Sphingosine-1-Phosphate as a Regulator of Hypoxia-Induced Factor-1α in Thyroid Follicular Carcinoma Cells

Sphingosine-1-phosphate (S1P) is a bioactive lipid, which regulates several cancer-related processes including migration and angiogenesis. We have previously shown S1P to induce migration of follicular ML-1 thyroid cancer cells. Hypoxia-induced factor-1 (HIF-1) is an oxygen-sensitive transcription factor, which adapts cells to hypoxic conditions through increased survival, motility and angiogenesis. Due to these properties and its increased expression in response to intratumoral hypoxia, HIF-1 is considered a significant regulator of tumor biology. We found S1P to increase expression of the regulatory HIF-1α subunit in normoxic ML-1 cells. S1P also increased HIF-1 activity and expression of HIF-1 target genes. Importantly, inhibition or knockdown of HIF-1α attenuated the S1P-induced migration of ML-1 cells. S1P-induced HIF-1α expression was mediated by S1P receptor 3 (S1P3), Gi proteins and their downstream effectors MEK, PI3K, mTOR and PKCβI. Half-life measurements with cycloheximide indicated that S1P treatment stabilized the HIF-1α protein. On the other hand, S1P activated translational regulators eIF-4E and p70S6K, which are known to control HIF-1α synthesis. In conclusion, we have identified S1P as a non-hypoxic regulator of HIF-1 activity in thyroid cancer cells, studied the signaling involved in S1P-induced HIF-1α expression and shown S1P-induced migration to be mediated by HIF-1.

A physiological concentration of S1P strongly increases migration of the ML-1 follicular thyroid cancer cell line [16], an effect which may have contributed to metastasis of the original tumor. We have also shown S1P and vascular endothelial growth factor (VEGF) signaling to cross-communicate in many ways in ML-1 cells. For example, S1P treatment increases both vascular endothelial growth factor receptor 2 (VEGFR-2) expression and VEGF-A secretion while inhibition of VEGFR-2 attenuates several S1P-induced effects [17], [18]. Since S1P and HIF-1 have many similar functions, we investigated whether extracellular S1P is able to affect HIF-1a expression in ML-1 cells. Interestingly, we were able to induce HIF-1a expression in normoxia with pro-migratory, physiological S1P concentrations. This finding led to several questions: does S1P also increase HIF-1 acivity, does S1Pinduced HIF-1a mediate S1P-induced migration, what are the signaling pathways involved and what is the mechanism of HIF-1a up-regulation.
In the present study we identify S1P as a non-hypoxic inducer of HIF-1a expression in thyroid cancer cells. S1P increases HIF-1 activity and HIF-1 is involved in S1P-induced migration. Additionally, we show that S1P regulates HIF-1a protein level through a signaling pathway including S1P 3 , G i , PI3K, mammalian target of rapamycin (mTOR), MAP kinase kinase (MEK) and protein kinase C bI (PKCbI). We suggest S1P to regulate HIF-1a stability by a pVHL-independent mechanism and HIF-1a synthesis through activation of translational regulators eIF-4E and p70S6K.

Cell Culture
ML-1 human follicular thyroid cancer cells were a kind gift from Dr. Johann Schönberger (University of Rosenburg, Germany). They were cultured in DMEM supplemented with 10% Bovine Serum Albumin (FBS), 2 mM L-glutamine and 100 U/ml penicillin/streptomycin. FTC-133 human follicular thyroid cancer cells were from Banca Biologica e Cell Factory, National Institute for Cancer Research (Genova, Italy). They were grown in Ham's medium and DMEM (1:1) supplemented with 10% FBS, 2 mM Lglutamine and 100 U/ml penicillin/streptomycin. Cells were cultured at 37uC in a water-saturated atmosphere containing 5% CO 2 and 95% air. During hypoxia experiments cells were incubated in an In vivo2 hypoxia workstation (Ruskinn Technology, Bridgend, UK) with 1% oxygen at 37uC. Before treatment with S1P, cells were lipid-starved in medium containing 5% charcoal/dextran treated FBS (lipid-stripped FBS). For migration experiments cells were serum-starved in medium containing 0.2% fatty acid-free BSA (serum-free medium).

Western Blotting
Cells were lipid-starved overnight before treatment. Whole cell lysates were obtained and Western blotting performed according to a protocol described elsewhere [17]. Proteins were detected with enhanced chemiluminescence using the Western Lightning Plus-ECL kit. Hsc70 or b-actin was used as a loading control. Levels of phosphorylated or hydroxylated proteins were normalized with the non-phosphorylated or non-hydroxylated form and with the loading control. Densitometric analysis of protein bands was done with the ImageJ program (http://rsbweb.nih.gov/ij/).

Cell Migration and Haptotaxis
Cellular migration and haptotaxis was studied with 6.5 mmdiameter Transwell Permeable Support inserts with 8-mm pore size. The protocols have been described elsewhere [16,17,18].

Proliferation
Cellular proliferation was studied with a [ 3 H]thymidine incorporation assay. Cells were lipid-starved overnight before treatment and the experiments were performed according to a protocol described elsewhere [16,17].

RNA Extraction, Reverse Transcriptase PCR and Quantitative Real-time PCR
RNA was isolated using the Aurum Total RNA Mini kit and RNA concentrations were determined using the RiboGreen RNA Quantitation Reagent. Reverse transcriptase PCR was performed with SuperScript III Reverse Transcriptase to produce cDNA. The quantitative PCR assays for HIF-1a, VEGF-A, AMF, TGFa and HPRT1 were designed using the Universal ProbeLibrary Assay Design Center (www.roche-applied-science.com). GAPDH and HPRT1 were used as reference genes. The primer and probe information are in Table S1. Reaction mixtures were prepared with ABsolute QPCR Rox Mix or with the KAPA Probe Fast qPCR Kit and real-time quantitative PCR was performed using the Applied Biosystems 7900HT Fast Sequence Detection System or the StepOnePlus Real-Time PCR system. The amplification results were analyzed with the SDS and RQ Manager programs (Applied Biosystems).

Luciferase Assays
Cells were co-transfected with a total of 20 mg of either TK-Luc or HRE-Luc plasmid together with a Ubi-Renilla plasmid. The HRE-Luc plasmid was from Addgene (plasmid 26731; [49]. The TK-Luc and HRE-Luc plasmids contain a TK or HRE promoter and the firefly luciferase gene whereas Ubi-Renilla contains the Ubi promoter and the Renilla Reniformis luciferase gene. Firefly luciferase luminescence was normalized with Renilla luciferase luminescence. Transfection was done with an Amaxa electroporation device and Amaxa Cell Line Optimization Nucleofector Kit according to the manufacturer's instruction. 24 h after transfection the cells were lipid-starved and the next day treated with S1P (100 nM) or CoCl 2 (150 mM) for 7 h. Luminescence was measured with the DualGlo Luciferase Assay System according to the manufacturer's instructions.

Immunoprecipitation
Lysates for immunoprecipitation (IP) were made with IP lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% NP-40, 0.2 mM PMSF, 0.5 mg/ml leupeptin). Lysates were adjusted to equal protein amount and volume and pre-cleaned with 20 ml of Protein A/G PLUS-agarose beads for 1 h at 4uC. Pre-cleaned lysates were incubated with 2 mg of antibody or IgG control overnight at 4uC and the next day incubated with 40 ml of Protein A/G PLUS-agarose beads for 2 h at 4uC. The agarose beads were washed five times with IP washing buffer (50 mM Tris-HCl pH 7.5, 250 mM NaCl, 0.1% NP-40), Laemmli sample buffer was added and the samples boiled.

Statistical Analysis
HIF-1a half-lives were determined with a non-linear curve fit of Chx chase data using the one phase exponential decay equation. Half-lives were calculated as ln(2)/k, where k is the rate constant, and compared with an extra sum-of-squares F test. For other experiments the data is presented as mean 6 SEM for at least three independent experiments and either Student's t-test, oneway ANOVA with Dunnett's post hoc test or one-way ANOVA with Bonferroni's post hoc test was used for statistical analysis. Analysis was performed and graphs were created with the GraphPad Prism 4 program (San Diego, CA, USA).

S1P is a Non-hypoxic Regulator of HIF-1a Expression
Since S1P treatment of ML-1 thyroid cancer cells strongly increases their migration [16] and HIF-1 is a known regulator of invasion and metastasis [7], [8], we investigated whether S1P could affect expression of the regulatory HIF-1a subunit in ML-1 cells. We found that S1P up-regulated HIF-1a protein in a timeand concentration dependent manner in normoxic conditions (Figs. 1A and 1B). As expected, hypoxia (1% O 2 ) up-regulated HIF-1a in ML-1 cells (Fig. 1C). Hypoxia-induced HIF-1a expression was stronger than S1P-induced expression but the kinetics of HIF-1a increase was similar in both cases. S1P did not affect HIF-2a protein expression (results not shown). To determine whether S1P-induced HIF-1a expression is a common feature in follicular thyroid cancer cells, we treated FTC-133 cells with S1P. S1P up-regulated HIF-1a in a time-dependent manner in these cells also (Fig. S1A).
These results show that S1P stimulates HIF-1a expression via S1P 3 and G i and their downstream effectors PKCbI, MEK, PI3K and mTOR and in ML-1 cells.

Effect of S1P on HIF-1a Synthesis and Stability
We attempted to determine whether S1P increases synthesis or stability of HIF-1a. The HIF-1a protein is up-regulated by S1P within 3 h but we saw no effect on HIF-1a mRNA during a 6-h treatment (Fig. S3A). Interestingly, we did see a small but significant increase in HIF-1a mRNA after 9 h of S1P treatment (Fig. S3B). We performed a classical chase experiment with the translation inhibitor cycloheximide (Chx) in order to compare the half-lives of basal, S1P-, hypoxia-, and CoCl 2 -induced HIF-1a (Fig. 4A). Half-life of basal HIF-1a was significantly lower than that of S1P-induced HIF-1a (0.4 and 3.0 min, respectively, **P , 0.01 with an extra sum-of-square's F test), indicating that S1P increases HIF-1a stability. Hypoxia and CoCl 2 were used as controls which are known to stabilize HIF-1a. Accordingly, halflives of hypoxia-and CoCl 2 -induced HIF-1a were high (9.7 and 41.5 min, respectively). We attempted to use [ 35 S]methionine pulse-chase labeling as an additional method to determine HIF-1a half-lives but were not able to immunoprecipitate sufficient amounts of labeled HIF-1a. When cells were pretreated with proteasome inhibitor MG-132 (20 mM, 1 h) to prevent HIF-1a degradation, S1P was not able to elevate the HIF-1a level (Fig.  S4A) also suggesting that S1P may affect HIF-1a stability. However, the approximately two-fold increase caused by S1P may have been lost during the over tenfold up-regulation seen in response to MG-132 treatment.
Taken together these results provide evidence for a S1P-induced effect on both synthesis and stability of HIF-1a. It is possible that both mechanisms are involved. Also, although HIF-1a transcription is not responsible for the initial HIF-1a up-regulation, it may mediate prolonged HIF-1a expression.
We also conducted migration experiments in hypoxia. We determined whether hypoxia could affect expression of the S1P receptors controlling migration. S1P 1 protein expression was increased in hypoxic conditions while S1P 2 and S1P 3 were not affected (Fig. 7A). However, hypoxia did not increase basal or S1P-induced migration or haptotaxis (Fig. 7B). Changes in proliferation did not interfere with the migration experiments since hypoxia did not decrease ML-1 proliferation during a 48-h incubation (Fig. 7C).

Discussion
In the current study we identify S1P as a non-hypoxic inducer of HIF-1a expression in thyroid cancer cells. We show that S1P increases HIF-1 activity and that HIF-1 mediates S1P-induced cell migration. We also present putative signaling pathways leading from extracellular S1P to increased HIF-1a.
Several studies have shown hypoxia to increase sphingosine kinase expression and activity [12], [30][31][32][33] and according to Ader et al. [12], SK1 regulates hypoxia-induced stabilization of HIF-1a via Akt and GSK3. S1P has also been shown to regulate HIF-1a transcription in mouse T cells [34] and macrophages [35]. The most relevant studies in comparison to our work are the identification of S1P as a non-hypoxic regulator of HIF-1a expression in vascular endothelial and smooth muscle cells [36] and in HepG2 liver carcinoma cells [37]. The focus of the former study was on the regulatory role of S1P and HIF-1a in the vascular system. In vascular cells S1P increased stability of the HIF-1a protein via activation of the anti-migratory S1P 2 but independently of G i proteins. In contrast, S1P-induced HIF-1a expression in ML-1 thyroid cancer cells is mediated by the pro-migratory S1P 3 as well as G i . In the latter study the focus was on identification of a S1P derivative (NHOBTD) and its effect on angiogenesis. They show S1P to increase HIF-1a expression in HepG2 cells and NHOBTD to prevent both S1P-induced HIF-1a up-regulation and S1P-induced VEGF secretion presumably mediated by HIF-1. In comparison to these studies we have also investigated signaling pathways mediating HIF-1a up-regulation and show S1P-induced HIF-1a expression to have a functional outcome in increased migration.
Burrows et al. [38] have compared basal and hypoxia-induced HIF-1a expression levels in normal thyroid tissues, primary thyroid tumors and thyroid cancer cell lines, including the follicular WRO and FTC-133 cell lines. They showed HIF-1a expression to be elevated in thyroid carcinomas and to correlate with malignancy, making it a potential target for thyroid cancer therapy. That we now identify S1P as a non-hypoxic regulator of HIF-1a in follicular ML-1 and FTC-133 cells suggests that S1Pinduced HIF-1a expression may be involved in thyroid tumor formation and cancer progression.
One central aim of the current study was to investigate the signaling leading to HIF-1a regulation. We found S1P 3 and G i to mediate S1P-induced HIF-1a expression via PKCbI, MEK, PI3K and mTOR. Additionally, we show S1P to activate translational regulators eIF-4E and p70S6K. While the MEK/ERK and PI3K/ Akt/mTOR cascades are known to regulate HIF-1a translation [7], PKC has been implicated in controlling HIF-1a transcription [39]. However, the initial S1P-induced HIF-1a up-regulation in ML-1 cells was not due to increased transcription. We have previously shown S1P-induced ERK1/2 phosphorylation in ML-1 cells to be mediated by PKCa rather than PKCbI [20]. Therefore, the exact role of PKCbI in S1P-induced HIF-1a expression remains unknown. Zhang et al. [40] showed nicotine-induced HIF-1a accumulation to be mediated by classical PKC isoforms as well as phosphorylation of Akt, ERK, 4E-BP1 and p70S6K in lung cancer cells. Therefore, nicotine may regulate HIF-1a expression in a similar PKC-dependent manner in lung cancer cells as S1P does in ML-1 cells. The signaling behind S1P-evoked HIF-1a also resembles IGF-1-induced HIF-1a expression in colon cancer cells and angiotensin II-evoked HIF-1a expression in vascular smooth muscle cells [39], [41]. Whether phosphorylation of eIF-4E actually activates it has been a controversial subject [21], [22], [42][43][44] but nonetheless, phosphorylation of 4E-BP1 is sufficient to activate eIF-4E [21].
We performed numerous experiments in order to determine whether S1P regulates HIF-1a synthesis or stability. According to protein half-life measurements S1P treatment stabilizes HIF-1a. The half-life of basal normoxic HIF-1a is commonly considered to be approximately 5 min but in ML-1 cells this half-life was as low as 0.4 min. Moroz et al. [45] have studied kinetics of HIF-1a degradation and showed the half-life of normoxic HIF-1a to be 3-6 min in their cell lines. Obviously, exact protein half-life is cell Figure 4. S1P stabilizes HIF-1a independently of pVHL binding. (A) S1P prolongs HIF-1a half-life. Cells were either left untreated, treated with S1P (100 nM) for 6 h, incubated in hypoxia (1% O 2 ) for 6 h or treated with CoCl 2 (150 mM) for 3 h before the cycloheximide chase (Chx, 5 mg/ml). S1P, hypoxic conditions or CoCl 2 were present throughout the chase. Time points are mean 6 SEM, n = 3-10. Curve fit was done with the one phase exponential decay equation. (B) S1P does not inhibit binding of pVHL to HIF-1a. Cells were treated with S1P (100 nM) for 6 h. The level of coimmunoprecipitated HIF-1a was compared with the level of immunoprecipitated pVHL and IgG bands were used as a loading control. **P , 0.01 indicates statistically significant difference between S1P treatment and vehicle control. doi:10.1371/journal.pone.0066189.g004 line specific. Half-lives of S1P-, hypoxia-and CoCl 2 -induced HIF-1a reflect the level of HIF-1a up-regulation seen in ML-1 cells: hypoxia and CoCl 2 are several fold stronger inducers of HIF-1a expression in ML-1 cells than S1P. We saw a S1P-induced decrease in Pro402-hydroxylation, which did not however inhibit binding of pVHL to HIF-1a. This is not necessarily contradictory since it has been shown that hydroxylation of either Pro402 or Pro564 is sufficient to promote pVHL binding [11]. Since Hsp90 inhibition prevented S1P-induced HIF-1a expression, HIF-1a stabilization might be mediated by decreased RACK1 binding and increased Hsp90 binding to HIF-1a. And as PI3K mediated S1Pevoked HIF-1a expression, the involvement of the Akt/GSK3 pathway is also possible. The signaling evoked by S1P in ML-1 cells is practically identical to the signaling induced by growth factors to increase HIF-1a translation through activation of p70S6K and eIF-4E. On the other hand, that S1P did not increase translation of mRNA containing the murine 59-UTR of HIF-1a points to S1P not affecting HIF-1a synthesis. However, changes in HIF-1a translation will readily affect HIF-1a levels because of the protein's low basal expression and short half-life whereas the effect on luciferase levels may not be as strong. Taken together, our data points to S1P stabilizing the HIF-1a protein but potentially also increasing its translation.
An important part of the project was to determine the significance of S1P-induced HIF-1a expression for the S1Pinduced migration of ML-1 cells. We were able to attenuate basal and S1P-induced ML-1 migration by HIF-1a inhibition. As a control we also conducted experiments in hypoxia. Surprisingly, we did not see a significant increase in either migration or haptotaxis in hypoxia. Thus, other factors induced or inhibited by hypoxic stress may have counteracted the migratory effect. Interestingly, although hypoxia significantly elevated expression of the pro-migratory S1P 1 receptor, S1P-induced migration was not increased either. However, hypoxia-induced up-regulation of S1P 1 is consistent with this receptor being essential for S1P Cells were preincubated with HIF-1 inhibitor (HIFi, 10 mM, 30 min) and S1P (100 nM, 30 min) and allowed to migrate towards serum for 8 h. (B) Down-regulation of HIF-1a decreases basal migration. Cells were transfected with HIF-1a siRNA and allowed to migrate towards serum and S1P (100 nM) for 8 h. (C) Down-regulation of HIF-1a attenuates S1P-induced migration. Cells were transfected with HIF-1a siRNA and allowed to migrate towards S1P (100 nM) for 20 h. (D) Inhibition of p70S6K decreases basal migration and prevents S1P-induced migration. Cells were preincubated with p70S6K inhibitor (p70i, 10 mM, 30 min) and S1P (100 nM, 30 min) and allowed to migrate towards serum for 8 h. (E) Down-regulation of S1P 3 attenuates S1P-induced migration. Cells were transfected with S1P 3 siRNA and allowed to migrate towards serum and S1P (100 nM) for 8 h. Results are mean 6 SEM, n $ 3. *P , 0.05 and ***P , 0.001 indicate statistically significant difference between S1P treatment and respective vehicle or siRNA control, o P , 0.05 and ooo P , 0.001 indicate statistically significant difference between siRNA treatment and control siRNA, between siRNA+S1P treatment and control siRNA+S1P or between inhibitor treatment and vehicle control. doi:10.1371/journal.pone.0066189.g006 or S1P-induced migration or haptotaxis. Cells were allowed to migrate in normoxia or hypoxia (1% O 2 ) towards serum and S1P in the migration experiments or towards collagen and S1P in the haptotaxis experiments for 8 h. (C) Hypoxia does not affect proliferation. Cells were incubated in normoxia or hypoxia (1% O 2 ) for the indicated times. Results are mean 6 SEM, n $ 3. **P , 0.01 and ***P , 0.001 indicate statistically significant difference between S1P treatment and respective control. doi:10.1371/journal.pone.0066189.g007 S1P and HIF-1a in Thyroid Cancer PLOS ONE | www.plosone.org function in vascular development [46][47][48], and S1P 1 expression being regulated via VEGF signaling in ML-1 cells [14].
In conclusion we identify S1P, a bioactive lipid readily available in blood, as a non-hypoxic regulator of HIF-1a expression in thyroid cancer cells. We show S1P to increase HIF-1 activity and to be a co-factor in S1P-induced migration. We also present signaling pathways involved in S1P-induced HIF-1a expression (Fig. 8). Altogether our work increases the knowledge of both the oncogenic function of S1P and normoxic regulation of HIF-1.