An Accurate-Mass-Based Spectral-Averaging Isotope-Pattern-Filtering

Jun 11, 2009 - AMSA-IPF will be another great tool to facilitate detection and ID of drug ... with FDA's most recent guidance on metabolites in safety...
0 downloads 0 Views 552KB Size
Download PDF
Anal. Chem. 2009, 81, 5910–5917

An Accurate-Mass-Based Spectral-Averaging Isotope-Pattern-Filtering Algorithm for Extraction of Drug Metabolites Possessing a Distinct Isotope Pattern from LC-MS Data Peijuan Zhu,* Wei Tong, Kevin Alton, and Swapan Chowdhury Drug Disposition, Pharmaceutical Sciences and Drug Metabolism, Schering Plough Research Institute, 2015 Galloping Hill Road, Kenilworth, New Jersey 07033-1300 Detection and identification (ID) of all drug metabolites following liquid chromatography (LC)/mass spectrometry (MS) analysis of complex biological matrixes are not trivial. To facilitate detection of drug-derived materials that possess highly diagnostic isotopic patterns (e.g., chlorineand bromine-containing compounds), we report an accurate-mass-based spectral-averaging isotope-patternfiltering (AMSA-IPF) algorithm developed in the computational language R. The AMSA-IPF algorithm offers three significant improvements over the traditional isotope filtering method often provided by instrument vendors. First, spectral averaging is performed before the IPF to reduce scan-to-scan variability of ion intensities. Second, the IPF process is strictly based on accurate mass typically obtained on high resolution mass spectrometers. The designated isotopic ion-pairs (e.g., M + 2:M or M + 1:M, where M is the molecular ion and M + 1 and M + 2 are the isotopic ions) must fall into the predefined accurate mass tolerance window (e.g., 5 ppm) and at the same time satisfy the predefined relative abundance criteria. Third, both M + 1:M and M + 2:M ion pairs are inspected in the filtering process. The inclusion of M + 1:M ion pair enhanced the specificity of this algorithm by removing background ions that form M:M + 2 ion pairs within predefined isotope ratios by coincidence. The algorithm demonstrated excellent effectiveness in detecting drug-related ions in in vivo samples (plasma, bile, urine and feces) obtained from rats orally dosed with 14Cloratadine. The ion chromatograms of the filtered LCMS data files showed near perfect qualitative correlation with the corresponding radioprofiles. AMSA-IPF will be another great tool to facilitate detection and ID of drug metabolites in complex LC-MS data without the help of radiolabels. The AMSA-IPF algorithm is applicable to not only compounds containing distinct natural isotopes (such as Cl and Br) but also compounds that contain synthetically incorporated isotopes (13C, 15N, etc) generating a distinct isotope pattern. The ability to detect and identify metabolites from nonradiolabeled studies will be extremely beneficial to * To whom correspondence should be addressed. E-mail: peijuan.zhu@ spcorp.com.

5910

Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

achieve compliance with FDA’s most recent guidance on metabolites in safety testing (MIST). The rapid advancement in ion physics technology has made available more sensitive LC-MS systems with high mass accuracy for metabolite ID work. Mass accuracy of less than 5 ppm can now be routinely obtained on FT-ion cyclotron resonance (ICR)MS, FT-Orbitrap-MS, and high-end time-of-flight (TOF) mass spectrometers. However, at present, detection and ID of metabolites from complex LC-MS data without a radiotracer is still challenging because of the interferences from a large excess of matrix ions. Because of the need for manual searching of metabolites, the rate of data mining is much slower than that of data acquisition. Rapid, accurate, and comprehensive computational tools for data mining are in great demand to address this issue. Recently, the development of a few computational tools for data acquired on the new generation of high resolution mass spectrometers such as Q-TOF and LTQ-Orbitrap have been reported.1-4 The most cited among them is the mass defect filter (MDF) or the fractional mass filter which is designed to detect drug-related ions by utilizing mass defect differences between drug-related ions and ions of endogenous origin.2,3,5,6 MDF and fractional mass filter have been applied to the studies of many low molecular weight compounds.2,3,5,6 Their discriminating power is limited when the mass defects of metabolites are similar to those of matrix ions or when the mass defects of unusual metabolites fall outside of the hypothesized ranges. Besides MDF, a new generation of background subtraction tools based on accurate mass was reported by Zhang et al.1,4 Later noise-reduction and ion-removal functions were added to further enhance its performance.7 The second generation of background subtraction tools addressed the pitfalls of LC retention time variability by subtracting background ions Zhang, H.; Yang, Y. J. Mass Spectrom. 2008, 43, 1181–1190. Zhang, H.; Zhang, D.; Ray, K. J. Mass Spectrom. 2003, 38, 1110–1112. Zhang, D. C.; Peter, T.; Zhang, H. Drug Metab. Lett. 2007, 1, 287–292. Zhang, H.; Ma, L.; He, K.; Zhu, M. J. Mass Spectrom. 2008, 43, 1191– 1200. (5) Zhu, M.; Ma, L.; Zhang, D.; Ray, K.; Zhao, W.; Humphreys, W. G.; Skiles, G.; Sanders, M.; Zhang, H. Drug Metab. Dispos. 2006, 34, 1722–1733. (6) Tiller, P. R.; Bateman, K. P.; Castro-Perez, J.; Mcintosh, I. S.; Kuo, Y.; Baillie, T. A. Rapid Commun. Mass Spectrom. 2008, 22, 3510–3516. (7) Zhu, P.; Ding, W.; Tong, W.; Ghosal, A.; Alton, K.; Chowdhury, S. Rapid Commun. Mass Spectrom. 2009, 23, 1563–1572. (1) (2) (3) (4)

10.1021/ac900626d CCC: $40.75  2009 American Chemical Society Published on Web 06/11/2009

from multiple scans within a retention time range. Superior effectiveness is achieved in extracting drug-related ions from complex LC-MS data compared to the traditional background subtraction method which operates in a scan-for-scan mode and in unit mass resolution. Many other postacquisition data processing approaches utilize distinct fragmentation patterns of drug metabolites, and those include constant neutral-loss and precursorion filters.8-11 For compounds that have a distinct isotope pattern, either due to the presence of natural isotopes (Cl, Br, etc.) or due to custom incorporation of stable-labeled isotopes (2H, 13C, 15N, etc.) at a known percentage, isotope pattern recognition tools have always been favorite choices. On-the-fly isotope-pattern-dependent MS/MS experiments are often used to capture drug-related ions and their fragments.9,12-14 This approach is effective for high abundance metabolites, but it tends to miss low abundance ions, especially in regions rich with matrix ions. Isotope pattern filters offered in commercial software packages can also be used to perform postacquisition data mining.11,13 The traditional isotope pattern filters are usually based on a single isotope ratio (e.g., M + 2:M) and operate under unit resolution. Lately, with the introduction of high resolution mass spectrometers such as Q-TOF and LTQ-Orbitrap, some isotope pattern filter/recognition tools, e.g., Metworks 1.2, have incorporated accurate mass options. However, their applications in processing complex LC-MS data obtained from biological matrixes have not been extensively reported, and their effectiveness compared with traditional isotope pattern filters has not been fully evaluated. In addition to the isotope filters described above, complex mathematical or statistical models have been used to perform feature extraction of complex isotope patterns, such as those of peptides or proteins in LC-MS data.15-18 For high-resolution MS data, these model-based approaches are usually computationally expensive and cannot deal well with overlapping features.15 In this study, we report a new and improved isotope pattern filter, which is an accurate-mass-based spectral-averaging isotopepattern-filter developed using R language, to extract ion signals containing simple isotope patterns such as those of chlorine- and bromine-containing compounds. The AMSA-IPF algorithm offers three major enhancements over conventional IPFs: spectral averaging, accurate mass based filtration, and inclusion of both M + 2:M and M + 1:M ion pairs in the filtering criteria. We tested (8) Ruan, Q.; Peterman, S.; Szewc, M. A.; Ma, L.; Cui, D.; Humphreys, G. W.; Zhu, M. J. Mass Spectrom. 2008, 43, 251–261. (9) Ma, L.; Wen, B.; Ruan, Q.; Zhu, M. Chem. Res. Toxicol. 2008, 21, 1477– 1483. (10) Xia, Y.; Miller, J. D.; Bakhtiar, R.; Franklin, R. B.; Liu, D. Q. Rapid Commun. Mass Spectrom. 2003, 17, 1137–1145. (11) Cuyckens, F.; Hurkmans, R.; Castro-Perez, J. M.; Leclercq, L.; MortishireSmith, R. J. Rapid Commun. Mass Spectrom. 2009, 23, 327–332. (12) Lim, H.-K.; Chen, J.; Cook, K.; Sensenhauser, C.; Silva, J.; Evans, D. C. Rapid Commun. Mass Spectrom. 2008, 22, 1295–1311. (13) Cuyckens, F.; Balcaen, L. I. L.; Wolf, K. D.; Samber, B. D.; Looveren, C. V.; Hurkmans, R.; Vanhaecke, F. Anal. Bioanal. Chem. 2008, 390, 1717–1729. (14) Ma, S.; Chowdhury, S. K.; Alton, K. B. Curr. Drug Metab. 2006, 7, 503– 523. (15) Noy, K.; Fasulo, D. Bioinformatics 2007, 23, 2528–2535. (16) Gras, R.; Muller, M.; Gasteiger, E.; Gay, S.; Binz, P. A.; Bienvenut, W.; Hoogland, C.; Sanchez, J. C.; Bairoch, A.; Hochstrasser, D. F.; Appel, R. D. Electrophoresis 1999, 20, 3535–3550. (17) Horn, D. M.; Zubarev, R. A.; McLafferty, F. W. J. Am. Soc. Mass Spectrom. 2000, 11, 320–332. (18) Berndt, P.; Hobohm, U.; Langen, H. Electrophoresis 1999, 20, 3521–3526.

this algorithm on LC-MS data generated from plasma, urine, bile, and feces samples obtained from rats orally dosed with 14Cloratadine (Claritin), a chlorine-containing compound. We also compared the results generated by AMSA-IPF with those generated by Metworks 1.2. Written in R language, the AMSAIPF program can be easily hosted on a linux server and provide service to end users through a web interface. AMSA-IPF will be another great tool to facilitate detection and identification of drug metabolites in exploratory drug metabolism studies in discovery settings, First in Human (FIH) studies, and in toxicology studies, where radiolabeled drugs are not usually used. EXPERIMENTAL SECTION Materials and Methods. Reagents. Acetonitrile, DMSO, and acetic acid were obtained from Fisher Scientific (Somerville, NJ). Water was purified using Millipore Milli-Q water purification system (Burlington, MA). Loratadine and 14C-loratadine were synthesized in-house. All other chemicals were purchased from Sigma Aldrich (St. Louis, MO). Loratadine Administration to Rats and Sample Collection. Two bile-duct cannulated male Sprague-Dawley rats were orally dosed with vehicle only on day 1 for the collection of control bile (0-6 h). On day 2, both animals were orally dosed with 14C-loratadine (15 mg/kg, 10 µCi/mg), and bile was similarly collected from 0-6 h. All samples were stored at -20 °C. Two groups of intact rats (7 each) were orally dosed with 14Cloratadine (15 mg/kg, 10 µCi/mg) or the vehicle (0.4% hydroxypropyl methylcellulose). Urine and feces were collected up to 6 h. At 6 h postdose, rats were euthanized by CO2 inhalation and terminal blood samples were collected by cardiac puncture. The blood samples were transferred to plastic tubes containing K2-EDTA as an anticoagulant on wet ice. The tubes were centrifuged for 3 min at 10 000g in a refrigerated centrifuge maintained at ∼4 °C. The resultant plasma was separated, transferred to plastic tubes, and pooled by dose group. Preparation of Plasma, Bile, and Urine Samples Obtained from Rats Treated with 14C-Loratadine. An aliquot of 10 mL of rat plasma (loratadine or vehicle treated) was extracted three times with 30 mL of 1% acetic acid in acetonitrile. The supernatants were combined, evaporated to 0.7 mL, and then reconstituted with 0.3 mL of DMSO. An aliquot (60 uL) was injected into the highresolution LC-MS system composed of Accela UPLC and LTQOrbitrap. Rat bile (20 uL) and urine (20 uL) pooled from individual rats were injected directly into the LC-MS system. Accurate Mass LC-MS. High-resolution accurate-mass LC-MS data were obtained using an LTQ Orbitrap Discovery FTMS (ThermoFisher Scientific, Bremen, Germany) coupled with a Thermo Accela LC system. The instrument was calibrated everyday before data acquisition and operated at 30 000 m/z resolution using the vendor-supplied Xcalibur data system. The data were acquired in centroid mode using Fourier transformed (FT) full scan. The scan range was from m/z 100 to 1000 in positive mode. MS/MS experiments were performed in centroid mode using LTQ part of the instrument. The typical settings were as follows: activation type CID, normalized collision energy at 35%, isolation width at 2 Da, default charge state at 1, activation Q at 0.25, and activation time at 30 ms. Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

5911

Figure 1. Schematic illustration of the AMSA-IPF. X stands for scan number. Step 1: ion intensities of the same ion (defined as within 5 ppm mass tolerance window) across five adjacent scans in the original data file are averaged and saved into the averaged data file. Step 2: the M, M + 1, and M + 2 ions are located within the each scan in the averaged data file, and the M + 1:M and M + 2:M ratios are calculated. If these ratios fall into the predefined ranges, M, M + 1, and M + 2 ions are flagged in the original data file. Step 3: all ions that are not flagged are assigned 0 intensities. Flagged ions remain untouched.

Chromatographic separation for loratadine plasma, urine, and feces samples was performed using a Luna Phenyl-hexyl column (250 × 4.6 mm, 5 µm) with a Phenyl guard column (4 × 3.0 mm) (Phenomenex, Torrance, CA). The mobile phases were 10 mM ammonium acetate in water (pH 6.0) (A) and acetonitrile (B). The gradient started at 5% B, ramped up to 45% B at 10 min, and then to 75% B at 25 min. After a 10 min wash at 95% B, the gradient was returned to the initial condition and equilibrated for 9 min prior to next injection. Chromatographic separation for loratadine bile samples was performed using a Inertsil C8 column (150 × 4.6 mm, 5 µm) with a Inertsil C8 guard column (4 × 3.0 mm) (Metachem Technologies, Torrance, CA) at a flow rate of 1 mL/ min. The mobile phases were 10 mM ammonium acetate in water (pH 6.0) (A) and acetonitrile (B). The gradient started at 10% B, ramped up to 28% B at 25 min, and then to 90% B at 40 min. After a 10 min wash at 90% B, the gradient was returned to the initial condition and equilibrated for 15 min prior to next injection. Following LC separation, a postcolumn Tee splitter was installed in both methods so that ∼80% of the LC flow was directed to a flow radiometric detector (IN/US, Tampa, FL) and the remaining ∼20% to the LTQ-Orbitrap mass spectrometer. Implementation of AMSA-IPF Algorithm. Centroid accurate mass LC-MS data were converted to NetCDF format in XCalibur (ThermoFisher) and processed using a program written in the statistical programming language R. The algorithm is implemented in three steps (Figure 1). First, spectral averaging is performed to reduce the variability of ion intensities across scans. For each ion (M) in a particular scan (X), the program searches in adjacent scans (e.g., two scans before and two scans after) for ions with m/z identical to that of ion M (typically within 5 ppm mass tolerance window). The intensities of these ions were added to 5912

Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

the intensity of ion M and then divided by the total scan number, e.g., five in this case. These new values are recorded into another data file referred to as the averaged data file. Second, the program searches for ions that match the predefined isotope pattern in the averaged data file. For each ion (M), the program searches in the same scan for M + 1 and M + 2 isotopic ions with m/z within the 5 ppm mass tolerance window of the predefined m/z for M + 1 and M + 2 ions. Typically, the predefined m/z for M + 1 ion is the m/z of M plus 1.0033 (according to the spectral simulation for loratadine generated by Qual Browser in Xcalibur). For chlorine- and bromine-containing compounds, the predefined m/z for M + 2 ion is the m/z of M plus 1.9979 (according to the spectral simulation for loratadine generated by Qual Browser in Xcalibur). If M + 1 and/or M + 2 isotopic ions are not found, their intensities are treated as 0s. The program then calculates the M + 2:M and M + 1:M isotope ratios. If both ratios fall into their predefined ranges (in this case, ± 10% around the theoretical isotope ratios of the parent drug loratadine simulated using Qual Browser in Xcalibur, e.g. 14-34% for M + 1:M and 22-42% for M + 2:M), the corresponding M, M + 1, and M + 2 ions in the original data file are flagged. In the last step, ions without flags are assigned 0 intensities, while ions with flags remain unchanged. Finally the modified NetCDF data file is converted back to a RAW file for easy reviewing with Qual Browser in XCalibur or is further processed for ion peak list generation as described below. Ion Peak List Generation. To compare the specificity of our AMSA-IPF algorithm with Metworks, an R script was written to generate a peak list of all ions observed after AMSA-IPF processing. The script utilizes the Bioconductor XCMS package available on www.bioconductor.org, and the calculations are carried out in two steps. In the first step, mass traces (characterized as regions

Figure 2. Structure and accurate masses of loratadine and desloratadine.

with m/z falling within defined mass tolerance window in consecutive scans) in the LC-MS map are located. In the second step, a continuous wavelet transform is performed on the mass traces found in the first step to locate ion peaks on different scales.19 Information on all ion peaks is summarized into a table containing the averaged accurate mass, retention time, maximum intensity, signal-to-noise ratio, etc. The parameters used in generating ion peak lists were as follows: mass tolerance window at 5 ppm, minimum peak intensity at 50 000, chromatographic peak width minimum 3 s, and maximum 60 s. Calculation of Isotope Ratio Statistics. An R script was written to calculate the M + 1:M and M + 2:M isotope ratios with and without spectral averaging when M, M + 1, and M + 2 were all present in the same scan. The calculated isotope ratios along with information for M, such as m/z, retention time, scan number, and intensity, were compiled into a list and exported into a txt file which was then imported into Excel. The list was sorted in ascending order according to m/z and retention time. As a result, ions belonging to multiple scans of the same peak were grouped together. Mean and standard deviations of the isotope ratios were calculated for each ion group that was confirmed to be from one of the major drug metabolites. RESULTS AND DISCUSSION Assessment of Effectiveness. The effectiveness of our AMSA-IPF algorithm was tested by processing accurate mass LCMS data of from plasma, urine, bile, and feces samples collected from rats dosed with 15 mg/kg 14C-loratadine (Figure 2), a compound containing a chlorine atom. The total ion chromatograms (TICs) before and after AMSA-IPF processing (panel A and B, respectively) and the corresponding radiochromatograms (panel C) are shown in Figure 3. For plasma, no drug-related ions were discernible in the unprocessed TIC because of the predominance of background matrix ions and elevated baseline (Figure 3, 1A). After applying AMSA-IPF algorithm, most matrix ions were removed and the TIC baseline was reduced from ∼3 × 107 to literally 0. As a result, drug-related ions were revealed as major peaks (Figure 3, 1B). The processed TIC is in excellent qualitative correlation with the radiochromatogram (Figure 3, 1C), suggesting that all major drug-related peaks have been detected. Similar results were observed after processing corresponding data from urine (Figure 3, 2A-2C) and feces (Figure 3,4A-4C) samples. A different LC condition was used for bile samples to separate bile acids from most of the metabolites and to avoid ion suppression. Looking at profiles from the bile sample, AMSA-IPF again (19) Du, P.; Kibbe, W. A.; Lin, S. M. Bioinformatics 2006, 22, 2059–2065.

demonstrated its power to remove the large excess of matrix ions (mostly bile acids) and highlight drug-related ions in processed TIC of the bile sample (Figure 3, 3A and 3B). The radiochromatogram in general showed good qualitative correlation with the AMSA-IPF-processed TIC (Figure 3, 3B and 3C). To ensure the sensitivity of our AMSA-IPF algorithm, all unprocessed data files were carefully searched manually and with the help of other tools (such as background subtraction7 and mass defect filter2) for potential drug-related ions that might have been erroneously filtered out. These searches have not yet found a single ion that was missed by our AMSA-IPF algorithm. All major drug metabolites detected by AMSA-IPF and their identifications are included in Table 1. Ease of Data Review and Metabolite Detection. The simplicity and cleanliness of the AMSA-IPF-processed data allows easy detection and ID of drug-related ions by a cursory scan of MS spectra in the TIC regions of interest. As an example, Figure 4, 1A and 1B show the mass spectra at 12.22 min of the LC-MS analysis of plasma before and after AMSA-IPF processing, respectively. Similarly, the MS spectra at 14.94 min of the urine analysis before and after AMSA-IPF processing are shown in Figure 4, 2A and 2B, respectively. The processed mass spectra became so clean that the drug-related ions could be identified effortlessly. Comparison with Metworks 1.2. To compare our AMSAIPF algorithm with commercially available software, we also processed the same data set using Metworks 1.2 with identical parameter settings: isotope ratio tolerance at 10%, minimum intensity at 50 000, and mass tolerance window at 5 ppm. Instead of producing a filtered data file, Metworks performs a scan-for-scan search of the data file and provides an ion list accompanied by spectra and extracted ion chromatograms. The total number of ion peaks and unique metabolites detected by Metworks and by our AMSA-IPF algorithm are compared in Table 2. The AMSA-IPF list includes both the molecular ion M and the two isotopic ions M + 1 and M + 2, while the list generated by Metworks only includes the molecular ion M, and occasionally M + 1 ion if the M + 3:M + 1 ratio also falls into the predefined range. The table indicates that Metworks produced many false positives evidenced by the comparatively large number of identified ions. The percent of positive hits by AMSA-IPF was almost 70% while that for the Metworks was less than 10%. Upon close examination, it was obvious that a large percentage of these false positives are due to coincidental occurrences of matrix ions forming M + 2:M ion pairs with isotope ratios falling inside the predefined ranges. AMSA-IPF not only detected all the drug-related ions found by Metworks, but also captured a number of other low intensity drug-related ions that were missed. Because both methods perform searches based on 5 ppm mass tolerance window, the extra specificity achieved in AMSA-IPF is likely due to spectral averaging and inclusion of M + 1:M isotope ratio in the filtering criteria. The Power of Accurate Mass. Just as accurate mass has revolutionized metabolite ID by drastically reducing the number of possible chemical formulas that match certain m/z values, it has also promoted the power of isotope pattern filtering to a completely new level. Because of the narrow mass tolerance window, accurate mass can remarkably reduce false positives Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

5913

Figure 3. TIC and radiochromatograms of rat plasma (1A-1C), urine (2A-2C), bile (3A-3C), and feces (4A-4C) samples. A: unprocessed TIC data. B: TIC after AMSA-IPF processing. C: radiochromatogram.

(usually caused by matrix ions) in the isotope pattern filtering process since there is a much lower possibility that two matrix ions can fall into the predefined mass tolerance windows by chance. This enables our AMSA-IPF algorithm to produce a near zero baseline (Figure 3), which has not been achieved previously. Operating in high resolution has also made it possible to discriminate between coeluting metabolites with overlapping isotopic envelopes. This is illustrated in Figure 4, 2C, which is a zoomed-in view of Figure 4, 2B, on the x axis around m/z 327. Obviously, two minor metabolites with m/z 325.1107 (metabolite A) and m/z 327.1263 (metabolite B) coeluted. The M + 2 ion (m/z 327.1078) from metabolite A was easily distinguished from the M ion from metabolite B at m/z 327.1263 with the power of high resolution. Thus, the potential interference that would have occurred under lower 5914

Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

resolution was avoided. If these two isobaric ions were not resolved, metabolite A would not have been detected since its M + 2:M ratio would not have met the criteria. A direct comparison of AMSA-IPF results over a range of mass tolerance settings (to simulate different mass resolution) is illustrated in Figure 5. As the mass accuracy is decreased from 5 ppm (2.5 mDa at 500 Da, Figure 5A) to 50 ppm (25 mDa at 500 Da, Figure 5B), a serious deterioration of the TIC chromatogram was observed due to an elevated and jagged baseline (Figure 5B). At 400 ppm, which is typical for unit resolution, almost no drug-related peak was discernible (Figure 5D). It is obvious that with higher resolution and mass accuracy, greater sensitivity and specificity is achieved in isotope pattern filtering. Advantage Offered by Spectral Averaging. Our AMSA-IPF algorithm minimizes scan-to-scan variability by averaging ion

Table 1. Measured Isotope Ratios of Major Drug Metabolites with and without Spectral Averaginga M + 2:M, % proposed metabolite definition

[M + H]+ RT (min) 311

17.4

311 323 325

30.9b 14.8 15.7

327

12.4

327 327 343

22.9b 21.7b 9.8

357 381 383 413 415 429 503 575

20.4b 29.8b 37.3b 30.2b 15.1 23.0b 28.1b 19.5b

M + 1:M, %

spectral no spectral spectral no spectral source theoretical averaging averaging theoretical averaging averaging

DL

plasma urine feces DL bile DL N-oxide piperidine ring aromatized plasma DL + O - 2H urine plasma feces DL + O plasma urine feces DL + O bile DL + O bile DL + 2O urine plasma DL + 3O - 2H bile L - 2H bile L bile L + 2O bile L CH3f COOH urine L + O and CH3f COOH bile DL + O + Glu bile L + O + Glu bile

32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0 32.0

31.3 ± 2.1 31.1 ± 1.7 31.6 ± 1.5 29.1 ± 3.4 30.8 ± 2.0 31.6 ± 1.3 30.6 ± 2.4 29.6 ± 2.9 30.1 ± 2.4 31.1 ± 1.8 30.6 ± 3.6 31.8 ± 1.1 31.5 ± 1.1 30.1 ± 2.2 29.0 ± 2.7 30.6 ± 1.1 27.9 ± 2.9 27.5 ± 1.4 28.7 ± 1.3 27.0 ± 2.8 28.0 ± 0.9 26.5 ± 1.1 29.8 ± 1.7

31.5 ± 4.1 31.3 ± 3.8 32.0 ± 4.2 30.2 ± 6.5 31.0 ± 3.5 32.1 ± 4.6 30.9 ± 4.6 29.5 ± 6.5 30.1 ± 4.2 31.2 ± 3.3 31.2 ± 7.6 31.7 ± 2.3 31.2 ± 2.9 30.2 ± 3.9 29.7 ± 5.1 31.7 ± 3.2 28.8 ± 6.4 29.0 ± 7.1 28.9 ± 2.3 27.3 ± 4.1 28.0 ± 2.0 26.9 ± 2.7 29.9 ± 3.8

20.5 20.5 20.5 20.5 20.5 20.5 20.5 20.5 20.5 23.8 23.8 23.8 23.8 23.8 27.0 30.3

19.0 ± 2.5 19.7 ± 1.4 19.3 ± 1.9 17.8 ± 2.8 19.1 ± 2.7 19.8 ± 1.9 19.3 ± 2.3 16.4 ± 3.9 18.3 ± 2.1 19.5 ± 1.3 16.5 ± 3.6 20.3 ± 0.9 20.1 ± 0.9 19.2 ± 1.8 18.2 ± 2.1 19.6 ± 1.5 21.0 ± 3.0 19.6 ± 3.7 23.2 ± 2.0 22.5 ± 2.7 22.7 ± 0.8 25.8 ± 1.6 26.3 ± 2.1

20.2 ± 3.3 19.7 ± 2.7 19.8 ± 2.7 19.8 ± 4.5 20.5 ± 3.3 20.4 ± 3.1 19.6 ± 4.2 19.4 ± 6.8 18.8 ± 3.1 19.6 ± 2.8 19.6 ± 5.9 20.2 ± 1.6 20.0 ± 2.2 19.6 ± 2.9 19.1 ± 4.6 19.8 ± 2.1 22.2 ± 4.3 20.6 ± 4.2 23.2 ± 3.2 22.8 ± 3.7 22.7 ± 2.0 26.0 ± 3.3 26.6 ± 3.7

a Ratios are expressed as mean ± standard deviation. RT: retention time. L: loratadine. DL: desloratadine (Figure 2) Glu: glucuronide. b Retention times obtained from the bile samples were based on a different LC gradient from that of plasma, urine, and feces samples.

Figure 4. Exemplary accurate mass spectra at 12.22 min (1A and 1B) and 14.94 min (2A and 2B) of rat plasma LC-MS data before and after AMSA-IPF processing. 1A and 2A: before processing. 1B, 2B, and 2C: after processing. 2C is a zoomed view of 2B around m/z 327.

intensities across multiple scans (typically five scans), as is obvious upon visual examination of the data before and after spectral

averaging. To our knowledge, this is the first application of spectral averaging used for isotope pattern filtering of LC-MS data. We have summarized the measured isotope ratios and their standard deviations for each identified major metabolite in Table 1. In the unprocessed data, the standard deviation for isotope ratio measurements was ∼5% for M + 2:M ion pairs and ∼4% for M + 1:M ion pairs. For some low-intensity ions the standard deviation could go beyond 7%. Because the theoretical isotope ratios are 32% and 24% for M + 2:M and M + 1:M ion pairs, respectively, the variability can be considered rather high. Spectral averaging is shown to reduce the standard deviation of measured isotope ratios by half (Table 1). This process is particularly helpful for lowintensity ions that suffer from excessively high variability. Without spectral averaging, some of these ions would fall out of their isotope ratio ranges and be filtered out erroneously. Figure 6 is a comparison of unsmoothed isotope filtered data from the plasma sample with or without prior spectral averaging. It is noticeable that many low-intensity data points are missing without spectral averaging, resulting in a noisy TIC trace in the low signal ranges (Figure 6B). It was also observed that the measured isotope ratios were consistently slightly lower than their theoretical values (Table 1), indicating a small systematic bias generated by our LTQOrbitrap toward the lower side. The bias was around -1% for high intensity ions (such as m/z 311 and m/z 327 in all matrix) and was -3% to -5% for low intensity ions (most metabolites in bile and m/z 415 in urine). Since most of the ions that exhibit higher bias were found in bile, it is not clear whether the bias was solely caused by poor signal quality or ion suppression by bile acids also played a role. Despite the systemic bias, most drug-related ions still fell within our predefined isotope ratio ranges. In future applications, the isotope ratio ranges can be adjusted accordingly to reflect the systemic bias. Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

5915

Table 2. Comparison of Results Generated by AMSA-IPF and Metworks 1.2 total number of ions detected bya

number of drug-related ions detected byb

number of metabolites detected byc

% positive hits byd

matrix

AMSA-IPF

Metworks

AMSA-IPF

Metworks

AMSA-IPF

Metworks

AMSA-IPF

Metworks

plasma urine bile feces

38 32 61 16

324 103 118 324

25 22 45 11

9 8 10 6

6 7 22 4

6 4 10 4

65.8 68.8 73.8 68.8

2.8 7.8 8.5 1.9

a Total number of ions includes all ions that were detected/retained after processing. For AMSA-IPF, it includes both the molecular ion M and the two isotopic ions M + 1 and M + 2. For Metworks, it includes the molecular ion M, and occasionally M + 1 ion if the M + 3:M + 1 ratio also falls into the predefined range. b Number of drug-related ions: includes all ions that were confirmed to be drug-related. Each metabolite could have multiple ion signals: M, M + 1, M + 2, in source fragment ions and their corresponding isotope ions. c Number of metabolites detected: number of compounds that were confirmed to be true metabolite by accurate mass and MS/MS. d % Positive hits: defined as ‘number of drug-related ions’ divided by ‘total number of ions detected’.

Figure 5. AMSA-IPF results of rat plasma sample under different mass tolerance window settings (to simulate different mass resolution). A: 5 ppm, B: 50 ppm, C: 150 ppm, D: 400 ppm.

Figure 6. Comparison of AMSA-IPF results of rat plasma sample with (A) or without (B) spectral averaging.

Improved Specificity by Inclusion of M + 1:M Ratio in the Filtering Criteria. Our AMSA-IPF algorithm distinguishes itself from many other commercial isotope pattern filters, such 5916

Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

Figure 7. Comparison of AMSA-IPF results of rat plasma sample with (1A-3A) or without (1B-3B) M + 1:M isotope ratio included in the filtering criteria. Panel 1: TIC chromatograms. Panel 2 and 3: exemplary accurate mass spectra at 3.56 min (panel 2) and 12.65 min (panel 3).

as those offered in Metworks and MS-Xelerator, by including a second isotope ratio (M + 1:M) in the filtering criteria. Very often, in complex LC-MS data derived from complex biological sample, matrix ions can form M + 2:M ion pairs with isotope ratios falling inside the range by sheer coincidence (Figure 7, 1B-3B). These matrix ions, however, usually do not simultaneously possess a proper M + 1:M ratio. The addition of M + 1:M ratio in the filtering criteria allowed removal of such matrix ions (Figure 7, 1A-3A), thus further improved the specificity already achieved by accurate mass. As shown in Figure 7, 1A, two extra matrix ion peaks (at 3.59 and 12.63 min) were removed by adding M + 1:M ratio to the filtering process, leaving only drug-related ions as major peaks in the processed TIC. The elimination of false positives should reduce the amount of time spent on confirming each positive hit and will improve the efficiency of metabolite detection and ID work.

The Latest MIST Guideline and the Potential Impact of the AMSA-IPF Algorithm. The issuance of the latest FDA guidance on MIST has greatly heightened the importance of detection and identification of all major metabolites from early clinical investigations (www.fda.gov/OHRMS/DOCKETS/98fr/ FDA-2008-D-0065-GDL.pdf). The MIST guidance requires that metabolites identified at disproportionately higher levels in humans than in any animal test species be considered for further safety assessment. The FDA strongly urges that human metabolites be detected and identified as early as possible, preferably in First in Humans (FIH) studies, where radiolabeled drugs are typically not used because of cost and logistic issues. From a compliance standpoint, missing any major metabolite would be unacceptable and could lead to significant program delays. The new guidance also stresses that the metabolite-to-parent-drug ratios be assessed under steady-state conditions. This requirement again brings up the analytical challenge of profiling metabolites in clinical and multiple-dose animal toxicology studies, which typically do not use radiolabeled drugs (it is not allowed to administer radiolabeled drugs in multiple doses to humans). At current state, it is generally not a trivial task to detect all metabolites from LC-MS data without radiotracers, especially when the LC-MS data are generated from complex biological matrixes where the large abundance of matrix ions can potentially mask drug-related ions. Traditionally, people use a predictive approach and search in extracted ion chromatograms for metabolites that are hypothesized to form based on the core structure. However, this strategy will almost always miss metabolites that are generated through unusual pathways. Should these wayward metabolites emerge as major human metabolites in the late phase of drug development, serious delays could occur in filing and approval of new drug applications. Lately, new computational tools that perform objective searching/filtering of LC-MS data to aid metabolite detection have gained popularity. Among them, MDF and retention-time-shift-tolerant background subtraction have been developed, tested, and proven to be effective in many cases. Unlike the background subtraction program, AMSA-IPF does not require background/control data and there is no prior assumption of mass defect range as would be the case with MDF. With enhanced power, the AMSA-IPF algorithm will be another add-on to the existing tool set to aid in metabolite detection and ID work in nonradiolabeled samples when dosed drug contains natural or synthetically incorporated isotope pattern.

CONCLUSIONS The three major improvements, accurate-mass-based processing, spectral averaging, and inclusion of M + 1:M ratio in the filtering criteria, greatly enhanced the sensitivity and specificity of AMSA-IPF algorithm in comparison with traditional isotope filtering tools. This algorithm is highly effective in facilitating metabolite detection and identification without the help of radiolabels. With the publication of the new MIST guideline, metabolite detection in nonradiolabeled studies such as FIH and multipledose toxicology studies will become more and more important. We believe the AMSA-IPF algorithm will greatly improve the efficiency of metabolite detection and ID work in compliance with the new MIST guideline and promote the application of isotope pattern filter in LC-MS data processing to a completely new stage. It should be noted that AMSA-IPF is applicable to not only chlorine- and bromine-containing drugs but also compounds that have synthetically incorporated isotope patterns. For example, it is a common practice in in vitro metabolism studies (such as glutathione binding studies) to use compounds premixed with a certain percentage of 13C- or 15N-labels.20-24 The AMSA-IPF algorithm could be applied in these cases to search for drugrelated ions that possess these synthetic isotope patterns, thus obviating the need to use constant neutral loss and precursor ion filters which are not sensitive enough and very often generate false positives.22,24 ACKNOWLEDGMENT We thank the radiochemistry group at Schering Plough Research Institute for the help with 14C-loratadine synthesis. We gratefully acknowledge the help of Danielle Vella (Drug Disposition, Pharmaceutical Sciences, Schering Plough Research Institute) for generating the in vivo samples. Received for review March 27, 2009. Accepted May 29, 2009. AC900626D (20) Mutlib, A.; Lam, W.; Atherton, J.; Chen, H.; Galatsis, P.; Stolle, W. Rapid Commun. Mass Spectrom. 2005, 19, 3482–3492. (21) Baillie, T.; Rettenmeier, A. J. Clin. Pharmacol. 1986, 26, 481–484. (22) Yan, Z.; Caldwell, G. W. Anal. Chem. 2004, 76, 6835–6847. (23) Lim, H. K.; Chen, J.; Sensenhauser, C.; Cook, K.; Subrahmanyam, V. Rapid Commun. Mass Spectrom. 2007, 21, 1821–1832. (24) Lim, H. K.; Chen, J.; Cook, K.; Sensenhauser, C.; Silva, J.; Evans, D. C. Rapid Commun. Mass Spectrom. 2008, 22, 1295–1311.

Analytical Chemistry, Vol. 81, No. 14, July 15, 2009

5917