Anal. Chem. 1999, 71, 2294-2300
High-Throughput Bioanalytical LC/MS/MS Determination of Benzodiazepines in Human Urine: 1000 Samples per 12 Hours Jerry Zweigenbaum, Katja Heinig, Simon Steinborner, Timothy Wachs, and Jack Henion*
Analytical Toxicology, New York State College of Veterinary Medicine, Cornell University, 927 Warren Drive, Ithaca, New York 14850
The analytical capabilities of liquid chromatography tandem mass spectrometry for sensitive and highly selective determination of target compounds in complex biological samples makes it well suited for high-throughput analysis. We report the fast separation of six benzodiazepines isolated from human urine via selected reaction monitoring liquid chromatography/mass spectrometry using short dwell times to accommodate fast-eluting chromatographic peaks. The analytes were extracted from human urine samples along with their deuterium-labeled internal standards by a simple liquid-liquid extraction in 96-well plates. Using four autosamplers coupled to one chromatographic column and one tandem mass spectrometer operated in the turbo ion spray mode with positive ion detection, 1152 samples (12 96-well plates) were analyzed in less than 12 h. Through an electronic switching box designed and constructed in-house, the autosamplers were synchronized with the mass spectrometer so that injections were made as soon as the mass spectrometer was ready to collect data. Each run required 30 s to complete with another 7-8 s for the data system to load the next data file to be collected. Chromatographic integrity and ion current response remained relatively constant for the duration of the analyses. The results show acceptable precision and accuracy and demonstrate the feasibility of using fast separations with tandem mass spectrometry for high-throughput analysis of biological samples containing multiple analytes. Recent advances in drug discovery and lead compound optimization in the pharmaceutical industry have lead to an everincreasing need to analyze more samples faster and cheaper. These needs include combinatorial chemistry syntheses1,2 requiring analysis of many samples for purity and structure verification3 as well as new drug discovery strategies such as n-in-one dosing,4,5 in vitro metabolism studies,6,7 and Caco2 cell absorption studies.8,9 * Corresponding author: (e-mail)
[email protected]. (1) Floyd, C. D.; Lewis, C. N.; Whittaker, M. Chem. Br. 1996, 32, 31-35. (2) Gordon, E. M.; Gallop, M. A.; Patel, D. V. Acc. Chem. Res. 1996, 29, 144154. (3) Zeng, L.; Kassel, D. B. Anal. Chem. 1998, 70, 4380-4388. (4) Beaudry, F.; Le Blanc, J. C. Y.; Coutu, M.; Brown, N. K. Rapid Commun. Mass Spectrom. 1998, 12, 1216-1222. (5) Frick, L. W.; Adkison, K. K.; Wells-Knecht, K. J.; Woollard, P.; Higton, D. M. Pharm. Sci. Technol. Today 1998, 1, 12-18.
2294 Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
New drugs entering clinical trials increase the demand for highthroughput sample analysis, and stringent time constraints demand much faster turnaround time requiring better sample preparation.10 Response by the analytical community has been largely through the development of equipment and processes via microtiter plate technology.11-19 The 96-well plate is the current format for high-throughput analysis and is being extended to 384well plates, 1536-well plates, and beyond.20 However, currently the only really fast way to analyze samples for screening and verification is bioassay with radiometric or optical detection.21-24 Although these techniques can be very sensitive, they lack the selectivity and ultimate molecular identification that is often required. The use of liquid chromatography tandem mass spectrometry (LC/MS/MS) is both sensitive and selective, but even with current strategies this technique can be the bottleneck in sample analysis. Fast liquid chromatographic methods have been introduced and applied to both conventional and mass spectrometric detection and the term fast or rapid covers a range from many minutes to (6) Obach, R. S.; Baxter, J. G.; Liston, T. E.; Silber, B. M.; Jones, B. C.; Macintyre, F.; Rance, D. J.; Wastall, P. J. Pharmacol. Exp. Ther. 1997, 283, 46-58. (7) Sanwald, P.; David, M.; Dow, J. J. Chromatogr., B: Biomed. Appl. 1996, 678, 53-61. (8) Gan, L.-S. L.; Thakker, D. R. Adv. Drug Delivery Rev. 1997, 23, 77-98. (9) Boulenc, X. S. T. P. Pharma Sci. 1997, 7, 259-269. (10) Henion, J.; Brewer, E.; Rule, G. Anal. Chem. 1998, 70, 650A-656A. (11) Kaye, B.; Herron, W. J.; Macrae, P. V.; Robinson, S.; Stopher, D. A.; Venn, R. F.; Wild, W. Anal. Chem. 1996, 68, 1658-1660. (12) Allanson, J. P.; Biddlecombe, R. A.; Jones, A. E.; Pleasance, S. Rapid Commun. Mass Spectrom. 1996, 10, 811-816. (13) Simpson, H.; Berthemy, A.; Buhrman, D.; Burton, R.; Newton, J.; Kealy, M.; Wells, D.; Wu, D. Rapid Commun. Mass Spectrom. 1998, 12, 75-82. (14) Goerlach, E.; Richmond, R.; Lewis, I. Anal. Chem. 1998, 70, 3227-3234. (15) Harrison, A. C.; Walker, D. K. J. Pharm. Biomed. Anal. 1998, 16, 777783. (16) Hegy, G.; Goerlach, E.; Richmond, R.; Bitsch, F. Rapid Commun. Mass Spectrom. 1996, 10, 1894-1900. (17) Hempenius, J.; Wieling, J.; Brakenhoff, J. P. G.; Maris, F. A.; Jonkman, J. H. G. J. Chromatogr., B: Biomed. Sci. Appl. 1998, 714, 361-368. (18) Janiszewski, J.; Schneider, R. P.; Hoffmaster, K.; Swyden, M.; Wells, D.; Fouda, H. Rapid Commun. Mass Spectrom. 1997, 11, 1033-1037. (19) Krakowski, K.; Bunville, J.; Seto, J.; Baskin, D.; Seto, D. Nucleic Acids Res. 1995, 23, 4930-4931. (20) Oldenburg, K. R.; Zhang, J.-H.; Chen, T.; Maffia, A., III; Blom, K. F.; Combs, A. P.; Chung, T. D. Y. J. Biomol. Screening 1998, 3, 55-62. (21) Llewellyn, L. E.; Doyle, J.; Negri, A. P. Anal. Biochem. 1998, 261, 51-56. (22) Stenroos, K.; Hurskainen, P.; Eriksson, S.; Hemmila, I.; Blomberg, K.; Lindqvist, C. Cytokine 1998, 10, 495-499. (23) Reichman, M.; Marples, E.; Lenz, S. Lab. Rob. Autom. 1996, 8, 267-276. (24) Schroeder, K. S.; Neagle, B. D. J. Biomol. Screening 1996, 1, 75-80. 10.1021/ac9813540 CCC: $18.00
© 1999 American Chemical Society Published on Web 05/07/1999
less than 1 min.25-31 Even turbulent flow chromatography with mass spectrometry has been reported with analysis times of less than 2 min.32 Each of these efforts has been directed at increased sample throughput. To further improve the sample analysis throughput of LC/ MS/MS techniques, we have developed fast separations of multiple compound mixtures in less than 30 s. In this report, a mixture of six benzodiazepines was selected to demonstrate feasibility of rapid biological sample preparation and fast analysis by selected reaction monitoring (SRM) LC/MS. These compounds represent an important class of anxiolytic-hypnotic drugs whose analysis in plasma and urine continues to be of concern and investigation33,34 and were among the first small-molecule combinatorial libraries to be generated.35-37 We report liquid-liquid extraction of these benzodiazepines (bromazepam, carbamazepine, estazolam, norfludiazepam, alprazolam, triazolam) in human urine in the 96-well format. Extracts in 12-96-well plates (1152 samples) were analyzed continually with 30-s run times completing the entire analysis in less than 12 h. The analytical system consisted of four autosamplers connected sequentially to one column and one tandem mass spectrometer. Analysis sequence control was accomplished through instrumental feedback from the mass spectrometer to the autosamplers. EXPERIMENTAL SECTION Materials. Extractions were performed in Matrix Technologies’ 1.1-mL 96-well disposable tube racks (Lowell, MA). The reconstituted filtered extracts were transferred to Beckman deepwell collection plates (Fullerton, CA). Bromazepam, carbamazepine, and estazolam were obtained from Sigma Chemical Co. (St. Louis, MO). Norfludiazepam, alprazolam, and triazolam were obtained from Radian International (Austin, TX). Deuterated benzodiazepines estazolam-d5, norfludiazepam-d4, alprazolam-d5, and triazolam-d4, used as internal standards, were from Radian International. HPLC grade acetonitrile and chloroform were from J. T. Baker (Phillipsburg, PA). Deionized water was generated in-house with a Barnstead Nanopure II filtration system (Boston, MA). Sodium carbonate was obtained from Sigma Chemical Co. and formic acid was 88% doubly distilled from GFS Chemicals (25) Murai, S.; Saito, H.; Nagahama, H.; Miyate, H.; Masuda, Y.; Itoh, T. J. Chromatogr. 1989, 497, 363-366. (26) Thomas, J. J. Chromatogr. 1989, 479, 430-436. (27) Graser, T. A.; Godel, H. G.; Albers, S.; Foeldi, P.; Fuerst, P. Anal. Biochem. 1985, 151, 142-152. (28) Kintz, P.; Lamant, J. M.; Mangin, P. Analyst (London) 1990, 115, 12691270. (29) Poetter, W.; Lamotte, S.; Engelhardt, H.; Karst, U. J. Chromatogr., A 1997, 786, 47-55. (30) Bischoff, K. LaborPraxis 1996, 20, 56-58. (31) Volmer, D. A.; Hui, J. P. M. Arch. Environ. Contam. Toxicol. 1998, 35, 1-7. (32) Ayrton, J.; Dear, G. J.; Leavens, W. J.; Mallett, D. N.; Plumb, R. S. Rapid Commun. Mass Spectrom. 1997, 11, 1953-1958. (33) Wolff, K.; Garretty, D.; Hay, A. W. M. Ann. Clin. Biochem. 1997, 34, 6167. (34) Azzam, R. M.; Notarianni, L. J.; Ali, H. M. J. Chromatogr., B: Biomed. Sci. Appl. 1998, 708, 304-309. (35) DeWitt, S. H.; Kiely, J. S.; Stankovic, C. J.; Schroeder, M. C.; Cody, D. M. R.; Pavia, M. R. Proc. Natl. Acad. Sci. U.S.A. 1993, 90, 6909-6913. (36) Bunin, B. A.; Plunkett, M. J.; Ellman, J. A. Proc. Natl. Acad. Sci. U.S.A 1994, 91, 4708-4712. (37) Bunin, B. A.; Plunkett, M. J.; Ellman, J. A. In Combinatorial Peptide and Nonpeptide Libraries; Jung, G., Ed.; VCH: Weinheim, Germany, 1996; pp 405-424.
Table 1. Benzodiazepines Analyzed, Their Range of Concentrations in the Standards, and the Transitions Monitored by SRM LC/MSa
a The structure of each analyte and its monoisotopic molecular weight is shown. b mw, monoisotopic molecular weight.
(Columbus, OH). The reconstituted urine extracts were filtered through 0.45-µm nylon membranes in 96-well 800-µL Unifilters from Polyfiltronics (Rockland, MA) using a Tomtec Quadra 96320 vacuum manifold (Hamden, CT). Sample Preparation. All solutions and solvents were dispensed to the 96-well plates using a Tomtec Quadra 96-320 workstation. Human urine, determined to be free of benzodiazepines by LC/MS/MS, was dispensed (400 µL) to each well of 12 96-well, 1.1-mL plates totaling 1152 samples. A stock solution of the benzodiazepines was diluted serially with 20:80 acetonitrilewater. The diluted standards (10 µL) were pipetted each to 400 µL of urine for final concentrations of 0.5, 1, 2, 5, 10, 20, 50, 100, and 200 ng/mL estazolam and alprazolam (and multiples of those concentrations for each of the others as indicated in Table 1) for calibration standards. Quality control (QC) samples were generated by diluting a different stock solution with control human urine for final concentrations of 7.5 (low QC), 75 (mid QC), and 150 ng/mL (high QC) estazolam and alprazolam (again with multiples of those concentrations for each of the other compounds as indicated in Table 1. Control human urine samples were spiked at levels corresponding to the lower standards to simulate unknowns that would be found in a typical benzodiazepine Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
2295
Figure 1. Diagram of the fluid control for multiplexing the mobile phase through each of the autosamplers.
administration. All fortification steps for standards, QCs, and spikes were added manually to each well using a Rainin EDPplus 25-µL pipettor (Emeryville, CA). Control urine (blank) was spiked to give a concentration of 50 ng/mL estazolam-d5, alprazolam-d5, and triazolam-d5 and 100 ng/mL norfludiazepam-d4. This control urine spiked with internal standards was added to all standards, QCs, samples, and blanks (100 µL), except double blanks, yielding a final concentration of internal standard at 10 and 20 ng/mL, respectively. To the double blanks, 100 µL of control urine not containing internal standard was added manually. Double blanks were dispersed randomly throughout the plates and contained neither the internal standard nor the targeted benzodiazepines. Extraction Procedure. The Tomtec robot was used for pipetting all solutions for the extraction procedure in the 96-well plates. To each well of the 12 plates containing 500 µL of urine standards, QCs, unknowns, blanks, and double blanks, 50 µL of 0.1 M sodium carbonate was added. The 12 plates were then covered with cap mats (Matrix Inc.) and shaken manually because a mixer was too gentle to agitate the aqueous solution. Centrifugation was performed at 2500 rpm (750g) for 2 min to ensure all of the solution returned to the bottom of each well using a Damon/ IEC Division HNS (Needham, MA) centrifuge. Next, 500 µL of chloroform was added to each well in two 250-µL aliquots and mixed by rotation (40 rotations/min) on a Fisher Brand hematological mixer for 10 min followed by centrifugation at 2500 rpm for 5 min. The chloroform solution (400 µL) was removed and dispensed to clean wells in 12 new plates. Since chloroform was the heavier layer, 50 µL of air was aspirated first followed by lowering the Tomtec workstation pipet tips to the bottom of the wells and then dispensing 20 µL of the air to remove any aqueous contamination. The chloroform was then evaporated to dryness under a gentle stream of dry nitrogen using an in-houseconstructed 96-nozzle blow-down manifold with the plates im2296 Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
mersed in a water bath maintained at 40 °C. The samples were reconstituted with 400 µL of acetonitrile-water (20:80) with the Tomtec and mixed by manual shaking followed by centrifugation at 2500 rpm for 2 min. Finally, each 400 µL of reconstituted sample was transferred to the filter plates via the Tomtec robot and suction filtered through 0.45-µm filters into clean 1.0-mL 96-well autosampler plates using the Tomtec vacuum manifold. The plates were sealed with adhesive plastic strips (3M, St. Paul, MN) and stored in a refrigerator at 4 °C. Each set of four plates was removed from the refrigerator and placed directly into the autosamplers for sequential analysis. Chromatography. The chromatographic system consisted of three SIL-10A/SCL-10A Shimadzu autosamplers (Shimadzu Scientific Intruments, Columbia, MD) converted to inject from one 96-well plate. The fourth autosampler was a Spark Holland Endurance (Emmen, The Netherlands). Each autosampler was equipped with a 50-µL injection loop. All tubing was stainless steel with 0.02-in. i.d. from the pump to the autosamplers and 0.01-in. i.d. from the autosampler to the column. A 0.005-in. i.d. PEEK tubing (Upchurch Scientific, Oak Harbor, WA) connected the analytical HPLC column to the turbo ion spray LC/MS interface. An isocratic mobile phase, consisting of 0.01% formic acid premixed in 33% acetonitrile-67% water, was delivered by one LC-10LS Shimadzu pump at a flow of 1.0 mL/min. The LC column was a Mac-Mod Rapid Resolution 2.1 mm × 15 mm cartridge packed with 3-µm particles of Zorbax SB-C18 (Hewlett-Packard Analytical, Chadds Ford, PA). The flow of the mobile phase from the pump to the autosamplers and the column was directed by a Rheodyne Selection Valve model 7066RV (Cotati, CA). Figure 1 shows the configuration of this switching system for multiple autosampler use. The switching valve was controlled via timed events on each of the autosamplers as described in instrument control (vide infra).
Table 2. Timed Sequence of Operation for Instrument Control of Multiple Autosamplers with a Single Column and One Tandem Mass Spectrometer sample no. 1 (well 1, plate 1) 2 (well 1, plate 2) 3 (well 1, plate 3) 4 (well 1, plate 4) 5 (well 2, plate 1) Figure 2. Block diagram of the control system designed to keep each analysis in synchronization between the mass spectrometer’s data collection system and each autosampler. In the diagram AS3 is running. At 25 s, AS3 switches Selection Valve to AS4. At 31 s run ends. At 38 s, MS ready signal causes ESB to start AS4, injection is made, and AS4 sends the signal to ESB to start MS.
Mass Spectrometry. The tandem mass spectrometer used in this work was a PE Sciex API 3000 (Concord, ON, Canada) equipped with a turbo ion spray LC/MS interface. The turbo ion spray LC/MS interface accepted the entire effluent from the LC column at 1.0 mL/min. Auxiliary, nebulizing, and collision gas was nitrogen obtained from a liquid nitrogen Dewar boil-off at 80 psi. The LC/MS interface was maintained at 450 °C, and the auxiliary gas flow was 7.5 L/min. The electrospray voltage was maintained at 4.0 kV. Nebulizing gas, curtain gas, and collision gas flows were at instrument settings of 12, 12, and 5, respectively. The mass spectrometer was operated in the positive ion SRM mode. Precursor ion selection was done with Q1 operating at 0.7Da peak width at half-height and product ions were selected by Q3 with 0.8-Da peak width at half-height. The six analytes and four internal standards were monitored with a dwell time of 25 ms plus a 2-ms pause each giving an overall data cycle of 270 ms. Optimal collision energy was determined and was set for each analyte. The ions to the transitions monitored are given in Table 1. LC/MS/MS acquisition time for each sample was 30 s with an additional 7-8 s for the mass spectrometer data system to load parameters and be ready to collect the next data file. Instrumental Control. Four autosamplers were required to eliminate delays between a complete SRM LC/MS run and the next injection. Each autosampler required a minimum of 1.5 min for each injection cycle. However, with data collection rates as fast as described in this work, there was a need to ensure that sample injection and data collection remained synchronous. To do so, each autosampler was set to load the next sample into the injection loop and wait for a “ready signal” from the mass spectrometric data system before making its next injection. A block diagram of this control is shown in Figure 2 and the sequence of events is given in Table 2. If either the mass spectrometer or the injector failed, analysis would stop at that point because the mass spectrometer could not send the MS ready signal; thus the injector would not start. As shown in Figure 2, if the autosampler did not inject, then it could not signal the MS to start. To achieve this, the MS ready signal was multiplexed using
time sample (min) time (min) 0:00 0:25 0:31 0:38 1:03 1:09 1:16 1:41 1:47 1:54 2:19 2:25 2:32 2:57
0:00 0:25 0:31 0:00 0:25 0:31 0:00 0:25 0:31 0:00 0:25 0:31 0:00 0:25
event autosampler 1 injects selection valve switches to AS 2 finish sample 1 autosampler 2 injects selection valve switches to AS 3 finish sample 2 autosampler 3 injects selection valve switches to AS 4 finish sample 3 autosampler 4 injects selection valve switches to AS 1 finish sample 4 autosampler 1 injects selection valve switches to AS 2
an in-house-designed and -constructed Boolean logic circuit made with common CMOS logic chips from Radio Shack (Fort Worth, TX). The circuit simply moved the signal to the next autosampler when the MS ready signal was output. As Table 2 indicates, this would then initiate the remaining sequence of events from the autosampler. At the end of each run, the autosampler would advance the solvent selector valve (Figure 1) so that liquid flow would be routed to the next autosampler. RESULTS AND DISCUSSION The fast separation of mixture components can be achieved with short columns having small particle size. Separation of the six benzodiazepines with some coelution of norfludiazepam and alprazolam was obtained in less than 25 s. Separation could be achieved even faster but at the expense of higher liquid flow rates delivered to the mass spectrometer. Increase in organic modifier caused loss of chromatographic resolution. Under the conditions used, polar compounds such as metabolites would elute in the column void and be separated from the analytes. Although the turbo ion spray interface can accommodate flows as high as 2.0 mL/min, sensitivity would likely be compromised at the higher flows. Also, the dwell time for each ion would need to be shorter to provide enough data points required to adequately represent the chromatographic peaks for area integration. The four autosamplers are necessary to eliminate wait times resulting from each autosampler preparing itself for the next injection, e.g., needle seat wash, needle wash, and injector loop loading. With the 30-s run time established in this example when each autosampler had injected, rinsed, and loaded the next sample, it was then ready to inject again about 10-20 s before the mass spectrometer had completed data collection for the autosampler injection preceding it. Thus, there is no wait time in this setup from the time the mass spectrometer sends the ready signal to the next injection. A Gilson 215 Autoprobe with 889 injector could accomplish this cycle time. If only one typical autosampler was used in this application, the mass spectrometer would be required to wait more than 60 s at the end of each 30-s run for the autosampler to be ready for the next injection. This could be shortened with some commercial autosamplers but at the expense of needle and injection port rinses. These rinses are necessary to reduce carry-over commonly observed in the sensitive detection that SRM LC/MS provides. Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
2297
Figure 3. Total selected reaction ions for the six benzodiazepines. Traces are from (A) plate 4, sample 24; (B) plate 8, sample 408; and (C) plate 12, sample 1132. The concentrations are bromazepam 100 ng/mL, carbamazepine 20 ng/mL, estazolam 10 ng/mL, norfludiazepam 50 ng/mL, alprazolam 10 ng/mL, and triazolam 30 ng/mL.
Although the number of samples that can be run in a day can be calculated from an individual sample analysis time, the only way to show real feasibility of this fast, high-throughput analysis is to actually carry out the experiment. We have set, in this initial study, a goal of greater than 1000 runs in 1 day and exceeded that goal in less than 12 h. Figure 3 shows the chromatographic response from sample 24 in the fourth plate, sample 408 in the eighth plate, and sample 1132 in the twelfth plate. These chromatograms show that the system maintained its integrity with respect to chromatographic peak shape and retention time through the course of the analysis. Note that sample 24 resides on plate four because each well is injected from each plate serially. Sample one is from plate one, well one; sample two is from plate two, well one, and so on. Interestingly, the only errors encountered were injection failures on run 1 and run 383. No other mechanical errors occurred. Figure 4 shows representative extracted ion chromatograms from sample 408. This plot shows the separation achieved in the 30-s analysis time. The coelution of alprazolam with norfludiazepam and triazolam (panels D-F in Figure 4) causes the three components to appear as one peak in the total ion chromatogram. However, if alprazolam were not present, norfludiazepam and triazolam would be observed as chromatographically separated in Figure 3. Column pressure was continually monitored and noted at the beginning of the analyses to be 140 bar. By the end of plate four, column pressure had increased from the initial value to 300 bar. When the first four plates were removed from the autosamplers and prior to the installation of the next four plates, the HPLC column was flushed with 100% acetonitrile for 5 min. The column was equilibrated with mobile phase whereupon analysis of the next four plates commenced. The pressure was then 160 bar and remained near that value for the analyses of the remaining eight plates. It is believed that these precautionary steps were not necessary, but these procedures can ensure a successful experience through such demanding experiments and show that monitoring of the system may be important whether by an operator or through automation. 2298 Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
Clearly, sample preparation must be such that matrix components will not build up on the column or eventually degrade the mass spectrometer’s performance. The liquid-liquid extraction procedure described in this work, with the necessary final filtering step for the samples, met these requirements. The inclusion of multiple standard spikes bracketing a concentration range from above the low calibration standard to near the high calibration standard also allowed for the examination of quantitative response of each analyte from plates 1 through 12. Figure 5 shows the calibration curves achieved for plates 1 and 12 for the standards at the beginning and the end of each plate. This represents the two different types of autosamplers, the Shimadzu systems and the Endurance. These data suggest the robustness available from these fast high-volume sample analyses. However, the autosamplers used did not perform at the same level of proficiency. The Shimadzu autosamplers (comparatively old models) showed carryover in the blanks and some chromatographic peak tailing probably associated with the cause of the carry-over. The calibration curves shown in Figure 5 are representative of the entire 12 plates. They show good reproducibility from the beginning to the end of the entire 1152-sample set when looking only at either standards at the beginning or end of each plate. They show a slight bias between the standards at the beginning of each plate and at the end of each plate. The linear fit (indicated by the correlation coefficient, r2) for the standards at the end of each plate is also better. The cause of the observed bias was not identified. Since the plates were not allowed to equilibrate to room temperature before they were run, the higher concentration standards may have had some precipitation that did not redissolve before they were analyzed at the beginning of each plate. No standards were rejected in the calculation and all samples and QCs concentrations for each plate were determined by the weighted (1/x) least-squares regression curve derived from the duplicate result of the standards at the beginning and end of each plate. Because of the bias, the QC results were skewed. Figure 5 does not represent the calibration used and is presented only to provide insight into the results obtained. The results shown are
Figure 5. Individual calibration curves for plates 1 and 12 of alprazolam, standards at the beginning of the plate and at the end. The unweighted linear least-squares regression for each curve is also shown. The correlation coefficient of the regressions for plate 1, standard one is r2 ) 0.985; for plate one, standard two is r2 ) 0.999; for plate 12, standard one is r2 ) 0.992; and for plate 12, standard two is r2 ) 0.998. These curves are presented for evaluation of the robustness of the methodology. They were not used for the calculation of QC sample concentrations (see text).
Figure 4. Selected reaction ion chromatograms of each of the benzodiazepines for run 408. The traces are (A) bromazepam 100 ng/mL, (B) carbamazepine 20 ng/mL, (C) estazolam 10 ng/mL, (D) norfludiazepam 50 ng/mL, (E) alprazolam 10 ng/mL, and (F) triazolam 30 ng/mL.
directed at demonstrating feasibility and not validating the method. However, the data treatment of the QC samples for precision and accuracy does follow accepted procedures. Retention times remained consistent throughout the entire 12-h run. Again, there was a difference between the Shimadzu autosamplers and the Endurance. At the flow rate used, the Shimadzu autosamplers experienced a pressure pulse at injection that shortened the retention times as compared to the Endurance. Despite this difference, the retention times for a given analyte showed good precision throughout the 1152 samples. This is represented in the plot shown in Figure 6 for norfludiazepam run on all the autosamplers. Blanks and samples where norfludiazepam was not detected were excluded from the graphic representation. To assess the analysis of this many samples by SRM LC/MS in the 12-h time period, QC samples interspersed throughout the plates were run with n ) 7 at each of three different levels. Table 3 shows the precision and accuracy obtained for all 12 plates at these three levels (n ) 84). Precision is represented by the relative standard deviation (RSD) as percent of the mean. Accuracy is given by the difference of the calculated concentration from 100%
Figure 6. Chart of the retention time for norfludiazepam run on the Endurance autosampler and the Shimadzu autosamplers. The total number of samples plotted was 697 for those run with the Shumadzu autosamplers and 239 for the Endurance autosampler. All blanks and samples with calculated concentrations below the low standard were excluded from the plot.
of the expected concentration and is shown as percent error. The tabulation shows the overall quality of the data for the entire 12-h run. Precision is acceptable for all but the low levels of bromazepam and carbamazepine. The error is high only for the low and mid levels of triazolam. The consistent positive bias is due to the low response obtained for high standards run at the beginning of every plate as described (vide supra). This is observed in Figure 5 and is representative of most of the plates. Each plate contained seven QC samples at each of three levels, near the detection limit, at the middle of the calibration range, and at the high end of the calibration range. There were 84 QC samples at each level. Outliers were not included in the overall determination of RSD and percent error using the high- and low-quartile value plus and minus 1.5 of the interquartile range criterion.38 Table 4 shows Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
2299
Table 3. Precision (% RSD) and Accuracy (% error) of the Quality Control (QC) Samples Included on Each Plate Determined Using Calibration Curves Constructed from Standards Located at the Beginning and End of Each Plate low QC
mid QC
high QC
% RSD % error % RSD % error % RSD % error bromazepam carbamazepine estazolam norfludiazapam alprazolam triazolam
23.4 22.4 12.8 15.5 12.2 13.4
3.0 15.7 0.6 15.0 7.0 43.4
16.7. 16.8 14.3 14.1 10.7 8.7
3.1 9.4 2.5 15.2 7.5 21.1
13.8 16.6 12.9 11.7 10.1 8.2
8.0 6.7 5.9 14.0 6.0 1.5
Table 4. Precision (% RSD) and Accuracy (% error) of the Quality Control (QC) Samples Included on Each Plate Determined Using Calibration Curves Constructed from Standards Located Only at the End of Each Plate low QC
mid QC
high QC
% RSD % error % RSD % error % RSD % error bromazepam carbamazepine estazolam norfludiazapam alprazolam triazolam
18.1 16.0 13.1 16.4 12.4 14.7
-4.2 11.1 -1.0 33.2 0.4 7.3
11.1 13.7 12.0 16.8 9.2 11.8
-4.0 6.8 4.5 10.5 -0.1 5.4
16.8 13.9 12.9 19.6 10.5 15.0
-0.4 4.8 5.9 -7.8 -1.4 4.1
these same results using only the standards at the end of each plate for calibration. The accuracy represented in this tabulation shows the randomness around 100% that would be expected in an unbiased calibration. The relatively high error value for the low QC of nurfludiazepam (33.2%) shown in Table 4 is due to the lower response at the high end of the calibration curve for the high standards resulting from saturation. The bias from the standards at the beginning of each plate compensated for the effect of these standards. However, no standards were omitted in the data treatment of the results shown in Table 4. Again, this treatment of the data was not directed at method validation, but to show feasibility of the fast analysis. Method validation would require the analysis of precision and accuracy separately for each plate with the QC samples on that plate. The data presentation in Table 3 summarizes all the results. Given the quantity of data obtained from this fast analysis, this presentation was done for brevity. The precision and accuracy of the QC samples show that the fast SRM LC/MS used can generate acceptable results for large numbers of samples. The acceptance criteria for method validation of fast methods may well be better suited for drug discovery and combinatorial libraries than clinical trials.39 However, we believe that a focused effort could improve the quality of results to meet modern GLP criteria. CONCLUSIONS We have reported the high-throughput analysis of 1152 human urine extracts in less than 12 h using fast SRM LC/MS. This initial (38) Ott, L. An Introduction to Statistical Methods and Data Analysis, 4th ed.; Duxbury Press: Belmont, CA, 1993. (39) Gilbert, J. D.; McLoughlin, D. A.; Olah, T. V. Methodol. Surv. Bioanal. Drugs 1998, 25, 235-245.
2300 Analytical Chemistry, Vol. 71, No. 13, July 1, 1999
feasibility study suggests that this approach is possible and that 120 samples/h throughput could become as routine as today’s 30-60 samples/h capability to meet increasing analytical demands. The major limitation of this technology is not the chromatography or the mass spectrometry, but rather sample handling, preparation, data analysis, and related issues concerning reduction of quality and error-handling procedures. The approach used here involved liquid-liquid extraction with filtration in the 96-well format and multiplexing four autosamplers to one SRM LC/MS system. This allowed continuous 30-s runs with an overall cycle time or duty cycle of ∼37 s. This could have been decreased by reducing the mass spectrometer’s data collection time to 23 s because both the chromatography and the autosamplers could have accommodated the shorter duty cycle. Instrumentation and postacquisition data analysis have not yet been designed to handle the fast cycle rate and high data production volume afforded by this approach. Improved data-processing systems will have to be developed to facilitate data reporting and quality assurance (QA) analysis of such high data volume in the future. The need to improve both sample handling and sample preparation for high volumes of samples has not been discussed. Additionally, the amount of data generated must be processed with reliable automated software that includes error checking for quality assurance. The criteria for this processing should be the subject of further investigation as these fast methodologies are promulgated. The system described does, however, facilitate the fast cycle times that have been presented. A single autosampler that can cycle at less than 15 s and eliminate carry-over would be highly desirable. Until the advent of an autosampler injection system that meets or exceeds these criteria, multiple autosamplers that have been optimized to eliminate carry-over can make maximum use of an expensive yet very capable LC/MS/MS system. The Gilson 215-889 multiple injection system may perform at this same level although we have no experience to verify this. It is clear to us there is a need for higher bioanalytical throughput and that SRM LC/MS techniques can provide this capability. By implementing modern sample preparation strategies and integrating simple changes to affect shorter chromatographic run times, such as short HPLC columns of smaller internal diameter maintained at mobile-phase flow rates 5-10 times their optimum linear velocity, we can increase sample throughput significantly. We intend to continue efforts toward routine very high throughput sample analysis by reengineering the entire process, e.g., sample handling, sample preparation, LC/MS analysis, data handling, and reporting. We believe further improvements may be realized via miniaturization of certain aspects of this process. ACKNOWLEDGMENT We gratefully thank the Eastman Kodak Co., HoffmanLaRoche, and SmithKline-Beecham for financial support. We also thank Tomtec, PE Sciex, and Jones Chromatography for their generous loans of the Quadra 96 Workstation, the API-3000 mass spectrometer, and the Endurance autosampler, respectively. Received for review December 7, 1998. Accepted March 17, 1999. AC9813540
