Anal. Chem. 2007, 79, 403-415
Integrative Metabolomics for Characterizing Unknown Low-Abundance Metabolites by Capillary Electrophoresis-Mass Spectrometry with Computer Simulations Richard Lee, Adam S. Ptolemy, Liliana Niewczas, and Philip Britz-McKibbin*
Department of Chemistry, McMaster University, 1280 Main Street West, Hamilton, Ontario L8S 4M1, Canada
Characterization of unknown low-abundance metabolites in biological samples is one the most significant challenges in metabolomic research. In this report, an integrative strategy based on capillary electrophoresis-electrospray ionization-ion trap mass spectrometry (CE-ESIITMS) with computer simulations is examined as a multiplexed approach for studying the selective nutrient uptake behavior of E. coli within a complex broth medium. Online sample preconcentration with desalting by CE-ESIITMS was performed directly without off-line sample pretreatment in order to improve detector sensitivity over 50-fold for cationic metabolites with nanomolar detection limits. The migration behavior of charged metabolites were also modeled in CE as a qualitative tool to support MS characterization based on two fundamental analyte physicochemical properties, namely, absolute mobility (µo) and acid dissociation constant (pKa). Computer simulations using Simul 5.0 were used to better understand the dynamics of analyte electromigration, as well as aiding de novo identification of unknown nutrients. There was excellent agreement between computer-simulated and experimental electropherograms for several classes of cationic metabolites as reflected by their relative migration times with an average error of 100 mM) can further delay the transition to zonal separation as examined in a recent report.36 Second, the desalting function of CE that is important to minimize ion suppression in ESI-MS is effective as long as the analyte coion mobility is lower than the stacked Na+ ions generated by t-CITP. Thus, sample desalting in CE is effective for cationic metabolites that have a lower effective mobility than Na+ (i.e., 5.19 × 10-4 cm2/V‚s). It is important to note that, due to the low EOF under these conditions, excess Cl- anions migrate directly out of the capillary inlet upon application of voltage and do not impact ESI-MS detection. Last, when computer simulations were performed under conventional t-CITP conditions (data not shown) without a pH discontinuity, on-line preconcentration of Trp occurred by a different process as reflected by stacking at the front portion of the BGE-sample boundary. In addition, the extent of sensitivity enhancement was reduced since only a fraction of the long sample plug was effectively focused. Thus, enhanced incapillary sample preconcentration with desalting was achieved in CE-ESI-MS by a dual t-CITP with dynamic pH junction process, where analyte stacking occurred distinctly at the back end of the sample-BGE boundary prior to zonal separation and subsequent ESI-MS detection. Prediction of RMT of Cationic Metabolites by Computer Simulations. Recently, Gas et al.33 have demonstrated that computer simulations using Simul 5.0 can be useful for modeling electromigration behavior of multivalent ionic analytes, as well as predicting their separations in CE and ITP. Computer simulations can also be used as a promising tool in metabolomics research for prediction of RMTs in order to confirm the identity of unknown metabolites among several candidates selected after CE-MS analyses. As a first step in this direction, Simul 5.0 was assessed to predict the separation of a group of 10 model amino Analytical Chemistry, Vol. 79, No. 2, January 15, 2007
409
Figure 2. Time-resolved electropherograms showing the dynamics of in-capillary Trp focusing by t-CITP with dynamic pH junction from experimental data and computer simulations using Simul 5.0 (insets). A long sample plug of 12.5 cm was placed at increasing distance from the distal end of the capillary (Ld) from about (a) 7.5, (b) 25, and (c) 40 cm. Conditions: BGE, 1 M formic acid, pH 1.8; sample, 200 mM ammonium acetate, pH 7.0, 15 mM NaCl. Analyte peak numbering corresponds to the concentration profiles of (1) Trp (experimental), (1*) Trp (simulated), (2) NH4+, (3) Na+, and (4) pH. Note that concentration profiles of NH4+ and Na+ co-ions were reduced relative to Trp for clarity purposes.
acids with MetS as the internal standard. Figure 3a shows an experimental overlay of EIEs from a 20 µM standard amino acid mixture under optimum conditions for on-line sample preconcentration with CE-ESI-ITMS. Note that the sample injection plug was reduced to 6.2 cm in order to have sufficient separation for the closely migrating isobaric species (MH+ ) 182 Da), Tyr (9) and MetS (11), due to the limited resolution of the mass analyzer. Also, it was clear that, among the amino acids, Gly had the lowest ionization efficiency, that was ∼3000-fold lower than either Lys or Trp. In general, the migration order in CE-MS under these conditions was observed as follows, cationic > small neutral > 410 Analytical Chemistry, Vol. 79, No. 2, January 15, 2007
aromatic ≈ acidic amino acids, which is reflective of their characteristic µo and pKa. Figure 3b shows the simulated electropherogram under the same electrolyte conditions based on µo and pKa parameters33 listed in Table 1. In the case of the internal standard, MetS, which did not have literature values reported, its two fundamental parameters (µo ) 2.60 × 10-4 cm2/V‚s and pKa ) 2.01) were determined as described in the Experimental Section. Briefly, µo was derived after computer molecular modeling and incorporation of V and zo parameters into the HubbardOnsanger expression in eq 1, whereas the ionic strength-corrected pKa was determined experimentally by CE-MS by apparent
Figure 3. Comparison of experimental and simulated electropherograms for model amino acids using on-line sample preconcentration by CE-MS. (a) An overlay of EIEs for 20 µM amino acids using the internal standard, MetS, and (b) predicted electropherogram using Simul 5.0 based on µo and pKa parameters listed in Table 1. Conditions: BGE, 1 M formic acid, pH 1.8; sample, 200 mM ammonium acetate, pH 7.0; sample injection, 6.2 cm; V, 23 kV; and Lc, 80 cm. Analyte peak numbering: (1) Lys, (2) His, (3) Gly, (4) Ala, (5) Ser, (6) Thr, (7) Glu, (7a) Lys (13C), (8) Trp, (9) Tyr, (10) Asp, and (11) MetS (IS).
mobility changes as a function of buffer pH using eq 2. It was evident that there was a good qualitative agreement between experimental and predicted electropherograms in Figure 3 based on the similarity in the overall migration order. Although there was an average relative error of ∼5.6% in terms of absolute migration times for amino acids due to run-to-run variances in EOF, a reduced error of only 1.7% was achieved when comparing predicted with experimental (n ) 20, 2 days) RMT of amino acids using MetS as an internal standard. A small negative bias was noticed by linear correlation of predicted and experimental amino acid RMTs, as reflected by a slope of 1.10, y-intercept of -0.106, and R2 of 0.990. Thus, computer simulations can be used to reliably predict RMTs of metabolites in CE-MS even within discontinuous electrolyte systems provided that µo and pKa are accurate. Subsequent figures in this study will use RMT as the normalized
time domain for improved reliability when applied to metabolomic studies of selective nutrient uptake behavior of E. coli by CE-MS. Differential Metabolite Uptake during Bacterial Growth. An integrative metabolomic strategy using CE-ESI-ITMS with computer simulations was next developed to study the differential extracellular uptake of nutrients by bacteria within an undefined broth solution. LB is a complex nitrogen-rich carbon source medium derived from bactotryptone, yeast extract, and salt mixtures, which is often used to accelerate bacterial growth in E. coli cultures for enhanced expression of recombinant wild-type or mutant protein.51 Since the growth rate of bacteria is highly dependent on the amount and specific type of carbon source, optimization of the composition of a broth medium for selected (51) Gavina, J. M. A.; Das, R.; Britz-McKibbin, P. Electrophoresis 2006, 27, 41964204.
Analytical Chemistry, Vol. 79, No. 2, January 15, 2007
411
Figure 4. Series of EIEs revealing selective nutrient uptake within complex LB broth media as a function of E. coli growth at (a) 0, (b) 1.5, and (c) 5 h by CE-MS. Conditions similar to Figure 3, except that analyte peak numbering corresponds to the following: (1) Lys, (1a) unknown (isobaric to Lys, 147 Da) 2-Ser, (3) unknown (267 Da), (4) Thr, (4a) unknown (isobaric to Thr, 120 Da), (5) Trp, (6) Glu, (7) Tyr, (8) Asp, (9) MetS (IS), and (10) unknown (284 Da).
microbes to maximize biomass yield is important in various biochemical and bioengineering applications. Although metabolomic studies are often directed at analysis of intracellular metabolites using glucose minimal media,11,22 this study was directed at identifying and quantifying extracellular metabolites that are taken up as key nutrients within an undefined growth media. Our earlier study of enantioselective metabolite flux using on-line sample preconcentration with chemical derivatization by CE52 revealed selective uptake of different amino acids by E. coli without evidence of D-amino acid efflux into the broth medium. However, the method was limited in selectivity due to comigration of several unknown interferences when using CE with UV absorbance detection. Figure 4 depicts a series of overlay EIEs representing cationic metabolites identified as undergoing significant uptake by E. coli during different stages of bacterial growth at (a) 0, (b) 1.5, and (52) Ptolemy, A.; Tran, L.; Britz-McKibbin, P. Anal. Biochem. 2006, 354, 192204.
412 Analytical Chemistry, Vol. 79, No. 2, January 15, 2007
(c) 5.0 h. The major advantage of using on-line sample preconcentration with CE-ITMS for comprehensive metabolite analyses is that it offers excellent full scan sensitivity with qualitative information provided by multistage MS experiments. In addition to several standard amino acids undergoing uptake from the medium, there were also several minor unidentified isobaric species (e.g., 1a, 4a), as well as other major unknown metabolites (e.g., 3, 10) detected. Figure 4 qualitatively demonstrated that E. coli cultures showed preferential uptake of specific nutrients (e.g., Asp, Ser) at early stages of growth until their depletion with subsequent usage of other metabolites despite their overall higher abundance. Also, some amino acids were not used significantly as nutrient sources even after 8 h of incubation (e.g., Tyr, Lys). This feature reflects selective bacterial cometabolism in complex media, where multiple carbon sources that can be most efficiently used as direct sources in central energy metabolism are preferentially influxed within cells, such as Ser. Prior to quantitative analysis of differential metabolite uptake by E. coli, CE-ESI-ITMS
Figure 5. Integrative metabolomics strategy for unknown metabolite identification based on preliminary (a) multistage MSn and (b) dHDX experiments in conjunction with (c) computer simulations by Simul 5.0. Identification of two potential isomeric candidates ((3a) G and (3b) isoG) from KEGG database related to unknown nutrient 10 in Figure 4. Confirmation of G based on comparison of experimental with predicted RMTs using parameters in Table 2. Simulations were also applied to readily distinguish other isobaric ((1a) Lys vs (1b) Gln) and isomeric ((2a) Thr vs (2b) hSer) cations unresolved by conventional MS.
in conjunction with computer simulations was first applied for identification of several unknown nutrients detected in Figure 4. De Novo Identification of Unknown Nutrients. CE-MS in conjunction with computer simulations was used to identify several unknown metabolites used as key nutrients in LB broth medium. For instance, metabolite 10 in Figure 4 that was observed to undergo rapid uptake at early stages of bacterial growth (
