate that ATGL is required for prostate cancer cell proliferation, survival, colony formation, and tumorgenicity. As such, we next wanted to assess ATGL as a potential therapeutic target in advanced prostate cancer. To test this, we used the ATGL-specific inhibitor atglistatin (29). While atglistatin is highly selective for ATGL over other lipases, it is specific to murine ATGL and will not inhibit human ATGL (30,35). Therefore, we tested the efficacy of atglistatin in murine prostate cancer models. Consistent with our genetic and molecular data in human cells, atglistatin increased neutral lipids as assessed by Oil Red O staining in murine RM-9 prostate cancer cells (Figure 3A; Supplementary Figure S9A). In addition, atglistatin decreased prostate cancer cell growth and colony formation in a dose-dependent manner (Figures 3B–C). Glucose starvation again sensitized cells to ATGL inhibition in both cell growth and colony formation assays (Figures 3B–C). The impairment in cell growth appeared to be due to both decreased proliferation (decreased p-HH3 levels) and increased cell death as atglistatin increased cleaved PARP, a marker of apoptosis (Supplementary Figure S9B). The additional observation of impaired proliferation, which was not observed following genetic ablation of PNPLA2/ATGL (Supplementary Figure S4), suggests that atglistatin may possesses some off-target effects.

Figure 3: ATGL can be molecularly or pharmacologically targeted in diverse prostate cancer models.

(A) Murine prostate cancer cells (RM-9) stained with Oil Red O to determine the TG content upon overnight treatment with atglistatin (40 μM). (B) 7-day cell growth curve of RM-9 cells treated with atglistatin in high or low glucose. (C) CFA of RM-9 cells stably expressing doxycycline (DOX)-inducible shRNAs targeting scramble control (shControl) or Pnpla2 (shATGL) grown in variable glucose concentrations. (D) Oil Red O staining and (E) CFA, upon knockdown and/or treatment with 10 μM atglistatin. (F-K) Hi-MYC model-derived organoids grown in (F-H) typical organoid media containing high or (I-K) physiological glucose concentrations. (L) Knockdown of ATGL decreases C4–2 colony formation in a glucose-dependent manner. (M) Knockdown of ATGL in 22Rv1 cells decreases colony formation. One-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = no significance.

A previous report in lung cancer cells observed opposing effects after atglistatin treatment compared to knockdown of ATGL(6), suggesting the potential for off-target effects. To test for potential off-target effects, we created stable RM-9 cells in which we could silence ATGL in a doxycycline (DOX)-dependent manner (Supplementary Figure S9C). Both ATGL knockdown and atglistatin increased neutral lipid accumulation and decreased colony formation (Figures 3D–E; Supplementary Figures S9D–E). Moreover, the effects of ATGL knockdown and atglistatin appeared mostly redundant or at most additive, consistent with atglistatin’s effects being on-target (Figures 3D–E; Supplementary Figures S9D–E). As an additional control, doxycycline-mediated induction of scramble control (shControl) did not impact neutral lipid levels or colony formation, indicating that the phenotypes observed are not due to doxycycline-mediated off-target effects (Figure 3D; Supplementary Figures S9C, E). Notably, ATGL knockdown increased colony formation when cells were switched to supraphysiological levels of glucose (25 mM) often found in cell culture media (e.g., DMEM) (Supplementary Figure S9F), which may help explain prior discrepancies in the field. To test if ATGL could be targeted in an orthogonal, more physiological setting, we treated organoids derived from Hi-MYC genetically engineered mice with atglistatin and observed a dose-dependent decrease in organoid diameter and cell growth, the efficacy of which was again increased when glucose concentrations were lowered from standard high-glucose organoid media (17 mM glucose) to physiological (5 mM) levels (Figures 3F–K). This effect was specific to Hi-MYC-derived organoids since atglistatin did not impair the growth of TRAMP organoids (Supplementary Figure S10), suggesting that ATGL may be important for certain subtypes of prostate cancer such as those driven by MYC and thus, ATGL inhibition is not broadly toxic to all cells. Lipidomics performed on the organoids revealed that atglistatin again decreased the levels of malignancy-associated phosphatidylcholines (34) such as PC32:1 and PC34:1 when the media was adjusted to physiological glucose levels (Supplementary Figure S11).

The species specificity of atglistatin limited testing of this compound to mouse models. To test the efficacy of temporal ATGL inhibition in human cells, we created human prostate cancer cell lines containing DOX-inducible shRNAs targeting scramble control (shControl) or PNPLA2 (shATGL) (Supplementary Figures S12A–D, S13A–C). Inducible knockdown of ATGL, but not scramble control, impaired C4–2 and 22Rv1 CRPC colony formation, effects that were again magnified in low glucose conditions (Figures 3L–M, Supplementary Figure S12A–D). Interestingly, knockdown of ATGL in hormone-sensitive LNCaP cells had minimal effects on basal and androgen-mediated colony formation (Supplementary Figure S13A–C), consistent with a heightened role for ATGL in CRPC relative to hormone-sensitive prostate cancer (HSPC).

Inhibition of ATGL creates a therapeutically targetable metabolic shift towards glycolysis

Because ATGL inhibition decreased proliferation/colony formation, we wanted to further understand the contribution of ATGL to prostate cancer metabolism. ATGL’s effects on prostate cancer biology were influenced by glucose concentration (Figures 1 and 3). Previous studies have described a metabolic shift towards glycolysis upon ATGL inhibition (36) but also upon ATGL activation (37,38). Due to the conflicting reports in adipose and other non-prostate tissues, we wanted to assess how ATGL inhibition impacted glycolysis in prostate cancer cells. We first tested our C4–2 ATGL KO and addback cell lines using metabolic flux analysis (Seahorse^™ glycolysis stress test) and observed an increase in the extracellular acidification rate (ECAR), which correlates with glycolysis, in ATGL KO cells (Figures 4A–B). These changes were accompanied by modest increases in oxygen consumption rate (OCR), reflective of oxidative phosphorylation levels, following ATGL KO (Supplementary Figure S14A–F). Like in cell culture, ATGL KO increased intratumoral levels of glucose and lactate, indicating that blockade of ATGL caused a similar metabolic shift towards glycolysis in vivo (Figures 4C–D). Pharmacological inhibition of murine prostate cancer cells with atglistatin also increased ECAR, which could be reversed upon treatment with the glycolysis inhibitor AZ-PFKFB3–26 (39) (Figures 4E–F). Together, these results indicated that targeting ATGL in CRPC creates a metabolic shift towards glycolysis that we hypothesized could be therapeutically exploited. To test this idea, we co-treated prostate cancer cells with atglistatin and AZ-PFKFB3–26 and assessed the effects on colony formation (long-term treatment) (Figure 4G). Coefficient of drug interaction (CDI) scores were calculated to be 0.563 between treatments suggesting a synergistic effect (CDI values <1) between the two treatments (Figure 4G). We also performed short-term (4 days) treatment cell growth assays with increasing doses of atglistatin and AZ-PFKFB3–26, which resulted in a highest single agent (HSA) combined synergy score of 8.133 with maximal scores (>15) between 1–2 μM AZ-PFKFB3–26 and 40 μM atglistatin (Figures 4H; Supplementary Figure S15). Recently, the first inhibitor of human ATGL, NG-497, was described (31). We first verified that it could induce neutral lipid accumulation in human CRPC cells (Supplementary Figure S16). We then tested the efficacy of NG-497 in combination with AZ-PFKFB3–26 in human CRPC cells (Figure 4I; Supplementary Figure S15). Like observed for atglistatin in murine prostate cancer cells, NG-497 synergized with AZ-PFKFB3–26 to kill C4–2 cells, further validating this treatment approach. Thus, glycolysis inhibitors further sensitize CRPC cells to ATGL inhibitors and vice versa. As such, the combination of glycolytic and ATGL inhibitors represents a potential novel therapeutic approach that we propose warrants further investigation.

Figure 4: ATGL KO or inhibition creates a metabolic shift towards glycolysis that can be therapeutically exploited.

(A) ECAR of glycolytic stress assay (2DG, 2-deoxyglucose). (B) Glycolysis (ECAR measurement after glucose injection), and glycolytic capacity (maximum ECAR after oligomycin injection). (C-D) DESI-MS quantification of CRPC xenografts first shown in Figure 2A, and the same H&E images are presented. (E-F) Metabolic flux analysis of RM-9 cells treated for 24h with atglistatin (A) and the glycolytic inhibitor AZ-PFKFB3–26 (AZ-26; P). (E) ECAR of glycolytic stress assay which was used to quantify (F) glycolysis and glycolytic capacity. (G) CFA of RM-9 cells co-treated with atglistatin and AZ-PFKFB3–26. CDI, Coefficient of drug interaction. (H-I) Short term co-treatment of atglistatin (H) or NG-497 (I) in combination with AZ-PFKFB3–26 was used to calculate synergy in RM-9 (H) and C4–2 (I) cells. One-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = no significance.

ATGL is highly phosphorylated in human mCRPC at Ser404, a target of AR-CAMKK2-AMPK signaling

In search of regulatory mechanisms that could impact ATGL’s functions in prostate cancer, we were interested in prior work demonstrating that AMP-activated protein kinase (AMPK), a master regulator of cellular homeostasis that has context-dependent roles in cancer (40), could directly phosphorylate and increase the lipolytic activity of ATGL’s ortholog in murine adipose tissue (41,42). These findings were notable because of data demonstrating that a direct target gene of AR, CAMKK2 (calcium/calmodulin-dependent protein kinase kinase 2) (43), can promote prostate cancer cell proliferation, survival, and migration in part through the phosphorylation and activation of AMPK (44). Hence, we hypothesized that ATGL could be a pro-cancer, downstream effector of AR-CAMKK2-AMPK signaling in prostate cancer.

To identify potential AMPK targets that could play a role in prostate cancer progression in an unbiased manner, we mined clinical phosphoproteomic data(10) using an improved AMPK substrate motif (11) to detect changes in the phosphorylation of predicted AMPK targets across benign, HSPC and mCRPC (isolated via a rapid autopsy program (10)) disease states. Through the application of an updated position weight matrix (PWM), a mathematical algorithm based on the improved AMPK target phosphorylation motif (11), we could nominate high-confidence AMPK targets across patient groups. This resulted in the identification of classic AMPK targets, such as the acetyl-CoA carboxylase (ACC), as well as potential new targets of AMPK in prostate cancer (Figure 5; Supplementary Figure S17). The substrate that exhibited the greatest change was ATGL, which was highly phosphorylated in mCRPC compared to treatment-naïve, localized prostate cancer and/or benign prostate tissue (Figure 5A). The levels of ATGL phosphorylated at S404 exhibited a ~40-fold increase in mCRPC patient samples compared to primary HSPC and/or benign prostate tissue (Figure 5A), higher than any other candidate AMPK substrate (Supplementary File 1). Prior work in murine adipose tissue demonstrated that the phosphorylation at S406, thought to be the murine ortholog of human S404 (Figure 5B), stimulated ATGL’s lipolytic activity (41,42). While human ATGL demonstrates high sequence similarity with murine ATGL (Figure 5B), the phosphorylation of human ATGL by AMPK has not been studied in detail. Previous studies have expressed murine ATGL constructs in human cell lines to demonstrate its regulation by AMPK (42) or have reported on human ATGL’s phosphorylation at S404 by AMPK using a murine-specific p-ATGL S406 antibody (45). When we tested the specificity of a popular murine-specific p-S406 antibody to detect changes in human p-ATGL (S404) levels, this murine-targeting antibody could not detect human p-ATGL (Supplementary Figure S17A). Thus, to determine if human ATGL was targeted by AMPK in human prostate cancer cells, we created and validated our own highly specific p-ATGL S404 antibody (Supplementary Figure S17B). While highly specific, this antibody was not sensitive enough to detect endogenous p-ATGL. Hence, we expressed V5-tagged ATGL constructs to test the regulation of ATGL at S404 by the AR-CAMKK2-AMPK signaling axis in human CRPC C4–2 cells. We immunoprecipitated C-terminal V5-tagged human ATGL, since the C-terminal tagging of ATGL does not interfere with its activity (46). Phosphorylation of ATGL at S404 was decreased following RNAi-mediated knockdown of the catalytic subunits of AMPK (PRKAA1/2) (Figure 5C; Supplementary Figure S17C) or CAMKK2 (Figure 5D; Supplementary Figure S17D). We further confirmed that AMPK mediated the phosphorylation of ATGL(S404) using site-directed mutagenesis and a second antibody that broadly recognizes proteins and peptides bearing the canonical LXRXXS AMPK substrate motif (Supplementary Figure S17E). Treatment of C4–2 cells with the synthetic androgen R1881 increased p-ATGL (S404) levels (Figure 5E; Supplementary Figure S17F), consistent with CAMKK2-AMPK signaling being a known downstream effector of AR in prostate cancer (44). Taken together, these data indicate that ATGL is highly phosphorylated at S404 in mCRPC in patients and that this site is targeted by AR-CAMKK2-AMPK signaling.

Figure 5: ATGL is highly phosphorylated at S404 in human mCRPC and targeted by AR-CAMKK2-AMPK signaling.

(A) ATGL is highly phosphorylated at S404 in tumor samples from men with mCRPC (n=31) compared to benign (n=12) or primary, treatment-naïve hormone-sensitive prostate cancer (HSPC) (n=10). (B) Murine S406/human S404 ATGL site homology. (C-E) ATGL-V5 was overexpressed in C4–2 cells and then cells were treated with siRNAs targeting scramble control, the AMPK α catalytic subunits (PRKAA1/2) or CAMKK2 (siAMPK and siCAMKK2) or synthetic androgen (R1881) for 72h (validation of siRNA efficacy and androgen-mediated effects on CAMKK2-AMPK signaling are shown in Supplemental Fig. S9). Immunoprecipitated ATGL was tested for phosphorylation status using a developed p-ATGL S404 antibody (validation in Supplemental Fig. S17). Images are representative results of n≥3.

Phosphorylation of human ATGL on S404 does not alter ATGL’s lipolytic activity

To investigate how the phosphorylation of ATGL on S404 impacted prostate cancer cell lipid metabolism, we created a series of isogenic CRPC PNPLA2 (ATGL) knockout (KO) and addback cell lines, including a S404A mutant as well as a lipolytic-inactive S47A mutant (46) (Figure 6A). ATGL KO consistently increased intracellular neutral lipid levels, an effect that was reversed by the re-expression of ATGL wildtype (WT) but not lipolytic-inactive S47A mutant in three different CRPC cell lines, including the AR− CPRC cell model PC-3 (Figure 6B). There was no difference in neutral lipid levels when comparing the WT and S404A addbacks (Figure 6B).

Figure 6: Phosphorylation of S404 does not impact ATGL’s lipolytic activity.

(A) Immunoblot of ATGL KO and add-back cell models. (B) Quantified Oil Red O staining. Data are representative results of n≥3. (C-G) Shotgun lipidomics of C4–2 ATGL KO and addback cell models. (C) Heatmaps of triglycerides (TG), diglycerides (DG) or phosphatidylcholine (PC) of detected lipid species in mol%. (D) Sum of TG and DG detected. (E) TG and DG lipid species with high abundance in samples detected. (F) Phosphatidylcholine lipids associated with prostate cancer in patient samples (Butler et al. 2021). (G) Additional lipid subclasses associated with prostate cancer (Butler et al. 2021): Phosphatidylethanolamine (PE), Phosphatidylinositol (PI), Phosphatidylglycerol (PG), Phosphatidylserine (PS). n = 3. One-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = no significance.

Since Oil Red O staining broadly stains neutral lipids, we could not rule out effects that phosphorylation of ATGL on S404 could have on individual lipid species. To test this, we performed shotgun lipidomics on C4–2 ATGL KO and addback cell models (Figures 6C–G; Supplementary Figures S18A–B,S19,S20). As expected, we observed an increase in detected TGs and decreases in select DGs upon ATGL knockout that could be reversed by re-expression of ATGL WT but not the catalytically inactive S47A mutant (Figures 6C–E). We also detected an increase in cholesterol esters (CE; Supplementary Figures S18A–B,S19) in the KO and ATGL S47A mutant addback cells, often observed when TGs are accumulated in the lipid droplet(47). Similar to our in vivo DESI-MS data (Figure 2), we observed decreases in various glycerophospholipids in our ATGL KO and ATGL KO/S47A add-back cells models (Figures 6F,G; Supplementary Figures S18A–B,S19,S20). This again included the most commonly found phosphatidylcholines associated with prostate cancer malignancy in patient samples (PC 32:1, PC 34:1, PC 36:2, PC 38:2)(34) (Figures 6C,F). Additional glycerophospholipids found to be increased in malignant prostate cancer patient samples (34) (PI 36:1, PI 38:2, PS 36:1, PS38:2, PG 34:1, PG36:2, PE 34:1 and PE 36:2) likewise were regulated by ATGL (Figures 6C,G, Supplementary Figures S18A–B,S20). Notably however, phosphorylation of S404 did not influence ATGL’s canonical lipolytic activity, at least with regards to the lipids profiled in this study. Together, these data confirm that ATGL is a major regulator of TG catabolism and glycerophospholipid synthesis in prostate cancer cells but that phosphorylation of human ATGL at S404 does not influence ATGL’s canonical lipolytic activity.

Prostate cancer cell proliferation, migration, and invasion is mediated by both ATGL lipolytic activity and S404 phosphorylation

We were surprised that phosphorylation of ATGL on S404, which was elevated in mCRPC tumors (Figure 5), did not impact ATGL’s canonical lipolytic activity (Figure 6). However, prior work indicates that ATGL has additional functions beyond its role in TG lipolysis (48–50). Thus, we next tested whether S404 phosphorylation could still impact prostate cancer cell biology and how this compared to canonical ATGL lipolytic activity (assessed by the S47A lipolytic-dead mutation). As before (Figure 1), ATGL KO decreased prostate cancer colony formation (Figure 7A). In C4–2 and C4–2B-LT cells, this impaired growth effect could again be rescued by the re-expression of ATGL WT, but not lipolytic-dead (S47A) ATGL and was only partially rescued by the S404-mutated ATGL (Figure 7A). Interestingly, similar effects were observed in AR− PC-3 cells, suggesting that CAMKK2-AMPK signaling may be intact and promoting prostate cancer cell growth independent of AR status. These data are consistent with a prior report demonstrating a pro-cancer role for CAMKK2 in AR− DU-145 cells (51). To test the roles of CAMKK2 and AMPK in models of AR− prostate cancer cell growth, we used molecular knockdown (siRNA) or pharmacological inhibition (STO-609 and SGC-CAMKK2–1) and found that inhibition of CAMKK2-AMPK signaling suppressed the growth of AR− PC-3 cells as well as the established AR− neuroendocrine prostate cancer (NEPC) cell models MDA-PCa-144–13 and NCI-H660 (Supplementary Figure S21A–D).

Figure 7: Inhibition of ATGL S404 phosphorylation impairs CRPC cell proliferation, migration, and invasion.

(A) CFA of CRPC models grown in 5 mM glucose. Images (left) and quantification (right) are representative results of n≥3. (B) Migration (wound healing) analyzed over 24h. Data are expressed as mean % scratch wound closure ± SEM. Images (left) and quantification (right) are representative results of n≥3. (C) Schematic of 3D invasion assay with a matrix metalloproteinase (MMP)-sensitive crosslinker embedded in hydrogel. (D) Representative images of clusters analyzed for invasion. High indicates overexpression, low indicates close to basal expression. (E) Quantification of invasion as invadopodia/cell (n=5). One-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = no significance.

Because of the disparate effects of ATGL S404A mutation on lipolysis and colony formation, we sought to test the impact of S404A mutation on other prostate cancer cell processes regulated by AR-CAMKK2-AMPK such as migration and invasion (43). These processes were of additional interest given the link between lipid metabolism and prostate cancer metastasis (34,52–54). Molecular and pharmacological inhibition of ATGL decreased human and murine prostate cancer cell migration (Supplemental Figure S22A–D). ATGL KO decreased CRPC cell migration in wound healing assays, an effect that could be reversed by the re-expression of ATGL WT but not by ATGL S47A or S404A mutants in C4–2 and C4–2B-LT cells (Figures 7B). While the addition of exogenous oleate influenced the magnitude of effects on migration like what was observed with cell growth (Supplemental Figure S3), the overall trends remained similar, indicating that oleate alone is not sufficient to mediate ATGL’s effects (Supplementary Figure S23A–D). Like observed for colony formation assays, re-expression of ATGL S47A or S404A mutants could only partially reverse the effects of ATGL loss on migration in PC-3 cells, suggesting roles for ATGL lipolytic and p-S404-regulated activity in AR+ and AR− CRPC cell migration. Similar to prior work in AR+ prostate cancer cell models (43) and as observed in cell growth assays (Supplementary Figure S21A–D), knockdown of CAMKK2 and AMPK also decreased AR− cell migration (Supplementary Figure S24). To test ATGL’s role in invasion, prostate cancer cells were grown in 3D spheroids in the presence of 5 mM glucose while encapsulated in a well-established hydrogel model, simulating an extracellular tumor matrix environment and allowing us to monitor invasion by determining the formed invadopodia per cell (Figure 7C). ATGL KO decreased prostate cancer cell invasion, reflected by a decrease in the number of invadopodia, an effect that could be rescued by the overexpression of ATGL WT in a dose-dependent manner, but not the ATGL S404A mutant (Figures 7D,E). Together, these data indicate functional roles for ATGL and p-ATGL (S404) in CRPC cell growth, migration, and invasion. These data also suggest that both ATGL lipolytic and non-lipolytic activity promote prostate cancer progression.

Mutation of S404 has minimal effects on gene expression, de novo lipogenesis, and ATGL’s subcellular localization, but does change ATGL’s interactome

In search of a mechanism to better understand how phosphorylation of ATGL on S404 impacts prostate cancer cell biology, we performed RNA-Seq analysis on the ATGL control, KO, and add-back C4–2 cell models (Supplementary Figure S25A–F). Surprisingly, mutation of S404 (ATGL KO + ATGL S404A vs ATGL KO + ATGL WT) did not significantly alter gene expression. In fact, ATGL WT had modest effects on global transcription. These data suggested that phosphorylation of ATGL on S404 may instead mediate its biological effects more at the posttranslational level.

Since lipid breakdown is inhibited in ATGL KO cells, we next asked whether alteration of ATGL expression or mutation of S404 impacted de novo lipogenesis in prostate cancer cells. Using ³H-acetate tracing, it was determined that ATGL KO and/or mutation had negligible effects on de novo lipogenesis (Supplementary Figure S26). These data were consistent with the modest effects that ATGL had on genes involved in de novo lipogenesis (Supplementary Figure S25A–F).

Next, we tested if phosphorylation of S404 regulated ATGL’s subcellular localization. To do this, we used immunofluorescence confocal microscopy to assess the subcellular localization of ATGL in ATGL control, KO and ATGL add-back cell models. For this experiment, we had to add exogenous oleate (75 μM) to help visual intracellular lipid droplets. As expected, ATGL WT was mostly present on the surface of intracellular lipid droplets (Supplementary Figure S27). Mutation of amino acid 404 from serine to alanine did not significantly change ATGL’s subcellular localization, indicating that phosphorylation of S404 does not promote prostate cancer cell biology through changes in ATGL subcellular localization.

Using unbiased immunoprecipitation followed by mass spectrometry, we then tested whether mutation of S404 disrupted ATGL’s interaction(s) with any proteins. To do this, we created CRPC cells that stably expressed V5-tagged ATGL WT or V5-ATGL S404A. We then immunoprecipitated V5-tagged ATGL using anti-V5 antibodies and quantified levels of proteins in the V5-pulled down complexes (Supplementary Figure S28, Supplementary File 2). Several proteins were clearly complexed with ATGL/PNPLA2 including STIP1 and HILPDA. Notably, mutation of S404 decreased some ATGL protein-protein interactions (e.g., IRAK1, RMDN3). Surprisingly, mutation of S404 also increased ATGL’s association with a new set of proteins (e.g., NPDC1, RAB25). Collectively, these data suggest that phosphorylation of serine-404 can regulate a protein-protein interaction surface on ATGL and thus, provide insight into how this posttranslational modification could impact prostate cancer cell biology. Hence, future studies investigating the individual roles of each of these ATGL-binding proteins in prostate cancer are warranted.

Discussion

ATGL’s role in cancer has remained enigmatic due to conflicting reports across different cancer types. In lung cancer, ATGL has been described as a tumor suppressor based on the early observation that ATGL expression is reduced in lung cancer (7). This study reported that 25% of the Pnpla2/ATGL knockout mice spontaneously developed lung adenocarcinoma, while ATGL WT mice were cancer-free. Two additional in vitro studies suggested that the knockout of ATGL in lung cancer results in an increase in lung cancer cell migration and proliferation (36,55). However, only a single cell line was used for both studies and a prior study found that inhibition of ATGL in the same cell line decreased proliferation and migration (56). A more comprehensive study looked at the role of ATGL in different cancer cell lines, including lung, melanoma, colon, and liver (6). While ATGL overexpression had tumor suppressive functions, shRNA-mediated knockdown of ATGL did not have any effects in vitro or in vivo. Adding to the confusion, in pancreatic cancer, ATGL levels are increased in obese patients in both the cancer and surrounding stroma and associated with worse outcome (57). Another study found that ATGL has a pro-tumor role in hepatocellular carcinoma in preclinical models, since the knockdown of ATGL reduced tumor size, unless supplemented with free fatty acids, and overexpression of ATGL increased tumor volume (58). In prostate cancer, the ATGL coactivator CGI-58 has a tumor suppressive function (9,59). However, CGI-58’s role as a tumor suppressor was uncoupled from ATGL (9). Notably, CGI-58, which is known to activate murine ATGL was shown to only have a minor impact on human ATGL’s activity (6,60). Prior studies utilizing prostate cancer cell culture models found that ATGL knockdown increased neutral lipids but had only minor effects on proliferation compared to DGAT1 (8,61). We speculate that the modest effects of ATGL in these studies may be attributable to the high glucose concentrations used in standard cell culture media that don’t reflect physiological levels of glucose in the tumor microenvironment. Finally, in contrast to the above findings, a previous study found a pro-cancer role for CGI-58 (8). Together, these data indicate potential context-dependent roles for ATGL that might be influenced by cancer type, disease stage, and/or experimental conditions. To the latter, our data suggests that using culture conditions that better mimic the nutrient-challenged tumor microenvironment is critical to assessing ATGL’s role in cancer. Our findings also raise the question of whether proteins such as ATGL have been overlooked in large CRISPR or RNAi loss-of-function screens due to the use of high glucose- and serum-containing media.

The data presented here provide additional insights into the regulation ATGL. ATGL is highly phosphorylated at S404 in mCRPC patient samples (Figure 5). The orthologous phosphorylation site was previously described in murine adipose tissue to be phosphorylated by AMPK, leading to an increase in its lipolytic activity (42). Since there is 87.5% homology between murine and human ATGL, it was largely assumed that this lipolytic-stimulating event would be conserved between species. However, this was never investigated in human cells using a human ATGL construct. Accordingly, murine-specific p-ATGL S404-targeted antibodies were used to determine changes in the phosphorylation of human ATGL. While we detected bands at the correct size for human p-ATGL using the most prominent, commercially available murine-specific p-ATGL (S406) antibody, siRNA-mediated knockdown of ATGL indicated that this antibody cannot recognize human p-ATGL (S404). We therefore created a new human-specific p-ATGL (S404) antibody and demonstrated that the AR-CAMKK2-AMPK axis phosphorylates this site in prostate cancer. To our knowledge, this is the first time that human ATGL has been shown to be phosphorylated by AMPK at S404 in human cells. In contrast to murine ATGL, we found that the phosphorylation by AMPK at S404 had negligible effects on ATGL’s lipolytic activity. By comparison, expression of ATGL appears to have a more prominent role in lipid metabolism. This is in line with previous findings investigating ATGL in non-adipose tissue, where the overexpression of human wildtype, but not lipolytically inactive (S47A), ATGL, decreased lipid size, suggesting that total ATGL activity may correlate best with ATGL expression (46). However, this study did not investigate the S404 site and its impact on ATGL lipolytic activity. Interestingly, beyond ATGL KO and expression of a catalytically inactive mutant of ATGL (S47A), CRPC cells expressing the ATGL (S404A) mutant exhibited decreased colony formation and invasion/migration compared to ATGL WT-expressing cells. These data suggest that the S404A mutation disrupts an important function of ATGL that is independent of its function in the breakdown of triglycerides. Our unbiased IP/MS data provided initial evidence indicating that phosphorylation of S404 likely impacts ATGL protein-protein interactions (Supplementary Figure S28, Supplementary File 2). Not only did mutation of S404 disrupt existing ATGL interactions, surprisingly, it also increased ATGL’s association with an independent set of proteins. Defining the roles of each of these newly described ATGL protein-protein interactions is an ongoing area of investigation. In addition to AR+ CRPC models, we also confirmed ATGL-dependent effects in AR− prostate cancer cell models. Likewise, CAMKK2 and AMPK inhibition also decreased AR− prostate cancer cell growth and migration, indicating that CAMKK2-AMPK signaling could be activated in prostate cancer cells independent of AR. Of note, disruption of ATGL had no effect on intracellular neutral lipid levels in non-transformed prostate epithelial cells and colony formation in hormone-sensitive prostate cancer cells, indicating that ATGL’s pro-cancer roles are more specific to advanced prostate cancer (Supplementary Figures S5,13). While ATGL has not been reported to have lipolytic specificity for TGs, we cannot rule out that this would never happen (e.g., perhaps modification of ATGL or other factors alters the access and/or availability of certain substrate pools). Further, additional functions of ATGL were described when it was first discovered including CoA-independent acyl-glycerol transacylase activity (two monoglyceride to one diglyceride or one monoglyceride and one diglyceride to one triglyceride) and phospholipase A₂ activity (48,62). However, these additional enzymatic activities have been poorly studied. ATGL has also been described to play a role in autophagy/lipophagy (49). Given the previous reports on the role of autophagy in CRPC (63), it would be of interest to test if ATGL plays a role in this process, and if the S404A mutant interferes with this function. Recently, a novel non-lypolytic function of ATGL was described to facilitate the biosynthesis of branched fatty acid (FA) esters of hydroxy FAs (50). Our lipidomic platforms did not explore these lipid species, but certainly this is an area of interest for future studies. A major challenge in the study of ATGL structure-function is that there is currently no published ATGL structure (28). Further, structure prediction models such as AlphaFold (64) cannot predict the structure of ATGL, and in particular the C-terminus where the S404 site is located. Most understanding about the role of ATGL’s C-terminus comes from mutational studies based on isolated murine and human ATGL, which suggest that mutations could impair its activity (65). Therefore, it is challenging to determine if the ATGL S404A mutant would lead to changes in other protein binding partners based on our current knowledge about this protein’s structure. A caveat of our work is that we did not examine any additional phosphorylation sites of ATGL that could, for example, interfere with its location as suggested in murine ATGL (66). We also did not test additional upstream kinases of ATGL that have been reported to phosphorylate murine ATGL at S406, such as PKA (67).

ATGL inhibition increased glycolysis in vitro and in vivo (Figure 4). This phenotype has been observed in lung cancer cells (36). This finding was important in the context of prostate cancer because while many types of cancer have been characterized as highly glycolytic, this is not the case for most prostate cancers (68). Our data indicate that the co-targeting of ATGL-mediated lipid metabolism and glycolysis could represent a novel approach to overcome the metabolic plasticity of mCRPC, a disease stage that currently has no cure.

One of the major concerns in the systemic targeting of ATGL is its potential impact on the heart. Systemic deletion of Pnpla2 leads to premature lethality unless ATGL expression is rescued in the heart (69–73). However, systemic targeting of ATGL in mice did not show any signs of cardiomyopathy during long-term treatment with atglistatin (30). While more studies are needed to evaluate the potential side effects of targeting ATGL systemically in the context of prostate cancer, it also still needs to be determined if enough of the inhibitor would reach the tumor. In our preclinical models, concentrations of atglistatin lower than those used in adipocytes to block ATGL lipolysis (80 μM) were sufficient to decrease prostate cancer colony formation, migration, invasion, and organoid growth (Figure 3, Supplementary Figures S9,S22). Whether an inhibitor of ATGL can safely block human prostate cancer progression in vivo remains to be tested. While preparing this manuscript, the first human-specific ATGL inhibitor, NG-497, was published (31), allowing further investigation for the potential treatment of mCRPC. Like atglistatin in murine prostate cancer cells, NG-497 synergistically killed human C4–2 CRPC cells in culture when combined with an inhibitor of glycolysis (Figure 4), underscoring ATGL’s potential as a therapeutic target in prostate cancer. To our knowledge, however, NG-497 has not yet been shown to have activity in vivo.

In this study we demonstrate clear pro-cancer roles for ATGL in advanced prostate cancer using multiple murine and human 2D and 3D models, as well as human xenograft mouse models. Moreover, early studies in mice have demonstrated that the systemic targeting of ATGL can counteract models of induced insulin resistance(30), a known co-morbidity of anti-AR therapy (74,75). As such, ATGL inhibitors could have dual benefit, targeting both the cancer itself and standard of care treatment-associated comorbidities.

Significance.

ATGL promotes prostate cancer metabolic plasticity and progression through both lipase-dependent and lipase-independent activity, informing strategies to target ATGL and lipid metabolism for cancer treatment.