A high-speed liquid chromatography/tandem mass spectrometry platform using multiplexed multiple-injection chromatography controlled by single software and its application in discovery ADME screening


Correspondence to: J. Zhang, Bristol-Myers Squibb, Research and Development, 5 Research Parkway, Wallingford, CT 06492, USA.

E-mail: jun.zhang1@bms.com



Multiplexed liquid chromatography (LC) coupled with multiple-injection-chromatogram acquisition has emerged as the method of choice for high-speed discovery bioanalysis, because it significantly reduces injection-to-injection cycle time while maintaining the chromatography quality. Historically, systems utilizing this approach had been custom built, and therefore relied on custom software tools to communicate with multiple vendor software for system control, which lacked transferability, flexibility and robustness.


In this study, we refined a multiplexed bioanalytical system previously reported, by implementing open-deck auto-sampler manifold and multiple-injection-chromatogram acquisition, all on a commercially available system with single software control.


As a result of these improvements, the developed LC/tandem mass spectrometry (MS/MS) method on the system was nearly three times faster than the previous method, while demonstrating comparable analytical accuracy, precision and robustness. This system has been evaluated for in vitro ADME screening assays including metabolic stability, CYP inhibition and Caco-2. The biological data generated on the developed system displayed good correlation with those from the previous LC/MS/MS approaches.


The developed platform demonstrated applicability to the in vitro screening assays evaluated and has been successfully implemented to support the high-throughput metabolic stability assay, with a significantly improved bioanalytical throughput, capacity and data turnaround. Copyright © 2013 John Wiley & Sons, Ltd.

Assessment of the drug-like properties of compounds, including absorption, distribution, metabolism and excretion (ADME) via in vitro screening assays, has become a frontline routine practice in drug discovery owing to its effectiveness in guiding lead optimization, and preventing selections of candidates with unfavorable properties for the later, more expensive drug development studies.[1-4] These plate-based profiling assays, commonly operated in a highly automated and parallel fashion, generate large numbers of samples for large numbers of discovery candidates, presenting challenges to the downstream bioanalysis using liquid chromatography/tandem mass spectrometry (LC/MS/MS), which is operated in a sequential manner for every sample injection. Therefore, there have been continuous efforts to improve bioanalytical speed in order to address the ever-increasing throughput and capacity requirements in ADME profiling. One such effort has been the application of multiplexed LC systems in place of the conventional single-stream LC prior to MS analysis.[5, 6] The multiplexed configuration allows for staggered sample injections and elution from multiple columns, followed by alternated MS detection of each of the multiplexed LC channels. The approach can theoretically reduce the injection-to-injection cycle time by n-fold (where n is the number of LCs multiplexed) by significantly reducing the mass spectrometer idle time, while at the same time maintaining the same chromatography integrity as on a single-column system. In addition to multiplexing, further speed-enhancing approaches include the reduction of auto-sampler overhead, utilization of ultra-high pressure liquid chromatography (UPLC), optimization of gradient, and the elimination of MS method downloading time for each injection by acquiring a multiple-injection-chromatogram (MIC) into the same data file. Taken together, these approaches have successfully reduced the bioanalytical cycle time from several minutes per injection to less than 1 min per injection for in vitro ADME support. Several custom-built systems with these features have been reported to date.[7, 8] However, these systems were typically built by integrating components from multiple vendors, and therefore the overall system control has to be performed by custom software that interacts with multiple vendor software. One additional complicating factor for ADME bioanalytical platforms is the integration with hardware/software solution for automated SRM method optimization, as that is the pre-requisite for SRM quantitation of a large number of compounds being profiled. The recently introduced Apricot Designs Dual Arm (ADDA) is the first commercially available system with all these features by integrating with DiscoveryQuant (DQ), a software tool by AB Sciex, to enable streamlined compound optimization, acquisition method creation and sample submission.[9]

The Aria system by Thermo Scientific has been extensively used for multiplexed bioanalysis with LC systems and mass spectrometers from all major vendors, with a single-point software control of multiplexing, running either multiplexed online solid-phase extraction via turbulent flow chromatography,[10, 11] or multiplexed LC without the online extraction.[12] When coupled with a Thermo mass spectrometer, Aria fully integrates with QuickQuan, a Thermo Scientific software tool for automated compound optimization, via infusion or flow injection analysis. We have previously reported the development and application of a multiplexed system employing Aria 1.6.2 on a Quantum Ultra mass spectrometer in support of a high-throughput protein-binding assay.[13] The version of Aria software used then only supported one data file per injection; however, the recently launched Aria 1.6.3 now supports MIC data acquisition. In this study, we report an optimized multiplexed bioanalytical platform by incorporating this new feature of MIC data acquisition to our system and developing a corresponding workflow to achieve fully automated and high-speed bioanalysis, all with single-point software control for the entire integrated system. The application of this system to high-throughput in vitro screening assays is discussed.


Chemicals and reagents

All assay control compounds, at their highest purity available, were purchased from Sigma Chemical Co. (St. Louis, MO, USA). HPLC grade water, acetonitrile, and methanol were purchased from EMD Chemicals (Gibbstown, NJ, USA). Formic acid, ammonium acetate, dimethyl sulfoxide (DMSO), sodium phosphate monobasic monohydrate, and sodium phosphate dibasic anhydrous were also purchased from EMD Chemicals. Hank's balanced salt solution (HBSS) was purchased from Sigma Chemical Co. Magnesium chloride and b-nicotinamide adenine dinucleotide phosphate (NADPH) were purchased from Sigma. Pooled human, rat, mouse, monkey and dog microsomes were purchased from BD Biosciences (San Jose, CA, USA).

Assay conditions

The assay conditions for metabolic stability, Caco-2 permeability and CYP inhibition have been previously described.[14-16] Briefly, in the metabolic stability assay, cytochrome P450 (CYP)-mediated metabolism of test compounds at 0.5 μM was evaluated in vitro using human, mouse and rat microsomes after 10 min incubation at 37°C. In the Caco-2 assay, the permeability of test compounds across differentiated monolayers of Caco-2 cells was measured to estimate the human intestinal permeability, and test compounds at 3 μM were incubated for 2 h before the assay samples were extracted and subjected to LC/MS/MS analysis. In the CYP inhibition assay, a specific probe substrate (midazolam at 2.5 μM) was co-incubated with test compounds in human liver microsomes for 5 min, and the formation of the CYP 3A4-specific metabolite (1'-hydroxymidazolam) was measured by LC/MS/MS to assess the inhibition of CYP 3A4 activity by test compounds. For all the assays evaluated, samples post-incubation were extracted using protein precipitation prior to LC/MS/MS analysis.

LC/MS/MS system and conditions

A modified Aria LX-2 multiplexed system (Thermo Scientific, CA, USA) was used as previously described.[13] Aria software (version 1.6.3, Thermo Scientific) controlled all the peripheral components, including two sets of independent 20AD binary pumps (Shimadzu Scientific Instruments, MD, USA), a valve interface module (VIM, Thermo Scientific) and a modified dual-arm CTC HTS PAL auto-sampler (Leap Technologies, NC, USA). Each arm of the auto-sampler had a dedicated injection valve mounted with a CTC Active Wash Station (Leap Technologies). After each injection, the injection valve was washed once respectively with water (wash solvent 1) and 50:50 water/ methanol (wash solvent 2). The auto-sampler sample holder in this study was an open-deck manifold containing eight positions for microtiter plates, instead of an enclosed cooling stack of sample drawers as previously described. A customized cooling solution by MeCour (Groveland, MA, USA) for the open-deck manifold was implemented to maintain temperature control and minimize solvent evaporation. The cooling manifold was located beneath the sample plate holder to maintain the plate temperature at the set point (10°C) without impacting the travel path or time of the auto-sampler arm. A TSQ Quantum Ultra triple quadrupole mass spectrometer (Thermo Scientific) equipped with a heated electrospray ionization (H-ESI) source was used. A universal set of source parameters was employed, including a spray voltage at 3.5 kV for positive ionization mode and –2.5 kV for negative mode, a vaporizer temperature at 450°C and a capillary temperature at 275°C. The sheath gas and auxiliary gas settings were 70 and 60 (arbitrary unit), respectively. The optimization of compound SRM MS/MS conditions was conducted on the system via automated infusion using QuickQuan software as previously reported.[14] For sample analysis, the column used was a fused-core Ascentis Express C18 (2.7 µm, 2.1 × 20 mm; Supelco, PA, USA) at a flow rate of 0.9 mL/min and the injection volume was 15 μL onto a 5 μL sample loop. Mobile phase A was 980:20:1 (v/v/v) water/acetonitrile/formic acid and mobile phase B was 980:20:1 (v/v/v) acetonitrile/water/formic acid. The gradient started at 5% B held for 6 s before ramping to 100% B in 6 s, which was held for 9 s at 1.1 mL/min, and ramped down to 5% B in 1 s at 0.9 mL/min, and held for 18 s for column re-equilibration. The total LC runtime was 40 s, and the active data window for mass spectrometric acquisition for each channel started from 11 s post-injection for 15 s.

Informatics workflow

The optimization process of compound SRM conditions, as previously reported,[14] was still an integral part of the current developed workflow, with no changes made from the previous system. The optimized MS/MS parameters were archived in the QuickQuan database and later retrieved for acquisition method creation and sample analysis. To address the change in acquisition mode from one injection per file to multiple injections per file, we revised the previously reported informatics workflow to accommodate the change. A scheme of the informatics workflow is presented in Fig. 1. An assay 'plate map' was first dispatched by the centralized compound management system, which contained the compound identity, molecular formula and well positions in the stock plate of DMSO. A sample injection sequence in text file was first created using a Visual Basic Application (VBA) macro developed in-house, defining the assay sample plate/well positions, and injection order. The text file was then imported into QuickQuan which subsequently created (1) MRM acquisition methods (.meth) for each individual compound, (2) processing methods (.pmd) for each compound, and (3) XCalibur sample sequence (.sld). During the sample analysis, Aria created a .cbt file with timing information for each injection. The chromatographic peak review was conducted using GMSU/QuickCalc (Alpharetta, GA, USA), which matched each analyte peak with the corresponding internal standard peak based on the timings within the Aria .cbt file. The resulted peak areas were exported to Excel for data uploading.

Figure 1.

Informatics workflow.


We previously reported an Aria 1.6.2 controlled, multiplexed LX-2 bioanalytical system consisting a twin-arm auto-sampler for injection, dual-column for staggered parallel chromatography, and LC/MS/MS acquisition via one data file per injection on a Thermo Quantum Ultra mass spectrometer.[13] In the development and implementation of the system, two areas were identified to be the rate-limiting steps to the injection-to-injection cycle time: the first was the auto-sampler pre-injection time (29 s) which included the time for opening and closing of the enclosed cool stack sample drawer for sample aspiration, and the second was the mass spectrometer reset delay (25 s) on the Quantum Ultra mass spectrometer for downloading an acquisition method and opening an acquisition file before each injection. To address the first problem, we replaced the drawer-style sample holders on the CTC auto-sampler with an open-deck manifold and eliminated the auto-sampler overhead time spent on opening and closing the drawers. The strategy effectively reduced the pre-injection time from 29 s to 14 s, leaving only the necessary traveling time of the auto-sampler arm to the designated wells for sample aspiration and to the injector for sample injection.

Collecting data of all the injections from a compound into a single file, or MIC data acquisition, could largely eliminate the 25 s mass spectrometer reset delay for each injection on the previous system using a Thermo Quantum Ultra mass spectrometer. Under this acquisition mode, a data file was created upon the first injection of each compound and remained open for all subsequent injections from this compound, thus eliminating the need to load the acquisition method or open a new data file for each injection. The detector reset delay was only applied to the first injection of each compound, cutting 25 s overhead for each subsequent injection. For many of the ADME assays when up to several dozens of samples are generated for the same compound, application of MIC data acquisition enabled significant runtime reduction. In terms of instrument operation, elimination of method downloading for each injection also reduced chances for communication errors between the peripheral modules and mass spectrometer, which in our experience could be problematic for analysis involving a large number of sample injections.

In addition to the aforementioned reduction of auto-sampler overhead and MS acquisition time, the chromatography gradient time was also trimmed down from 91 s on the previous system to 40 s on the current, with a corresponding reduction in column dimension from 2.1 × 30 to 2.1 × 20 mm to speed up re-equilibration. With all these implemented, the injection-to-injection cycle time achieved was 23 s, which approaches the theoretical maximum efficiency of the LX-2 system running this gradient (40 s/2 = 20 s).

The chromatographic review of MIC data thus acquired was not trivial, since the number of injections and the corresponding injection time had to be taken into consideration by the reviewing software to perform proper peak detection and integration. This was accomplished by using GMSU/QuickCalc software, which parsed the Aria sequence file (.cbt) to obtain this information and subsequently applied it to raw data files during peak integration. The unique data format allowed display of all the chromatographic peaks for a compound and corresponding internal standard altogether in a single view, greatly facilitating the data review process. QuickCalc provided flexibility in peak integration, allowing users the option of modifying chromatographic integration for an individual peak manually, or applying globally a universal set of integration parameters for all the injections of a compound, just as in the case of single file per injection. The utilization of GMSU/QuickCalc has been reported for reviewing MIC data generated on both a multiplexed LC/MS/MS system and an online SPE/MS/MS system.[8, 17]

To evaluate the performance of developed analytical methods and system in terms of reproducibility and carryover, repetitive injections of a reserpine solution in 50:50 water/ methanol at 100 nM were made using the multiplexed LC/MS/MS analysis and MIC data acquisition. The %CV for twelve consecutive injections of reserpine was 4.7% and 7%, respectively, on the two multiplexed channels; a blank solvent injection subsequent to the reserpine injections represented a carryover of 2.9% and 1.8%, respectively; both were comparable to the multiplexed platform we reported previously before the upgrade. The bioanalytical method accuracy, precision and dynamic range were assessed using propranolol prepared at five different concentrations in extracted microsomal matrix spiked with d7-propranolol as internal standard. Four sets of calibration curves of such were separately prepared, each sample of a given set was injected onto both of the multiplexed channels, and measurements from both channels were combined for the intra-assay calibration curve construction. As listed in Table 1, the method demonstrated adequate linear dynamic range (5–2000 nM), intra-assay accuracy, and inter-assay accuracy (96–114%) as well as inter-assay %CV (<6%). Figure 2 plots the internal standard areas acquired in a 13 h overnight run on both of the multiplexed channels. The robustness of the system and method was manifested by a %CV of 8% on each of the multiplexed channels for the measured signal of the internal standard during the entire course of the sample run.

Table 1. Performance of propranolol standard curve
  Propranolol nominal concentration (nM)
1Intra-assay mean (n=4)4.9564929292010
2Intra-assay mean (n=4)4.9584589851903
3Intra-assay mean (n=4)5.0585379681838
4Intra-assay mean (n=4)5.0554929481933
AllInter-assay mean (n=4)5.0564949571921
Figure 2.

Alprenolol internal standard variability from a CYP 3A4 inhibition assay.

The applicability of the platform for metabolic stability assay was evaluated with 288 randomly selected proprietary compounds incubated in six separate assays, each containing 48 compounds. The assay incubation was conducted in 96-well plates and protein precipitation was employed post-incubation for sample extraction, the resulted supernatant was transferred to 384-well plates for LC/MS/MS analysis. The assay generated 12 samples to be injected for each compound, which were grouped into a single file using MIC data acquisition. The %Remaining results were derived from the ratio of analyte to internal standard signal at T10 and T0 of incubation. For all the compounds screened, an MS acquisition window of 15 s was able to capture all the test compounds eluted by the generic chromatographic method with a gradient time of 40 s. Figure 3 represents the typical MIC chromatograms for a test compound. For analysis of each compound, the sample injection sequence was purposely arranged to have T0 and T10 samples of the same replicate injected to the same channel of the multiplexed system, to avoid any differential responses between the two channels for the analyte and the non-isotopically labeled internal standard. In addition, all the T10 samples for all three species were injected prior to the T0 samples to minimize the impact of carryover. The assay samples generated for all the test compounds were subjected to analysis with both the previous LX-2 multiplexed platform, and the current system acquiring data in MIC fashion. The %Remaining values obtained from both systems were compared and the results are presented in Fig. 4, which shows an excellent correlation between the two measurements. Bioanalytical runtime for the metabolic stability assay using the current platform was reduced significantly, from 14.6 to 4.6 min per compound, as illustrated in Fig. 5. This shortened runtime is a combined effect of reduced pre-injection time and MS reset delay, and it translates directly to an appreciable improvement in assay data turnaround, assay throughput and capacity. As illustrated in Table 2, the required runtime of a metabolic stability assay on the previous system allowed for only one assay analysis per overnight run on each instrument, and currently the overnight run can complete three assays.

Figure 3.

MIC data acquired for erythromycin from a metabolic stability assay.

Figure 4.

%Remaining correlation from metabolic stability assay.

Figure 5.

Bioanalytical runtime reduction for metabolic stability assay.

Table 2. Comparison of platforms
  Current platformPrevious platform
  • * MS reset delay of 25 s is only applied to the first injection of each compound.
Auto-samplerFormatOpen-deck plate holderEnclosed cool stack
Pre-injection time14 s29 s
LC methodMethod time40 s91 s
MS detectionFormatMICOne file per injection
Reset delay0 s*25 s
Acquisition window15 s42 s
Metabolic stability assayInjection-to-injection cycle time23 s73 s
Assay analytical time4.1 h11.7 h

Besides the metabolic stability assay, the applicability of the system was also evaluated for in vitro Caco-2 and CYP inhibition (CYP3A4, midazolam substrate) assays using respectively 48 and 224 randomly selected proprietary compounds. All the assay samples generated (in 384-well plates for CYP3A4, and 96-well plates for Caco-2) were injected onto both the current system and a comparing system routinely used for the bioanalytical support of the assays in our lab. Figure 6 illustrates correlations of apparent permeability for the PAMPA assay determined using the current system for analysis with those using the conventional non-multiplexed, single file per injection acquisition. Figure 7 presents correlation of %Inhibition for the CYP inhibition assay between the current multiplexed platform and a RapidFire (Agilent, Santa Clara, CA, USA) direct online SPE/MS/MS system. For both assays the results displayed acceptable linear correlation, demonstrating the feasibility of the platform to additional in vitro ADME assays.

Figure 6.

Correlation of apparent permeability from Caco-2 assay.

Figure 7.

Correlation of %Inhibition from CYP 3A4 inhibition assay.

The developed system and methods using multiplexed LC coupled with MIC data acquisition significantly reduced the injection-to-injection cycle time (23 s) than our previously reported platform. This cycle time approaches those reported for the new ADDA system[9] and RapidFire system,[17, 18] with both running direct online SPE in MIC mode. In comparison, advantages of this platform reported herein include its relatively lower cost, its compatibility with various LC and MS components from major vendors, and its single point software control by Aria. In addition, the current system offers flexibility in performing better chromatographic separation without any system modifications. Figure 8 illustrates partial separation of an analyte from its hydroxyl metabolite using the generic chromatographic conditions used for the metabolic stability assay. This chromatography resolution can be further optimized if needed, which would not be possible on a dedicated online SPE system.

Figure 8.

Separation of analyte from its hydroxyl metabolite from a metabolic stability assay sample.


A bioanalytical platform was successfully developed by adopting an open-deck manifold for auto-sampler, multiplexed LC/MS/MS with multiple-injection-chromatogram for data acquisition, all under a single point software control for the system operation. The platform displayed satisfactory bioanalytical performance and also generated comparable biological results for metabolic stability, Caco-2 and CYP3A4 inhibition assays to those from the previous LC/MS/MS platform. The implementation of this platform for the bioanalytical support of a metabolic stability assay greatly increased bioanalytical speed, throughput and capacity, and it could also be easily applied to the bioanalytical support of other in vitro ADME assays.