ARTICLE pubs.acs.org/ac
High-Throughput, Accurate Mass Metabolome Profiling of Cellular Extracts by Flow Injection Time-of-Flight Mass Spectrometry Tobias Fuhrer,† Dominik Heer,† Boris Begemann,†,‡ and Nicola Zamboni*,† † ‡
Institute of Molecular Systems Biology, ETH Zurich, Switzerland Life Science Zurich PhD Program on Systems Biology, ETH Zurich, Switzerland
bS Supporting Information ABSTRACT: Direct injection of samples on high-resolving mass spectrometers is an effective way to maximize analytical throughput and yet allow analyte discrimination in complex samples by mass-to-charge ratio. We present a platform of flow injection electrospray time-of-flight mass spectrometry to profile small molecules in >1400 biological extracts per day at native mass resolution. We comprehensively benchmark the performance with more than 5000 injections of chemically defined standards and Escherichia coli cellular extracts obtained from miniscale cultivations. For at least 90% of tested compounds, we attain a linear response over 3 decades of concentration, interday coefficient of variation of 1000 samples per week (5 min per run). However, the low resolving power of the instrument and binning into unit masses during processing did not allow for discrimination of compounds with the same nominal mass. Although some of these caveats can be compensated by lowthroughput fragmentation studies,22,23 there is an obvious interest Received: May 20, 2011 Accepted: August 10, 2011 Published: August 10, 2011 7074
dx.doi.org/10.1021/ac201267k | Anal. Chem. 2011, 83, 7074–7080
Analytical Chemistry in exploiting the excellent mass resolution of MS, i.e., Fourier transform and time-of-flight (TOF) instruments. As outlined by many studies before, accurate mass in the ppm range massively decreases the ambiguity of ion identification24,25 also in the absence of chromatographic separation.23,26 28 For instance, Madalinski et al. demonstrated reproducible detection of 400 signals in yeast extracts by flow injection on a LTQ-Orbitrap at a cycle time of 3 min per sample and with intraday coefficient of variations of 0.9 and annotated within 0.4 ( 0.3 10 3 amu accuracy (Table S-2, Supporting Information). In positive mode, 38 compounds were annotated on the basis of accurate mass within 0.001 amu deviation, assuming [M + H+] was the predominant ion. A linear response with a correlation coefficient of >0.9 was obtained for 36 compounds. The average mass error was 0.3 ( 0.2 10 3 amu accuracy. Calibration slopes ranged from a few hundred to 10 000 and 80 000 counts/μM in negative and positive mode, respectively. Lower limit of detection were 0.01 to 2 μM (Table S-2, Supporting Information). The estimated lower limits of detection and the high linear response at low concentrations allow for qualitative but robust detection of shifts in metabolite concentration. Reproducibility. Reproducibility was tested with a set consisting of E. coli wild-type and four gene deletion mutants lacking either pgi, pgl, purM, or aroD. We grew each strain in 16 replicate wells (80 in total) in a 96 well plate to mid exponential phase and extracted the water-soluble metabolites. The growth experiment was repeated on three different days starting from independent precultures. All extract plates (240 samples in total) were analyzed by flow injection TOF MS measured in triplicates. Similarly, the analysis was repeated for all samples on three different days to a total of 2160 injections per polarity. The analytical reproducibility for triplicate injections of the same sample plate was within 10% (relative standard deviation, Figure S-3, Supporting Information). Technical reproducibility was analyzed comparing intensities from the same sample plates measured on different days, while biological reproducibility was based on intensities recorded on the same day for independent extractions. Standard deviations were typically within 10 and 20% (Figure S-3, Supporting Information) and particularly worse for ions with 1.5; p-value, 0.90). This hypothesis is trivial for isotopologues as determined by the natural abundance of stable isotopes.22,28 In the case of adducts and neutral losses, the same claim is supported by the fact that both instrument settings and matrix composition are constant and, thus, also the ratio between electrospray derivatives is likely to be conserved. Third, we systematically tested for double charged ions and homodimers, as well as heterodimers. Albeit rare in chromatography-based systems, the latter are more probable than homodimers with flow injection. In the resulting computational analysis with a tolerance of 0.003 amu and the three aforementioned additions, 10% of the ions were assigned to singly deprotonated metabolites or peptides, and 20% matched the mass of dimers, adducts, or secondary isotopes (Table S-4, Supporting Information). To verify the putative annotations solely based on accurate mass and correlation, we selected representative candidates featuring sufficient abundance to allow for collection of fragmentation spectra. This was obtained by continuous infusion of cell extracts at nanoflow and collisional induced fragmentation. Out of 15 ions annotated as deprotonated metabolites, 12 were positively confirmed by comparing the observed fragmentation patterns against an internal spectral library and the MassBank database.37 Isotope shifts could be observed for 12C 13C and confirmed in two cases. In contrast, the identity of putative adducts, dimers, or doubly charged ions was proven valid in only 2 cases out of 14 tested. Because of the high false positive rate, we decided to remove dimers and multiple charges from our annotation procedure. The low fraction of verifiable adducts motivated us to further leverage on the size of the data set and mass accuracy to discover recurring adducts. We searched for ions that over all samples correlated strongly and then built a histogram of the mass differences measured for each pair of ions found (Figure S-6, Supporting Information). The resulting plot shows that the mass shifts specific for 12C 13C, CO2, H/Na exchange, H3PO4, and C2H4 were frequently observed. To our surprise, the neutral gains of H2PO4Na, H2PO4K, HPO4Na2, (H2PO4Na)2, and (H2PO4)2H3PO4 were also strikingly prominent. Hence, we included these in our list of potential neutral gains. On the basis of these notions, we finally annotated the ions obtained for both 7077
dx.doi.org/10.1021/ac201267k |Anal. Chem. 2011, 83, 7074–7080
Analytical Chemistry
ARTICLE
Table 1. Annotation Summary model KEGG ecob
Feist 2009b
mode
mode
positive
negative
positive
negative
3222 57996
3222 57996
884 15912
884 15912
ions (centroids) detected
3241
1531
3241
1531
unknown ions
2525
841
2832
1038
compounds in database theoretical ions derived from database
annotated ionsa
716
690
409
493
annotated ions matching a (de)protonated compound
163
168
65
123
matched compounds including isomers
309
453
85
210
total (de)protonated compounds (without overlap)
670
244
positive negative mode overlap
92
51
a
Annotation included M H+, + Cl , + OH or M + H+, + K+, + Na+in negative or positive mode, respectively, and neutral mass-shifts for both ionization modes: 12C1 13C1, 12C2 13C2, H+ + Na+, H+ + K+, H2O, CO2, NH3, HPO3, H3PO4, + H2PO4Na, + H2PO4K, + HPO4Na2, + HPO4K2, + (H2PO4Na)2, + (H2PO4K)2, + (H2PO4)2NaH, and + (H2PO4)2KH. b All compounds contained within KEGG database for E. coli or within the genome-scale model by Feist et al.36
polarities with a very stringent m/z tolerance of 0.001 amu and a routine that combinatorially seeks for all relevant ions, gains, losses, and isotopes starting either with the genome-scale metabolic model36 or the KEGG repository38 for E. coli (Table 1). In about 5% of the cases, the mass tolerance was relaxed to 0.005 amu because the precision of the m/z measurement was worse than 0.001 amu. This can occur for low intensity ions in the higher m/z range. In total, in the two modes, we could detect 244 (27%) and 670 (21%) unique compounds from either the E. coli genome model or the KEGG repository, respectively. Given that isobaric compounds are counted individually although they cannot be separated, these numbers represent the best case scenario. We quantified the occurrence of these mass overlaps based on our experimentally measured mass resolution. For the genome model and KEGG, we obtained that 694 (for 884 compounds, 79%) and 2024 (for 3223 compounds, 63%) masses are resolvable, respectively. At nominal mass, the number of resolvable ions drops to 504 (57%) and 774 (24%), respectively. Notably, a change in a single metabolite within a cluster of isomers with identical molecular weight remains detectable if it is sufficient to globally affect the total ion count. Considering that not all metabolites listed in theoretical libraries are amenable to electrospray ionization, present in polar extracts, or sufficiently abundant, the experimental coverage positively matches our expectations despite the massive increase in throughput. The large fraction of ions that cannot be assigned to any metabolite in the reference list might be ascribed to background ions originating from solvents and gases. We experimentally assessed the origin of ions by growing isogenic E. coli strains on minimal medium containing either U 12C glucose or U 13C glucose. For each medium, triplicate extracts were generated and analyzed by flow injection TOF MS. We expected that all ions originating from glucose would exhibit a shift in mass equivalent to the number of carbon atoms contained. In total, 75% of the 1514 detectable ions exhibited a significant change in intensity, i.e., a fold change of >1.33 at a p-value of 2 at p-values of
