Virtual Chromatographic Resolution Enhancement in Cryoflow LC

Mar 30, 2007 - resulting in a “virtual” chromatographic resolution en- hancement visualized directly in the NMR spectral projec- tion. We demonstr...
0 downloads 0 Views 590KB Size
Download PDF
Anal. Chem. 2007, 79, 3304-3311

Virtual Chromatographic Resolution Enhancement in Cryoflow LC-NMR Experiments via Statistical Total Correlation Spectroscopy Olivier Cloarec,† Alison Campbell,† Li-hong Tseng,‡ Ulrich Braumann,‡ Manfred Spraul,‡ Graeme Scarfe,§ Richard Weaver,§ and Jeremy K. Nicholson*,†

Department of Biomolecular Medicine, Division of Surgery, Oncology, Reproductive Biology and Anaesthetics, Faculty of Medicine, Imperial College London, South Kensington, London SW7 2AZ, UK, Bruker BioSpin GmbH, Silberstreifen, D-76287 Rheinstetten, Germany, and Department of Drug Metabolism, Division of Pharmacokinetics & Metabolism, Servier Research & Development Ltd., Rowley, Wexham Springs, Framewood Road, Wexham, Slough, SL3 6PJ, UK

A new approach to enhancing information recovery from cryogenic probe “on-flow” LC-NMR spectroscopic analyses of complex biological mixtures is demonstrated using a variation on the statistical total correlation spectroscopy (STOCSY) method. Cryoflow probe technology enables sensitive and efficient NMR detection of metabolites onflow, and the rapid spectral scanning allows multiple spectra to be collected over chromatographic peaks containing several species with similar, but nonidentical, retention times. This enables 1H NMR signal connectivities between close-eluting metabolites to be identified resulting in a “virtual” chromatographic resolution enhancement visualized directly in the NMR spectral projection. We demonstrate the applicability of the approach for structure assignment of drug and endogenous metabolites in urine. This approach is of wide general applicability to any complex mixture analysis problem involving chromatographic peak overlap and with particular application in metabolomics and metabonomics. The rapid rise in popularity of metabonomics1-9 and metabolomfor sample classification and disease biomarker detection has led to the development of a novel range of analytical ics10-14

* To whom correspondence should be addressed. E-mail: [email protected]. † Imperial College London. ‡ Bruker BioSpin GmbH. § Servier Research & Development Ltd. (1) Nicholson, J. K.; Lindon, J.C.; Holmes, E. Xenobiotica 1999, 29 (11), 11819. (2) Nicholson, J. K.; Connelly, J.; Lindon J. C.; Holmes, E. Nat. Rev. Drug Discovery 2002 1 (2), 153-61. (3) Clayton, T. A.; Lindon, J. C.; Antti, H.; Charuel, C.; Hanton, G.; Provost, J. P.; Le Net, J. L.; Baker, D.; Walley, R. J.; Everett, J. E.; Nicholson, J. K. Nature 2006, 440 (20), 1073-7. (4) Robertson, D. G. Toxicol. Sci. 2005, 85 (2), 809-22. (5) Robertson, D. G.; Reily, M. D.; Baker, J. D. Expert Opin. Drug Metab. Toxicol. 2005, 1 (3), 363-76. (6) Brindle J. T.; Antti, H.; Holmes, E.; Tranter, G.; Nicholson J. K.; Bethel H. W. L.; Clarke, P. M.; Scofield, E.; McKillikghan D. E.; Mosedale, D.; Grainger, D. Nat. Med. 2002, 8 (12), 1439-44. (7) Keun, H. C. Pharmacol. Ther. 2006, 109 (1-2), 92-106. (8) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Pharm. Res. 2006, 23 (6), 107588. (9) Nobeli, I.; Thornton, J. M. Bioessays 2006, 28 (5), 534-45.

3304 Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

approaches to analyze complex biological mixtures. Both LCMS and LC-NMR methods have been extensively applied to compound structure elucidation, but a remaining and significant limitation in LC-NMR applications is its relatively low sensitivity compared with LC-MS; this is especially true for “on-flow” applications, where the number of scans per spectrum is typically limited to 4-16, which also limits the observed chromatographic resolution.15-18 In recent years, NMR detection limits have improved significantly with the advances in cryogenics and the introduction of cryoprobes, allowing more rapid on-flow detection rates.19,20 In cryo-NMR, the electronic components and transmitter-receiver cells are cooled to 25 K while vacuum isolation keeps the sample at ambient temperatures and in the liquid state. The electronic noise in the radio frequency system is typically reduced by a factor proportional to the square root of the temperature ratio (in K), thus resulting in an increase in the signal-to-noise ratio for cryoprobes over that of comparable conventional probe of ∼×4 for a single scan, giving a real time improvement of ∼×16 for multiscan experiments. This real-time improvement is particularly significant in LC-NMR applications, for example, in detection of low-level metabolites in biological samples. The greatest benefit of cryoprobes comes in the LC-NMR experiment where the mass of analyte in the flow probe is low and the analytes flow through the cell in less than 20 s, depending on exact flow rate and cell (10) Wang, Q. Z.; Wu, C. Y.; Chen, T.; Chen, X.; Zhao, X. M. Appl. Microbiol. Biotechnol. 2006, 70 (2), 151-61. (11) Shulaev, V. Brief Bioinformatics 2006, 7 (2), 128-39. (12) Hall, R. D. New Phytol. 2006, 169 (3), 453-68. (13) Wishart, D. S. Am. J. Transplant. 2005, 5 (12), 2814-20. (14) Weckwerth, W.; Morgenthal, K. Drug Discovery Today 2005, 10 (22), 15518. (15) Spraul, M.; Hofmann, M.; Wilson, I. D.; Lenz, E.; Nicholson, J. K.; Lindon, J. C. J. Pharm. Biomed. Anal. 1993, 11 (10), 1009-15. (16) Shockcor, J. P.; Unger, S. E.; Wilson, I. D.; Foxall, P. J.; Nicholson, J. K.; Lindon, J. C. Anal. Chem. 1996, 68 (24), 4431-5. (17) Lindon, J. C.; Nicholson, J. K.; Sidelmann, U. G.; Wilson, I. D. Drug Metab. Rev. 1997, 29 (3), 705-46. (18) Wilson, I. D.; Plumb, R.; Granger, J.; Major, H.; Williams, R.; Lenz, E. M. J. Chromatogr., B: Anal. Technol. Biomed. Life Sci. 2005, 817 (1), 67-76. (19) Spraul, M.; Freund, A. S.; Nast, R. E.; Withers, R. S.; Maas, W. E.; Corcoran, O. Anal. Chem. 2003, 75 (6), 1536-41. (20) Keun, H. C.; Beckonert, O.; Griffin, J. L.; Richter, C.; Moskau, D.; Lindon, J. C.; Nicholson, J. K. Anal. Chem. 2002, 74 (17), 4588-93. 10.1021/ac061928y CCC: $37.00

© 2007 American Chemical Society Published on Web 03/30/2007

Figure 1. Schematic of procedures for LC-NMR-STOCSY. (a) 1D NMR spectrum corresponding to the retention time t2 during the LC-NMR experiment. The star shows a triplet of a compound of interest. (b) Simulated LC-NMR pseudo-2D spectrum of a sample. On the side, the total UV chromatogram shows a strong overlap between three peaks. (c) NMR spectrum of the sample before separation (d) Result of a 1D STOCSY analysis driven from the center of the ″unknown″ triplet: a quartet is shown highly correlated with this indicating that they are signals from the same molecule. (e) Spectra of the pure compounds with confirmed STOCSY assignments.

Figure 2. LC-cryo-NMR pseudo-2D spectrum of a urine sample from a rat treated with thiabendazole. The spectrum contains information on both endogenous and drug metabolites, but many compounds coelute or overlap.

Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

3305

Figure 3. (a) 1D NMR spectrum and expansion collected in on-flow LC-NMR run corresponding to a retention time of 76 min and (b) chromatographic profiles of two peaks (δ 7.20 and 7.57) selected from the spectrum corrsponding to the retention time of 76 min.

volume.19,21 Using the cryoflow probes not only increases the signal-to-noise ratio but also enables rapid characterization of metabolites that would otherwise have been undetectable in onflow mode. Biological fluids are inherently complex with a diverse range of chemical classes present. Therefore, extensive chromatographic peak overlap is normal in LC-NMR and LC-MS studies, which can interfere with the interpretation of the spectra and subsequent structure elucidation. Curve resolution methods with complex algorithms are potentially powerful enough to try solving the

overlapping problem in LC-NMR profiles.22,23 However, we hypothesized that an elegant approach to “virtual” chromatographic resolution enhancement of partially overlapped peaks could involve using the property correlation between the various NMR signals of molecules as they sequentially enter and leave the flow cell. The different signal intensities from protons of different functional groups from each molecule should be highly correlated as the concentration varies across a chromatographic peak and by “back-projecting” the correlation coefficient (via a color coding) into a representative 1D LC-NMR spectrum; signals

(21) Lewis, R. J.; Bernstein, M. A.; Duncan, S. J.; Sleigh, C. J. Magn. Reson. Chem. 2005, 43 (9), 783-9.

(22) Wasim, M.; Brereton, R. G. J. Chromatogr., A 2005, 1096 (1-2), 2-15. (23) Wasim, M.; Brereton, R. G. J. Chem. Inf. Model 2006, 46 (3), 1143-53.

3306

Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

Figure 4. STOCSY spectra for the driver chemical shifts δ4.17 hippuric acid (A) and δ3.64 4-hydroxyphenylacetic acid (B) taken from on-flow LC-NMR corresponding to a retention time of 76 min.

for the same molecule could be highlighted in the presence of confounding signals from partially coeluting metabolites. We have previously shown the application of statistical total correlation spectroscopy (STOCSY)24 to complex mixture analysis using statistical covariation of peaks across multiple sample spectra. Previously, these variations were induced by the difference in concentration between the different samples. In this application, LC-cryo-NMR, these concentration variations are generated by the elution of the different molecules in function of time. This method uses the intrinsic correlation between the intensities of the various NMR signals on a given molecule present at varying concentration across a set of complex mixtures to highlight “statistical coupling” to other resonances from the same molecule and improve information recovery. Thus, considering each NMR acquisition in the LC-cryo-NMR run as a separate sample, the relative concentration of molecules will vary sequentially with the chromatographic run time; hence, the STOCSY methodology can be used to mathematically deconvolute the different molecular signals. In this way, virtual resolution enhancement is achieved via statistical selection of peaks with intramolecular connectivity and so facilitates the improved identification of the metabolites. Here we exemplify the application of STOCSY to LC-cryo-NMR data for the first time by solving the structures of selected (24) Cloarec, O.; Dumas, M. E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gauguier, D.; Lindon, J. C.; Holmes, E.; Nicholson, J. Anal. Chem. 2005, 77 (5), 1282-9.

chromatographically overlapped endogenous and drug metabolites in urine. METHODS Sample Collection and Preparation. Thiabendazole, purity >99%, was obtained from Sigma Aldrich. Six male Wistar rats (3 intact and 3 bile-cannulated) were acclimatized in metabolism cages for 24 h prior to dosing. Rats were permitted free access to food and water throughout the study, with bile-cannulated animals receiving water enriched with glucose. A 100-mg sample of thiabendazole was suspended in corn oil (10 mL) to give a suspension containing 10 mg/mL thiabendazole. The animals were dosed by oral gavage via a gastric tube (2.5 mL/kg), with the thiabendazole dose prepared as a suspension in corn oil, to give a nominal dose of 25 mg/kg. Urine was collected from all animals for the 24 h prior to dosing (control), 0-8, 8-24, and 24-48 h postdose into sterile containers, with these samples containing both drug-related and endogenous compounds and providing a suitable challenge to evaluate the HPLC-cryo-NMR-STOCSY approach to virtual chromatographic resolution enhancement. The samples were centrifuged prior to NMR analysis at 10 000 rpm for 10 min, with the supernatant being collected and centrifuged again under the same conditions. HPLC-cryo-NMR Hyphenation. Urine samples (100 µL) were injected onto a Synergi Hydro-RP C18 column (250 × 4.6 i.d. mm) (Phenomenex) with 4-µm particles. The chromatographic flow rate was 140 µL‚min-1, and the run time was 180.0 min, with Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

3307

Figure 5. 1D NMR spectrum corresponding to a retention time of 54 min showing the driver chemical resonance (a) with (b) being a 1D STOCSY spectrum for the driver chemical shift δ 7.13: the highlighted metabolite is phenyl glucuronide.

the separation monitored at 254 nm. The mobile phases used were 10 mM ammonium formate in D2O (solvent A) (Goss Scientific) and acetonitrile-d3 (solvent B) (Cambridge Isotope Laboratories). The starting conditions were 10% solvent B from 0 to 157 min a gradual increase from 10 to 50% solvent B, from 157.0 to 158.0 min 50 to 90% solvent B, from 158.0 to 168.0 min isocratic at 90% solvent B, and from 168.0 to 169.0 min a decrease back to the starting conditions (10% solvent B), continuing until 180.0 min. Continuous flow NMR data were acquired on a Bruker Avance 600, operating at 600.13 MHz, equipped with a 30-µL LC-CryoFit using a 5-mm TCI probe head. The chromatographic method and on-flow NMR were controlled by Bruker Hystar software. NMR spectra were acquired using 1D 1H WET solvent suppression (D2O and acetonitrile-d3) with shape pulse and C-13 decoupling in the f2 domain during WET and acquisition time, with an intermediate preparation scan into a second data set for LC gradient runs with updated shapes. The data were recorded continuously throughout the experiment with the spectra being acquired into 32K data points every 0.8 min. A pseudo-2D NMR spectrum was obtained for each sample with each slice corresponding to a 1D spectrum. Spectra were referenced to the residual acetonitrile peak at δ1.95. All the 1D NMR spectra were phased and baseline corrected individually using the standard Bruker software prior to STOCSY analysis. STOCSY. This approach to enhance information recovery from one-dimensional NMR spectra of several samples was developed 3308

Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

by Cloarec et al.24 and has already been successfully applied to a number of real metabolite biomarker identification problems.25 The principle of STOCSY is based on the fact that the intensities of the NMR signals of an individual molecule should be perfectly correlated in a series of samples when the concentration of the molecule varies. In order to display the correlation results in a way that respects the peak shape, the output of the STOCSY result is displayed using the method proposed by Cloarec et al.24 The covariance between the selected variable and all the other variables of the spectrum is plotted with a color code corresponding to the correlation between the selected variable and all the other variables of the spectrum. By this way, the shape of the different peaks is respected and the correlated peaks are highlighted by the projection of a color coding to show degree of correlation. RESULTS Simulated Example. Before applying the proposed methodology on a real biological sample, a simulated example is presented in order to present the application of STOCSY on LC-NMR. Simulated NMR spectra of three compounds (Figure 1e) have been used to compose a mixture with the corresponding NMR (25) Holmes, E.; Cloarec, O.; Nicholson, J. K. J. Proteome Res. 2006, 5 (6), 131320. (26) Cloarec, O.; Dumas, M. E.; Trygg, J.; Craig, A.; Barton, R. H.; Lindon, J. C.; Nicholson, J. K.; Holmes, E. Anal. Chem. 2005, 77 (2), 517-26.

Figure 6. (a) Standard 1H NMR spectra corresponding to the urine sample (top) and to a standard aqueous solution of thiabendazole (bottom [1 mM]). b) Magnified LC-cryo-NMR pseudo-2D spectrum of a urine sample for the chemical shift area of thiabendazole (a, b, and c are the peaks used to drive the subsequent 1D STOCSY analysis).

spectrum presented in Figure 1c. These compounds have different retention times, but their chromatographic profiles are still significantly overlapping (Figure 1b). For this reason, the NMR spectrum corresponding to the retention time t2 shows all the peak resonances from all the compounds (Figure 1a). In previous applications of STOCSY, the variations of the peak intensity were induced by the difference in concentration between different samples. Here, these concentration variations are generated by the elution of the different molecules in function of time (Figure 1b). From the NMR spectrum corresponding to a specific retention time, a peak is then selected. A time window is also defined around this retention time. The size of this window depends of the retention time resolution, but the whole corresponding chromatographic peak must be included in the window. The NMR spectra acquired in this time window were used to compute the covariance and correlation between the selected peak and all the other variables in the measured chemical shift interval. The representation of the covariance and correlation is shown in

Figure 1d. This plot allows the identification of the other peaks corresponding to the compound B. Endogenous Metabolites. A typical example of LC-cryo-NMR pseudo-2-D map from a dosed animal urine sample is presented in Figure 2. Many polar molecules are not retained on the C18 column under these chromatographic conditions and hence elute close to the solvent front (from 20 to 30 min). The residual signal from the partially suppressed solvent signals and formate peak change shifts slightly with the gradient variation as reported previously.27 Even though there is reasonable dispersion in the chromatographic time scale, many compounds still coelute and make the interpretation of the NMR spectral time slices and the identification of molecules more difficult. For instance, in Figure 3a, the NMR spectrum profile corresponding to a retention time of 76 min shows different resonances, with the chromatographic profiles (27) Spraul, M.; Hofmann, M.; Dvortsak, P.; Nicholson, J. K.; Wilson, I. D. Anal. Chem. 1993, 65 (4), 327-30.

Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

3309

Figure 7. Illustration of drug metabolite structure elucidation via LC-NMR-STOCSY. Standard 1D NMR and 1D STOCSY spectra for the identified metabolites of thiabendazole in a rat urine sample seen eluting at 59 min (5-hydroxythiabendazole (a and b); driver resonance at δ9.25) and the peak seen electing at 33.6 min (glucuronidated 5-hydroxythiabendazole (c and d) at δ9.204). Color scales identify degree of correlation between peaks on the same molecule highlighting the 5-hydroxythiabendazole metabolite in (b) and its glucuronide in (d).

of the resonances at δ7.2 and δ7.57 (Figure 3b) showing that at least two molecules were coeluting at this time point. A basic NMR assignment attributes these peaks to two aromatic spin systems, with one being a para disubstituted aromatic ring and the other being a monosubstituted aromatic ring, and to the formic acid present in the mobile phase. However, the exact assignment of one of the doublets was difficult because the relative intensity of the doublets and their coupling constants are very similar and the signals at δ3.64 and δ4.17 show no spin-spin couplings. The successive applications of the 1D-STOCSY method, described above, to both former singlets (δ3.64 and δ4.17) shows that both aromatic spin systems are part of two different molecules (Figure 4). The first molecule (Figure 4A) presents two triplets and one doublet (3J coupling ) 7.2 Hz) with the respective integral ratio 2:1:2 typical of a monosubstituted aromatic ring. The remaining two doublets are typical of a para disubstituted aromatic ring (Figure 4B). The monosubstituted and disubstituted aromatic 3310

Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

ring signals were shown to be correlated with the two singlets at δ4.19 and δ3.64, respectively. The separation of both molecules signals shows clear connectivities for hippuric acid and 4-hydroxyphenylacetic acid, which aids assignment. These molecules are indeed very well-known, and their identification would have been possible without the use of this methodology. However, this example has been used to show the potential of the method to separate the signal of coeluting molecules in a real sample using STOCSY. The following example illustrates how STOCSY can also be used to enhance the signal-to-noise ratio of spectra using its intrinsic averaging of the signal. In Figure 5, the NMR spectrum of the peak seen eluting at 54 min shows many different peaks with a low signal-to-noise ratio. Performing a STOCSY calculation based on the top computer point of the peak at δ7.13 not only revealed the resonances of phenyl glucuronide, but also attenuated

the background noise by averaging the noise while increasing the signal of the molecule. Example with Drug Metabolites. The 1H NMR spectrum of the exemplar drug, thiabendazole, acquired from a standard solution in methanol and the 1H NMR spectrum of the urine collected from a rat during the time interval of 0-8 h postdosing are shown on Figure 6a. A comparison of these spectra revealed that, if the thiabendazole is present in the urine sample, its signal will overlap with signals from other metabolites (both xenobiotic and endogenous), especially in the area around δ7. Figure 6b presents a magnified region of a LC-cryo-NMR pseudo-2D spectrum acquired for the same 0-8 h postdose urine sample. It shows three sets of resonances close to those from thiabendazole, but with different retention times. The respective concentrations of these metabolites are very different. The first compound (labeled c in Figure 6b) has the strongest resonance intensity and has a retention time of 59.0 min. The 1D NMR spectrum corresponding to this retention time is shown on Figure 7a. The application of STOCSY on the δ9.256 singlet revealed one other singlet at δ8.691, two doublets at δ7.849 and δ7.539 with the same coupling constant (J ) 7.2 Hz), and one other singlet at δ7.769, which led to the assignment of this set of resonances as one of the major metabolites of thiabendazole, 5-hydroxythiabendazole (Figure 7b). The second compound (labeled b) has a retention time of 45.8 min and has the lowest concentration. This compound has not been surely identified through STOCSY although it presents a resonance pattern very similar to 5-hydroxythiabendazole, but this molecule signal is strongly correlated to a doublet at δ1.178 (not shown). After application of STOCSY on the singlet at δ9.204 eluting after 33.6 min (the NMR spectrum corresponding to this retention time is shown on Figure 7c), the last compound (labeled a in Figure 6b) has been identified as glucuronidated 5-hydroxythiabendazole because it presents a resonance pattern similar to 5-hydroxythiabendazole but signals are also highly correlated to a structure typical of a glucuronide conjugation (doublet at δ5.221, doublet at δ4.122, and multiplet at δ3.615) (Figure 7d). Here again, the added value of STOCSY analysis on

LC-cryo-NMR data is demonstrated when compared with the 1D NMR spectrum corresponding to the retention times of each of the molecules identified. CONCLUSIONS We have shown that statistical correlation spectroscopy in combination with LC-cryo-NMR experiments can be used to provide virtual resolution enhancement in chromatography by statistically highlighting self-correlated proton resonances from compounds eluting in overlapped peaks. Cryoflow probe technology enables efficient NMR detection of metabolites on-flow, and the rapid spectral scanning allows multiple spectra to be collected over chromatographic peaks containing several species with similar but nonidentical retention times. The inherent correlation between the signals of the same molecule along the chromatographic profile can be used to isolate the signal of individual molecules and by this way to solve complicated overlap problems. Furthermore, STOCSY methods inherently improve observed sensitivity because multiple spectral sets contribute to the analysis, and hence, the absolute chromatographic requirement of optimized analyte resolution is reduced, which could lead to more robust and speedier method development and faster chromatographic separations. This approach is of wide general applicability to any complex mixture analysis problem involving hyphenation of chromatographic technique with spectroscopic or spectrometric method. ACKNOWLEDGMENT The authors thank Servier Research and Development Ltd., Slough, for the provision of the funding for A.C.’s studentship and the samples utilized in this work, the Wellcome Trust (BAIR) for funding O.C., and finally Bruker BioSpin GmbH for use of their HPLC-cryo-NMR system Received for review January 25, 2007.

October

11,

2006.

Accepted

AC061928Y

Analytical Chemistry, Vol. 79, No. 9, May 1, 2007

3311