The metabolic network of sphingolipids plays important roles in cancer biology. Prominent sphingolipids include ceramides and sphingosine-1-phosphate that regulate multiple aspects of growth, apoptosis, and cellular signaling. Although a significant number of enzymatic regulators of the sphingolipid pathway have been described in detail, many remained poorly characterized. Here we applied a patient-derived systemic approach to identify and molecularly define progestin and adipoQ receptor family member IV (PAQR4) as a Golgi-localized ceramidase. PAQR4 was approximately 5-fold upregulated in breast cancer compared with matched control tissue and its overexpression correlated with disease-specific survival rates in breast cancer. Induction of PAQR4 in breast tumors was found to be subtype-independent and correlated with increased ceramidase activity. These findings establish PAQR4 as Golgi-localized ceramidase required for cellular growth in breast cancer.
Induction of and cellular dependency on de novo sphingolipid synthesis via PAQR4 highlights a central vulnerability in breast cancer that may serve as a viable therapeutic target.
Sphingolipids are key regulatory bioactive molecules that play important roles in cancer biology (1). Altered sphingolipid metabolism are linked to several cancer types including liver, colon (2), endometrial (3), and breast cancer (4, 5). The levels of sphingolipid pathway metabolites, including ceramides and sphingosine-1-phosphate (S1P), are significantly deregulated in breast tumor tissue compared with normal tissue (4, 5), and elevated tumor ceramide levels have been shown to be associated with higher tumor grades (4, 6). The conversion of ceramides to sphingosine and subsequent phosphorylation generates S1P. The balance between the intracellular levels of proapoptotic ceramide and prosurvival S1P determines the fate of cancer cells (7). By mediating apoptosis, growth arrest, and senescence, ceramides function as tumor-suppressor lipids, whereas S1P is a key tumor-promoting lipid that enhances cell proliferation, migration, and angiogenesis (8–10).
In this study, we identify PAQR4 as a Golgi-localized ceramidase that is highly expressed in breast cancer tissues and whose expression is required for tumor growth. We find that enhanced PAQR4 expression endows cancer cells with a dual selective advantage through the combined effects of lowering cytotoxic ceramides and generating S1P.
Materials and Methods
MDA-MB-231, HCC1806, MCF7, and T-47D cells were purchased from ATCC and the murine cell line EO771 from Ch3Biosystems. Ishikawa cells were kindly provided by Dr. C. Krakstad. H293T, SUM149, and MCF10A cells were a kind gift from Dr. J. Lorens. Cells were maintained using standard tissue culture procedures, according to the manufacturer's instructions, and grown at 37°C with 5% CO2 and atmospheric oxygen.
Genetic fingerprinting and short tandem repeat profiling cell line authentication of MDA231, MCF7, and MCF10A was performed by Eurofins Genomics laboratory.
Patient tumor samples
Survival analysis was performed on a quartile normalized cDNA microarray data from 203 breast patient's tumor samples based on Illumina HumanHT-12v4 Expression platform. The samples were treatment naïve (11). The study was approved by the regional committees for medical and health research of Western Norway (REK-Vest; approval number 273/96-82.96).
Female breast cancer patient specimens, used for ceramidase assays, were obtained at the Department of Surgery (Haukeland University Hospital, Bergen, Norway) under written informed consent, and anonymized. The tumor and normal tissue samples were matched (N = 7 pairs), meaning that the tumor and normal sample were collected from the same patient during mastectomy (12). The matched normal tissue was collected from a region of the breast with a clear physical separation from the malignant lesion.
Animal experiments were approved by the Norwegian Animal Research Authority and conducted according to the European Convention for the Protection of Vertebrates Used for Scientific Purposes, Norway. The Animal Care and Use Programs at University of Bergen are accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. The laboratory animal facility at University of Bergen was used for the housing and care of all mice.
Female NOD-SCID gamma (005557, RRID:IMSR_JAX:005557) and C57BL/6J (000664, RRID: IMSR_JAX:000664) mice were purchased from Jackson laboratories. Mice were kept in IVC-II cages (Sealsafe IVC Blue Line 1284L, Tecniplast). For both strains, 5–6 mice were housed together and maintained under standard housing conditions at 21°C ± 0.5°C, 55% ± 5% humidity, and 12-hour artificial light-dark cycle. Mice were provided with standard rodent chow (Special Diet Services RM1, 801151, Scanbur BK) and water ab libitium.
Syngeneic and xenograft model
PAQR4-depleted cells were orthotopically implanted in the inguinal mammary fat pad of 8 (C57BL/6J) or 6 (NOD-SCID) weeks old female mice. A total of 5 × 104 viable EO771 or 5 × 105 viable MDA-MB-231 cells in PBS were mixed 1:1 by volume with Matrigel (Corning, 356231) and injected in a total volume of 50 μL. At endpoint, the mice were humanely euthanized, and tumors harvested for weight measurements and histology.
Total transcriptomics profiles of paired normal and tumor tissues breast invasive patients with carcinoma were obtained from The Cancer Genome Atlas (v1.5.2 TCGA, n = 111) and Gene Expression Omnibus (GSE70947, n = 147). To identify the differentially expressed genes (DEG, with cut-off adjusted P value 0.01 and log2 tumor/normal fold-change expression ±1.5), a supervised paired sample t test was implemented using limma Bioconductor package in R (13). To investigate the DEGs involved in sphingolipid metabolism, we compiled a gene set containing 203 genes involved in the sphingolipid and ceramide biosynthesis, catabolism, and signaling pathways based on the Molecular Signatures Database (v6.1 MsigDB; ref. 14). As adiponectin receptor 1 and 2 (AdipoR1 and AdipoR2) are known as ceramidases, all 11 genes in the progestin and adipoQ receptor (PAQR) family were included in the gene set.
Gene set enrichment analysis
Functional analysis of the DEGs comparing the transcriptomics of the paired normal and tumor tissues from TCGA was performed using Gene Set Enrichment Analysis (GSEA) software (14). The significant enriched gene sets were selected as FDR < 0.25.
Disease-specific survival (DSS) and relapse-free survival (RFS) were calculated with GraphPad Prism software using log-rank test. Patients with 25% quartile of PAQR4 expression were considered as low PAQR4 group and patients with ≥25% quartile were classified as patients with highly expressed PAQR4 tumors. In addition, Gene Expression–Based Outcome for Breast Cancer Online (GOBO, v.1.0.3; ref. 15) was used to graph the distant metastasis-free survival based on PAQR4 expression in 1,176 available breast tumor.
Structural prediction and molecular dynamic simulation
PAQR2 and PAQR4 amino acid sequence alignment was performed using Clustal Omega. Swiss-Model (16) was used to prepare a homology-model of PAQR4, based on the structure of AdipoR2 (PDB ID: 5LXA). The PAQR4 model was embedded in a lipid bilayer using the Charmm membrane builder (17) and subjected to molecular dynamics simulations with the Amber molecular modeling package as described below.
Simulations of PAQR4 in complex with ceramide were based on docking of ceramide to the homology model of PAQR4. The PAQR4-ceramide complex was embedded in an equilibrated bilayer of POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine). The system was solvated in a water box encompassing the protein with extensions in the XY-dimensions identical to the lipid bilayer. Starting coordinates for simulations of PAQR4 with the three zinc-coordinating histidine residues mutated to alanine were prepared by removing zinc and modifying histidine to alanine in the coordinate file. Simulations were performed using the GPU-accelerated Particle Mesh Ewald Molecular Dynamics (PMEMD) module (18) implemented in the AMBER18 molecular dynamics software package (19), applying the ff14SB (20) and lipid14 force fields (21). Periodic boundary conditions with particle mesh Ewald summation (22) of electrostatic interactions were applied and van der Waals interactions were truncated with a 10 Å cutoff. SHAKE (23) was used to constrain bonds involving hydrogen atoms. After 10,000 steps of minimization, the systems were subjected to 5 ps of gradual constant volume heating (NVT) from 0 to 100 K followed by 100 ps gradual constant pressure heating (NPT) applying anisotropic Berendsen coupling (24) with reference pressure set to 1 bar. Temperature was regulated with the Langewin thermostat (25) gradually over 100 ps from 100 to 303 K. Weak restraints maintained by a force constant of 10 kcal mol−1 Å−2 were applied to the protein and lipids throughout both heating steps. The systems were finally simulated at constant pressure conditions (NPT) without restraints for 1,500 ns at 303 K and the resulting coordinates were saved every 100 ps for analysis.
Generation of knockdown and overexpression cell lines
Short hairpin RNA (shRNA) constructs targeting PAQR4 and scramble (shCtrl) were purchased from Sigma (Supplementary Table S1). Full-length PAQR4 (Isoform 1) was PCR amplified from cDNA of Ishikawa cells and cloned into the retroviral pBabe-puro vector (Addgene, catalog no. 1764) by conventional restriction enzyme-based cloning. Sanger DNA sequencing analysis confirmed successful cloning. Lentivirus (knockdown) and retrovirus (overexpression) production were performed as described previously (26). After selection, cells were allowed to recover for at least 48 hours before tested for overexpression or knockdown of PAQR4. Knockdown cells were never passaged for more than three passages after infection.
Site-directed mutagenesis was performed using the PCR-based QuickChange Lightning Multi Site-Directed Mutagenesis Kit (Agilent Technologies, 210515) according to the manufacturer's instructions. Primers are listed in the Supplementary Table S1. Sanger sequencing analysis was performed to confirm mutated PAQR4 products.
Total RNA extraction and cDNA synthesis were performed using total RNA purification Kit (Norgen Biotek, 37500) and Superscript III reverse transcriptase (Invitrogen, 18080051). qRT-PCR was performed using designed primers (Supplementary Table S1) and SYBR Green (Roche, 4707516001) on the LC480 instrument (Roche). The relative amount of cDNA was calculated by the ΔΔCt method using human Hypoxanthine-guanine phosphoribosyltransferase (hHPRT) and mouse actin (mActin) RNA expression as controls for human and mouse cell lines, respectively.
Proliferation of cells was determined in triplicates by high content imaging using the IncuCyte Zoom (Essen Bioscience) according to the manufacturer's instructions. In all experiments, four fields were imaged per well under 10× magnification every 2 hours for 3–5 days. The IncuCyte Zoom (v2018A) software was used to calculate confluency values.
Tissue IHC for PAQR4 and immunofluorescence for cleaved caspase-3
About 5-μm tissue sections were deparaffinized and antigen retrieval performed using a high pH buffer (Vector Laboratories, H-3301). Sections were blocked (Vector Laboratories, PK-4001) and incubated overnight at 4°C with PAQR4 primary antibody (Supplementary Table S1) diluted in 1% BSA in PBS. Following signal amplification by the Vectastain ABC reagent, the HRP signal was developed by incubating the sections with HRP substrate (Vector Laboratories, SK-4105). Counterstaining was performed by haematoxylin. The images were acquired using Hamamatsu slide scanner and Aperio ImageScope software.
For immunofluorescence of cleaved caspase-3, mouse tumor tissues were fixed in 4% paraformaldehyde (PFA) overnight and embedded in paraffin for sectioning. After deparaffinization and rehydration, antigen unmasking was performed in pH 6.0 (Vector Laboratories, H-3300). The sections were blocked in 4% goat serum in PBS with 1% BSA and stained with cleaved caspase-3 antibody. Images were acquired using the Leica SP5 confocal microscope. Cleaved caspase-3 staining was quantified as stained area per field (∼0.15 mm2) in five different fields per tumor (n = 5 per group).
Cell immunofluorescence and confocal microscopy
Cell staining for immunofluorescence was performed as described previously (27). The type, source, and dilution of antibodies are described in the Supplementary Table. Immunostained samples were imaged using Leica SP5 confocal microscope. Images for PAQR4 localization were acquired using Leica SP8 confocal microscope. Image analysis was performed using ImageJ Fiji. The surface-rendering tool in the Imaris 9.1.2 Bitplane software was used to generate Fig. 5A.
Apoptosis and cell-cycle assays were determined by the Annexin V assay (Thermo Fisher Scientific, A13201) and incubation of the cells with PI and RNase (BD Biosciences, 550825), respectively, according to the manufacturer's protocol. Flow cytometry was performed using the Accuri C6 (BD Biosciences) and data were analyzed with FlowJo software (Tree Star, Inc).
MDA-MB-231 and MCF7 cells were seeded in triplicates and allowed to adhere overnight. The following day the media were renewed with fresh culture media. For cells subjected to S1P measurements, the media were replaced with DMEM phenol-free media (Sigma, D1145) containing 0.02% FBS. Two days after plating, cells were washed once in PBS and harvested by scraping. Snap-frozen cell pellets were quantified by LC/MS-MS as described (28) by the VCU Lipidomics/Metabolomics Core. The measured sphingolipid levels were normalized to protein concentration as determined using Pierce Bicinchoninic Acid (BCA) Protein Assay Kit (Thermo Fischer Scientific, 23225).
Ceramidase activity assay and sphingolipid measurement
Tumor biopsies and frozen cell pellets were homogenized in 500 μL cold Dubelco PBS solution containing calcium and magnesium and EDTA-free protease inhibitor cocktail (Roche) with a mechanical tissue homogenizer (tissue biopsies) and tissue dismembrator probe (cell pellets, Thermo Fisher Scientific Model FB50, 8 short 1-second pulses at 15% of amplitude). After incubation of the samples on ice, 10 μL of an ethanolic mixture of deuterated ceramides-d7 was added to each sample [Ceramide d18:1-d7/16:0 (Cayman Chemicals, 22787, 4.9 μmol/L), Ceramide d18:1-d7/18:0 (Cayman Chemicals, 22788, 4.6 μmol/L), Ceramide d18:1-d7/24:1 (Avanti Polar Lipids, 10 μmol/L), Ceramide d18:1-d7/24:0 (Avanti Polar Lipids, 860679, 10 μmol/L)]. The homogenates were incubated at 37°C with continuous shaking for 3 hours and the reaction was quenched by adding 2 mL of organic extraction solvent (Isopropanol/Ethyl Acetate 1:2, vol/vol). Immediately afterward, 20 μL of organic internal standard solution was added (Ceramide/Sphingoid Internal Standard Mixture II diluted 1:10 in ethanol, Avanti Polar Lipids). After short-vortexing, a two-phase liquid–liquid extraction was performed. The organic phases were combined and dried under nitrogen stream at 40°C. In the case of tissue samples, the dried residue was reconstituted in 200 μL of methanol. The aqueous phase was dried in a SpeedVap solvent evaporation system and the dried residue was reconstituted in 500 μL RIPA buffer containing 5% TritonTM X-100 and total soluble protein content was determined by the BCA assay.
About 5 μL samples were injected into an LC/MS-MS system for the analysis of ceramides and sphingoid bases, 1 μL injection was required for the analysis of sphingomyelins. The system consisted of a Shimadzu LCMS-8050 triple quadrupole mass spectrometer with the dual ion source operating in electrospray positive ionization mode in the case of sphingoid bases and ceramides and in negative mode in the case of sphingomyelins. The mass spectrometer was coupled to a Shimadzu Nexera X2 UHPLC system equipped with three solvent delivery modules LC-30AD, three degassing units DGU-20A5R, an autosampler SIL-30ACMP and a column oven CTO-20AC operating at 40°C (Shimadzu Scientific Instruments). Analysis of sphingolipid species was achieved using selective reaction monitoring scan mode. Sphingoid lipid separation was achieved by reverse phase LC on a 2.1 (i.d.) × 150 mm Ascentis Express C8, 2.7 μm (Supelco) column under gradient elution, using three different mobile phases: eluent A consisting of methanol/water/formic acid, 600/400/0.8, vol/vol/vol with 5 mmol/L ammonium formate, eluent B consisting of methanol/formic acid, 1,000/0.8, vol/vol with 5 mmol/L ammonium formate, and eluent C consisting of CH3OH/CH2Cl2 350/650. The relative concentration of each metabolite was determined using the peak-area ratio of analyte versus corresponding internal standard. Data are reported as analyte peak area/internal standard peak area and normalized according to the protein content.
C6-NBD ceramidase assay
C6 ceramide-NBD (Thermo Fisher Scientific, N-22561) was prepared according to manufacturer's instruction. Cells grown on coverslips were washed twice in Hanks' Balanced Salt Solution (Sigma, H8264) supplemented with 10 mmol/L HEPES (HBSS/HEPES), before adding 5 μmol/L C6-NBD ceramide dissolved in HBSS/HEPES. The cells were then incubated at 4°C for 30 minutes and washed five times in ice cold HBSS/HEPES before incubated at 37°C for 30 minutes. Finally, cells were washed four times before fixation with 4% PFA for 15 minutes. For immunofluorescence staining of the Golgi, cells were permeabilized with 0.03 μg/mL digitonin for 10 minutes at 4°C and stained with RCAS1 as described above. Images were acquired using the Leica SP5 confocal microscope. ImageJ Fiji was used to measure the C6 ceramide-NBD signal in RCAS1-positive areas.
C6-ceramide treatment of cells
C6-ceramide/cholesteryl phosphocholine formulations were prepared following the manufacturer's instructions (Avanti Polar Lipids, 640001). For assessing CDK4 expression and cytochrome C release upon C6-ceramide treatment, cells were treated with 20 μmol/L (48 hours) and 5 μmol/L (overnight) C6-ceramide before collection/fixation of cells, respectively.
Cells were lysed in RIPA buffer containing EDTA-free protease inhibitor cocktail (Pierce RIPA buffer, Thermo Fisher Scientific, 89901; Sigma, 5892791001). Protein samples were immunoblotted using standard methods. Flag-tagged PAQR4 expression was detected using an anti-Flag antibody conjugated to HRP (Supplementary Table S1) for 1 hour at room temperature, whereas CDK4 expression was detected by incubating membranes overnight at 4°C with anti-CDK4 antibody. β-actin served as a loading control.
Quantification and statistical analysis
Statistical computation on the transcriptomics data and principle component analysis on clinical ceramidase assay data were performed in RStudio. Several R packages were used to prepare the data and analyze the RNA-seq. The in-house script is available upon request.
For the ceramidase assays (Fig. 4C and D) on the clinical patient samples, the results were analyzed by Wilcoxon matched-pair signed rank test using GraphPad Prism 7 software. For all the comparisons, Student two-tailed t test was used in GraphPad Prism 7 software. The significant symbols used in the figures are as *, P < 0.05; **, P < 0.01; ***, P < 0.001.
The sphingolipid metabolism–related gene PAQR4 is required for breast cancer cellular growth and negatively correlated with breast cancer patient survival
GSEA of DEGs between 111 paired tumor and nontumor tissue revealed an enrichment of genes in the ceramide signaling pathway (NES = 1.67; FDR = 0.075) and sphingolipid metabolism (NES = 1.62; FDR = 0.053) in tumor tissues (Fig. 1A; Supplementary Fig. S1A). To uncover key regulatory members of the sphingolipid pathway, we performed a supervised differential expression analysis of sphingolipid-related genes (203 genes) between matched tumor and nontumor tissue samples. We identified PAQR4 as the most upregulated gene transcript (fold change ∼5, Padj < 0.01; Fig. 1B). PAQR4 deregulation was breast cancer subtype-independent, as enhanced PAQR4 expression was detected in tumors irrespective of estrogen, progesterone receptor, and HER2 receptor profile in both the TCGA and a validation dataset (Supplementary Fig. S1B and S1C). Consistent with the transcriptomic analysis, detection of PAQR4 by IHC in breast cancer tissue sections showed enhanced staining intensity in the tumor compartment compared with the stromal/normal tissue areas (Fig. 1C). PAQR4 expression was further negatively correlated with patient RFS (log-rank P value = 0.0082; Fig. 1D) and DSS (log-rank P value = 0.0301; Fig. 1E) in an in-house dataset (11), as well as metastasis-free survival in an independent validation set (Supplementary Fig. S1D). These findings establish that PAQR4 expression is induced in tumor tissues and negatively correlates with patient survival.
To investigate the mechanism by which PAQR4 affect tumor cells, we next sought to determine the cellular phenotypes it governs. Depletion of endogenous PAQR4 in triple-negative (MDA-MB-231 and HCC1806) and estrogen receptor–positive (MCF7 and T47D) human breast cancer cell lines as well as the murine breast cancer cell line EO771, using lentiviral delivered shRNAs (Supplementary Fig. S2A), significantly reduced cellular growth as determined by high content imaging (Fig. 2A–C). Such conserved effect across tumor cell genotypes is consistent with the observed PAQR4 induction between breast cancer subtypes. In contrast to the findings of Zhang and colleagues (29), the observed reduction of cellular growth in PAQR4-depleted cells was independent of changes in cell-cycle status (Supplementary Fig. S2B). Consistent with the observed growth inhibition in vitro, PAQR4 depletion in human MDA-MB-231 and mouse EO771 triple-negative breast cancer cells abrogated orthotopic tumor growth in immune-compromised NOD-SCID and immunocompetent C57BL/6J mice, respectively (Fig. 2D). Finally, ectopic expression of PAQR4 in SUM149 cells increased the cellular growth (Fig. 2E; Supplementary Fig. S2C). Collectively, these data establish that the sphingolipid metabolism–related gene PAQR4 is required for breast cancer cell growth.
Homology modeling suggests that PAQR4 functions as a ceramidase
To investigate the mechanism by which PAQR4 regulates tumor growth, we next sought to determine the structural basis of its function. PAQR4 belongs to the PAQR protein family consisting of 11 members characterized by a 7-transmembrane domain. The close family member PAQR2 (also known as AdipoR2) is known to possess ceramidase activity and has been cocrystalized with oleic acid, that is, the byproduct of its enzymatic activity (30, 31). Within its catalytic site, a zinc ion coordinated by three histidine residues (His202, His348, and His352), is essential for catalysis. In concert with the zinc ion, an aspartic acid residue (Asp219) facilitates the cleavage of ceramide into sphingosine and a fatty acyl chain. Interestingly, these histidine residues and the aspartic acid are conserved in PAQR4 (His100, His240, His244, and Asp119; Supplementary Fig. S3A and S3B).
We therefore constructed a homology model of PAQR4 based on the resolved crystal structure of AdipoR2. We then did a structural superimposition of the AdipoR2 structure to our PAQR4 model to position oleic acid inside the channel of PAQR4 and performed a 100 ns molecular dynamics simulation of the complex embedded in a lipid bilayer to relax the structure. This model revealed conservation of a 7-transmembrane structure that forms a barrel domain with an amphipathic pore (Fig. 3A). The interactions of the fatty acid with PAQR4 were essentially the same as observed for AdipoR2. The carboxylic group of the fatty acyl locates at the proximity of the zinc coordination in PAQR4 (Fig. 3A and B). We then repeated the simulation by docking a C18:1-ceramide molecule in the PAQR4 model using the oleic acid as an anchor. Independent simulations over 1.5 μs showed that PAQR4, as was shown for PAQR2 (31), readily accommodated the fatty acyl part of the ceramide molecule inside the pore, with the cleavage site (amide bond) in close proximity to the active site zinc center (Fig. 3C; Movie S1). Interestingly, in both structures, the sphingosine moiety of the ceramide molecule is predicted to localize outside of the pore.
We further found that the homology model was remarkably stable over time (Fig. 3D). To better understand the role of the zinc molecule, we performed another simulation of C18:1-ceramide docked in PAQR4 wherein the three zinc-binding histidine residues were mutated to alanine, thereby excluding zinc from the structure. Interestingly, this simulation showed that both the fatty acid and the sphingosine arm of ceramide can be accommodated inside the amphipathic pore (Fig. 3E). Although the overall structure of mutated PAQR4 was stable (Fig. 3F), the more relaxed structure revealed that ceramide can interact in two different conformations. In sum, these modeling experiments suggest that PAQR4 possess ceramidase activity.
PAQR4 is a Golgi-associated ceramidase
To experimentally test whether PAQR4 harbors intrinsic ceramidase activity, we incubated deuterium-labeled ceramides with cellular lysates from control and PAQR4 knockdown cells. Following the incubation period, ceramidase activity was assessed by measuring the abundance of exogenous deuterated ceramide species by LC/MS-MS. Consistently, we found that lysates from PAQR4-depleted cells displayed reduced ceramidase activity when compared with control cells (Fig. 4A). We did not detect any PAQR4-related biases for specific ceramide fatty acyl chain length, as the levels of both ceramide species with long (C16:0 and C18:0) and very long fatty acyl chains (C24:1 and C24:0) were increased in knockdown lysates (Fig. 4A).
Given that we found PAQR4 to be induced in breast cancer, we wondered whether this induction was paralleled by higher ceramidase activity in tumor compared with nontumor tissue. We therefore performed the ceramidase assay in matched normal and breast cancer tissue samples. Interestingly, a principle component analysis was able to differentiate tumor tissues from the corresponding normal tissues (Fig. 4B). We further found that tumor tissue lysates exhibited increased ceramidase activity for C16:0, C18:0, and C24:1 ceramide compared with their paired normal tissues lysates (Fig. 4C). However, C24:0 ceramide abundance was unchanged. In further support of increased ceramidase activity in tumors, we found significant increase of labeled sphingosine—the downstream product of the ceramidase reaction—in the tumor compared with the matched control tissues (Fig. 4D). Combined, our studies strongly suggest that PAQR4 possesses ceramidase activity. In addition to the aforementioned PAQR2, cancer cells contain a number of other ceramidases. A transcriptional analysis of these ceramidases revealed that PAQR4 is induced to a larger extent compared with other ceramidases (Fig. 4E).
Sphingolipid metabolism is highly compartmentalized within the cell (32); thus, we next used three-dimensional (3D) confocal imaging to determine the subcellular localization of PAQR4. Detection of expressed flag-tagged PAQR4 revealed a perinuclear staining that colocalized with the Golgi marker RCAS1 both in MCF10A cells and in the breast cancer cell line SUM149 (Fig. 5A; Supplementary Fig. S4A and S4B). To further determine whether PAQR4 possesses ceramidase activity in the Golgi compartment, we took advantage of the fact that NBD fluorescent-labeled C6 ceramide readily accumulates in the Golgi when added exogenously to cells (33, 34). We therefore added NBD-labeled ceramides to control cells and PAQR4 overexpressing cells and measured Golgi-localized fluorescence intensities by confocal imaging. Interestingly, we found that overexpression of PAQR4 significantly reduced the NBD-labeled ceramide in the Golgi as determined by cosignal between NBD and RCAS1 (Fig. 5B). Furthermore, this effect was significantly reduced when the three histidine residues coordinated with the zinc ion in the catalytic domain were mutated to alanine (PAQR4HtriAOE; Fig. 5B; Supplementary Fig. S4C). At the cellular level, mutation of the three histidine residues, ablated the ability of PAQR4 to promote cancer growth (Fig. 5C).
Ceramides accumulation in the Golgi has been shown to induce local Golgi fragmentation (35). Consistent with PAQR4 having ceramidase activity, we observed a significant Golgi fragmentation in the PAQR4-depleted cells compared with the control cells by confocal immunofluorescence (Fig. 5D). Collectively, our findings suggest that PAQR4 acts as a ceramidase in the Golgi compartment.
PAQR4 depletion causes accumulation of de novo sphingolipid intermediates and ceramide-induced apoptosis
Having identified PAQR4 as a ceramidase, we next sought to determine how PAQR4 depletion alters cellular sphingolipid homeostasis. To this end, we performed an unbiased sphingolipidomics analysis of both MDA-MB-231 and MCF7 cells with depleted levels of PAQR4. Consistent with a ceramidase activity, we found overall elevated levels of ceramides in the PAQR4 knockdown cells in both cell lines (Fig. 6A). Similar to what we observed for ceramidase activity (Figs. 4A and 5B), we did not detect any PAQR4-related biases for specific ceramide fatty acyl chain length. Interestingly, we also detected an accumulation of upstream intermediates (i.e., dihydroceramides and sphinganine) in the de novo sphingolipid synthesis pathway (Fig. 6B and C), but not the downstream metabolites (i.e., glycosylceramides and sphingomyelins; Fig. 6D and E). This suggests that the de novo synthesis pathway is highly active in breast cancer cells.
Accumulation of cellular ceramides is tightly linked to apoptosis (36, 37). We therefore asked whether the reduced growth rates observed in the PAQR4-depleted cells is caused by increased apoptosis by assessing externalized phosphatidylserine in MDA-MB-231, HCC1806, MCF7, and T-47D PAQR4 knockdown cells by flow cytometry. Accordingly, we found increased rates of apoptosis across all cell lines (Fig. 7A and B). Furthermore, immunostaining of cleaved caspase-3 in tumor sections from control and PAQR4 knockdown tumors grown in vivo, confirmed activation of the apoptotic pathway in PAQR4-depleted cells (Fig. 7C).
Ceramides are linked to apoptosis by forming pores in the outer mitochondrial membrane to facilitate the release of cytochrome C from mitochondria into the cytoplasm (38, 39). Consistent with our hypothesis and cellular buildup of ceramides, we found increased levels of cytoplasmic cytochrome C in PAQR4-depleted cells as well as in cells treated with exogenous C6-ceramides (Fig. 7D and E). Furthermore, ectopic overexpression of PAQR4 abolished C6-ceramide induced release of cytochrome C to the cytoplasm (Fig. 7F; Supplementary Fig. S5). Considering the selective advantage of PAQR4 induction in tumors, we postulated that in addition to reducing cytotoxic ceramides, induction of PAQR4 could provide intermediates for S1P production. In support of this, we found that overexpression of wild-type PAQR4 increased cellular S1P levels, while overexpression of the ceramidase-dead HtriA mutation did not (Fig. 7G). Taken together, we find that PAQR4 depletion results in ceramide-induced apoptosis and an accumulation of the de novo sphingolipid synthesis intermediates.
Beyond their structural role in membrane homeostasis, sphingolipids control critical signaling transduction pathways within cancer cells to drive growth, proliferation, migration, and invasion. Altered levels of ceramide species and other metabolites in the sphingolipid metabolism network of cancer cells are related to the deregulation of expression of the genes encoding the enzymes within this network (4). Here, we identify and molecularly define the role of the ceramidase PAQR4 in breast tumors. We find that PAQR4 is highly expressed in the tumor tissue of the patients with breast cancer compared their corresponding normal tissue and that its expression negatively correlates with patient survival. On the basis of our findings, we propose that PAQR4 acts as a Golgi-localized ceramidase that provides a selective advantage to cancer cells through reducing cytotoxic ceramides as well as providing the building blocks for S1P production. Consistent with the work by Zhang and colleagues (29, 40, 41), we find that downregulation of PAQR4 reduces cancer cell growth. Their studies suggested that PAQR4 controls cellular growth through stabilization of cyclin-dependent kinase 4 (CDK4) and hence cell-cycle state. We were not able to experimentally observe cell-cycle arrests in PAQR4-depleted cells. Instead, through a combination of structural homology modeling, enzymatic, and lipidomics approaches, we propose that PAQR4 functions as a ceramidase. Consistent with the previous findings, we also observe reduced CDK4 expression in PAQR4 knockdown cells (Supplementary Fig. S6A). However, this is phenocopied by incubation of breast (Supplementary Fig. S6B) and lung (42) cancer cells with C6-ceramide, suggesting that the reduced CDK4 abundance in PAQR4-depleted cells is an indirect consequence of increased ceramide levels.
The de novo synthesis of ceramides is initiated in the endoplasmic reticulum (ER) by condensation of serine and fatty acids by the serine palmitoyltransferase complex (43). Ceramides are then converted enzymatically into different classes of sphingolipids in the ER, cis-, and medial-Golgi network that subsequently are integrated in organelle membranes (1). In this study, we show that PAQR4 is localized to the Golgi apparatus where it degrades ceramides into sphingosine. Although PAQR4 is homologous with other PAQR family members, in particular, PAQR1-3, PAQR4 appear to be selectively induced in breast cancer tissue compared with patient-matched nontumor tissue. We speculate that the selective advantage of PAQR4 is related to its placement in the Golgi. Other members, PAQR1 and 2 (commonly known as AdipoR1 and 2), are placed in the plasma membrane, where they serve as receptors for the antidiabetic hormone adiponectin (44, 45). By intervening early in the sphingolipid pathway, we suggest that the tumor cells gain a selective advantage by being able to reduce apoptotic ceramide species while at the same time creating the intermediates required for S1P synthesis. We did not detect PAQR4 in the plasma membrane, but it is tempting to speculate that PAQR4 activity could be regulated by an intracellular ligand.
Despite the local placement in the Golgi apparatus, we find evidence of global cellular accumulation of ceramides. In particular, the release of mitochondrial cytochrome C to the cytoplasm strongly suggests a mitochondrial buildup of ceramides. Future work will determine whether this is a result of direct Golgi-mitochondrial interaction sites (46) or vesicular lipid transfer (47).
In summary, we here describe how PAQR4 may provide cancer cells with an adaptive advantage for growth through modulation of the balance between ceramides and S1P. This suggests that development of small-molecule inhibitors of PAQR4 could provide a feasible path for a novel class of targeted therapies for breast cancer, either as a stand-alone treatment or in combination with chemotherapies known to alter ceramide levels.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: P. Panahandeh, P.E. Scherer, K. Teigen, N. Halberg
Development of methodology: L. Pedersen, P. Panahandeh, M.I. Siraji, A. Molven, K. Teigen, N. Halberg
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Pedersen, P. Panahandeh, S. Knappskog, P.E. Lønning, R. Gordillo, K. Teigen, N. Halberg
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Pedersen, P. Panahandeh, R. Gordillo, P.E. Scherer, K. Teigen, N. Halberg
Writing, review, and/or revision of the manuscript: L. Pedersen, P. Panahandeh, S. Knappskog, P.E. Lønning, R. Gordillo, P.E. Scherer, A. Molven, K. Teigen, N. Halberg
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Panahandeh, S. Knappskog
Study supervision: P. Panahandeh, A. Molven, K. Teigen, N. Halberg
N. Halberg is supported by the Trond Mohn Foundation Starting Grant. We thank James Lorens and Claudio Alacórn for comments on previous versions of the manuscript. We thank the VCU Lipidomics/Metabolomics Core, the NIH-NCI Cancer Center Support Grant P30 CA016059 to the VCU Massey Cancer Center, as well as a shared resource grant (S10RR031535) from the NIH for assistance with lipidomics analysis. We also acknowledge the Flow Cytometry Core Facility, Department of Clinical Science, and the Molecular Imaging Center, Department of Biomedicine, University of Bergen. N. Halberg is supported by the Trond Mohn Foundation Starting Grant.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.