Comparison of extraction solvents
There have been extensive reviews that compare various different extraction strategies for cellular metabolic analyses [27–29]. Here, we have focused on comparing the extraction efficiency of both polar and nonpolar metabolites with biphasic solvent MeOH/H2O/CHCl3 (1:1:2) in the BD method  to two other extraction solvents, MeOH/EtOH (1:1) and MeOH/EtOH/H2O (2:2:1). BD extraction is commonly used for comprehensive metabolomics to ensure full coverage of both polar and nonpolar metabolites by running both polar and nonpolar phases separately on HILIC and RPLC. Similar to the BD method, MeOH/EtOH and MeOH/EtOH/H2O were able to extract both polar and nonpolar metabolites, but with all metabolites present in one single phase. Therefore, the latter two extraction strategies, when compared to BD, were able to minimize sample handling and also shorten the analysis time by a factor of two while still maintaining comprehensive coverage of both polar and nonpolar metabolites.
The scalable extraction method was tested on a 100-μL S. meliloti (2 × 109) cell culture grown in M9 growth medium in a 96-well microtiter plate. A bead-beating technique was adopted instead of vortex mixing to ensure full cell disruption. Sonication, though allowing complete cell disruption, was not used in order to avoid overheating which may degrade thermally labile metabolites. All three extraction solvents were able to precipitate protein, DNA and RNA as well as providing good recovery for a wide range of both hydrophilic and hydrophobic metabolites.
Based on the methionine-d3, the recoveries for the polar fraction of BD (BD polar), MeOH/EtOH, and MeOH/EtOH/H2O after three extraction procedures were 77 ± 2 %, 59 ± 5 % and 79 ± 2 %, respectively, for a sextuplicate experiment. A much lower recovery was obtained for MeOH/EtOH when compared to the two other extraction methods. MeOH/EtOH failed to maintain a compact protein pellet during the extraction process; therefore, an extra centrifugation was required for MeOH/EtOH extraction to remove particulates in the sample extract. Moreover, based on the selected 24 endogenous metabolites in S. meliloti, the extra centrifugation step in MeOH/EtOH could also have caused lower extraction efficiencies when compared to MeOH/EtOH/H2O solvent (see Electronic Supplementary Material Fig. S4). Minimal sample handling and short-time extraction procedures are critical for large-scale metabolomic studies in order to achieve greater reproducibility and sensitivity and to prevent metabolite modification and degradation with time [14, 30]. In terms of ease of performance, the MeOH/EtOH/H2O extraction procedure outperformed MeOH/EtOH and the two-staged extraction procedure of BD for comprehensive polar and nonpolar metabolite analyses. Hence, the MeOH/EtOH/H2O was further optimized to achieve better extraction efficiencies and robustness.
The optimal numbers of extraction processes required in order to reach a minimal of 95 % extraction efficiency of the selected endogenous metabolites was determined by extracting a 100-μL S. meliloti cell culture using MeOH/EtOH/H2O seven times. Each extraction was performed in sextuplicate. The percentages of recoveries of 20 endogenous metabolites from S. meliloti at each extraction step were calculated by dividing the relative abundance of each individual metabolite at each step with its sum in all seven extractions (Fig. 1). Most metabolites showed greater than 80 % recovery upon the first extraction. Among them, N-acetyl-aspartic acid, adenosine monophosphate (AMP), methylhistidine and acetylcarnitine were entirely (100 %) recovered in the first extraction. The second extraction was able to recover the remaining 10–15 % for most of the metabolites. Metabolites such as γ-aminobutyric acid, adenine, adenosine and proline were persistent and were still detected after the seventh extraction. Therefore, a minimum of two MeOH/EtOH/H2O extraction steps were required to ensure at least 95 % extraction efficiency for the major endogenous metabolites. We recommend extracting cells three times to ensure great extraction efficiency and reproducibility.
The dissolution solvent has a significant impact on peak shapes in HILIC chromatography . The samples were concentrated in order to improve the detection limit. Sample volume was reduced to 50 μL by drying with a gentle stream of nitrogen, and the remaining solvent was primarily composed of water as it was the least volatile solvent in the MeOH/EtOH/H2O extracts. Water is not an appropriate dissolution solvent for HILIC gradients with a high percentage of ACN because it causes peak broadening and, consequently, reduces sensitivity . Therefore, samples were dried completely to remove all residual water and reconstituted in a solvent mix low in water to also minimize irreproducibility due to inconsistent sample volumes. We have adopted the use of 60 %v/v ACN/H2O to ensure adequate peak shape and sensitivity while still allowing full dissolution of the highly polar metabolites.
HILIC/MS for simultaneous detection of both polar and lipid metabolites
HILIC is typically used to separate polar compounds via hydrophilic partitioning mechanism. In 2010, HILIC was reported to be able to retain lipids, especially phospholipids, according to the polarity of the lipid heads [32, 33]. Therefore, since HILIC can simultaneously separate polar and lipid metabolites, it was selected as the chromatographic method for high-throughput comprehensive metabolomic analyses. RPLC is often used to retain and separate nonpolar analytes . Separation of polar compounds can also be achieved with RPLC with an ion-pairing agent in the mobile phase ; however, the ion-pairing reagents often lead to contamination in the MS instrument  and, therefore, were not preferred.
Unlike RPLC, small changes in pH and buffer ionic strength can often cause large retention deviations in HILIC . To improve reproducibility and minimize retention deviation for better retention time alignment using XCMS, consistent preparation of the mobile phase was critical. For large-scale comprehensive metabolomic analyses, all samples should be run using the same batch of mobile phases. The IS and RS spiked in each biological samples should also be used to correct retention time drift of metabolites when assigning metabolite identification based on the retention time of authentic standards. HILIC separations are less tolerant of fast gradients and require a longer equilibrium time compared to RPLC. Though the starting gradient at 98 % ACN was able to retain a greater amount of metabolite features, it required much longer equilibration time (more than 10 min) than starting at 95 % ACN (8–10 min). Running blanks and pooled samples at the beginning of the HILIC sequence is critical in order to condition the column to minimize variation in peak shape, retention time and ionization response.
HILIC is able to retain phospholipids or other polar lipids via an adsorption mechanism . Silica HILIC was chosen specifically for optimized phospholipid separation at low buffer strength instead of other commonly used zwitterionic or diol HILIC [32, 38]. The low buffer ionic strength of 10 mM ammonium acetate in mobile phase B allowed secondary interactions between the HILIC stationary phase and the polar lipid head groups via hydrogen bonding and electrostatic interaction. Therefore, our optimized HILIC gradient was able to separate phospholipids by classes based on their polar head group. Though each lipid class eluted within a very narrow time window (often within 1 min), there were still separations within each lipid class based on hydrophobicity (carbon chain length) and unsaturation (number of C=C bonds) (Fig. 2). The lipid class separation achieved with HILIC in combination with the sub-5-ppm mass accuracy attained with internal calibration was able to accurately identify lipid metabolites without running copious authentic standards. Isomers between phosphatidylcholine (PC) and phosphatidylethanolamine (PE) could be accurately identified because PCs and PEs were chromatographically separated. Figure 3 summarizes 2,125 metabolite features (after data reduction) detected in the intracellular extracts of murine macrophage. Different classes of phospholipids including phosphatidylglycerol, PCs, PEs and lyso-PCs were detected along with small polar metabolites such as nucleosides, amino acids and organic acids. The ability of HILIC to separate both polar and lipid compounds combined with our extraction methodology allowed simultaneous analyses of both polar and lipid metabolites for enhanced sample throughput.
Mass accuracy can be significantly improved by the usage of sodium formate as an internal calibrant which was formed by the endogenous sodium ions in the cell and the formic acid in mobile phase B. The presence of sodium formate adducts with retention times at 7.2 min was used for internal mass calibration in both ESI+ and ESI− modes which dramatically improved mass accuracies for all three cell types grown in different biological media (see Electronic Supplementary Material Table S2). The confidence of metabolite identification was improved significantly with the sub-5-ppm mass accuracy attained after internal calibration with sodium formate.
The endogenous sodium formate also caused minor ion suppression regardless of the biological matrix of interest (see Electronic Supplementary Material Fig. S5). The IS, phenylalanine-d8, eluted in the ion suppression region and was used to normalize peak areas of metabolite features eluting in the region to correct for varying degrees of ion suppression in different samples.
The current method was applied to the comprehensive metabolomic analyses of S. meliloti, S. intermedius and murine macrophages. Over a continuous 7-day injection series of S. meliloti extracts, the retention time deviation was less than 7 s in a 24-min LC run with approximately 260,000 metabolite features in 137 samples (1,900 features per sample over 137 samples in ESI+) analysed by XCMS using the centWave method (see Electronic Supplementary Material Fig. S6). The peak area deviations of IS were all below 10 %. The QC and sample data were analysed with PCA with QC samples clustered tightly in the centre of the score plot, indicating that instrumental variability was minimal. The optimized HILIC-TOF-MS method was highly robust and reproducible.
Metabolome coverage from the MeOH/EtOH/H2O extraction compared to the two fractions of the Bligh and Dyer method
Untargeted comprehensive large-scale metabolomics demands that the experimental method have high sample throughput, high robustness to sustain long LC sequence and excellent metabolome coverage. We propose using a MeOH/EtOH/H2O extraction in combination with HILIC-TOF-MS to encompass both polar and nonpolar metabolites in a single analysis. Compared to the conventionally used BD method in which polar and nonpolar fractions are analysed separately, the proposed method doubles the throughput and minimizes the sample handling time with comparable reproducibility. The metabolome coverage of the proposed method was compared to both of the polar and nonpolar fractions obtained using BD methods.
Traditionally, the BD polar fraction was run using HILIC, and the BD nonpolar fraction was run using RPLC [18, 19, 39]. However, in order to directly compare the extraction efficiency of nonpolar metabolites, the BD nonpolar fraction was also run using the same optimized HILIC method as used for the MeOH/EtOH/H2O extracted samples and BD polar extracts. Evaluating all three extract samples using the same LC method has also allowed us to compare metabolite features that were found in common between all three extracts. However, more features were expected when analysing the BD nonpolar fraction with RPLC in comparison to HILIC. Triacylglycerols, diacylglycerols and fatty acids, which were commonly analysed with RPLC, cannot be retained using HILIC, and were eluting in the unquantifiable dead volume with retention time below k
app′ 0.7. Gram-negative bacteria S. meliloti was used to compare the metabolome coverage and extraction efficiency of MeOH/EtOH/H2O to BD polar and BD nonpolar extracts.
The XCMS centWave algorithm in combination with CAMERA has deconvoluted a total of 3,378 metabolite features. All those features were present in at least one of the MeOH/EtOH/H2O, BD polar and BD nonpolar extracts. There were more features detected in the ESI- mode (1,900) compared to ESI+ mode (1,478). Metabolite features from solvent contamination, instrumentation noise and spiked IS and RS that were shared in the extracted samples and the standard mixture containing IS and RS were removed (unpaired heteroscedastic t test, p > 0.05 between all extracts and standard mixtures). Any features that were eluted in the dead volume with k
app′ < 0.7 were removed because they were unquantifiable due to severe ion suppression. The isotopic ions annotated by CAMERA were also removed along with ions associated with sodium formate clusters. After data reduction, a final list of 1,059 metabolite features was attained. The data reduction process was important to reduce the quantity of redundant data and false positives during statistical analyses.
Multivariate analysis using OPLS-DA revealed that all three types of extracts had unique metabolome profiles (Fig. 4a). The model was robust with Q
2(cum) = 0.936 and describes nearly all variables with R
X(cum) = 0.926 and R
Y(cum) = 0.986. All extracts from the MeOH/EtOH/H2O extraction were clustered in between the BD polar and BD nonpolar extracts, indicating shared metabolome profiles between MeOH/EtOH/H2O extracts and BD polar extracts as well as nonpolar extracts. Since MeOH/EtOH/H2O extracts were not centered in the OPLS-DA score plot, these extracts contained some unique metabolite features that were absent in the BD polar and BD nonpolar extracts.
Among the 1,059 detectable endogenous metabolite features of S. meliloti, 59.4, 53.9 and 92.2 % were detected in BD polar, BD nonpolar and MeOH/EtOH/H2O extracts, respectively (Fig. 4b). Of the 7.8 % of the detectable metabolome not covered by the MeOH/EtOH/H2O method, 2.9 % were only detected in BD polar, 4.5 % were only detected in BD nonpolar and 0.3 % were detected in both BD polar and nonpolar fractions. There were 34.7 % of the features shared between MeOH/EtOH/H2O and BD polar that were undetected in BD nonpolar; among them, polar metabolites such as amino acids, organic acids, sugar phosphates, nucleotides and nucleosides were detected. There were 27.6 % of the features shared between MeOH/EtOH/H2O and BD nonpolar that were not detected in BD polar; lipids such as phosphatidylglycerols (PGs), PEs, PCs and phosphatidylserines (PS) were among those that were identified. There were 8.4 % of the metabolite features that could only be detected in MeOH/EtOH/H2O extracts, and they ranged in polarity and m/z values. A few of the identified metabolites and their relative responses are shown in Fig. 4c, and the ionization responses of all metabolites in MeOH/EtOH/H2O were normalized to one as a reference. The MeOH/EtOH/H2O method had equivalent recoveries for vast majority of those endogenous polar and lipid metabolites as compared to the separately analysed BD polar or BD nonpolar extracts.
There were also 21.5 % of the features shared between MeOH/EtOH/H2O, BD polar and BD nonpolar extracts. Among those shared features, 59.6 % of the shared metabolite features were equally extracted using MeOH/EtOH/H2O in comparison to the most pronounced BD fractions based on unpaired heteroscedastic Student’s t test of p < 0.05 (Table 1). In Fig. 5, the ionization responses of some selected shared metabolite features were normalized against IS and the sum of ionization response of BD polar and BD nonpolar which was normalized to 1. The normalization was done under the assumption that BD polar and BD nonpolar in combination have a net 100 % recovery of all metabolites. MeOH/EtOH/H2O had lower extraction efficiency in 14.6 % of the shared features than at least one of the BD fractions (Fig. 5a). However, 15.8 % of the shared features had a greater recovery in MeOH/EtOH/H2O than both of the BD fractions (Fig. 5b). Many of those metabolite features were partially recovered in both BD polar and BD nonpolar, but were more efficiently recovered using MeOH/EtOH/H2O. If a conventional comprehensive metabolomic method was used with each of the BD polar and nonpolar fractions run on either HILIC or RP, then features that were partially extracted and present in both fractions would be considered as different metabolites and would be quantified separately and result in bias during multivariate analyses. Moreover, recovering these metabolites in full using MeOH/EtOH/H2O results in higher injected concentrations and facilitates their detection. Therefore, the 8.4 % of the metabolite features detected exclusively in the MeOH/EtOH/H2O method would likely be low abundant metabolites in the cells which were under a detection limit when partially recovered in either of the BD fractions, but detectable when more efficiently recovered in the MeOH/EtOH/H2O extraction.
Based on all these results, MeOH/EtOH/H2O in combination with HILIC-TOF-MS provides a very robust, high-throughput and comprehensive approach for cellular metabolomic analyses. The metabolomic coverage of MeOH/EtOH/H2O was comparable to the combined coverage of BD polar and BD nonpolar yet was twice as efficient in terms of data acquisition speed. The method was unbiased towards neither of the polar or nonpolar metabolites.