Estradiol and mTORC2 cooperate to enhance prostaglandin biosynthesis and tumorigenesis in TSC2-deficient LAM cells

To examine the possible effects of estradiol on metabolic pathways in TSC2-deficient rat uterus–derived (ELT3) cells, we performed a metabolomic screen. A significant increase in prostaglandins including PGE₂, PGD₂, and 6-keto-PGF_1α, was seen in estradiol-treated cells (Fig. 1 a). Furthermore, estradiol increased p44/42–MAPK T202/Y204 phosphorylation, COX-2 expression, and PGE₂ levels in TSC2-deficient cells at 2 and 24 h (Fig. 1, b and c). Analysis of LAM patient–derived TSC-deficient 621–101 cells (Yu et al., 2004) showed that estradiol also increased PGE₂ levels at 24 h (Fig. 1 d), confirming the results in the TSC2-deficient ELT3 cells.

To define the impact of estradiol on activation of signaling pathways, we first compared the basal levels of phospho-S6, phospho-MAPK, and phospho-Akt S473 for the TSC2^+/+ and TSC2^−/− cells. TSC-deficient cells exhibited lower levels of phospho-Akt S473 and higher levels of phospho-MAPK and phospho-S6, relative to TSC-addbackcells (TSC2⁺; Fig. 1 e). We next performed a time-course analysis of the effect of estradiol on activation of these pathways in both TSC-deficient and TSC-addback LAM-derived cells. We found that estradiol activated Akt S473 within 6 h, MAPK (T202Y204) at 2, 4, 6, 8, and 24 h, but not S6 in TSC2-deficient cells (Fig. 1 f, left). In comparison, estradiol stimulated Akt S473 and MAPK (T202Y204) to a lesser degree in TSC2-addback cells (Fig. 1 f, right). Our data indicate that estradiol activates MAPK and Akt pathways in the absence of TSC2.

To determine whether the effect of estradiol on cells is dependent on TSC defects, we examined the levels of COX-2 using immunoblotting in estradiol-stimulated TSC2-deficient and TSC2-addback cells. Estradiol treatment did not affect COX-2 expression in TSC2-addback LAM patient–derived cells (Fig. 1 g). To address how estradiol exerts its action on COX-2 expression and prostaglandin production, we examined the activation of p44/42–MAPK and PI3K–Akt, which are known pathways promoting COX-2 expression (Wang and Dubois, 2010). We found that estrogen activates both p44/42–MAPK and PI3K–Akt in TSC2-deficient cells, assessed by phosphorylation at T202/204 and S473 sites, respectively, but not in TSC2-addback cells (Fig. 1, g and h). Inhibition of p44/42–MAPK using PD98059 or PI3K–Akt using PI-103 blocked estrogen-enhanced COX-2 expression (Fig. 1, i and j). Collectively, these data indicate that estradiol activates COX-2 expression via p44/42–MAPK and PI3K–Akt pathways.

To determine the effect of estradiol on cellular metabolism in vivo, we used xenograft tumors of TSC2-deficient ELT3 cells (Yu et al., 2009) from placebo or estradiol-implanted ovariectomized female mice, in which p44/42–MAPK phosphorylation was evident (Fig. 1 k). A metabolomic screen showed that levels of PGE₂, PGD₂, and 6-keto-PGF_1α was significantly increased in xenograft tumors from mice treated with estradiol (Fig. 1 l), Estradiol-treated mice bearing ELT3 xenograft tumors also exhibited higher levels of urinary PGE₂ relative to placebo controls (Fig. 1 m). These data demonstrate that estradiol stimulates prostaglandin biosynthesis by TSC2-deficient cells in vitro and in vivo.

Prostaglandins are products of prostaglandin-endoperoxide synthases (PTGSs) 1 and 2, or more commonly known as COX-1 and COX-2 (Fig. 2). COX-1 and COX-2 convert arachidonic acids released from membrane phospholipids into PGH₂. Prostacyclin (PGI₂) is produced by prostacyclin synthase (PTGIS) from PGH₂ (Fig. 2 a). To define the molecular mechanisms responsible for estradiol-enhanced COX-2 expression and prostaglandin production, we analyzed our previous expression array of TSC2-deficient LAM patient–derived cells (Lee et al., 2010) and found that both COX-2 (PTGS2) and prostacyclin synthase (PGTIS) expression were significantly increased, by 2- and 40-fold, respectively (Fig. 2 b and Table 1), in TSC2-deficient cells relative to TSC2-addback cells. To validate the findings of the expression array, we first performed real-time RT-PCR analysis. TSC2-deficient LAM patient–derived cells exhibited a 102-fold increase of PTGS2, and a 15-fold increase of PTGIS (P < 0.0001; Fig. 2 c). Importantly, rapamycin treatment did not affect the levels of PTGS2, but increased the levels of PTGIS by 6.6-fold (P < 0.05; Fig. 2 c), consistent with the changes identified in the expression array profile (Lee et al., 2010; Fig. 2 b). We next examined the protein levels of COX-2 and PTGIS using immunoblotting analysis. COX-2 protein levels were higher in TSC2-deficient cells compared with TSC2-addback 621–101 cells (Fig. 2 d). Rapamycin treatment suppressed phosphorylation of the ribosomal protein p70S6 (S6), but did not affect the level of COX-2 expression, although it altered the migration of COX-2 (Fig. 2 d). COX-2 occurs as 72- and 74-kD isoforms, with the larger form a result of glycosylation at Asn580 (Sevigny et al., 2006). LAM cells treated with tunicamycin, an inhibitor of N-acetylglucosamine transferases, showed a marked reduction in the slower migrating form of COX-2, but showed similar production of PGE₂ (Fig. 2 e). This implies that this gel shift of COX-2 is caused by increased glycosylation.

Prostacyclin synthase (PTGIS) catalyzes the conversion of PGH₂ to prostacyclin (PGI₂). PTGIS protein levels were markedly higher in TSC2-deficient cells compared with TSC2-addback 621–103 cells (Fig. 2 f). Rapamycin treatment suppressed phosphorylation of the ribosomal protein p70S6 (S6), and decreased PTGIS levels in TSC2-deficient ells to a minor extent, but had no appreciable effect on TSC2-addback cells (Fig. 2 f), suggesting that TSC2 also regulates PTGIS expression independent of mTORC1.

To determine the functional impact of altered expression COX-2 and PGTIS on prostaglandin production, secreted levels of PGEM (a stable PGE₂ metabolite₎, 6-keto-PGF_1α (a stable prostacyclin PGI₂ metabolite), and PGF_2α were measured from TSC2-deficient LAM patient–derived cells treated with rapamycin or control. Levels of PGE₂, 6-keto-PGF_1α, and PGF_2α were all significantly higher in TSC2-deficient cells relative to TSC2-addback cells, and were also insensitive to rapamycin treatment (Fig. 2 g). Similar results on COX-2 expression were also seen in TSC2-deficient ELT3 cells (Fig. 2 h). PGE₂ levels were elevated by 2.5-fold in ELT3 cells in comparison to TSC2-addback cells, and this again was not affected by rapamycin treatment (Fig. 2 i). To determine whether this phenomenon of rapamycin-insensitive prostaglandin production was seen in other cancer cells with intact TSC2 levels but mTORC1 activation, we examined HeLa, U2OS, and OVCAR5 cells. Although PGE₂ production was variable among these cell lines, rapamycin did not affect COX-2 expression or PGE₂ production, although it did reduce phospho-S6 (Fig. 2, j and k). Together, these data indicate that up-regulation of COX-2 and prostaglandin production is rapamycin-insensitive in cells with mTORC1 activation.

To determine whether TSC2 regulates COX-2 and prostaglandin production in vivo, we used xenograft tumors from mice inoculated with TSC2-deficient ELT3-V3 (vector-control) cells and TSC2-addback ELT3-T3 cells. COX-2 levels were significantly higher in the ELT3-V3 xenograft tumors with elevated phospho-S6 relative to TSC2-addback ELT3-T3 tumors (Fig. 3 a). In addition, levels of PGE₂ in xenograft tumors were significantly higher in mice with the ELT3-V3 tumors in comparison to mice with the TSC2-addback ELT3-T3 tumors (Fig. 3 b). Moreover, urinary PGE₂ level was significantly higher in mice with the ELT3-V3 tumors in comparison to mice with the TSC2-addback ELT3-T3 tumors (Fig. 3 c). Similar results on PGTIS expression were also observed in xenograft tumors of TSC2-deficient ELT3 cells (Fig. 3 d). Urinary 6-keto-PGF1α levels were also significantly higher in mice with the ELT3-V3 tumors in comparison to mice with the TSC2-addback ELT3-T3 tumors (Fig. 3 e). Together, these data indicate that up-regulation of COX-2 and prostaglandin production is rapamycin-insensitive in cells with mTORC1 activation

To identify additional molecular determinants responsible for rapamycin-insensitive COX-2 expression and prostaglandin biosynthesis, we first treated Tsc2^−/−p53^−/− MEFs with Torin 1, a potent ATP-competitive mTORC1 and mTORC2 inhibitor for 24 h. Torin 1 decreased the level of COX-2 and phosphorylation of S6 (S235/236) and 4EBP1. Interestingly, cells treated with LY294002, a PI3K inhibitor, for 24 h exhibited reduced level of COX-2. However, rapamycin treatment did not affect the level of COX-2 (Fig. 4 a). Our data suggest that mTORC2 might play a role in mediating COX-2 expression. To further test the impact of mTORC2 on COX-2 expression, we generated stable knockdown of Rictor, an mTORC2 component, in Tsc2^−/−p53^−/− MEFs. Rictor silencing profoundly decreased the level of COX-2 and phosphorylation of S6, supporting the notion that mTORC2 is a critical mediator of COX-2 expression in TSC2-deficient cells (Fig. 4 b). Moreover, we found that Rictor knockdown markedly reduced phosphorylation of Akt (S473), a direct target of mTORC2 (Fig. 4 c), indicating the potential impact of mTORC2–Akt axis on COX-2 expression in Tsc2^−/−p53^−/− MEFs. Why does 4a look so different from 4d? Note that COX-2 is reduced a little with Torin1 in 4a, whereas it’s gone in 4d.

To examine the effect of Torin 1 on COX-2 expression in LAM patient–derived cells, we treated 621–101 TSC22 cells with Torin 1 for 24 h. Torin 1 markedly decreased the level of COX-2. However, rapamycin treatment did not affect COX-2 expression in TSC2-deficient 621–101 cells. NS398, a COX-2 inhibitor, strongly decreased the levels of COX-2 in TSC2-deficient cells (Fig. 4 d). Interestingly, Torin 1 treatment decreased phosphorylation of both p44/42–MAPK T202/Y204 and PI3K–Akt S473 (Fig. 4 e). To prove that Akt is a critical mediator of COX-2 expression in LAM patient–derived cells, we treated 621–101 cells with PI-103 (a dual inhibitor of PI3K and mTOR) or Akt-VIII (a selective Akt inhibitor), for 24 h, and found that both inhibitors strongly decreased the levels of COX-2 without affecting COX-1 (Fig. 4 f).

It has been reported that TSC2 does not regulate mTORC2 activation (Dalle Pezze et al., 2012). We have found that mTORC2 is involved in COX-2 expression in TSC2-deficient cells. In another experiments, we found that COX-2 expression was decreased partially by treatment of PD98059 (MEK1/2 inhibitor) or PI-103 (Akt inhibitor), but more strongly by the combination of PD98059 and PI-103 (Fig. 4 g), suggesting that both MEK1/2-MAPK and PI3K–Akt pathways contribute to COX-2 expression. Collectively, our data indicate that mTORC2-Akt plays a critical role in mediating COX-2 expression in TSC2-deficient cells.

To further define the mechanism by which TSC2 acts as a negative regulator of COX-2 protein expression, we focused on epidermal growth factor receptor (EGFR), a known regulator of COX-2 expression and prostaglandin production (Mann et al., 2005; Brand et al., 2013). We found that TSC2-deficient cells treated with the anti-EGFR agents Afatinib and Gefitinib exhibited decreased levels of COX-2 protein, compared with vehicle control (Fig. 4 h). Interestingly, using confocal microscopy, we observed that TSC2 deficiency resulted in abundant nuclear localization of EGFR in comparison with TSC2-addback in two cell types, rat uterine leiomyoma–derived and LAM patient–derived cells (Fig. 4 i). Furthermore, EGFR nuclear localization was not affected by rapamycin treatment (Fig. 4 i), supporting rapamycin-insensitive up-regulation of COX-2 expression and prostaglandin production. Concomitantly, treatment with two PI3K–Akt inhibitors, Akt VIII or PI-103, suppressed COX-2 expression (Fig. 4 f, g). Together, these data support a model in which TSC2 negatively regulates COX-2 expression via effects on nuclear and cytoplasmic localization of EGFR.

To determine whether inhibition COX-2 impacts the proliferation of TSC2-deficient cells, we treated 621–101 cells with Sulindac (a COX-1 inhibitor), NS398 (a COX-2 inhibitor), or aspirin (an irreversible COX-1 and COX-2 inhibitor that works by blocking enzymatic activities, and also acetylates COX-2, resulting in production of 15-epi-LXA₄, an antiinflammatory/antitumor compound) for 24 h. NS398 and aspirin reduced COX-2 and COX-1 levels without affecting phosphorylation of p44/42–MAPK or S6 (Fig. 4 j). Aspirin significantly decreased PGE₂ levels (Fig. 4 k) and reduced proliferation of TSC2-deficient 621–101 cells (Fig. 4 l).

To determine the efficacy of inhibition of COX-2 in vivo, first, we used the COX-2–specific inhibitor Celecoxib to treat Tsc2^+/− mice, which develop spontaneous renal cystadenomas at age of 4 mo. We found that treatment with Celecoxib strongly suppressed microscopic renal lesions in Tsc2^+/− mice by ∼50% after 4 mo of treatment (0.1% wt/wt in chow) relative to vehicle control (Fig. 5, a and b). These data indicate that COX-2 is responsible for renal tumorigenesis.

We next assessed the possible benefit of aspirin in a xenograft tumor model of TSC2-deficient ELT3-luciferase-expressing cells. Aspirin treatment for 3 wk decreased the intensity of bioluminescence (Fig. 5, c and d), and decreased the tumor size (Fig. 5 e). Tumors also had reduced expression of COX-2 and c-Myc, and increased levels of cleaved-caspase-3 and cleaved-PARP (Fig. 5 f). Aspirin-treated mice bearing ELT3 xenograft tumors had markedly reduced urinary levels of PGE₂ (Fig. 5 g), consistent with strong inhibition of each of COX-1 and COX-2, as expected (Fig. 4 j). These data indicate that aspirin reduces PGE₂ production and tumor growth in this TSC2-deficient xenograft model.

We then examined whether these in vitro and xenograft model findings were relevant to human LAM. Accordingly, LAM lungs were found to express higher levels of COX-2 (PTGS2) in comparison with control lungs (Fig. 6 a). Furthermore, immunohistochemistry analyses showed that COX-2 expression was specifically increased in pulmonary LAM nodules (arrows), which also expressed both smooth muscle actin and phospho-S6 (Fig. 6 b). To examine the functional effects of COX-2 expression in LAM, 15-epi-lipoxin A₄ (LXA₄), a product of aspirin-acetylated COX-2, was measured in exhaled breath condensate (EBC) from three LAM subjects (Table 2). LXA₄ was detected in EBCs and the levels were increased with aspirin (Fig. 6 c). LXA₄ decreases proliferation of A549 cells (Human lung adenocarcinoma; Clària et al., 1996) and also reduced the growth of LAM patient–derived 621–101 cells at 100 nM (Fig. 6 d). In a separate experiment, the 15-epi-LXA₄ integrity was confirmed by UV-Vis spectrophotometry to ensure the presence of the diagnostic tetraene chromophore and accurate quantitation and HPLC to ensure that only a single peak was present without evidence for isomerization. (Fig. 6 e). We next performed a dose–response of LXA₄ and found that LXA₄ at 500 nM exhibited the strongest inhibitory effect on cell growth (Fig. 6 e).

Urinary PGE₂ levels were measured in 29 LAM patients and in 18 healthy women; however, these levels were not significantly different between two groups (Fig. 6 f). Because renal prostaglandin production can significantly influence urinary prostaglandin levels, we next measured PGE₂ and 6-keto-PGF_1α levels in sera from 14 LAM patients and 13 healthy women. The mean serum PGE₂ level of LAM patients (27.8 ± 1.8 pg/ml) was higher than that of healthy women (19.6 ± 1.4 pg/ml; P = 0.0021; Fig. 6 g). In addition, the mean serum 6-keto-PGF1α level of LAM patients (192.2 ± 64.9 pg/ml) was also higher compared with the mean in healthy women (82.6 ± 6.3 pg/ml; P = 0.0006; Fig. 6 h).

LAM is a neoplastic disorder in which smooth muscle-like cells with biallelic inactivation of TSC2 leads to hyperactivation of mTORC1. This fundamental mechanistic insight led to the recent reported success in treating LAM patients with rapamycin. However, mTORC1-independent effects of TSC2 loss are also known (Yu and Henske, 2010; Neuman and Henske, 2011). The marked female predominance of LAM suggests that estradiol promotes disease progression. In this study, we identified an estradiol-induced prostaglandin metabolic signature in TSC2-deficient ELT3 cells in vitro and in vivo. Prostaglandins, synthesized via COX-1/COX-2, appear to play important roles in cancer progression (FitzGerald and Patrono, 2001; Müller, 2004; Wang et al., 2007; Wang and Dubois, 2010), although their precise role is still hotly debated. Estradiol-dependent prostaglandin production has been studied in other models. Estradiol activated COX-2 and increased prostacyclin synthesis in mouse aortic smooth muscle cells (Egan et al., 2004). Increased levels of PGE₂ causes enhanced aromatase expression and local estradiol production in breast tissue (Subbaramaiah et al., 2012). However, the relationship between prostaglandins and TSC/LAM has not been investigated previously. Remarkably, we found that estradiol enhances the expression of COX-2 and induces production of PGE₂ in TSC2-deficient cells in vitro and in vivo. Furthermore, we identified a novel function of TSC2 as a negative regulator of COX-2 expression and prostaglandin biosynthesis. Interestingly, this regulation appears to be rapamycin-insensitive but sensitive to Akt inhibition and Rictor knockdown, suggesting that COX-2 may be regulated by Akt in TSC2-deficient cells.

We also demonstrated that COX-2 is abundant in LAM lesions, and that serum levels of PGE₂ are elevated in LAM patients. Collectively, our data suggest that COX-2 may play an important role in LAM pathogenesis.

NS398 and aspirin are direct enzyme inhibitors of COX-2 and/or COX-1. As expected, we have found that cells treated with NS398 or aspirin exhibited reduced levels of PGE₂ (Fig. 4 k), consistent with this inhibition. However, we also found that both NS398 and aspirin decreased the levels of COX-2 protein. Although this was initially surprising to us, multiple experiments confirmed this effect, and our results are in agreement with previous reports that COX-2 protein expression is reduced by classical nonsteroidal antiinflammatory drugs (NSAIDs), including aspirin and NS398 (Takada et al., 2004; Galamb et al., 2010). Therefore, we think that both enzymatic inhibition and reduction in expression of COX-2 occurs in response to NS398 and aspirin treatment. Aspirin, the prototypical NSAID, covalently modifies both COX-1 and COX-2 by acetylation. We demonstrated that aspirin treatment significantly reduced the growth of xenograft tumors of TSC2-deficient ELT3 cells, and led to reduced urinary levels of PGE₂ and 6-keto-PGF1α, which is consistent with strong inhibition of COX-1/COX-2 function. Furthermore, xenograft tumors from aspirin-treated mice exhibited higher levels of apoptosis compared with vehicle treatment (Fig. 5 f). Although aspirin-acetylated COX-2 inhibits prostaglandin formation, the enzyme is not completely inactivated. Rather, aspirin-acetylated COX-2 catalyzes the conversion of arachidonate to 15-epi-LXA₄ (Clària et al., 1996). In this manner, aspirin both inhibits prostaglandin production and triggers the formation of 15-epi-LXA_4, a potent inhibitor of malignant cell proliferation (Clària et al., 1996). 15-epi-LXA₄ was present in EBCs from LAM patients, and was increased with oral aspirin ingestion. 15-epi-LXA₄ also decreased LAM-patient–derived cell proliferation (Fig. 6, d and e) Together, these findings suggest that aspirin may have rapamycin-independent beneficial effects in LAM.

LAM is often a progressive disease which leads to respiratory failure and death in the absence of lung transplantation. The recent demonstration that rapamycin has clinical benefit in LAM is a major success. However, not all patients respond to rapamycin, and upon rapamycin withdrawal, lung function decline resumes (McCormack et al., 2012). Hence, lifelong treatment of LAM patients with rapamycin may be required to maintain benefit, with unknown long-term toxicities. Our findings suggest that aspirin and/or other COX-1/COX-2 inhibitors may have significant benefit in slowing LAM progression. The well-known side-effect and toxicity profile of these drugs make them attractive candidates for long-term therapy in LAM patients. It is also possible that other neoplastic conditions associated with mTOR hyperactivation could be responsive to these agents. Further preclinical and clinical investigation is warranted to explore these possibilities.

ELT3 cells (Eker rat uterine leiomyoma-derived cells; Howe et al., 1995a,b) and LAM patient–derived 621–101, 621–102, and 621–103 cells were cultured in IIA complete medium. Tsc2^−/−p53^−/− MEFs, HEK293, HeLa, U2OS, and OVARC5 cells were cultured in DMEM supplemented with 10% FBS. 17-β-estradiol (10 nM; Sigma-Aldrich), rapamycin (20 nM; Biomol), Torin 1 (250 nM; Tocris), LY294002 (20 µM; Cell Signaling Technology), NS398 (50 µM; Cayman Chemical), Sulindac (50 µM; Cayman Chemical), aspirin (450 µM; Sigma-Aldrich), Celecoxib (Novartis), PI-103 (5 µM; Tocris), and AktVIII (5 µM; Millipore), and Wortmannin (1 µM; Cell Signaling Technology) were used as indicated. 15-epi-LXA₄ (10–500 nM) was obtained from Millipore. The integrity of 15-epi-LXA₄ was determined by UV-Vis spectrometry and HPLC. Because of the volatile nature of this product, stock solutions were stored at −80°C.

All animal work was performed in accordance with protocols approved by the IACUC-CHB. For xenograft tumor establishment, 2 × 10⁶ ELT3-luciferase cells were inoculated bilaterally into the posterior back region of female intact CB17-SCID mice (Taconic) as previously described (Yu et al., 2009; Parkhitko et al., 2011; Liu et al., 2012). 5 wk after cell inoculation, mice bearing subcutaneous tumors were randomized into two groups: vehicle control (n = 5) and aspirin (n = 5; 100 mg/kg/day, in drinking water). Tumor area (width × length) was measured weekly using calipers. Urine specimens were collected from mice for PGE₂ and 6-keto-PGF1_α measurement. For spontaneously arising renal tumor model, Tsc2^+/− mice, originally generated in this laboratory (Onda et al., 1999), were serially crossed with A/J mice for over five generations. These pure strain mice were used in all experiments. Celecoxib (Novartis) was administered by mouse chow (0.1%) for 4 mo, beginning at 1 mo of age, and then sacrificed for renal tumor assessment at the end of treatment (5 mo).

Mouse kidneys were removed rapidly after euthanasia and fixed overnight in 10% formalin. Kidneys were then prepared for histological evaluation in stereotypical fashion by cutting the kidney into sections at 1-mm sections.

Microscopic kidney tumor scores were determined in a semiquantitative fashion by a single blinded observer. The set of 1-mm sections were prepared as H&E-stained 8-µm sections. Each tumor or cyst identified was measured to determine its length and width in two dimensions, as well as the percentage of the lumen filled by tumor (this was 0% for a simple cyst, and 100% for a completely filled, solid adenoma). These measurements were converted into a measurement of tumor volume per lesion using the following formula: tumor volume = maximum (tumor percent, 5)/100 × 3.14159/6 × 1.64 × (tumor length × tumor width) **1.5 (Auricchio et al., 2012). The total tumor volume per kidney was then equal to sum of the tumor volume of each lesion identified. Comparisons between sets of mice for tumor measurements were made using the nonparametric Mann-Whitney test in Prism (GraphPad Software, Inc.).

10 min before imaging, animals were injected with luciferin (120 mg/kg; i.p.; Xenogen). Bioluminescent signals were recorded using the Xenogen IVIS System. Total photon flux of tumors was analyzed as previously described (Yu et al., 2009).

ELT3 cells and LAM-derived cells were plated on glass coverslips in 12-well tissue culture plates overnight. Cells were serum starved overnight, and then treated with 20 nM rapamycin for 24 h. Cells were rinsed with PBS twice, fixed with warm 4% paraformaldehyde, permeabilized with 0.2% Triton X-100, blocked in 3% BSA/PBS for 1 h, incubated with primary antibody 1% BSA in PBS for 1 h at room temperature, and then incubated with secondary antibodies for 1 h. Images were captured with a FluoView FV-10i Olympus Laser Point Scanning Confocal Microscope.

Re-analysis of previously published expression array data (GEO accession no. GSE16944; Lee et al., 2010) was performed using the online tool GEO2R. Transcript levels of PTGS2 (COX-2; ID: 238018) and PTGIS (prostacyclin synthase; ID: 454010) were compared between TSC2-deficient and TSC2-addback (TSC2⁺) cells, or rapamycin- and vehicle-treated TSC2-deficient cells.

RNA from cultured cells and xenograft tumors was isolated using RNeasy Mini kit (QIAGEN). Gene expression was quantified using One-Step qRT-PCR kits (Invitrogen) in the Applied Biosystems Step One Plus Real-Time PCR System and normalized to β-actin.

Cells were lysed in m-PER buffer (Thermo Fisher Scientific). The following antibodies were used: COX-1, COX-2, EGFR, phospho-p44/42–MAPK (T202/Y204), phospho-S6 (S235/236), phospho-Akt (S473), cleaved caspase 3, and cleaved PARP (Cell Signaling Technology); tuberin and c-Myc (Santa Cruz Biotechnology, Inc.); smooth muscle actin (BioGenex); β-actin (Sigma-Aldrich); and Alexa Fluor 488 goat anti–rabbit IgG (H+L) antibody (Invitrogen).

100 µg of frozen biopsy tissue was submitted to Metabolon, Inc. for sample extraction and analysis. In brief, Metabolon performed cold methanol extraction of mechanically disaggregated tissue samples and these extracts were split into three aliquots. The reproducibility of the extraction protocol was assessed by the recovery of xenobiotic compounds spiked into every tissue sample before extraction. These aliquots were processed and characterized by one of the three analytical methods previously described (Evans et al., 2009): UHPLC-ESI-MS/MS in the positive ion mode, UHPLC-ESI-MS/MS in the negative ion mode, and sialylation followed by GC-EI-MS. Chromatographic timelines were standardized using a series of xenobiotics that elute at specified intervals throughout each chromatographic run. The technical variability of each analytical platform was assessed by repeated characterization of a pooled standard that contained an aliquot of each sample within the study. Metabolon’s global screening platform utilizes a nontargeted approach (Evans et al., 2009). For LC-MS/MS, after peak identification and quality control filtering (signal greater than 3x background, retention index within a prespecified platform-dependent window, quant ions to library match within 0.4 m/z, and the MS/MS forward and reverse scores), the metabolites’ relative concentrations were obtained from median-scaled day-block normalized data for each compound. In vitro data were normalized to total protein levels as measured by Bradford assay. Eicosanoids identified by the Metabolon platform in this study were measured by LC-MS/MS run in the negative mode. Fragmentation of the molecule’s quant ion generated a pattern of ions that matched in mass and relative intensities to a purified standard as described by DeHaven et al. (2010). In this study, the data were derived using a nontargeted method that does not incorporate labeled standards. However, in cases where targeted methods have been created and used to validate Metabolon screening platform, the results from the screening platform have shown a high degree of correlation with the targeted/absolute quantification methods and clinical measurements, when available (Sreekumar et al., 2009; Zhang et al., 2011).

Cell viability was determined by MTT assay (Sigma-Aldrich).

293T packaging cells were transfected with Rictor shRNA or non-targeting shRNA vectors using Mirus Trans-IT TKO Transfection reagent (Mirus). Tsc2^−/−p53^−/− MEFs were infected with lentivirus containing Rictor shRNA, or non-Targeting shRNA. Cells were harvested 48 h after transfection and then selected against puromycin. Stable clones were harvested for future experiments.

Sections were deparaffinized, incubated with primary antibodies and biotinylated secondary antibodies, and counterstained with Gill’s Hematoxylin.

Levels of PGEM (a stable metabolite of PGE₂), 6-keto-PGF_1α (a stable metabolite of prostacyclin), and creatinine were measured using ELISA kits (Cayman Chemical). Levels of secreted prostaglandins were normalized to vehicle control and expressed as fold change. Urinary levels of prostaglandins were normalized to creatinine levels and expressed as nanograms per milliliter.

Statistical analyses were performed using Student’s t test when comparing two groups for in vitro and in vivo studies. Welch’s t tests and Wilcoxon rank sum tests were performed for metabolomic profiling. Two-way ANOVA test was performed in xenograft tumor-aspirin studies. Mann-Whitney tests were used for prostaglandin quantification in clinical data.

We are grateful for Dr. C Walker (Texas A&M Health Science Center) for providing ELT3 cells, Ms. E. Peters (The Center for LAM Research and Clinical Care - Brigham and Women’s Hospital) for human specimen acquisition, Mr. B. Ith, and Dr. M Perrella for specimen preparation, Dr. C. Jiang (Peking Union Medical College, China) for providing access to research facility, and Dr. I. Rosas for providing non-LAM lung tissues. We thank Dr. A. Choi for valuable discussion and critical review of the manuscript. We gratefully acknowledge LAM patients and healthy volunteers for their valuable contribution to this study.

A. Csibi and X. Gu are LAM Foundation postdoctoral fellows. J. Li is a Tuberous Sclerosis Alliance postdoctoral fellow. J. Blenis is a LAM Foundation established investigator. This study is supported by The LAM Foundation, The Adler Foundation, The LAM Treatment Alliance (to E.P. Henske)., National Institutes of Health Grant GM51405 (to J. Blenis), the National Heart Lung and Blood Institute grants HL68669 (to B.D. Levy) and HL118760 (to E.P. Henske), the National Cancer Institute grants 1P01CA120964 (to D. Kwiatkowski) and HL098216 (to J.J. Yu), the Department of Defense Exploratory Idea Development Award (W81XWH-12-1-0442 to J.J. Yu), and Biomedical Research Institute-BWH Microgrants to C. Li and J.J. Yu.

B.D. Levy is an inventor on patents on lipoxins assigned to Brigham and Women’s Hospital, some of which are licensed for clinical development. The interests of B.D. Levy were reviewed and are managed by the Brigham and Women’s Hospital and Partners HealthCare in accordance with their conflict of interest policies. The other authors have no conflicting financial interests.