metabolomics
Koek, Maud Marijtje
Citation
Koek, M. M. (2009, November 10). Gas chromatography mass spectrometry: key technology in metabolomics. Retrieved from https://hdl.handle.net/1887/14328
Version: Corrected Publisher’s Version
License: Licence agreement concerning inclusion of doctoral thesis in the Institutional Repository of the University of Leiden
Downloaded from: https://hdl.handle.net/1887/14328
Note: To cite this publication please use the final published version (if applicable).
105
5
Higher mass loadability in GC×GC–MS:
improved analytical performance for metabolomics analysis
ABSTRACT
A major challenge in metabolomics analysis is the accurate quantification of metabolites in the presence of (extremely) high-abundant metabolites. Quantification of metabolites at low concentrations can be complicated by co-elution and/or peak distortion when these metabolites elute close to high abundant metabolites. To increase the separation capacity a comprehensive two-dimensional gas-chromatographic method (GC×GC-MS) was set up, in which a polar first-dimension column and an apolar second-dimension column were used to maximize the peak capacity. The feasibility of using wider bore, thicker film columns in the second dimension to improve the mass loadability and inertness of the analytical system was investigated. Several column combinations with varying second-dimension column dimensions were compared with a setup with a narrow bore column (0.1 mm ID) in the second dimension. With a wider bore column (0.32 mm ID) in the second dimension the mass loadability was improved 10-fold, and the more inert column surface of the thicker film second dimension column resulted in a more accurate (automated) quantification and improved linearity in the presence of high concentrations of matrix compounds or metabolites. This amply compensated for the observed decrease in peak capacity of 40% when using a wider bore thicker film column. Compared to GC-MS and conventional GC×GC-MS, better performance for quantification of metabolites for typical metabolomics samples was achieved.
Based on: Koek, M.M.; Muilwijk, B.; van Stee, L.L.P.; Hankemeier,T. J. Chromatogr.
A 2008, 1186 (1-2), 420-429. DOI: 10.1016/j.chroma.2007.11.107
106
INTRODUCTION
Metabolomics involves the non-targeted analysis of changes in the complete set of metabolites (the metabolome) in cells, tissues or body fluids.1 The non-targeted approach can lead to new insights in the functioning of biological systems. However, the development of sensitive, quantitative and precise comprehensive analytical methods needed to achieve this goal is very challenging.2 In chapter 3 the development and validation of a gas-chromatography‒mass-spectrometry method is described, consisting of an oximation and silylation derivatization reaction and subsequent analysis with GC-MS. This method allows the analysis of a large range of small medium-polar to polar metabolites in cells.3
In metabolomics, the concentration differences of different metabolites within one sample and between different samples are very large. For example, in human-blood- plasma samples, normal glucose concentrations are as high as 700 – 1100 mg/L4, while the majority of metabolites have concentrations ranging from ng/L up to mg/L level.
Differences in composition or composition changes occur, for example, when analyzing urine samples or blood plasma obtained at different time points after drug administration or other perturbations. As the comprehensive approach in metabolomics requires a non-selective and generic sample pretreatment, the concentrations of metabolites in the sample solutions after sample pretreatment vary from sub-µg/L level up to 1000 mg/L or more. Ideally, all metabolites in a sample can be quantified, but no MS detector currently covers such a large range of concentrations. Therefore a major challenge in metabolomics with GC-MS is to maximize the total-mass tolerance, i.e.
the total amount that can be injected without disturbing the quantification. In other words, our goal is to maximize the difference between the lowest concentration that can be quantified and the highest concentration that can be present without disturbing the quantification and to facilitate the accurate and automated quantification of metabolites in the presence of high-abundant metabolites.
With the one-dimensional GC-MS method3 it was not possible to accurately quantify metabolites that co-elute with or elute close to high-abundant overloaded metabolites due to peak distortion. With comprehensive two-dimensional gas chromatography mass spectrometry (GC×GC-MS) higher separation capacities can be obtained and this technique should be a valuable tool to reduce the problems with co-eluting metabolites and improve their quantification compared to a one-dimensional approach. Another advantage of GC×GC-MS is the improved detectability, i.e. less co-elution and sharper peaks due to cryo-focussing lead to a higher probability of detection and correct assignment of identity of metabolites.
The current work was focused on the development and optimization of a suitable GC×GC-MS method for the analysis of metabolomics samples based on the earlier published GC-MS method.3 In GC×GC-MS, peak capacity, robustness, inertness, mass
107 loadability and analysis time are important performance parameters. All these parameters are influenced by, among other things, the stationary-phase types in the first (1D) and second dimension (2D), the column lengths, the internal diameters, the film thicknesses and the oven temperature program of the 1D and 2D column. Due to the nature of metabolomics samples, inertness and mass loadability, i.e. the amount of one component that can be injected on the column without influencing the separation efficiency, are especially important for the analytical performance. To minimize adsorption and degradation of the relatively polar derivatized metabolites at active sites on the column surface, a very inert analytical system is required. In addition, the large differences in concentrations within and between samples require a high mass loadability to reduce overloading effects. When overloading occurs, peaks eluting in the vicinity of the overloaded peak shift in retention time and very high concentrations of certain metabolites can even lead to peak shape deterioration of other metabolites due to changes in the stationary-phase characteristics caused by large amounts of the overloaded component.5,6 To facilitate accurate and automated quantification the analytical system has to be inert, to avoid degradation and adsorption of the derivatized metabolites, and sharp peaks and minimal retention time shifts are required for automated data processing.
The inertness of a GC column is generally increasing with the stationary-phase film thickness, as the surface of the column becomes less accessible. The mass loadability can be increased by using columns with wider internal diameters, greater film thicknesses and longer lengths or by using higher oven temperatures. Also the stationary phase type influences the mass loadability, the higher the solubility of the analyte in the liquid stationary phase, the higher the mass loadability.7
In conventional GC×GC, short and usually very narrow bore (0.1 - 0.18 mm ID) 2D columns are used that have limited sample capacities. Therefore overloading of the 2D column occurs very easily, especially when cryogenic modulation is used to achieve small peak widths in the second dimension.8 Moreover, the thin stationary-phase film in these columns could pose problems with degradation and/or adsorption of metabolites from our samples. Thus, while it is true that narrow-bore columns provide the best efficiency per meter and allow very fast separations in the second dimension, these columns might not always be the best choice for the analysis of those metabolites we are interested in. Especially when high concentrations and large composition differences are present in the samples, which applies to most metabolomic samples, longer columns with larger internal diameters and thicker stationary phase thicknesses may be better suitable.
Several papers have been published using GC×GC-MS with conventional setup for the analysis of metabolites in human urine9, human serum10, yeast cells11, mouse tissue12,13, and plant samples.14,15 In all these studies narrow bore (0.1 mm – 0.18 mm), thin film
2D columns were used, limiting the mass loadability in the second dimension. Some
108
work has been published about increasing the mass loadability in the second dimension using GC×GC-FID. Harynuk et al.16 used 2D columns with different internal diameters (0.1 mm – 0.32 mm) and film thicknesses (0.1 µm – 0.25 µm) and investigated the effects on peak asymmetry and peak widths of n-alkanes with varying concentrations.
Another approach to increasing the mass loadability was published recently by the same research group.17 They used a thicker stationary-phase film (up to 1 µm) column in the first dimension to broaden the 1D peaks and consequently inject a smaller amount into the second dimension column to reduce overloading. In both studies the lengths of the different 2D columns were one meter, so that the separation efficiency in the second dimension was higher for the narrow bore columns compared to the wider bore columns. To increase the mass loadability and maintain a high efficiency in the second dimension at the same time, one possibility is to use longer lengths and wider bore columns. However, when operating at near-optimum linear-velocity conditions, the column dead time and total separation time in the second dimension would be too high, restricting the use of long(er) 2D columns in GC×GC-FID. As a result of the vacuum outlet pressure in GC×GC-MS, the average column pressure in the 2D column is lower, resulting in increased gas-phase diffusion coefficients and increased optimum linear velocities.18 Consequently, it should be possible to use longer and wider bore 2D columns in GC×GC-MS at operating conditions near the optimum linear velocities in the first and second dimension and maintaining acceptable column dead times and separation times in the second dimension.
In this chapter the development of a GC×GC-MS method based on the previously validated 1D-GC-MS method3 is described. The method is suitable for the accurate automated quantification of metabolites in the presence of high abundant metabolites.
In analogy with the use of thicker film columns in 1D-GC-MS, the feasibility of using wider bore thicker film columns in the second dimension to improve the mass loadability and inertness of the analytical system was investigated. For this purpose first the configuration of the column set (type of stationary phases of first and second dimensions) was optimized. The relevant analytical performance (i.e. accurate quantification, linearity, limit of detection, mass loadability and peak capacity) of the different column sets with varying 2D column dimensions (internal diameters, stationary phase thicknesses and lengths) were then evaluated and compared to a conventional setup with a 0.1 mm ID 2D column, by using standards mixtures with varying amounts of glucose (50 – 2500 ng injected on-column). Obviously, the sample preparation was not further validated as this was done before (Chapter 3).3 The improved performance of the optimized method was demonstrated for the analysis of fetal-bovine-serum samples with different amounts of biomass.
109
EXPERIMENTAL
Chemicals
Pyridine (Baker analyzed) was purchased from Mallinkrodt Baker (Deventer, The Netherlands), methanol (p.a.) and dichloromethane (A.R.) were purchased from Biosolve (Valkenswaard, The Netherlands). A solution of 56 mg ethoxyamine hydrochloride (> 99%, Acros Organics, Geel, Belgium) per mL pyridine was used for oximation and N-methyl-N-trimethylsilyl trifluoroacetamide (MSTFA; Alltech, Breda, The Netherlands) was used for silylation.
Metabolite mixture (M-mixture)
A standard mix of isotopically labeled amino acids (20 different amino acids labeled with 2H, 15N) was purchased from Cambridge Isotope Laboratories (Andover, Massachusetts, USA). All other standards used were purchased from Sigma-Aldrich (Zwijndrecht, The Netherlands). A derivatized standard mixture of eighty-three metabolites (Supplement, Table S-1) from different metabolite classes and with varying volatilities was prepared. Stock solutions were prepared in pyridine (approximately 1000 ng/µL), or when metabolites were insoluble in pyridine, alternatively in methanol/water (1:4 v/v).
The metabolites in water/methanol were spiked to a tube and evaporated to dryness under nitrogen flow and subsequently the standards in pyridine were added. 240 µL of ethoxyamine reagent were added (total volume of standards + oximation reagent was 2500 µL) and the mixture was oximated for 90 min at 40°C. Then 2500 µL MSTFA were added and the mixture was silylated for 45 min at 40°C. The concentration for each metabolite in the derivatized mixture was approximately 8 – 10 ng/µL. The concentrations of the various 2H, 15N labeled amino acids ranged from 0.3 – 15 ng/µL.
To determine the detection limits and linear ranges the mixture was diluted to various extents, i.e. 2 to 500 times with pyridine:MSTFA 1:1 v/v mixture, resulting in
concentrations of the metabolites of about 20 pg – 5 ng/µL. Dicyclohexylphthalate (10 ng/µL) was added as an internal standard. More details on the method performance in Chapter 3.3
ine metabolite mixtures with different levels of glucose (G-mixtures)
To test the method performance in the presence of varying amounts of glucose, different amounts of a derivatized glucose standard (5000 ng/µL) were spiked to 500 µL M-mixture (above) and all spiked mixtures were subsequently diluted to a volume of 1 ml with pyridine:MSTFA 1:1 v/v. As a result, the nine different mixtures had the same concentrations for all metabolites, but contained different concentrations of glucose, i.e. 50, 100, 250, 500, 750, 1000, 1500, 2000 and 2500 ng/µL.
Dicyclohexylphthalate (10 ng/µL) was added as an internal standard.
110
Fetal-bovine-serum samples
US Defined Fetal Bovine Serum was purchased from Hyclone (Logan, Utah, USA).
Stock solutions of the metabolites for spiking (see below) of the serum extracts prior to extraction, were prepared in water. The stock solution of the deuterated octadecanoic acid was prepared in dichloromethane.
Three samples containing three different amounts of serum, i.e. 135 µL, 500µl and 750 µL, were spiked with the same amount of 2H, 15N labeled amino acid mix standard, octadecanoic acid and 6-fluoro-6-deoxyglucose. Also three blank serum samples with the same amounts of biomass as above were prepared. Methanol was added to precipitate the proteins; a volume of three times the sample volume was used. Then the samples were centrifuged for 10 min at 10,000 rpm. The supernatant was transferred to an autosampler vial and evaporated to dryness under a stream of nitrogen. The dry extracts were oximated with 200 µl of ethoxyamine reagent for 90 min at 40°C and subsequently silylated with 300 µl MSTFA for 45 min at 40°C.
Each derivatized spiked serum sample was diluted with the corresponding derivatized blank extract with the same amount of biomass, in order to determine the linear range.
Five dilutions per biomass concentration were prepared resulting in spiked metabolite concentrations from approximately 50 pg/µL to 5 ng/µL in the derivatized extract.
Dicyclohexylphthalate was added as internal standard. In total 15 samples were prepared with three different biomass concentrations (five each), glucose was the most abundant metabolite in the samples with a concentration of 250, 1000 and 1500 ng/µL in the samples containing the lowest, medium and highest biomass concentrations respectively.
Optimization of the types of stationary phases (
1D and
2D) for GC×GC
The following stationary-phase combinations were evaluated to determine the best column set for the derivatized metabolomics samples. HP-5MS (30 m x 0.25 mm x 0.25 µm) coupled to three different columns with polar phases in the second dimension, e.g. BPX50 (1 m x 0.1 mm x 0.1 µm), BP20 (1 m x 0.1 mm x 0.1 µm) and BPX70 (1 m x 0.1 mm x 0.2 µm). And a reversed setup with BPX50 (30 m x 0.25 mm x 0.25 µm) coupled to BPX5 (1 m x 0.1 mm x 0.1 µm) was tested. All columns were produced by SGE (Austin, Texas, USA).
GC×GC-MS analysis
The derivatized samples were analyzed with an Agilent 6890 gas chromatograph fitted with a dual-stage, four-jet (two liquid-nitrogen-cooled and two hot-gas jets) cryogenic modulator and a secondary oven (LECO, Monchengladbach, Germany) and coupled to a time-of-flight mass spectrometer (Pegasus III, LECO).
After the first optimization (above) all further experiments were carried out with the most suitable stationary phase column set. The 1D column used for all column
111 combinations was a 30 m x 0.25 mm x 0.25 µm BPX50 column (SGE). BPX5 columns (SGE) were used in the second dimension. The lengths of 2D columns were chosen so that approximately the same number of plates was obtained on the different columns used in the second dimension. The optimum linear velocities for these columns in the first and in the second dimension were calculated using GCcalc software18,19; a proper inlet pressure was calculated (in-house software) to obtain as optimal as possible flow rates in both dimensions with a maximum dead time of the 2D column of approximately 1 s (Table 1).
Table 1 Method parameters for different column sets
Column set A B C D
1D column (BPX50) a) 30 m x 0.25 mm x 0.25 µm
2D column (BPX5) b) 1 m x 0.1 mm x 0.1 µm
2 m x 0.25 mm x 0.25 µm
2 m x 0.32 mm x 0.25 µm
2 m x 0.32 mm x 0.5 µm
Theoretical plates/m c) 8127 4244 3505 3209
Optimum linear velocity d) 147 164 190 180
Linear velocity 1D e) 22 60 65 65
Linear velocity 2D e) 194 199 204 204
2D dead time (s) f) 0.52 1.01 0.98 0.98
Abs. inlet pressure (kPa) 400 300 300 300
Secondary oven (°C) g) + 5 + 30 + 30 + 40
Modulator offset (°C) h) + 20 + 40 + 40 + 50
Modulation time (s) 4 5.3 5 7
Hot pulse time (s) 1.6 1 1 1
a) 1D column: BPX50, 30m x 0.25mm x 0.25µm, optimum linear velocity: 30 cm/s, practical linear velocity: 50 cm/s (k = 10, T = 100°C).
b) Length of the 2D column excluding the transferline; an extra 20 cm of column was used as transferline.
c) According to performance report (test chromatogram) from manufacturer.
d) Average linear velocity in cm/s, calculated values with GCcalc, column temp: 100°C, and k- factors of 3, 3, 2.3 and 4.7 (to compensate for difference in phase ratio) for column A, B, C and D, respectively.
e) Average linear velocity in cm/s at the applied inlet pressure; calculated with in-house software.
f) 2D column dead time (s) calculated with in-house software.
g) Temperature offset of secondary oven compared to temperature of 1st dimension GC-oven.
h) Temperature offset of modulator compared to temperature of 1st dimension oven.
112
1-µL aliquots of the samples were injected in splitless mode (splitless time: 3 min) using PTV injection (Gerstel CIS4 injector; Mülheim an der Ruhr, Germany). The initial PTV temperature was 70°C and 0.6 min after injection the temperature was raised to 280°C at a rate of 2°C/s and held at 280°C for 10 min. The initial GC-oven temperature was 70°C, 3 min after injection the temperature was raised to 300°C with a rate of 5°C/min and held at 300°C for 10 min. Helium was used as carrier gas and the analyses were carried out in constant pressure mode (Table 1). The MS transfer line was set at 325°C and the ion source temperature was 280°C. The detector voltage was set at -1600V; only for the determination of the detection limits and the fetal-bovine- serum samples the detector was set at -1900V. Data acquisition rates were 50 Hz for column set A and 40 Hz for column sets B,C and D. All other operating conditions for the different column combinations are listed in Table 1.
Quantification of metabolites
The peak areas of the metabolites were determined using an appropriate reconstructed ion chromatogram per metabolite and all peak areas were corrected for variations in injection volumes and MS response with the internal standard dicyclohexylphthalate.
Peak areas of the metabolites in the standard mixtures with varying amounts of glucose (G-mixtures, above), as shown in Figure 3 and Table 2 were determined using the Chemstation software (D.01.02.16, Agilent Technologies). For this purpose, the parts of the chromatograms that contained the model metabolites were exported to ANDI files and imported in the Chemstation software to facilitate manual integration.
For all other automated quantifications (i.e. the data of Table 3 and Figure 4) the LECO ChromaTOF software (V2.32) was used.
Calculation of detection limits
The signal-to-noise ratios (S/N) of the model metabolites were determined manually in the standard with the lowest concentration that was still detectable, using the appropriate reconstructed ion chromatograms of selective ions for the metabolites. The detection limits were calculated by determining the S/N ratio and extrapolating to the S/N = 3 level. To compare the detection limits of the different column combinations with a one-dimensional GC-MS setup the diluted samples were also measured with combination B (Table 1) without using the modulator, the instrumental parameters were set as in the method for combination B; only a data acquisition rate of 2 Hz was used.
Calculation of peak capacity
The peak capacity, achieved by separation only and not taking MS detection into account, in the first dimension was calculated by dividing the total separation space in the 1st dimension (Tend – Tm; where Tm is the column holdup time and Tend the maximum retention time of elution of the last peak) by the average peak width (4σ) of
113 all detectable peaks in the M-mixture as calculated with the LECO software with a signal-to-noise ratio of over two-hundred. In the second dimension the separation is virtually isotherm, so the peak capacity was calculated using the formula: nc = 1 + (√N/4) · ln(tr/tm)20, where tm is the column holdup time and tr is the maximum retention time for elution of the last peak. The peak width and retention time of derivatized mannitol were used to estimate the number of plates (N) in the second dimension and subsequently the peak capacity, as this metabolite had the largest k-factor in the second dimension, and is therefore expected to provide the most reliable (and least favorable) results.
114
RESULTS AND DISCUSSION
Choice of stationary phases for GC×GC
The best column combination for metabolomics samples was determined by evaluating the peak capacity and the distribution of the 84 test metabolites from the M-mixture over the available chromatographic space. For all tested column combinations with an apolar column in the first dimension and a polar column in the second dimension, the retention of the 84 metabolites in the second dimension was very limited (retention factor k was 0.5 < k < 2.5, only for internal standard k > 2.5) (Figure 1A). For a polar x apolar setup (Figure 1B), the metabolites were much better spread across the chromatogram resulting in a better use of the available peak capacity compared to the apolar x polar setup (retention factors were 2 < k < 4). This can be explained by the relatively apolar character of the metabolites after derivatization with limited differences in polarity. In the setup with the polar column in the second dimension, the metabolites have very similar and low retention factors on the 2D column at their elution temperature, resulting in short retention times and limited differences in retention time. In the system with the polar column in the first dimension apolar metabolites elute at relatively low temperatures, and at these temperatures these metabolites have relatively high retention times on the apolar 2D column. More polar metabolites elute from the 1D column at relatively high elution temperatures and will have relatively low retention on the apolar 2D column. Although one might lose some selectivity in the first dimension, the improved selectivity in the second dimension in the non-orthogonal polar x apolar system resulted in a better use of the total chromatographic space.
The peak capacity achieved by separation only and not taking MS detection into account (calculated as described in Experimental) was approximately 8000 for the apolar x polar setup and 9000 for the polar x apolar setup, so no large difference in peak capacity was observed. In addition, the number of critical metabolite pairs, i.e.
peaks in the M-mixture (84 metabolites) remaining unresolved (Resolution 2Rs < 1) were determined: in the polar x apolar setup two critical pairs remained while for the apolar x polar combination five critical pairs were found.
The polar x apolar (BPX50 x BPX5) column combination was chosen for the remaining experiments, because the metabolites were better distributed over the available chromatographic space using this setup and the peak capacity was comparable with a conventional apolar x polar setup. In addition, the mass loadability in the second dimension was higher for the polar x apolar configuration, as the solubility of the derivatized metabolites in the apolar 2D stationary phase could be assumed to be better compared to a polar stationary phase (larger solute-liquid phase specific factors7).
115 Figure 1 Chromatograms of standard mixture on two column combinations; A (apolar x polar) and B (polar x apolar)
Comparison of second dimension columns with different lengths, diameters and film thickness
Once the polarities of the 1D and 2D columns were chosen, the dimensions of the 2D column were optimized (Table 1). The analytical performance was evaluated for all different column combinations, by investigating the following parameters, (1) influence of high abundant matrix compounds or metabolites on the accuracy of quantification of other metabolites, (2) linearity, (3) mass loadability, (4) detection limit and (5) peak capacity. Obviously, all these parameters were significantly influenced by the peak shape and eventually adsorption and/or degradation of each metabolite.
Figure 2 The model metabolites and their distribution across the 2D chromatogram (polar x apolar).
1: Succinic acid, 2: Fumaric acid, 3: 4-Aminobutanol, 4: Citric acid, 5: Glutamine, 6: Lysine, 7:
Fructose, 8: Glucose, 9: 9-Octadecenoic acid, 10: Octadecanoic acid, 11: Fructose-6-phosphate, 12:
Mannose-6-phosphate.
1 2 3
4 5 6 7 8
9 10 11
12
A B
116
To assess the performance, twelve metabolites of different compound classes and with different positions in the 2D-chromatogram were chosen as model metabolites for further quantification out of the standard mixtures (M- and G-mixtures) (Figure 2).
Among the 12 metabolites were cases of (partly) coeluting peaks in the first and/or second dimension, so-called critical pairs (Figure 2, metabolites 1,2,3 and 9,10,11,12), or elution in the vicinity of glucose (metabolites 4,5,6 and 7).
Influence of stationary film thickness in the second dimension on peakshape and quantification
The peak areas of the twelve model metabolites from the G-mixtures (same amount of metabolites but different amounts of glucose) were determined with all four different column sets. Two different effects were observed when a thicker stationary-phase film in the second dimension was used. First, the peak shapes of metabolites eluting close to a high-abundant peak (i.e. lysine, glutamine) were much less distorted and retention- time shifts were limited when a thicker film column was used. Secondly, the peak areas of some metabolites were higher compared to the thinner film setup when almost no glucose was present. This was probably due to the higher inertness of the surface of the thicker film column, which resulted in fewer active sites and, therefore, less adsorption on the silica surface of the column.
Both effects are illustrated in Figure 3. Because the concentrations of the test metabolites were the same in all mixtures, the same peak area (normalized on the internal standard dicyclohexylphthalate) of fumaric acid, lysine and citric acid at the various glucose concentrations would be expected, if no disturbances were present. For fumaric acid this is the case in both the conventional setup (Table 1, A) and the high loadability setup (Table 1, C). However in setup A, the peak areas of lysine and citric acid are influenced by the amount of glucose in the sample. At low levels of glucose the peak areas of lysine and citric acid were lower, probably due to adsorption at active sites on the 2D column, whereas in setup C this effect was much less pronounced due to the (more inert) thicker stationary phase film.
At high levels (> 750 ng on-column) of glucose the peak area of lysine (eluting close to glucose) in setup A was also lower (Figure 3). Due to the peak-shape distortion and to some extent co-elution with the large amount of glucose, the peak area of lysine could not be quantified accurately. In addition, at very high concentrations (>1500 ng) of glucose the peaks of lysine and glutamine, which co-eluted with glucose in the first dimension, even shifted to lower 2D retention times. At these amounts of glucose the stationary phase characteristics of the second dimension column of setup A were obviously changed and caused the change in retention time of these metabolites. In the setup C the shifts in retention time and distortion did not occur, because the characteristics of the thicker stationary phase were not influenced at that level of
117 glucose. In other words, the total-mass tolerance was improved; the quantification of lysine was still accurate, even at high levels of glucose (up to 2500 ng on-column).
Figure 3 Normalized peak areas (all peak areas corrected by internal standard; peak areas of the metabolites in the G-mixture containing 1000 ng/µL glucose was set at 100) of three metabolites with a concentration of ± 4 ng/µL in standards with increasing amounts of glucose measured on setup A (above) and setup C (below).
In Table 2 the influence of glucose on the quantification of metabolites is described as the RSD of peak areas of the same metabolites in samples with different glucose amounts (nine G-mixes, 50 – 2500 ng/µL glucose). For all setups using a thicker film and a 2D column with a larger ID (Table 1: B,C and D) a more accurate quantification was achieved than with the conventional setup A. Overall the most accurate quantification was achieved using column set C, with a 2 m x 0.32 mm x 0.25 µm BPX5 column in the second dimension. Even in setup D with a thicker film (0.5 µm) in the second dimension in the first dimension better results than with setup A were obtained. Only lysine could not be quantified due to co-elution with glucose as a result of the decreased efficiency in the second dimension compared to the other setups. For combination C it was also possible to use the LECO software for automated quantification of the model metabolites, which was less straightforward for setup A.
Due to the larger retention time shifts (compared to combination C) in the second dimension and the distorted peak shapes, the quantification of, especially, lysine and
0 20 40 60 80 100 120
0 1000 2000 3000
Glucose conc. in m atrix (ng/ul)
Normalized MS response
Fumaric acid Lysine Citric acid 0
20 40 60 80 100 120
0 1000 2000 3000
Normalized MS response A
C
118
glutamine could not be performed in an automated fashion with setup A. The quantification data for setup C from the LECO software were comparable to the results obtained with manual integration with the Chemstation software (data not shown).
Table 2 Relative standard deviations (n = 9) of peak areas of metabolites in samples with the different amounts of glucose
RSDs of peak areas for different column sets (%) a)
Metabolite M/z b) A c) B c) C c) D c)
Succinic acid 247 7 7 3 3
Fumaric acid 245 6 5 3 4
4-Aminobutanol 174 11 14 14 28
Citric acid 273 13 4 3 5
Fructose 307 11 9 8 10
Glutamine 161 69 23 10 30
Lysine 177 40 20 6 nd
9-Octadecenoic acid 339 13 6 4 7
Octadecanoic acid 341 8 8 4 9
Fructose-6-phosphate 315 9 6 3 11
Mannose-6-phosphate 387 12 6 4 9
a) Amount injected approximately 4 ng/µL for all metabolites. Glucose concentration ranged from 50 – 2500 ng/µL in the nine G-mixes, for details see Experimental.
b) Ion used for quantification in extracted ion chromatogram (manual integration).
c) Dimensions of 2D column, A: 1 m x 0.1 mm x 0.1 µm, B: 2 m x 0.25 mm x 0.25 µm, C: 2 m x 0.32 mm x 0.25 µm and D: 2 m x 0.25 mm x 0.5 µm.
Linearity
The linear ranges for the model metabolites were compared for setup A and setup C by injecting the M-mixture at varying concentrations (Table 3). The better linearity observed for some test metabolites for setup C may probably be assigned to less adsorption on the 2D column due to the thicker 2D stationary-phase film. The linear range of lysine and the sugar-phosphates started at much lower concentrations compared to setup A and the correlation coefficient (R2) for glutamine was higher in setup C. Also the intercepts of the calibration lines of lysine, citric acid and the sugar- phosphates were significantly smaller in setup C, indicating that less adsorption occurred.
Detection limits
The detection limits (LODs) were determined as described in Experimental. Detection limits obtained for setup A and C were comparable (Table 3), even though the average peak widths of the metabolites for combinations C were 1.4 times greater compared to the average peak widths in the setup A. Typical peak widths for setup A were 140 ms for metabolites with short second dimension retention times (2tR) up to 200 ms for
119 metabolites with long 2tR. For setup C peak widths of 170 ms up to 340 ms were obtained (data not shown).
The detection limits of glutamine and citric acid were better for setup C, due to less adsorption on the column surface, as discussed above. In general, the detection limits for all column sets tested were between 0.2 pg – 30 pg on-column (Supplement, Table S-2). Only the detection limits of combination D were slightly higher for some of the metabolites, as average peak widths on this column set were 2 times greater than on set A. For all combinations the detection limits were 2 to 20 times better than for 1D-GC- MS using the same mass spectrometer, even though a lower data acquisition rate was used for the one-dimensional separations (Supplement, Table S-2).
Table 3 Linearity and limit of detection of model metabolites a)
Compound Set-up A
Linear from
(ng/µL) n R2 Intercept
(ng/µL)
LOD (pg)
Succinic acid ≥ 0.08 9 0.9997 0.04 1
Fumaric acid ≥ 0.12 9 0.9933 0.1 1
4-Aminobutanol ≥ 0.08 7 0.9960 0.04 0.2
Citric acid ≥ 0.5 6 0.9956 0.4 15
Fructose ≥ 0.08 7 0.9971 0.03 5
Glutamine ≥ 0.5 5 0.9735 0.2 280
Lysine ≥ 0.75 7 0.9955 0.4 0.5
9-Octadecenoic acid ≥ 0.09 9 0.9994 0.04 2
Fructose-6-phosphate ≥ 0.9 5 0.9956 0.6 15
Mannose-6-phosphate ≥ 2 3 0.9984 1.3 30
Set-up C Linear from
(ng/µL) n R2 Intercept
(ng/µL) LOD
(pg)
Succinic acid ≥ 0.08 9 0.9989 0.06 2
Fumaric acid ≥ 0.12 9 0.9979 0.1 1
4-Aminobutanol ≥ 0.08 7 0.9983 0.02 0.2
Citric acid ≥ 0.2 6 0.9970 0.03 3
Fructose ≥ 0.08 7 0.9993 0.02 7
Glutamine ≥ 0.5 5 0.9915 0.25 110
Lysine ≥ 0.1 9 0.9989 0.07 1
9-Octadecenoic acid ≥ 0.09 8 0.9998 0.05 2
Fructose-6-phosphate ≥ 0.25 7 0.9978 0.2 15
Mannose-6-phosphate ≥ 0.4 6 0.9980 0.25 35
a) The experimental details are given in Experimental. Data for test metabolite octadecanoic acid are not shown, as this compound was also present in the septum, disturbing the measurements especially at low level of concentration.
120
Mass loadability
The peak shapes of deuterated octadecanoic acid-d35 in the 2D chromatogram at increasing concentration levels were evaluated for column combinations A and C, using the 2D chromatogram with the highest peak area of octadecanoic acid-d35 (Table 4). Peaks were considered overloaded when the peak asymmetry (As) measured at 10%
peak height was below 0.8 or above 1.2, or when a significant shift in the 2rT occurred (relative retention time shift > 1%) compared to the retention time of the standard with the lowest concentration. In setup A, the second dimension column was already overloaded when 25 ng of octadecanoic acid-d35 was injected on-column (approximately 90% of the peak area in the largest slice). In setup C the overloading of the 2D column occurred at a concentration level of 150 ng octadecanoic acid-d35 on- column (approximately 90% of the peak area in the largest slice). As expected, the mass loadability using a wider-bore, thicker-film 2D column was much better than using a narrow-bore thin-film 2D column; an increase in mass loadability of approximately a factor of 10 was achieved. More importantly, the limited shift in 2rT in setup C compared to setup A is a major advantage when automated data analysis is applied.
The loadability of a column is not only dependent on its dimensions, but also on the elution temperature and the compatibility of the column stationary phase and the eluting metabolite.7 In this case the metabolite tested for both columns was the same and the elution temperatures in the second dimension for both combinations was comparable, so that these parameters had no significant influence on the difference in loadability.
Table 4 Peak asymmetries and retention-time shifts of octadecanoic acid-d35 at varying concentrations a)
Setup A (2D: 1 m x 0.1 mm x 0.1 µm) Setup C (2D: 2 m x 0.32 mm x 0.25 µm) ng on-column As b) rT shift (%) ng on-column As rT shift (%)
2 0.86 0 10 1.1 0
10 0.83 0.9 50 1.0 0
25 0.83 3.6 75 1.0 0.6
50 0.59 6.2 100 1.1 0.6
75 0.60 7.1 150 1.3 1.9
a) For details, see Experimental.
b) As; Asymmetry factor.
Peak capacity
The peak capacity for all setups as achieved by separation only and not taking MS detection into account was calculated from peak-width data of standard mixtures as described in Experimental. In the conventional setup (A), the peak capacity was higher than in the setups B, C and D with thicker films, as the peak widths in the second dimension using this setup provided the narrowest peaks (Table 5). However, in
121 practice, the actual peak capacity using this setup was lower due to overloading of the metabolites present at higher concentrations, and their effect on metabolites present at lower concentration.
Table 5 Peak capacities of the different column sets a)
Peak width (s) Peak capacity
Column set 1D b) 2D c) 1D 2D Total
A 8.5 0.20 330 30 9900
B 8.8 0.34 320 20 6400
C 9.6 0.34 290 20 5800
D 9.6 0.51 290 20 5800
a) For calculations see Experimental
b) Average peak width (4σ) of all peaks in standard mixture with S/N > 200
c) Peak width (4σ) of mannitol
A decrease in peak capacity of 40% for setup C compared to setup A was observed.
Still setup C was preferred as a result of the improved mass loadability and inertness of the analytical system. Obviously, the actual peak capacity when using appropriate selected-ion chromatograms as acquired by MS detection is much higher than the calculated value, because in principle a separation of only a few data acquisition points is enough to deconvolute two peaks.
Performance in serum samples
Fifteen fetal-bovine samples with three different biomass concentrations (five each) spiked with labeled metabolites at concentration levels ranging from 0.05 – 5 ng/µL were measured with setups A and setup C. The results were consistent with the results in standard solutions. The peak shapes of metabolites eluting close to high-abundant matrix compounds or metabolites were improved and a more accurate quantification was achieved with setup C compared to setup A. As shown in figure 4, the quantification of lysine and 6-fluoro-6-deoxyglucose was less accurate at the highest level of biomass (corresponding with a glucose level of 1500 ng/µL in the serum extract).
The linearities of most model metabolites were acceptable (R2 > 0.996), however the results for setup A were less good than for setup C (Supplement, Table S-3). For example, the linear dynamic range of valine in setup A started at a higher concentration and the regression coefficients for valine and isoleucine were lower compared to setup C. This was probably caused by adsorption on the 2D column. Due to the shielding of the active sites in the 2D column by the metabolites from the serum, less adsorption was observed compared to standard solutions for both setups (Table 3). Still the performance of the wider-bore, thicker-film column set (C) was better than that of the narrow-bore thin-film setup.
122
Figure 4 Normalized peak areas (peak areas corrected for internal standard, peak areas of the metabolites in the serum containing 1000 ng/µL glucose were set to 100%) of lysine, 6-fluoro-6- deoxyglucose and octadecanoic acid (spike level: 2.5 ng/µL) in serum samples with increasing amount of biomass measured on setup A (above) and setup C (below).
Figure 5 show2 two full-scan chromatograms of a fetal-bovine sample analyzed with setup A and setup C. One of the benefits of using a wider-bore thicker-film column in the second dimension is illustrated; the higher mass loadability in the second dimension in setup C reduced the overloading effects significantly. As a result, serine and threonine were baseline separated from the large overloaded urea peak with setup C, whereas in setup A these metabolites could not be separated from urea.
0 20 40 60 80 100 120 140
0 500 1000 1500 2000
Normalized MS response
0 20 40 60 80 100 120 140
0 500 1000 1500 2000
Glucose conc. in m atrix (ng/µl)
Normalized MS response
lysine fluoroglucose octadecanoic acid
A
C
123 Figure 5 Full scan chromatograms of fetal-bovine-serum sample with conventional setup A (left) and high loadability setup C (right). Enlargements (100 s x 1.5 s): positions of serine (1), threonine (2) and urea (3). Color intensity settings for both chromatograms are the same.
A 1 2
3
C
3
2
1
124
CONCLUSION
The difference in metabolite concentrations in metabolomics samples are generally extremely large (≥ 9 decades). As a result, one of the major challenges is the accurate automated quantification of (low-abundant) metabolites in the presence of extremely high-abundant matrix compounds or metabolites, which are typically resulting in overloaded peaks in GC-MS. By using a non-conventional column setup in GC×GC- MS, i.e. polar x apolar combination with a wider-bore and thicker-film 2D column, a better quantification was achieved compared to conventionally used GC×GC-MS setups and 1D-GC-MS.
By the use of a wider-bore thicker-film 2D column, the most important drawbacks of the conventionally used narrow-bore thin-film 2D columns were overcome, i.e. the limited mass loadability and limited inertness towards the metabolites of interest. The enhanced total-mass tolerance, and thereby, improvement of peak shapes and reduction of retention-time shifts, allowed for automated data analysis of metabolites at concentration levels as low as 50 pg/µL, even when they co-eluted with a very-high abundant (2500 ng on-column) metabolite in the first separation dimension. With the new setup, data analysis for metabolomics studies is more straightforward compared to conventional GC×GC-MS: automated quantification using a target table of all (or as many as possible) detectable metabolites becomes feasible. In future research the possibilities of this approach will be further explored.
In conclusion, due to the higher mass loadability and inertness of the second dimension column, and the higher separation efficiency compared to one-dimensional GC-MS, the developed GC×GC-MS method is particularly suitable for metabolomic profiling of complex samples with large differences in composition within and between samples, such as microbial, plant or mammalian (blood, urine, tissue) metabolomic samples.
125 Reference List
1. Fiehn, O. Plant Mol. Biol. 2002, 48 (1-2), 155-171
2. van der Werf, M. J.; Hankemeier, Th.; Jellema, R. H. J. Ind. Microbiol. and Biotechnol. 2005,
3. Koek, M. M.; Muilwijk, B.; vanderWerf, M. J.; Hankemeier, T. Anal. Chem.
2006, 78 (4), 1272-1281
4. Kusmierz, J.; DeGeorge, J. D.; Sweeney, D.; May, C.; Rapoport, S. I.
J.Chromatogr.B 1989, 497, 39-48
5. Deans, D. R. Anal. Chem. 1971, 43 (14), 2026-2029
6. Cazes, J.; Scott, R. P. W. Preparative chromatography. In Chromatography Theory, 1 ed.; CRC: 2002; pp 419-444.
7. Ghijsen, R. T.; Poppe, H.; Kraak, J. C.; Duysters, P. E. Chromatographia 1989, V27 (1), 60-66
8. Ong, R.; Shellie, R.; Marriott, P. J. Sep. Science 2001, 24 (5), 367-377
9. Sinha, A. E.; Hope, J. L.; Prazen, B. J.; Nilsson, E. J.; Jack, R. M.; Synovec, R.
E. J. Chromatogr. A 2004, 1058 (1-2), 209-215
10. O'Hagan, S.; Dunn, W. B.; Knowles, J. D.; Broadhurst, D.; Williams, R.;
Ashworth, J. J.; Cameron, M.; Kell, D. B. Anal. Chem. 2007, 79 (2), 464-476 11. Mohler, R. E.; Dombek, K. M.; Hoggard, J. C.; Young, E. T.; Synovec, R. E.
Anal. Chem. 2006, 78 (8), 2700-2709
12. Welthagen, W.; Shellie, R.; Spranger, J.; Ristow, M.; Zimmermann, R.; Fiehn, O. Metabolomics 2005, 1 (1), 65-73
13. Shellie, R. A.; Fiehn, O.; Welthagen, W.; Zimmermann, R.; Zrostlikovβ, J.;
Spranger, J.; Ristow, M. J. Chromatogr. A 2005, 1086 (1-2), 83-90 14. Hope, J. L.; Prazen, B. J.; Nilsson, E. J.; Lidstrom, M. E.; Synovec, R. E.
Talanta 2005, 65 (2), 380-388
15. Pierce, K. M.; Hope, J. L.; Hoggard, J. C.; Synovec, R. E. Talanta 2006, 70 (4), 797-804
16. Harynuk, J.; Gorecki, T.; Zeeuw, J. d. J. Chromatogr. A 2005, 1071 (1-2), 21- 27
17. Zhu, Z.; Harynuk, J.; Gorecki, T. J. Chromatogr. A 2006, 1105 (1-2), 17-24 18. Hail, M. E.; Yost, R. A. Anal. Chem. 1989, 61, 2402-2410
19. Hail, M. GCCalc v2.1 2001, Novatia LCC, internet communication viewed 9 August 2009, http://www.enovatia.com/gccalc.
20. Poole, C. F. The Essence of Chromatography; 1st ed.; Elsevier Science B.V.:
2003.
126
SUPPLEMENT
Table S-1 List of eighty-three metabolites present in metabolite mixture (Experimental)
Amino Acids Sugars Fatty acids (continued)
Alanine 2-Deoxyglucose Docosanoic acid
Asparagine 2-Deoxyribose-5-phosphate Dodecanoic acid
Aspartic Acid 6-Phosphogluconic acid Hexadecanoic acid
Cysteine Fructose Icosanoic acid
Glutamine Fructose-6-phosphate Octadecanoic acid
Glutamic Acid Galactonic acid Pentadecanoic acid
Glycine Gluconic acid Tetradecanoic acid
Histidine Glucosamine Others
Isoleucine Glucosamine-6-phosphate 2-Amino-1-butanol
Leucine Glucose 2-Amino-1-propanol
Lysine Glucose-6-phosphate 2-Methoxyphenol
Methionine Maltose I 3-Amino-1,2-propanediol
Phenylalanine Mannitol 3-Amino-1-propanol
Proline Mannose-6-phosphate 3-Hydroxybenzylalcohol
Serine Myo-inositol 3-Phosphoglyceric acid
Threonine Ribose 4-Amino-1-butanol
Tryptophan Ribose-5-phosphate 4-Aminobenzoic acid
Tyrosine Salicin 4-Hydroxybenzylalcohol
Valine Xylitol 4-Hydroxyproline
Organic acids Fatty acids 7-Dehydrocholesterol
Cholic acid 11-Amino-undecanoic acid Adenine
Citric acid 2,6-Diaminoheptanedioc acid Cholic acid methyl ester
Fumaric acid 2-Aminoadipic acid Dihydrocholesterol
Lactic acid 2-Hydroxycaproic(C6) acid Glycerol
Malic acid 9,12,15-Octadecatrienoic acid Glycerol-3-phosphate Nicotinic acid 9,12-Octadecadienoic acid Histidinol
Oxo-glutaric acid 9-Hexadecenoic acid N-Acetyl-L-aspartic acid
Pyruvic acid 9-Octadecenoic acid N-formyl-methionine
Succinic acid Decanoic acid
127 Table S-2 Comparison of detection limits (S/N= 3) of standard mixtures (in pg on-column) a)
Column set
Metabolite 1D GC-MS A b) B b) C b) D b)
Succinic acid 25 1 1 2 4
Fumaric acid 20 1 0.5 1 1
4-Amino-butanol 1 0.2 0.3 0.2 0.3
Citric acid 50 15 2 3 20
Fructose 15 5 3 7 2
Glutamine 120 280 190 110 70
Glucose 3 1 0.5 1 1
Lysine 3 0.5 0.5 1 n.d. c)
Pentadecanoic acid 30 2 0.5 0.5 1
9-Octadecenoic acid 35 2 2 2 20
Fructose-6-phosphate 55 15 10 15 8
Mannose-6-phosphate 220 30 25 35 20
11-Amino-undecanoic acid 6 0.5 0.5 0.5 3
Cholic acid 120 15 10 5 40
a) For details see Experimental.
b) Dimensions of 2D column, A: 1 m x 0.1 mm x 0.1 µm, B: 2 m x 0.25 mm x 0.25 µm, C: 2 m x 0.32 mm x 0.25 µm and D: 2 m x 0.25 mm x 0.5 µm.
c) n.d. not detected.
Table S-3 Correlation coefficients of model metabolites in fetal bovine serum at different levels of biomass a)
R2 Setup A R2 Setup C
Biomass conc. b) Low Medium High Low Medium High
Valine 0.9725 0.9977 0.9959 0.9995 0.9976 0.9984
Isoleucine 0.9902 0.9997 0.9964 0.9991 0.9995 0.9968
6-Fluoro-6-deoxyglucose 0.9997 0.9998 0.9991 0.9995 0.9997 1.0000
Lysine 0.9999 0.9999 0.9986 0.9994 0.9998 0.9998
Octadecanoic acid-d35 0.9963 1.0000 0.9991 0.9998 1.0000 1.0000
a) For details see Experimental.
b) Samples were prepared with three different biomass concentrations, corresponding to a concentration of 250 (low), 1000 (medium) and 1500 (high) ng/µl glucose in the injected samples.
128
