A key role for transketolase-like 1 in tumor metabolic reprogramming

Metabolic reprogramming, a crucial cancer hallmark, shifts metabolic pathways such as glycolysis, tricarboxylic acid cycle or lipogenesis, to enable the growth characteristics of cancer cells. Here, we provide evidence that transketolase-like 1 (TKTL1) orchestrates aerobic glycolysis, fatty acid and nucleic acid synthesis, glutamine metabolism, protection against oxidative stress and cell proliferation. Furthermore, silencing of TKTL1 reduced the levels of sphingolipids such as lactosylceramide (a sphingolipid regulating cell survival, proliferation and angiogenesis) and phosphatidylinositol (which activates PI3K/Akt/mTOR signaling). Thus, in addition to its well-known roles in glucose and amino acid metabolism, TKTL1 also regulates lipid metabolism. In conclusion, our study provides unprecedented evidence that TKTL1 plays central roles in major metabolic processes subject to reprogramming in cancer cells and thus identifies TKTL1 as a promising target for new anti-cancer therapies.


INTRODUCTION
Because an evident feature of cancer cells is their accelerated proliferation rate, signals mediating the response to mitogenic stimuli and alterations in (proto-)oncogenes were among the first processes to be mechanistically addressed and identified in cancer. More recently, it has become evident that cancer cells must also undergo a metabolic reprogramming to sustain the abnormal growth characteristics associated with malignant transformation [1][2][3]. Consequently, different metabolic genes have been proposed to function as (proto-) oncogenes [4,5]. Among them, glucose-6-phosphate dehydrogenase (g6pd) and transketolase-like 1 (tktl1) are of special interest because they regulate the pentose phosphate pathway (PPP), which generates NADPH for redox homeostasis and lipid synthesis and balances the flux of glucose derivatives between production of pentoses for nucleotide synthesis and supply of pyruvate for TCA cycle replenishment, or lactate production [6,7].

Research Paper www.impactjournals.com/oncotarget
Transketolase (TKT) is the key enzyme of the non-oxidative PPP branch (non-oxPPP). Via genome duplication and exon skipping, higher vertebrates acquired transketolase-like 1 (TKTL1), which compared to TKT protein lacks 38 amino acid residues [8,9]. TKTL1 is upregulated in various cancer cells and tissues [4,5,[10][11][12][13], and TKTL1 overexpression in cancer patients is correlated with malignancy, invasiveness, therapeutic resistance and poor prognosis [11,[14][15][16]. TKTL1 is required for rapid growth and viability of tumor cells [17], and protects against apoptosis induced by growth factor withdrawal [18], oxidative stress [17][18][19], cytotoxic therapy or targeted therapies [16,20], while TKTL1 silencing hinders cancer cell proliferation [17,21]. In vivo, silencing of TKTL1 in cancer cells reduces tumor growth, while overexpression of TKTL1 has opposite effects [7,22]. However, little is known about how TKTL1 participates in all these processes that guarantee the growth and survival of cancer cells. Apart from the aforementioned effects, TKTL1 could also have a role in the reactions of the PPP. However, the precise reaction scheme of the non-oxPPP still remains debated, as the degree and distribution of 14 C-isotope labeling differ from what has been predicted by the scheme of the individual reactions [23].
Unlike TKT, which catalyzes the common twosubstrate reaction by using xylulose-5-phosphate (X5P) as donor and ribose-5-phosphate (R5P) as acceptor, it has been proposed that TKTL1 may perform a one-substrate reaction by converting X5P to glyceraldehyde-3-phosphate (GA3P) and acetate, similar to phosphoketolase, an enzyme that performs a similar one-substrate reaction in heterofermentative lactic acid bacteria [8,24]. The acetate, generated in this reaction, can then be converted to acetyl-CoA (AcCoA).
Cancer cells often have increased rates of fatty acid synthesis and therefore require large amounts of AcCoA. However, the issue of the supply of AcCoA for fatty acid synthesis in glycolytic cancer cells is still unresolved, since AcCoA production from pyruvate in cancer cells is often reduced due to: i) a switch from pyruvate kinase (PK) M1 to the less active PKM2 that reduces pyruvate production [25,26] and ii) a decreased entry of pyruvate into the TCA cycle due to inactivation of pyruvate dehydrogenase (PDH) [27], which can be compensated by the entry of pyruvate into the TCA cycle via the anaplerotic conversion to oxaloacetate by pyruvate carboxylase (PC) [28] or glutamine catabolism [29][30][31]. Given that, in cancer cells, the bulk of pyruvate is directed to lactate and the flux from glutamine to AcCoA is low, it is challenging to explain how the large amounts of AcCoA necessary for fatty acid synthesis are generated. We hypothesized that the one-substrate reaction catalyzed by TKTL1 may help to explain this enigma. Our findings support this hypothesis and further contribute to clarify additional mechanisms by which TKTL1 coordinates metabolic reprogramming and tumor growth.

TKTL1 contributes to overall cellular transketolase activity
As an initial approach to characterize the relevance of TKTL1 in the metabolism and growth properties of cancer cells, we analyzed 4 cell lines with variable levels of TKTL1 (THP-1 and PC-3S, high expression; HCT-116 and PC-3M, low expression). Transduction of human THP-1 leukemia cells with a lentiviral vector expressing a TKTL1-specific shRNA (THP-1 KD ) specifically downregulated TKTL1 ( Figure 1A) but not TKT (changed by 4.0 ± 1.2%; n = 3; p = NS). SiRNA-mediated silencing reduced TKTL1 expression by 86 ± 13% in human PC-3S prostate cancer cells, by 81 ± 13% in human PC-3M prostate cancer cells and by 60 ± 28% in human HCT116 colorectal carcinoma cells (n = 5; p < 0.01 for all). Consistent with the contribution by both TKT and TKTL1 to the overall transketolase activity according to their relative expression [10,32], silencing of TKTL1 lowered but did not eliminate total transketolase activity (measured as GA3P production from R5P and X5P). Indeed, total transketolase activity was reduced by 60% in THP-1 KD cells ( Figure 1B) and by 37% in PC-3S KD cells (mU/mg protein: 8.0 ± 1.0 in PC-3S WT cells versus 5.0 ± 0.4 in PC-3S KD cells; n = 3; p < 0.05), but not in HCT116 KD or PC-3M KD cells (changed by 7 ± 1% in HCT116 KD cells and by 1 ± 3% In PC-3M KD cells; n = 5; p = ns), indicating that TKTL1 minimally contributed to the transketolase activity in the latter cells.
In HCT116 KD cells, in which TKTL1 did not significantly contribute to the overall transketolase activity, no differences in glucose consumption, lactate production (not shown), glucose oxidation ( Figure 1H) or PDH/PC activities ( Figure 2C, 2D) were observed. Thus, TKTL1 drives the glucose metabolism and contributes to the Warburg effect in cancer cells through its transketolase activity.

TKTL1 contributes to the non-oxidative pentose phosphate pathway
We analyzed the role of TKTL1 in regulating the incorporation of glucose in the PPP by measuring metabolic fluxes via isotopomer analysis ( Figure 3). After [1,2-13 C 2 ]-glucose incubation, the PPP flux was determined by measuring ribose enrichment: [1,2-13 C 2 ]glucose generates m1 ribose if metabolized through the oxidative branch (oxPPP), whereas it produces m2 ribose via the non-oxPPP. Further recombination of labeled intermediates leads to m3 (oxPPP) and m4 (non-oxPPP) ribose. In THP-1 KD cells, TKTL1 silencing increased the pool of m1 and m3 ribose, while decreasing the pool of m2 and m4 ribose, indicating that the non-oxPPP flux was reduced ( Figure 1I). Moreover, the decrease in total enrichment of ribose indicated that TKTL1 silencing reduced the net PPP flux by 10% ( Figure 1J) and that the relative contribution of the oxPPP over the non-oxPPP flux, calculated as m1/m2, was increased in THP-1 KD cells ( Figure 1K). Phase plane analysis of normalized m1 and m2 ribose isotopologues graphically illustrates these results ( Figure 4A). Interestingly, similar results were obtained in HCT116 KD cells ( Figure 4B), even though TKTL1 does not contribute to the total transketolase activity in these cells.

TKTL1 silencing enhances glutamine consumption
In addition to glucose, glutamine is a major source of carbon atoms for the cell, anaplerotically replenishing the TCA cycle to maintain an active mitochondrial metabolism when pyruvate supply is compromised and enabling the production of AcCoA either through the TCA cycle or by forming citrate via reductive carboxylation [30,31]. TKTL1 silencing in THP-1 KD cells did not affect glutamate accumulation or glutamine consumption ( Figure  4C, 4D). In contrast, TKTL1 silencing in HCT116 KD cells increased glutamate and glutamine consumption ( Figure  4E, 4F). However, even though glucose and glutamine metabolism were modulated differently in THP-1 and HCT116 cells, TKTL1 silencing induced similar changes in the ratio of glucose/glutamate consumption rates in both cell lines. This ratio was decreased by 19% in THP-1 KD cells and by 25% in HCT116 KD cells, indicating   that TKTL1 regulates the balance between glucose and glutamine metabolism regardless of how it modulates each of these metabolic pathways individually or the extent of its contribution to overall transketolase activity.
On the other hand, THP-1 KD cells reduced the consumption of isoleucine (Ile), leucine (Leu) and valine (Val), i.e. amino acids that are also catabolized to replenish the TCA cycle and generate AcCoA ( Figure 5), indicating that TKTL1 favors these AcCoA production pathways.

TKTL1 contributes to fatty acid synthesis
Given the importance of fatty acid synthesis in rapidly proliferating cells and the role of TKTL1 silencing on several pathways leading to AcCoA production, we analyzed whether TKTL1 silencing affects lipid synthesis. We determined [1,2-13 C 2 ]-glucose incorporation in C16-palmitate and C18-stearate as a surrogate readout of [1,2-13 C 2 ]-glucose incorporation into AcCoA, and measured the fraction of de novo synthesized lipids (FNS) and the proportion of lipids synthesized by chain elongation. FNS levels and the contribution of [1,2-13 C 2 ]-glucose to fatty acid synthesis were reduced in THP-1 KD cells ( Figure 6A, 6B), indicating that TKTL1 participated in the conversion of glucose carbon into fatty acids via generation of AcCoA. Similar effects were observed in HCT116 KD cells ( Figure 6C, 6D). The decreased contribution of glucose to lipid synthesis was not attributable solely to a slower growth rate, since the incorporation of [1,2-13 C 2 ]-glucose into fatty acids was reduced more than the [1,2-13 C 2 ]-glucose incorporation into nucleotides (Table 1).

Evidence in support of one-substrate activity by TKTL1
We explored if TKTL1 might display one-substrate activity by isotopologue analysis of m1 lactate, as this provides insight into the TKT/TKTL1-dependent glucose carbon flux through the PPP. We considered two possibilities: (i) both TKT and TKTL1 catalyze only the two-substrate reaction ( Figure 7A), and (ii) TKT catalyzes the two-substrate reaction while TKTL1 catalyzes the one-substrate reaction ( Figure 7B). To test this model, we analyzed the enrichment of [1,2-13 C 2 ]-glucose in m1 lactate, which can only be generated in the PPP via the two-substrate reaction, since the label is donated to AcCoA in the one-substrate reaction ( Figure 4B, red pathway). Hence, in THP-1 cells, where TKTL1 accounts for 60% of the overall transketolase activity, TKTL1 silencing should decrease m1 lactate levels if it would only have the classical two-substrate activity ( Figure 7A), but should increase m1 lactate levels if it had one-substrate activity ( Figure 7B) since phosphorylated pentoses would be metabolized only via TKT generating more m1 lactate molecules.
This analysis revealed that the levels of m1 lactate were increased by 13 ± 7% (n = 6; p < 0.005) in THP-1 KD cells, which perfectly fits with theoretical predictions for the one-substrate reaction model ( Figure 7B). In HCT116 cells, TKTL1 silencing decreased m1 lactate levels by 46 ± 10% (n = 4; p < 0.001). Given that TKTL1 minimally contributed to the overall transketolase activity in these cells, the effects on m1 lactate are unlikely to be due to its enzyme reaction but rather to the reduction of the non-oxPPP flux in TKTL1-silenced cells ( Figure 4B).

ACLY complements TKTL1 in the promotion of fatty acid synthesis
ATP-citrate lyase (ACLY) is a key enzyme regulating lipid synthesis by converting cytosolic citrate to AcCoA. To evaluate the possible role of TKTL1 in producing AcCoA in comparison with ACLY, we silenced ACLY in HCT116 (HCT116-ACLY KD ) cells. SiRNAmediated silencing reduced ACLY transcript levels by 45 ± 10% (n = 4; p < 0.001) without affecting proliferation, glucose consumption or lactate production ( Figure 8A-8C), or altering the glycolytic rate or the PC/PDH ratio (not shown). ACLY silencing did not affect glutamate levels or glutamine consumption (not shown), nor did it alter glutamate enrichment ( Figure 1H), but decreased fatty acid enrichment ( Figure 6D), as expected. Indeed, when ACLY's activity is reduced, less citrate is processed to generate AcCoA and fewer labeled substrates are incorporated into palmitate without however affecting stearate enrichment ( Figure 6D). These results, together with the abovementioned results in TKTL1 silenced cells, suggest that ACLY and TKTL1 play cooperative roles in fatty acids synthesis.

DISCUSSION
In this study, we have found that TKTL1 plays broader roles in cell metabolism than previously anticipated. First, in agreement with previous data [7], we show that TKTL1 enhances glycolysis and sustains accelerated proliferation. These findings implicate TKTL1 in aerobic glycolysis (Warburg effect), the mechanism on which cancer and other proliferative cells rely to produce glycolytic intermediates for biomass synthesis and energy [25,26,35,36]. TKTL1 also enhances the PPP flux for biosynthetic reactions, as the PPP generates ribose-5phosphate for nucleic acid synthesis and NADPH for fatty acid synthesis and maintaining redox homeostasis to protect cells against apoptosis [37].
Second, TKTL1 affects glutamine metabolism, the other main carbon source in cancer cells. In contrast to its effects on glucose metabolism, we show that depletion of TKTL1 enhances glutamine consumption rates independent of its transketolase activity. Interestingly, the effect of TKTL1 on the ox-PPP versus non-ox PPP ratio is also independent of its contribution to the overall transketolase activity. Given that both glutaminolysis and    Values are expressed as pmol/10 6 cells (mean±SD; n=7; p-values are adjusted using a Holm method, *p<0.05, **p<0.01, ***p<0.001). To calculate the totals ± standard deviations per lipid species, we first calculated per repeat the sums of all chain length values, and then calculated the average and standard deviation of these sums. Due to the fact that in this high-throughput analysis, a peak was not present for each chain length in every repeat, some totals listed do not equal the mathematical sum of the individual chain length values. R, total number of carbons:total number of insaturated acyl moieties. • Some species are detected with two independent peaks in the MS.  PPP contribute to the maintenance of redox homeostasis, our results suggest that the previously described role of TKTL1 in protecting cells against oxidative stress [38] is independent of its contribution to the overall transketolase activity.
Although our results suggest that TKTL1 affects cell metabolism via mechanisms that are either dependent or independent of its enzymatic activity, they also unambiguously show that TKTL1 has transketolase enzymatic activity, supporting previous reports [10,20,32], and refuting studies claiming that TKTL1 lacks enzymatic activity [39,40]. In our studies, TKTL1 silencing lowered the overall cellular transketolase activity, supporting that TKTL1 cleaves a two-carbon fragment from a pentose-phosphate to generate GA3P. Failure to detect in vitro transketolase activity of an artificial TKT protein variant with a 38 amino acid deletion was used in previous studies as an argument against an intrinsic enzymatic function of TKTL1, although no experiments were performed with the wild-type TKTL1 protein [39]. We speculate that the lack of transketolase activity by this TKT deletion variant might be due to defective interactions with other cellular components or cofactors that may be required for its enzymatic activity. Alternatively, why TKTL1 activity is detected in vivo, but not in vitro, might be attributed to the weak binding of TKTL1 to its cofactor, thiamine diphosphate, which retains the two-carbon fragment upon cleavage of X5P until it is transferred to R5P.
Our characterization of the oncogenic role of TKTL1 also sheds light on two largely unexplained phenomena in cancer cell metabolism. First, the TKTL1 activity described here better reconciles with the proposed hypothetical reaction scheme [23]. Second, our observations may help resolve the outstanding enigma of how rapidly proliferating cells can boost lipid synthesis [41] when AcCoA production is limited by impaired pyruvate production from phosphoenolpyruvate (because of PKM2 overexpression) [25,26] and inhibition of PDH (which drives pyruvate to lactate, thereby reducing pyruvate entry into the TCA cycle) [27,42,43]. Taking into account that i) silencing of TKTL1 or ACLY in HCT116 cells reduced the expression of the corresponding protein to a similar extent (around 50%), and ii) the decrease in palmitate enrichment from labeled glucose was similar for both knock downs (around 40-60%), it can be estimated that a fully active TKTL1 can contribute to up to 50% of the flux from glucose to fatty acid synthesis. Indeed, our study shows that TKTL1 plays a significant role in the synthesis of fatty acids and phospholipids.
The observed enrichment of [1,2-13 C 2 ]-glucose in m1 lactate is consistent with the one-substrate reaction previously suggested by Coy et al. [8]. In this regard, the 38-amino acid segment in TKT missing in TKTL1 contains a critical histidine residue, homologous to His103 in the S. cerevisiae transketolase protein. Site-directed mutagenesis of this residue in the yeast transketolase converts its activity from the classical two-substrate reaction to a one-substrate reaction [44]. We speculate that, by analogy, the absence of the equivalent His residue in mammalian TKTL1 may favor the one-substrate reaction, whereby X5P is converted to GA3P and a twocarbon fragment. This fragment could be AcCoA or acetate (that subsequently generates AcCoA via acetyl-CoA synthetase 2) [45,46]. Thus, TKTL1 may link the synthesis of nucleotide precursors (R5P) to the production of fatty acid precursors (AcCoA), providing an alternative explanation for how cancer cells can generate new lipids when AcCoA production from glucose metabolism is restricted. This TKTL1-specific one-substrate reaction would also explain why TKTL1 silencing in HCT116 KD and PC-3M KD cells reduces cell proliferation, while not affecting overall transketolase activity. We thus speculate that the evolutionary removal in TKTL1 of the 38 amino acid residues present in TKT, which includes the His residue equivalent to yeast His103, has created a new enzyme capable of catalyzing a one-substrate reaction that enables the fermentation of glucose to lactic acid and AcCoA even in the presence of oxygen.
Our study therefore shows that TKTL1 is an active transketolase that modulates the Warburg effect to provide energy and precursors for DNA replication and membrane biogenesis depending on its enzymatic activity. Moreover, our results suggest that TKTL1 controls redox homeostasis and protects cells against oxidative stress and apoptosis independent of its transketolase activity. Finally, we have shown that TKTL1 participates in the synthesis of lactosylceramides, a sphingolipid regulating cell survival, proliferation and angiogenesis [47,48], and phosphatidylinositols, which activate PI3K/Akt/mTOR signaling [49].
In conclusion, we provide unprecedented evidence that TKTL1 confers a growth advantage to proliferating cells and is essential in cancer metabolic reprogramming, suggesting that TKTL1 might be a promising target for the design of future anti-cancer therapies.

RT-PCR
Total RNA was isolated from cultured cells (approximately 1-2 x 10 6 cells) using the TRIzol reagent (Roche Diagnostics). For quantitative RT-PCR, cDNA was synthesized from 1 μg of total RNA using random hexamers and M-MLV-reverse transcriptase (Invitrogen) and subjected to real-time PCR with specific TaqMan probes (Assays-on-demand TKTL1: Hs00202061_ m1; ACLY: Hs00153764_m1; reference probe, PPIA: Hs99999904_m1) (Applied Biosystems, Foster City, CA) on a ABI Prism 7700 Sequence Detector System (Applied Biosystems). All reactions were performed in triplicate. Fold changes in gene expression were calculated using the ΔΔCt method, using as PPIA as a reference.

Cell proliferation
Cells were seeded in 6-well plates at a concentration of 75,000 cells per well and counted using a Scepter Handheld Automated Cell Counter (Millipore). THP-1 WT and THP-1 KD cells were seeded in 75 cm 2 flasks at a concentration of 200 x 10 3 cells/mL in a total volume of 15 mL medium. Every 24 h until 120 h an aliquot of 1 mL medium was removed and cells were counted. Duplication times were determined as the slope of the graph obtained by plotting the napierian logarithm of cell concentration (cells/mL) versus the incubation time in hours corresponding to the period of cellular exponential growth.

TKT enzymatic activity
Fresh cell pellets containing approximately 5-10 x 10 6 cells were resuspended in lysis buffer (20 mM HEPES, pH 7.5, 10% glycerol, 0.4 mM NaCl, 10 mM EGTA, 5 mM EDTA, 0.4% Triton X-100, 25 mM NaF, 1 mM DTT, 0.4 mM Pefabloc SC, 20 μg/mL pepstatin), sonicated in ice-cold buffer (4°C) and immediately followed by centrifugation at 15,000 × g for 15 min at 4°C. The supernatant was used for the determination of enzyme activity with a Cobas Mira Plus analyzer (HORIBA ABX, Montpellier, France). Transketolase activity was determined as described [51]. Briefly, the TKT activity product, glyceraldehyde-3-phosphate, is isomerized to dihydroxyacetone-phosphate and its subsequent conversion to glycerol-phosphate produces 1 molecule of NADH that absorbs at 340 nm. Initially, 40 μL of sample were added to a cuvette containing 150 μL of buffer (50 mM Tris-HCl buffer, pH 7.6; 5 mM MgCl 2 ; 0.1 mM thiamine pyrophosphate (TPP); 0.2 mM NADH; 0.2 U/mL triose phosphate isomerase and 0.2 U/mL glycerol dehydrogenase). The reaction mixture was kept at 37°C for 15 min in order to convert the trioses-phosphate into glycerol-phosphate. Finally, 75 μL of a solution containing 25 mM R5P and 25 mM X5P was transferred to the cuvette and absorbance at 340 nm was recorded for 5 min.

Amino acid quantification
Amino acid concentrations in cell culture medium were determined by ion-exchange chromatography with a Biochrom 30 amino acid analyzer (Pharmacia Biochrom Ltd, Cambridge, UK). 70 μL of 150 μM norleucine was used as an internal standard and added to 500 μL of medium. Solvent was evaporated to complete dryness using a Savant SpeedVac (Thermo Scientific, Waltham, MA, USA). The samples were redissolved in 500 μL lithium citrate buffer (pH 2.2), filtered through a 0.22 μm filter and 30 μL per sample injected onto the Biochrom 30 lithium system. A set of lithium citrate buffers were used as a mobile phase for separation for 115 min and post column derivatization with ninhydrin allowed amino acid quantification at 570 and 440 nm.

Isotopologue analysis by GC-MS
For isotopologue distribution analysis, 3 x 10 5 cells/ mL were seeded in 75 cm 2 Petri dishes in the presence of RPMI 1640 medium spiked with 10 mM of [1,2-13 C 2 ]glucose (Sigma-Aldrich) (50% isotope enrichment), for 48 h. Following incubation, cells were counted with a Scepter Handheld Automated Cell Counter (Millipore) and centrifuged (1,500 rpm for 5 min). Medium was collected for glucose, lactate and glutamate/glutamine analysis, and cell pellets were frozen for RNA ribose and fatty acids analysis.
Glucose was isolated from 200 μL medium by ionexchange chromatography and converted to its aldonitrile acetate derivative by adding 100 μL hydroxylamine in pyridine (2% wt) at 100°C for 30 min followed by 75 μL acetic anhydride at 100°C for 1 h. The reaction mixture was evaporated to complete dryness and redissolved in 50 to 200 μL of ethyl acetate. 1 μL of the derivatized product was loaded to a GC-2010 gas chromatograph (Shimadzu) equipped with a HP-5MS capillary column (30 m length, 250 μm diameter and 0.25 μm film thickness), connected to a GCMSQP2010 mass selective detector (Shimadzu) [54]. Lactate was extracted from the medium by ethyl acetate after acidification with HCl and 1 mL ethyl acetate at 20°C. The propylamideheptafluorobutyric derivative of the lactate was prepared and analyzed as described [55]. RNA ribose was isolated by acid hydrolysis of cellular RNA after Trizol (Invitrogen) extraction from cell pellets. Ribose was derivatized to its aldonitrile acetate form by an analogous derivatization procedure as described for glucose. After derivatization, the reaction mixture was evaporated to complete dryness and redissolved in 50 to 200 μL of ethyl acetate. Glutamate was isolated from 1 mL of cell medium using ion-exchange chromatography and converted to its n-trifluoroacetyl-n-butyl derivative by adding 200 μL of butanolic HCl (0.4 N), the mixture incubated at 100°C for 1 h and dried under nitrogen stream. Next, 100 μL methylene chloride and 25 μL trifluoroacetic acid were added. The solution was kept at 20°C for 20 min and dried under nitrogen stream. The pellet was dissolved in 50 to 200 μL methylene chloride and 1 μL was injected for analysis [34,55]. Fatty acids were extracted with methanol from the organic phase obtained during RNA isolation with Trizol (Invitrogen).

Lipidomic analysis
Two standard mixes of phospholipids (PL) and sphingolipids (SL) were prepared. The PL mix contained PC 34:0, PE 34:0, PG 34:0, PI 34:2, PS 34:0 and LysoPC 17:0 at a concentration of 20 μM each. The SL mixture contained Cer 12:0, GlcCer 12:0, LacCer 12:0, SM 12:0 and sphinganine 17:0 at a concentration of 20 μM each. These mixes were used as internal standards for PL and SL identification and quantification at a final concentration of 200 pmol/sample. Two pellets containing 750 x 10 3 cells each were re-suspended in 100 μL of milliQ water and 500 μL pure methanol (MeOH), these suspensions were transferred to glass vials. 250 μL chloroform (CHCl 3 ) was added to obtain an extraction solvent MeOH:CHCl 3 :H 2 O (2:1:0.4). 10 μL of PL or SL standard mix was added and lipids were extracted overnight at 48°C. For sphingolipid analysis, 75 μL of 1 M potassium hydroxide (KOH) dissolved in MeOH was added and samples were sonicated. After 2 h of incubation at 37°C, 75 μL of 1 M acetic acid was added. Following extraction, PL and SL samples were dried under nitrogen stream and stored at -20°C until further usage. Prior to analysis on an UPLC-MS, samples were dissolved in 500 μL MeOH and transferred to Eppendorf tubes. The volume was reduced by evaporation under gentle nitrogen steam to a final volume of 150 μL MeOH, prior to analysis samples were centrifuged at 10,000 x g for 3 min and 150 μL of supernatant was infused into the UPLC. The liquid chromatography-mass spectrometer (Waters Acquity UPLC system, connected to a Waters LCT Premier orthogonal accelerated time of flight mass spectrometer; Waters, Millford, MA) was operated in positive electrospray ionisation mode. Full scan spectra from 50 to 1,500 Da were acquired and individual spectra were summed to produce data points every 0.2 s. Mass accuracy and reproducibility were maintained by using an independent reference spray via LockSpray interference. The analytical column, maintained at 30°C, was a 100 x 2.1 mm id, 1.7 μm C8 Acquity UPLC BEH (Waters). The mobile phase consisted of solvent A: MeOH/H 2 O/ HCOOH (74:25:1 v:v:v); and solvent B: MeOH/HCOOH (99:1 v:v), both containing 5 mM ammonium formate. A gradient was programmed -0.0 min, 80% B; 3 min, 90% B; 6 min, 90% B; 15 min, 99% B; 18 min, 99% B; 20 min, 80% B; 22 min, 80% B, with a 300 μL/min flow rate. Quantification was carried out using the ion chromatogram extracted for each compound in 50 mDa windows. The linear dynamic range was determined by injecting standard mixtures. Positive identification of compounds was based on the accurate mass measurement with an error < 5 ppm and its LC retention time, compared to that of a standard (± 2%). Lipid species were assigned from the m/z values detected in the MS using the MassLynx software (v4.1 SCN627 Waters). Positive ionization allowed detection of PC, PC-pmg, Lyso-PC and Lyso-PC-pmg from the phospholipid extract and Cer, dhihydroceramides, SM, dihydrosphingomyelins, GlcCer, glucosyldihydroceramides, LacCer and lactosyldihydroceramides from the sphingolipid extract. Negative ionization mode allowed detecting the rest of phospholipid classes: PE, PE-pmg, Lyso-PE, Lyso-PEpmg, PS, Lyso-PS and PI.

Statistical analysis
In vitro experiments were performed in triplicate for each treatment in at least two independent experiments. Mass spectrometric analyses were carried out in three independent 1 μL injections of each sample by the automatic sampler in a linear set-up. Raw data were considered valid only if the sample deviation was < 1% of the normalized peak intensity. Statistical analyses were performed using the parametric unpaired, two-tailed independent sample Student's t-test, assuming normality of distribution and homogeneity of variance based on past experience in similar measurements [56,57]. When a family of statistical hypotheses was tested in the same www.impactjournals.com/oncotarget series of experiments, the Holm method for P-value adjustment was used [58]. P < 0.05 (*) was considered statistically significant.

ACKNOWLEDGMENTS
Authors acknowledge Dr. Vitaly Selivanov (Department of Biochemistry and Molecular Biomedicine, Faculty of Biology, Universitat de Barcelona) for his help in the statistical analyses performed in the manuscript.

CONFLICTS OF INTEREST
JF Coy is a shareholder of Zyagnum AG ( Frankfurt am Main, Germany) and TAVARGENIX GmbH, (Frankfurt am Main, Germany). All other authors declare no conflicts of interest.