Characterization of an easy-to-use method for the ...

Characterization of an easy-to-use method for the routine analysis of the central metabolism using an affordable lowresolution GC–MS system: application to Arthrospira platensis Myriam Phélippé, Rémy Coat, Camille Le Bras, Lorene Perrochaud, Eric Peyretaillade, Delphine Kucma, Abdellah Arhaliass, et al. Analytical and Bioanalytical Chemistry ISSN 1618-2642 Volume 410 Number 4 Anal Bioanal Chem (2018) 410:1341-1361 DOI 10.1007/s00216-017-0776-x

1 23

Your article is protected by copyright and all rights are held exclusively by SpringerVerlag GmbH Germany, part of Springer Nature. This e-offprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to self-archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com”.

1 23

Author's personal copy Analytical and Bioanalytical Chemistry (2018) 410:1341–1361 https://doi.org/10.1007/s00216-017-0776-x

RESEARCH PAPER

Characterization of an easy-to-use method for the routine analysis of the central metabolism using an affordable low-resolution GC–MS system: application to Arthrospira platensis Myriam Phélippé 1 & Rémy Coat 1 & Camille Le Bras 1 & Lorene Perrochaud 1 & Eric Peyretaillade 2 & Delphine Kucma 1 & Abdellah Arhaliass 1 & Gérald Thouand 1 & Guillaume Cogne 1 & Olivier Gonçalves 1 Received: 4 July 2017 / Revised: 13 November 2017 / Accepted: 21 November 2017 / Published online: 18 December 2017 # Springer-Verlag GmbH Germany, part of Springer Nature 2017

Abstract We developed an easy-to-use method for the routine analysis of the central metabolism using an affordable low-resolution GC– MS system run in SIM mode. The profiling approach was optimized for the derivatization protocol of some 60 targeted metabolites. The performance of two silylation reagents (MSTFA and BSTFA) that allowed the comprehensive derivatization of 42 key intermediary metabolites of the 60 initially targeted (organic acids, phosphate derivatives, monosaccharides and amino acids) was measured. The experimental results unequivocally showed that the MSTFA reagent met mandatory criteria including ease of handling (a very simple one-step protocol was developed), comprehensiveness of derivatization (the 42 compounds covered the extended metabolic pathways of the central carbon metabolism, with a coverage percentage ranging from 17% for the worst to 90% for the best result), optimized response coefficient of the whole derivatives (median value greater than the others by one order of magnitude) and repeatability of the protocol (RSD value below 25% for the whole procedure). When tested in real conditions (cyanobacteria polar extract), the experimental results showed that the profiling methodology was adequately repeatable (RSD = 35%) to ensure quantification results comparable with much more sensitive analytical techniques (capillary electrophoresis/mass spectrometry and liquid chromatography/triple quadrupole mass spectrometry system), while needing only about twice the quantity of biomass. Keywords Gas chromatography–mass spectrometry . Central carbon metabolism . Elementary bricks . Metabolic precursors . Intermediary metabolites . Cyanobacteria

Abbreviations BSTFA GC–MS ISD ISND

N,O-Bis(trimethylsilyl)trifluoroacetamide Gas chromatography–mass spectrometry Internal standard for the control of derivation and absolute quantification calculations Internal standard for the normalization of data

Electronic supplementary material The online version of this article (https://doi.org/10.1007/s00216-017-0776-x) contains supplementary material, which is available to authorized users. * Olivier Gonçalves [email protected] 1

Université de Nantes, GEPEA, UMR CNRS-6144, Bât.CRTT, 37 boulevard de l’Université, BP406, 44602 Saint-Nazaire Cedex, France

2

Université Clermont Auvergne, CNRS, LMGE UMR CNRS 6023, 63000 Clermont-Ferrand, France

MSTFA MTBSTFA PPP RSD SIM TCA TIC TMCS TMIS TMSO

N-Methyl-N-(trimethylsilyl)trifluoroacetamide N-(t-Butyldimethylsilyl)-N-methyltrifluoroacetamide Pentose phosphate pathway Relative standard deviation Selected ion monitoring Tricarboxylic acid cycle Total ion current Trimethylchlorosilane trimethyliodosilane Trimethylsilyloxime

Introduction Metabolic engineering is a historical approach used to increase the production of desired metabolites by cell factories [1–4]. It has been widely used for a large variety of organisms including bacteria [5], yeasts [6] and in recent years more specifically for microalgae or cyanobacteria given their vastly

Author's personal copy 1342

underexploited chemodiversity [7, 8]. This approach relies mainly on knowing the studied organism’s chemical phenotype and, when available, its central carbon metabolism. Central carbon metabolites are key entities for understanding the fate of metabolic fluxes in microorganisms at the level of the metabolic networks [9]. Depending on the metabolic pathway architecture, these entities cannot play the same roles and have therefore been classified according to their functions [10]. Precursor metabolites (first class) are located at key positions in the central carbon metabolism and are considered essential to characterize the biosynthesis routes for proteins, sugars, nucleic acids and lipids as starting points of those pathways, leading to the formation of the building blocks required by polymerization and assembly reactions. These last are also called elementary bricks (second class) and include amino acids, monosaccharides, nucleotides and fatty acids. The link between all those metabolites is made by a third class of entities named intermediate metabolites, which allow the structuration of the chemical networks. In the case of the cyanobacteria, there are 12 precursor metabolites, provided by glycolysis (5), the tricarboxylic acid cycle (3), the pentose phosphate pathway (2) and the Calvin cycle (2), plus 30 elementary bricks of interest and 20 intermediary products (Fig. 1) [11]. The comprehensive analysis of these 60 or so entities constitutes a great challenge, but one that offers a large prize, since it can enable the identification of the solicited energy pathways, or of the metabolic nodes under modulated environmental conditions for cyanobacteria metabolic engineering purposes [12–14]. Moreover, their precise quantification can produce data that when post-computed can be used for metabolic modelling [15] or, when associated with 13C labelling, for carbon flux calculations [16]. Usually, such entities are analysed using liquid chromatography in tandem with highresolution mass spectrometry (LC–HRMS) [17]. However, LC–HRMS techniques are neither affordable nor easy to use for non-specialists, and need systematic strong scientific and technical support to be exploited for routine measurements. These limitations could be bypassed by using simpler analytical systems such as gas chromatography in tandem with mass spectrometry (GC–MS) as proposed by Wittmann [18]: GC– MS is a robust, versatile technique capable of analysing a wide variety of compounds from a wide variety of matrices such as bacteria, yeasts, fungi, mammalian cells or intact tissues [19]. However, accurate profiling of the small molecules related to the central carbon metabolism requires comprehensive derivatization of these non-volatile polar entities. Numerous derivatization protocols are described in compendium guides involving straightforward or combined reactions, including silylation (for alcohols, phenols, amines, amino acids, carboxylic acids, α-keto acids, carbohydrates, etc.), alkylation (for phenols, amino acids, carboxylic acids, thiols, etc.) or acetylation (for phenols, amines, amino acids, thiols,

Phélippé M. et al.

carbohydrates, etc.) [20]. It is usually performed by substitution on the polar function of the targeted compounds. Alkylation diminishes the polarity of the compounds by replacing labile hydrogens by aliphatic groups [21]. For acylation [22], compounds with labile hydrogens are transformed into esters, thioesters or amides via the reaction of the carboxylic acid. However, the presence of a residual acid precludes their direct injection into the GC system, and so purification before injection is mandatory. In silylation reactions [23], hydrogens from acids, alcohol, thiols, amines, amides or ketones and aldehydes (enolizable) are replaced by a trimethylsilyl or tert-butyldimethylsilyl group depending on the silylating reagent. This is achieved through a nucleophilic attack of the SN2 type, with reaction yield strongly enhanced by the presence of a very good leaving group. The derivatized products generally gain volatility and thermal stability. A non-negligible advantage of silylation over acylation is that the derivatives do not require any further purification, allowing their direct injection into the GC system. Moreover, numerous possible derivatization reagents exist, such as the most commonly found trimethylchlorosilane (TMCS), trimethylsilylimidazole (TMSI), N-methyl-trimethylsilyltrifluoroacetamide (MSTFA), N,O-bis-(trimethylsilyl)trifluoroacetamide (BSTFA) or N-(tbutyldimethylsilyl)-N-methyltrifluoroacetamide (MTBSTFA). This offers a wide range of utilization, explaining why the silylation procedure ranks among the most prevalent derivatization techniques, and is useful when dealing with a very wide diversity of compounds. In the context of central carbon metabolite profiling by GC–MS, it is therefore reasonable to propose silylation as a preferred technique, since it has the potential, with the numerous available reagents, to ensure the comprehensive derivatization of a vast range of targeted chemical classes: organic acids, phosphates derivatives, monosaccharides and amino acids [24]. Each of the aforementioned silylation reagents has known advantages or disadvantages for the derivatization of specific chemical functions. The BSTFA reagent, for example, has the advantage of being less volatile than MSTFA, but performs less well for the derivatization of the dicarboxylic acid family. Derivatized carbohydrates obtained using BSTFA are quite unstable, producing numerous fragments on the MS spectra [25]. Finally, when targeting the functional groups of the secondary alcohols and amines, it is mandatory to use TMCS as catalyst to enhance derivatization yields [26]. MTBSTFA is of great interest for the derivatization of numerous chemical families including carboxyls, hydroxyls, primary and secondary amines, and thiols [27]. The derivatized compounds obtained are less sensitive to humidity, and reaction conditions are gentler than those producing TMS derivatives [28]. Moreover, when used with TBDMCS as catalyst [26], the TBDMS derivatives are 103 to 104 times more stable than the classical TMS derivatives [29], leading to fewer derivatized compounds. However, the use of this reagent

Author's personal copy Characterization of an easy-to-use method for the routine analysis of the central metabolism using an...

1343

Fig. 1 Simplified extended map of cyanobacterial central carbon metabolism indicating the ca. 60 substances classified as metabolic precursors, buildings blocks and metabolic intermediates (adapted from [11]. Abbreviations are identified in Table 1)

implies taking into account the steric hindrance of the targeted compound: the analytical response of metabolites with weak steric hindrance will be enhanced, and conversely severe loss of sensitivity will be seen for compounds with higher steric hindrance such as sterols or carbohydrates, thus limiting its use when a general derivatization reagent is sought [30]. The latest silylation reagent to be considered, MSTFA, appears to be of great promise, having been reported to be able to derivatize large chemical families [31]. Moreover, its polarity allows it to be used with no additional solvent, thereby simplifying the experimental protocol. Moreover, among all the trimethylsilyl ethers obtained, TMS derivatives from MSTFA are the most volatile (along with their by-product N-methyltrifluoroacetamide), explaining why it is so often recommended for very polar derivatization compounds such as amine groups or amino acids [25, 32]. Finally, if one compares the reactivities of silylating reagents, their selectivity towards the targeted compounds, the stability of the derivative and the abundance/nature of reaction by-products, the MSTFA and the BSTFA reagents (with or without catalysts

such as TMCS or TMSI) appear to have the greatest potential to best achieve the comprehensive derivatization of the central carbon metabolites as partially described by Orata [33]. However, until now no experimental studies have shown quantitatively whether these reagents could be conveniently used for that purpose. Any such study would have to embrace the derivatization of a large chemical family comprising more than 60 different compounds (Table 1) belonging to organic acids and phosphate derivatives—sugar and organic acids— for the precursors, monosaccharides and amino acids for the elementary bricks, and organic acids and phosphate derivatives for the intermediate metabolites [11]. In the present work the development of an affordable and easy-to-use GC–MS method for the routine characterization of the central metabolism is proposed. The use of a lowresolution MS system for this purpose is very challenging, and has never been attempted before to encompass such broad chemodiversity. There was a need to devise an original strategy to find the best analytical compromise with biological utility. The strategy was developed in an integrated manner,

Author's personal copy 1344 Table 1 List of the ca. 60 compounds representative of the central metabolism of cyanobacteria


Compounds

ID

CAS

Phosphate derivatives Precursor metabolite

D-Glucose-6-phosphate

Precursor metabolite Precursor metabolite Precursor metabolite

D-Fructose-6-phosphate

G6P F6P

54010-71-8 26177-86-6

Glyceraldehyde-3-phosphate Dihydroxyacetone phosphate

GAP DHAP

591-59-3 102783-56-2

D-(−)-3-Phosphoglyceric

D-Ribose-5-phosphate

3PG E4P R5P

80731-10-8 103302-15-4 207671-46-3

Phosphoenol-pyruvate α-D-Glucose-1-phosphate D-fructose-1,6-bisphosphate D-(−)-2-Phosphoglyceric acid D-(−)-6-Phosphogluconic acid D-Gluconic acid, D-lactone, 6-(dihydrogenphosphate) D-Sedoheptulose-1,7-bisphosphate D-Sedoheptulose-7-phosphate D-Xylulose-5-phosphate D-Ribulose-5-phosphate D-Ribulose-1,5-bisphosphate

PEP G1P 1,6-FBP 2PG 6PGC 6PGL SBP S7P Xu5P Ru5P RuBP

Precursor metabolite Precursor metabolite Precursor metabolite Precursor metabolite Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product Organic acids Precursor metabolite

acid

D-Erythrose-4-phosphate

Pyruvic acid

Pyr

4265-07-0 56401-20-8 38099-82-0 70195-25-4 53411-70-4 2641-81-8 815-91-8 2646-35-7 138482-70-9 93-87-8 14689-84-0 1.1.1.1.1.1 113-24-6

Precursor metabolite Precursor metabolite

Acetyl coenzyme A Oxaloacetic acid

AcCoA OAA

102029-73-2 328-42-7

Precursor metabolite Intermediate product Intermediate product Intermediate product Intermediate product Intermediate product

α-Ketoglutaric acid Citric acid DL-Isocitric acid L-(−)-Malic acid Fumaric acid Succinic acid D-Galacturonic acid D-Glucuronic acid Glyoxylic acid L-(+)-Lactic acid DL-3-Hydroxybutyric acid Indole-3-acetic acid

αKG Cit ICit Mal Fum Suc Galu Glca Glx Lac 3HB IAA

L-Glutamic

Glu Gln Ala Arg Asp

328-50-7 77-92-9 1637-73-6 97-67-6 623-91-6 150-90-3 685-73-4 207300-70-7 563-96-2 79-33-4 150-83-4 87-51-4 1.1.1.1.1.2 142-47-2 56-85-9 56-41-7 74-79-3 56-84-8

Cys Gly His Hse Ile Leu Lys Met

52-90-4 56-40-6 71-00-1 672-15-1 73-32-5 61-90-5 56-87-1 63-68-3

Intermediate product Intermediate product Intermediate product Other product Other product Other product Amino acids Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks

acid

L-Glutamine L-Alanine L-Arginine L-Aspartic

Elementary bricks

L-Cysteine

Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks

L-Glycine

acid

L-Histidine L-Homoserine L-Isoleucine L-Leucine L-Lysine L-Methionine

Author's personal copy Characterization of an easy-to-use method for the routine analysis of the central metabolism using an... Table 1 (continued)

Compounds Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Monosaccharides Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Elementary bricks Intermediate product

1345

ID

CAS

Phe Pro Ser Thr Trp Tyr Val Pga

63-91-2 147-85-3 56-45-1 72-19-5 73-22-3 60-18-4 72-18-4 98-79-3

L-Fucose

Glc Ara Xyl Man Gal Fru Fuc

1.1.1.1.1.3 50-99-7 10323-20-3 58-86-6 3458-38-4 59-23-4 57-48-7 2438-80-4

L-Rahmnose myo-Inositol

Rham Myo

10030-85-0 87-89-8

L-Phenylalanine L-Proline L-Serine L-Threonine L-Tryptophan L-Tyrosine L-Valine L-Pyroglutamic

acid (pidolic acid)

D-Glucose D-Arabinose D-Xylose D-Mannose D-Galactose D-Fructose

and considers ease of handling, the comprehensiveness of the analysable information, expressed as targeted metabolic pathway coverage, and the sensitivity and repeatability of the protocol. The results obtained for the optimal choice of the derivatization reagents (MSFTA and BSTFA) are first detailed here, since this is the main bottleneck to pass when seeking to encompass such chemodiversity with a low-resolution GC– MS system. This last point is emphasized because it greatly influences the quality of the data, as it determines in fine the representativeness of the chemical picture, in terms of both quality and absolute abundance. The results of the optimization of the method using the best silylating reagent are then presented, focusing more narrowly on the improvement of the extraction process on a real matrix (the cyanobacterium Arthrospira platensis). It presents a real picture of the metabolite recovery for the method developed. Finally, in the last part, the fully optimized method is applied on a real biological matrix to verify its robustness. The performance of the method is compared with the quantitative results of early published data on Arthrospira platensis, but using much more resolutive and sensitive analytical technologies.

Materials and methods

Bois, France). All the compounds used in this study are listed in Table 1. Derivatization grade N-methylN-(trimethylsilyl)trifluoroacetamide (MSTFA), N,Ob i s ( t r i m e t h y l s i l y l ) t r i f l u o r o a c e t a m i d e ( B S T FA ) , trimethyliodosilane (TMIS), trimethylchlorosilane (TMCS) and methoxyamine hydrochloride were purchased from Sigma-Aldrich (Saint-Quentin-Fallavier, France).

Preparation of standard stock and working solutions Standard stock solutions of phosphate derivatives, organic acids, amino acids and monosaccharides were prepared in ultrapure water at concentrations ranging from 1 to 5 g L−1. Standard working solutions were prepared at concentrations ranging from 30 mg L−1 to 2 g L−1 from stock solutions by consecutive dilution in ultrapure water when needed. A solution of ribitol in water at 1 g L−1 and a solution of squalane in pyridine at 4 g L−1 were used as internal standard for the derivation plus the absolute quantification calculations (ISD), and as internal standard for normalization of data (ISND) respectively. The preparation of the solutions and the derivatization methods described hereafter were always performed in a fume hood and under constant nitrogen stream to limit effects due to moisture, and using sealed vial caps.

Standards and reagents

Preparation of silylation solutions

GC grade solvents and standards were purchased from Sigma-Aldrich (l’Isle-d’Abeau, France), Acros Organics (Noisy-le-Grand, France) or Prolabo (Fontenay-sous-

Different silylation solutions were prepared to test the effect of the silylating reagents of high donor strength in the presence or absence of catalysts. MSTFA or BSTFA reagents were


used, which also acted as solvents. They were also prepared with added catalysts, giving the following derivatization solutions: MSFTA + TMIS (1000:1, v/v), BSTFA + TMIS (1000:1, v/v), MSFTA + TMCS (1000:1, v/v) and BSTFA + TMCS (1000:1, v/v).

Derivatization methods Preparation of standards for derivatization Working standard solutions were used for the comparison of the derivatization protocols. Depending on the targeted concentration, varying proportions of working standard solution and ultrapure water volumes (for a constant volume of water + standard working solution of 900 μL) were systematically combined with 100 μL of the ISD stock solution for a final volume of 1000 μL. The prepared solutions containing the targeted compounds and the ISD then underwent the derivatization procedures described below. Derivatization protocols The following derivatization procedures were studied to screen the method best adapted to the targeted compound classes. Reaction parameters, e.g. solvent, volume ratios, incubation time and temperature, were not optimized for that purpose. Instead, general manufacturers’ recommendations were applied. However, once the best derivatization reagent was selected, reactions parameters were optimized for time (15 min, 30 min, 60 min and 120 min), temperature (30 °C, 60 °C and 90 °C) and reagent (R) to standard (S) ratio (100:1, 200:1, 300:1, excess = 500:1). For the silylation procedure, 100 μL of standard solution prepared above was dried under nitrogen at 40 °C (Nevap 111 OA-SYS Nitrogen Evaporator, Organomation, France). A 190-μL aliquot of silylation reagent solution was then added, agitated and heated at 60 °C for 30 min (Block heater SBH130D, Stuart Scientific, France) to form derivatives. A constant volume of 10 μL of the ISND stock solution was then added before the direct GC–MS analysis. For the sequential methoximation combined with the silylation procedure and leading to the formation of trimethylsilyl oximes (TMSO), 100 μL of standard solution was dried under nitrogen at 40 °C (N-evap 111 OASYS Nitrogen Evaporator, Organomation, France). First, 95 μL of neutral methoxyamine hydrochloride solution at 4 mg mL−1 in pyridine was added, agitated and heated at 60 °C for 30 min. Second, 95 μL of silylation reagent solution (MSTFA) was directly added to the crude derivatized compounds, agitated and heated at 60 °C for 30 min to form derivatives. A constant volume of 10 μL


of the ISND stock solution was then added before the direct GC–MS analysis. The effect of the pH was also estimated as previously addressed in the work of Nguyen et al. [34] for the derivatization of keto acids. For this purpose standard working solutions were prepared at 30 mg L−1 to 2 g L−1 from stock solutions by consecutive dilution in ultrapure water with pH adjusted (HCl) to ca. less than 2 and 6.5. Derivatization protocols were then performed as described above.

Chromatography–mass spectrometry analysis The derivatized samples were analysed using a Trace GC Ultra (Thermo Fisher Scientific, Boston, MA, USA) gas chromatograph interfaced to a Trace ISQ single quadrupole mass spectrometer (Thermo Fisher Scientific, Boston, MA, USA). The optimized separation conditions for measuring the substances of interest were set up as follows: 1 μL of the derivatized sample was injected onto an HP-5MS column (apolar phase 5% phenyl methylpolysiloxane, internal diameter 30 m × 250 μm, film thickness 0.25 μm; Agilent Technologies, Santa Clara, CA, USA) using programmed temperature vaporization (PTV) injection set at constant temperature (CT) mode at 260 °C, with a split mode of 1/25 (25 mL min−1). The flow rate of the carrier gas (hydrogen produced by a generator; WM-H2, F-DGSi, Evry, France) during the analysis was set to 1.2 mL min−1. The oven temperature gradient for the separation of the derivatized compounds was 50 °C for 5 min, ramped at 5 °C min−1 to 280 °C for 10 min, and finally at 100 °C min−1 to 340 °C for 4 min for cleaning. The temperature of the transfer line to the MS was 250 °C, and the ion source was set at 300 °C. Electron ionization was conducted at 70 eV. For the exploratory qualitative measurements (used for the development phase of the present study), the MS was used in full scan (FS) mode throughout the mass range of 50– 800 amu at a rate of 0.2 scan s−1. For the SIM mode, a specific timetable was used to target the primary metabolic intermediates (see Electronic Supplementary Material (ESM) Table S1). The mass resolution was 1 mass unit over the mass range of 50–800 amu. The total GC–MS analysis time was ca. 65 min. Data were postprocessed and analysed using the Xcalibur 2.1 software (Thermo Fisher Scientific, Waltham, MA, USA ThermoFisher). The identity and purity of the eluted compounds were confirmed using the AMDIS v2.1 software (Automated Mass Spectral Deconvolution and Identification System, National Institute of Standards and Technology, Gaithersburg, MD, USA), the NIST 08 mass spectral library and by spiking experiments using the pure standards listed in Table 1.


Definitions and calculations

1347

Trial of the method on a cyanobacterium: biomass cultivation, harvesting and metabolite extraction

Definitions Arthrospira platensis biomass production One of the main difficulties in the present work was finding a compromise for the comprehensive derivatization of the greatest number of targeted compounds (Table 1). It required choosing those criteria that would best allow the study aim to be achieved. The following criteria were specified. The response coefficient was defined as the ratio of the chromatographic peak areas integrated from the mass spectrometric total ion current (TIC) chromatogram or selected ion recordings to the quantity of substance injected into the GC–MS system expressed in nanograms. The amount of injected substance was determined by initial weighing, the dilution factor and the final volume of the injected derivative. The metabolic coverage was defined here as the ratio of the number of substances detected to the number targeted for a given metabolic pathway. The complexity of the TIC was defined as the ratio of the number of derivatized forms to the number of detected metabolites. The repeatability (relative standard deviation or RSD) was calculated as the ratio of the standard deviation of a peak area to the mean of the integrated area values for at least five replicates. It was expressed as a percentage.

Calculations and graph plotting The chromatograms were compared after normalization of each chromatogram of interest to the ISND peak abundance. The quantification calculations were performed using the ISD for the internal standard calibration approach. The absolute concentrations of the primary metabolic intermediates were determined from their respective peak abundance relative to the peak abundance of the ISD, and expressed in micromoles of compound per catalytic biomass dry weight as in [13]. For the purpose of developing an affordable and easy-to-use profiling methodology we deliberately excluded quantification methods based on labelling approaches, since this would dramatically increase the price per analysis (one commercial labelled compound would be necessary for each targeted compound, ca. 60 compounds) or making the protocol more difficult to handle in the case of a home-made mixture of labelled compounds, for example (IDMS approach as already described in [13] for autotrophic organisms). Statistical analysis, data fitting and graph plotting were performed using R 3.2.1 and SigmaPlot 12.5 software. A metabolic map was drawn using the Omix 1.9.9 software [35].

The cyanobacterium Arthrospira platensis strain 8005 (Institut Pasteur, Paris, France) was axenically grown in a modified Zarrouk medium as described in [11] using a torus photobioreactor with 1.5 L of working volume developed in [13] to allow homogeneous sampling of the biomass. Continuous cultures were run with an incident photon flux density set at 100 μmol m−2 s−1. Standard culture conditions were pH 9.5 ± 0.1, temperature 36 ± 1 °C. The dilution rate was 0.0312 h−1, and air was supplied continuously at a flow rate of 12 L h−1.

Biomass harvesting and metabolism quenching Once the culture reached 1 g L−1, at least five independent cultivation volumes of 20 mL (i.e. 20 mg of biomass) were sampled on polyamide filters (Sartorius, Goettingen, Germany, 25007–47N, 0.2 mm) using the automatic sampling device of the torus photobioreactor [13]. The biomass was then washed with 3 volume equivalents of physiological water directly on the filters. Once the biomass was washed (in less than 10 s), the filters were immediately immersed in liquid nitrogen (−196 °C) to stop enzymatic reactions and preserve cell integrity.

Biomass metabolite extraction The filters plus the biomass were then submitted to various solvent extraction systems using 5 volume equivalents of the organic solvent for 1 h stirring. For the organic solvent method, pure methanol (M) and (10:3:1) methanol–chloroform–water mixture (MCW) were used; for the hot extraction system, i.e. boiling water (BW) or (25:75) water/ethanol or boiling ethanol (BE) mixtures were used for various contact times of 15, 30 and 60 s. Once extraction was performed the mixture was chilled in methanol/water (3:1) at −80 °C. The extract was then separated from the biomass using centrifugation (3600g, 15 min, 4 °C, Micro R22, Hettich) according to the protocol of Courant et al. [12]. The extraction procedure was repeated four times for the various mixtures. The extracts were then independently dried under nitrogen at 40 °C (N-evap 111 OA-SYS Nitrogen Evaporator, Organomation, France), submitted to a derivatization procedure and analysed with the GC–MS in FS and SIM mode.



Results and discussion

MSTFA + TMIS or TMCS BSTFA+TMIS or TMCS BSTFA MSTFA

1.3 1.3

Selection of best derivatization reagent The first part of this study focused on screening for the best one-step silylation reagents. It consisted in comparing the results obtained for the profiling of the ca. 60 standards chosen to represent the targeted compounds of interest, namely organic acids and phosphate derivatives—sugar and organic acids—for the precursors, monosaccharides and amino acids for the elementary bricks, and organic acids and phosphate derivatives for the intermediate metabolites (Table 1). For that purpose, an integrative selection procedure was developed consisting in seeking the best compromise between the comprehensiveness of the analysable information expressed as targeted metabolic pathway coverage and mandatory analytical parameters i.e. sensitivity, repeatability of the protocol and finally ease of handling. Metabolism coverage was first assessed to estimate potential biological reliability of information gathered with tested derivatization reagents The detectability of the targeted compounds representative of the cyanobacterial central metabolism was estimated as a priority, since it would guide the whole comparison procedure: the more compounds detected, the more information is collected on the metabolic pathways of interest and their potential regulatory nodes, directly with precursors or indirectly with the building blocks and the intermediates. This biological qualitative criterion was taken into consideration as a mandatory analytical requirement: here this meant the number of obtained derivatives to avoid selecting a protocol that caused the formation of too many side products (too many peaks on the chromatograms), and therefore added too much complexity to the recorded signal. Figure 2 summarizes the data obtained for the initial screening conditions of the retained silylating reagents of high donor strength, i.e. MSTFA and BSTFA. The MSTFA protocol with no catalyst increased the total number of detectable compounds with the fewest silylated products (i.e. minimum chromatogram complexity) compared with the protocol using BSTFA as silylating reagent alone or with the combined use of TMIS as catalyst. Identical results were obtained using a better leaving group catalyst such as TMCS, suggesting that the MSFTA should be used alone, since it was reactive enough towards the targeted substances. This was moreover supported by the degraded MSTFA derivatization performance in the presence of TMIS or TMCS catalysts, which may act as competing reagents. The results summarized in Figs. 3 and 4 indicate that the targeted metabolic pathways were not evenly covered, with a global prevalence observed for the biosynthesis pathways (especially carbohydrate and amino acid

Chromatogram.complexity 1.2 1.2

47 50 Total.number.of.resolved.peaks 45 51

36 38 Number.of.detected.compounds 38 42

0

10

20

30

40

50

Fig. 2 Metabolite coverage and chromatogram complexity. The numbers of analysed compounds and resolved peaks are presented respectively in the second and third group of bars in the chart. The chromatogram complexity (first group of bars in the chart) corresponds to the ratio of the total number of resolved peaks to the number of detected compounds. The derivatization methods tested are indicated in the top legend

synthesis) and the central metabolic pathways (glycolysis, TCA cycle, Calvin cycle and pentose phosphate pathway). On close examination, the MSTFA derivatization protocol with no catalyst seemed to offer the best metabolic coverage results with key substances not detectable with the other conditions tested, therefore offering the best potential in terms of metabolic snapshot quality. For glycolysis it included notably glyceraldehyde-3-phosphate (situated at the junction of the glycolysis and the PPP reductive branch), pyruvic acid (end of the glycolysis and entrance point of the TCA cycle) and αketoglutaric acid, one of the two precursors of the amino acid biosynthesis of the TCA cycle. The MSTFA derivatization also gave access specifically, under the tested conditions, to crucial intermediates such as glucose-1-phosphate, which is solicited for both reserve and exopolysaccharide sugar biosynthesis, or elementary bricks such as xylose, mannose, glutamic acid, cysteine, lysine, tryptophan for carbohydrates or amino acid biosynthesis. Although the metabolic coverage did not cover all 12 crucial precursors or 21 intermediate products, the position of the analysable compounds still suggests good metabolic comprehension using such observables, mainly because of their key position in the simplified carbon central metabolic pathways or of the other analysable compounds located downstream of these key positions (see ESM Fig. S1). The observed prevalence is of prime importance, since future metabolomics interpretation based on this protocol should always allow for a bias in the representativeness of the targeted metabolites.

Author's personal copy Characterization of an easy-to-use method for the routine analysis of the central metabolism using an... Chemical classes

1349

Metabolic Pathways BSTFA BSTFA+TMIS or TMCS MSTFA + TMIS or TMCS MSTFA

25 0

Pentose.phosphate

25 25 29

Phosphate derivatives

29

Calvin.cycle

0 29 72 72

Glycolysis

43 87 64

Organic acids

75

TCA.cycle

75 75 33

Amino acids

55

Amino.acids

55 67 50

Monosaccharides

80

Carbohydrates

80 89

0

20

40

60

80

% of identified molecules vs the number of known compounds in each metabolic pathways

Fig. 3 Metabolic pathway and chemical class coverage. The coverage of the targeted metabolic pathways is expressed as the percentage of identified substances out of the number of known compounds in each

metabolic pathway. The compounds are grouped according to the pathways and to the chemical classes of interest. The derivatization methods tested are indicated in the top legend

From a chemical point of view, the MSTFA protocol seemed to offer preferred derivatization of the chemical classes of the monosaccharides, amino acids, organic acids and phosphate derivatives (Fig. 3). BSTFA and MSTFA were selected, first for their difference in reactivity toward hindered functional groups when used with a catalyst, and second for their difference in volatility. Steric hindrance seemed to play a modest role here in the derivatization performance of the tested metabolites, no significant increase in the total number of detectable metabolites being observed for BSTFA used with a catalyst. Figure 3 shows, however, that it significantly increased the number of metabolites for the specific classes of monosaccharides, amino acids, organic acids and phosphate derivatives. But BSTFA was still less efficient than MSTFA. For the latter, in the presence of a catalyst a negative effect could even be observed irrespective of the chemical class considered. This behaviour could not be associated with the experimental design, standard solution preparation biases or interactions between substances in the various chemical classes, since similar prevalence had already been observed on microalga cellular extract. The metabolic profiling results of the snow microalga Chlamydomonas nivalis performed by Lu et al. [36] using GC/TOF-MS with a more complicated two-

step derivatization protocol (combination of methoximation plus silylation using MSTFA with no catalyst) indicated the same chemical class representativeness. The order observed in both cases, which could be hypothesized to be matrix-independent, could finally be mostly explained by the reactivity of the targeted chemical group toward silylation: carboxyl groups found in amino acids and organic acids and other active hydrogen groups found in phosphate derivatives were described to react in that observed order by Orata [33]. For the specific case of the carbohydrates, the measured qualitative prevalence could not be simply explained with the above classical description of reactivity: hydroxyl groups found in monosaccharides are known to be less reactive than the other groups and to have their own specificity [37]. Steric and electronic effects are known to drive the relative reactivity of the hydroxyl groups, which may compete favourably or unfavourably in the efficient formation of TMS derivatives [38]. The fact that almost all the monosaccharides tested reacted positively may merely illustrate that these reactive groups are all evenly accessible to MSTFA or BSTFA. Concerning the volatility of the two reagents and their byproducts, known to interfere only slightly with early eluting peaks, no clear conclusion could be proposed to differentiate

Author's personal copy 1350 Fig. 4 Metabolite precursor, intermediate and elementary brick coverage. The coverage of the key metabolites is expressed as the percentage of identified substances out of the number of known compounds in each category. The derivatization methods tested are indicated in the top legend

Phélippé M. et al. MSTFA BSTFA BSTFA + TMIS or TMCS MSTFA + TMIS or TMCS

66 66 Other.products

66 100

44 44 Elementary.bricks

38 67

31 31 Intermediate.products

31 40

41 41 Metabolite.precursors

41 67

0

20

40

60

80

100

% of identified molecules vs the number of known compounds in each category

the efficiencies of the two reagents: the chromatogram complexity was systematically low when using BSTFA or MSTFA alone, although MSTFA was more volatile than BSTFA. However, the reactivity of MSTFA seemed to make it more efficient than BSTFA in the derivatization conditions tested, significantly increasing the number of detected compounds in the monosaccharide, amino acid and organic acid classes. These results are consistent with previous reports showing that MSTFA was superior to BSTFA in reactions with amines and amino acids [32, 39, 40]. Moreover, when compared with the available data, the original results obtained in the present study point to an extended domain of usage for MSTFA for the simultaneous derivatization of a complex mixture of compounds including amino acids, monosaccharides, organic acids and some phosphate derivatives. Response coefficient of derivatized metabolites was also taken into account to optimize chances of developing a sensitive method The response coefficient of the targeted metabolites was considered to assess the quantitative efficiency of the tested derivatization procedures towards the targeted chemical classes. It should allow an optimal detectable quantity of matter to be obtained for each tested compound, and in fine optimal

sensitivity. The response coefficient of each compound tested was expressed as the ratio of the area of the corresponding peak to the quantity of the injected mass (ESM Fig. S2). Table 2 summarizes the position parameters (median and mean) of the response coefficient distribution observed for all the detectable metabolites according to the chosen procedure (ESM Fig. S3). The response coefficient distributions were statistically different according to Kruskal–Wallis test population comparison results. At the 0.05 significance level the null hypothesis of identical population was rejected with a p value found close to zero (9 × 10−13). The MSTFA with no catalyst protocol seemed here to offer the best global response coefficients, greater by one order of magnitude than the others (median values in Table 2), consistent with the qualitative results in the previous section. Response coefficients were found not to be evenly distributed according to targeted chemical class (Fig. 5): a different rank order could be observed for the response coefficients obtained with MSTFA and BSTFA alone or with catalysts. Without a catalyst MSTFA still presented response coefficients greater than that for BSTFA, but following a common order: organic acids > amino acids > phosphate derivatives or monosaccharides. With a catalyst, observations were slightly different. MSTFA presented response coefficients similar to that for BSTFA, and followed a common order, but with switched monosaccharides and amino acids: organic acids >

Author's personal copy Characterization of an easy-to-use method for the routine analysis of the central metabolism using an... Table 2 Median and mean response coefficients

1351

Response coefficient (ng-1)

BSTFA

BSTFA + TMIS or TMCS

MSTFA

MSTFA + TMIS or TMCS

Median

3.1E+06

7.0E+06

4.0E+07

7.0E+06

Mean

2.1E+07

7.9E+07

1.1E+08

7.8E+07

The mean and median of the response coefficients distributions were extracted from Fig. S3 and ordered according to the derivatization protocols. The mean or median response coefficients are expressed per nanogram of analysed compound

monosaccharides > amino acids > phosphates derivatives. These differences were statistically significant according to Kruskal–Wallis test population comparison results (at the 0.05 significance level, p ≤ 2.1 × 10−14). The measured response coefficients for MSTFA and BSTFA perfectly fitted what was described by Orata [33] in terms of reactivity: carboxyl groups should react more efficiently than amine groups and other hydroxyls or active hydrogen groups. The use of a catalyst, however, demonstrated a utility for the derivatization of the monosaccharides: the reactivity of the derivatization reaction was significantly enhanced, irrespective of the derivatization reagent used.

Repeatability of derivatization protocols was included in screening strategy to select a robust protocol The last criterion used to estimate the efficiency (in term of robustness) of the derivatization protocols was repeatability. To ensure the development of an optimized quantification protocol with a controlled error, it was important to estimate the sample preparation variability. RSD was calculated for all the detectable metabolites in all the conditions tested (ESM Fig. S4). The RSD distribution values are summarized in Fig. S5 (see ESM) and the extracted median or mean values in Table 3. Using the MSTFA as sole reagent was unequivocally

2e+08

1e+08

1e+08

0e+00

0e+00 Phosphate derivatives

Monosaccharides

MSTFA + TMIS

4e+08

3e+08

3e+08

2e+08

2e+08

1e+08

1e+08

0e+00

0e+00

Fig. 5 Distribution of the response coefficients grouped per chemical class (i.e. phosphate derivatives, organic acids including uronic acids, amino acids and monosaccharides) as a function of the derivatization protocol tested. The response coefficients are expressed per nanogram

Amino acids

5e+08

4e+08


6e+08

5e+08

Organic acids

6e+08

Monosaccharides

7e+08

Amino acids

7e+08

Monosaccharides

BSTFA + TMIS


3e+08

2e+08


4e+08

3e+08

Organic acids

5e+08

4e+08

Amino acids

6e+08

5e+08

Organic acids

7e+08

6e+08

Amino acids

7e+08

Organic acids

MSTFA

Monosaccharides

BSTFA

of analysed compound, and the calculations were performed for five replicates per compound. Median values are shown as blue bars and mean values as red crosses

Author's personal copy 1352 Table 3 Median and mean RSD values of the derivatization protocols


Coefficient of variation (RSD in %) mean median

BSTFA

83 67

BSTFA + TMIS or TMCS 84 82

MSTFA

MSTFA + TMIS or TMCS

19 18

36 23

The mean and median of the RSD distributions were extracted from Fig. S5 and ordered according to the derivatization protocols. The mean or median RSD are expressed in %

the most repeatable protocol for the targeted metabolites: compared with the BSTFA protocol with added catalyst (TMIS or TMCS) or with no catalyst, it was ca. four times more repeatable. The use of a catalyst even seemed to cause a statistically significant degradation of the protocol repeatability (Kruskal– Wallis test at the 0.05 significance level, p ≤ 1.4 × 10−11), irrespective of the derivatization reagent considered. On analysing repeatability according to targeted chemical class (Fig. 6), the RSD values calculated for BSTFA were found to be systematically higher than those calculated for MSTFA, and were even greater with the use of catalysts. Concerning the MSTFA RSD values, it is interesting to note that catalyst use degraded the repeatability performance of the amino acid and phosphate derivative classes by at least a factor of 2. Similar behaviour was described in the literature for acidic sugar derivatization, but for BSTFA with TMCS [41]. TMCS was suspected to form ammonium chloride, which interfered with the sample drying, so modifying the derivatization yields. A similar hypothesis generalized to the TMIS or the TMCS could be advanced for the substances in the present study, lowering the reactivity of the targeted groups towards silylation, and therefore strongly impacting the repeatability of the protocol. This hypothesis could be also supported by van Look [42], who described the formation of dehydrated side products when using TMIS as a reagent, thereby diminishing the derivatization yields.

Derivatization reagent was chosen to obtain the best compromise between biological and analytical requirements Finally, the choice of the best protocol relied mainly on finding a compromise between all the tested parameters, i.e. metabolite coverage, response coefficients of the derivatized metabolites and repeatability of the derivatization protocols. Using the MSTFA as sole derivatization reagent, i.e. with no catalyst, gave the best number of detectable metabolites—42 out of the ca. 60 targeted—and by extension the best metabolic pathway coverage ranging from 17% to 90%, the best median response coefficients with almost one log more than the other tested protocols, and the best median repeatability with RSD below 20%, four times better than the worst measured result. It was also relatively simple to use, therefore meeting the ease-of-handling qualitative criterion.

MSFTA derivatization protocol to quantify targeted metabolites Optimization of derivatization procedure Before envisaging using the MSTFA derivatization procedure for the quantification of the targeted metabolites, it was necessary to optimize the derivatization reaction parameters for temperature, time, and ratio of reagent to metabolites, and to estimate the recovery of the targeted compounds. Tests were first performed on compounds representing the targeted chemical classes: glucose for the monosaccharides, lysine for the amino acids, G6P for the phosphate derivatives and citric acid for the organic acids. The ratio of reagent to compounds was tested for the following values: 100:1, 200:1, 300:1 and in excess (500:1, reagent used as solvent). For each ratio, temperature of derivatization was tested for 30, 60 and 90 °C, with reaction times of respectively 15, 30, 60 and 120 min for each temperature. Results indicated that the tested ratio gave no significant difference (measured by the Kruskal–Wallis test at the 0.05 significance level (p > 0.05)) for the tested times and temperatures. For the tested temperatures, maximum recovery (calculated on the basis of the cumulated signal measured for the tested compounds) was obtained for 30 and 60 min of derivatization reaction (> 90%). Optimal signals were obtained for 60 °C and started to decrease significantly at 90 °C (data not shown). Finally, the best compromise consisted in performing the derivatization reaction using MSTFA as solvent to avoid stoichiometric problems, at 60 °C for 30 min to ensure the best compromise between thermostability and compound recovery. Finally, the chosen parameters conformed to the manufacturer’s recommendations. Trimethylsilyl oximes (TMSO) have been widely used, especially for the analysis of keto acids and ketose and aldose monosaccharides, since their derivatives present good GC properties associated with simple chromatograms [34, 43, 44]. In a complex mixture like the one in the present study, it was of interest to estimate the effect of such a protocol for optimization purposes, even if a two-stage derivatization protocol could influence the final repeatability. The acidification effect was also considered to estimate whether acidic or neutral pH conditions could influence the stability of the targeted compounds. The results indicate that very little improvement could be detected with the tested conditions. The metabolic


1353

50

0

0 Amino acids 50

0

0 Phosphate derivatives

50

Amino acids

100


100

Organic acids

150

Monosaccharides

150

Amino acids

200

Organic acids

CV MSTFA + TMIS

200

Monosaccharides

CV BSTFA + TMIS


50


100

Organic acids

100

Monosaccharides

150

Amino acids

150

Organic acids

CV MSTFA 200

Monosaccharides

CV BSTFA 200

Fig. 6 Distribution of the RSD grouped per chemical class (i.e. phosphate derivatives, organic acids including uronic acids, amino acids and monosaccharides) as a function of the derivatization protocol tested.

The RSDs are expressed in %, and the calculations were performed for five replicates per compound. Median values are shown as blue bars and mean values as red crosses

coverage remained unchanged (data not shown) compared at identical pH TMSO with TMS derivative detection. The chromatogram complexity decreased as expected, but not greatly (from 1.13 to 1.08 for the best results). The response coefficient distributions remained statistically unchanged whatever the derivatization protocol considered (Fig. 5). No significant differences could be found by the Kruskal–Wallis test at the 0.05 significance level (p > 0.05). However, for the same derivatization protocol (MSTFA or methoximation + MSTFA), the data obtained at low pH presented slightly enhanced response coefficients for citric, isocitric and αketoglutaric acids (respectively ca. four, three and two times) with a respective repeatability of the same order of magnitude. Concerning the repeatability of the procedures tested, the distributions of the RSD values were statistically unchanged (Kruskal–Wallis test at the 0.05 significance level, p > 0. 5) (Fig. 7), even if the dispersion was systematically higher for the two-step protocol, with a factor of ca. two orders of magnitude. This was not surprising, since the addition of a second step in the derivatization protocol will cause additional experimental errors, resulting in increased RSD value ranges.

Finally, the use of TMSO derivatives did not provide any significant optimization over the whole analysis procedure. Derivatization using MSTFA as sole derivatization reagent, i.e. without a catalyst, was definitively chosen for the whole procedure. Repeatability of whole process The repeatability of the optimized MSTFA derivatization process was estimated using constant amounts of ISD and ISND to select, at each stage of the protocol, the best technical choice, i.e. the one considered to be associated with the least variability. It included the choice of the sampling vials, the sample drying mode and the influence of the method of storage at −20 °C for 1 week. The best combination (i.e. the one described in BMaterials and methods^) proposed RSD values (calculated for 15 independent replicates) of 19%, 11% and 11% for the sampling vials, drying and storage, respectively. It allowed calculation of the global repeatability of the derivatization procedure by error propagation law calculation [45]. The value was found to be equal to ca. 25%, and


Phélippé M. et al. Response coefficients

Repeatability 150

7e+08 6e+08 5e+08

100

4e+08 3e+08 50

2e+08 1e+08 0e+00

0 neutral_TMS

acid_TMS

neutral_TMSO

acid_TMSO

neutral_TMS

acid_TMS

neutral_TMSO

acid_TMSO

Fig. 7 Distribution of the response coefficients grouped per derivatization protocol (i.e. neutral TMS, MSTFA silylation in neutral conditions; acid TMS, MSTFA silylation in acid conditions; neutral TMSO, methoximation + MSTFA silylation in neutral conditions; acid TMSO, methoximation + MSTFA silylation in acid conditions). The

response coefficients are expressed per nanogram of analysed compound, and the calculations were performed for all the detectable compounds in five replicates. Median values are shown as blue bars and mean values as red crosses

considered acceptable for quantification purposes and in line with the values commonly found for metabolomic profiling (ca. 25–35%) [12].

sensitivity of the quantitative analysis of those two compounds was among the worst results obtained, they are situated at key positions on the central metabolism map, where they are connected to better quantifiable compounds, i.e. fructose-6-phosphate, 3-phosphoglycerate and glucose-6-phosphate at a maximal distance from three metabolic nodes (ESM Fig. S1). The two main precautions recommended here would mainly be to first check the consistency of the measured data relative to the connected metabolites, and second to limit the interpretation of the data to biological fold changes greater than ca. three times the analytical variability when working on real cyanobacterial samples. The calibration curves (Fig. 8), which followed a fourparameter sigmoidal relation, were globally extended to the nanogram range, and presented an upper detection limit at 20– 160 ng of injected compound, offering a dynamic range of one or two orders of magnitude.

Calibration and limits of detection and quantification Detection and quantification limits of the developed protocol were measured in SIM mode with only 1/25 of the final sample volume injected in the GC–MS operating in split mode. The detection limits are expressed as amounts of injected compounds calculated on the basis of the amount required before derivatization with the MSTFA reagent (Table 4). The amino acids, monosaccharides and organic acids encompassing mainly the elementary bricks and the metabolic intermediates presented the lowest limit of detection (