Therapeutic Targeting of the Secreted Lysophospholipase D Autotaxin Suppresses Tuberous Sclerosis Complex-Associated Tumorigenesis Free

Tuberous sclerosis complex (TSC), an autosomal dominant disease characterized by multisystem hamartomas, including benign tumors of the brain, kidney, heart, and lung, affects one in 8,000 live births. About 30% of women with TSC develop lymphangioleiomyomatosis (LAM), a cystic lung destruction associated with diffuse proliferation of smooth muscle actin-positive cells that can progress to pulmonary failure requiring oxygen supplementation and lung transplant. Sporadic LAM can also occur, characterized by somatic mutations in the TSC1 or TSC2 gene and frequently associated with renal angiomyolipomas (1, 2). TSC2 deficiency due to inactivating mutations in the TSC genes leads to hyperactivation of mTOR Complex 1 (mTORC1), which integrates growth factor and nutrient signaling to stimulate cell growth, proliferation, and metabolism (3–8). Clinical trials of TSC and LAM with the mTORC1 inhibitor rapamycin showed heterogeneous response of tumor lesions and stabilization of pulmonary function; however, tumor growth and pulmonary function decline resumed when treatment was stopped (9, 10). Similarly, in laboratory studies, rapamycin exerts a cytostatic effect in TSC2-deficient cells. These studies highlight the need for additional therapeutic regimens in TSC and LAM.

Choline phospholipid metabolism is dysregulated in TSC2-deficient cells, and distinct lysophosphatidylcholine (LPC) species are significantly increased in LAM patient plasma (6) and suppressed by treatment with rapamycin and chloroquine (11), supporting the hypothesis that circulating LPC may participate in TSC/LAM pathogenesis. LPC is the major substrate of autotaxin (ATX), a secreted lysophospholipase D that degrades LPC to lysophosphatidic acid (LPA), a bioactive lipid known to play roles in cell proliferation, angiogenesis, and tumor metastases via specific G protein–coupled receptors (GPCR; ref. 12). ATX also degrades sphingosylphosphorylcholine (SPC), converting it into sphingosine-1-phosphate (S1P), a metabolite regulating cell motility (13). ATX is involved in wound healing, inflammation, and angiogenesis, and was identified among the top 40 upregulated genes in a model of metastatic mammary carcinoma (14).

Here, we show the impact of inhibiting the ATX pathway on the biology of TSC2-deficient cells in vitro and in vivo. GLPG1690 (developed by Galapagos NV) is a compound that specifically targets ATX and has progressed to phase III clinical trial for idiopathic pulmonary fibrosis (IPF; ClinicalTrials.gov Identifier: NCT03711162). We found that ATX is upregulated in TSC2-deficient cells, and that GLPG1690 inhibits the oncogenic potential of TSC2-deficient cells in vitro and in vivo. Short-term treatment with GLPG1690 inhibits the phosphorylation of AKT and ERK1/2 in TSC2-deficient cells, whereas long-term treatment suppresses lipid synthesis and promotes fatty acid oxidation, leading to lower neutral lipid content in TSC2-deficient cells. TSC-associated renal angiomyolipomas express significantly higher levels of LPA receptor 1 (LPAR1) and S1P receptor 3 (S1PR3) compared with normal kidney. Consistent with these results, ATX products LPA and S1P rescue the proliferation, survival, and transcriptome of human renal angiomyolipoma-derived TSC2-deficient cells treated with GLPG1690.

In summary, our data support a role for the ATX–LPA/S1P pathway in TSC-associated tumorigenesis with potential therapeutic implications.

The following cell lines were used: (i) isogenic derivatives of LAM patient renal angiomyolipoma-derived TSC2-deficient 621-101 cells (gift of Dr. Elizabeth Henske). These cells were derived from a LAM patient renal angiomyolipoma (15) and carry the same somatic bi-allelic TSC2 gene inactivating mutations as the patient's LAM cells (G1832A missense mutation of one allele, and loss of the other allele) (16). The isogenic derivative pair includes empty vector 621-102 cells and TSC2 add-back 621-103 cells (Supplementary Fig. S1); and (ii) Tsc2^−/− and Tsc2^+/+ mouse embryonic fibroblasts (MEF, gift of Dr. David Kwiatkowski; ref. 17).

All cell lines were grown in DMEM supplemented with 10% FBS, 100 IU/mL of penicillin, and 100 μg/mL of streptomycin, unless specified otherwise. 621-102 and 621-103 cells were grown under antibiotic selection pressure with zeocin (30 μg/mL). Zeocin was removed before each experiment.

TSC2 deficiency, constitutive activation of mTORC1, and rapamycin sensitivity were validated after each thawing by immunoblotting for tuberin/TSC2 and phospho-S6 kinase or phospho-S6 ribosomal protein in the presence or absence of FBS. Mycoplasma testing (MycoAlert Mycoplasma Detection Kit; Lonza) was conducted after each thawing and at least monthly. Cells were no longer used in experiments after reaching passage 40.

Tsc2^−/− MEFs were infected with pBabe-Puro-Myr-Flag-AKT1 (18) and/or transfected with pCMV-myc-ERK2-L4A-MEK1_fusion (gift from Melanie Cobb; Addgene plasmid #39197; http://n2t.net/addgene:39197; RRID:Addgene_39197) using Fugene HD (Promega).

For CRISPR gene editing, Tsc2^−/− MEFs were transfected with a predesigned TrueGuide sgRNA targeting Enpp2 (assay ID CRISPR480928_SGM) or TrueGuide sgRNA Negative Control nontargeting 1 and TrueCut Cas9 v2 (Invitrogen) following the manufacturer's recommendations. Because of low transfection efficiency, single cell clones were grown and screened for on-target genome editing using the Alt-R Genome Editing Detection Kit (IDT). T7EI assay results were analyzed by visualizing the cleavage products and the full-length amplicon (forward primer: 5′-GAATCTCTCCGATCACTACCATTT; reverse primer: 5′-AGGCAGGTGGTGTTTCATAG) on a 2% agarose gel.

GLPG1690 was obtained from Medkoo Biosciences and dissolved in DMSO. LPA and S1P were obtained from Sigma and Avanti Polar Lipids and preconjugated with 2% fatty acid-free BSA at 37°C for 20 to 30 minutes prior to each experiment. Rapamycin (LC Laboratories), MK2206 (Selleckchem), and SCH772984 (Cayman Chemical) were dissolved in DMSO.

Cells were plated on 12-well plates and treated with increasing doses of GLPG1690 or rapamycin (20 nmol/L) in medium supplemented with 10% FBS unless otherwise specified. After 68 to 96 hours of incubation, cells were fixed with formalin and stained with crystal violet, then dissolved in methanol and read on a Synergy HT BioTek plate reader.

Oris assays (Catalog no. CMA5.101; Platypus Technologies) use a stopper to create a cell-free detection zone in the center of each well of a 96-well plate. Assays were performed according to the manufacturer's instructions. Briefly, 30,000 cells were seeded in DMEM containing 2% FBS per well around the stoppers. After cells attached overnight, the stoppers were removed (except for 0 hour control wells) and GLPG1690 (3 μmol/L for Tsc2^−/− MEFs and 6 μmol/L for the human TSC2-deficient cells) or DMSO vehicle was added. Cells were allowed to migrate to the center of wells for 18 hours before the 96-well plate was scanned on a Celigo imager. Migration was quantified by measuring the % wound healing (t_end – t₀, 40% well mask) and normalized to vehicle control.

Cells (10,000/well) were mixed in a layer of 0.4% Noble agar (BD Biosciences) in DMEM with 10% FBS (1 mL) and plated on top of a layer of 0.6% agar in DMEM with 10% FBS (3 mL) in 6-well plates. After agar solidified, cells were treated with GLPG1690 (6 μmol/L) or DMSO control (0.06%) in 1 mL of medium, twice a week for 6 weeks. Images of the entire wells were taken with an Olympus SZH10 Research Stereo Microscope and colonies were counted.

Human TSC2-deficient or TSC2 add-back cells were plated on 10 cm dishes and treated with vehicle or GLPG1690 (6 μmol/L) in DMEM with 2% FBS, 0.18% DMSO, and 0.1% BSA. LPA (6 μmol/L) or S1P (6 μmol/L) was supplemented to human TSC2-deficient cells. After 24-hour treatment, cells were washed with cold PBS (6 mL) and RNA was collected with PureLink RNA Mini Kit (Invitrogen) following the manufacturer's instructions. The concentration of purified RNA was measured using Nanodrop. Two micrograms of RNA were submitted for Illumina RNA-sequencing (RNA-seq), which was conducted by the Molecular Biology Core Facilities, Dana-Farber Cancer Institute.

Libraries were prepared using Kapa strandedmRNA Hyper Prep sample preparation kits from 100 ng of purified total RNA according to the manufacturer's protocol. The finished dsDNA libraries were quantified by Qubit fluorometer, Agilent TapeStation 2200, and RT-qPCR using the Kapa Biosystems Library Quantification Kit according to manufacturer's protocols. Uniquely indexed libraries were pooled in equimolar ratios and sequenced on an Illumina NextSeq500 with single-end 75 bp reads.

Sequenced reads were aligned to the UCSC hg19 reference genome assembly and gene counts were quantified using STAR (v2.5.1b; ref. 19). Differential gene expression testing was performed by DESeq2 (v1.10.1; ref. 20) and normalized read counts (FPKM) were calculated using cufflinks (v2.2.1; ref. 21). RNA-seq analysis was performed using the VIPER snakemake pipeline (22).

Gene set enrichment analysis (GSEA) was performed using the R package GSEABase (23). Entrez IDs ranked by decreasing fold changes from DESeq2 results table were used as input and evaluated against the mdsig database v6.2 (24–26). Gene ontology enrichment analysis was performed by VIPER using on genes selected from the DESeq2 results table that had a fold change >2 or fold change <−2 and an adjusted P value <0.05 against a background of all genes detected in the dataset.

Published RNA-seq data (27) were obtained through dbGap. Differential gene expression (DESeq) analysis was performed for 12 renal angiomyolipomas and 4 normal kidney tissue samples using R. Transcripts per million (TPM) values for LPA and S1P receptors were obtained and plotted.

Two micrograms of total RNA (PureLink RNA Mini Kit; Invitrogen) were retrotranscribed with the SuperScript IV First-Strand Synthesis System (Invitrogen). The following TaqMan probes (Applied Biosystems) were used: ENPP2 (HS00905125_m1), Enpp2 (mouse Mm00516572_m1), FASN (Hs01005622_m1), PCYT1A (Hs00192339_m1), ACACA (Hs01046047_m1), SCD (Hs01682761_m1), LPAR1 (HS00173500_m1, LPAR2 (HS01109356_m1), LPAR3 (HS00173857_m1), LPAR4 (Hs01099908_m1), LPAR5 (Hs01054871_m1), LPAR6 (Hs05006584_m1), S1PR3 (Hs01019574_m1), and S1PR5 (Hs00924881_m1).

For BrdUrd incorporation, Tsc2^+/+ and Tsc2^−/− MEFs cells were plated on 10 cm plates and incubated with GLPG1690 (3 μmol/L) or DMSO control for 68 hours in DMEM supplemented with 10% FBS. Two hours before collecting the cells, 10 μmol/L BrdUrd (Upstate) was added. Adherent cells were trypsinized and combined with floating cells. Cells were pelleted (1,000 rpm, 5 minutes), resuspended in 50 μL of PBS, and fixed with 6 mL of precooled 70% ethanol at room temperature for 30 minutes. Cells were pelleted, washed with 1 mL of 0.5% BSA in PBS, pelleted again, and resuspended with 500 μL of 2M HCl in PBS. After 20-minute incubation at room temperature, 1 mL of 0.5% BSA in PBS was added immediately to each sample. Cells were pelleted, resuspended in 50 μL of BrdUrd-FITC Ab (BD-Biosciences; Catalog no. 556028) or mouse IgG negative control, and incubated in the dark at room temperature for 30 minutes. One milliliter of 0.5% BSA in PBS was added to each sample. Cells were again pelleted and resuspended in 500 μL of PI (10 μg/mL in distilled H2O) with 10 μL of RNase A (10 μg/μL). After 30-minute incubation at room temperature, cells were kept on ice and analyzed on a flow cytometer.

For BODIPY493/503 staining, 621-102 and 621-103 cells were seeded on six-well plates (200,000 cells per well) in DMEM with 2% FBS. After attachment, cells were treated with vehicle control (0.18% DMSO + 0.1% BSA), GLPG1690 (6 μmol/L), GLPG1690 (6 μmol/L) + LPA (6 μmol/L), LPA (6 μmol/L), GLPG1690 (6 μmol/L) + S1P (6 μmol/L) or S1P (6 μmol/L) for 70 hours. Cells were then washed with PBS, incubated with 2 mL of 4 μmol/L BODIPY 493/503 (Catalog no. D3922; Invitrogen) in PBS at 37°C in the dark for 30 minutes, rinsed with PBS, trypsinized, and resuspended in 300 μL of PBS. Flow cytometry was performed to obtain a minimum of 10,000 events per condition.

Total proteins were extracted through 30-minute incubation on ice with Nonidet P-40 lysis buffer containing protease and phosphatase inhibitors, and resolved on Bolt Bis-Tris Plus polyacrylamide gels (Life Technologies). Antibodies against PARP (Catalog no. 9532S), phospho-AKT (S473; Catalog no. 4060S), AKT (Catalog no. 4685S), phospho-ERK (T202/Y204; Catalog no. 9101S), ERK1/2 (Catalog no. 9102S), phospho-S6 ribosomal protein (S235/236; Catalog no. 2211S), total S6 ribosomal protein (Catalog no. 2317S), phospho-S6 kinase (Catalog no. 9234S), total S6-kinase (Catalog no. 2708S), tuberin/TSC2 (Catalog no. 4308S), phospho-RSK (S380; Catalog no. 9335), total RSK (Catalog no. 9355), fatty acid synthase (Catalog no. 3180S), acetyl-CoA carboxylase (Catalog no. 3676S), Stearoyl-CoA desaturase 1 (Catalog no. 2794S), CCTα (Catalog no. 6931S), and BrdUrd (5292S) were obtained from Cell Signaling Technology. Anti-β-actin antibody (Catalog no. A5316) was obtained from Millipore Sigma and anti-CPT1A antibody (Catalog no. ab128568) from Abcam.

TSC2 add-back (300,000/well for 24 hours and 150,000/well for 72 hours) and TSC2-deficient cells (200,000/well for 24 hours and 100,000 for 72 hours) were seeded in 12-well plates and treated with GLPG1690 (6 μmol/L) or control (0.06% DMSO) in DMEM with 10% FBS for 24 or 72 hours. Cells were then incubated for 3 hours at 37°C with 1 μCi/mL of [U-¹⁴C]palmitate (PerkinElmer Inc.). 3M perchloric acid was added to the cell culture medium and the wells were sealed with Whatman filter paper saturated with phenethylamine (Sigma-Aldrich) to capture ¹⁴C-CO₂. The plates were gently shaken for 3 hours at room temperature and the filter paper was removed and placed into Ultima Gold F Scintillation Fluid (PerkinElmer Inc.). Radioactivity was counted on a Packard Tri-Carb Liquid Scintillation Analyzer. Data were normalized against the protein mass (total micrograms from three independent wells).

Human TSC2-deficient cells (400,000/well for 24 hours and 200,000 for 72 hours) and TSC2 add-back cells (600,000/well for 24 hours and 300,000/well for 72 hours) were seeded in six-well plates and treated with GLPG1690 (6 μmol/L) or control (0.06% DMSO) in DMEM with 10% FBS for 24 or 72 hours. Cells were then labeled with [1-¹⁴C]acetic acid (0.5 μCi/mL; Perkin-Elmer) for 4 hours, washed two times with PBS and collected for lipid extraction using isopropanol (500 μL). Radioactivity from 20 μL of the lipid extract was counted on a Packard Tri-Carb Liquid Scintillation Analyzer. Data were normalized against the protein mass (total micrograms from three independent wells).

Subcutaneous tumors were generated by injecting Tsc2^−/− MEFs in female NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice (The Jackson Laboratory, JAX stock no. 005557). Three independent trials were conducted. 2.5 × 10⁶ cells/mouse were resuspended in 100 μL of PBS and injected with matrigel (1:1) in a single flank. When tumors reached a palpable size, mice were treated with GLPG1690 (60 mg/kg/day) or control (DMSO) diluted in sterile vehicle (0.25% Tween 80/0.25% PEG 200 in distilled water) through intraperitoneal injection using a 27G needle for 30 days. Tumors were harvested 4 hours after BrdUrd injection (1 mg/mouse, i.p.) and the last treatment, and submitted for histopathologic analyses.

The animal studies were conducted under a protocol approved by the Institutional Animal Care and Use Committee at the Brigham and Women's Hospital.

Dissected tumors were fixed in formalin for 24 hours. Hematoxylin and eosin and IHC staining were performed on 5-μm sections of formalin-fixed and paraffin-embedded (FFPE) samples. IHC was performed by InvivoEx Inc.

Paraffin sections of tissue were dewaxed using xylene and rehydrated in graded ethanol. DNA hydrolysis was performed using HCl and neutralized with sodium borate buffer. Sections were then incubated in a 1:200 dilution of the mouse monoclonal anti-BrdUrd primary antibody (Bu20a; Cell Signaling Technology) at 4°C overnight, and then incubated in a 1:200 dilution of biotinylated goat anti-mouse IgG secondary antibody (Vector Laboratories) for 1 hour at room temperature. Immunoreactivity was visualized using streptavidin-alkaline phosphatase followed with substrate Vector blue, which resulted in a blue immunoreactive signal; sections were then counterstained with nuclear fast red and mounted.

Small tumor fragments (1–2 mm³) were fixed in formaldehyde-glutaraldehyde-picric acid fixative (2.5% glutaraldehyde, 1.25% paraformaldehyde, and 0.03% picric acid in 0.1M sodium cacodylate buffer, pH 7.4) at 4°C. Three tumor samples from each group (drug or control) were analyzed at the Electron Microscopy Core Facility (Harvard Medical School).

Tissue samples were postfixed with 1% osmium tetroxide (OsO₄)/1.5% potassium ferrocyanide (KFeCN₆) for 1 hour, washed in water two times, one time in 50 mmol/L maleate buffer (pH 5.15, MB) and incubated in 1% uranyl acetate in MB for 1 hour followed by one wash in MB, two washes in water and subsequent dehydration in grades of alcohol (10 minutes each; 50%, 70%, 90%, 2 × 10 minutes 100%). The samples were left in propyleneoxide for 1 hour and infiltrated overnight in a 1:1 mixture of propyleneoxide and TAAB Epon (TAAB Laboratories Equipment Ltd., https://taab.co.uk). The following day the samples were embedded in TAAB Epon and polymerized at 60°C for 48 hours. Ultrathin sections (about 80 nm) were cut on a Reichert Ultracut-S microtome, picked up on to copper grids stained with lead citrate, and examined in a JEOL 1200EX Transmission electron microscope or a TecnaiG² Spirit BioTWIN. Images were recorded with an AMT 2k CCD camera.

Statistical analyses were performed using GraphPad Prism 5. Data are reported as median ± 95% CI unless otherwise noted. Statistical significance was defined as P < 0.05.

TSC2-deficient cells produce increased levels of LPC compared with TSC2-expressing cells (6). Because LPC is a preferential substrate of the secreted lysophospholipase D ATX, we assayed the expression of this enzyme in TSC2-deficient and TSC2-expressing cells. ATX mRNA expression was nearly four-fold higher in LAM patient renal angiomyolipoma-derived TSC2-deficient cells (621-102) compared with the isogenic TSC2 add-back control cells (621-103; Fig. 1A).

This result was validated in Tsc2^−/− MEFs, which showed ∼30-fold increase in ATX expression compared with Tsc2^+/+ MEFs (Supplementary Fig. S2A). Treatment with rapamycin (20 nmol/L, 24 hours) significantly reduced ATX expression selectively in TSC2-deficient 621-102 cells (Fig. 1A), suggesting that ATX expression may be regulated downstream of mTORC1 in this patient-derived cell line.

Next, to determine the role of ATX in the proliferation and survival of TSC2-deficient cells, we used a specific ATX inhibitor, GLPG1690, which is currently being tested in phase III clinical trial for IPF (NCT03711162). First, using a synthetic ATX substrate, FS-3 (LPC analogue), in a fluorogenic activity assay, we confirmed that GLPG1690 inhibits the enzymatic activity of recombinant ATX in vitro with an IC₅₀ of ∼64 nmol/L (Supplementary Fig. S2B), validating its potency. Second, we tested GLPG1690 in both TSC2-deficient models, the human renal angiomyolipoma cells and Tsc2^−/− MEFs, in dose-dependent experiments. GLPG1690 inhibited the proliferation of TSC2-deficient cells at a significantly lower IC₅₀ (5.46 ± 0.24 μmol/L for human TSC2-deficient cells; 2.84 ± 0.22 μmol/L for Tsc2^−/− MEFs) than TSC2-expressing cells (7.34 ± 0.15 μmol/L for human TSC2 add-back cells, Fig. 1B; 4.71 ± 1.17 μmol/L for Tsc2^+/+ MEFs; Supplementary Fig. S2C) by crystal violet staining.

Consistent with these results, inhibition of ATX via CRISPR sgRNA suppressed TSC2-deficient cell proliferation by ∼75% (Supplementary Fig. S2D). On-target genome editing was confirmed using T7 endonuclease I (T7EI), which recognizes and cleaves mismatched DNA heteroduplexes. T7EI assay results were analyzed by visualizing the cleavage products and the full-length amplicon on a 2% agarose gel (Supplementary Fig. S2D).

In addition, GLPG1690 induced moderate apoptosis levels selectively in the TSC2-deficient cells, as shown by PARP cleavage (an apoptosis marker) in the presence of the drug (6 μmol/L, 6 or 72 hours) in immunoblotting analysis (Fig. 1C). As expected, rapamycin alone did not induce apoptosis.

We tested the impact of GLPG1690 on other oncogenic properties of TSC2-deficient cells. GLPG1690 inhibited the migration of LAM patient renal angiomyolipoma-derived TSC2-deficient cells by 73% (Fig. 1D) and that of Tsc2^−/− MEFs by 37% (Supplementary Fig. S2E) in the presence of 10% FBS, as shown by 18-hour Oris migration assays.

Next, we found that GLPG1690 inhibited the anchorage-independent growth of TSC2-deficient cells (621-102) by 82% (Fig. 1E). TSC2 add-back cells (621-103) formed 74% less colonies in soft agar than TSC2-deficient cells at baseline, and were not affected by the drug (Fig. 1E).

To understand the mechanisms through which inhibition of the ATX pathway by GLPG1690 suppresses TSC2-loss associated oncogenicity, we performed RNA-seq analysis on human TSC2-deficient and TSC2 add-back cells treated with the drug or vehicle, in the presence of LPA or S1P, the two lipid products of ATX. Treatment with GLPG1690 induced substantial gene expression changes in the TSC2-deficient cells: 5,116 genes were differentially expressed (P_adj < 0.05), including 294 genes with absolute log2 (fold change) > 1.0 (205 upregulated genes, and 89 downregulated genes; Fig. 2A; Supplementary File S1). Only 280 differentially expressed genes (P_adj < 0.05) including one gene with absolute log2 (fold change) > 1.0 were found in the TSC2 add-back cells (Fig. 2A; Supplementary File S1). Interestingly, ATX mRNA levels were reduced by GLPG1690, selectively in TSC2-deficient cells, suggesting that the drug suppresses not only the activity but also the transcription of ATX (Fig. 2B). To identify transcriptional changes at the pathway level, GSEA was conducted, revealing 50 significantly enriched KEGG gene sets and 36 significantly enriched Hallmark gene sets in the TSC2-deficient cells. These included cell cycle (KEGG), focal adhesion (KEGG), oxidative phosphorylation (Hallmark), adipogenesis (Hallmark), apoptosis (Hallmark), and fatty acid metabolism (Hallmark; Fig. 2C; Supplementary File S1).

To test whether the effects of GLPG1690 were mediated by the ATX lipid products, we supplemented the culture media of drug-treated human renal angiomyolipoma-derived TSC2-deficient cells with LPA or S1P and tested these conditions in RNA-seq, proliferation, and survival experiments.

In the RNA-seq analysis, of the 294 genes impacted by GLPG1690 in 621-102 cells [P_adj < 0.05, log2 (fold change) > 1.0], expression of 147 genes was reversed by adding back LPA, expression of 64 genes was reversed by adding back S1P, and expression of 15 genes was reversed by both LPA and S1P. The rescue was defined as a significant change (with opposite sign) in gene expression in GLPG1690 + LPA or S1P-treated cells versus GLPG1690-treated cells (Fig. 3A; Supplementary Fig. S3A; Supplementary File S1). These results suggest that LPA and S1P drive nonredundant transcriptional programs in TSC2-deficient cells, differentially contributing to ATX signaling. Gene Ontology (GO) term enrichment analysis of the RNA-seq data revealed that LPA mainly regulated inflammatory-associated pathways and adhesion-associated genes, whereas S1P regulated lipid metabolism-associated genes (Supplementary File S1).

To validate the role of LPA and S1P in mediating the effects of GLPG1690 on the biology of TSC2-deficient cells, these cells were treated with GLPG1690 (6 μmol/L), LPA (6 μmol/L), or S1P (6 μmol/L), or the combination of both in the presence of 2% FBS for 72 hours. Crystal violet staining showed that either LPA or S1P could partially rescue the proliferation of TSC2-deficient cells upon treatment with GLPG1690 (Fig. 3B). In line with the RNA-seq data, supplementation of both LPA and S1P (3 μmol/L + 3 μmol/L) fully rescued proliferation under the same conditions (Supplementary Fig. S3B). Immunoblotting revealed that supplementation of either LPA or S1P could prevent PARP cleavage (apoptosis) in the TSC2-deficient cells treated with GLPG1690 (Fig. 3C).

RNA-seq experiments corroborated a role for the ATX products, LPA and S1P, in the response to treatment with GLP1690. These lipids act through specific GPCRs, LPARs and S1PRs. We tested the expression of these receptors in human TSC2-deficient and TSC2 add-back cells in the RNA-seq database and by qPCR. LPAR1 and S1PR3 were significantly overexpressed in TSC2-deficient cells (Fig. 3D and E). Treatment with GLPG1690 led to an increase in the expression of LPAR1 and a decrease in the expression of S1PR3 in these cells (Fig. 3E, top). LPAR1 expression was also enhanced by treatment with rapamycin, whereas S1PR3 expression was not affected (Fig. 3E, bottom). Lower or unchanged expression levels in TSC2-deficient versus TSC2 add-back cells were found for S1PR5, LPAR2, and LPAR3 (Supplementary Fig. S4).

Importantly, LPAR1 and S1PR3 were also significantly overexpressed in TSC-associated renal angiomyolipomas, as tested in a published RNA-seq dataset (Fig. 3F).

To assess GLPG1690-induced cell signaling changes, we screened 43 P-kinase sites and 2 related proteins in the LAM patient-derived TSC2-deficient cells and the TSC2 add-back control cells treated with GLPG1690 (6 μmol/L, 6 hours) or DMSO. Twenty-four of these P-kinase sites (or proteins) showed greater than 25% suppression by GLPG1690 treatment specifically in the TSC2-deficient cells; eight of them showed greater than 50% change with the inhibitor, including ERK1/2 (T202/Y204, T185/Y187) and AKT1/2/3 (S473; Supplementary Fig. S5A), which are known to mediate signaling downstream of LPAR/S1PR (28–34). We confirmed the effect of GLPG1690 on AKT and ERK phosphorylation by immunoblotting: 6-hour treatment with GLPG1690 (6 μmol/L) led to a decrease in P-AKT (S473) by 68 ± 10% and in P-ERK (T202/Y204) by 56 ± 12% in the human TSC2-deficient cells (Fig. 4A and B). P-S6 (S235/236), a direct target of mTORC1, was not affected under this condition. Consistent results were obtained in Tsc2^−/− MEFs (Supplementary Fig. S5B).

Intriguingly, a differential effect of LPA and S1P supplementation on AKT/ERK activation was found. P-AKT levels were rescued by supplementation of LPA, whereas P-ERK levels were rescued by supplementation of S1P (Fig. 4C).

Next, to ask whether suppression of AKT and ERK signaling plays a role in GLPG1690 proapoptotic and antiproliferative effects, we used two approaches. First, we treated cells with a specific AKT or ERK inhibitor in combination with GLPG1690. The human TSC2-deficient cells were pretreated for 30 minutes with AKT inhibitor MK2206 (4 μmol/L) or ERK inhibitor SCH772984 (2 μmol/L), and then incubated with GLPG1690 (6 μmol/L) for 18 hours in the presence of 10% FBS. Immunoblotting was performed to detect cleaved PARP. Either inhibitor induced low levels of apoptosis as single agent and worked synergistically in combination with GLPG1690 to enhance apoptosis (Fig. 4D). Second, to test whether constitutive activation of AKT or ERK would prevent the impact of GLPG1690 treatment on TSC cell proliferation, we expressed myristoylated-AKT (myr-AKT) or a fusion of ERK2 with the low activity form of its upstream regulator, the MAP kinase MEK1 (35), in Tsc2^−/− MEFs (Fig. 4E). Cells were treated with GLPG1690 (3 μmol/L) or vehicle for 92 hours. The proliferation rate upon drug treatment (drug/DMSO, each normalized to its own baseline) was significantly higher in the presence of coexpression of myr-AKT and constitutively active ERK compared with the empty vector control (Fig. 4E).

These data support a role for AKT and ERK signaling in TSC2-deficient cell proliferation, including effects downstream of the ATX/LPA/S1P axes.

The RNA-seq analysis revealed substantial changes in genes of fatty acid metabolism in 621-102 cells treated with GLPG1690. Specifically, in the gene sets of fatty acid metabolism and adipogenesis, 63 of 146 genes and 86 of 186 genes were significantly altered transcriptionally. Four enzymes involved in fatty acid oxidation, including acyl-CoA dehydrogenase short chain (ACADS), acyl-CoA thioesterase 8 (ACOT8), and malonyl-CoA decarboxylase (MLYCD), were upregulated, whereas seven enzymes involved in lipid synthesis, including fatty acid synthase (FASN), acetyl-CoA carboxylase alpha (ACACA), and acyl-CoA synthetase long chain family member 1 (ACSL1), were downregulated (Supplementary File S1).

To validate the metabolic reprogramming suggested by the transcriptome changes, we used flow cytometry-based neutral lipid quantification and ¹⁴C labeling experiments to trace fatty acid oxidation and de novo lipid synthesis. TSC2-deficient cells were treated with GLPG1690 (6 μmol/L), LPA (6 μmol/L), S1P (6 μmol/L), the combination of GLPG1690 with either lipid, or vehicle (DMSO), in the presence of 2% FBS for 70 hours. Cellular neutral lipids were stained with BODIPY493/503. GLPG1690 decreased neutral lipid content by 29% (P < 0.01) in TSC2-deficient cells, which have higher neutral lipid content than TSC2 add-back cells (P < 0.01), as expected. Intriguingly, the decrease in neutral lipid content was rescued by adding back either LPA or S1P (Fig. 5A).

To further determine the causes of these changes, we performed ¹⁴C-palmitate oxidation (fatty acid oxidation) and ¹⁴C-acetate lipid incorporation (de novo lipid synthesis) assays upon ATX inhibition for 24 or 72 hours in TSC2-deficient and TSC2 add-back cells. GLPG1690 significantly promoted β-oxidation selectively in TSC2-deficient cells at 24 hours and in both cell lines at 72 hours (Fig. 5B). The drug also significantly downregulated de novo lipid synthesis in both cell lines at 72 hours (Fig. 5B).

Consistent with these results, the protein expression of the lipogenic enzyme CCTα (CTP:phosphocholine cytidylyltransferase α), which is involved in lipid droplet biogenesis (36) was suppressed by ∼50% following 72-hour treatment with GLPG1690, whereas no change in CCTα protein expression was found in TSC2 add-back cells (Fig. 5C; Supplementary Fig. S6). Minor changes in the expression (∼30%) of two enzymes involved in de novo fatty acid synthesis, fatty acid synthase (FASN) and acetyl-CoA carboxylase α (ACCα), occurred in TSC2-deficient cells (Fig. 5C; Supplementary Fig. S6), and expression of the desaturase stearoyl-CoA desaturase 1 (SCD1) was suppressed in both TSC2-deficient and TSC2 add-back cells (Fig. 5C). FASN and SCD1 were confirmed to be regulated transcriptionally (Fig. 5D). Expression of the mitochondrial fatty acid oxidation rate-limiting enzyme carnitine palmitoyltransferase 1A (CPT1A) was not affected by drug treatment.

Treatment with GLPG1690 led to a reduction in tumor burden by ∼40% (P = 0.016; Fig. 6A). Mouse body weight was not affected by GLPG1690 treatment (Fig. 6B) and no drug toxicity was found. Pathologic analysis revealed clusters of more differentiated, fibroblast-like cells. As an observation, subcutaneous fat around the tumors appeared to be less abundant in the drug-treated mice and tumors seemed to infiltrate less into the muscle (Fig. 6C). Interestingly, electron microscopy images revealed inflated endoplasmic reticulum and confirmed a reduction in lipid droplets in the tumors treated with GLPG1690 (Fig. 6D), consistent with the BODIPY493/503 staining and fatty acid oxidation/de novo lipid synthesis assays results. Finally, consistent with the effect of GLPG1690 on Tsc2^−/− MEFs in vitro (Fig. 6E), we found a decrease in BrdUrd incorporation in tumor cells by 38% (P = 0.034; Fig. 6F).

This study identifies a novel mode of metabolic dysregulation in the TSC tumor microenvironment, the ATX–LPA/S1P pathway (Fig. 6G).

ATX regulates availability of two bioactive lipids, LPA and S1P, for activating specific membrane GPCRs. ATX generates LPA and S1P through its lysophospholipase D activity and binds and delivers these lipids to their receptors, protecting them from phosphatase degradation (37). The ATX pathway has been associated with cancer progression and metastasis (38, 39). LPA and S1P regulate several physiological processes, including cell proliferation, cell migration/invasion, angiogenesis, and inflammation. These lipids activate a series of GPCRs, at least six for LPA (LPAR1-6) and five for S1P (S1PR1-5), stimulating a wide variety of downstream signaling including PI3K/AKT and RAS/ERK pathways (40–47).

Importantly, two of these GPCRs, LPAR1 and S1PR3, are upregulated in TSC-associated renal angiomyolipomas (Fig. 3F), consistent with TSC2-deficient human cells (Fig. 3D and E).

GLPG1690 is a potent and specific ATX inhibitor currently in phase III clinical trials for IPF. Its safety and target engagement was shown in phase I and II trials (48, 49). We found that inhibition of the ATX pathway using GLPG1690 suppresses the oncogenicity of TSC2-deficient cells, including cell proliferation, cell migration, anchorage-independent growth, and tumor growth in vivo. Consistent with GLPG1690 effects, ATX gene editing via CRISPR sgRNA dramatically suppressed the proliferation of Tsc2^−/− MEFs.

Taken together, these data suggest a substantial role for ATX–LPA/S1P signaling pathway in TSC tumorigenesis. Mechanistically, AKT and ERK signaling were affected in cells treated with GLPG1690 and combination of the ATX inhibitor with either AKT or ERK1/2 specific inhibitors led to enhanced apoptosis in TSC2-deficient cells; on the contrary, expression of constitutively active AKT and ERK rendered the cells significantly less sensitive to the antiproliferative effect of GLPG1690.

Moreover, ATX inhibition led to LPA and S1P-dependent transcriptomic and metabolic reprogramming. Our RNA-seq experiments uncovered specific roles for LPA and S1P in the context of TSC2 loss. These bioactive lipids reversed differential changes in the transcriptome of TSC2-deficient cells treated with GLPG1690, with major involvement of LPA in cell adhesion/motility and inflammatory processes, and of S1P in sterol and lipid biosynthesis.

We found that inhibition of ATX by GLPG1690 led to a reprogramming of lipid metabolism via multiple mechanisms. One mechanism included reduction in the mRNA and/or protein expression of lipogenic enzymes, CCT, ACC, FASN, and SCD1, with associated decrease in lipid droplet content and de novo lipid synthesis in cells treated with the drug. CCT is the rate-limiting enzyme in the CDP-choline pathway for phosphatidylcholine biosynthesis. This enzyme participates in nuclear membrane, nucleoplasmic reticulum, and lipid droplet biogenesis, and contributes to phospholipid homeostasis. ACC, FASN, and SCD1 mediate fatty acid synthesis. Interestingly, TSC2-deficient cells have the ability to upregulate expression of the lipogenic enzyme FASN over time in culture (72 hours compared with 24 hours), likely to enhance fatty acid synthesis when exogenous availability decreases; however, treatment with GLPG1690 prevented this increase, potentially making these cells more vulnerable to nutrient depletion. Another mechanism involves lipid catabolic processes. Treatment with GLPG1690 led to enhanced fatty acid oxidation selectively in the TSC2-deficient cells at 24 hours and in both TSC2-deficient and TSC2 add-back human renal angiomyolipoma cells at 72 hours.

These data suggest a role for the ATX pathway in the regulation of the intracellular lipidome of TSC2-deficient cells.

Surprisingly, although suppression of LPA and S1P levels through ATX inhibition would be expected to upregulate ATX expression in tissues due to feedback regulation (50), we found that treatment with GLPG1690 suppressed the expression of ATX in human TSC2-deficient cells, suggesting that this compound acts through multiple mechanisms to suppress ATX activity.

In summary, our studies suggest that dysregulated ATX–LPA/S1P pathways are critical players in TSC2-deficient cell fitness and in TSC tumorigenesis, and that ATX could be tackled for novel therapeutic modalities in TSC and LAM.

No potential conflicts of interest were disclosed.

Conception and design: Y. Feng, C. Priolo

Development of methodology: Y. Feng, C. Priolo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Feng, W.J. Mischler, A.C. Gurung, T.R. Kavanagh, G. Androsov, P.M. Sadow, Z.T. Herbert, C. Priolo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Feng, W.J. Mischler, A.C. Gurung, T.R. Kavanagh, G. Androsov, P.M. Sadow, Z.T. Herbert, C. Priolo

Writing, review, and/or revision of the manuscript: Y. Feng, A.C. Gurung, T.R. Kavanagh, P.M. Sadow, C. Priolo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.C. Gurung, C. Priolo

Study supervision: C. Priolo

We thank Maria Ericsson (Electron Microscopy Facility, Harvard Medical School) for providing assistance with electron microscopy, and Victor Barrera Burgos of the Harvard Chan Bioinformatics Core, Harvard T.H. Chan School of Public Health, Boston, MA, for assistance with the analysis of RNA-seq datasets obtained through dbGap. This work was supported through NIH R01HL130336 and funds from the Department of Defense (W81XWH-16-1-0165) and the Tuberous Sclerosis Alliance (50K Crowdfunded Research Challenge) to C. Priolo. Y. Feng was supported by a Postdoctoral Fellowship co-funded by the Tuberous Sclerosis Alliance and The LAM Foundation and a microgrant from the Brigham Research Institute. Work by Victor Barrera Burgos was partially funded by the “HSCI Center for Stem Cell Bioinformatics.” We are grateful to the Engles Program in TSC and LAM Research for supporting publication of this work.

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.