Development of a Rapid Microbore Metabolic Pro ﬁ ling Ultraperformance Liquid Chromatography − Mass Spectrometry Approach for High-Throughput Phenotyping Studies

: A rapid gradient microbore ultraperformance liquid chromatography − mass spectrometry (UPLC − MS) method has been developed to provide a high-throughput analytical platform for the metabolic phenotyping of urine from large sample cohorts. The rapid microbore metabolic pro ﬁ ling (RAMMP) approach was based on scaling a conventional reversed-phase UPLC − MS method for urinary pro ﬁ ling from 2.1 mm × 100 mm columns to 1 mm × 50 mm columns, increasing the linear velocity of the solvent, and decreasing the gradient time to provide an analysis time of 2.5 min/sample. Comparison showed that conventional UPLC − MS and rapid gradient approaches provided peak capacities of 150 and 50, respectively, with the conventional method detecting approximately 19 000 features compared to the ∼ 6 000 found using the rapid gradient method. Similar levels of repeatability were seen for both methods. Despite the reduced peak capacity and the reduction in ions detected, the RAMMP method was able to achieve similar levels of group discrimination as conventional UPLC − MS when applied to rat urine samples obtained from investigative studies on the e ﬀ ects of acute 2-bromophenol and chronic acetaminophen administration. When compared to a direct infusion MS method of similar analysis time the RAMMP method provided superior selectivity. The RAMMP approach provides a robust and sensitive method that is well suited to high-throughput metabonomic analysis of complex mixtures such as urine combined with a 5-fold reduction in analysis time compared with the conventional


* S Supporting Information
ABSTRACT: A rapid gradient microbore ultraperformance liquid chromatography−mass spectrometry (UPLC−MS) method has been developed to provide a high-throughput analytical platform for the metabolic phenotyping of urine from large sample cohorts. The rapid microbore metabolic profiling (RAMMP) approach was based on scaling a conventional reversed-phase UPLC−MS method for urinary profiling from 2.1 mm × 100 mm columns to 1 mm × 50 mm columns, increasing the linear velocity of the solvent, and decreasing the gradient time to provide an analysis time of 2.5 min/sample. Comparison showed that conventional UPLC−MS and rapid gradient approaches provided peak capacities of 150 and 50, respectively, with the conventional method detecting approximately 19 000 features compared to the ∼6 000 found using the rapid gradient method. Similar levels of repeatability were seen for both methods. Despite the reduced peak capacity and the reduction in ions detected, the RAMMP method was able to achieve similar levels of group discrimination as conventional UPLC−MS when applied to rat urine samples obtained from investigative studies on the effects of acute 2-bromophenol and chronic acetaminophen administration. When compared to a direct infusion MS method of similar analysis time the RAMMP method provided superior selectivity. The RAMMP approach provides a robust and sensitive method that is well suited to high-throughput metabonomic analysis of complex mixtures such as urine combined with a 5-fold reduction in analysis time compared with the conventional UPLC−MS method. I ncreasingly untargeted metabolic phenotyping (metabotyping), of the type performed in metabonomic/metabolomic studies is being applied to large scale investigations, often comprising several thousands of samples obtained in preclinical metabolism/toxicological, 1,2 clinical, and epidemiological investigations. 3−5 In pursuit of the analysis of these samples ultraperformance liquid chromatography−mass spectrometry (UPLC−MS) has become an indispensable tool for exploratory metabolic phenotyping 6,7 as it provides 3−5-fold increase in throughput compared to conventional HPLC−MS. However, even with the combination of sensitivity, reproducibility, and throughput provided by UPLC−MS, there remains an unresolved tension between the desire to have both rapid analysis and comprehensive global metabolite coverage. Currently UPLC analysis times of 10−20 min/sample are used to provide a compromise between throughput and the number of features detected. 8,9 Faster analyses can be achieved by eliminating the chromatographic separation using direct infusion/injection of samples into the MS (DIMS). 10−12 However, although data acquisition via these approaches is fast, DIMS data processing and interpretation is a timeconsuming and complicated step due to the presence of multiple components such as molecular ions, adducts, in-source fragments and multiply charged species, all present in the same spectrum. In addition, DI methods remain prone to ion suppression/enhancement effects and are unable to distinguish isomeric species. While DIMS is undoubtedly useful as a rapid diagnostic technique, LC−MS, particularly UPLC−MS, still provides a more comprehensive metabolic phenotyping tool. 13−15 A halfway house between the current UPLC−MS profiling methods and DIMS is to use short chromatographic analysis times, thereby ameliorating some of the disadvantages of DIMS, while accepting the loss of some metabolome coverage that inevitably results from reduced chromatographic resolution. Indeed, an early study on the metabolite profiling of rodent urine that compared short chromatographic separations with "conventional" UPLC−MS-based methods clearly showed the viability of this approach, maintaining class separation even with a reduction in the number of features detected. 16 Faced with the need for higher sample throughput, we have therefore reinvestigated the use of this type of rapid UPLC−MS, but in addition have combined the approach with microbore LC, which we have recently shown to combine the robustness required for large scale LC−MS-based metabotyping with enhanced sensitivity and reduced solvent usage. 17−19 This was undertaken with the aim of developing methodology that provides similar performance to conventional LC that, when applied to screening large batches of samples of rodent urine, also reduces analysis time, cost per sample, and environmental impact while providing the same results in terms of major biomarkers.
Here we evaluate the utility of a rapid microbore metabolic profiling (RAMMP) method based on the use of a microscale reversed-phase UPLC separation, coupled to a fast data acquisition rate high-resolution accurate QTOF MS. The samples were obtained from rats administered with either a single, acute dose of 2-bromophenol or chronic administration of acetaminophen (7 days) from studies that formed part of those undertaken by the Consortium of Metabonomic Toxicology (COMET). 20,21 The positive ion mode results of the RAMMP methodology are also compared with the outcomes obtained via both "conventional" UPLC−MS and chip-based nanoelectrospray DIMS analysis. A 120 μL aliquot of urine was mixed with 120 μL of water to dilute the salt concentration (before protein removal with acetonitrile (1:3 v/v). These samples were vortex mixed and left at −20°C overnight before centrifugation for 5 min at 15 000 g at 4°C. For analysis, 50 μL was removed and added to 200 μL water in 350 μL 96-well plates. A pooled quality control (QC) sample was prepared by combining 100 μL of each sample and diluting 1:4 with water. 8,22,23 The 96-well plates were stored at −20°C until analysis, and the plates were centrifuged again for 5 min at 700g before being placed into the autosampler at 4°C.
Chronic Acetaminophen Study. Urine samples were collected from four groups of male Sprague−Dawley (10 rats/group) rats dosed orally with acetaminophen (APAP) by gastric intubation once daily, for 7 days. The rats were treated with 200 mg/kg, 400 mg/kg, or 800 mg/kg in 0.2% carboxymethyl-cellulose and a control group was treated with 0.2% carboxymethylcellulose alone. The animals were housed in metabolic cages and urine was collected for metabonomic analysis at −16, 0, 8,24,32,48,56,72,96,80,104,120,144,168,192,216,240,264,288, and 312 h after administration. Samples were stored at −40°C prior to analysis.
For LC−MS, a 20 μL aliquot of each urine sample was mixed with 60 μL of methanol for protein removal (methanol was used for this study, rather than acetonitrile, because of occasional instances of phase separation that were observed when the latter was used). The samples were vortex mixed and left at −20°C overnight before centrifugation for 5 min at 15 000 g at 4°C. For analysis, 20 μL was removed and added to 180 μL of water in 250 μL 384-well plates. A pooled quality control (QC) sample, used for system conditioning and assessment of analytical performance was prepared by combining 5 μL of each sample and diluting 1:9 with water. 8,22,23 The well plates were centrifuged again for 5 min at 700g before being placed into the autosampler at 4°C.
For DIMS analysis 20 μL of both protein precipitated rat urine samples and the pooled QC sample were diluted with ultrapure water by a factor of 80. An aliquot of 50 μL of each diluted sample was pipetted, in a randomized order, into the 96 well-plates followed by 100 μL of ultrapure methanol/0.1% formic acid (v/v) to give a sample of 1:2 water−methanol. The sample plates were sealed with foil and then centrifuged at 1500g and 4°C for 10 min before DIMS analysis.
Liquid Chromatography−Mass Spectrometry (LC− MS) and Direct Infusion-Mass Spectrometry (DIMS). Liquid Chromatography. Liquid chromatographic analysis was performed on an Acquity I-class UPLC system, equipped with a binary solvent manager, sample manager, and column heater (Waters Corp., Milford, MA), interfaced with a Synapt G2-S HDMS mass spectrometer (Waters Corp., Wilmslow, U.K.). The chromatographic separations were performed on a HSS T3 1.8 μm stationary phase of either 2.1 mm × 100 mm for the conventional analysis or 1 mm × 50 mm for the RAMMP analysis (Waters Corp., Milford, MA). The chromatographic mobile phase was composed of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B). The column temperature was maintained at 40°C and linear gradient elution was performed at either 0.5 mL/min for the 2.1 mm × 100 mm column or 0.4 mL/min for the 1 mm × 50 mm column. For the conventional method, the starting composition was 1% B, held for 1.0 min before increasing to 15% at 3.0 min, 50% at 6.0 min, 95% at 9.0 min for a 1.0 min wash, and returning to 1% B for a 2 min re-equilibration step (total cycle time 12 min). A 2 μL injection of sample was performed using the flow through needle. For the RAMMP method, the starting For each study, prior to the analysis of the samples themselves, 50 consecutive injections (2 μL volume) of the pooled samples were made at the start of the chromatographic run to "condition" the column to ensure that the analytical system was fully equilibrated. 9,24 The QC sample was injected at the beginning of the analytical run and every 10 injections thereafter to monitor instrument stability. For use of 1 mm i.d. columns, minor system modifications were necessary, as described in detail elsewhere, 19 to reduce peak dispersion. Briefly, for the microbore separations, the standard outlet tubing i.d. (0.004 in.) was reduced to 0.0025 in. of minimum length and the ESI stainless steel capillary (125 μm i.d.) was replaced with a narrow bore variant (50 μm i.d.).
LC−Mass Spectrometry. Mass spectrometry for metabolic profiling was performed on a Synapt G2-S HDMS accurate mass instrument (Waters Corporation, Wilmslow, U.K.) operated with electrospray ionization (ESI) in positive (ESI +) ion mode. The capillary voltage was 1.0 kV, cone voltage was 25 V, source temperature was set at 120°C with a cone gas (nitrogen) flow rate of 50 L/h, a desolvation gas temperature of 600°C, and a nebulization gas (nitrogen) flow of 1000 L/h. The Synapt G2-S was operated in resolution (V optics) mode and was set to acquire data over the m/z range 50−1200 with an acquisition rate of 0.1 s for the conventional method or 0.05 s for the RAMMP method. All mass spectral data were collected in centroid mode using the MS e data acquisition 25 function to obtain fragmentation data simultaneously. In function one, a low collision energy (4 eV) was used and in the second function a high collision energy (ramp 15−45 eV) was used for fragmentation. For mass accuracy, leucine enkephalin (MW = 555.62) was used as a lock mass at a concentration of 200 pg/μL (in 50:50 CH 3 CN/H 2 O, 0.1% formic acid) infused at a flow rate of 20 μL/min via a lock spray interface. Lockmass scans were collected every 15 s and averaged over 3 scans to perform mass correction. The instrument was calibrated before analysis with 0.5 mM sodium formate solution. These data were collected using MassLynx V 4.1 software (Waters Corp., Manchester, U.K.).
Chip-Based Nanoelectrospray DIMS. Chip-based nanoelectrospray infusion analysis was performed using the TriVersa NanoMate system (Advion BioSciences, Ithaca, New York) coupled to a Waters Synapt G2-S (Waters MS Technologies, U.K.). The nanoelectrospray was created and maintained by applying 1.4 kV high voltage and 0.8 psi nitrogen flow controlled by ChipSoft software (version 8.3.1). The sample plate temperature was maintained at 4°C. The data were with an acquisition rate of 1 s over the mass range of 40−600 m/z in negative (ESI−) and positive (ESI+) ion modes with automatic polarity switch infusing 5 μL of sample. The MS e data independent acquisition function was used with the low collision energy of 4 eV and a high collision energy ramp 15−45 eV in the second function for fragmentation. The sampling cone voltage was set at 40 V, and the source offset at 80 V.
Total data acquisition time was of 40 s for each ionization mode (first ESI− and then ESI+) but the overall turnaround time for each sample was 2 min in order to let the instrument automatically switch the polarity and enable the voltage to settle before acquiring the data in the second ion mode. The data for the negative and positive mode were acquired in two separate files in the MassLynx software. The total time for the analysis of a 96 well-plate was 4.5 h.
Sodium formate solution was used to calibrate the mass spectrometer on a daily basis. The lock-mass function was turned off but data in both modes were recalibrated postacquisition by in-house software using reference signals of well-characterized endogenous metabolites present in all urine samples. MS/MS was performed on discriminant peaks in DIMS in resolution mode using a pooled urine sample. The optimal CID energy was selected for each peak between 10 and 30 eV. For the endogenous and drug-related metabolites, the spectra were compared to the metabolite fragmentation patterns available in online databases (HMDB 26 and Metlin 27 ) and to the spectra acquired in-house using DIMS for the standards in neat methanol containing 0.1% formic acid (v/v).
Data Analysis. The raw data obtained via UPLC−MS in positive ion mode were processed by Progenesis QI data analysis software (Nonlinear Dynamics, Newcastle, U.K.) for peak picking, alignment, and normalization, to produce peak intensities for retention time (RT) and m/z data pairs. In the case of DIMS, basic data visualization and quality control was achieved using MassLynx 4.1 software (Waters Corporation, U.K.). For DIMS data analysis, the raw data (ESI+) were  converted to the mzML format with ProteoWizard software 28 followed by processing using in-house scripts as described previously. 29 In both cases, further statistical analysis was performed on the resulting normalized peak intensities using SIMCA P 14.0 (Umetrics, Umea, Sweden).

■ RESULTS AND DISCUSSION
As shown previously, 16 the use of a relatively short chromatographic run is likely to result in some loss of metabolome coverage compared to the conventional 12 min UPLC−MSbased method routinely used by us for metabolic profiling of urine. 9 Here we have employed a 1 mm i.d. column format, 19 previously optimized to be equivalent to our generic 2.1 mm i.d. separation, 9 and combined it (ensuring appropriate scaling of chromatographic factors) with a reduction in column length to 50 mm, an increased linear velocity and a short 2.5 min gradient (with both 2.1 and 1.0 mm-based separations employing a gradient profile of 12 column volumes). This combination of column geometry, flow rate, and reduced run time thereby enables significant gains in both sensitivity and reduced solvent consumption compared to the existing "conventional" 2.1 and 1 mm methods to be obtained. In order to evaluate the utility of the RAMMP method and determine how it compared to the conventional UPLC−MS method, a subset of samples of rat urine from animals dosed with 2-bromophenol were analyzed using both approaches. Subsequent analysis of these data showed that, as would be expected, the use of the RAMMP method resulted in a decrease in the number of ions detected from 18 823 features seen using conventional UPLC−MS to 6 188. However, as shown in Figure 1, when principal component analysis (PCA) was performed, the higher throughput RAMMP analysis (shown in Figure 1B) gave a similar degree of discrimination between the dose groups as provided by the longer conventional method (shown in Figure 1A). So, although a higher number of detected metabolites may be seen as advantageous in metabolic phenotyping, with the chance of detecting potential biomarkers thereby increased, discrimination between groups may still be observed with fewer metabolites as seen when comparing parts A and B of Figure 1. In the case of both the conventional and RAMMP methods, the QCs clustered tightly in the center of the PCA plots indicating good analytical stability for both platforms.
Example chromatograms of the same urine sample, taken 24 h post-dose from a high dose (200 mg/kg) animal, are shown in Figure 2 to illustrate the two types of separation. The chromatographic peaks generated using the RAMMP method had an average width at the base of 1.5 s, which compares with the base width of 3.6 s generated by the conventional method. Clearly, however, the much shorter run time (2.5 vs 12 min) resulted in reduced peak capacity, going from ∼150 for the conventional method to 50 in the case of the RAMMP analysis, and this is reflected in the reduced number of features detected, which parallels the reduced peak capacity to a similar extent.
As the aim of the RAMMP method was to increase sample throughput, and thereby decrease analysis time for large sample numbers, the robustness and repeatability of the RAMMP assay was investigated by the repeated analysis of a single 96-well plate of rat urine samples from the 2-bromophenol study. This 96-well plate was profiled 5 times in a single analytical run and the pooled quality control (QC) sample, used to assess analytical variation throughout the analysis, analyzed every 10 injections. In addition, as has been discussed else-where 8,22,24,30,31 column "conditioning", whereby the LC−MS system is modified by the injection of matrix, was performed to stabilize retention times and signal intensities prior to the start of an analysis. Michopoulos et al. 31 have previously shown for human plasma that increasing the amount of matrix injected on column, and using a more rapid gradient for the conditioning phase, represents an efficient means of reducing the time required to achieve reasonable repeatability. Through the use of a rapid gradient, such as that investigated here, a large number of pooled matrix (QC) injections can be made to improve instrument stability without extending the length of the analysis significantly. Thus, to ensure that the best possible results were obtained, extensive column conditioning was employed before the analysis of the samples was commenced. In addition, in order to maximize system conditioning, following on from the work of Michopoulos et al., 31 the amount of sample injected on column was increased from the 1 μL of sample used for analysis to 2 μL, effectively doubling the number of conditioning injections. In this instance, 50 conditioning injections of the QC sample were performed prior to the commencement of sample analysis, with conditioning completed in ∼2 h. Sample analysis was then performed immediately after the conditioning step. The high degree of similarity of the chromatograms obtained for the last QC sample injected, from each of the 5 replicate analyses of the 96-well plate (shown in Supporting Information Figure S1) provides a good indication of the robustness of the assay for the ∼500 injections made over the course of the analysis (∼20 h). The repeatability of the technique is further exemplified in Figure 3 by the PCA plot illustrating the agreement between the replicate analysis (n = 5) of selected samples from the control animals in the 2bromophenol study (this repeatability is also demonstrated in Table S1 for a number of ions selected from both QC samples and selected control animals).
The QC samples are also useful in determining the reliability of a particular ion as a potential biomarker by examining the stability of factors such as retention time and response in each  of the QC injections throughout the analysis. Assessing the coefficient of variance (CV) of the peak area/height of a particular ion across the QC samples gives an indication of how repeatable it is likely to be if used as a potential biomarker (although, clearly features selected in this way require identification and reanalysis by a validated and specific method which is then applied to study samples to confirm their utility as actual, rather than potential, biomarkers). As we have noted elsewhere, the acceptance criteria for reliability in exploratory metabolic phenotyping are still evolving, but using the QC samples to calculate ion intensity CV throughout a run is now a widely adopted technique, with a CV filter of between 15 and 30% used to determine acceptable metabolite precision, depending on the rigor of analysis required. 32 Ion intensity stability is typically greater the more intense the ion (providing it is not saturating the detector) and those features that are subject to ion suppression/enhancement will show greater variation than those that are resolved from interfering matrix components. It might, therefore, be supposed that the conventional method offering, as it does, greater chromatographic resolution would also result in a higher proportion of stable features than the RAMMP method, where greater coelution could result in greater potential for ion suppression/ enhancement. Using the CV filter approach here to examine the    Figure S2 and Table S1). A similar trend is observed if lower % CV cutoff levels are used, with 51% and 40% of features detected by the conventional and the RAMMP methods, respectively, displaying <10% CV (Supporting Information Figure S2 and Table  S2). The reduction in repeatability for the RAMMP, compared to the conventional, method is therefore modest in comparison to the ∼5-fold increase in sample throughput. Another obvious question that follows on from the partial discrimination of dose and control groups by PCA of the data obtained from both RAMMP and conventional UPLC−MS methods is how similar the features driving the statistical separation are for each method. To determine whether the discriminating ions responsible for the PCA separation of the samples were the same, orthogonal projections to latent structures discrimination analysis (OPLS-DA) plots were generated for the control and high dose groups 96 h postdose.
The OPLS-DA loadings S-plots generated from these data analysis are shown in Figure 4, indicating the similarity in the ions responsible for the differences in these samples detected by both analytical platforms (see also Figure S3 and Table S3). This is further evidence to support the fact that even though some information is lost with the reduction of peak capacity, the features of significance, responsible for discrimination of the different sample groups, are conserved.
Method Applicability to Large Sample Sets. The application of conventional UPLC−MS-based methods toward increasingly large sample sets can require a significant amount of time to perform the analysis (several weeks). This often results in a need to split sample analysis across multiple blocks because of the need to perform instrument maintenance, etc. at regular intervals. The generation of multiple batches can lead to challenges when attempting to combine data obtained from different analytical LC−MS runs. Correcting these features to enable data to be combined can be difficult and timeconsuming, although statistical correction procedures have been proposed to deal with this problem, 33,34 often dependent on the use of the data from the QC samples to correct for drift.

Article
We were therefore interested in evaluating the use of the RAMMP method as a means of not only increasing sample throughput but also as a method for reducing, or even eliminating, the need for analysis of multiple batches of samples such as those derived from a multidose level, chronic acetaminophen (APAP) study in the rat. Consequently, the 680 urine samples obtained in this COMET acetaminophen study, together with QCs, were prepared in 384-well plate format (thereby enabling this number of samples to be stored in a conventional, two-plate autosampler). After column conditioning, as described previously, the study samples were analyzed in a single analytical run and the results illustrated in Figure 5. This PCA scores plot shows the QCs tightly clustered (perhaps as a result of the reduction in analytical drift obtained using extensive matrix conditioning) and clear differentiation of the various dose groups (for comparison the data from the conventional analysis of these samples in four batches is shown in Figure S4).
As illustrated previously, despite the reduction in peak capacity and feature detection resulting from the use of the 2.5 min rapid gradient, the discrimination between the different dose levels was comparable to that obtained with the conventional approach. However, the major advantage accruing from this methodology was that it was possible to analyze all of the samples as a single analytical run eliminating the need for batch correction. When the major ions discriminating the dose groups were identified they were found to correspond to acetaminophen itself together with its major sulfate and glucuronide metabolites and the N-acetylcysteinlyl (mercapturate) metabolite that results from the initial production of a glutathione conjugate from a reactive metabolite of the drug. The detection of large quantities of drug-related material using untargeted metabonomic profiling is not unusual in preclinical rodent toxicology studies and can be used to the advantage as a means for detecting and identifying drug metabolites in the toxicology species. Indeed this approach of using of LC−MSbased untargeted metabolic profiling for investigating xenobiotic metabolism, for both in vitro and in vivo investigations, is well accepted, (e.g., see ref 35 and review by Chen et al. 36 ). While acetaminophen and its major sulfate and glucuronide metabolites (but not the mercapturate) share a common ion (m/z 152.0712), corresponding to the parent drug itself, the short separation used still enabled them to be resolved in the RAMMP method. This is shown in Figure 6 where the separation of acetaminophen (m/z 152.0712), acetaminophen glucuronide (m/z 328.1032), and acetaminophen sulfate (m/z 232.0280) is illustrated for both the RAMMP ( Figure 6A) and conventional ( Figure 6B) UPLC−MS methods. The corresponding mass spectra are shown for APAP glucuronide ( Figure 6C), APAP sulfate ( Figure 6D), and APAP ( Figure  6E). The presence of an ion at m/z 152 in the mass spectra of both conjugates indicates a degree of decomposition to the parent compound in the ion source demonstrating the value of the chromatographic separation.
When the ions derived from acetaminophen and its metabolites were removed from the data, all the dose groups showed a much smaller degree of discrimination from the controls. This result indicates little disruption of the metabolic profiles of the test animals, irrespective of dose, and was consistent with the results from conventional methods of toxicity assessment (clinical chemistry and histopathology data not shown). While the study therefore clearly demonstrates that the doses of acetaminophen employed here were sub toxic, the detection of the mercapturate of APAP would, in the case of, e.g., a novel drug, be a trigger for further in-depth investigations. This is because of the association of mercapturates with the production of glutathione conjugates from reactive metabolites and, when detected during metabolic phenotyping experiments, these metabolites have been suggested as "biomarkers" indicative of compounds with the potential for causing toxicity. 37,38 Indeed, the utility of monitoring the mercapturic acids N-acetyl-S-(carbamoylethyl)-L-cysteine and N-acetyl-S-(1-carbamoyl-2-hydroxyethyl)-Lcysteine as biomarkers of the dietary exposure of humans to the genotoxic carcinogen acrylamide has been demonstrated. 39,40 Again, the QC samples were used to determine the CV of each feature and a 30% cutoff was used for further analysis, with 64% of the peaks from this RAMMP analysis fulfilling this criterion. PCA and OPLS-DA was performed between the control and various dose groups at each time point to determine the markers contributing to the class separation noted in the PCA plot ( Figure 5). An example is shown in Figure 7 for the control and high acetaminophen dose (800 mg/kg) rat urine samples 24 h post-dose, illustrating separation

Analytical Chemistry
Article of the groups by PCA and OPLS-DA scores plots and the OPLS-DA loadings S-plot to observe the major features responsible for the separation. The significant features highlighted correspond to the major acetaminophen metabolites (APAP-glucuronide, APAP-sulfate, APAP, and N-acetylcysteinyl (APAP-NAC), the properties of which, taken from the QC data obtained by the RAMMP method, are detailed in Table 1 (with the equivalent data from the conventional method also provided for comparison).
Comparison of RAMMP with DIMS. As discussed in the introduction, an obvious alternative to the RAMMP method for the rapid analysis of samples for metabolic phenotyping is DIMS and indeed, when DIMS was used for the metabolic phenotyping of a subset of the urine samples from the APAP study this technique was also able to discriminate between the various dose groups (see the Supporting Information Figure S5 for PCA scores plots). DIMS was also able to detect the APAPsulfate, glucuronide, and mercapturate metabolites as significant discriminating ions. However, the results from DIMS PCA look rather different to the RAMMP result for the same samples (in contrast to comparison of RAMMP with the conventional UPLC−MS methodology). A number of samples were highlighted as "outliers" by DIMS and inspection of their spectra showed the presence of several intense peaks, possibly a result of a high salt content in these rat urines, causing ion suppression. An example of a typical and an "outlier" spectrum are shown in Supporting Information Figure S6, demonstrating signal suppression of, e.g., creatinine, due to matrix components. Additionally, a DIMS spectrum acquired in positive ion mode for a typical high dose sample detected a relatively much lower amount of APAP itself (m/z = 152.072) compared to the same sample analyzed using RAMMP (Supporting Information Figure S7). The peaks of Na + and K + adducts were also seen to be more intense in the DIMS spectra compared to RAMMP, presumably due to the inherently high salt content of rodent urine. Desalting prior to DIMS may be necessary to remove such matrix components and reduce these effects, as has previously been described elsewhere, 41 but clearly introduces additional sample processing. A summary of the main discriminating ions from DIMS and RAMMP is given in Supporting Information Table S4, and while a difference in adduct formation was noted, the major metabolites of significance were found to be the same in both techniques (though it is unclear from the DIMS data if the ion corresponding to APAP itself at m/z 152.0712 resulted from the in source fragmentation of conjugates, as was seen in UPLC−MS, or the parent compound itself). DIMS showed similar repeatability to RAMMP with 85% of the features in the QC samples having a CV < 30% compared with 82% for LC− MS over the course of the analysis. These results demonstrate that, while DIMS can also be used as a rapid diagnostic tool for analyzing this type of sample there was, as would be expected, a loss of information related to some of the ions observed with a short separation step prior to MS that were shown to be important for group discrimination.
As metabolic phenotyping moves from small scale "proof-ofprinciple" studies to much larger scale biomarker discovery and patient stratification applications, the requirement for rapid and economical methods of analysis for large numbers of samples provides a strong driver for robust high-throughput methods. Combining this need with miniaturized systems that also reduce the amount of sample consumed and minimize environmental impact by lowering solvent consumption provides an added benefit. The pragmatic approach to analysis described here, using a microbore column, rapid gradient reversed-phase UPLC separation, and a fast data acquisition rate high-resolution accurate QTOF mass spectrometer for screening large batches of samples, resulted in a method offering comparable analysis time to DIMS techniques, where typical infusion times range from 2 to 5 min. 42,43 In addition, the RAMMP method provided the ability to resolve isomeric metabolites, as well as those that readily fragment to a common ion, and also reduced the risk of ion suppression and adduct formation. Clearly, a consequence of decreasing the analysis time in this way was a reduction in metabolome and xenometabolome coverage, which brings with it the potential to miss both potential biomarkers and xenobiotic metabolites. However, the use of this rapid gradient methodology as a screening approach for large sets of samples does not preclude further, more comprehensive, analyses being performed subsequently to better define the metabolomes and seek further potential biomarkers.

■ CONCLUSIONS
Microbore reversed-phase UPLC−MS performed using a rapid gradient provides a suitable analytical platform with which to perform high-throughput endogenous and xenobiotic metabolic phenotyping with a similar analysis time to DIMS but advantages in method selectivity. Compared to a conventional UPLC−MS method, the RAMMP approach of reducing column diameter and length combined with a short gradient provided a 5-fold reduction in analysis time and a 6-fold reduction in sample consumption, while maintaining the ability to detect critical discriminating features between different sample groups. The methodology was robust, enabling the analysis of over 700 urine samples and QCs as a single batch with good reproducibility and sensitivity. It can be concluded, therefore, that this rapid gradient UPLC−MS approach is well suited to the analysis of large numbers of urine samples while maintaining analytical integrity, reducing sample and solvent

Analytical Chemistry
Article requirements, and overcoming analytical drift associated with lengthy analysis times.
■ ASSOCIATED CONTENT

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.6b00038.

Notes
The authors declare no competing financial interest.

■ ACKNOWLEDGMENTS
The MRC-NIHR National Phenome Centre is supported by the UK Medical Research Council (in association with National Institute of Health Research (England)) Grant MC_PC_12025. The financial support of Bruker Biospin, Waters Corporation, Metabometrix LTD, and Imperial College is also gratefully acknowledged by the NPC. The views expressed are those of the authors and not necessarily those of the NHS, NIHR, or the Department of Health.