Assay Development and Screening of Human DGAT1 Inhibitors with an LC/MS-Based Assay

Reagents and materials

The reagents and materials used for this study are listed in the following. 1,2-Dioleoyl-sn-glycerol (diolein) and oleoyl coenzyme A (oleoyl-CoA) were obtained from Avanti (Alabaster, AL). Pentanol and DMSO were obtained from Fisher Biotech (Subiaco, WA). 1,2,3-Tri(cis-9-octadecenoyl)glycerol (or glyceryl trioleate/triolein) and glyceryl tripalmitate (or tripalmitin) were obtained from Sigma (St. Louis, MO). The internal standard 1,1,1-¹³C-glyceryl trioleate (or 3-¹³C-glyceryl trioleate or ¹³C-triolein) was obtained from Cambridge Isotope Laboratories (Andover, MA). Deep-well polypropylene 384-well plates were purchased from Greiner Bio-One (Longwood, FL). All other chemical reagents and solvents were obtained from Sigma or stated otherwise.

Enzyme solution and substrate mixture preparation

The enzyme used in this assay was prepared as membrane fraction from Sf9 cells overexpressing human (His)₆DGAT1 (hDGAT1). An aliquot of frozen recombinant hDGAT1 protein membrane was thawed and diluted 1:10 with the enzyme buffer (100 mM HEPES, 100 mM MgCl₂, pH 7.5) and homogenized briefly 5 times on ice by a Wheaton Potter-Elvehjem homogenizer. After homogenization, the enzyme was further diluted in the same enzyme buffer to 0.025 mg/mL as working concentration for the DGAT1 assay. Diolein and oleoyl-CoA were dissolved to a stock concentration of 10 mM in 100% ethanol and 10 mM sodium acetate (pH 4.5), respectively. Then, these 2 portions were diluted and mixed gently to give a substrate mixture with working concentrations of diolein and oleoyl-CoA both at 100 µM.

Assay development and validation

The assay procedure and initial condition were adapted from an earlier report for DGAT with radioactive detection. ¹⁷ The DGAT assay was modified and conducted in 384-deep well plates. The amount of hDGAT1 enzyme used in the assay was determined by a 9-point enzyme titration series (1:2 serial dilution [v/v] using the enzyme buffer in a range from 0.125-0.0009 mg/mL). Substrate titrations were carried out with varied amount of each substrate (diolein or oleoyl-CoA, both from 400-3.12 µM). After the enzyme and substrate titrations, a time course was retested to confirm the linear range of reaction response under the selected enzyme and substrate conditions. In addition, a known DGAT1 inhibitor, LCD344, was examined for inhibition of hDGAT1 activity, and IC₅₀ values were obtained from an 11-point dilution series, ranging from 3.3 µM to 0.056 nM (1:3 serial dilutions). To further validate the assay for its hit-finding capability and data reproducibility in screening, assay plates with randomly spiked LCD344 at 3 concentrations (corresponding to ~ 90% inhibition, 50% inhibition, and 10% inhibition) were tested twice in 2 separate runs, and the results were compared.

Procedure of DGAT1 inhibitor screening

First, 384-deep well assay plates were predotted in each well with 2 µL of compound (or DMSO in control wells in columns 23 and 24). Then, 25 µL of the enzyme working solution and, 10 min later, 25 µL of substrate mixture were dispensed to each well using a WellMate™ liquid dispenser (Matrix Technology, Hudson, NH). The final screening concentration for each compound was 8 µM for the orthogonally pooled mixtures (10 compounds/mixture) or 10 µM for compounds in singleton. The reaction was allowed to proceed at room temperature for 30 min. After incubation, reaction was stopped by adding 10 µL of 10% acetic acid containing 5 µM 3-¹³C-triolein (internal standard) to each well. Afterward, 50 µL pentanol was added in all wells using a WellMate™. The plates were sealed with “nonsticky” aluminum seal with a PlateLoc™ plate sealer (Velocity11, Menlo Park, CA) and then put on a plate shaker (Barnstead lab, Dubuque, IA) and agitated with maximum speed for 5 min to extract triolein into the organic phase. In negative control wells (in column 24), the same stop solution was added before the reaction started. The positive control wells (in column 23) contained all the components of reaction with only the corresponding amount of DMSO. The final concentration of DMSO in all wells was 3.8%, at which concentration DMSO has no significant effect on the enzymatic reaction.

LC/MS/MS setup, analysis methods, and data processing

LC/MS/MS analyses were performed on a MUX-LC/MS/MS system operating in a 4-way parallel mode, and the MS operating parameters were the same as described previously. ¹⁵ This system consisted of a Gilson 215/889 parallel autosampler, 4 Agilent binary high-performance liquid chromatography (HPLC) pumps, and a Micromass Quattro Micro with a 4-way MUX source ( Fig. 1 ). For each sample, a 5-µL plug of the organic phase (i.e., pentanol) from liquid-liquid extraction in the assay plates was injected onto a Thermo Betabasic C4 guard column (2.1 mm inner diameter × 20 mm total length, 5-µm particle size) and eluted isocratically (1.6 mL/min) using the mobile phase (51.5% isopropyl alcohol, 35% acetonitrile, and 13.5% water containing 0.1% ammonium formate). Two 6-port 2-way Rheodyne valves were used to divert the mobile phase for 0.2 min, at which time the primary Agilent HPLC pump triggered the valve. Multiple-reaction monitor (MRM) chromatograms of triolein (m/z 903>604) and 3-¹³C-triolein (m/z 906>606) (monitoring the corresponding [triolein-NH₄ ⁺] and [3-¹³C-triolein-NH₄ ⁺] ions) were collected for 0.55 min. The analysis was formatted for 384-well plates, with each plate analyzed in 110 min (i.e., 17 s per sample, including cycle time). Two such LC/MS/MS detection systems were used for the HTS sample analysis. The chromatograms were processed using Quanlynx software, which divided the area of individual triolein peaks by the area of the internal standard 3-¹³C-triolein peaks. The resulting values were reported as the relative response for the corresponding sample.

OPEN IN VIEWER

Fig. 1.

Schematic diagram for the MUX-LC/MS/MS system setup used in DGAT1 screening. The insert shows the multiple-reaction monitor chromatograms of product detected from each injection cycle.

Compound percent inhibition was calculated by the following formula: Inhibition % = 1 – [(Compound Response – Low Control Response)/(High Control Response – Low Control Response)] × 100%. The screening data were processed with the Novartis internal data analysis software XTS and also visualized by Spotfire™ (Somerville, MA). For the orthogonally pooled compound mixture, the primary hits at or above the designated hit inhibition cutoff values were deconvoluted in a similar fashion as described previously. ²²

LC/MS-based DGAT1 assay development

The enzymatic assay was optimized for both the reaction kinetics and facilitation of the LC/MS detection. First, suitable DGAT1 enzyme and substrates concentrations used in the assay were determined via titration. The selected substrate concentrations for diolein and oleoyl-CoA, 50 µM each, were in the vicinity of the apparent K_m values determined experimentally (data not shown). The enzyme titration and reaction kinetics are shown in Figure 2 . From these experiments, the enzyme concentration and the assay reaction time were set at 25 µg/mL and 30 min, respectively, based on the linear range of the response curves. The hDGAT1 enzyme activity was quite tolerant to DMSO and ethanol and was not significantly affected by up to 10% DMSO or 20% ethanol (data not shown). However, the reaction was sensitive to acidic conditions, as 2% acetic acid would completely stop the reaction.

OPEN IN VIEWER

Fig. 2.

(A) Enzyme titration curve and (B) time course for DGAT1 assay. In (A), the experiment was performed in the reaction with substrate concentration at 100 µM of both diolein and oleoyl-CoA. Enzyme concentration at 0.025 mg/mL was selected for the assay. Each point represent mean ± SD for n = 6 replicates. In (B), reactions were performed for 0, 5, 10, 20, and 30 min at 0.025 mg/mL of total protein. Results are mean ± SD for n = 3 replicates.

Reduction of assay data variation

With the originally adopted assay and detection protocol, the triolein measurement by LC/MS/MS showed wide data variation among replicate wells in an assay plate, with a coefficient of variation (CV) typically in the range of 20% to 30% for the full-reaction response signal (see Fig. 3A ). Therefore, the data variation had to be reduced significantly before the assay could be used for screening. Realizing that there were several liquid handling steps involved in the assay, efforts had been focused on optimizing the process for enhancing the reproducibility of the measurement. Further tests showed that liquid dispensing, solvent extraction, and internal standard all affected the data variation. A comparison of the assay data variation before and after optimization is shown in Figure 3A,B . The data variation (CV) in the finalized assay process is reduced to typically 5% to 10% for the positive control signal in 384-well plates ( Fig. 3B ). A few key determinants in process optimization are described and discussed below.

OPEN IN VIEWER

Fig. 3.

Assay data variation reduction and reproducibility test. Scatter plot of an assay plate data (A) before and (B) after assay optimization is shown. The high and low signal control data are highlighted in blue and green, respectively. In (C), 1 blinded plate was tested twice to check the detection repeatability. The data in wells with spiked inhibitor LCD344 in 3 (high, medium, and low) concentrations are represented in red, blue, and green, respectively, and data in DMSO wells (high controls) are shown in black. Dashed line represents the diagonal line of the correlation plot. Experiments were performed under the final assay conditions (see text for details).

Assay validation for HTS

The IC₅₀ value of a known DGAT1 inhibitor, LCD344, a pyrrolo(1,2-B)pyridazine derivative originally discovered at Tularik (San Francisco, CA), was determined to be in the range of 32 to 50 nM, which was comparable to values obtained by using the original assay protocol. ¹⁷ To evaluate the assay hit-finding capability and reproducibility, a plate with randomly spiked LCD344 at 3 concentrations (corresponding to ~90%, 50%, or 10% inhibition) was tested blindly in 2 separate runs. Figure 3C showed the result of this test. The result suggested that the assay was capable of identifying hits of a varied inhibition range with good reproducibility. To further evaluate the data variability and data uniformity, several plates with positive and negative control wells were tested. The CV in the positive control data was below 8%, and the assay quality statistic Z′ factors ²⁴ for these plates were all above 0.7, indicating the assay data quality being excellent for screening. In addition, a pilot screening of a small subset of compounds (at 10 µM compound) also yielded similar Z′ values.

Screening singleton vs. mixture compounds with LC/MS-based assay

The throughput of an LC/MS-based assay has been a major technical hurdle in screening a large chemical library consisting of several hundred thousands to a million compounds in a reasonably short time (e.g., several weeks). One approach to increase the throughput of LC/MS analysis is via parallel analysis with multiple LC setups. Other improvements, including the use of an optimized and simplified assay procedure and screening logistics, can also enhance the throughput. With our current 4-way parallel MUX-LC/MS/MS detection at 17 s per sample (slightly less than 2 h per 384-well plate) in the DGAT1 assay, further enhancement of throughput is needed for large-scale screening. To achieve higher throughput for the LC/MS-based screening, we have employed the use of orthogonally pooled compound mixtures. ²³ In this compound compression strategy, each compound in the orthogonally pooled plates is present twice in the pooled library, in 2 different mixture contexts (one by row and one by column). Therefore, compounds are tested essentially in duplicate, but the overall throughput in the case of 10 compounds per mixture is still 5-fold higher than testing compounds in singleton. The primary hits can then be deduced and identified by a self-deconvolution process, based on the compounds present in the 2 different active mixtures.

Figure 4 shows the scatter plots of DGAT1 screening data from a set of screening plates of both single compound (10 µM) and pooled mixture (10 compounds per mixture, 8 µM each compound). The slightly different compound concentrations used in this case (i.e., 10 µM vs. 8 µM) were mainly due to logistic consideration in compound addition. The scatter plot of the pooled mixture showed noticeably wider activity dispersion, and the mean activity was skewed slightly toward the inhibition side. This phenomenon was due to the so-called bystander effect of mixture (biological activity of a mixture affected by compound interaction, interference, or crowdedness in the mixture). Although the exact nature of this effect is not clear, it seems to be a common phenomenon, with severity of the effect varying in different assays. The pooled matrix strategy usually works well in screening with high-quality biochemical assays. ²² In this DGAT1 LC/MS screening, there was clearly an adequate activity window to identify hits from the pooled compound screening data ( Fig. 4B ). The use of the pooled mixtures also made LC/MS-based screening more cost-effective and efficient.

OPEN IN VIEWER

Fig. 4.

Scatter plots of DGAT1 primary screening data. The negative response values in this plot represent inhibition. Overlay of (A) the first 25 plates (384-well) as screened in singleton and (B) the first 200 plates as screened in the orthogonal mixture of 10 compounds/well (B) is shown. The compound sample wells and full inhibition control wells are dots in green and blue, respectively. Each compound was screened at 8 or 10 µM, as described in the text.

Results from DGAT1 inhibitor screening

A total of 773,000 compounds were tested in the primary screen in this LC/MS-based screening format, with 16,200 compounds tested in singleton (screening at 10 µM) and 756,800 compounds tested in pooled mixtures (screening at 8 µM for each compound). With 2 identical MUX-LC/MS/MS apparatus setups for sample analysis, the screening was completed in 8 weeks. The overall data quality of the screening was high, with a Z′ factor for the majority of assay plates in the 0.7 to 0.8 range. A very small number of plates with Z′ factors <0.5 were repeated.

From the primary screening, approximately 4600 primary active wells were identified with a hit limit cutoff set at 60% inhibition. After hit deconvolution and a triage process, which applied several in silico compound filters to eliminate compounds with less attractive structures or undesired properties, 2022 compounds were selected and retested in dose-response curves. In total, 1119 hits yielded reasonable dose-dependent inhibition curves and were subjected to further evaluation. The hit confirmation rate for this LC/MS-based screening was approximately 51%. Among these confirmed hits, 196 compounds (~17% of the confirmed hits) showed sub-micromolar potency (IC₅₀ <1 µM). Figure 5 shows the dose-response curves from a few selected inhibitors identified and confirmed from this screening. The confirmed hits contained a variety of structural scaffold series, including chemotypes with both known (identified from a previous in-house DGAT1 screening with FlashPlate™ assay format) and novel structures (data not shown), indicating good hit-finding capability with the current LC/MS-based screening format.

OPEN IN VIEWER

Fig. 5.

Dose-response curves of 4 representative example hits with different potencies from DGAT1 liquid chromatography/mass spectrometry screening. Negative activity values in this plot represent inhibition.

Through optimization of assay conditions and screening process, a high-quality LC/MS-based DGAT1 assay has been developed, and the assay procedure and logistics have been optimized and simplified. Furthermore, by combining parallel LC/MS/MS detection with the use of a orthogonally pooled compound mixture, this assay is apt for high-throughput mode and has been applied successfully in the DGAT1 inhibitor screening campaign with nearly 800,000 compounds. These results demonstrate the feasibility of using LC/MS-based biochemical assays for large-scale primary screening. This screening technology platform is particularly useful for biological targets for which the conventional HTS assays are difficult to implement.