Anal. Chem. 2008, 80, 1073
Heteronuclear 19F-1H Statistical Total Correlation Spectroscopy as a Tool in Drug Metabolism: Study of Flucloxacillin Biotransformation Hector C. Keun,*,† Toby J. Athersuch,† Olaf Beckonert,† Yulan Wang,† Jasmina Saric,†,‡ John P. Shockcor,§ John C. Lindon,† Ian D. Wilson,| Elaine Holmes,† and Jeremy K. Nicholson*,†
Department of Biomolecular Medicine, Division of Surgery, Oncology, Reproductive Biology and Anaesthetics, Faculty of Medicine, Imperial College London, Sir Alexander Fleming Building, South Kensington Campus, London SW7 2AZ, U.K., Department of Public Health and Epidemiology, Swiss Tropical Institute, P.O. Box CH-4002, Basel, Switzerland, Waters Corporation, 34 Maple Street, Milford, Massachusetts 01757, and AstraZeneca, Department of Drug Metabolism and Pharmacokinetics, Mereside, Alderley Park, Macclesfield, Cheshire SK10 4TG, U.K.
We present a novel application of the heteronuclear statistical total correlation spectroscopy (HET-STOCSY) approach utilizing statistical correlation between onedimensional 19F/1H NMR spectroscopic data sets collected in parallel to study drug metabolism. Parallel onedimensional (1D) 800 MHz 1H and 753 MHz 19F{1H} spectra (n ) 21) were obtained on urine samples collected from volunteers (n ) 6) at various intervals up to 24 h after oral dosing with 500 mg of flucloxacillin. A variety of statistical relationships between and within the spectroscopic datasets were explored without significant loss of the typically high 1D spectral resolution, generating 1H-1H STOCSY plots, and novel 19F-1H HET-STOCSY, 19F-19F STOCSY, and 19F-edited 1H-1H STOCSY (XSTOCSY) spectroscopic maps, with a resolution of ∼0.8 Hz/pt for both nuclei. The efficient statistical editing provided by these methods readily allowed the collection of drug metabolic data and assisted structure elucidation. This approach is of general applicability for studying the metabolism of other fluorine-containing drugs, including important anticancer agents such as 5-fluorouracil and flutamide, and is extendable to any drug metabolism study where there is a spin-active X-nucleus (e.g., 13C, 15N, 31P) label present.
largely absent from the human metabolome, and it has been shown that 19F NMR spectroscopy of biofluids (e.g., urine) provides an efficient method of monitoring fluorinated drug metabolism, as it is a highly sensitive (83% of 1H), I ) 1/2 nucleus present at 100% natural abundance. 19F NMR has been used to aid pharmacokinetic and pharmacodynamic studies of several medically important fluorinated drugs including chemotherapeutic antineoplastic fluoropyrimidines, 7,8 the nonsteroidal anti-inflammatory agent niflumic acid,9 and the antipsychotic drug haloperidol.10 Importantly, such studies translate well into a clinical setting as 19F magnetic resonance imaging allows the noninvasive monitoring of drug metabolism and pharmacokinetics in vivo.11,12 Resonances in 19F NMR spectra are usually well resolved at high frequency because the large chemical shift range renders the 19F resonance frequency exquisitely sensitive to structural change yet they typically provide little structural information that can be directly used for metabolite identification or characterization.13-15 In contrast, 1H NMR spectra are typically complex and contain a great deal of structural information. However, positively differentiating metabolite resonances in 1H spectra, especially those arising from xenobiotics, can be difficult due to the presence of dominant, overlapping resonances from endogenous compounds.
NMR spectroscopy has been widely applied in drug metabolism research using a range of spin-active (I ) 1/2) nuclear magnetic probes.1-6 Organic molecules containing fluorine are
(6) Foxall, P. J. D.; Lenz, E. M.; Neild, G. H.; Lindon, J. C.; Wilson, I. D.; Nicholson, J. K. Ther. Drug Monit. 1996, 18, 498-505. (7) Malet-Martino, M.; Bernadou J.; Martino, R.; Armand, J. P. Drug Metab. Dispos. 1988, 16, 78-84. (8) Desmoulin, F.; Gilard, V.; Malet-Martino, M.; Martino, R. Drug Metab. Dispos. 2002, 30, 1221-1229. (9) Kitamura, K.; Omran, A.; Takegami, S.; Tanaka, R.; Kitade, T. Anal. Bioanal. Chem. 2007, 387, 2843-2848. (10) Shamsipur, M.; Shafiee-Dastgerdi, L.; Talebpour, Z.; Haghgoo, S. J. Pharm. Biomed. Anal. 2007, 43, 1116-1121. (11) Wolf, W.; Presant, C. A.; Waluch, V. Adv. Drug Delivery Rev. 2000, 41, 55-74. (12) Malet-Martino, M.; Gilard, V.; Desmoulin F.; Martino, R. Clin. Chim. Acta 2006, 366, 61-73. (13) Wade, K. E.; Wilson, I. D.; Troke, J. E.; Nicholson, J. K. J. Pharm. Biomed. Anal. 1990, 8, 401-410. (14) Ghauri, F. Y. K.; Blackledge, C. A.; Wilson, I. D.; Beddell, C.; Lindon, J. C.; Glen, R.; Nicholson, J. K. Biochem. Pharmacol. 1992, 44, 1935-1946. (15) Spraul, M.; Hoffman, M.; Wilson, I. D.; Lenz, E.; Nicholson, J. K.; Lindon, J. C. J. Pharm. Biomed. 1993, 11, 1009-1016.
* To whom correspondence should be addressed. E-mail: j.nicholson@ imperial.ac.uk (J.K.N),
[email protected] (H.C.K.). † Imperial College London. ‡ Swiss Tropical Institute. § Waters Corporation. | AstraZeneca. (1) Nicholson, J. K.; Wilson, I. D. Prog. Nucl. Mag. Res. Spectrosc. 1989, 21, 449-501. (2) Bales, J. R.; Sadler, P. J.; Nicholson, J. K.; Timbrell, J. A. Clin. Chem. 1984, 30, 1631-1636. (3) Bales, J. R.; Nicholson, J. K.; Sadler, P. J. Clin. Chem. 1985, 31, 757-762. (4) Preece, N.; Nicholson, J. K.; Timbrell, J. A. Biochem. Pharmacol. 1991, 41, 1319-1324. (5) Nicholls, A.; Caddick, S.; Wilson, I. D.; Farrant, R. D.; Lindon, J. C.; Nicholson, J. K. Biochem. Pharmacol. 1995, 49, 1155-1164. 10.1021/ac702040d CCC: $40.75 Published on Web 01/23/2008
© 2008 American Chemical Society
Analytical Chemistry, Vol. 80, No. 4, February 15, 2008 1073
1H
NMR and other X-nuclei including 19F have been widely used in drug metabolism studies and are of considerable interest because of the conferred ability to study drug metabolism without the use of radiolabels.1 Drug metabolite resolution has been improved by the use of two-dimensional (2D) NMR methods (both homonuclear and heteronuclear)3,13 and by LC-NMR11,16 and LCNMR-MS17 approaches; however, these methods can be highly time-consuming and, in some cases, require specialized instrumentation. We have shown previously that information recovery from spectroscopic data can be enhanced by the use of statistical total correlation spectroscopy (STOCSY), which exploits correlation between the intensities of spectral features over multiple spectra in order to discern either physical or biological relationships.18 In particular, features from the same molecule will exhibit strong positive intensity correlations and this information can be exploited in spectral/structural assignments. This technique is of particular value in metabolic profiling for several reasons. First, such studies typically generate a substantial number of qualitatively similar but quantitatively different spectra for analysis, i.e., those data needed to detect structural correlations and reduce spurious ones. In addition, STOCSY can retain the full spectral resolution at which the original one-dimensional (1D) spectra were recorded, whereas conventional 2D correlation NMR spectroscopy has many practical limitations that prevent the indirect dimension reaching the resolution of a directly acquired spectrum. This is an important advantage, as there is often significant crowding and resonance overlap in biofluid and tissue NMR spectra. As we are also often interested in characterizing compounds in relatively low concentration compared with the major endogenous metabolites, we can exploit the better sensitivity of 1D NMR over multidimensional NMR, especially when looking for long-range couplings. STOCSY, unlike conventional NMR correlation experiments, which are limited by the strength of the scalar couplings or internuclear proximity, clearly does not suffer any reduction in sensitivity to structural correlations as the distance between the relevant nuclei increases. A recent development has been the integration of heteronuclear NMR spectra (HET-STOCSY),19 a technique which is applicable to paired spectra following parallel or serial acquisition. Using parallel 31P{1H} and 1H magic angle spinning (MAS) NMR data collection it was possible to interrogate the metabolism of phosphorus-containing compounds in intact liver samples and make structural assignments based on statistical correlations between the data sets.19 STOCSY has previously been used to investigate correlations within a variety of 1H NMR datasets including urine spectra from different mouse strains,18 rats exposed to toxins,20 drug metabolites (16) Sidelmann, U.; Braumann, U.; Hoffmann, M.; Spraul, M.; Hofmann, M.; Lindon, J. C.; Nicholson, J. K.; Hansen, S. H. Anal. Chem. 1997, 69, 607612. (17) Shockcor, J.; Unger, S.; Wilson, I. D.; Foxall, P. J. D.; Nicholson, J. K.; Lindon, J. C. Anal. Chem. 1996, 68, 4431-4435. (18) Cloarec, O.; Dumas, M.-E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gaugier, D.; Lindon, J. C.; Holmes, E.; Nicholson, J. Anal. Chem. 2005, 77, 1282-1289. (19) Coen, M.; Hong, Y-S.; Cloarec, O.; Rhode, C.; Reily M. D.; Robertson, D.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Anal. Chem. 2007, 79, 89568966. (20) Holmes, E.; Cloarec, O.; Nicholson, J. K. J. Proteome Res. 2006, 5, 13131320.
1074
Analytical Chemistry, Vol. 80, No. 4, February 15, 2008
in urines from human epidemiological studies,21 from urine processed by liquid chromatography with coupled NMR spectroscopy,22 and diffusion-edited spectra from human urine and plasma.23 The method has also been adapted to identify signals arising from the same molecule across different types of analytical platforms, e.g., LC-MS and 1H NMR (statistical heterospectroscopy, SHY24). We have now used a HET-STOCSY approach with 1H and X-nucleus (19F) datasets, recorded in parallel, to assign the metabolites of the fluorinated antibiotic flucloxacillin, the metabolism of which has been previously investigated in humans by HPLC25,26 and 1H/19F NMR spectroscopy.27,28 This study is the first demonstration of 19F and 1H NMR spectra being directly integrated by correlation analysis for drug metabolite analysis and shows the novel use of statistically based heteronuclear editing of homonuclear correlation spectra to give a statistical equivalent of the three-dimensional (3D) HSQCTOCSY NMR experiment.29 We illustrate how 19F-1H HETSTOCSY allows the spectroscopist to exploit the resolution and sparsity of 19F NMR spectra in the exploration of complex, but information-rich 1H NMR metabolic profiles. We also show how intermetabolite correlations aid the interpretation of STOCSY data and can assist the structural characterization of drug metabolites by providing information on the routes of biotransformation exploited in vivo. EXPERIMENTAL SECTION Sample Generation. Flucloxacillin was obtained from Generics [U.K.], Ltd. (Herts. U.K.) and used in the supplied tablet formulation taken as a 500 mg oral dose by six human volunteers. Urine samples were collected just prior to administration of the dose and then intermittently over a 24 h period. On collection, samples were immediately frozen and stored at -40 °C. Typically, four urine samples were collected from each volunteer during the course of the study. Sample Preparation. Samples were thawed at ambient room temperature and vortex-mixed prior to preparation for NMR analysis. Urine (300 µL) was added to 300 µL of phosphate buffer (0.2 M, pH 7.4) containing 1 mM 3-(trimethylsilyl)-[2,2,3,3-2H4]propionic acid sodium salt (TSP), 3 mM sodium azide, and 20% (v/v) D2O. Samples were vortex-mixed and centrifuged (16 000 g, 10 min, RT) to remove suspended solids before analysis by NMR. NMR Acquisition and Processing. NMR spectra were acquired using a DRX 800 US2 NMR spectrometer (Bruker (21) Holmes, E.; Loo, R. L.; Cloarec, O.; Coen, M.; Tang, H.; Maibaum, E.; Bruce, S.; Chan, Q.; Elliot, P.; Stamler, J.; Wilson, I. D.; Lindon, J. C.; Nicholson, J. K. Anal. Chem. 2007, 79, 2629-2640. (22) Cloarec, O.; Campbell, A.; Tseng, L. H.; Braumann, U.; Spraul, M.; Scarfe, G.; Weaver, R.; Nicholson, J. K. Anal. Chem. 2007, 79, 3304-3311. (23) Smith, L. M.; Maher, A. D.; Cloarec, O.; Rantalainen, M.; Tang, H.; Elliot, P.; Stamler, J.; Lindon, J. C.; Holmes, E.; Nicholson, J. K. Anal. Chem. 2007, 79, 5682-5689. (24) Crockford, D. J.; Holmes, E.; Lindon, J. C.; Plumb, R. S.; Zirah, S.; Bruce, S. J.; Rainville, P.; Stumpf, C. L.; Nicholson, J. K. Anal. Chem. 2006, 78, 363371. (25) Murai, Y.; Nakagawa, T.; Yamaoka, K.; Uno, T. Int. J. Pharm. 1983, 15, 309-320. (26) Thjissen, H. H. W. J. Chromatogr. 1980, 183, 339-345. (27) Everett, J. R.; Jennings, K.; Woodnutt, G. J. Pharm. Pharmacol. 1985, 37, 869-873. (28) Everett, J. R.; Tyler, J. W.; Woodnutt, G. J. Pharm. Biomed. Anal. 1989, 7, 397-403. (29) Krishnamurthy, V. V. J. Magn. Reson. 1995, 106, 170-177.
Figure 1. (A) Overlay of urine 19F NMR spectra obtained from control (black) and flucloxacillin-exposed individuals (red). I ) flucloxacillin, II ) (5R)-flucloxacillin penicilloic acid, III ) 5′-hydroxymethylflucloxacillin, IV ) (5S)-flucloxacillin penicilloic acid, V ) unknown. (B) Overlay of urine 1H NMR spectra obtained from control (black) and flucloxacillin-exposed individuals (red). Resonances relating to the gem-methyl protons of flucloxacillin and metabolites are clearly observable in the post-dose spectrum.
BioSpin, Rheinstetten, Germany) operating at 18.81 T (800.32 MHz 1H frequency, 753.05 MHz 19F frequency), equipped with a 5 mm SEI 800S6 H-F-D-05 dual channel probe at 298 K. TOPSPIN (version 2.0.a, Bruker BioSpin) was used for spectrometer control. 1H NMR Spectroscopy. 1H spectra were acquired using a 1D pulse sequence (RD-90°-t1-90°-tm-90°-AQ), with 64 scans being collected into 64 K data points, following 8 dummy scans (t1 ) 3 µs). The acquisition (AQ) time was 3.36 s. Presaturation of the water signal was applied during a relaxation delay (RD) of 2 s and the mixing time (tm) of 100 ms. The spectral width was 12.2 ppm (9766 Hz), giving the FID resolution of 0.149 Hz. An 0.3 Hz exponential line broadening function was applied to the FID prior to Fourier transformation. Phase and baseline correction were conducted manually using TOPSPIN (version 2.0.a., Bruker BioSpin). Spectra were referenced to the TSP resonance (δH ) 0.00 ppm). 19F NMR Spectroscopy. 1H-decoupled 19F (19F{1H}) spectra (n ) 21) were acquired using a 1D pulse sequence (RD-90°-AQ), with 64 scans being collected into 8 K data points, following 8 dummy scans. The AQ time was 1.09 s, during which a WALTZ16 decoupling scheme was applied at the 1H resonance frequency to eliminate the effects of heteronuclear 19F-1H scalar couplings and to generate a singlet at each 19F resonance. A 10 s relaxation delay (RD) was used to allow relaxation of 19F spins to equilibrium between scans. Backward linear prediction of the first 64 data points of the FID was conducted using 100 coefficients to eliminate low-frequency baseline effects in the spectra (TOPSPIN version 2.0.a., Bruker BioSpin). Spectra were referenced to the flucloxacillin parent peak at ∂F ) -110.36 ppm (from CFCl3). Statistical Analysis. All NMR data were imported into the MATLAB computing environment (R2007a, The MathWorks, Inc., MA) using MetaSpectra 4.0 (in-house MATLAB routine written by Dr. O. Cloarec) with spectra reconfigured to a common ppmscale by spline interpolation to 11 000 data points (1H, δH )
Figure 2. 1D 19F-1H HET-STOCSY (driven by the peak at ∂F ) -110.36 ppm)si.e., correlation between 1H NMR data and the 19F intensity for flucloxacillin (parent, I), highlighting all the 1H resonances of flucloxacillin. The correlation coefficient (r) is superimposed on each data point of the median spectrum using the color scheme indicated. Correlation cutoff set at r > 0.5 for display purposes.
Analytical Chemistry, Vol. 80, No. 4, February 15, 2008
1075
Table 1. 1H and
19F
NMR Chemical Shifts (δ in ppm) for Compounds I-V in Human Urine
compound number
compound name
δF
δH gem-CH3
I II III IV V
flucloxacillin (5R)-flucloxacillin penicilloic acid 5′-hydroxymethylflucloxacillin (5S)-flucloxacillin penicilloic acid 5′-hydroxymethyl-(5R)-flucloxacillin penicilloic acida
-110.36 -110.41 -110.51 -110.07 -110.82a
1.48, 1.52 1.22, 1.54 1.49, 1.54 1.12, 1.58 1.12a, 1.59a
aProposed
from
19F-1H
HET-STOCSY analysis.
Single linkage analysis was applied using Euclidean distance as a metric for assessing similarity.
Figure 3. Distinguishing between flucloxacillin and one of its metabolites by 1D 19F-1H HET-STOCSY with back-projected 19F correlations onto the median 1H NMR spectrum (colored by the Pearson correlation (r) to the 19F peak of interest). (A) Correlations to 19F peak of metabolite IV (δF ) -110.07 ppm). (B) Correlations to 19F peak of parent (I, δ ) -110.36 ppm). Resonances relating to F the gem-methyl protons of the respective compounds are indicated with arrows.
-1 to 10 ppm) and 4500 data points (19F, δF ) -113 to -108.5 ppm). This procedure resulted in a final spectral resolution of 0.8 Hz/pt and 0.75 Hz/pt for 1H and 19F, respectively. Of the 26 samples analyzed, 5 were rejected from statistical analysis due to significant water suppression artifacts in the spectra obtained. Normalization of spectral intensity values was only performed for the generation of the homonuclear 1H correlation matrix using the method described by Dieterle et al.30 Hierarchical clustering was conducted on the 19F peak intensities correlation matrix using an in-house MATLAB routine written by Dr. Timothy Ebbels. (30) Dieterle, F.; Ross, A.; Schlotterbeck, G.; Senn, H. Anal. Chem. 2006, 78, 4281-4290.
1076 Analytical Chemistry, Vol. 80, No. 4, February 15, 2008
RESULTS AND DISCUSSION In a previous investigation of flucloxacillin metabolism in the rat, administered at high doses (200 mg kg-1), 19F{1H} NMR spectroscopy at 9.4 T (400 MHz 1H resonance frequency, 376 MHz 19F resonance frequency) indicated the presence of three major detectable metabolites: 5′-hydroxymethylflucloxacillin (III) and both 5R (II) and 5S (IV) epimers of the β-lactam ring-opened penicilloic acid.27,28 In our investigation in man, using a single therapeutic dose, urine samples were measured at 18.81 T (800.32 MHz 1H frequency, 753.05 MHz 19F frequency) and all of these metabolites were readily observed and could be tentatively assigned by the relative frequency of the resonances to that of the parent compound (I, Figure 1A) and using literature values.27 In addition, one other low-frequency resonance from an unknown metabolite (V) was clearly visible. From an overlay of the corresponding 1H spectra from each urine sample, it was less straightforward to identify flucloxacillin-related resonances among the large number of endogenous features, with the exception of the gem-methyl signals (Figure 1B). 1D Heteronuclear Correlation Analysis. By displaying the correlation coefficient between the intensity of the parent 19F resonance (I) with the values of each data point in the 1H spectrum (1D correlation analysis), all resonances of flucloxacillin could be clearly identified, including resonances from protons up to 13 bonds from the fluorine atom (Figure 2). This is in contrast to previous efforts using 1H-19F COSY experiments28 which could only detect cross-peaks between nuclei on the aromatic ring of the parent compound, where there were significant scalar couplings of JHF > 2 Hz. The detected resonances were assigned on the basis of characteristic chemical shift and literature values,27 and a summary of the 19F and methyl-region 1H chemical shifts is given in Table 1. In addition, features characteristic of metabolites of flucloxacillin, such as the appearance of extra, strongly correlated methyl resonances were also highlighted. The profile of correlations produced was dependent on the 19F resonance selected to generate the 1D correlations. For example, the correlation profile of (5S)-flucloxacillin penicilloic acid (IV) highlighted resonances originating from the parent and other metabolites, but indicated that among the highest correlations (r > 0.9) were those to the gem-methyl resonances specific to IV (Figure 3A). By contrast, correlation to the parent compound did not highlight these resonances at all (Figure 3B), indicating that the underlying correlation structure of metabolite excretion is sufficiently modulated to allow 1H signals to be differentiated.
Figure 4. 2D 19F-1H HET-STOCSY analysis. Expansion of 19F NMR, 1H NMR, and 19F-1H HET-STOCSY plots showing fine detail of correlations between spectra, determined using the correlation between peak intensities in the two sets of spectra. Correlation cutoff set at r ) 0.8. Contours are colored according to the Pearson correlation (r) of the 19F and 1H signal intensities. I ) flucloxacillin and III ) 5′-hydroxymethylflucloxacillin.
Figure 5. Pseudo-2D 19F-edited 1H-1H X-STOCSY correlation spectra. (A) unnormalized data; (B) normalized data; (C) normalized data edited by 19F peak I (parent, δF ) -110.36 ppm); (D) normalized data edited by peak IV (δF ) -110.07 ppm). The process of generating an X-STOCSY data matrix removes all signals from regions with r < 0.5 to the 19F peak.
2D Heteronuclear 19F-1H Analysis (HET-STOCSY). In addition to peak-selective 1D correlation analysis, we explored ways of fully integrating both 19F{1H} and 1H datasets via statistical correlation. The most direct approach was to calculate the correlation coefficient between each individual spectral point in the 1H dataset with those in the 19F dataset, i.e., the SHY strategy.
The result of this 2D heteronuclear correlation analysis could then be displayed as a conventional 2D NMR dataset, using colored contours (Figure 4). In the example shown, the gem-methyl resonances of the parent compound and III can be distinguished by the intensity level (degree of correlation) between individual 19F resonances and 1H resonances. Analytical Chemistry, Vol. 80, No. 4, February 15, 2008
1077
Figure 6. 19F-19F signal intermetabolite correlation connectivities. Hierarchical cluster analysis (A) and correlation matrix of 19F peak intensities (B) indicating relative and absolute similarities between excretion profiles of the five 19F NMR-detectable flucloxacillin metabolites. The correlation matrix is colored by the Pearson correlation (r). ? ) structure of metabolite V proposed as 5′-hydroxymethyl-(5R)-flucloxacillin penicilloic acid.
Figure 7. Proposed metabolism scheme for flucloxacillin. I ) flucloxacillin, II ) (5R)-flucloxacillin penicilloic acid, III ) 5′-hydroxymethylflucloxacillin, IV ) (5S)-flucloxacillin penicilloic acid. A putative structure for metabolite V is shown (5′-hydroxymethyl-(5R)-flucloxacillin penicilloic acid).
X-Edited 1H-1H STOCSY Analysis (X-STOCSY). An alternative strategy was to calculate a homonuclear correlation spectrum in a similar manner but then only to display regions of the spectrum with an appreciable correlation to a 19F resonance. This approach is conceptually analogous to the spectral editing of conventional homonuclear NMR experiments for single samples by magnetization transfer to a heteronucleus, e.g., 3D HSQCTOCSY.29 One advantage of this approach is that the optimum data preprocessing used for generating the 1H spectral data can be used for STOCSY analysis. In the present example, independent normalization of the 1H NMR dataset greatly simplified the visual detection of features in the 1H-1H correlation spectrum that were likely to originate from structural relationships (Figure 5A,B). Subsequent 19F-editing of the 1H-1H correlation spectrum via the 19F resonance of the parent (I, Figure 5C) and IV, (Figure 5D) shows how highly metabolite-specific 1H-1H cross-peaks can be highlighted using this mode of analysis. 1078 Analytical Chemistry, Vol. 80, No. 4, February 15, 2008
Intermetabolite Correlations. Repeating the same 1D correlation analysis as described above, but using the intensities of 19F resonances of compounds II and III produced qualitatively similar results to the parent compound while 1D correlation analysis to the 19F resonance of the unknown metabolite (V) highlighted additional resonances characteristic of IV. This pattern of similarities and differences between the correlation profiles could be predicted from and is a direct consequence of correlations between the urinary excretion of individual fluorinated metabolites. Visual inspection and hierarchical clustering of the 19F-19F intensity correlation matrix indicated that the urinary concentrations of II and III showed a high degree of correlation to the parent (I, Figure 6). By contrast, concentrations of IV exhibited a weaker correlation to the parent and II and III, but did possess a reasonable correlation to levels of the unknown metabolite V. By similar arguments to the 1D analyses, variation between the 2D heteronuclear correlation analyses can be explained by the correlation structure originating from flucloxacillin and its metabolites; i.e., the more distant (in pathway connectivity terms) two metabolites are, the weaker their overall correlation. In principle, this gives a new means of assessing the likely metabolic conversion sequence between drug metabolites and could be applied for pathway reconstruction purposes (Figure 7). Formation of III and II requires hydrolysis of the β-lactam ring or hydroxylation of flucloxacillin, respectively, each of which are single-step biotransformations which readily occur in vivo (Figure 7).27,28 However, IV cannot be formed directly from the parent and requires epimerization from the 5R isomer, an additional reaction which occurs spontaneously in aqueous solution.31,32 This metabolic pathway is consistent with the notion that the degree of correlation between the urinary concentrations of flucloxacillin and one of its metabolites can be rationalized in terms of the number of biological interconversions required to transform one to the other. On the basis of this assumption, it is reasonable (31) Davis, A. M.; Jones, M.; Page, M. I. J. Chem. Soc., Perkin Trans. 2 1991, 1219-1223. (32) Davis, A. M.; Layland, N. J.; Page, M. I.; Martin, F.; O’Ferrall, R. M. J. Chem. Soc., Perkin Trans. 2 1991, 1225-1229.
to suggest that the unknown fluorinated metabolite (V) is a further conversion of either III and IV as it exhibits a better correlation to both metabolites than to the parent. The compound 5′hydroxymethyl-(5R)-flucloxacillin penicilloic acid is one metabolic step from III and two from IV, consistent with this rationale, and is likely to be formed as it is the product of both known biological transformations in vivo. The presence of this metabolite in urine has previously been reported following flucloxacillin exposure in man.26 While we cannot make a definitive assignment from spectral correlations alone, the analysis presented demonstrates how intermetabolite correlations can be used in an integrated manner with heteronuclear NMR datasets to facilitate the structural assignment of unknown metabolites. CONCLUSION Using the metabolism of the antibiotic flucloxacillin as an example, we have shown how statistical correlation (HETSTOCSY) can be used to integrate 19F and 1H heteronuclear NMR datasets, combining the benefits of direct 1H detection for structural assignment with the resolution and simplicity of 19F spectra. Additionally, the use of heteronuclear editing of homonuclear statistical correlation data (X-STOCSY) to aid interpretation has been demonstrated. This approach is generic and would be extendable to include other drug types containing X-nuclei (e.g., 13C) if appropriate labels were present. Clear advantages over traditional heteronuclear spectroscopy were exemplified, including
the preservation of spectroscopic resolution in 2D spectra and the detection of structural relationships across an effectively unlimited number of chemical bonds between nuclei. It has also been demonstrated how a detailed examination of intermetabolite correlations can aid the interpretation of heteronuclear spectral correlations. While these approaches have obvious benefit for the study of drug metabolism, there are important implications for endogenous metabolic profiling (metabonomics/metabolomics), as the accurate identification and comprehensive exclusion of exogenous signals is a prerequisite for the correct characterization of pharmacodynamics and/or the systemic response to other forms of xenobiotics. ACKNOWLEDGMENT The first two authors (H.C.K. and T.J.A.) contributed equally to this work. The authors wish to thank Pete Simpson (CFNMR Centre, Imperial College London) for his help and advice acquiring 800 MHz NMR spectra. Tim Ebbels kindly provided MATLAB scripts for conducting HCA analysis. T.J.A. is supported by the European Union carcinoGENOMICS project (Contract No. PL037712).
Received for review October 1, 2007. Accepted November 7, 2007. AC702040D
Analytical Chemistry, Vol. 80, No. 4, February 15, 2008
1079
