Tailored liquid chromatography–mass spectrometry analysis improves the coverage of the intracellular metabolome of HepaRG cells

Accepted Manuscript Title: Tailored LC–MS analysis improves the coverage of the intracellular metabolome of HepaRG cells Authors: Matthias Cuykx, Noelia Negreira, Charlie Beirnaert, Nele Van den Eede, Robim Rodrigues, Tamara Vanhaecke, Kris Laukens, Adrian Covaci PII: DOI: Reference:

S0021-9673(17)30135-8 http://dx.doi.org/doi:10.1016/j.chroma.2017.01.050 CHROMA 358229

To appear in:

Journal of Chromatography A

Received date: Revised date: Accepted date:

14-9-2016 15-1-2017 22-1-2017

Please cite this article as: Matthias Cuykx, Noelia Negreira, Charlie Beirnaert, Nele Van den Eede, Robim Rodrigues, Tamara Vanhaecke, Kris Laukens, Adrian Covaci, Tailored LC–MS analysis improves the coverage of the intracellular metabolome of HepaRG cells, Journal of Chromatography A http://dx.doi.org/10.1016/j.chroma.2017.01.050 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Tailored LC-MS analysis improves the coverage of the intracellular

metabolome of HepaRG cells

Matthias Cuykx1*, Noelia Negreira1, Charlie Beirnaert2, Nele Van den Eede1, Robim Rodrigues3,

Tamara Vanhaecke3, Kris Laukens2, Adrian Covaci1*

1

Toxicological Center, Department of Pharmaceutical Sciences, University of Antwerp,

Universiteitsplein 1, 2610 Wilrijk-Antwerp, Belgium

2

Department of Mathematics & Computer Science, University of Antwerp, Middelheimlaan 1, 2020

Antwerp, Belgium

3

Department of In Vitro Toxicology and Dermato-Cosmetology, Center for Pharmaceutical Research,

Vrije Universiteit Brussel, Laarbeeklaan 103, 1090 Jette, Belgium

*

Corresponding Author Contacts: [email protected]; [email protected]

1

Highlights 

Fractionation and optimised analysis improves the metabolomic coverage



The lipidome was analysed with a tailored RP-LC gradient for each ionisation mode



HILIC-LC has better separation properties for neutral and basic polar compounds



Anionic polar compounds were better retained and separated with ion paring RP-LC



A four-run analysis provides coverage of 2200 high precision metabolites (mRSD<10%)

Abstract Metabolomics protocols are often combined with Liquid Chromatography-Mass spectrometry (LCMS) using mostly reversed phase chromatography coupled to accurate mass spectrometry, e.g. quadrupole time-of-flight (QTOF) mass spectrometers to measure as many metabolites as possible. In this study, we optimised the LC-MS separation of cell extracts after fractionation in polar and nonpolar fractions. Both phases were analysed separately in a tailored approach in four different runs (two for the non-polar and two for the polar-fraction), each of them specifically adapted to improve the separation of the metabolites present in the extract. This approach improves the coverage of a broad range of the metabolome of the HepaRG cells and the separation of intra-class metabolites. The non-polar fraction was analysed using a C18-column with end-capping, mobile phase compositions were specifically adapted for each ionisation mode using different co-solvents and buffers. The polar extracts were analysed with a mixed mode Hydrophilic Interaction Liquid Chromatography (HILIC) system. Acidic metabolites from glycolysis and the Krebs cycle, together with phosphorylated compounds, were best detected with a method using ion pairing (IP) with tributylamine and separation on a phenyl-hexyl column. Accurate mass detection was performed with the QTOF in MS-mode only using an extended dynamic range to improve the quality of the dataset. Parameters with the greatest impact on the detection were the balance between mass accuracy and linear range, the fragmentor voltage, the capillary voltage, the nozzle voltage, and the nebuliser pressure. By using a tailored approach for the intracellular HepaRG metabolome, consisting of three different LC techniques, over 2,200 metabolites can be measured with a high precision and acceptable linear range. The developed method is suited for qualitative untargeted LC-MS metabolomics studies.

Keywords: HepaRG, Metabolomics, Optimization, Liquid Chromatography-Mass Spectrometry, Hydrophilic Liquid Interaction Chromatography, Ion Pairing

2

1. Introduction At the introduction of metabolomics as a new “–omics” domain in 1999, the field of research proved to be a valuable source of downstream information of the actual phenotype or state of organisms [1]. In comparison to conventional analytical methods where a small number of known compounds are quantitatively measured, untargeted metabolomics measure hundreds of metabolites, resulting in a fingerprint profile representing the composition of endogenous metabolites. For successful metabolomics applications, a reliable protocol should reduce variation as much as possible [2]. A full analytical validation with parameters as selectivity, accuracy, precision, range, linearity, LOD and matrix effects for each metabolite separately would be tedious and time consuming [3]. In untargeted studies, the metabolites observed are too numerous and their exact identity is in most cases unknown. Nonetheless, the recommendations can be used as a guideline to improve the quality of metabolomics research [2,4,5]. Multiple standards are often merged into a reference mix which is then validated as a representation of a complex sample [4]. Another common way to include an extra level of quality assurance during general untargeted studies is the use of pooled samples as a Quality Control, i.e. a sample injected multiple times to obtain information about the quality of the acquisition and the dataset [6–8]. Since this common pool of extracts is identical to real, complex samples, it is also an ideal tool to optimise LC-MS parameters before real toxicological experiments are to be made. The chemical properties are extremely variable between different classes of endogenous metabolites and a method covering the entire metabolic window is hard to design, if not impossible. For example, there is a general division in polar metabolites, such as amino acids, nucleic acids, sugars, etc. (further referred to as the polar fraction) and the lipidome consisting out of fatty acids, phospholipids, triglycerides, etc. (further referred to as the non-polar fraction). Analysing all metabolites in one experiment requires performant instruments and, even with advanced technology, optimal separation cannot be achieved [6,9–11]. A good mass resolution is essential to differentiate between two molecules with the same nominal mass; High Resolution mass spectrometers, like QTOF, Orbitrap or Fourier Transform Ion Cyclotron Resonance (FT-ICR) have a resolution ranging from 10,000 (QTOF) and 100,000 (Orbitrap) to even 1,000,000 (FT-ICR). This improves mass accuracy, reducing the amount of possible chemical formulas for the compound of interest. However, such systems alone cannot make the difference between isomers. Isomers have the same molecular formula, but they have different structures leading to different biological functions. Alterations between these molecules can go unnoticed (false negative result), or might falsely suggest a significant difference (false positive). Another issue is ion suppression or enhancement: the presence of multiple metabolites makes them interfere during their evaporation, resulting in a respectively lower or higher signal in comparison to pure standards. This interference can cause a form of bias and reduces the quality of the dataset [12,13]. One way to improve detection sensitivity and accuracy is the use of hyphenated systems such as liquid chromatography coupled to mass spectrometry (LC-MS), a common technique in metabolomics [14–16]. A chromatographic separation also reduces in-source interactions, such as ion suppression, allowing to represent the actual metabolome more accurately in comparison to methods which suffer from ion suppression effects [17]. Another advantage of chromatographic separation is the generated second dimension, which can be used to improve peak picking [18]. This process is then able to remove noise and signals related to the solvent. However, the most important advantage of chromatographic separation hyphenated to MS-detection is that isomeric features are measured separately, increasing confidence in identification. As a result of this hyphenation, the

3

quality of the dataset during the statistical analysis can be improved, resulting in less false positive results. Generally, a metabolomics study is performed using one single generic analytical platform, with one column type and one mobile phase composition for both ionisation modes, trying to measure as many metabolites as possible [19–21]. Popular methods use either reversed phase (C18 or C8 stationary phase) columns or HILIC (bare silica, amide, or zwitterionic stationary phase) approach [22,23]. These methods require more time than direct-infusion systems, and although LC-separation is a powerful separation technique, the current technologies are not able to separate all metabolites, often leading to coelution of similar compounds. Another limitation is the selectivity of the stationary phases: reversed phase systems are useful for (slightly) hydrophobic compounds, but fail when extremely polar metabolites should be separated (e.g. nucleotides). HILIC systems offer a solution, but usually cannot separate intra-class metabolites (e.g. different chain lengths and positions in phospholipids) [22,23]. Fractionation of extracts in a polar and non-polar fraction followed by separate analysis provides an opportunity to increase metabolomics coverage in untargeted studies [4,14,24,25]. However, this approach increases sample preparation, causes possible degradation and introduces variation. Analysis time also increases, reducing sample throughput [19,26–29]. An optimal protocol establishes a balance between throughput and sensitivity without introducing too much variance. The main goal in this study was to optimise a LC-MS platform to cover the metabolome of cultured HepaRG cells to the fullest extent. HepaRG cells are an in vitro hepatic cell line derived from a hepatocellular carcinoma; their metabolic competence make them an emerging tool in hepatotoxicological research [30,31]. Several LC- and MS-parameters were evaluated in order to obtain high-quality LC-MS data, specific for the fractions obtained. Different stationary and mobile phases of both RP and HILIC systems were tested in order to improve chromatographic efficiency. Different MS-parameters were compared to improve detection sensitivity and resolution. For the final method, priority was given to conditions that resulted in a linear response, high sensitivity and high intra-batch repeatability for a large number of features.

2. Methods 2.1. Materials Cryopreserved differentiated HepaRG® cells and recommended culture media (Basal Hepatic medium®, HepaRG Thaw Seed and General Purposes Supplement® and Metabolism and Maintenance supplement®) were purchased from Biopredic International (Rennes, France). Phosphate Buffer Saline (PBS) was bought from Gibco Life Sciences (Gent, Belgium). 12 Well cellculture plates were bought at Greiner bio-one (Vilvoorde, Belgium). Isotope labelled standards trypthophane-2’,4’,5’,6’,7’-d5 (98 %), lauric acid-12,12,12-d3 (99 %) were obtained from CDN Isotopes (Pointe-Claire, Quebec, Canada). Polar lipid standards PC-17:0, PA-17:0, Cer-17:0 and LPC-17:0 were bought from Avanti Lipids (Alabaster, Alabama, US). ATP-13C10, TG-(12C15 13 C:0)3, lysine-13C6-15N2, glucose-13C6, ADP, ATP, stearic acid, folic acid, mono-, di- and trioleylglycerol, misoprostol, cholic acid, cholic acid-d4, phospho-enol-pyruvic acid, ornithine, glutamate-d4, leucined3, adenine, glucose phosphate, citric acid, caffeine, N-acetyl glucosamine, pyridoxal-d3, dopamined4, palmitoylcarnitine, cholesterylpalmitate and a standardised mix of amino acids (AAS18) were bought from Sigma Aldrich (st. Louis, Missouri, USA). Ammonium acetate for analysis (>98 %) (NH4Ac), formic acid (>98 %) (FA), Acetic Acid (>98 %) (HAc), chloroform (for analysis, 99.8 %), 4

isopropanol (IPA) (for analysis) and methanol (MeOH) (gradient grade for liquid chromatography, Lichrosolv® >99.9 %) were bought from Merck (Darmstadt, Germany). Ultrapure (milliQ) water was obtained from an Elga Pure Lab apparatus (Tienen, Belgium). Acetonitrile (for LC-analysis) (ACN) was bought from Fisher (Loughborough, UK). Ammonium hydroxide (30 %) (NH4OH) and ammonium formate (NH4F) were bought at Sigma Aldrich (Steinheim, Germany). Ion pairing agents, tributylamine (99 %) (TBA), triethylamine (99 %) (TEA) and methyl-piperidine (98 %) (1-MP) were bought at Sigma-Aldrich, Acros (Geel, Belgium) and Merck, respectively.

2.2. Protocol 2.2.1. Column screening The injection of standards at a concentration of 1,000 pg/µL was prepared in the mobile phase composition and used to assess a first separation method where the standards representing different metabolic classes are spread across the chromatogram. Further adjustments were made and reassessed on cell extracts to improve the separation of the complex extracts. The optimisation with these cell extracts include matrix effects which can have an influence on the separation and guarantees an optimal separation of within- and between-class metabolites. 2.2.2 Sample preparation of cell extracts HepaRG® cells were thawed and seeded according to the manufacturers’ guidelines. HepaRG® cells were thawed and seeded in 12 well-plates at a concentration of 0.89 x 106 cells per well. Plates were incubated for 7 days at 37 °C and 5 % CO2. Medium was entirely renewed every three days. After seven days, the differentiated cells were ready for extraction. The extraction is performed using liquid-liquid extraction (LLE) according to our previous protocol [32,33]. Cells were frozen on liquid nitrogen and remaining metabolism was quenched with 80 % (v/v) MeOH in water. After two minutes, the cells were scraped and transferred to vials containing water and chloroform (final solvent ratio 2/3/2 water/MeOH/CHCl3). The mixture was vortexed and centrifuged before a polar and a non-polar fraction was recovered. Both fractions were analysed separately. Details of the entire protocol (seeding and extraction) are available in Supplementary Information 1. 2.2.3. Assessment LOD and linearity After optimisation of the LC and MS conditions, LOD and linearity were investigated with standards prepared in mobile phase composition at concentrations between 10 and 2,000 pg/µL; precision of the method was evaluated on the cell extracts (section 2.3). 2.2.4. LC-analysis of the non-polar fraction Cell extracts were reconstituted with 35/65 IPA/MeOH and separated on a Kinetex® XB-C18 100 Å column (150 x 2.10 mm, 1.7 µm particle size) (Phenomenex, Utrecht, the Netherlands) using an Agilent 1290 Infinity UPLC (Agilent, USA). Injection volume was varied between 1.5 and 3 µL. Mobile phases tested were mixtures of milliQ water, MeOH, ACN and IPA using ammonium formate (NH4F), formic acid (FA), ammonium acetate (NH4Ac) and acetic acid (HAc) as pH-buffers. Separation of metabolites was tested on a gradient elution with flows of 0.2, 0.3 and 0.4 mL/min at a temperature of 55 °C to reduce column backpressure. Gradient parameters were adjusted to improve peak width, feature density over the chromatogram and the number of features detected. Detection was performed using an Agilent 6 530 QTOF-MS with Agilent-Jet-Stream-Electrospray Ionisation (AJS-ESI, 5

Agilent technologies). The analysis was performed in positive and negative ionisation modes. Source temperature and gas flows were tested between 300 °C to 350 °C and 8-11 L/min, respectively. Capillary voltage and fragmentor voltage were also varied between respectively 2,500 V to 5,000 V and 125 V to 225 V. Acquisition was obtained in both 4 GHz (High Resolution mode) and 2 GHz (Extended Dynamic mode) using multiple scan rates ranging from 50 to 500 ms per scan. LCoptimizing runs were obtained in centroid mode; MS-optimizing runs were acquired in profile mode. 2.2.5. LC-analysis of the polar fraction Cell extracts were reconstituted with mixtures of ACN/water or MeOH/water depending on the mobile phase used. Mobile phases were composed of mixtures of milliQ water and either ACN or MeOH. Different pH-ranges were tested using NH4F, FA, NH4Ac, HAc and NH4OH. Different columns were tested using identical LC-MS parameters for both ionisation modes. As a result, differences could only be related to the difference in column or mobile phase composition. A selection of columns included a Kinetex HILIC (stationary phase: bare silica, Phenomenex), a Luna HILIC (stationary phase: diol, Phenomenex), a Luna NH2 (stationary phase: aminopropyl, Phenomenex) and an iHILIC Fusion (mixed mode, HILICON AB, Umeå, Sweden). The optimal column and mobile phase composition were selected by using an initial gradient of 95 % organic phase for 2.5 min, decreasing to 10 % organic phase at 20 min and a 5 min rinse. The best performing system was further adapted to obtain better chromatographic results. For the negative mode, additional experiments using trimethylamine (TEA), tributylamine (TBA) or 1methylpiperidine (1-MP) as ion pairing agents on a Synergi-polar and a Kinetex PFP column were evaluated to compare its performance against the HILIC conditions. When optimising, concentrations ranging from 1 mM till 10 mM were tested at different pH values between 2 to 9. An initial gradient of 1 % B to 95 % B was performed, followed by tailoring the gradient to an optimal performance. Since the optimal pH of the optimal mobile phase was not compatible with the silica carriers, a Gemini phenyl-hexyl (Phenomenex), which also provides pi-selectivity at high pH stability, was investigated as an alternative. 2.2.6. Final methods 2.2.6.1. Non-polar fraction, positive ionisation mode Separation was performed on a Kinetex XB-C18 (150 x 2.1 mm; 1.7 µm particle size) RP-column. A buffer of 10 mM NH4AC and 0.1 % HAc in milliQ water (pH 4.2) was prepared to stabilise the pH of the mobile phases. Mobile phase A was composed of ACN/buffer (1/1), mobile phase B was made of 2/10/88 buffer/ACN/IPA. A gradient elution with following program was performed: 55 % B for 1 min, increasing to 70 % at 5 min. Afterwards, the percentage of B was increased to 98 % at 25 min and kept for a 4 min rinse at 100 % B to complete the analytical separation. The column was equilibrated for 9 min at starting conditions, resulting in a total run time of 40 min per sample. The column was heated to 55 °C, flow rate was kept constant at 0.25 mL/min. MS conditions were constant during the entire run: drying gas temperature and flow were 325 °C and 8 L/min respectively. Capillary, nozzle and fragmentor voltage were 3,500 V, 500 V and 175 V, respectively. 2.2.6.2. Non-polar fraction, negative ionisation mode Separation was performed on a Kinetex XB-C18 (150 x 2.1 mm; 1.7 µm particle size) RP-column. Mobile phase A was composed of 1/1 MeOH and 10 mM NH4AC in milliQ water (pH 6.7), mobile phase B was made of 10mM NH4Ac in milliQ water/MeOH/IPA (2/10/88). A gradient elution with following program was performed: 55 % B for 1 min, increasing to 70 % B at 5 min. Afterwards, 6

percentage of B was increased to 91 % at 20 min and 6 min rinse at 100 % B to complete the analytical separation, the column was re-equilibrated for 9 min at starting conditions before the next injection. The column was heated to 55 °C, and the flow rate was 0.25 mL/min. MS source conditions were constant during the entire run: drying gas temperature and flow were 350 °C and 8 L/min respectively. Capillary, nozzle and fragmentor voltage were 3,750 V, 0 V and 175 V, respectively. After 21 min, the capillary and nozzle voltage were slightly altered to 3,500 V and 500 V, respectively. 2.2.6.3. Polar fraction, positive ionisation mode An iHILIC Fusion (100 x 2.1 mm; 1.8 µm particle size) column was used in the final method. Mobile phase A was made of 10 mM NH4F and 0.1 % FA v/v (pH 3.15) and mobile phase B was ACN/MeOH (98/2, v/v), flow rate was 0.3 mL/min. A gradient elution started with 95 % B. After 2 min, the percentage of B decreased linearly to 65 % at 8 min, and decreased further to 25 % at 13 min. A rinse was performed with 25 % of B for 2 min and the column was re-equilibrated at starting conditions for 6 min. Column was heated to 30 °C to prevent temperature related changes in chromatography. MS conditions were constant throughout the run. Drying gas temperature and flow rate were 250 °C and 8 L/min. Sheet gas was 350 °C at 11 L/min. Capillary and fragmentor voltages were 2,000 V and 150 V, respectively. The nebulizer was set at 45 psig. 2.2.6.4. Polar fraction, negative ionisation mode A Gemini® Phenyl-hexyl (150 x 2 mm, 3 µm particle size) was used in the final method. An isocratic pump with isolated tubing prevented contamination of pairing agents on the conventional hardware. The mobile phase consisted of 25 % MeOH with 10 mM TBA and 0.02 % FA (v/v) (pH = 9) and was pumped through the column at a flow rate of 0.2 mL/min. Runtime was set at 20 min. The temperature and flow rate of drying and sheath gasses were similar: 250 °C and 10 L/min. Capillary and fragmentor voltages were 2 000 V and 100 V, respectively. The nebulizer was set at 45 psig. 2.2.7 Comparison to a generic LC-MS metabolomics platform A generic analysis was performed to evaluate the benefits of the developed platform. Since the generic platform is based on a single LC-system, no liquid-liquid extraction was performed: after scraping, the extract was evaporated to dryness and reconstituted in 100 µl of 20 % (v/v) MeOH/water. Chromatographic separation was performed on a Kinetex XB-C18 (150 x 2.1 mm; 1.7 µm particle size) RP-column. Mobile phase A was composed of a buffer of 10 mM NH4AC in milliQ water (pH 6.7) and mobile phase B was made of ACN. A gradient elution with following program was implemented: 10 % B for 1 min, increasing to 70 % at 12 min. Afterwards, the percentage of B was increased to 98 % at 30 min and kept for a 10 min rinse to complete the analytical separation. The column was equilibrated for 5 min at starting conditions, resulting in a total run time of 45 min per sample. The column was heated to 55 °C, flow rate was kept constant at 0.25 mL/min. MS conditions were constant during the entire run: drying gas temperature and flow were 325 °C and 8 L/min respectively. Capillary, nozzle and fragmentor voltage were 3,500 V, 500 V and 175 V, respectively. These parameters were applied for both ionisation modes.

2.3. Data treatment Evaluation of LC-and MS- parameters was performed using the Mass-Hunter (version 2.06.00, Agilent technologies) qualitative software and R [34]. Standards were extracted from the chromatogram by 7

the “Find by Formula”-algorithm (FBF) of Mass-Hunter and integrated by the mono-isotopic mass of the most abundant adduct. When using the Mass-Hunter “Molecular Feature Extractor”-algorithm (MFE, Agilent Technologies) for the complex cell extracts, deconvolution parameters were set as follows: threshold of 3,000 and quality score >80 %. The algorithm searches the raw data for different m/z chromatograms and groups signals which correlate to isotopes or adducts into molecular features. The extracted molecular features are thus groups of several isotopes and adducts, representing the m/z signals of a single metabolite. For intermediary precision, consecutive analytical runs were merged into a data frame using the Mass Profiler (Agilent Technologies), the columns represent the samples (identified by the injection number) and the rows represent the extracted molecular features (identified as a retention time and a mass representing a single molecule with multiple m/z values). The sum of the areas of all corresponding ions of the molecular feature is considered as the dependent value of the variable. Results were exported to CSVs which were summarized using R [34]. During the evaluation, general peak shape, distribution, chromatographic peak width, chromatographic resolution, and the number of detected compounds were compared between the different LC-conditions without changing the MS-parameters. MS-performance was evaluated by comparing the number of features detected, full width at half maximum resolution (FWHM), signal intensities and the number of features reaching detector saturation. For the optimal LC-MS method, linearity and LOD were further studied on standards in order to comply with metabolomics instrumental consensus [2,5,35]. Because of the large differences in response between metabolites and the great differences in intracellular concentration, it is not possible to develop a method that is ideal for all features, since no standards are available for each metabolite. Nonetheless, when validating compounds of multiple classes, these can represent the quality of the method for the represented class, providing more confidence than no validation at all. The standards represent different classes of metabolites, providing insights (but not a guarantee) in the quality of the method. Precision values are calculated with raw data obtained from HepaRG extracts: no internal standard correction, scaling, filtering or normalisation has been performed, the only bias towards improved quality of the dataset might occur during Feature Extraction. This approach represents the real introduced variation during the LC-MS analysis, further processing (e.g. scaling and normalisation) would lower the obtained value, resulting in an underestimation of the real variation. The data were filtered to obtain only molecular features present in 80 % of the samples. The precision of the dataset is evaluated according to Parsons et al. [36]. Of the retained molecular features, the relative standard deviation (RSD) was calculated, giving a value for the variance. Since there are many molecular features in metabolomics, the RSDs are considered as a group. These RSDs show a distribution which is skewed to the right. All RSDs were shown into a histogram or a boxplot to evaluate the distribution of the RSD of all molecular features. To generate a single value for each dataset, the median of these RSDs (mRSD) is taken as a representative value for the general quality of the dataset [36].

8

3. Results and Discussion The main goal in the development of untargeted metabolomics methods is the separation of the highest number of metabolites in order to obtain a reliable representation of the phenotype. When fractionating the entire sample in different subsets, each of them analysed with a tailored approach, the platform will improve the metabolomics coverage and precision in comparison to a one-shot analysis. The use of standards representing different classes of the metabolome is necessary to obtain optimal parameters, but they are not enough to evaluate chromatographic conditions. When using cell extracts containing hundreds to thousands of metabolites, the chromatographic conditions can be further optimised since the performance can be evaluated for conditions which are identical to the sample composition during the real application. The final parameters are summarised in Table 1. Each optimisation step is discussed in the following sections.

3.1. Non polar fraction A reversed phase Kinetex® XB-C18-UPLC column provided good chromatographic separation. Differences were observed depending on the type of cosolvent (methanol vs. acetonitrile) due to their different selectivity properties: acetonitrile has more affinity for unsaturated bonds and pisystems, speeding up the elution of unsaturated lipids. As a result, double-bond selectivity would be reduced in the positive run, but the separation between lipid species in general increased, resulting in improved detection. Differences were also noted with the additives (10 mM ammonium acetate, pH: 6.7 vs. 0.1 % acetic acid and 5 mM ammonium acetate, pH: 4.2). Neutral pH made acidic species anionic, increasing their polarity. This lowers their hydrophobicity and increases the importance of the hydrophobic chain, resulting in improved separation. For neutral to basic species, acidic buffers improved chromatographic resolution, but also increased the ionization of these compounds, improving sensitivity. The optimal mobile phases were different between positive and negative mode. In positive mode, acetonitrile proved to be the best cosolvent in combination with a buffer of 5 mM NH4Ac and 0.1 % (v/v) HAc in milliQ water, giving an output of 1,200 compounds (Figure 1A). In negative mode, optimal separation was obtained with methanol as a cosolvent while the optimal buffer was 10 mM NH4Ac and 700 compounds were detected (Figure 1B). Total ion current chromatograms (TIC) and peak widths of the detected features are given in Figures SI 1 and SI 2, respectively. Figures SI 3 and SI 4 represent the normalised chromatograms of the standards analysed in the positive and negative ionisation mode, respectively. More details are provided in Tables SI 1 and SI 2 for the positive and negative ionisation mode, respectively. When using reversed phase systems, we noticed significant carry-over for multiple high-lipophilic analytes, belonging to mainly the triglyceride family. Longer rinsing times with relative lipophilic solvents (IPA) reduced carry-over significantly [37]. This precaution is also important in the negative mode although triglycerides are not detected in negative mode, they are still injected on the column and therefore need to elute prior to a new injection in order to ensure reproducible results. Injection of a standard mix provided an overall precision < 5 %, the RSD is inversely correlated to the abundance of the lipophilic compounds. Linearity and limits of detection (LOD) are detailed in Table 2. The lower LOD ranged between 10 and 100 pg/µL, providing a (semi-)quantification range of 2 to 3 orders of magnitude (up to 1 - 2 ng/µL, saturation at higher concentration). Linearity showed coefficients of determination (R²) higher than 0.99 for most standards in the negative ionisation mode. In positive mode, saturation was observed at 2 ng/µL, compromising the linearity. Exclusion of this highest concentration point resulted in better linearity (e.g. ceramide (17:0): R² 0.86  0.99). Although the results are not sufficient for absolute quantification, the linear response is sufficient for 9

qualitative comparison when two sample classes are compared against each other (e.g. control vs. disease). Because the same extract was injected multiple times, precision could be assessed for the entire dataset, shown as a boxplot and represented by a single value, such as the mRSD (Figures 2A and B) [36]. Precision of the method was within the range of the metabolomics standards: for positive and negative ionisation mode, mRSDs were of 5.8 % and 5.1 %, respectively. Only 13 % (positive mode) and 6 % (negative mode) of the extracted and aligned features had a RSD > 15 %. This good intermediary precision is in agreement with the requirements stated in metabolomics consensus [2,5]. The precision of other analytical batches was evaluated by the QC-samples confirming the quality of the described methods. It is important to note that these values are calculated with raw data: no internal standard correction, scaling, filtering or normalisation has been performed yet. The only bias towards improved quality of the dataset can be correlated to the Feature Extraction. This step was essential to assess the quality of the LC-MS method and could not be excluded. Cell extracts were stable in a cooled autosampler (4 °C) up to 24 h, while degradation occurred during further storage, as shown in Figure SI-5. We recommend keeping the number of thawed samples to a minimum as stability is better at -80 °C. However, reconstitution of dry samples stored at -80 °C for more than two weeks does not guarantee identical profiles, since degradation may occur [38]. The observed degradation was considered acceptable up to ten days after storage. Longer storage times were observed to induce too much variation in comparison to the original extract.

3.2. Polar fraction For the polar fraction in positive mode, HILIC systems provided better selectivity and retention in comparison to different RP-stationary phases (pentafluoro-phenyl, C8 and biphenyl). When analysing polar compounds, RP systems are generally less performant in comparison to HILIC methods [4,16]. In positive mode, a straightforward method with a bare silica column provided a good chromatographic separation. Although the exact interaction between the analytes and the stationary phase is complex, most literature explain the retention based on hydrophilic interactions and H-bonds between the accepting stationary phase (negatively charged silica) and donor metabolites (e.g. protonated basic metabolites) [4,16,39,40]. The main advantage of bare silica HILIC columns is the easy interpretation of the retention times: the later a compound elutes, the more polar it is. With a neutral pH, the silica is negatively charged, providing an extra retention factor for basic compounds [40]. Major drawbacks were the unspecific interactions and the observed tailing for basic compounds. The unspecific interactions were explained by the absence of organic groups on the silica surface. Therefore, only general polarity and proton bonding properties are the main factors for chromatographic separation. As a result, there were regions where a lot of coelution occurred: at 5 min, a cluster of neutral amino acids could be observed; basic structures typically coeluted at 22 min. As an alternative solution, the iHILIC® Fusion (HILICON, Umeå, Sweden) was suggested since the column chemistry is composed of different functional groups used in popular HILIC systems (amide, hydroxyl, sulphate, phosphate and bonded trimethylamine) [41]. As a result, the interactions were more complex, showing better selectivity than a bare silica column. Datawise, the same number of features could be detected in a shorter run, generally having a better peak distribution and providing better separation between similar compounds. Feature distribution is shown in Figure 1C, TCC and 10

peak widths are shown in Figure SI-1C and SI-2C respectively. A normalised chromatogram of the standards (Figure SI 6) and a table with the details of these standards (Table SI 3) are also provided. For example, this column was able to chromatographically resolve leucine and isoleucine (Figure SI7), which only differ by the position of the methyl group on the aliphatic chain of the amino acid. Table 3 shows the linearity and LOD for standards. The method was able to detect 700 compounds with mRSD of 8 % (Figure 2C), which was a better precision than using the bare silica column, which had mRSD of 11 % (data not shown,). Other advantages were the improved sensitivity and linear response. The only disadvantage is the risk for ‘overloading’ the column, the stationary phase can get oversaturated and chromatographic precision will be compromised. This effect has been noticed with a standard mix at a concentration of 5 ng/µL. This issue should not invoke major problems when avoiding too concentrated samples. The polar cell extracts were stable in a cooled autosampler (4 °C) up to 18 h; degradation occurred during further storage. Reconstitution of dry samples stored at -80 °C for up to 10 days did not invoke major alterations. Longer storage times were observed to induce too much variation in comparison to the ‘fresh’ cell extracts. For the negative mode, a Luna NH2 (Phenomenex) with an aminopropyl stationary phase was suggested, since it provided better retention as its mechanism is based on ionic interaction between the basic stationary phase and the anionic (acidic) analytes [16,42,43]. However, injection of standards resulted in poor chromatographic results: very acidic compounds (e.g. phosphates, nucleotides, or glycolytic cycle metabolites) were retained too strongly, eluting only with 100 % of water at high pH (9.4) after at least 30 min of rinsing, resulting in broad peaks with reduced sensitivity. Figure 3 shows the extracted ion chromatogram (EIC) for aspartate, while the EICs of ADP, ATP and citric acid are shown in supplementary information (Figures SI-8, SI-9 and SI-10). For other standards, similar effects could be observed. In HILIC systems, this extreme polarity (having a log D below -7 to even -11 at basic pH according to Chemspider (Royal Society of Chemistry; UK) resulted in long analytical runs where the polar compounds eluted when using isocratic conditions at 100 % water. The output of this HILIC system was not adequate since peaks were broad, reproducibility and sensitivity were bad and some compounds were even not detected. Furthermore, these conditions were time-consuming, highly stressful for the column and MS and provide bad analytical results. Proposed solutions, such as the addition of chelating agents and fresh solvents (<1 h) did not improve peak shape and sensitivity [43,44]. When applying ion pairing in metabolomics, the rationale is that the basic pairing agents interact with the acidic metabolites, the improved hydrophobicity results in a longer retention. [45,46] Most studies used C18-columns [46–48]. Indeed, very polar compounds were retained and separated, but a stationary phase with greater aromatic (pi-bonding) interaction would improve the separation of aromatic metabolites, e.g. nucleic acids and nucleotides [49]. Best results were obtained with ion pairing (IP) agents on a Synergi column. Since the analytes eluted with a percentage of organic phase, ionisation was improved significantly, resulting in quantifiable peaks. When using ACN, no retention was observed, but MeOH greatly improved retention and provided chromatographic resolution in short runtimes [50]. As a result, MeOH provided better separation and was preferred as mobile phase. The choice of pairing agent significantly contributed to retention improvement; tributylamine provided better retention and separation in comparison to less hydrophobic triethylamine and methyl-piperidine. Different buffer concentrations and pH changes indicated that best separation efficiency and sensitivity was observed with 10 mM of IP agent combined with 0.02 % (v/v) FA. However, these conditions compromise column lifetime, since high pH dissolves the stationary phase of the Synergi column. The Gemini phenyl-hexyl has a pH stable column with comparable 11

interactions. The detected features eluted still early, but as shown in Figures 1D and SI-1D, there is a chromatographic separation. Peaks become wider during the separation, giving a larger distribution shown in Figure SI-2D. A normalised chromatogram of the standards and a table with details are provided in Figure SI 11 and Table SI 4. The number of detected features was 190 in a 15 min elution program and mRSD of 4.6 %. RSD distribution is shown in Figure 2D. The LOD ranged between 100 and 500 pg/µL, which was not sensitive in comparison to the other separation methods. Although it is proposed as a viable alternative method, ion pairing is not an ideal option since the ‘sticky’ IP agent contaminates the column, tubing, and MS. This can cause ion suppression of the analytes and solving this issue requires thorough and extensive cleaning procedures. Possible solutions include thorough rinsing procedures before other analyses. Since this maintenance is done off-line, the analysis of polar acids should be implied at the end of the analytical workflow, allowing the analysis of the non-polar cell extracts and of polar neutrals/bases before the use of IP agents. Another solution is an analytical set-up specifically dedicated to analytical methods using IP agents, which is highly recommendable when a high number of samples have to be analysed. Using this method, major contamination is avoided, but long rinsing procedures were still required to remove remaining IP agent in the common parts of the autosampler and the MS. Nonetheless, this method proved to be more reliable than HILIC mechanisms and is therefore still preferred. The main advantage in comparison to regular RP is the improved retention of highly polar acidic metabolites. Both separation methods (i.e. positive and negative mode) have their limits regarding certain classes of the polar metabolome, but they excel in the chromatographic separation of the metabolites compatible with the ionisation mode (e.g. basic compounds are generally better separated on the iHILIC Fusion column, hyphenated to a MS run in positive mode; the acidic metabolites can be assessed using IP with detection in negative mode). As a result, both methods for polar compounds are highly complementary and a broad range of both acids and bases can be analysed accurately.

3.3 Comparison to a generic LC-MS method The designed platform has four different analytical runs to improve the metabolomic coverage. To evaluate the gain of the investment, the crude extract has been analysed using a generic run. The crude extract is identical to the extract before the liquid-liquid extraction, no differences in metabolic composition could introduce any bias towards either the generic or the designed methods. After the molecular feature extraction, only 700 and 300 molecular features were detected in the positive and negative mode, respectively. The developed 4-run platform improved the number of molecular features detected significantly to 1800 and 600 metabolites in positive and negative ionisation mode, respectively. The major impact has been noticed on the very polar compounds, which on a generic C18 column would coelute, since they are poorly or not retained. The proposed orthogonal separation improved their detection and guaranteed a better dataset for these important metabolites. The lipid classes were also better resolved because the gradient of the new method was designed specifically for an optimal separation between and within lipid classes. Limited coelution also reduced ion suppression, which further improved the overall sensitivity and precision. The overall impact is an improved detection of compounds belonging to different classes of the metabolome. Importantly, improved separation was essential to improve the peak picking (and grouping), since the molecular feature extraction is less performant when many metabolites coelute. The improved 12

LC-MS approach is thus expected to deliver higher quality data through better data analysis. As a final result, the output of the statistical analysis will be more reliable. 3.4. Critical considerations for the optimisation of mass spectrometric parameters When optimising the MS-parameters, both sensitivity and m/z resolution were affected when altering source parameters such as the voltages and temperatures. Ideally, a balance between sensitivity and specificity should be achieved. Drying gas flow and temperature did not affect sensitivity as long as the spray was stable. The greatest impact was observed when varying the nebulizer pressure and the voltages of the fragmentor, nozzle and capillary. For the non-polar methods, a fragmentor voltage of 175 V seemed to yield the best sensitivity; capillary voltage was optimal at 3 500 V in positive ionisation mode and 3 750 V in negative ionisation mode. Lipids have rather similar structures and although these parameters were not optimal for each specific analyte, the average sensitivity was optimal and no metabolites suffered great sensitivity loss. After 20 min in negative mode, a nozzle voltage was applied to improve ionisation of late eluting compounds. Concerning the MS-parameters in the metabolomics approach, lower capillary voltages (2,000 V) gave better sensitivity in comparison to higher voltages. This is in contrast with most methods reported, which measure at higher voltages of 3,500 V to even 4,500 V [4,46,48,51]. The authors state that care should be taken for the chemical differences between metabolites and that, during optimisation, the sensitivity of essential metabolites should be checked as their detection can be crucial when assessing a metabolic profile for toxicity studies. General mass spectral resolution was better at 4 GHz in comparison to 2 GHz Extended Dynamic Range (average resolution: ± 15,000 vs 8,000), but the dynamic range of the intensities measured was severely compromised, reaching over 10 % of the detected metabolites being saturated. Reducing the concentration injected lowered the number of saturated peaks, but this approach led to a loss of information for the low-abundant metabolites. Measuring at 2 GHz extended dynamic range provided better semi-quantifying results (less than 1 % saturated), but compromised m/zresolution and possible identification. The improved dynamic range of the 2 GHz-mode can be explained to the functions of the processors. In the High Resolution (4 GHz) mode, the machine dedicates the processors to the m/z calculations, resulting in improved mass-accuracy. In the extended dynamic range (2 GHz) mode, the detection cells are divided into a low and a higher gain, improving the signal capacity. For untargeted metabolomics applications, this extended dynamic range is important since saturation decreases the linearity of the response during the analysis [52]. With saturated peaks, the differences between the two groups would be less noticeable, resulting in more false negative results. For statistical reasons, a better dynamic range was the first priority and 2 GHz in MS mode was preferred. Analytical methods for statistical analysis are best performed in MS-mode only, since MS/MS- and MSE-modes require slower scan cycles, possibly compromising chromatographic peak shape, which should have at least 10 and preferably >15 data points before being chromatographically acceptable for statistical purposes [53]. When reducing the scan time, a loss of sensitivity was observed resulting in less extracted features concerning mainly low-abundant metabolites. For early eluting peaks with a bandwidth of 6 s, less than 400 ms per scan cycles are preferred. Faster scan rates are achievable, but loss of sensitivity for the lower abundant metabolites was observed resulting in less molecular features detected. The MS scan rate was set at 4 scans/s for the non-polar fraction. For the polar fraction, a scan rate of 2 scans/s was implied to increase mass accuracy and chromatographic peak 13

shape. Since chromatographic peak shapes were broader in comparison to the non-polar fraction approach, enough data points could be guaranteed for the detection of analytes. 3.5. Application Untargeted metabolomics experiments try to monitor the metabolome of the organism in an unbiased way, searching for fingerprints which differentiate between two or more exposed groups. When this approach is used in toxicological investigations, the protocol is designed to monitor as many endogenous compounds as possible in order to efficiently search for biomarkers. These endogenous metabolites will provide information about the key events in the toxicological pathways and could be used for biomarker discovery. Cell extracts contain thousands of metabolites and are an excellent method to improve the method in order to obtain High-quality results. This study was conducted to improve the coverage of the HepaRG metabolome. The HepaRG cell line is a promising in vitro alternative for toxicological studies concerning the hepatotoxicity of novel compounds [30,31,54,55]. The cell line differentiates in hepatic and biliary like cells, mimicking the in vivo histological structures [31]. Furthermore, these cells are metabolic competent and more stable than other cell lines and are therefore considered to be a reproducible, consistent source of cells for toxicological experiments [31,56,57]. Since the liver is a complex organ with multiple functions in the human body, its metabolic interactions are more expanded than the common primary metabolism, resulting in a more complex biochemistry [58]. Because of its unique functions, the metabolome of liver cell cultures will be different from other cell types. Nonetheless, the method can be applied to non-liver culture applications since the primary metabolism is a common physiological activity for all eukaryotic cells.

4. Conclusions Untargeted LC-MS metabolomics generally use a single gradient for the entire metabolome. We have applied a fractionation of cell extracts to improve the metabolomics coverage of the HepaRG cells. Two fractions were generated by liquid-liquid extraction. For the non-polar fraction, two RP-LC lipidomic methods were developed in positive and negative mode, allowing the detection of 1,200 and 500 metabolites, respectively. For the polar fraction, a HILIC method with a mixed mode column performed best for positive mode. For the negative mode, an ion pairing approach with a phenyl hexyl column proved to give the best results regarding separation, sensitivity, and precision. The metabolomics yield was respectively 700 and 200 polar metabolites. This approach with four tailored LC-MS runs allowed the detection of chemically diverse metabolites. The intra-batch precision was excellent for the non-polar fraction (mRSD < 5 %) and good for polar fraction (mRSD < 10 %). The entire analytical platform provides a reliable analysis of over 2,000 metabolites which are present in most metabolic pathways.

5. Acknowledgments Matthias Cuykx and dr. Robim Rodrigues were funded by Fonds voor Wetenschappelijk Onderzoek (FWO) (Brussels, Belgium) and Charlie Beirnaert was funded by the Geconcerteerde OnderzoeksActie project (GOA) of the University of Antwerp (Belgium). Noelia Negreira acknowledges the University 14

of Antwerp for her postdoctoral fellowship. The authors thank Biopredic for the availability of HepaRG® cells. Further thanks to Dr. Wen Jiang, CEO from HILICON AB, for his intensive collaboration regarding the iHILIC Fusion column. The authors also thank Ronny Blust, Freddy Dardenne and Femke De Croock for their logistic assistance during the cultivation of the HepaRG cells during the first tiers of the experiments.

Bibliography [1] J.K. Nicholson, J.C. Lindon, E. Holmes, “Metabonomics”: understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data, Xenobiotica. 29 (1999) 1181–1189. doi:10.1080/004982599238047. [2] S. Naz, M. Vallejo, A. García, C. Barbas, Method validation strategies involved in non-targeted metabolomics, J. Chromatogr. A. 1353 (2014) 99–105. doi:10.1016/j.chroma.2014.04.071. [3] O. González, M.E. Blanco, G. Iriarte, L. Bartolomé, M.I. Maguregui, R.M. Alonso, Bioanalytical chromatographic method validation according to current regulations, with a special focus on the non-well defined parameters limit of quantification, robustness and matrix effect, J. Chromatogr. A. 1353 (2014) 10–27. doi:10.1016/j.chroma.2014.03.077. [4] H.G. Gika, G. Theodoridis, U. Vrhovsek, F. Mattivi, Quantitative profiling of polar primary metabolites using hydrophilic interaction ultrahigh performance liquid chromatography-tandem mass spectrometry, J. Chromatogr. A. 1259 (2012) 121–127. doi:10.1016/j.chroma.2012.02.010. [5] L.W. Sumner, A. Amberg, D. Barrett, M.H. Beale, R. Beger, C.A. Daykin, et al., Proposed minimum reporting standards for chemical analysis: Chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI), Metabolomics. 3 (2007) 211–221. doi:10.1007/s11306007-0082-2. [6] W.B. Dunn, I.D. Wilson, A.W. Nicholls, D. Broadhurst, The importance of experimental design and QC samples in large-scale and MS-driven untargeted metabolomic studies of humans., Bioanalysis. 4 (2012) 2249–64. doi:10.4155/bio.12.204. [7] H.G. Gika, C. Zisi, G. Theodoridis, I.D. Wilson, Protocol for quality control in metabolic profiling of biological fluids by U(H)PLC-MS, J. Chromatogr. B. 1008 (2016) 15–25. doi:10.1016/j.jchromb.2015.10.045. [8] J. Godzien, V. Alonso-Herranz, C. Barbas, E.G. Armitage, Controlling the quality of metabolomics data: new strategies to get the best out of the QC sample, Metabolomics. (2014) 518–528. doi:10.1007/s11306-014-0712-4. [9] T. Kind, O. Fiehn, Metabolomic database annotations via query of elemental compositions: mass 15

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18] [19]

[20]

[21]

[22]

[23]

[24]

[25]

accuracy is insufficient even at less than 1 ppm, BMC Bioinformatics. 7 (2006) 234. doi:10.1186/1471-2105-7-234. L. Burton, G. Ivosev, S. Tate, G. Impey, J. Wingate, R. Bonner, Instrumental and experimental effects in LC-MS-based metabolomics, J. Chromatogr. B Anal. Technol. Biomed. Life Sci. 871 (2008) 227–235. doi:10.1016/j.jchromb.2008.04.044. A. Gil, D. Siegel, H. Permentier, D. Reijngoud, F. Dekker, Stability of energy metabolites — an often overlooked issue in metabolomics studies : a review, Electrophoresis. 36 (2015) 2156– 2169. doi:10.1002/elps.201500031. T. Tohge, T. Mettler, S. Arrivault, A.J. Carroll, M. Stitt, A.R. Fernie, From models to crop species: caveats and solutions for translational metabolomics., Front. Plant Sci. 2 (2011) 61. doi:10.3389/fpls.2011.00061. S. Tulipani, R. Llorach, M. Urpi-Sarda, C. Andres-Lacueva, Comparative analysis of sample preparation methods to handle the complexity of the blood fluid metabolome: When less is more, Anal. Chem. 85 (2013) 341–348. doi:10.1021/ac302919t. U. Sommer, H. Herscovitz, F.K. Welty, C.E. Costello, LC-MS-based method for the qualitative and quantitative analysis of complex lipid mixtures, J. Lipid Res. 47 (2006) 804–814. doi:10.1194/jlr.M500506-JLR200. T.J. Waybright, Q.N. Van, G.M. Muschik, T.P. Conrads, T.D. Veenstra, H.J. Issaq, LC-MS in metabonomics: optimization of experimental conditions for the analysis of metabolites in human urine, J. Liq. Chromatogr. Relat. Technol. 29 (2006) 2475–2497. doi:10.1080/10826070600914638. S.U. Bajad, W. Lu, E.H. Kimball, J. Yuan, C. Peterson, J.D. Rabinowitz, Separation and quantitation of water soluble cellular metabolites by hydrophilic interaction chromatography-tandem mass spectrometry, J. Chromatogr. A. 1125 (2006) 76–88. doi:10.1016/j.chroma.2006.05.019. C. Böttcher, E. von Roepenack-Lahaye, E. Willscher, D. Scheel, S. Clemens, Evaluation of matrix effects in metabolite profiling based on capillary liquid chromatography electrospray ionization quadrupole time-of-flight mass spectrometry, Anal. Chem. 79 (2007) 1507–13. doi:10.1021/ac061037q. N.J. Lindon J Holmes E., Handbook of Metabonomics and Metabolomics, Elsevier, Amsterdam, Nederland, 2007. E. Rico, O. González, M.E. Blanco, R.M. Alonso, Evaluation of human plasma sample preparation protocols for untargeted metabolic profiles analyzed by UHPLC-ESI-TOF-MS, Anal. Bioanal. Chem. 406 (2014) 7641–7652. doi:10.1007/s00216-014-8212-y. M.H. Sarafian, M. Gaudin, M.R. Lewis, F.P. Martin, E. Holmes, J.K. Nicholson, et al., Objective set of criteria for optimization of sample preparation procedures for ultra-high throughput untargeted blood plasma lipid profiling by ultra performance liquid chromatography-mass spectrometry, Anal. Chem. 86 (2014) 5766–5774. doi:10.1021/ac500317c. F. Fei, D.M.E. Bowdish, B.E. McCarry, Comprehensive and simultaneous coverage of lipid and polar metabolites for endogenous cellular metabolomics using HILIC-TOF-MS, Anal. Bioanal. Chem. 406 (2014) 3723–3733. doi:10.1007/s00216-014-7797-5. J.-L. Wolfender, G. Marti, A. Thomas, S. Bertrand, Current approaches and challenges for the metabolite profiling of complex natural extracts, J. Chromatogr. A. 1382 (2015) 136–164. doi:10.1016/j.chroma.2014.10.091. P. Yin, G. Xu, Current state-of-the-art of nontargeted metabolomics based on liquid chromatography-mass spectrometry with special emphasis in clinical applications., J. Chromatogr. A. 1374 (2014) 1–13. doi:10.1016/j.chroma.2014.11.050. S. Liu, J.D. Brown, K.J. Stanya, E. Homan, M. Leidl, K. Inouye, et al., A diurnal serum lipid integrates hepatic lipogenesis and peripheral fatty acid use., Nature. 502 (2013) 550–554. doi:10.1038/nature12710. A. Ruiz-Aracama, A. Peijnenburg, J. Kleinjans, D. Jennen, J. van Delft, C. Hellfrisch, et al., An untargeted multi-technique metabolomics approach to studying intracellular metabolites of HepG2 cells exposed to 2,3,7,8-tetrachlorodibenzo-p-dioxin, BMC Genomics. 12 (2011) 251. 16

doi:10.1186/1471-2164-12-251. [26] D. Vuckovic, Solid-phase microextraction as sample preparation method for metabolomics, University of Waterloo (Canada), 2010. [27] D. Vuckovic, Current trends and challenges in sample preparation for global metabolomics using liquid chromatography-mass spectrometry, Anal. Bioanal. Chem. 403 (2012) 1523–1548. doi:10.1007/s00216-012-6039-y. [28] B. Bojko, N. Reyes-Garcés, V. Bessonneau, K. Goryński, F. Mousavi, E. a. Souza Silva, et al., Solidphase microextraction in metabolomics, Trends Anal. Chem. 61 (2014) 168–180. doi:10.1016/j.trac.2014.07.005. [29] Z. Leon, J.C. Garcia-Canaveras, M.T. Donato, A. Lahoz, Mammalian cell metabolomics: experimental design and sample preparation, Electrophoresis. 34 (2013) 2762–2775. doi:10.1002/elps.201200605. [30] P. Gripon, S. Rumin, S. Urban, J. Le Seyec, D. Glaise, I. Cannie, et al., Infection of a human hepatoma cell line by hepatitis B virus., Proc. Natl. Acad. Sci. U. S. A. 99 (2002) 15655–60. doi:10.1073/pnas.232137699. [31] A. Guillouzo, A. Corlu, C. Aninat, D. Glaise, F. Morel, C. Guguen-Guillouzo, The human hepatoma HepaRG cells: A highly differentiated model for studies of liver metabolism and toxicity of xenobiotics, Chem. Biol. Interact. 168 (2007) 66–73. doi:10.1016/j.cbi.2006.12.003. [32] N. Van den Eede, M. Cuykx, R. Rodrigues, K. Laukens, H. Neels, A. Covaci, et al., Metabolomics analysis of the toxicity mechanism of triphenyl phosphate in HepaRG cells and comparison to oxidative stress mechanisms by acetaminophen, Toxicol. Vitr. 29 (2015) 2045–2054. doi:10.1016/j.tiv.2015.08.012. [33] H. Wu, A.D. Southam, A. Hines, M.R. Viant, High-throughput tissue extraction protocol for NMRand MS-based metabolomics, Anal Biochem. 372 (2008) 204–212. doi:10.1016/j.ab.2007.10.002. [34] R.C. Team, R: A Language and Environment for Statistical Computing, R A Lang. Environ. Stat. Comput. (2014). http://www.r-project.org/. [35] T. Cajka, O. Fiehn, Towards merging untargeted and targeted methods in mass spectrometrybased metabolomics and lipidomics, Anal. Chem. 88 (2016) 524–45. doi:10.1021/acs.analchem.5b04491. [36] H.M. Parsons, D.R. Ekman, T.W. Collette, M.R. Viant, Spectral relative standard deviation: a practical benchmark in metabolomics, Analyst. 134 (2009) 478–485. doi:10.1039/b808986h. [37] K. Sandra, A. dos S. Pereira, G. Vanhoenacker, F. David, P. Sandra, Comprehensive blood plasma lipidomics by liquid chromatography/quadrupole time-of-flight mass spectrometry, J. Chromatogr. A. 1217 (2010) 4087–4099. doi:10.1016/j.chroma.2010.02.039. [38] A.M. Zivkovic, M.M. Wiest, U.T. Nguyen, R. Davis, S.M. Watkins, J.B. German, Effects of sample handling and storage on quantitative lipid analysis in human serum, Metabolomics. 5 (2009) 507–516. doi:10.1007/s11306-009-0174-2. [39] T. Zhang, D.J. Creek, M.P. Barrett, G. Blackburn, D.G. Watson, Evaluation of coupling Reversed Phase (RP), Aqueous Normal Phase (ANP) and Hydrophilic Interaction (HILIC) liquid chromatography with orbitrap mass spectrometry for metabolomic studies of human urine, Anal. Chem. 84 (2012) 1994–2001. doi:10.1021/ac2030738. [40] P. Jandera, Stationary and mobile phases in hydrophilic interaction chromatography: A review, Anal. Chim. Acta. 692 (2011) 1–25. doi:10.1016/j.aca.2011.02.047. [41] W. Jiang, iHILIC Fusion: General information, (2016). [42] R. Zhang, D.G. Watson, L. Wang, G.D. Westrop, G.H. Coombs, T. Zhang, Evaluation of mobile phase characteristics on three zwitterionic columns in hydrophilic interaction liquid chromatography mode for liquid chromatography-high resolution mass spectrometry based untargeted metabolite profiling of Leishmania parasites, J. Chromatogr. A. 1362 (2014) 168–179. doi:10.1016/j.chroma.2014.08.039. [43] T.M. Khin, T. Uehara, K. Aoshima, Y. Oda, Polar anionic metabolome analysis by nano-LC/MS with a metal chelating agent, Anal. Chem. 81 (2009) 7766–7772. doi:10.1021/ac901269h. [44] T. De Vijlder, J. Boschmans, E. Witters, F. Lemière, T. De Vijlder, Study on the loss of nucleoside 17

[45]

[46]

[47]

[48]

[49]

[50] [51]

[52] [53] [54]

[55]

[56]

[57]

[58]

mono-, di- and triphosphates and phosphorylated peptides to a metal-free LC-MS hardware, Int. J. Mass Spectrom. 304 (2010) 1–8. doi:10.1016/j.ijms.2010.05.022. L. Gong, J.S.O. McCullagh, Comparing ion-pairing reagents and sample dissolution solvents for ion-pairing reversed-phase liquid chromatography/electrospray ionization mass spectrometry analysis of oligonucleotides., Rapid Commun. Mass Spectrom. 28 (2014) 339–50. doi:10.1002/rcm.6773. L. Coulier, R. Bas, S. Jespersen, E. Verheij, M.J. Van Der Werf, T. Hankemeier, Simultaneous quantitative analysis of metabolites using ion-pair liquid chromatography-electrospray ionization mass spectrometry, Anal. Chem. 78 (2006) 6573–6582. doi:10.1021/ac0607616. D. McCloskey, J. a. Gangoiti, B.O. Palsson, A.M. Feist, A pH and solvent optimized reverse-phase ion-paring-LC–MS/MS method that leverages multiple scan-types for targeted absolute quantification of intracellular metabolites, Metabolomics. 11 (2015) 1338–1350. doi:10.1007/s11306-015-0790-y. J.M. Knee, T.Z. Rzezniczak, A. Barsch, K.Z. Guo, T.J.S. Merritt, A novel ion pairing LC/MS metabolomics protocol for study of a variety of biologically relevant polar metabolites, J. Chromatogr. B. 936 (2013) 63–73. doi:10.1016/j.jchromb.2013.07.027. A. Contreras-Sanz, T.S. Scott-Ward, H.S. Gill, J.C. Jacoby, R.E. Birch, J. Malone-Lee, et al., Simultaneous quantification of 12 different nucleotides and nucleosides released from renal epithelium and in human urine samples using ion-pair reversed-phase HPLC, Purinergic Signal. 8 (2012) 741–751. doi:10.1007/s11302-012-9321-8. A.R. Johnson, M.F. Vitha, Chromatographic selectivity triangles, J. Chromatogr. A. 1218 (2011) 556–586. doi:10.1016/j.chroma.2010.09.046. Y. Zhao, G. Liu, Y. Liu, L. Yuan, D. Hawthorne, J.X. Shen, et al., Improved ruggedness of an ionpairing liquid chromatography/tandem mass spectrometry assay for the quantitative analysis of the triphosphate metabolite of a nucleoside reverse transcriptase inhibitor in peripheral blood mononuclear cells, Rapid Commun. Mass Spectrom. 27 (2013) 481–8. doi:10.1002/rcm.6473. Agilent Technologies, Agilent Q-TOF LS/MS techniques and operation course number R1904A Student Manual, Agilent Technologies Inc., Alpharetta GA, USA, 2009. L. Snyder, J. Kirkland, J. Dolan, Introduction to modern Liquid Chromatography, third edit, Wiley and sons, New Jersey USA, 2010. S. Antherieu, A. Rogue, B. Fromenty, A. Guillouzo, M.A. Robin, Induction of vesicular steatosis by amiodarone and tetracycline is associated with up-regulation of lipogenic genes in HepaRG cells, Hepatology. 53 (2011) 1895–1905. doi:10.1002/hep.24290. M. Szabo, Z. Veres, Z. Baranyai, F. Jakab, K. Jemnitz, Comparison of human hepatoma HepaRG Cells with human and rat hepatocytes in uptake transport assays in order to predict a risk of drug induced hepatotoxicity, PLoS One. 8 (2013) e59432. doi:10.1371/journal.pone.0059432. M. Lubberstedt, U. Muller-Vieira, M. Mayer, K.M. Biemel, F. Knospel, D. Knobeloch, et al., HepaRG human hepatic cell line utility as a surrogate for primary human hepatocytes in drug metabolism assessment in vitro, J Pharmacol Toxicol Methods. 63 (2011) 59–68. doi:10.1016/j.vascn.2010.04.013. C. Aninat, A. Piton, D. Glaise, T. Le Charpentier, S. Langouet, F. Morel, et al., Expression of cytochromes P450, conjugating enzymes and nuclear receptors in human hepatoma HepaRG cells, Drug Metab Dispos. 34 (2006) 75–83. doi:10.1124/dmd.105.006759. P. Godoy, N.J. Hewitt, U. Albrecht, M.E. Andersen, N. Ansari, S. Bhattacharya, et al., Recent advances in 2D and 3D in vitro systems using primary hepatocytes, alternative hepatocyte sources and non-parenchymal liver cells and their use in investigating mechanisms of hepatotoxicity, cell signaling and ADME, Arch Toxicol. 87 (2013) 1315–1530. doi:10.1007/s00204-013-1078-5.

18

Figure 1: Number of features detected during the analysis of the non-polar fraction in positive (A) and negative (B) ionisation modes and of the polar fraction in positive (C) and negative (D) ionization modes.

19

Figure 2: RSD distribution of raw-extracted features of the non-polar fraction in positive (A) and negative (B) ionisation modes and of the polar fraction in positive (C) and negative (D) ionization modes.

20

A

B

C

Figure 3: EIC of aspartate: HILIC mechanisms (Kinetex HILIC (A), Luna NH2 (B)) in comparison to the Synergi column with ion pairing (C).

21

Table 1: Summary of the LC parameters for the final methods.

Run

Column

Mobile phase A

Mobile phase B

Lipid positive

Kinetex C18 150 mm x 2.1 mm, 1.7 µm

1/1 ACN/buffer

2/10/88 buffer/ACN/IPA

Lipid negative

Kinetex C18

1/1

2/10/88

150 mm x 2.1 mm, 1.7 µm

MeOH/buffer

buffer/MeOH/IPA

Polar positive

HILICON iHILIC 100 mm x 2.1 mm, 1.8 µm

buffer

Polar negative

Phenyl Hexyl 150 mm x 2.1 mm, 3 µm

75/25 buffer/MeOH

Buffer (milliQ water)

Run time Temp (min) (°C)

5 mM NH4Ac 45 0.1 % (v/v) HAc

No. of Peak width features < 12 s – med

55

1,100

36 % - 15 s

10 mM NH4Ac

36

55

400

37 % - 13 s

98/2 ACN/MeOH

10 mM NH4F 0.1 % (v/v) FA

25

30

700

99 % - 5 s

isocratic

10 mM TBA 0.02 % (v/v) FA

16

30

150

56 % - 10 s

22

Table 2: Linearity and LOD of standards representing different lipidome classes in positive and negative mode.

Name Lauryl acid-d3 STEARIC ACID LPC 17:0 PC 17:0 PA 17:0 Ceramide 17:0 Carnitine 16:0 4-pregnen-21-ol-20-dione-d8 Corticosterone-d8 Cholic acid Cholic acid d4 Misoprostol Mono-oleyl glycerol Di-oleoyl glycerol Di-oleoyl glycerol Trioleylglycerol CE 16:0

Formula C12H21D3O2 C18H36O2 C25H52NO7P C42H84NO8P C37 H73 O8P C35 H69NO3 C23 H45NO4 C21H22D8O3 C21H22D8O4 C24H40O5 C24D4H36O5 C22H38O5 C21H40O4 C39H72O5 C39H72O5 C57H104O6 C43H76O2

Positive ionisation r² (1pg - 1ng) LOD (pg) ND NA ND NA 100 0.999 10 0.920 10 0.994 10 0.999 100 0.979 100 0.998 100 0.999 1000 NA 1000 NA 10 0.999 100 NA 10 0.999 10 coelution 10 0.954 1000 NA

Negative r² (1 pg-1 ng) 0.998 0.923 NA NA NA 0.992 NA NA 0.999 0.995 0.998 0.997 0.999 0.999 0.986 NA NA

LOD (pg) 10 10 1000 1000 NA 10 ND ND 10 100 100 10 10 10 10 ND ND

NA: Not applicable (because only highest two concentrations were detected) ND: not detected

23

Table 3: Linearity and LOD of standards representing different metabolite classes in positive mode with iHILIC Fusion. Name

Formula

r² (10 pg -1 ng)

LOD (pg)

Lysine 13C6 15N2 Glucose 13C6 Tryptofaan d5 Folic acid Misoprostol Glycine

[13C]6H14[15N]2O2 [13C]6O6H12 C11D5H7N2O2 C19H19N7O6 C22H38O5 C2H5NO2

0.962 0.971 0.948 0.901 0.941 0.992

10 50 10 10 10 10

Alanine Serine Aspartate Threonine Valine Methionine Ornithine Adenine Proline leucine d3 Cystine Leucine Isoleucine Glucose phosphate Lysine Arginine Caffeïne N acetyl glucosamine

C3H7NO2 C3H7NO3 C4H7NO4 C4H9NO3 C5H11NO2 C5H11NO2S C5H12N2O2 C5H5N5 C5H9NO2 C6D3H10O2N C6H12N2O4S2 C6H13NO2 C6H13NO2 C6H13O9P C6H14N2O2 C6H14N4O2 C8H10N4O2 C8H15NO6

0.991 0.970 0.840 0.940 0.991 0.974 0.950 0.987 0.863 0.986 0.979 0.983 0.988 0.971 0.951 0.961 0.969 0.937

10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 50 10

Pyridoxal d3 Dopamine d4

C8H6D3NO3 C8H7D4NO2

0.983 0.930

10 50

Phenylalanine Tyrosine

C9H11NO2 C9H11NO3

0.995 0.975

10 10

24