Comprehensive Two-Dimensional Gas Chromatography Coupled to Time-of-Flight Mass Spectrometry in Human Metabolomics

Metabolomics is a discipline aiming to characterize a phenotype by means of metabolome analysis. In recent years, it has developed into an accepted and valuable tool in life sciences and its use has been growing rapidly in the study of microbial, plant, and mammalian metabolomes. It has been shown to be an effective tool in characterizing cancer cells and their response to anticancer drugs (Griffiths & Chung, 2008). To assess the effects of drugs on important pathways in clinical trials of innovative therapies, metabolomic approaches might be more cost-effective than those that measure specific molecular targets (Workman et al., 2006). The derived biomarkers applied in early clinical trials are expected to help identify appropriate patients, provide proofs of concepts, aid decision making, and ultimately reduce the high level of attrition and costs of drug development (Sarker & Workman, 2007).


Introduction
Metabolomics is a discipline aiming to characterize a phenotype by means of metabolome analysis.In recent years, it has developed into an accepted and valuable tool in life sciences and its use has been growing rapidly in the study of microbial, plant, and mammalian metabolomes.It has been shown to be an effective tool in characterizing cancer cells and their response to anticancer drugs (Griffiths & Chung, 2008).To assess the effects of drugs on important pathways in clinical trials of innovative therapies, metabolomic approaches might be more cost-effective than those that measure specific molecular targets (Workman et al., 2006).The derived biomarkers applied in early clinical trials are expected to help identify appropriate patients, provide proofs of concepts, aid decision making, and ultimately reduce the high level of attrition and costs of drug development (Sarker & Workman, 2007).
The biological specimens used in human metabolomic studies are e.g.urine (Weiss et al., 2007), blood plasma (Boernsen et al., 2005), and saliva (Walsh et al., 2006).All of them can play an important role in the diagnostic processes of problem illnesses.The individual metabolome of human biofluids is defined by genetic factors but it can also be affected by diet, age, disease, etc.In this respect, the use of human cell cultures offers a good alternative, since the influence of the above-mentioned factors is minimized in a culture where a defined extracellular environment takes place (Rabinowitz et al., 2006).
As concerns the analytical techniques applied in metabololmics, chromatographic techniques coupled to mass spectrometry play an important role.For the analysis of organic acids, amino acids, and sugars gas chromatography-mass spectrometry (GC-MS) after derivatization is widely applied.Since biological materials represent a very complex matrix, second dimension column has to be completed before the next modulation cycle starts.In this way, the separation obtained in the first dimension is preserved and additional separation on the second column is obtained.The re-injection process is called modulation and is enabled by an interface device called modulator, often referred to as the "heart" of the system.A GC × GC modulating interface can be placed at the end of the first dimension ( 1 D) or at the beginning of the second dimension ( 2 D), e.g. for thermal modulators, or between the columns (valve-based modulators).Nowadays, cryo-modulators which trap primary column effluent below ambient temperatures with the use of various cooling mechanisms are most commonly used (Edwards et al., 2011).
As a typical column set-up, a nonpolar column is used as the 1 D and polar one in the 2 D. Under these conditions, separation according to the volatility of the compounds occurs in the 1 D column while "polarity" separation dominates in the 2 D column.In principle, the column arrangement can be inverted.Both set-ups have their advantages and disadvantages and therefore the right set-up must be a result of optimization of a particular application.As concerns the column dimension, the first column is relatively long (typically 30-60 m) and normal bore (typically 0.25 mm with 0.25 µm film thickness).The second column has to be very short and narrow (typically 1-2 m of a 0.1 mm column with 0.1 µm film thickness) to perform a very fast separation.The use of such narrow-bore column results in limited sample capacity and easy 2 D column overloading, especially for biological samples.This limitation has been overcome by the use of wider-bore columns in the second dimension (Koek et al., 2008).
Primary GC × GC data are a series of second dimension chromatograms registered by the detector.Appropriate software reconstructs the second dimension chromatograms into three-dimensional plots or contour plots as shown in Figure 1.As concerns the detectors applicable in GC × GC, the peak width generated by this technique must be taken into account.Typical GC × GC peaks are 0.1-0.3s wide, thus to get a sufficient number of data points to accurately describe the shape of the peak (at least 10 points per peak) a detector must collect data at a rate of at least 100 Hz.For the coupling of GC × GC with MS detection, a high-speed TOF-MS is the technique of choice in most studies.Quadrupole MS have been also applied in GC × GC (Adahchour et al., 2006) with some compromises in the data density and mass range.
Since GC × GC generates large quantities of data, appropriate software tools become very important, not only for data acquisition but also visualization and interpretation.Depending on the data itself, various operations may be required, typically including background removal, mass spectral deconvolution, peak finding, combination of modulated peaks, peak height (area, or other characteristics) computation, identification of the found peaks by the comparison with mass spectral databases (or custom libraries), and finally export of analysis report (Reichenbach et al., 2004).

Application of GC × GC and its current trends
Since its introduction in 1991 (Liu & Philips, 1991), the GC × GC technique has gone through the years of rapid development.Today, GC × GC is widely used in many diverse areas which cover an interesting variety of applications.In general, the usage possibilities of GC × GC can be divided into three areas -fingerprints of very complex matrices, target analyses, and identification of unknown compounds.An excellent review on the GC × GC applications written by Adahchour et al. maps the usage of the technique from its introduction till 2008 in following fields -petrochemical products, environmental studies (soils and sediments, airs and aerosols, cigarette smoke), organohalogen compounds, food analysis (fats and oils, essential oils, alcoholic beverages), and also biological samples (Adahchour et al., 2008).
As regards the recent developments in GC × GC, the utilization of new stationary phases such as ionic liquids (Zapadlo et al., 2011), development of new modulators, e.g.(Panic et al., 2011), and improvement of data handling and evaluation (Wang et al., 2010;Kim et al., 2011;Koek et al., 2011) should be mentioned.

Derivatization via chloroformates
Analytes derivatization is employed in many analytical methods that utilize GC as a final step.The dominant reasons for derivatization in GC are either to increase analytes volatility, to improve their chromatographic behaviour by decreasing of polarity or to increase the detector sensitivity of the target analytes.The group of silylation procedures is by far the dominant derivatization methods.Silylation is almost universal technique and the silyl groups increase the total ion current which leads to increase the sensitivity using positive ion MS (Wells, 1999).
On the other hand, this derivatization procedure has several disadvantages.First, it is timeconsuming and requires elevated temperature.Further, the silylation reactions result in the formation of many artifacts as well as forming of multiple peaks for the same compound or the presence of unexpected peaks in the chromatogram.Also, in the electron impact mass spectra of silylated compounds, the non-specific masses belonging to the silyl group prevail, while the molecular and other characteristic ions have low intensity, which complicates spectra interpretation.Finally, it is also important to mention the instability of the silylated compounds and its sensitivity even to the traces of moisture which makes sample preparation and storage more demanding (Little, 1999;Ong et al., 2010).
As a promising alternative derivatization technique, an indirect alkylation via chloroformates appears.The sample preparation procedure is as follows.A portion of pyridine (serving as a catalyst) and an alcohol (to form esters) is added to the sample that is present in basic-aqueous environment.The reaction itself starts with the addition of an appropriate alkyl chloroformate.The reaction is fast (seconds) and no heating is needed.An illustration of derivatization reaction is shown in Figure 2.After the reaction, the derivates are directly extracted into the water-immiscible organic solvent (chloroform, isooctane) that can be immediately (or after drying by e.g.anhydrous sodium sulfate) injected.In this way a biological material (plasma, urine, cell extract) can be derivatized without any pretreatment.

Fig. 2. Derivatization method using methyl chloroformate (MCF). 2-Aminopropanoic acid (alanine) is converted to methyl 2-[(methoxycarbonyl)amino]propanoate
Using this procedure, due to the reaction conditions (pH, presence of other solvents), many polar-functional groups are converted to the corresponding forms, i.e. -carboxy, amino, hydroxyl, and thiol groups to esters, carbamates, carbonates, and thiocarbonates, respectively.The reactions are robust with low-cost reagents and in majority cases (>95%) produce a single stable derivate that has an easily interpretable MS spectrum (Figure 3).

Urine samples
For the analysis of organic acids, urine samples were acidic-extracted into ethyl acetate, ethoxymated and silylated by MSTFA.We analyzed 10 healthy urines and urines from patients with inherited metabolic disorder.For more detail see (Wojtowicz et al., 2010).
Analytes response was normalized to the concentration of creatinine measured by the common Jaffe rate method.

Fibroblasts
Human skin fibroblasts were cultured by standard protocol (Dulbecco's Modified Eagle's Medium, supplemented with 10% of fetal bovine serum and amphotericin 100 g/mL, 37 °C, 5% CO 2 , 25 cm 2 flasks) to confluence.The cells were harvested by quenching or trypsinization (see below).
Quenchinq procedure: The cells were quenched by spraying-out of 20 mL of 60% aqueous methanol (v/v) pre-cooled to -50 °C using plastic syringe with bent needle.The flasks with quenched cells were kept on dry ice and the cells were extracted with 1 mL of cold (-50 °C) methanol solution (80%, v/v) while scraping.The cell debris in the methanol solution was drained out with pipette and another 1 mL of cold extraction solution to wash the flask was used.Both methanol fractions were combined, sonicated (1 min), and centrifuged (1800g, 5 min) to remove the cell pellet and the supernatant was freeze-dried, silylated or alkylated via MCF.For the experiment with 5-fluorouracil (5-FU), the growing medium was supplemented with the drug (50 μmol/L) for 6, 24, and 48 hours before quenching procedure, respectively.
Trypsinization: Before trypsinization the cells were washed twice with 0.9% sodium chloride solution and then the trypsin/EDTA solution was added.After 2 min, trypsin was deactivated by adding of 5 mL of a cultivation medium.The cell suspension was centrifuged (260g, 5 min) and the pellet was extracted twice by 1 mL of 80% methanol.Combine extracts were freeze-dried.
Freeze-dried intracellular metabolite extracts were derivatized for analysis by two-stage silylation procedure based on the method described previously (Koek et al., 2006).The dry extracts were derivatized with ethoxyamine hydrochloride (10 µL, 56 mg/mL in pyridine) and 20 µL of pyridine for 60 min at 40 °C.Subsequently, the extracts were silylated for 50 min at 40 °C with 40 µL of MSTFA and 30 µL of pyridine.
The MCF derivatization was as follows: Dry metabolite extract was dissolved in 100 µL of water and 200 µL of sodium hydroxide (0.5 mol/L).The mixture was transferred into the glass tube containing 20 µL of internal standard norvaline (0.1 mmol/L).After, 200 µL of methanol and 50 µL of pyridine as a catalyst were added and the mixture was briefly vortexed.The derivatization reaction was started by adding 20 µL of MCF and the mixture was then vortexed for 30 s. Another 20µL portion of MCF was added again followed by shaking for 30 s.To separate the MCF derivatives from the reactive mixture a 300 µL of chloroform was added and shaken 10 s followed by the addition of 300 µL of sodium bicarbonate solution (50 mmol/L) and shaking for an additional 10 s.For better layering, the tubes were centrifuged (1000g, 1 min).The upper aqueous layer was discarded and the chloroform phase was dried by adding a small portion of anhydrous sodium sulfate.The dry organic solution was transferred to a GC vial with an insert which was tightly capped and then analyzed.

Plasma samples
The control and patient plasma samples were from infants from routine diagnostic processes performed in the laboratory of authors.The diagnoses had been previously confirmed by biochemical, enzyme or molecular-genetic analyses in all the patients.We analyzed 10 control samples and 19 samples with amino acids defects (phenylketonuria, PKU, maple syrup urine disease, MSUD, tyrosinemia I, TYR I, homocystinuria, HCYS, carbamoyl phosphate synthetase deficiency, CPS, ornithine transcarbamylase deficiency, OTC, and non-ketotic hyperglycinemia, NKH) (Janeckova et al., 2011).
For the MCF derivatization procedure, 50 µL of plasma were pipetted into the glass tube containing 20 µL of internal standard norvaline (0.1 mmol/L).After addition of 200 µL of sodium hydroxide (0.5 mol/L), the derivatization procedure was the same as in the case of fibroblasts.

Optimized analyses conditions
A Pegasus 4D system consisting of an Agilent 7890A gas chromatograph (Agilent Technologies, Palo Alto, USA), a MPS2/CIS4/ALEX system (Gerstel, Mülheim an der Ruhr, Germany), and a Pegasus HT time-of-flight mass spectrometer (LECO Corporation, St. Joseph, USA) was used.The GC × GC system employed a dual-stage, quad-jet modulator and a secondary oven, both built-in to the Agilent GC oven.A consumable-free option of the modulator was employed.Compressed air was used for both hot and cold modulation jets.For the hot jets the air was resistively heated, while for the cold jets the air passed through a moisture filter and was cooled by immersion cooling (-80 °C).

Data processing and quantification
An automated data processing based on the so called "Reference" was performed by the ChromaTOF software.The analyte concentration was calculated from the deconvoluted total ion current (DTIC) peak area.

Deconvoluted total ion current
Reference materials of rarely occurring pathological metabolites are not always easily available and/or their cost is considerably high.Therefore, it is a common practice to use the total ion chromatogram (TIC) signal for quantification along with the internal standard use.This approach was used also in this work.However, TIC quantification can overestimated results in case of chromatographic coelutions, which cannot be completely avoided even with GC × GC.This obstacle was overcome in our work by using deconvoluted total ion chromatogram (DTIC) for the quantification of the analytes.Deconvolution is a mathematical algorithm that is based on the absence of spectral skew and faster acquisition rates for peak apex definition in TOF-MS data.This algorithm mathematically separates mass spectra of compounds that chromatographically co-elute.In addition to producing deconvoluted spectra, the ChromaTOF software also allows the calculation of DTIC peak area.Deconvoluted TIC is the portion of TIC area corresponding to a particular analyte in a coelution.An example is shown in Figure 4. Fig. 4. Creation of DTIC for suberylglycine 2TMS (analysis of urine extract).On the right the raw peak apex spectrum (up) and the spectrum after deconvolution (down) are shown.The similarity between the deconvoluted spectra and the spectra from the library was 895.

Using the reference
When a classical GC-MS is used, the chromatogram is typically reviewed manually by plotting characteristic masses in the segment of the expected retention time.To avoid such a time-consuming procedure, a ChromaTOF feature called "Reference" was applied.This procedure consisted of the following steps: i.The sample is processed using a general peak finding method (e.g.S/N 200).ii.Peaks of interest are exported to the Reference, which is a set of information containing the retention times and mass spectrum of each analyte, among other data.Criteria such as the retention time window-width in both dimensions and minimum spectral match are defined by the user.iii.The Reference is applied to target search for each analyte in the unknown sample and for the quantification of positively identified analytes.iv.If some new interesting analyte is found in newly processed sample, it can be added to the existing Reference.
The retention time tolerances were determined based on the repeatability of retention times.The appropriate relative standard deviations (n=10) of the retention times of selected urine metabolites were under 0.17% and 1.74% for 1 D and 2 D, respectively (Wojtowicz et al., 2010).
Using this approach, two References have been made -one for TMS and one for MCF-based derivatization which will be described in following sections.

Analysis of samples derivatized by silylation
EZfaast standard solution (contains 32 amino acids and dipeptides), 10 healthy and 14 pathological urine extracts, and 15 quenched fibroblast extracts derivatized by TMS were analyzed by a GC × GC-TOF-MS method and subjected to data processing procedure.After manual inspection of these peaks and confirmation of their identity by means of retention indices, a Reference was created from these compounds.This TMS compounds Reference contains 268 analytes at this moment.

Analysis of urine acidic extract in relation to organic acidurias
A large subgroup of Inherited Metabolic Disorders called organic acidurias is characteristic by increased levels of organic acids in urine or the presence of pathological ones not appearing in healthy urine.The diagnosis of organic acidurias is commonly performed by the analysis of urine after acidic extraction.Besides organic acids, some other types of metabolites also serving as pathological markers, are extracted.Automated data processing with peak finding above S/N 200 was applied to the data, which resulted in the detection of 1353-3420 peaks in the set of studied urines.After sorting out GC column bleed peaks and the peaks belonging to the derivatization reagent, some 60% of the peaks remained.They are naturally occurring metabolites, pathological metabolites, nutrition artifacts, and drug artifacts, but most of the compounds are still not clearly identified.Our TMS Reference contains 153 identified and confirmed urine chemical species (from all 268 in the TMS Reference).In the Figure 5 is shown a 2D contour plot of healthy urine.
To evaluate the newly developed GC × GC method as well as the data processing strategy described above, we selected external quality control samples (ERNDIM, http://www.erndim.unibas.ch/)and one sample from an asymptomatic patient with medium-chain acyl-CoA dehydrogenase deficiency, who had been diagnosed by neonatal screening sixth days after birth.Since the patient was in a non-crisis state, many biochemical markers of the disease were in normal levels, so he could be missed by GC-MS.Using GC × GC-TOF-MS with the TMS Reference we found both hexanoylglycine peaks, which, although present at low concentrations, are obviously pathological markers and confirm the presence of the disease.Figure 6 illustrates the major benefit of our approach -separation of main markers of the disease from an excess of naturally occurred metabolite.For more details see (Wojtowicz et al., 2010).Fig. 6.Enlarged part of a contour plot from the analysis of urine from a patient with medium-chain acyl-CoA dehydrogenase deficiency.Two peaks of hexanoylglycine derivatives (1 and 2TMS) are resolved from a large peak of 4-hydroxyphenylacetic acid 2TMS.The deconvoluted spectrum (up) and the library hit belonging to the 2TMS derivate are shown (similarity 798)

Analysis of cultured human skin fibroblasts -comparison of trypsin treatment vs. quenching
The influence of most of external factors on the cell metabolomes is minimized in a culture where a defined extracellular environment takes place.To collect reliable metabolome data sets, culture and sampling conditions are crucial.The accurate analysis of intracellular metabolites requires a reliable sampling technique.Metabolites are generally labile species and their reactions occur on a time scale much shorter than that of large molecule synthesis or degradation, with substantial changes in small molecule concentrations possible during a time scale of seconds (Canelas et al., 2008).
The metabolomic analysis of adherent cell cultures presents a complex challenge, mostly given by limited sample sizes.The rapid quenching of the intracellular metabolism, simultaneously with the considerable removal of superabundant growing medium, are the main prerequisites.As a result of a known rapid turnover of intracellular metabolites, the sampling process should be very fast and cause minimal metabolome loss as a result of cell leakage.In this study, we compared trypsinization as the classical procedure for adherent cell harvesting (Dettmer et al., 2011) with the simple quenching procedure developed in our laboratory.
A set of studied fibroblasts has been analyzed by a GC × GC-TOF-MS method and the data was subjected to automated data processing with peak finding above S/N 200, which resulted in the detection of 431-601 peaks.Similarly as described in the section 5.1, our TMS Reference contains 72 identified and confirmed intracellular chemical species, from all 268 in the TMS Reference (Table 1).In the Figure 7 1.The list of 72 unique metabolites identified and confirmed in extracts of human cultured skin fibroblasts by GC × GC-TOF-MS In order to evaluate the effect of quenching vs. trypsinization, a cellular metabolome was determined in fibroblast cultures (n=6 for each method) from a single cell line.Data from analyses were corrected to percentages of a sum.The obtained compositional data were statistically analyzed.The geometric means for the single compounds were calculated.In order to be able to deal with Gaussian distribution of single compounds for variance analysis the data were transformed by the following equation [1/sqrt(2)*log(x/(1-x))], that represents a special case of so-called isometric log-ratio transformation (Egozcue et al., 2003).The results from trypsin and quenching approaches were compared as the natural logarithms of ratios of means and variations (Figure 8) and multivariate statistics -PCA and CA (Figure 9).It is clearly visible that the quenching technique substantially affects the concentrations of a number of metabolites.Several metabolites (e.g.citrate, lysine) differ by an order of magnitude.From the variations it is also evident that sample preparation by means of conventional trypsinization provides substantially more variable data in comparison to quenching by means of spraying.

MCF approach
EZfaast standard solution (contains 32 amino acids and dipeptides), 20 intracellular extracts from quenched fibroblasts, and 20 plasma samples derivatized by MCF were analyzed by a GC × GC-TOF-MS method and subjected to data processing procedure similarly as described in previous sections and the MCF Reference has been made.
Our MCF Reference currently contains 185 analytes -they are mostly organic acids and their derivates (the same as in the TMS approach) and compounds containing amino group (the exception is arginine because of the thermal instability of its MCF derivative that carries of thymidylate synthase activity or incorporation into RNA and/or DNA.The catabolism of 5-FU has been better understood only in recent years.The products of 5-FU catabolism have been linked to several 5-FU toxicities, including neurotoxicity (Grem, 2000;Kuhn, 2001).
This study was to show the differences in the fibroblasts metabolome after cultivation with addition of 5-FU against the non-treated controls (n=3 for each 6, 24, and 48h treatment, and without treatment, respectively), all from a single cell line.Data processing with peak Fig. 10.PCA projection of the cases (up) and the loading plot (down) for the comparison of the effect of 5-FU on the fibroblast cultivation.Triplicates (a, b, c) for 6, 24 or 48 hours of cultivation with presence of 5-FU and controls (C) are shown finding above S/N 200 was applied to the data, which resulted in the detection of 393-451 peaks in the set of studied fibroblasts.Our MCF Reference contains 78 identified and confirmed intracellular chemical species (from all 185 in the MCF Reference).Data from analyses were corrected to percentages of a sum, normalized to the unit standard deviation, and statistically analyzed through PCA (Figure 10). Figure 11 shows the differences for the most deviating metabolites.From the presented graphs it is clearly visible that cultivation with 5-FU influenced fibroblasts' metabolism.Automated data processing with peak finding above S/N 50 was applied to the data, which resulted in the detection of 408-594 peaks in the set of studied plasma samples.Our MCF Reference contains 65 identified and confirmed plasma chemical species (from all 185 in the MCF Reference).
Prior to statistical analysis the centred logratio (clr) transformation was applied.Data were evaluated using PCA and CA based on hierarchical clustering with a complete-linkage Euclidian distance method (visualized as a heat map) -Figure 12.
All the patients' samples were discriminated from the controls by appropriate metabolites in the PCA analysis.Patients with identical disease were recognized using the PCA approach and also clustered together.

Conclusion
This technique has been shown to be very powerful for the purpose of comprehensive sample profiling.On selected samples we demonstrated that higher separation power of GC × GC can help in removing co-elutions occurring in one dimensional approach.GC × GC is a valuable tool in metabolomic analysis of many biological matrices and enables diagnosing metabolic disorders.The great benefit is data processing that can be fully automated, what strongly simplify operator's effort and increases the sample throughput.

Acknowledgement
Infrastructural part of this project (Institute of Molecular and Translational Medicine) was supported from the Operational programme Research and Development for Innovations (project CZ.1.05/2.1.00/01.0030).

Fig. 1 .
Fig. 1. Construction of a contour plot.The 1 D effluent containing not fully separated analytes is modulated to create a series of short 2 D chromatograms.They are reconstructed by the software form a three-dimensional view.For the practical purposes the contour plot view is more feasible.

Fig. 5 .
Fig. 5. GC × GC-TOF-MS contour plot from the analysis of healthy urine (A) and enlarged part (B)

Fig. 11 .
Fig. 11.Graph of the most increasing (up) and decreasing (down) intracellular metabolites influenced by 5-FU.Differences are quantified as percentages of influenced response to the non-influenced ones.

Table
a 2D contour plot of analysis of the intracellular metabolites extracted from spray-quenched human cultured skin fibroblasts is shown.