Metformin induces lipid changes on sphingolipid species and oxidized lipids in polycystic ovary syndrome women

Metformin is one of the treatments used for PCOS pathology decreasing body weight, plasma androgen, FSH and glucose levels. Unfortunately, there is little known about metformin’s effects on lipid metabolism, a crucial process in PCOS pathology. We have employed a lipidomic approach to explore alterations in the plasma lipid profile of patients with PCOS following metformin treatment. The aim is to offer new insights about the effect of metformin in PCOS patients. Plasma samples were obtained from 27 subjects prior to and following 12 weeks of metformin treatment. A detailed biochemical characterization and lipidomic profile was performed. Metformin reduces BMI, HOMA-IR, FSH and androstenedione and increases DHEA-S but no changes were found in glucose levels after treatment. Multivariate statistics revealed a specific lipidomic signature due to the effect of 12 weeks of metformin treatment in PCOS patients. This signature includes changes in sphingolipid metabolism suggesting a crosstalk between these lipid species and the androgenic metabolism and a decrease in oxidized lipids reinforcing that metformin treatment improves oxidative stress status. Our study confirms the specific effect of metformin in lipid metabolism on women with PCOS after 12 weeks of treatment.

plasma lipidome of pcoS patients under metformin treatment. In order to assess how 12 weeks of metformin treatment effected the global plasma lipidomic profile of PCOS patients, we carried out a non-targeted lipidomic approach, focusing on low molecular-weight ionizable lipid molecules (m/z of 300-3000). In the case of both polarities (negative and positive), 1950 features were aligned after applying the MFE algorithm 31 . Then, we selected only those features present in at least 50% of the samples of the same group (312 lipid species ionized with positive polarity and 64 lipid species in negative polarity).
To determine whether the detected lipidome was altered in the PCOS patients after the 12-week of treatment multivariate statistics were applied ( Fig. 1 and Fig. S1). All the analyses reflected in Fig. 1 were carried out with lipid species ionized with a positive ESI polarity, while the results of the multivariate analyses of lipidome with negative ESI are represented in Fig. S1. Non-supervised PCA (Fig. 1A) revealed clustering of both groups, thus suggesting that the effect of metformin in PCOS patients has a specific plasma lipidomic signature. In line with this, the machine learning algorithm Random Forest correctly classified all of the patients after the treatment (classification error PRE: 0.111, classification error POST: 0.000, out of bag error: 0.0556) based on the lipidome detected using the positive polarity mode (Fig. 1B). All of the multivariate analyses performed with the negatively charged lipid species showed no differences attributable to the metformin treatment in PCOS patients (Fig. S1).
www.nature.com/scientificreports www.nature.com/scientificreports/ To better characterize the effect of metformin on the plasma lipidome we performed a parametric paired t-test for equal variances. Of the 376 lipid species detected, 45 were found to differ in a statistically significant manner between groups. The hierarchical analysis of these 45 lipid species clearly showed a differential regulation after and before metformin treatment (Fig. 1C). Interestingly, all of the 18 identified lipid species (based on exact mass, retention time, isotopic distribution and/or MS/MS spectrum) were decreased after metformin treatment (Table 2), while the remaining 27 were unidentified (Table 3).
Among the lipid species that could be identified, 8 belonged to the glycerophospholipid family, four of which (ethanolamine and choline GP-based on) were oxidized and one was a plasmalogen polyunsaturated species. We also identified 8 sphingolipids with a differential regulation due to the metformin treatment. Specifically, three polyunsaturated ceramides and five sphingomyelins, two of which were monounsaturated and three polyunsaturated. Finally, levels of one oxidized triacylglycerol and one cholesterol ester were also found to be lower after the treatment.
In order to further determine to what extent these metabolites can predict the effect of treatment of PCOS with metformin, ROC analyses were performed using MS peak areas. The details of the area under the curve, as well as the specificity and sensitivity of each lipid species, are presented in Table 2. The lipid species with the highest AUC were oxPE(17:3), with a value of 0.90123 and 90% specificity and sensitivity, and SM(d36:6), with 0.90672 AUC, 90% specificity and 80% sensitivity.
When plasma fatty acid composition was analysed, differences were found due to the effect of the 12-week metformin treatment (Table 4). There was a significant increase in the content of 20:2n-6, whereas 20:3n-6, 20:4n-6, 22:1n-9 and 22:5n-3 were found to have decreased. This led to a slight but significant alteration of the average chain length (ACL). When we evaluated other parameters based on fatty acid composition, such as SFA, UFA, MUFA, PUFA, PUFAn-3, PUFAn-6, DBI, PI, and AI, no differences were found in PCOS patients after the metformin treatment.

Discussion
PCOS is a heterogenic and multigenic metabolic disorder which affects women of a reproductive age and is characterized by hyperandrogenism and is commonly associated with other pathologies such as IR and obesity 32 . Currently, there is no universal treatment for PCOS patients, and most available drugs are directed at treating symptoms such as androgen excess, IR or oligo-ovulation 32,33 . Among available insulin-sensitizer drugs, metformin is one of the most used, having similar effects as lifestyle changes in terms of decreasing body weight/ BMI and more marked effects in terms of decreasing androgen concentrations 34 . Furthermore, previous studies have demonstrated that metformin reduces plasma glucose levels and FSH in PCOS subjects, and also improves the oxidative stress status of patients, among other benefits 18 . By decreasing gluconeogenesis and lipogenesis and enhancing glucose uptake in the liver, skeletal muscle, adipose tissue and ovaries metformin increases insulin sensitivity. One potential mechanism proposed for the glucose-lowering effect of metformin is the inhibition of www.nature.com/scientificreports www.nature.com/scientificreports/ mitochondrial complex I and the consequent stimulation of AMP kinase 18,25,34 . On the other hand, the intracellular lipid-lowering effects of metformin have been associated with an increase in the mitochondrial channelling of fatty acids. Treatment with metformin enhances the mitochondrial β-oxidation process, directing excess intracellular FA towards β-oxidation, which reduces the supply of substrates for the synthesis of bioactive lipids, which in turn affects the insulin signalling pathway 35 . Furthermore, metformin blocks increments in the fatty acid transport protein CD36 and in aberrant ceramide and diacylglycerol content in the skeletal muscle of diabetic rats 36 . However, the molecular mechanisms through which metformin induces these changes are not fully understood. As far as we know, the effect of metformin on plasma lipid profile in PCOS has been basically focused on the study of lipoprotein profile. Therefore, the objective of this study was to characterize how a 12-week metformin treatment affects the plasma lipidomic profile of women with PCOS.
Anthropometric and metabolic parameters in PCOS patients prior to and following 12 weeks of metformin treatment showed a decrease in BMI values, whereas no changes in total triglycerides, total cholesterol, LDLc and HDLc levels were detected. This reduction in body weight could be partially responsible for the lipidomic changes in plasma described after metformin treatment. In terms of insulin resistance parameters, we found a slight decrease in plasma glucose and insulin levels after treatment, though these changes were not statistically significant. However, the HOMA-IR index decreased after the 12-week treatment, thus indicating an improvement in insulin resistance. These results are in line with most previously published studies on the subject and confirm the beneficial effect of metformin on body weight and insulin sensitivity 25 .
Endocrine measurements revealed several rearrangements after metformin treatment; namely, decreased levels of FSH and androstenedione and increased levels of DHEA-S. The FSH levels found in plasma patients after treatment were very similar to those detected previously in non-PCOS women, suggesting a reversion of PCOS pathology after metformin treatment 21 . With respect to androstenedione levels, after treatment we found very similar levels to those detected in "pure" PCOS women, suggesting that this parameter is affected by PCOS-associated pathologies 21 . Total testosterone concentrations are very much influenced by SHBG   www.nature.com/scientificreports www.nature.com/scientificreports/ concentration, as 65% of plasma testosterone is bound to SHBG 37 . Previous studies in PCOS patients without associated pathologies showed no changes in SHBG and testosterone levels compared with control subjects, thus reinforcing the idea that PCOS-associated pathologies could be partially responsible for the changes observed in androgenic metabolism 21 . In the present study, we did not observe differences in testosterone and SHBG levels after metformin treatment.
Once we characterized the classical parameters of PCOS patients, we analyzed the whole plasma lipidomic profile using an LC-MS-based technique. Furthermore, fatty acid composition was also measured in order to better characterize the composition of the plasma lipidome. The results revealed the existence of a plasma lipidomic profile specifically associated with a 12-week metformin treatment in PCOS patients, which is the first time this has been reported. This signature is defined mostly by glycerophospholipids and sphingolipids; we detected 45 lipid molecular species whose concentration was altered by the effect of metformin treatment (7 lipid species were up-regulated, while 38 were down-regulated).
The interaction between sphingolipids and steroid hormones has previously been described, and this interaction modulates the steroidogenic signaling pathway. Specifically, these lipids can modulate steroidogenesis, acting at different levels as a second messengers, paracrine/autocrine regulators and/or ligands for nuclear receptors 38 . For example, ceramides have been shown to regulate progesterone and testosterone production, although the precise molecular mechanisms underlying this process are unclear and require further study 39 .  Table 4. Total fatty acid composition of plasma from subjects at the baseline and at the end of 12-weeks follow-up. Abbreviations: ACL, average chain length; SFA, saturated fatty acids; UFA, unsaturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; PUFAn-3, polyunsaturated fatty acids series n-3; PUFAn-6, polyunsaturated fatty acids series n-6; DBI, double bond index; PI, peroxidizability index; AI, anti-inflammatory index.

PRE-TREATMENT
In the present work, we described lower levels of 3 ceramides and 4 sphingomyelins after metformin treatment, suggesting that the effect of metformin on androgenic metabolism is partially mediated by sphingolipid metabolism regulation. Moreover, the role of sphingolipids, and specially ceramides, in the mediation of insulin resistance [40][41][42] has been reported previously, so the effect of metformin on parameters of insulin resistance in women with PCOS could also be a result of the interaction of these species with the insulin signaling pathway.
Concerning glycerophospholipids, we identified 8 species that were down-regulated after metformin treatment, 4 of which were oxidized. Surprisingly, we also identified an oxidized TAG. The effect of metformin on cell redox status has been described previously 18 . The present results reaffirm the improvement of oxidative stress in PCOS patients after metformin treatment and reveal that an improvement also takes place in the lipid metabolism, thus decreasing lipoxidative damage.
Globally, fatty acid analyses indicated minor changes in fatty acid composition after metformin treatment. Although we observed changes in C20-based polyunsaturated fatty acids (C20:2 (n-6), C20:2 (n-6), C20:6 (n-6)), C22:1 (n-9) and C22:5(n-3), the variations only represented about 0.2% of ACL reduction, while no changes were detected in the other parameters calculated. The present results may indicate a modulation of elongase and desaturase activity, especially in the elongases ELOVL3, 5 and 2 and desaturases Δ8 and 5 43 . Fatty acid elongation and desaturation are crucial processes in the biosynthesis of saturated, monounsaturated and polyunsaturated fatty acids. Furthermore, the deregulation of these enzymes' activity has been previously related to IR and diabetes 44,45 . Specifically, previous studies in mice have reported a relationship between low ELOVL5 activity in the liver and glucose intolerance and insulin resistance 46,47 . In line with this, our results suggest that the minor alterations found in some PUFA could have been a response to slight changes in insulin levels and insulin sensitivity induced by metformin.
The principle limitation of the present study is the number of patients (n = 27) and the fact that they came from a reduced geographical area. Furthermore, we have used the Rotterdam criteria for this study, and not the last international guideline, due to the fact that our patients were part of a previously recruited cohort. Moreover, as we have stated before, the PCOS condition is usually associated with other pathologies, such as IR or obesity. In the present study we included 11 IR patients and 6 non-IR patients, and so we cannot be sure that IR was not a cofounding factor. Further studies with a larger cohort would guarantee a "purer" PCOS population. In addition to these population and clinical factors, our study has not enabled us to unravel the biological significance of this compositional complexity. Moreover, although sampling was scheduled to minimize the potential influence of diet on the lipidome, we cannot rule out the possibility that some compounds appearing as a result of the metabolism of nutrients affected the plasma lipidome in general, although the fact that our cohort was from a small geographic area would suggest that they were relatively homogeneous in terms of lifestyle and dietary habits.
All in all, we can conclude that metformin treatment induces a specific plasma lipidomic profile in PCOS women that is characterized mainly by a decrease in sphingolipids and glycerophospholipids and a reduction of lipoxidative species. We believe that this work could aid future research in the exploration of the molecular mechanisms involved in sphingolipid -steroid interaction. Our study confirms that lipid profile highlights a specific effect of metformin on PCOS women after 12 weeks of treatment.

Subjects.
This study was carried out in the Service of Endocrinology at the University Hospital Dr. Peset (Valencia, Spain). Plasma samples of twenty-seven women with PCOS before and after treatment for 12 weeks with metformin were analysed. Following the criterion of our Hospital and according to a previous study published 18 treatment with metformin was initiated at 500 mg per day (during the first 2 weeks). After 2 weeks it was increased to 1000 mg/day (weeks 3 and 4) and then to 1500 mg/d during weeks 5-12. Patients did not take any other medication.
Diagnosis of PCOS was confirmed using the Rotterdam criteria 48 . In brief, presence of oligoovulation (cycles longer than 35 days or less than 26 days) 49 ; free testosterone levels higher than 0.5 ng/dl (this cut-off level was estimated as the mean ± 2 SD according to the levels in healthy women); hirsutism (Ferriman-Gallwey score > 7) and polycystic ovaries (presence of at least 12 small −2 to 9 mm-follicles in each ovary), assessed by trans-vaginal ultrasonography. Ultrasound scans were performed and scored independently by one of two experienced and blinded reviewers.
None of the subjects had any condition affecting her reproductive physiology or any systemic or endocrine disease or galactorrhea. Exclusion criteria were malignant neoplasia, active infectious diseases, anemia, diabetes mellitus, history of ischaemic heart disease, thromboembolism, stroke and the taking of antihypertensive or lipid-lowering drugs. It was confirmed for all participants the absence of any medication that might have affected the hypothalamic-pituitary-gonadal axis during the previous semester. Approval by the ethics committee of the University Hospital Dr. Peset was obtained and the study was performed in accordance with the declaration of Helsinki. All participants provided their informed consent as required by these institutions.
Biochemical determinations. All participants were subject to an anthropometric evaluation to measure weight (kg), height (m) and waist circumference (cm) and to calculate the body mass index (BMI = weight (kg)/ height (m) 2 ). The weight was determined without footwear and with light clothing using electronic scales with an approximation of 0.1 kg and a capacity of up to 200 kg. The height was measured with a stadiometer with an approximation of 0.5 cm. The BMI was calculated by dividing the weight in kilograms by the square of the height in metres. The circumference of the waist was measured at the natural indentation between the 10th rib and the iliac crest and the circumference of the hips at the height of the major trochanter, using a metric tape with approximations of 0.5 cm. All measurements were made by the same nurse-study. Blood was collected from the antecubital vein on the follicular phase of the menstrual cycle or after 3 months of amenorrhea; after 12 hours of fasting. To separate serum and plasma from blood cells, samples were immediately centrifuged at 1500 g for 10 min at 4 _C.
www.nature.com/scientificreports www.nature.com/scientificreports/ Fresh samples were used to measure biochemical parameters and the remaining aliquots were stored at −80 °C for subsequent measurement of lipidomic parameters.
Total cholesterol and triglycerides were measured using enzymatic assays (Beckman Coulter, La Brea, CA, USA). High density lipoprotein cholesterol (HDLc) levels were obtained by a direct method using a Beckman LX-20 autoanalyser (Beckman Coulter, La Brea, CA, USA). The intraserial variation coefficient was < 3.5% for all determinations. Levels of low density lipoprotein cholesterol (LDLc) were calculated using the Friedewald formula 50 . Insulin concentration was determined by means of an enzymatic luminescence technique. Glucose levels were obtained with a Dax-72 autoanalyzer using enzymatic techniques (Bayer Diagnostic, Tarrytown, NY, USA). The homeostasis model assessment of insulin resistance (HOMA-IR) was calculated using baseline glucose and insulin levels: 51 HOMA = (fasting insulin (μU/ml) × fasting glucose (mmol/L)/22.5. Concentration of high sensitivity C-reactive protein (hsCRP) was assessed by an immunonephelometric assay (Behring Nephelometer II, Dade Behring, Inc., Newark, DE, USA) which had an intra-assay coefficient of variation of 8.7% and a sensitivity of 0.01 mg/L. Follicle-stimulating hormone (FSH) and luteinizing hormone (LH) were measured using a 2-site monoclonal non-isotopic system (Architect, Abbott Laboratories, Abbott Park, IL). Androstenedione, testosterone and sex hormone binding globulin (SHBG) were measured in our hospital's Clinical Analysis Service using specialized chemiluminiscence techniques.
Differences among anthropometric and metabolic parameters were analyzed by paired paired t-test for equal variances (p < 0.05 with Benjamini-Hochberg Multiple Testing Correction) and differences were considered significant when p < 0.05). Lipid extraction. A lipidomic analysis was performed to the plasma sample based on a previously validated method 52 . Briefly, to precipitate the protein fraction, 5 μl of miliQ water and 20 μl of methanol were added to 10 μl of plasma sample. After the addition, samples were shaken for 2 min. Then, 250 μl of MTBE plus internal standards were added and samples were ultra-sounded in a water bath (ATU Ultrasonidos, Valencia, Spain) with a frequency and power of 40 kHz and 100 W, respectively, at 10 °C for 30 min. Then, 75 μL of miliQ water were added to the mixture and organic phase was separated by centrifugation (1,400 g) at 10 °C for 10 min. The upper phase, containing all the extracted lipid species, was collected and subjected to mass-spectrometry. A pool of all lipid extracts was prepared and used as quality controls (QC) as previously described 53 .
LC-MS/MS method. Lipid extracts were subjected to liquid chromatography coupled to mass-spectrometry (LC-MS) using an Agilent UPLC 1290 coupled to the Q-TOF MS/MS 6520 (Agilent Technologies, Barcelona, Spain) basing on previously published method 54 . Sample compartment was refrigerated at 4 °C and, for each sample, 10 μl of lipid extract was applied onto 1.8 μm particle 100 × 2.1 mm id Waters Acquity HSS T3 column (Waters, Mildord, MA, USA) heated at 55 °C. The flow rate was 400 μl/min with solvent A composed of 10 mM ammonium acetate in acetonitrile-water (40:60, v/v) and solvent B composed of 10 mM ammonium acetate in acetonitrile-isopropanol (10:90, v/v). The gradient started at 40% B and reached 100% B in 10 min and held for 2 min. Finally, the system was switched back to 60% B and equilibrated for 3 min. Duplicate runs of the samples were performed to collect positive and negative electrospray ionized lipid species in a TOF mode, operated in full-scan mode at 100 to 3000 m/z in an extended dynamic range (2 GHz), using N 2 as nebulizer gas (5 L/min, 350 °C). The capillary voltage was set 3500 V with a scan rate of 1 scan/s. Continuous infusion using a double spray with masses 121.050873, 922.009798 (positive ion mode) and 119.036320, 966.000725 (negative ion mode) was used for in-run calibration of the mass spectrometer. For MS/MS analyses, we applied a previously described method 55 .
Data analyses. The MassHunter Data Analysis Software (Agilent Technologies, Barcelona, Spain) was used to collect the results and the MassHunter Qualitative Analysis Software (Agilent Technologies, Barcelona, Spain) to obtain the molecular features of the samples, representing different, co-migrating ionic species of a given molecular entity (i.e. ion adducts) using the Molecular Feature Extractor algorithm (Agilent Technologies, Barcelona, Spain) 31 . We selected samples with a minimum absolute abundance of 5000 counts and with a minimum of 2 ions. Compounds from different samples were aligned using a RT window of 0.1% ± 0.15 min and a mass window of 10.0 ppm ± 2.0 mDa. Only common features (found in at least 50% of the samples of the same condition) were analyzed, correcting for individual bias and excluding possible contaminants and artefacts. Finally, MassHunter Mass Profiler Professional Software (Agilent Technologies, Barcelona, Spain) was used to perform a non-targeted lipidomic analysis over the extracted features. Only common features (found in at least 50% of the samples of the same condition) were taken into account to correct for individual bias. Multivariate statistics (Hierarchical Clustering, PCA and Random Forest analyses) and biomarker analysis were done using both MassHunter Mass Profiler Professional and Metaboanalyst softwares 56,57 . The masses representing significant differences by the paired t-test for equal variances (p < 0.05 with Benjamini-Hochberg Multiple Testing Correction) were searched against the LIPID MAPS database (exact mass ppm < 20) and identified with the R-based tool LipidMatch 58 . Finally, the MS/MS spectra were checked using the LipidBlast software 59 . targeted Lipidomic analysis: plasma fatty acids composition. Fatty acid preparation. After lipid extraction, fatty acyl groups were analyzed as methyl esters derivatives by gas chromatography (GC) 55 . Briefly, fatty acids were transesterified by incubation in 2 ml of 5% methanolic HCl at 75 °C for 90 min. The resulting fatty acid methyl esters (FAMEs) were extracted by adding 2 ml of n-pentane and 1 ml of saturated NaCl solution. The n-pentane phase was separated, evaporated under N2 gas, re-dissolved in 80 μl of carbon disulfide and 2 μl were used for GC analysis.
GC method. The analysis was performed on a GC System 7890A with a Series Injector 7683B and a flame ionization detector (FID) (Agilent Technologies Inc., Barcelona, Spain) equipped with a DBWAX capillary column (length 30 m × inner diameter 0.25 mm × film thickness 0.20 μm; Agilent Technologies Inc., Barcelona, Spain). The injections were performed in the splitless mode. The temperature of the injector was 220 °C. The flow rate of helium (99.99%) carrier gas was maintained at a constant rate of 1.8 ml/min. The column temperature was held at 145 °C for 5 min; subsequently, the column temperature was increased by 2 °C/min to 245 °C for 50 min, and held at 245 °C for 10 min, and with a post-run of 250 °C for 10 min.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.