Article pubs.acs.org/jpr
A Simultaneous Metabolic Profiling and Quantitative Multimetabolite Metabolomic Method for Human Plasma Using GasChromatography Tandem Mass Spectrometry Otto I. Savolainen,* Ann-Sofie Sandberg, and Alastair B. Ross Division of Food and Nutrition Science, Department of Biology and Biological Engineering, Chalmers University of Technology, Gothenburg, Sweden S Supporting Information *
ABSTRACT: For the first time it is possible to simultaneously collect targeted and nontargeted metabolomics data from plasma based on GC with high scan speed tandem mass spectrometry (GC-MS/MS). To address the challenge of getting broad metabolome coverage while quantifying known biomarker compounds in highthroughput GC-MS metabolomics, we developed a novel GC-MS/MS metabolomics method using a high scan speed (20 000 Da/second) GC-MS/MS that enables simultaneous data acquisition of both nontargeted full scan and targeted quantitative tandem mass spectrometry data. The combination of these two approaches has hitherto not been demonstrated in metabolomics. This method allows reproducible quantification of at least 37 metabolites using multiple reaction monitoring (MRM) and full mass spectral scan-based detection of 601 reproducible metabolic features from human plasma. The method showed good linearity over normal concentrations in plasma (0.06−343 to 0.86−4800 μM depending on the metabolite) and good intra- and interbatch precision (0.9−16.6 and 2.6−29.6% relative standard deviation). Based on the parameters determined for this method, targeted quantification using MRM can be expanded to cover at least 508 metabolites while still collecting full scan data. The new simultaneous targeted and nontargeted metabolomics method enables more sensitive and accurate detection of predetermined metabolites and biomarkers of interest, while still allowing detection and identification of unknown metabolites. This is the first validated GC-MS/MS metabolomics method with simultaneous full scan and MRM data collection, and clearly demonstrates the utility of GC-MS/MS with high scanning rates for complex analyses. KEYWORDS: metabolomics, targeted, nontargeted, gas chromatography tandem mass spectrometry
■
metabolome6−11 so here we underline that GC-MS has advantages of far greater chromatographic resolution12−14 when compared to liquid chromatography mass spectrometry (LC-MS) based methods, large spectral libraries,15 and good retention of small compounds that tend to elute early with the solvent front in traditional reverse-phase LC-MS methods. The merits of GC-MS metabolomics also include electron impact (EI) ionization that produces a reproducible fragmentation pattern, which can be repeated across different GC-MS instruments. In addition, GC-MS metabolomics benefits from the use of retention indexes that makes comparison of retention times with different instruments possible when column stationary phase is kept constant. The combination of the reproducible fragmentation and retention indexes provides the possibility of identification of unknown molecules from databases with better accuracy than is currently available for LC-MS based methods. There has been limited method development and instrumental advances for GC-MS compared to LC-MS
INTRODUCTION Metabolomics is increasingly becoming a standard research tool in biosciences and biomedical research for helping to generate wider understanding of biological mechanisms. There are two main methodological approaches for metabolomics: targeted metabolomics and nontargeted metabolomics. Targeted metabolomics focuses on analyzing a predefined set of molecules, often relevant to the study question and usually gives quantitative or semiquantitative information. In contrast, nontargeted metabolomics aims to analyze the global metabolite profile in a sample allowing unhypothesized associations to be made between metabolites and the study question. Combining targeted and nontargeted metabolomics could provide the reliability and comparability of quantification with the possibility of novel discoveries, yet none of the main mass spectrometry methods used for metabolomics allow simultaneous acquisition of both targeted and nontargeted data. Presently there are numerous methods for metabolic profiling of blood plasma or serum using gas chromatography coupled to either a single quadrupole1 or a time-of-flight mass spectrometer (GC-MS).2−5 Many articles have outlined the relative merits of the different techniques for analyzing the © XXXX American Chemical Society
Received: August 24, 2015
A
DOI: 10.1021/acs.jproteome.5b00790 J. Proteome Res. XXXX, XXX, XXX−XXX
Article
Journal of Proteome Research
standard (concentration), from which the other four calibration standards were made by serial dilution with 90% ethanol (see SI Table S1 for concentrations). 90 μL of each standard solution was transferred into GC-vials, evaporated into dryness using a vacuum centrifuge evaporator (MiVac Duo concentrator, Genevac Ltd., Ipswich, UK) followed by addition of 270 μL of extraction solution (see below) and evaporation to dryness before derivitisation (see below). Internal standards were prepared in either water or methanol as 500 ng/μL stock solutions. The final concentration of all of the internal standards in the extraction solution was 0.0625 ng/ μL in MeOH:H2O (90:10 v/v). Plasma samples were prepared as using the method of A et al.2 with some modifications (SI Figure S1). Frozen plasma samples were thawed on an ice bath or at 4 °C and a 100 μL aliquot of each sample was transferred into a 1.5 mL microcentrifuge tube. 900 μL of extraction solvent at 4 °C containing the internal standards (SI Table S2) was added into the tube and the mixture was shaken at 30 Hz for 3 min in a bead shaker (Retsch GmbH, Germany) and incubated at +4 °C for 2 h. Samples were then centrifuged for 10 min at 17g. 300 μL of the supernatant was transferred into GC-vials, evaporated into dryness using a vacuum evaporator system (MiVac) and derivatized (see below).
metabolomics over the past decade. The use of GC coupled to tandem mass spectrometry (MS/MS) has great potential for improving sensitivity and quantification as the multiple reaction monitoring (MRM) mode allows a considerable improvement in sensitivity by essentially removing background noise for compounds of interest. This also enables more sensitive and specific identification and quantification of metabolites. Until recently, most GC-MS/MS systems have not been able to run MRM and full scan modes simultaneously and only recently developments in data collection electronics have allowed simultaneous collection of full scan and MRM data in some instruments.16,17 With quadrupole scan speed of up to 20 000 mass units/second in recent GC-MS/MS systems it is theoretically possible to obtain sufficient points over a peak when running both full scan spectra, and several MRM transitions simultaneously. This would enable acquisition of both nontargeted metabolic profiling (full scan mode) and targeted metabolomics (MRM mode) simultaneously in one analytical run. The combination of these offers the possibility for direct quantification of predefined metabolites relevant to the study question, with greater accuracy in addition to the ability of detection of unexpected changes in full scan data. The use of MRM also allows the targeted analysis of compounds present at low concentrations that could be missed using nontargeted methods. The use of “targeted metabolomics” methods have gained popularity over recent years, as they can be semi- or fully quantitative, allowing comparison of results with published literature using other quantitative methods. GCMS/MS with simultaneous full scan and MRM data collection may allow the best of both worlds. Here, we describe the development of the first method to utilize fast quadrupole GC-MS/MS to combining both targeted quantitative and nontargeted metabolomics methodologies into one single accurate, reproducible and reliable method.
■
Derivatization and GC-QqQ Analysis
Derivatization was performed as follows: Dried plasma extracts and standards were first methoxymated with 30 μL of methoxyamine in pyridine (15 μg/μL). After adding the derivatization agent, samples were vortexed for 5 min and put into an oven held at 70 °C for 60 min. The methoxymation reaction was left to continue overnight at room temperature. Samples were then silylated with 30 μL of MSTFA with 1% TMCS for 60 min at room temperature. After derivatization, 30 μL of heptane with 15 ng/μL of methyl stearate was added and samples analyzed by GC-MS/MS. Analysis were carried out on a Shimadzu GCMS TQ-8030 GC-MS/MS system consisting of a Shimadzu GC-2010 Plus gas chromatograph, Shimadzu TQ-8030 triple quadrupole mass spectrometer and a Shimadzu AOC-5000 Plus sample handling system (Shimadzu Europa GmbH, Duisberg, Germany). Data was acquired with Shimadzu GCMSsolutions software version 4.2. One μL of sample was injected with a split ratio of 1:4 and separation was done on a 15 m × 0.25 mm × 0.25 μm Rxi-5Sil MS column (Restek Corporation, Bellefonte, PA). Injection port temperature was set to 270 °C and oven temperature program was as follows: 70 °C for 2 min and ramped to 330 °C at 30 °C/min where it was held for 2 min. GC was operated in constant linear velocity mode set to 80 cm/sec. Septum purge flow was set to 3 mL/min. Interface and ion source temperature were 290 and 200 °C respectively. The autosampler was kept at 15 °C. Helium was used as the GC carrier gas, and argon was used as the MS/MS collision gas. Optimized conditions for MS data collection were scan time (50−750 m/z) 75 ms, MRM transitions 45 ms, for a total loop time of 120 ms. GCMSsolutions SmartMRM function automatically allocated time windows to minimize overlapping MRMs.
EXPERIMENTAL SECTION
Chemicals, Standards, and Reference Materials
All the reference compounds (Supporting Information (SI) Table S1) were from Sigma-Aldrich (St. Louis, Mo). The stable isotope labeled internal standards (IS) 13C5-proline, 2H4succinic acid, 13C5, 15N-glutamic acid, 1,2,3-13C3-myristic acid, 2 H7-cholesterol, 13C4-α-ketoglutarate, 13C12−sucrose, 13C4-hexadecanoic acid, 13C6-glucose 2H4−butanediamine, and 2H6salicylic acid were from Cambridge Isotope laboratories (Tewksbury, MA). Anhydrous pyridine, LC-MS grade methanol, LC-MS grade heptane, methyl stearate and methoxyamine hydrochloride were from Sigma-Aldrich (St. Louis, MO). Nmethyl-N-trimethylsilyltrifluoroacetamide (MSTFA) with 1% trimethylchlorosilane (TMCS) was from Thermo Fischer Scientific (Waltham, MA) or from Sigma-Aldrich (St. Louis, MO). Ethanol was from Solveco (Rosersberg, Sweden). Blood plasma collected using sodium citrate as an anticoagulant used for method development and as a standard control was from Sahlgrenska University Hospital (Gothenburg, Sweden). The standard reference material (SRM)1950 for human plasma was from National Institute of Standards and Technology (NIST, Gaithersburg, MD).
MRM Optimization
Standard Solutions and Sample Preparation
Each standard was analyzed in full scan mode (50−750 m/z) to identify precursor ions for optimization of the MRM events. The selected precursor ions were fragmented with collision energies between 9 and 45 eV in 3 eV intervals. Product ions were selected based on the abundance of the transition from
The individual standard stock solutions were prepared either in HPLC-grade water or in ethanol depending on compound solubility. The stock solutions were then further diluted into one solution with ethanol to make the highest calibration B
DOI: 10.1021/acs.jproteome.5b00790 J. Proteome Res. XXXX, XXX, XXX−XXX
Article
Journal of Proteome Research
Figure 1. Utilizing fast quadrupole technology with a gas chromatography triple quadrupole mass spectrometer it was possible to simultaneously acquire both nontargeted full scan data (A) and tandem mass spectrometric detection of key metabolites of interest (B).
using a targeted data processing method (see below). Intrabatch precision was also assessed for the number of features detected in full scan mode (see below). All tests were done by extracting and analyzing a plasma sample in triplicate over three different batches on three separate days spread over 2 weeks, together with samples from two clinical trials to simulate normal working conditions for this method. A fivepoint calibration curve was established in triplicate for each batch before starting analyzing the samples. Intrabatch reproducibility for the MRM channels and the corresponding targeted full scan data was obtained from triplicates analyzed during 1 day and interbatch reproducibility from all the runs on the three separate days (n = 9). Variation was expressed as RSD %. Precision of the detected features was evaluated by assessing the number of features that pass three defined RSD% thresholds (20, 25, and 30 RSD%). Matrix free blank samples were also evaluated for the number of features with the same reproducibility threshold limits. Method accuracy was tested by quantifying NIST SRM1950 plasma in triplicate and comparing the measured values with the reference values.
precursor to product ion. Dwell times were defined automatically by the software. Full Scan and Loop Time Optimization
Full scan (50−750 m/z) time was optimized between 75 and 50 ms in 5 ms intervals with five repeated injections of derivatized plasma sample (one vial for one scan time) while holding a constant loop time of 100 ms, with the remaining time used for scanning MRMs. Loop times were tested between 100 and 150 ms in 5 ms intervals with four repeated injections of derivatized plasma. A loop was created by using a fixed scan time of 75 ms and adjusting a random MRM event time to achieve the desired loop time. The different loop and scan times were evaluated for the average amount of spectral features detected in AMDIS18 deconvolution software with medium sensitivity and default settings. Linearity, Plasma Volume, and Limit of Detection
Standard linearity was tested with a five-point calibration curve prepared in triplicate that was designed to cover the normal biological variation in plasma based on data from the Human Metabolome Database.19 Intra- and interbatch variation were corrected by using normalization of the analyte response with IS response (SI Table S2). IS were selected on the basis of providing the lowest variation observed as RSD% and good linearity. Linearity in plasma was evaluated by preparing and analyzing plasma samples by using different amounts of plasma for extraction and generating calibration curves for the MRM channels. The volumes used were 40, 60, 80, 100, 120, 140, and 160 μL and samples were prepared in triplicates while keeping the amount of extraction solvent constant. The same samples were used to evaluate the effects of plasma volume on reproducibility of the MRM channels. Limit of detection (LOD) was estimated for each MRM analyte by using the following formula: LOD = 3s + b, where s is the standard regression error and b is the intercept of the calibration line.
Data Processing
A Matlab (Mathworks, Natick, MA) script and database developed at the Swedish Metabolomics Centre (Umeå, Sweden) was used for targeted analysis of the full scan data.20 The script identifies metabolites by searching for a specific m/z value (usually the base peak or molecular ion) within a given retention index window. Normalization and quantification was done as for MRMs to enable comparison between the two data collection modes. Peak areas were normalized by dividing the peak area with the corresponding IS area (SI Table S2) and quantification was done by generating standard calibration curves using the same methodology. Detection of spectral features for the precision study was done with XCMS-online21 using predefined settings (GC/ single Quad (centwave)) with the following settings; min peak width 0.6 s and max peak width 5 s. The results were exported into Excel (Microsoft Co., Redmond, WA) for quantification and variability calculations. Internal standard normalization of
Precision and Accuracy
Intra- and interbatch precision was assessed for both the analytes measured by MRM and if detected by full scan data C
DOI: 10.1021/acs.jproteome.5b00790 J. Proteome Res. XXXX, XXX, XXX−XXX
Article
Journal of Proteome Research raw data was performed using the method of Jonsson et al.20 IS variables from each sample were scaled to unit variance and modeled together using principle components analysis (SIMCA +, Umetrics AB, Sweden) to describe overall variation in multivariate space. Vectors from the first component were then used to correct for analytical variation between samples.
Linearity, Plasma Volume, and Limit of Detection
Linearity was assessed for the compounds measured by MRM both in standard solution and in the plasma matrix (SI Table S1) and was found to be good in both with R2 values >0.99 in most cases, The analytes that were structurally similar to the labeled IS used had the best linearity (e.g., proline and glutamic acid with their respective deuterated standards) and in these cases R2 values were greater than 0.999. The analytes with a R2 under 0.99 in standard solution were citric acid and 3hydroxybutyric acid, which had R2 values of 0.95 and 0.97 respectively. In the plasma matrix, α- and γ-tocopherol had inverse U shaped calibration curves, with reduced response when larger volumes of plasma were used. A coeluting compound from the plasma matrix is the most likely reason for the lack of linearity over a wide range for the two tocopherols as the phenomenon was similar for both, even though there is a 10-fold difference in plasma concentration (25 μmol/L vs 2 μmol/L). Both compounds were linear up to the plasma volume used in the final method (100 μL). Limit of detection was assessed from calibration curves for each analyte from both deconvoluted scanning data and MRM data. Overall the estimated LODs varied between 0.05 and 102.8 μmol/L for the MRM data and between 0.23 and 2118.4 μmol/L for the scanning data (SI Table S1). In most cases the estimated LODs were lower for the MRM method as can be expected due to the more selective detection of analytes when applying tandem mass spectrometry. The optimal amounts of plasma for extraction were 100 and 120 μL, with average % RSD between 3.5 and 4.1 respectively for the MRM channels. Use of only 40 μL led to the greatest variation between extractions (12%). These results match those found during the original development of the extraction protocol for GC-TOFMS.2
■
RESULTS AND DISCUSSION The aim of this method development was to combine the selectivity, sensitivity and reproducibility of tandem mass spectrometry with the potential of nontargeted full-scan profiling to detect unexpected and unknown metabolites in biological samples (Figure 1 and SI Figure S2). This is made possible by the development of a GC-MS/MS system with a faster scan rate than previously available.22 Here we demonstrate the possibilities of high scan speed GC-MS/MS to perform parallel MRM and full scan analyses with 37 selected molecules that cover a range of different metabolite chemistries that will be of interest in future studies. This includes organic acids, amino acids, and biomarkers of food intake that are normally present in concentrations below that detectable by normal GC-MS metabolomics methods. MRM Optimization
MRM conditions, including both qualifier and quantifier transitions, were optimized for the most intense and selective transition for each standard (SI Table S2). For the majority of compounds both the quantifier and qualifier transition were defined successfully. However, in the case of the two alkylresorcinols the molecular ion was found to be too weak for a reasonable transition, and were detected using single ion monitoring for a characteristic alkylresorcinol fragment at m/z 268.23 Dwell and measurement times were defined automatically in the software and ranged between 1.5 and 20.5 ms depending on the number of MRM channels measured simultaneously. These settings provided enough points per peak for reproducible quantification. MRMs for new compounds of interest can easily be added to the method so that it is easily customizable based on the research questions.
Precision and Accuracy
To evaluate both intra- and interbatch precision for the method, control plasma samples together with plasma samples from two separate clinical trials were analyzed over three separate days with at least 1 week in between to simulate realistic running conditions. Intrabatch variation ranged between 1.0 and 16.7% with 31 analytes 0.99), for some compounds inter- and intrabatch variation was high. While this is expected for a method covering a wide range of compound chemistries, and similar to that found with other GC-MS methods, work using labeled derivitization reagents to make a wide range of internal standards was able to reduce variation to
