MS System for

Anal. Chem. 2001, 73, 5691-5697

An Integrated Ten-Pump, Eight-Channel Parallel LC/MS System for Automated High-Throughput Analysis of Proteins Bingbing Feng,* Michael S. McQueney,† Tara M. Mezzasalma,‡ and J. Randall Slemmon§

Department of Gene Expression and Protein Biochemistry, Discovery Research, GlaxoSmithKline, King of Prussia, Pennsylvania 19406

An integrated 10-pump eight-channel LC/MS system has been developed for automated high-throughput analysis of intact proteins in recombinant protein purification processes. The key features of the system include (1) a compact 10-pump HPLC module that uses two pumps to generate a binary gradient and 8 pumps to deliver the mixed gradient to eight independent flow channels; (2) a TOF mass spectrometer with an eight-channel multiplexed ESI interface, which records separate data for all eight channels over each HPLC run cycle; and (3) highly automated data processing software that allows unattended calculation of protein molecular weight (in Da) from original mass spectral data (in m/z). This system was used in the routine screening of fractions from preparative scale chromatography to monitor the purification process with the required mass accuracy and throughput. As an example, the production and purification of an acylated protein with a molecular weight of 9 kDa is described. Using this off-line approach, it is practical to fully characterize protein-containing fractions from column chromatography with an overall analytical throughput of 1 min/protein sample with minimum operator involvement.

INTRODUCTION Drug discovery is dependent upon the production and purification of recombinant proteins at a relatively large scale. Additionally, the quality of the purified protein must be high, because it is used for a number of tasks, including activity screens and crystallography. This is in contrast to traditional proteomics, in which a number of elegant mass-spectrometry-based strategies have been developed for identifying small amounts of many proteins in a short period of time.1-3 The identification and analysis of recom* Corresponding author. Phone: 610-270-4991. Fax: 610-270-7359. E-mail: [email protected]. † Current affiliation: Department of Systems Research, Discovery Research, GlaxoSmithKline. ‡ Current affiliation: Department of High Throughput Screening, 3-Dimensional Pharmaceuticals, Exton, PA 19341. § Current affiliation: Department of Genomics, Pharmacia Corporation, Skokie, IL 60077. (1) Godovac-Zimmermann, J.; Brown, L. R. Mass Spectrom. Rev. 2001, 20, 1-57. (2) Aebersold, R.; Goodlett, D. R. Chem. Rev. 2001, 101, 269-295. (3) Pandey, A.; Mann, M. Nature 2000, 405, 837-846. 10.1021/ac0106187 CCC: $20.00 Published on Web 10/26/2001

© 2001 American Chemical Society

binant proteins differs in that a single known protein is tracked starting with its production during the fermentation process and then continuing through the many fractions created during the multistep chromatography-based purification. Traditionally, the purification steps include affinity-tag-assisted, ion-exchange, and reverse-phase column chromatography at a preparative or semipreparative scale. The critical requirement for a mass-spectrometry based strategy that can support the purification of a recombinant protein is rapid throughput of the many protein fractions so that the recombinant protein can be found quickly and the presence of contaminants can be determined. The ultimate goal is to obtain a recombinant protein product of high purity, because contaminants can significantly influence the results of compound screening, assay development, crystallography, or other operations. Conventional bioanalytical techniques, such as polyacrylamide gel electrophoresis, protein assay, and spectroscopy can be timeconsuming or lack the resolution needed to distinguish the correct protein from similar-sized contaminants. Commercial MALDI (matrix-assisted laser desorption ionization) MS instruments often have inadequate resolution in the identification of small chemical modifications on large intact proteins, and single column LC/MS strategies have insufficient throughput for fraction analysis. A number of pharmaceutical companies and contract research organizations (CROs) have developed high-throughput parallel LC/MS and LC/MS/MS approaches for ADME/Tox (absorption, distribution, metabolism, excretion, and toxicity) research and DMPK (drug metabolism and pharmacokinetics) studies to characterize organic compounds in biological matrixes.4-16 In terms of analytical precision, mass accuracy, lower limit of (4) Yang, L.; Mann, T. D.; Little, D.; Wu, N.; Clement, R. P.; Rudewicz, P. J. Anal. Chem. 2001, 73, 1740-1747. (5) Bayliss, M. K.; Little, D.; Mallett, D. N.; Plumb, R. S. Rapid Commun. Mass Spectrom. 2000, 14, 2039-2045. (6) De Biasi, V.; Haskins, N.; Organ, A.; Bateman, R.; Giles, K.; Jarvis, S. Rapid Commun. Mass Spectrom. 1999, 13, 1165-1168. (7) Van Pelt, C. K.; Corso, T. N.; Schultz, G. A.; Lowes, S.; Henion, J. Anal. Chem. 2001, 73, 582-588. (8) Gao, V. C. X.; Luo, W. C.; Ye, Q.; Thoolen, M. J. Chromatogr., A 1998, 828, 141-148. (9) Wang, T.; Zeng, L.; Cohen, J.; Kassel, D. B. Comb. Chem. High Throughput Screen. 1999, 2, 327-334. (10) Peng, S. X.; King, S. L.; Bornes, D. M.; Foltz, D. J.; Baker, T. R.; Natchus, M. G. Anal. Chem. 2000, 72, 1913-1917. (11) Watt, A. P.; Morrison, D.; Locker, K. L.; Evans, D. C. Anal. Chem. 2000, 72, 979-984. (12) Wu, J.-T.; Zeng, H.; Qian, M.; Brogdon, B. L.; Unger, S. E. Anal. Chem. 2000, 72, 61-67.

Analytical Chemistry, Vol. 73, No. 23, December 1, 2001 5691

quantitation, and data reproducibility, these approaches were shown to be practical means of eliminating analytical bottlenecks for the small molecule screenings. However, high-throughput LC/ MS characterization of recombinant proteinssoften the drug targetsduring purification and scale-up production processes has not been fully explored. We recently extended the application of a high-throughput, eight-channel, parallel LC/MS system for fast off-line characterization of intact proteins in column fractions in protein purification processes.17 Using the commercial system, we were able to analyze column fractions of an acylated protein (8.9 kDa) and to identify a contaminating byproduct that differs in size by less than 20 Da with a throughput of 50 samples an hour. In summary, the system used one binary HPLC unit to deliver the gradient elution buffer, through an eight-way splitter, to an eight-port autosampler/ injector. In each HPLC run, eight samples were simultaneously injected into eight reverse-phase columns linked to eight UV detectors. After UV, each channel was connected to the corresponding port of an eight-way multiplexed ESI interface. A timeof-flight (TOF) mass spectrometer was used to acquire the raw mass spectral data, from which molecular weights of the desired protein and coeluting modified proteins were calculated. Although it was a high-resolution and high-throughput approach, there were certain disadvantages associated with the system. First, a simple eight-way splitter generally cannot provide the desired even split of the elution buffer across all eight channels. It has been recognized as a challenge to obtain an even split across just four channels with a splitter in a similar setup.4 Second, as a result of composition variations of fraction samples, it has become more difficult to achieve HPLC retention time and MS data reproducibility of standards after a relatively short period of time (sometimes after less than a hundred analyses), which resulted in increased system maintenance and optimization and, thus, reduction of actual “in-service” time. Third, unlike smallmolecule identification by LC/MS or LC/MS/MS, in which the targeted molecular weights and single/multiple reaction monitoring transitions can be easily displayed without intensive data processing, the conversion of the charge series (in m/z) of a protein into its molecular weight (in Da) is a calculation-intensive process. The earlier edition of a commercial software (MassLynx, version 3.4, Micromass U.K. Ltd., Manchester, U.K.) could process individual raw mass spectral data, but it did not allow automated batch processing of such mass spectral deconvolution for a set of data files generated by the high-throughput analyses. It was not uncommon that we sometimes spent more time and effort on manually operated data processing than automated data acquisition. In this paper, we describe a new integrated 10-pump, eightchannel, parallel LC/MS system equipped with improved software that can overcome several disadvantages of the earlier systems, (13) Bakhtiar, R.; Tse, F. L. S. Rapid Commun. Mass Spectrom. 2000, 14, 11281135. (14) Korfmacher, W. A.; Veals, J.; Dunn-Meynell, K.; Zhang, X.; Tucker, G.; Cox, K. A.; Lin, C.-C. Rapid Commun. Mass Spectrom. 1999, 13, 1991-1998. (15) Xia, Y.-Q.; Whigan, D. B.; Powell, M. L.; Jemal, M. Rapid Commun. Mass Spectrom. 2000, 14, 105-111. (16) Eichhold, T. H.; Bailey, R. E.; Tanguay, S. L.; Hoke, II, S. H. J. Mass Spectrom. 2000, 35, 504-511. (17) Feng, B.; Patel, A. H.; Keller, P. M.; Slemmon, J. R. Rapid Commun. Mass Spectrom. 2001, 15, 821-826.

5692

Analytical Chemistry, Vol. 73, No. 23, December 1, 2001

such as flow-split instability and HPLC retention time irreproducibility, as well as labor-intensive data processing. The key features of the system include (1) a compact 10-pump HPLC module that uses 2 pumps to generate a binary gradient and 8 pumps to deliver the well-mixed solvent to eight independent flow channels; (2) a TOF mass spectrometer with an eight-channel multiplexed ESI interface, which records separate data for all eight channels over each HPLC run cycle; and (3) an automated data processing protocol that allows unattended calculation of protein molecular weight (in Da) from raw mass spectral data (in m/z). This highly automated system has significantly shortened analytical time for identifying desired proteins with mass accuracy, throughput, and operational convenience unachievable by other traditional bioanalytical methods or previously documented parallel LC/MS instrumentations. This highly automated parallel LC/MS system allows much faster characterization of fractions from column chromatography, thereby enabling a general approach for using LC/MS in the detection and analysis of proteins. As an example, we were able to optimize the production and purification of an acylated protein substrate with an overall analytical throughput of 1 min/column fraction sample using the off-line LC/MS approach. EXPERIMENTAL SECTION Reagents and Samples. HPLC-grade water, acetonitrile, ACSgrade trifluoroacetic acid (TFA), horse heart myoglobin, and alcohol dehydrogenase from bakers yeast were purchased (SigmaAldrich, Milwaukee, WI) and used without further purification. The acyl carrier protein (ACP) samples used in LC/MS were ionexchange chromatography column fractions collected during routine protein purification processes at GlaxoSmithKline. Protein fractions were added sequentially to wells in a Costar 300-µL V-bottom 96-well plate (Corning, Corning, NY) immediately before parallel LC/MS analysis. Water and myoglobin solution were added to the first and second eight wells of the 96-well plate in each sample group as blank and calibration standard, respectively, to monitor data accuracy. Synthesis and Purification of Crotonyl-ACP. Crotonyl-ACP was prepared in vitro by enzymatic coupling of Escherichia coli apo-ACP and crotonyl-CoA using E. coli ACP synthase.18 The reaction progress was monitored at 30-min intervals by LC/MS. After 2 h, the final reaction mixture was diluted to 250 mL with picopure water, and a portion was loaded onto a 6-mL Resource Q anion-exchange column (Amersham Pharmacia Biotech, Piscataway, NJ) for purification. The column was eluted with a shallow linear gradient from 200 to 600 mM NaCl over 10 column volumes. One-milliliter fractions were collected and analyzed by parallel LC/MS to identify those containing crotonyl-ACP. Ten-Pump Eight-Channel Parallel HPLC Module. The Jasco PAR-1500 (Jasco U.K. Ltd., Essex, U.K.) system is a highpressure gradient parallel pumping system designed for highthroughput parallel chromatography at the semipreparative and analytical scale (Figure 1). It utilizes two high-flow-rate (20 mL/ min maximum) semipreparative HPLC pumps (model PU-1586, PA and PB, Figure 1) to deliver a binary gradient to two sequentially connected dynamic mixers (HG-1580, M1 and M2). (18) Fawcett, T.; Copse, C. L.; Simon, J. W.; Slabas, A. R. FEBS Lett. 2000, 484, 65-68.

Figure 1. Flow diagram of the 10-pump eight-channel parallel HPLC system. This module contains two gradient pumps (PA and PB), eight delivery pumps (P1 through P8), two dynamic mixers (M1 and M2), two back-pressure regulators (B1 and B2) and one in-line filter (F1). It draws degassed solvents (buffers A and B) and delivers the gradient in eight streams to the eight injectors of the autosampler.

The first mixer (M1) is used to form the gradient buffer at low pressure, which is transferred to the second one with an in-line filter (F1) and back-pressure regulator (B1) (Upchurch Scientific, Oak Harbor, WA) between the two. The second mixer (M2) further blends the solvent in a chamber with 10 ports, one port for incoming mixed buffer, one for waste, and the other eight to the inlets of eight analytical HPLC delivery pumps (PU-1585, P1 through P8). Another back-pressure regulator (B2) is connected at the waste port to ensure pressurized buffer to the delivery pumps and that only excess solvent is diverted to the waste. The mixing chambers and solvent transfer lines between the pumps and injection ports of the autosampler were modified in-house to minimize dead volume. Each of the eight analytical HPLC pumps individually delivered the well-mixed gradient buffer to one of the eight ports of the Gilson 215/889 eight-channel liquid handler and multiple injection module with 50-µL injection loops (Gilson, Middleton, WI). The two gradient pumps provided a combined flow of 5 mL/min, and each of the delivery pumps operated at 0.5 mL/min. The additional 1 mL/min flow went to the waste. The pumps and mixers were stacked (Figure 2), occupying only a 19-in. (48 cm, width) by 20-in. (51 cm, depth) bench space and a total of 25 in. (64 cm) height. High-Throughput Parallel Liquid Chromatography/Mass Spectrometry. The parallel eight-channel LC/MS system was configured in a sequential fashion: HPLC module, autosampler, columns, UV detectors, splitters, and MUX-ESI-TOF mass spectrometer (Figure 2). The Gilson autosampler with eight injection needles can draw eight samples from a 96-well plate and inject them into eight separate reverse-phase columns (POROS R1/10, 2.1 × 30 mm or 2.1 × 100 mm, PerSeptive Biosystems, Framingham, MA) simultaneously. Eight Waters 2487 dual-wavelength UV detectors (Waters, Milford, MA) were used to monitor the UV absorbance of the protein elution in each channel at 214 nm wavelength. After the UV detectors, each LC stream was split with a Valco T-splitter (Valco Instruments, Houston, TX) so that only ∼150 uL/min per channel went to the inlet of an eight-channel multiplexed (MUX) ESI interface. An LCT time-of-flight mass spectrometer (Micromass U.K. Ltd.) was used to acquire the mass

spectral data. Separate UV data for each channel were also recorded using the MassLynx software, which controls the entire analysis: hardware operation, data acquisition, and data processing. The system hardware (excluding control computer and monitor) occupies a bench area about 9 ft long and over 2 ft deep, as indicated by the dimensions of each unit, drawn to scale in Figure 2. Note that columns and splitters take up minimal space. The eight-channel MUX ESI interface consists of eight sprayers, which are mounted on a support assembly; a desolvation gas heater probe, which delivers heated gas to the ionization enclosure; and a sampling rotor, which is positioned coaxially to the ion sampling cone at the entrance of the mass spectrometer. The schematic diagrams of a similar four-channel MUX interface have been published previously.4-6 Although all of the HPLC streams are electrosprayed simultaneously and ions are continually formed at the tip of each sprayer, the MUX interface allows sprayers to be sampled sequentially, one-by-one, in rapid succession. At any given time, only ions from one stream are admitted to the ion sampling cone, and ions from the other seven sprays are shielded. With a 100-ms dwell time for mass spectral data acquisition at each sprayer and an interscan delay time of 50 ms (minimum required) for the rotor to move from one sprayer to another, the TOF mass analyzer can acquire separate data for all eight channels over the entire HPLC run with a duty cycle of around 1.2 s. The mass spectrometer was operated in positive ion mode with 3 kV applied to the ES ionization sprayers, and sample cone voltage was set at 30 V. The source temperature was at 140 °C, and the desolvation gas was heated to 350 °C. Industrial-grade high purity nitrogen was used as the nebulizer and desolvation gas, introduced to the MUX source at a pressure of 100 psi, with flow rates of 300 L/hr and 1200 L/hr, respectively. Reverse-phase HPLC parameters were as follows: initial condition, 80% buffer A (water 95%, acetonitrile 5%, TFA 0.025%) and 20% buffer B (acetonitrile 80%, water 20%, TFA 0.025%); linear gradient 2090% buffer B in 2.8 min; total run length, 5 min. Automated Data Processing. As mentioned earlier, protein molecular weight must be calculated from its multiply charged ion series, obtained by summing mass spectra acquired over the corresponding LC peak when proteins eluted. A deconvolution program, MaxEnt, which is part of the MassLynx software, was used to provide a zero-charge MaxEnt spectrum from the unprocessed raw data.19 Deconvolution parameters, such as spectral resolution and damage model conditions, were identical for standards and proteins of interest in order to ensure calculation uniformity. To maximize the usage of the MS instrument, we used a standalone networked high-performance workstation (Compaq SP 750, 866 MHz, 2.0 GB RAM, 18 GB hard drive) to process calculationintensive MaxEnt deconvolution, so that the instrument control computer (Compaq AP 200, 500 MHz, 256 MB RAM, 18 GB hard drive) can be used to acquire additional data while previously acquired data are being processed off-line. Automated data processing protocols were set up using the OpenLynx program of the MassLynx software. For each acquired data file (one file/ HPLC channel) the program is able to identify the “peak” in the total ion current (TIC) chromatogram within a set time frame. A (19) Ferrige, A. G.; Seddon, M. J.; Jarvis, S. A. Rapid Commun Mass Spectrom. 1991, 5, 374-379.

Analytical Chemistry, Vol. 73, No. 23, December 1, 2001

5693

Figure 2. Schematic diagram of the 8-channel parallel LC/MS system. It consists of four major components: a 10-pump 8-channel HPLC module (Figure 1), an 8-channel autosampler, eight UV detectors, and a TOF mass spectrometer equipped with an 8-way MUX ESI source. The diagram is drawn to scale with dimensions (width, depth, and height).

raw mass spectrum was then summed across that peak, and the molecular weight of the protein from the charge series could be calculated. This latest edition (version 3.5) of the program allows the MaxEnt data processing routine be repeated on all of the acquired data, automating this previously labor-intensive calculation. Although the actual computing time was not dramatically reduced, there was no need to manually open each data file and repeat the routines, as the earlier versions of the software required. This is a major saving in manpower and makes the data processing facile. The OpenLynx program can also generate a final report for easy review of processed data using either computer. RESULTS AND DISCUSSION Evaluation of the 10-Pump Eight-Channel Parallel LC/ MS System. The 10-pump eight-channel parallel LC/MS system was evaluated using a mixture of two proteins, myoglobin (MYO) and alcohol dehydrogenase (ADH). To illustrate the independent flow characteristics for each individual channel, two groups of columns (four in each group, both 2.1 mm in diameter, but different in length, 30 mm or 100 mm) were used in the same HPLC elution. In each parallel LC/MS run of eight channels, UV absorption data (Figure 3) and TIC chromatograms (one example in Figure 4A) were obtained for all samples. Under our conditions, for 30-mm-long columns, MYO and ADH elution times were around 2.74 and 2.98 min, respectively, with baseline resolution, but for 100-mm-long columns, MYO and ADH elution times were around 3.26 and 3.44 min, respectively. Within each group, each channel can be individually optimized to obtain highly reproducible and consistent protein elution profiles, as shown. The different elution profiles between the two groups reflect primarily the difference in column length. An independently controlled flow path for each channel eliminated undesired cross-channel interference, which was often observed if a front-end multiway flow splitter was used, as in the original commercial design. For method development, this new system provides enhanced flexibility and uncompromised controllability; thus, it leads to highly reliable HPLC elution for all channels and allows simultaneous and independent evaluation of different columns for the same sample on a particular gradient. An example (Figure 4A) of a TIC chromatogram profile (30mm-long column) indicated band-broadening and delay in time 5694 Analytical Chemistry, Vol. 73, No. 23, December 1, 2001

Figure 3. Eight-channel parallel LC/MS chromatogram for one HPLC run of eight samples. UV absorbance chromatograms at 214 nm are shown. The 2.1 × 30 mm columns were used in channels 1-4; 2.1 × 100 mm columns were used in channels 5 through 8. Myoglobin (MYO) and alcohol dehydrogenase (ADH) eluted at 2.74 and 2.98 min, respectively, for the 30-mm columns; their retention times were 3.26 and 3.44 min, respectively, for the 100-mm columns.

(comparing to UV data) as a result of postcolumn flow-splitting between UV detectors and the MUX interface. Note that the fast 5-min HPLC gradient (20-90% buffer B in less than 3 min) was not optimized for the separation of these two particular proteins (MYO 17 kDa, ADH 37 kDa), but was used primarily as a procedure for sample desalting. In fact, this fast gradient has been routinely and successfully applied to the majority of our protein samples (either 5 or 50 kDa) from ion-exchange or size-exclusion chromatography in which each column fraction often contained one protein or a few proteins of similar size along with a large amount of salt. In our applications, the ability to quickly evaluate

Figure 4. (A) Corresponding total ion current (TIC) chromatogram of one of the UV traces (channel 1, Figure 3). Mass spectral data for myoglobin and alcohol dehydrogenase were recorded at 2.84 and 3.08 min, respectively, for the 30-mm column. (B) Raw mass spectrum obtained by summing individual mass spectra between 2.75 and 3.00 min. The ion series of 9+ through 24+ charge states of myoglobin is shown. (C) Deconvoluted mass spectrum showing the molecular weight of the measured myoglobin at 16 951.0 Da. A few other minor species were also found in the 16 500-17 500 Da mass range.

Analytical Chemistry, Vol. 73, No. 23, December 1, 2001

5695

the overall composition profile of each fraction is more important than the baseline separation of each individual protein in a particular fraction. By summing the individual mass spectra acquired during the narrow time frame during which a protein elutes, for example, the TIC peak around 2.84 min (Figure 4A), a raw mass spectrum is obtained (Figure 4B) that shows the multiply charged ion series of a specific protein. Here, a charge series of myoglobin ions, consecutively from 9+ to 24+, was recorded between the 500 and 2000 m/z regions. Unlike small-molecule applications, in which the m/z value is often the protonated (or de-protonated) molecular species of interest, this spectrum does not show the composition profile of proteins, such as its molecular weight, or the relative abundance of other modified species. Using the aforementioned deconvolution program, the molecular weight of a protein can be calculated (Figure 4C), in this example, that of myoglobin (measured, 16951.0 Da; theoretical, 16951.5 Da). A few minor species, such as salt adducts, were also found in this sample. The calculation generated a composition profile of this particular sample in the 16 500-17 500 Da molecular weight range, which can be modified as needed. This calculation routine, from TIC to molecular weight profile can be manually applied repeatedly to reveal the composition profile of all of the analyzed samples in each eight-channel parallel LC/MS analysis cycle.17 The automated calculation protocols we recently set up (see below) permitted us to monitor the final results of the composition profile of all analyzed fractions, bypassing the labor-intensive manual data treatment, as illustrated (Figure 4). High-Throughput Parallel LC/MS Analyses of Chromatography Fractions from Crotonyl-ACP Purification. In bacteria, ACP functions as the carrier protein upon which fatty acids are synthesized before subsequent incorporation into membranes and other acylated moieties. Crotonyl-ACP is an important intermediate in the synthesis of fatty acids.20 In our study, the created crotonyl-ACP serves as a substrate for assaying enzymes involved in the fatty acid biosynthetic pathway. As such, gram quantities of highly pure crotonyl-ACP are required to support enzymology and high-throughput screening efforts. Purification of the crotonyl-ACP from the crude reaction product is accomplished by high-resolution anion-exchange chromatography (Figure 5). The purification of crotonyl-ACP from starting materials and other contaminants requires LC/MS analytical capabilities to monitor the identity of species within column fractions, because traditional ultraviolet and SDS-PAGE methods lack the requisite sensitivity, accuracy, and precision. In this example, fractions 8 through 39 in the purification were analyzed by high-throughput parallel LC/MS. The OpenLynx program allows automated batch processing of mass spectrum deconvolution after the completion of LC/MS analyses. We have set up a protocol by modifying processing parameters so that the program is instructed to identify a protein elution peak, from which a raw mass spectrum is then summed. Subsequent deconvolution calculates all present species across a preset molecular weight range (e.g., 8500-9500 Da). The protocol was executed in batch mode to automatically process all individual (20) Garrett, R. H.; Grisham, C. M. Biochemistry, 2nd ed.; Saunders College Publishing: New York, 1999; Chapters 24 and 25.

5696 Analytical Chemistry, Vol. 73, No. 23, December 1, 2001

Figure 5. Ion exchange chromatogram of crotonyl-ACP purification. Fractions 1-50 were collected; fractions 8-39 (wells 3A-6H, Figure 6) were analyzed by parallel LC/MS.

data files, and the final results can be displayed for easy review (Figure 6). The data display consists of three panels: a 96-well plate diagram and sample information; a deconvoluted mass spectrum of a sample in a selected well; and a TIC of that sample. The software will highlight the well(s) in the 96-well plate diagram if a target mass is identified. In our purification, crotonyl-ACP (MW 8917 Da, target mass) is the product of interest. In the final results display, the top panel indicated that this target molecule was found in wells 4C through 5E (fractions 18 through 28). The composition profile of a selected well (e.g., fraction 23 in well 4H) was shown in the center panel. Evidently, along with the desired product, a contaminant (9155 Da, later identified as holo-ACP) of significant abundance was also present in that fraction. In addition, another species, 8933 Da, was also found, although at a relatively low intensity. The molecular weight difference indicated oxidation or hydroxylation as a likely modification. It would be very difficult to obtain such detailed information using commercial MALDIMS (data not shown). The bottom panel is the TIC of fraction 23, from which the results shown in the center panel were calculated. By clicking each well in the diagram, the composition profile of all fractions can be quickly reviewed, and correct actions can be taken regarding further process of fractions that contain the product of interest. The overall analytical throughput, from the injection of the first eight samples to the final results display, is about 1 min/sample. With this integrated high-throughput parallel LC/MS system and automated data processing software, we have been able to quickly and easily provide accurate information about these purification fractions in a manner unachievable in the past by other means. Future Directions. As a tradeoff of high-throughput, the eightchannel MUX ESI source offers limited sensitivity compared to the interchangeable standard single-sprayer ESI source, most likely as a result of shortened spectral acquisition time and fixed sprayer tip position. Under our conditions, out of the 1.2-s MUX duty cycle, only 100 ms (about 8% of 1.2 s) is available for ion signals from a particular channel to be recorded, but during the remaining 1.1 s (about 92% of time), ions generated from the same channel are shielded by design. The complicated MUX interface tolerates virtually no optimization of sprayer tip positioning, because all hardware components are fixed in their positions. On the basis of our experience with the standard single-sprayer ESI

Figure 6. Display of automated data processing results of the parallel LC/MS analyses. The color-coded 96-well plate diagram (top panel) allows easy review of the final results. By highlighting each individual well, the composition profile (i.e., the deconvoluted mass spectrum over a preset range) of that particular fraction is illustrated (center panel). Here, the product, crotonyl-ACP (MW 8917 Da), the modified product (8933 Da), and a contaminant (9155 Da) are all present in fraction 23. The TIC of the corresponding fraction (bottom panel), from which the final results are calculated, is also displayed.

source, significantly better ion signals can be obtained when the sprayer tip is positioned at other locations under simulated MUX source conditions. Nevertheless, the MUX source served our purposes well, because our proteins were generally present at relatively high concentrations (often above 0.1 mg/mL) in a fraction. Improvements in the MUX design may resolve these concerns. For our system setup, the latest version of the MassLynx software cannot control all of the HPLC individual pumps or additional switching valves, which can be used to divert unwanted flow to waste at the beginning of a gradient. To the best of our knowledge, another deconvolution algorithm in the MassLynx software, named Transform, can also be used to generate a zerocharge molecular-weight spectrum from a multiply charged ion series. Although relatively less accurate in comparison to MaxEnt, the final results can be calculated much faster, in a fraction of the time that MaxEnt requires. If the algorithm can be improved and the process automated for batch analyses, the overall analytical throughput of the system will be further enhanced for the characterization of proteins. CONCLUSION We have demonstrated a new integrated 10-pump eight-channel parallel LC/MS system that combines the strengths of an

independently controlled multichannel HPLC, high-throughput parallel mass spectral data acquisition, and automated data processing for the characterization of intact proteins in a purification process. The system allows LC/MS analyses of samples in a highly automated fashion, providing results of unmatched accuracy in a short period of time. It is expected that this system platform can also be used in small-molecule high-throughput parallel LC/MS applications. ACKNOWLEDGMENT We thank Dr. Arun H. Patel, Department of Gene Expression and Protein Biochemistry, and Paul M. Keller, Department of Assay Methodology Development, GlaxoSmithKline, for their help in the synthesis and purification of various related ACPs used in this paper. We appreciate helpful suggestions regarding HPLC components from Dr. Muneo Saito, Jasco Inc., U.S. We also thank Dr. Andrew J. Organ, Department of Computational, Analytical, and Structural Sciences; and Gilbert F. Scott, Department of Gene Expression and Protein Biochemistry, GlaxoSmithKline, for critical reading of the manuscript and valuable suggestions. Received for review June 1, 2001. Accepted September 17, 2001. AC0106187

Analytical Chemistry, Vol. 73, No. 23, December 1, 2001

5697