Anal. Chem. 2007, 79, 1507-1513
Evaluation of Matrix Effects in Metabolite Profiling Based on Capillary Liquid Chromatography Electrospray Ionization Quadrupole Time-of-Flight Mass Spectrometry Christoph Bo 1 ttcher, Edda v. Roepenack-Lahaye, Edith Willscher, Dierk Scheel, and Stephan Clemens*
Leibniz Institute of Plant Biochemistry, Weinberg 3, 06120 Halle (Saale), Germany
The coupling of liquid chromatography to electrospray ionization quadrupole time-of-flight mass spectrometry can be a powerful tool for metabolomics, i.e., the comprehensive detection of low molecular weight compounds in biological systems. There have, however, been doubts about the feasibility and reliability of this approach, because LC-MSsespecially with electrospray ionizations can be subject to matrix effects. We evaluated matrix effects for our metabolomics platform in three ways: (i) postextraction addition of a set of reference compounds to different complex biological matrixes to determine absolute and relative matrix effects, (ii) postcolumn infusion of two reference compounds, and (iii) mixing of two complex matrixes. Our data demonstrate that there are indeed significant absolute matrix effects when comparing highly divergent samples. However, relative matrix effects are negligiblesunless extremely divergent matrixes are comparedsand do not compromise the relative quantification that is aimed for in nontargeted metabolomics studies. In conclusion, employing LC-coupled ESI-QTOFMS for metabolomics studies is feasible yet rigorous validation is necessary. Metabolomics is emerging as a key technology of functional genomics and systems biology. This is illustrated, for instance, by the fact that “metabolomics technology development” was recently placed on the NIH roadmap as a component of “new pathways to discovery”.1 Metabolomics approaches seek to comprehensively detect and quantify the low molecular weight compounds in a cell or an organism.2,3 The information resulting from such nontargeted analysis will be of tremendous value for gene function analysis, will provide molecular insights into physiological states and biological networks, and will allow one to identify disease markers or to study the impact of environmental perturbations on a biological system. Being complementary to transcriptomics and proteomics, metabolomics will be an integral * Corresponding author. Tel. 49-345-55821420. Fax: 49-345-55821409. E-mail:
[email protected]. (1) Zerhouni, E. Science 2003, 302, 63-72. (2) Oliver, S. G.; Winson, M. K.; Kell, D. B.; Baganz, F. Trends Biotechnol. 1998, 16, 373-378. (3) Fiehn, O. Plant Mol. Biol. 2002, 48, 155-171. 10.1021/ac061037q CCC: $37.00 Published on Web 01/13/2007
© 2007 American Chemical Society
part of any attempt to progress toward a modeling of biological systems.4,5 Metabolomics is still largely in a technology development phase, again apparent from the listing on the NIH roadmap. Because of the chemical diversity of the metabolome, which for any given multicellular species comprises a mixture of thousands of compounds differing in size, polarity etc., and varying in abundance by several orders of magnitude, the need for multiparallel analytical techniques is obvious and well-accepted.6,7 Most projects to date have been employing GC/MS platforms, which offer the advantages of robustness, relatively low hardware costs, and the availability of software tools for deconvolution and metabolite identification.7,8 However, only volatile compounds are directly amenable to GC/MS analysis, and other, more polar nonvolatile compounds need to be derivatized. LC-MS is in principle a more versatile analytical tool. It covers a much wider mass range and allows one to target many compound classes not detectable by GC/MS. Furthermore, there is usually no need for derivatization, and modern LC-MS setups offer superior options to structurally elucidate unknown metabolites, namely, accurate mass determination and tandem MS. Because of this potential to complement GC/MS-based metabolite profiling, the number of metabolomics platforms employing LC-MS techniques is rapidly increasing. In particular, the capacity to analyze many classes of secondary metabolites makes LC-MS an attractive option in plant metabolomics. Metabolic plasticity and the synthesis of an enormous array of secondary metabolites, which serve important developmental functions and are essential for plant survival in ever-changing environments, can be considered hallmarks of plant biology. Consequently, many of the early metabolomics applications of LC-MS were developed for projects on Arabidopsis thaliana, tomato, and other plants.9-12 Initial results demonstrated the potential of LC coupled to (4) Fernie, A. R.; Trethewey, R. N.; Krotzky, A. J.; Willmitzer, L. Nat. Rev. Mol. Cell Biol. 2004, 5, 763-769. (5) Bino, R. J.; Hall, R. D.; Fiehn, O.; Kopka, J.; Saito, K.; Draper, J.; Nikolau, B. J.; Mendes, P.; Roessner-Tunali, U.; Beale, M. H.; Trethewey, R. N.; Lange, B. M.; Wurtele, E. S.; Sumner, L. W. Trends Plant Sci. 2004, 9, 418-425. (6) Goodacre, R.; Vaidyanathan, S.; Dunn, W. B.; Harrigan, G. G.; Kell, D. B. Trends Biotechnol. 2004, 22, 245-252. (7) Hall, R. D. New Phytol. 2006, 169, 453-468. (8) Dunn, W. B.; Ellis, D. I. TrAC, Trends Anal. Chem. 2005, 24, 285-294. (9) Tolstikov, V. V.; Lommen, A.; Nakanishi, K.; Tanaka, N.; Fiehn, O. Anal. Chem. 2003, 75, 6737-6740.
Analytical Chemistry, Vol. 79, No. 4, February 15, 2007 1507
electrospray ionization quadrupole time-of-flight MS to detect a large number of metabolites in plant extracts. However, doubts about the feasibility and reliability of LCMS-based metabolite profiling have been raised repeatedly, because LC-MSsespecially with electrospray ionizationscan be subject to matrix effects.3,4,13 This was observed for the first time by Tang and Kebarle.14 The term matrix effects refers to alterations of ionization efficiency of analytes by the presence of coeluting substances.15 This could severely compromise quantification. Mechanistically, matrix effects are not really understood. One probable cause is a competition of analytes and coeluting matrix components for ionization at the electrospray interface.16,17 It was shown, for instance, that the flow ratescontrolled by the aperture size of the spray needlesstrongly influences signal ratios for coeluting compounds. A critical factor apparently is the surface activity of the analytes, which determines their partitioning during droplet fission. Consequently, a lower flow rate favors ionization and reduces ion suppression because the initial droplet size is smaller.18 Other possible causes of ion suppression that are discussed in the literature include the presence of nonvolatile material and the influence of interfering compounds on the viscosity and surface tension of droplets. Nonvolatile material can decrease the efficiency of droplet formation;19 an increase in surface tension would lead to reduced solvent evaporation and lower probability of analytes to reach the gas phase.20 LC-MS/MS has owing to its high sensitivity and selectivity become the method of choice for a large number of pharmaceutical and clinical analyses. Often, matrix effects are not taken into consideration. That is why recently many authors have been proposing to make the evaluation of matrix effects a mandatory step in LC-MS method development.17,21 We decided to provide the needed rigorous validation4 for metabolomics applications and to systematically assess the influence of matrix effects in CapLCESI-QTOF-MS profiling in three ways: (i) We chose representatives of several compound classes detectable in A. thaliana extracts and performed spiking experiments in a diverse range of plant and fungal matrixes (postextraction addition method). Both absolute and relative matrix effects17 were determined. (ii) Two analytes differing in polarity were infused into the eluate after CapLC separation of complex samples (postcolumn infusion method). These two methods are considered standard approaches to assess matrix effects.15 In addition, (iii) we performed interfer(10) von Roepenack-Lahaye, E.; Degenkolb, T.; Zerjeski, M.; Franz, M.; Roth, U.; Wessjohann, L.; Schmidt, J.; Scheel, D.; Clemens, S. Plant Physiol. 2004, 134, 548-559. (11) Bino, R. J.; Ric, de Vos, C. H.; Lieberman, M.; Hall, R. D.; Bovy, A.; Jonker, H. H.; Tikunov, Y.; Lommen, A.; Moco, S.; Levin, I. New Phytol. 2005, 166, 427-438. (12) Clemens, S.; Bo¨ttcher, C.; Franz, M.; Willscher, E.; von Roepenack-Lahaye, E.; Scheel, D. In Plant Metabolomics; Saito, K., Dixon, R.A., Willmitzer, L., Eds. Springer: Berlin, 2006; pp 65-79. (13) Kell, D. B. Curr. Opin. Microbiol. 2004, 7, 296-307. (14) Tang, L.; Kebarle, P. Anal. Chem. 1993, 65, 3654-3668. (15) Taylor, P. J. Clin. Biochem. 2005, 38, 328-334. (16) Niessen, W. M. J. Chromatogr., A 1999, 856, 179-197. (17) Matuszewski, B. K.; Constanzer, M. L.; Chavez-Eng, C. M. Anal. Chem. 2003, 75, 3019-3030. (18) Schmidt, A.; Karas, M.; Dulcks, T. J. Am. Soc. Mass Spectrom. 2003, 14, 492-500. (19) Annesley, T. M. Clin. Chem. 2003, 49, 1041-1044. (20) Mallet, C. R.; Lu, Z.; Mazzeo, J. R. Rapid Commun. Mass Spectrom. 2004, 18, 49-58. (21) Rogatsky, E.; Stein, D. J. Am. Soc. Mass Spectrom. 2005, 16, 1757-1759.
1508
Analytical Chemistry, Vol. 79, No. 4, February 15, 2007
ence experiments with two very different matrixes and compared signal strength for more than 170 features after dilution of an extract with either pure solvent or another type of extract. EXPERIMENTAL SECTION Chemicals and Materials. All solvents used were of LCMS grade quality (Chromasolv). DL-R-Phenylglycine (1), 3-indolylacetonitrile (6), and formic acid (puriss. p.a. for mass spectrometry)werepurchasedfromFluka.BiochaninA(7),2-methoxybenzoic acid (4), kinetin (2), and N-(3-indolylacetyl)-L-valine (5) were obtained from Sigma-Aldrich, and quercetin-3-rutinosid (3) was from Acros. A. thaliana plants were grown in hydroponic culture.10 Roots and leaves of 6-week-old plants were harvested separately. Ten plants were pooled per sample. Arabidopsis halleri leaves were harvested from individual greenhouse-grown plants. Schizosaccharomyces pombe was cultivated as described.22 All biological material was frozen in liquid nitrogen and stored at -80 °C. Instrumentation. An API QSTAR Pulsar Hybrid Q-TOF-MS (Applied Biosystems/MDS Sciex) equipped with an ion spray source, a capillary LC system (Ultimate; Dionex), and an autosampler (Famos; Dionex) were used for LC-MS analyses. The mass spectrometer was operated under Analyst QS 1.0. LC Conditions. Chromatographic separations were performed on a modified C18 phase (GROM SIL ODS-4 HE, 150 × 0.3 mm, particle size 3 µm, pore size 120 Å) at 20 °C using water/formic acid, 99.9/0.1 v/v (A) and acetonitrile/formic acid, 99.9/0.1 v/v (B) as eluents. In order to assess the influence of chromatographic resolution on the extent of matrix effects, the following two gradient programs were used: gradient 1, 0-5 min, 5% B; 5-45 min linear from 5 to 95% B; 45-55 min 95% B; 55-65 min, 5% B; gradient 2, 0-5 min, 5% B; 5-25 min linear from 5 to 95% B; 2535 min 95% B; 35-45 min 5% B. Flow rate was 5.0 µL min-1 and injection volume 2.0 µL. MS Conditions. The mass spectrometer was operated in positive ion mode according to settings typically applied in a metabolite profiling experiment: ion spray voltage, 5.5 kV; nebulizer gas, N2, 25 au; curtain gas, N2, 20 au; declustering potential 1, 50 V; declustering potential 2, 15 V; focusing potential, 220 V; collision gas, N2, 4 au; ion release delay 10 µs; ion release pulse 10 µs; pulser frequency, 9.986 kHz; pulse duration, 9 µs; accumulation time, 2 s. Ions were detected from m/z 100 to 1000 at a resolution of R(fwhm) ) 8000-9000 within three Q1 transmission windows: m/z 100-400, 33% scan time; m/z 400-700, 33% scan time; m/z 700-1000, 34% scan time. Void time was t0 ) 6.4 min. Determination of Linear Range, Sensitivity, and Detection Limit. Methanolic stock solutions (2, 3.3 mM; 3-7, 10 mM) of each analyte were prepared, except 1, which was dissolved in water/formic acid, 99.5/0.5 v/v. Three independent mixtures containing each analyte at a concentration of 100 µM were prepared and serially diluted with methanol/water, 1/9 v/v, (50, 20, 10, 5, 1, and 0.5 µM). The samples (3 × 7) were analyzed by LC-ESI(+)-MS applying gradient 2. Analytes were quantified using the corresponding quantifier ions from Table 1. For each analyte, the linear range was estimated by plotting peak area-to-concentration ratios against log concentration and inspecting the plots (22) Clemens, S.; Bloss, T.; Vess, C.; Neumann, D.; Nies, D.; Zur Nieden, U. J. Biol. Chem. 2002, 277, 18215-18221.
Table 1. Analytical Parameters for the Set of Test Compounds Used in the Postextraction Addition Experiments gradient Aa
no.
analyte name
tr (min)
t1/2b (min)
1 2 3 4 5 6 7
R-phenylglycine kinetin rutin 2-methoxybenzoic acid N-(3-indolylacetyl)valine 3-indolylacetonitrile biochanin A
7.5 15.4 24.5 26.1 31.4 34.4 40.9
0.64 0.74 0.36 0.27 0.27 0.27 0.28
gradient Ba
Asc
tr (min)
t1/2b (min)
Asc
2.2 2.1 0.8 1.9 1.2 1.2 1.2
7.5 11.8 20.8 22.3 25.3 27.5 30.8
0.65 0.79 0.33 0.30 0.26 0.23 0.22
1.9 2.0 1.0 1.3 1.1 1.1 1.0
quantifier ion (m/z)
linear range (pmol)
sensitivityd (counts pmol-1)
LODe (pmol)
135.05 [M+H-NH3]+ 216.09 [M+H]+ 611.16 [M+H]+ 135.04 [M+HsH2O]+ 275.14 [M+H]+ 130.07 [M+H-HCN]+ 285.08 [M+H]+
10-100 1-200 10-200 2-200 1-40 1-100 1-20
991 998 93 3980 550 736 3680
5 0.5 2.5 0.5 0.5 0.5 0.2
a t ) 6.4 min. b Peak width at 50% peak height. c Peak asymmetry at 10% peak height. d Slope of the calibration curve in the linear range. e Limit 0 of detection estimated for S/N ≈ 3.
Figure 1. Evaluation of absolute matrix effects by postextraction addition. A set of seven reference compounds, structurally related to the major biosynthetic classes of secondary metabolites in A. thalianxa, were selected for a good distribution of retention times across the chromatographic run. They were spiked into four heterogeneous matrixes: A. thaliana leaf (black bars) and root (white bars) extracts, A. halleri leaf extracts (light gray bars), and S. pombe extracts (dark gray bars). Signal strength was compared to that in pure solvent. Shown are percent ion suppression or enhancement as determined in eight independent experiments and with two different gradients (A, total run time, 55 min; B, total run time, 35 min). * 6 was difficult to integrate in A. thaliana root extracts due to the presence of additional coeluting indole-derived fragments at m/z 130.07.
visually. Within the estimated linear range, the slopes of the calibration curves were determined by linear regression (R > 0.996). Limits of detection were estimated for a signal-to-noise ratio of 3. Preparation of Extracts. Freshly ground tissue (130 ( 5 mg) was subjected twice to the following extraction procedure: mixing with 200 µL of methanol/water, 4/1 v/v, sonication at 20-22 °C for 15 min and centrifugation at 19000g for 10 min. Both extracts were combined and evaporated under reduced pressure in a
Figure 2. Evaluation of relative matrix effects by postextraction addition. The test set of seven reference compounds was added to pure solvent (dark gray bars) and to extracts derived from either eight different pools of plants (A. thaliana leaf and root, black and white bars, respectively) or eight individual plants (A. halleri leaf, light gray bars). Relative standard deviations were determined for two different gradients (A, total run time, 55 min; B, total run time, 35 min). * 6 was difficult to integrate in A. thaliana root extracts due to the presence of additional coeluting indole-derived fragments at m/z 130.07.
vacuum centrifuge at room temperature. The remaining residue was redissolved in 40 µL of methanol, diluted with 360 µL of water, and filtered through a PTFE syringe filter (0.45 µm). Postextraction-Spiking Experiments. A spiking mixture of 1-7 was prepared daily from stock solutions with the following concentrations: 1, 250 µM; 2, 4, 7, 50 µM; 3, 200 µM; 5, 6, 100 µM. For the construction of a sample set, 10 µL of this mixture was spiked in 90 µL of methanol/water, 1/9 v/v (n ) 8) and in 90 µL of the appropriate matrix (n ) 8). In the case of A. thaliana, leaf and root extracts were obtained from eight different pools, Analytical Chemistry, Vol. 79, No. 4, February 15, 2007
1509
Figure 3. Evaluation of matrix effects by postcolumn infusion. Compounds 2 and 7 (kinetin, upper panel; biochanin A, lower panel) were continuously delivered into the column eluate. Shown is a comparison of solvent injection (A), A. thaliana leaf extract injection (B), and A. thaliana root extract injection (C).
Table 2. Limits of Detection for Kinetin (2) and Biochanin A (7) in Pure Solvent and A. thaliana Leaf Extract As Determined by CapLC-ESI(+)-MS and FIA-ESI(+)-MS LOD (pmol) CapLC-ESI(+)-MS
FIA-ESI(+)-MS
compd
injected in solvent
injected in matrix
injected in solvent
2 7
0.5 0.2
0.5 0.2
1 1
injected in matrix 10 5
while A. halleri leaf extracts originated from eight individual plants. S. pombe extracts were prepared from one single cell pellet. The sample set (2 × 8 samples) was analyzed by LC-ESI(+)-MS in random order starting with a blank injection using gradient program 1 or 2, respectively. For each analyte and each of the four matrixes, absolute matrix effects were calculated as [1 (mean peak area in the matrix)/(mean peak area in the neat solvent)]. For the three plant-derived matrixes, which were prepared from different pools/individuals, relative matrix effects were expressed as relative standard deviation of peak area. Interference Experiments. Leaf and root extracts were prepared from four different sample batches and pooled separately. Two sample sets were constructed by crosswise dilution. Set 1: 200 µL (100 µL) of the leaf extract was diluted either with 200 µL (300 µL) of methanol/water, 1/9 v/v (n ) 4) or with 200 µL (300 µL) of the root extract (n ) 4). Set 2: 200 µL (100 µL) of the root extract was diluted either with 200 µL (300 µL) of methanol/water, 1/9 v/v (n ) 4) or with 200 µL (300 µL) of the leaf extract (n ) 4). Both sample sets (4 × 4 samples each) were analyzed by LCESI(+)-MS using gradient 1. Extraction and integration of mass signals appearing either in the root or in the leaf extract were performed by XCMS.23 Settings for peak detection were as (23) Smith, C. A.; Want, E. J.; O’Maille, G.; Abagyan, R.; Siuzdak, G. Anal. Chem. 2006, 78, 779-787.
1510 Analytical Chemistry, Vol. 79, No. 4, February 15, 2007
follows: snr ) 6, fwhm ) 30, step ) 0.1, mzdiff ) 0.2, profmethod ) ”binlinbase”. Grouping and nonlinear retention time correction was accomplished in three iteration cycles with descending bandwidth. Integration results were manually checked for 20 randomly selected signals. Postcolumn-Infusion Experiments. A solution of 2 and 7 at 5 µM in methanol/water/formic acid, 80/19.9/0.1 v/v was prepared and continously infused via a syringe pump at a flow rate of 2.0 µL min-1 into the eluent between column and mass spectrometer. After injection of a leaf/root extract or pure solvent, 2 and 7 were continously monitored along gradient 1 using their protonated ions at m/z 216.09 and 285.08, respectively. Flow Injection Experiments. Separate dilution series of 2 and 7 (1, 2, 5, 10, 20, 50, 100, 200 µM) in pure solvent (methanol/ water/formic acid, 49.95/49.95/0.1 v/v) and in A. thaliana leaf extract (in methanol/water/formic acid, 49.95/49.95/0.1 v/v) were prepared. Flow injection ESI(+)-MS analysis (FIA-ESI(+)-MS) was performed using the same experimental setup as described for LC-ESI(+)-MS but omitting the chromatographic column. A flow rate of 5.0 µL min-1 of methanol/water/formic acid, 49.95/49.95/ 0.1 v/v and an injection volume of 2.0 µL was used. RESULTS AND DISCUSSION An evaluation of compounds tentatively identified in A. thaliana methanolic extracts showed that molecules from all but one (terpenes) of the biosynthetic classes known to date in A. thaliana secondary metabolism24 can be detected via CapLC-ESI(()-QTOFMS: glucosinolates and their breakdown products (isothiocyanates, nitriles), N-containing compounds (indole derivatives), phenylpropanoids, flavonoids/polyketides, and fatty acid derivatives.12 From these classes, we chose various structurally related test compounds that were commercially available in sufficient purity and not detected in A. thaliana extracts. These were initially screened for a good distribution of retention times. The final set (24) D’Auria, J. C.; Gershenzon, J. Curr. Opin. Plant Biol. 2005, 8, 308-316.
Table 3. Distribution of Signal Ratios (Mean Peak Area Matrix-Diluted Sample/Mean Peak Area Solvent-Diluted Sample) from A. thaliana Leaf and Root Extract Interference Experiments (See Figure 4) no. of signals with log2(signal ratio) within the interval exp
total no. of signals
[0;+0.5]
[0;+1.0]
[0;+2.0]
[-0.5;0]
[-1.0;0]
[-2.0;0]
84
12
18
19
51
64
65
84
10
16
20
35
60
63
93
23
30
33
47
60
60
93
24
31
37
35
45
53
root/leaf 1:1 (v/v) root/leaf 1:3 (v/v) leaf/root 1:1 (v/v) leaf/root 1:3 (v/v)
comprised DL-R-phenylglycine (1), kinetin (2), quercetin-3-rutinosid (3), 2-methoxybenzoic acid (4), N-(3-indolylacetyl)-L-valine (5), 3-indolylacetonitrile (6), and biochanin A (7) (Table 1). Retention times ranged from 7.5 to 40.9 min with the long gradient 1 (total run time 55 min, void time 6.4 min) and from 7.5 to 30.8 min with the short gradient 2 (total run time 35 min, void time 6.4 min). For the complete set of reference compounds, dilution series were analyzed to determine the linear range, the limit of detection, and the response factors (Table 1). All compounds showed a linear response over at least 1 order of magnitude. There was no significant difference in linearity when dilution series were prepared in A. thaliana leaf extracts (data not shown). Limits of detection ranged from 0.2 to 5 pmol; recovery rates were at ∼80%. Response factors differed by up to 43-fold. Metabolite profiling experiments are per se restricted to relative quantification. Calibration is not possible since identities of most analytes are unknown.7 All the more important is in the absence of standards or isotopically labeled reference extracts,25 the evaluation of quantification and, specifically, how matrix effects might compromise quantification of signals in complex samples. We performed and analyzed postextraction addition experiments in a way that allowed one to differentiate between absolute and relative matrix effects.17 Absolute matrix effects for the test set, i.e., the ratio between signal strength in a complex matrix and signal strength in pure solvent, were determined for four very heterogeneous matrixes: leaf and root extracts of A. thaliana, leaf extracts of the relative A. halleri, and cell extracts of the yeast S. pombe. Of higher immediate relevance for metabolomics experiments are relative matrix effects, i.e., the variation in matrix effect between samples derived from different experiments or
individuals that would in a scientific study be compared.17 In order to measure these, we carried out experiments with different batches (pools) of independently grown plants (A. thaliana matrixes) or with extracts derived from several individual plants (A. halleri matrixes). Absolute matrix effects are shown in Figure 1. For six of the seven compounds, we observed some degree of ion suppression; for 3, we found significant ion enhancement in the two A. thaliana matrixes. For 2, 4, 5, and 7, ion suppression was below 30% in all of the heterogeneous matrixes. Compound 6 was subject to stronger ion suppression in two of the matrixes. Massive ion suppression of >80% was found for 1. Overall, the results obtained for the two chromatographic gradients were very similar. Relative matrix effects for the test set were determined by comparing the relative standard deviation of signal strength after postextraction addition to eight independent A. thaliana or A. halleri samples (Figure 2). Technical variation was between 3.4 and 12.7% as was apparent from the relative standard deviation for addition to neat solvent ()blank). For 2-7, addition to different pools of A. thaliana root or leaf material did not result in any increase in variation. Thus, relative matrix effects were practically zero. This was found for both the longer and the shorter gradients. Slight relative matrix effects (increase in variation to between 5.6 and 25.1%) were determined for the comparison in eight different A. halleri extracts. This most likely is due to the fact that the A. thaliana matrixes were less diverse. Plants grown in a climate chamber were pooled, while for A. halleri, individual plants grown under less controlled conditions in a greenhouse were harvested. Remarkably, even for 1, which was subject to massive ion suppression, variation was within acceptable limits, i.e., in the range of typical biological variation,10,26 at least across different A. thaliana matrixes. The full set of postextraction addition experiments was performed at one analyte concentration close to the lower limit of the linear range. In order to check for concentration dependence of matrix effects, we spiked the test set into A. thaliana leaf extract at concentrations between 1 and 50 µM. No effect of concentration on the degree of ion suppression or enhancement was detectable (data not shown). It was apparent from the postextraction addition that compounds retained on the column showed only minor matrix effects in the complex matrixes used, while, not surprisingly, we observed
(25) Birkemeyer, C.; Luedemann, A.; Wagner, C.; Erban, A.; Kopka, J. Trends Biotechnol. 2005, 23, 28-33.
(26) Roessner, U.; Wagner, C.; Kopka, J.; Trethewey, R. N.; Willmitzer, L. Plant J. 2000, 23, 131-142.
Table 4. Signal Loss through Interference of A. thaliana Leaf and Root Extracts
sample root extract root extract leaf extract leaf extract a
dilution 1:1 1:3 1:1 1:3
total no. of signalsa in solvent-diluted sample
no. of no longer detectable signalsa in matrix-diluted sample
703 515 1487 850
158 (23%) 137 (27%) 314 (21%) 233 (27%)
S/N > 3, minfrac ) 1 (n ) 4).
Analytical Chemistry, Vol. 79, No. 4, February 15, 2007
1511
Figure 4. Interference experiments to evaluate matrix effects independently of reference compounds. Two different complex samples (A. thaliana leaf and root extracts) were diluted either with pure solvent or with the respective other sample in two different ratios (1:1, 1:3, v/v). Following alignment using the XCMS algorithm, intensity ratios of matrix-diluted sample to solvent-diluted sample were determined for robust signals (93 in leaf extracts, 84 signals in root extracts), log-transformed, and plotted against retention time (n ) 4). Identified compounds are numbered: 1, coniferin m/z 365.12 [M + Na]+; 2, syringin m/z 395.13 [M + Na]+; 3, scopolin m/z 193.05 [M + H - C6H10O5]+; 4, methoxy1H-indol-3-ylmethylglucosinolate m/z 399.12 [M + H - SO3]+; 5, methoxyascorbigen m/z 374.06 [M + K]+; 6, methoxy-1H-indole-3-carbaldehyde (isomer 1) m/z 176.07 [M + H]+; 7, methoxy-1H-indole-3-carbaldehyde (isomer 2) m/z 176.07 [M + H]+; 8, 2-(methoxy-1H-indol-3-yl)acetonitrile 160.07 [M + H - HCN]+; 9, 9-(methylthio)nonanenitrile m/z 186.13 [M + H]+; 10, kaempferol-3-O-R-L-rhamnopyranoside-7-O-R-Lrhamnopyranoside m/z 579.17 [M + H]+; 11, 4-(methylsulfinyl)butylisothiocyanate m/z 200.02 [M + Na]+; 12, sinapoylmalate m/z 207.07 [M + H - C4H6O5]+; 13, 5-(methylsulfinyl)pentylisothiocyanate m/z 214.03 [M + Na]+.
strong matrix effects in the solvent front.15 In order to investigate this further, we determined limits of detection for two differentially polar compounds, kinetin (2) and biochanin A (7), either following chromatographic separation or when using flow injection. Both compounds were analyzed in solvent and in A. thaliana leaf matrix. As shown in Table 2, the limits of detection were higher by factors of 10 and 5 for kinetin and biochanin A, respectively, when LC separation was omitted. Next, we applied postcolumn infusion, an evaluation method that is regarded as more dynamic since matrix effects are monitored over the entire chromatographic run and not just at the elution time points of spiked reference compounds.15 Kinetin (2) and biochanin A (7) were delivered into the column eluate. As shown in Figure 3 for injection of two matrixes, there was strong suppression within the solvent front between 6 and 10 min. Slight suppression was apparent in leaf extracts between 18 and 24 min, when several major metabolites such as sinapoylmalate are eluting. Noise was mostly a consequence of fluctuations in analyte flow. Also, there was no obvious difference between the more polar and the more nonpolar compound. Thus, we concluded from the results of the classical matrix effect assessment that in (27) Trethewey, R. N.; Krotzky, A. J.; Willmitzer, L. Curr. Opin. Plant Biol. 1999, 2, 83-85.
1512 Analytical Chemistry, Vol. 79, No. 4, February 15, 2007
the analyzed range of medium polarity compounds, which are amenable to extraction with methanol/water and separable on reversed-phase material, there is no indication for matrix effects to an extent that would be unacceptable for metabolomics studies, which are inherently compromised with regard to quantification:27 large numbers of compounds are detected simultaneously, most of them unknown and differing widely in abundance and chemical behavior. Still, we decided to perform a third set of experiments, this time abandoning the addition of reference compounds. Instead, two different complex samples were diluted either with pure solvent or with the other sample in two different ratios (1:1, 1:3, v/v). Using the XCMS algorithm,23 signals were identified among the several hundred detected ones in leaf and root extracts, which were sufficiently robust (S/N > 6), eluted at tR > 10 min, were according to signal strength in the two different dilutions detected in the dynamic range, and were consistently (n ) 4) absent from the respective other sample matrix. Ninety-three signals in leaf extracts and 84 signals in root extracts met these criteria (for information on m/z, retention times, intensities; see Supporting Information file 1). Several of the respective compounds were identified (numbered in Figure 4; MS data are summarized in Supporting Information file 2.). Quantification was checked manu-
Figure 5. Frequency distribution of signal intensity in a solventdiluted root extract (1:1 dilution) (A, white bars) and solvent-diluted leaf extract (1:1 dilution) (B, white bars). Black bars indicate frequency distributions of signal intensities that are “lost” due to interference of the respective other matrix in a 1:1 dilution.
ally for 10 randomly chosen signals in each root/leaf mixture. Deviation from automatic peak detection and integration was
