High-Throughput Screening for Enzyme Inhibitors Using Frontal

This work presents new frontal affinity chromatography (FAC) methodologies for ..... even though the column assay was fully saturated with ligand in t...
0 downloads 0 Views 506KB Size
Download PDF
Anal. Chem. 2005, 77, 6125-6133

High-Throughput Screening for Enzyme Inhibitors Using Frontal Affinity Chromatography with Liquid Chromatography and Mass Spectrometry Ella S. M. Ng,† Feng Yang,‡ Akihiko Kameyama,‡ Monica M. Palcic,§,‡ Ole Hindsgaul,§ and David C. Schriemer*,†

Department of Biochemistry and Molecular Biology, University of Calgary, Calgary, Alberta, Canada, Department of Chemistry, University of Alberta, Edmonton, Alberta, Canada, and Carlsberg Laboratory, Valby, Denmark

This work presents new frontal affinity chromatography (FAC) methodologies for high-throughput screening of compound libraries, designed to increase screening rates and improve sensitivity and ruggedness in performance. A FAC column constructed around the enzyme N-acetylglucosaminyltransferase V (GnT-V) was implemented in the identification of potential enzyme inhibitors from two libraries of trisaccharides. Effluent from the FAC column was fractionated, sequentially processed via LC/MS, and referenced to a similar analysis through a control FAC column lacking the enzyme. The resulting multidimensional data sets were compared across corresponding sample and control fractions to identify binders, in a semiautomated approach. A strong binder in the protonated form at m/z 795 was identified from the first library of 81 compounds, exhibiting an estimated Kd value of 0.3 µM. Other binders yielded Kd values ranging from 0.35 to 3.35 µM. To demonstrate the improvement in performance of this FAC-LC/MS approach over the conventional online FAC/MS approach, 15 compounds from this library were blended with a second library of 1000 synthetic trisaccharides and screened against GnT-V. All ligands in the 15-compound set were identified in this larger screen, and no ligands of greater affinity than compound 1 were found. Our results show that FAC-LC/MS is a reliable method for screening large compound libraries directly and useful for large-scale ligand discovery initiatives.

Most high-throughput screening tools applied during ligand discovery rely on massively parallel individual compound assays and require archives of individually stored library molecules. The scale of these screening events creates a significant barrier in the routine application of chemical diversity for ligand discovery, for developing drugs or affinity reagents in general. A goal of our laboratory is to simplify the screening of large mixtures and, in turn, relax the requirements for library creation. Key objectives * To whom correspondence should be addressed. Phone: 403-210-3811. Fax: 403-270-0834. E-mail:[email protected]. † University of Calgary. ‡ University of Alberta. § Carlsberg Laboratory. 10.1021/ac051131r CCC: $30.25 Published on Web 08/31/2005

© 2005 American Chemical Society

include screening mixtures using equipment that does not require scaling according to library size and developing robust methods for obtaining compound binding data independent of compound concentration. We are exploring the attributes of frontal affinity chromatography (FAC) in this regard. FAC operated online with electrospray mass spectrometry (FAC/MS) has been utilized to screen small mixtures of compounds to discover ligands for antibodies,1 lectins,2 and proteases.3 FAC, first developed by Kasai et al.,4,5 involves the preparation of a column with an immobilized receptor such as a protein. A mixture of compounds containing potential ligands is continuously infused into the column to the breakthrough point. Under an equilibrium model, the order of elution of compounds from the column represents their affinities for the receptor, with the tightest binding ligands eluting last. A multidimensional detector, such as a mass spectrometer, is necessary to detect specific binding events and discriminate between the coeluting compounds in a complex mixture, as MS allows the resolution of compounds in the effluent according to the uniqueness of their mass to charge values.1,6 Other approaches for mixture screening have been reported, including surface plasmon resonance,7 NMR,8 and an affinity selection-mass spectrometry method recently described by Annis et al.9 The unique strength of FAC rests in its ability to detect binding events in a concentration-independent fashion, making it applicable to the detection of ligands at concentrations significantly below their respective Kd values.6 FAC is a robust and scaleable technology; it is simple, generates highly precise binding data, and is tolerant of complex multiligand samples. A FAC column can be reused and prepared with very low capacities, thus minimizing protein and compound consumption.1,3 (1) Schriemer, D. C.; Bundle, D. R.; Li, L.; Hindsgaul, O. Angew. Chem., Int. Ed. 1998, 37, 3383-3387. (2) Zhang, B. Y.; Palcic, M. M.; Mo, H. Q.; Goldstein, I. J.; Hindsgaul, O. Glycobiology 2001, 11, 141-147. (3) Schriemer, D. C.; Hindsgaul, O. Comb. Chem. High Throughput Screening 1998, 1, 155-170. (4) Kasai, K.; Ishii, S. J. Biochem. 1975, 14, 261-264. (5) Kasai, K.; Oda, Y.; Nishikata, M.; Ishii, S. J. Chromatogr. 1986, 376, 3347. (6) Schriemer, D. C. Anal. Chem. 2004, 76, 440A-448A. (7) Malmqvist, M. Biochem. Soc. Trans. 1999, 27, 335-340. (8) Meyer, B.; Peters, T. Angew. Chem., Int. Ed. 2003, 42, 864-890. (9) Annis, D. A.; Nazef, N.; Chuang, C.; Scott, M. P.; Nash, H. M. J. Am. Chem. Soc. 2004, 126, 15495-15503.

Analytical Chemistry, Vol. 77, No. 19, October 1, 2005 6125

Scheme 1. Reaction Products for the Solid-Phase Synthesis of Library B

While FAC coupled to MS has been applied to library screening, the approach is inadequate for screening mixtures of hundreds or thousands of compounds. The rate of false negatives increases because of buffer incompatibilities, limitations in mass resolving power, and the occurrence of ionization suppression in general.6 Increasing compound concentration leads to problems of solubility and increased compound consumption. In this work, we report on the inclusion of a zonal liquid chromatography step applied to fractions of the FAC column effluent, prior to MS analysis, as a means of improving the compound screening rate and minimizing false negatives. Our work demonstrates that substantially lower compound concentrations can be processed, supporting the reliable screening of mixtures of at least 1000 compounds/assay. We assess the approach using a mixture of ∼1000 modified triasccharide acceptor analogues targeting N-acetylglucosaminlytransferase V (GnT-V), an enzyme regulating the branching pattern of N-linked oligosaccharides on glycoproteins.10 Increased expression of active GnT-V in mammary, hepatocellular, and pancreatic cancer11-13 has been reported, and it has been suggested that inhibition of GnT-V represents a potential for treatment of cancer.14 EXPERIMENTAL SECTION Materials. Recombinant soluble human GnT-V was produced as previously reported.15 The biotinylation reagent, sulfo-NHS-LCbiotin was obtained from Pierce. BA-ID self-pack medium, a slurry solution composed of polystyrene-based porous beads with im(10) Dennis, J. W.; Granovsky, M.; Warren, C. E. Biochim. Biophys. Acta 1999, 1473, 21-34. (11) Fernandes, B.; Sagman, U.; Auger, M.; Demetrio, M.; Dennis, J. W. Cancer Res. 1991, 51, 718-723. (12) Yao, M.; Zhou, D. P.; Jiang, S. M.; Wang, Q. H.; Zhou, X. D.; Tang, Z. Y.; Gu, J. X. J. Cancer Res. Clin. Oncol. 1998, 124, 27-30. (13) Nan, B. C.; Shao, D. M.; Chen, H. L.; Huang, Y.; Gu, J. X.; Zhang, Y. B.; Wu, Z. G. Glycoconjugate J. 1998, 15, 1033-1037.

6126

Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

mobilized streptavidin, was from Applied Biosystems. UDP N-acetylglucosamine (Na2UDP-GlcNAc) was obtained from Sigma. The acceptor substrate for GnT-V, β-D-GlcNAc-(1f2)-R-D-Man(1f6) β-D-Glc-O-octyl (OGMG) was from Chemica Alta Ltd. (Edmonton, Alberta, Canada). The following buffers were used: GnT-V storage buffer (50 mM MES), 150 mM NaCl, 0.005% NaN3, pH 6.5), and bicarbonate biotinylation buffer (50 mM sodium bicarbonate, 150 mM NaCl, pH 8.5). The elution buffer and makeup solution used in conventional FAC/MS studies were 25 mM ammonium acetate (pH 6.6) and 0.05% formic acid in acetonitrile, respectively. Two libraries containing 81 (library A) and 1000 (library B) trisaccharides were prepared in-house (University of Alberta). All compounds were synthesized through solid-phase protocols with modifications of the trisaccharide analogues occurring at the C-4′ position, as reported previously.16 Briefly, the azide-modified trisaccharide was loaded on trityl chloride polystyrene resin and converted to an amine with all hydroxyl groups protected. The amine was reacted with a range of acyl chlorides (library A) or extended via FMOC chemistry (library B, Scheme 1). Library A was prepared by pooling six sets of mixtures, each containing 10-15 compounds, and library B was synthesized as a single crude mixture. All compounds were cleaved and deprotected with trifluoroacetic acid, dissolved in water, filtered with a 0.22-µm nitrocellulose filter, lyophilized, and weighed. Library B was not tested for overall yield; thus, we can only claim a nominal 1000-compound library. Subsequent LC/MS analyses demonstrated a distribution of products with molecular weights ranging from 700 to 1450, indicating significant conversion of starting (14) Ko, J. H.; Miyoshi, E.; Noda, K.; Ekuni, A.; Kang, R. J.; Ikeda, Y.; Taniguchi, N. J. Biol. Chem. 1999, 274, 22941-22948. (15) Zhang, B.; Palcic, M. M.; Schriemer, D. C.; Alvarez-Manilla, G.; Pierce, M.; Hindsgaul, O. Anal. Biochem. 2001, 299, 173-182. (16) Lu, P. P.; Hindsgaul, O.; Li, H.; Palcic, M. M. Carbohydr. Res. 1997, 303, 283-291.

Figure 1. High-precision nano-FAC system. It consists of two positive-displacement pumps capable of regulated flow rates down to 10 nl/min, and a six-port valve.

material into intended product, but also into side products; the low end of the mass range indicates incomplete coupling. Nonetheless, 959 chromatographic peaks were observed within the targeted mass range (708-1183 u), suggesting a reasonable conversion of starting material into product. An aqueous stock solution of library A (1.7 mg/mL) was prepared by dissolution in HPLC grade water (Fisher Scientific) representing a nominal percompound concentration of 25 µM assuming an averaged molecular weight of 818. An aqueous stock solution of library B (22 mg/mL) was diluted similarly to 25 µM, assuming an average molecular weight of 902. In both cases, stocks were diluted to 0.25 µM with elution buffer in preparation for FAC analysis. Methods. Preparation of the GnT-V Column. Biotinylation and immobilization of GnT-V were performed as previously described,15 using 1.7 nmol of GnT-V. Biotin incorporation was estimated by matrix-assisted laser desorption/ionization mass spectrometry (Voyager-DE STR, Applied Biosystems, Foster City, CA) and shown to average a 10:1 stoichiometry of biotin to GnT-V. The activity of biotinylated GnT-V, determined by a previously reported radiochemical assay,17 was ∼20% of the unlabeled protein. Biotinylated GnT-V (100 µL) was diluted with MES storage buffer to 665 µL immediately after the biotinylation reaction and mixed with 15 µL of aqueous slurry of BA ID self-pack medium (50% w/v beads in H2O). The mixture was gently rotated at room temperature for 20 min and then incubated overnight at 4 °C. It was essential to immobilize biotin-GnT-V immediately, as the labeled enzyme loses activity rapidly in solution. (17) Palcic, M. M.; Ripka, J.; Kaur, K. J.; Shoreibah, M.; Hindsgaul, O.; Pierce, M. J. Biol. Chem. 1990, 265, 6759-6769.

BA beads containing streptavidin-immobilized GnT-V were packed in microcolumn hardware obtained from Upchurch Scientific (Oak Harbor, WA). It consisted of 5.9 cm × 250 µm i.d. fused-silica tubing (360-µm o.d.) fritted with a 1-µm stainless steel nanofilter capsule (Upchurch Scientific) on each end of the column. The column was packed using a pressure cell (Brechbuhler, Inc.), which allows slurry packing of microcapillary columns at a high pressure. Typically, a 10-µL slurry solution was diluted with water in an Eppendorf tube and placed in a reservoir of the pressure cell. The end of the capillary from the column was inserted into the tube through a nut at the top of the pressure cell and packed at a pressure of 1000 psi. An identical blank column was prepared similarly by packing the streptavidin beads without GnT-V and used as a control. FAC Fluidic System. The FAC system used in this work was an integrated module consisting of two high flow rate precision nanofluidic delivery systems and a six-port valve (Figure 1). Elution buffer was delivered by pump A at a flow rate of 1 µL/min through the six-port injection valve, where the library compounds were injected. The FAC column was connected from the outlet of the injection valve to a mixing tee. Makeup solution was introduced through pump B of the FAC system at a flow rate of 4 µL/min. In the online mode, combined flow was directed into an Agilent 1100 MSD single quadrupole electrospray mass spectrometer operating with Chemstation software (Agilent Technologies, Palo Alto, CA). The spectrometer was operated in selected ion monitoring in positive-ion mode. The capillary voltage was 3500 V, nitrogen drying gas flow rate was 5 L/min, the nebulizer pressure was 6 psi, and the gas temperature was set at 150 °C. Breakthrough volumes Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

6127

Figure 2. Schematic diagram of FAC-LC/MS configuration. FAC was performed offline prior to LC/MS analysis. FAC fractions were collected, evaporated to dryness, and reconstituted in mobile phase. These fractions were then injected on a HILIC column and gradient eluted into a QStar mass spectrometer.

were measured at the midpoints of the extracted ion chromatograms. In offline mode, fractions were collected for processing via LC/MS as indicated in Figure 2. Mixtures of library A or B (nominally 0.25 µM/compound) were prepared with 0.2 µM OGMG in the elution buffer and infused into the GnT-V column at a flow rate of 1 µL/min. Fractions were collected every 5 µL, evaporated to dryness in a SpeedVac (Savant, Inc.), and reconstituted in 10 µL of acetonitrile prior to analysis by LC/MS analysis. The same procedure was followed using a blank column identical in all respects, minus GnT-V. Compound separation was performed with hydrophilic-interaction chromatography (HILIC) with an 1100 Series Capillary LC system (Agilent Technologies) using a polyhydroxyethyl aspartamide column (0.32 mm i.d. × 50 cm, 3 µm, 100 Å) from PolyLC (Columbia, MD). Liquid chromatography was performed using acetonitrile containing 2% 6.5 mM ammonium acetate, pH 5.5 (mobile phase A), and 6.5 mM ammonium acetate, pH 5.5 (mobile phase B), at a flow rate of 15 µL/min with a gradient of 0-30% B over 30 min. Samples (5 µL) were injected by an autosampler. MS detection was accomplished with a QSTAR Pulsar i Hybrid quadrupole time-of-flight (TOF) mass spectrometer fitted with a TurboIonSpray source (Applied Biosystems). The ion spray voltage was set at 4500 V. TOF MS experiments were performed in positive ion mode over an m/z range of 300-1450. The LC/ MS system was operated under Analyst QS software (service pack 8). The LC/MS data were processed using Advanced Chemistry Development MS Processor software (ACD MS Processor, Version 8.11), to identify possible “hits” as described below. RESULTS AND DISCUSSION Function and Stability of the GnT-V Column Assay. Prior to the screening of the library mixtures, the active column capacity (Bt) and stability of the GnT-V column assay was determined by online FAC/MS using 10 µM Tris-HCl as a nonbinding reference compound and 1 µM OGMG as the binding ligand. The stability of the column capacity was determined by multiple infusions of the above mixture, interleaved with washes of 100 µL of elution 6128

Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

Figure 3. (A) XICs of the void marker Tris-HCl and OGMG at 10 and 1 µM respectively. (B) Multiple assessments of the column capacity (Bt) of the GnT-V column were performed over the course of a day. The FAC-MS run was repeated three times between each successive wash with 100 µL of elution buffer for the determination of Bt. A value of 13.99 ( 0.29 pmol was obtained. Kd for OGMG, 12.8 µM.15

buffer (34 column volumes). Figure 3A demonstrates a FAC/MS chromatogram for OGMG during the first infusion. Reassessing the column capacity after the washing step indicates the breakthrough time for OGMG remained unchanged between each infusion (Figures 3B). Assuming a dissociation constant (Kd) of 12.8 µM for the OGMG-GnT-V interaction,15 a Bt of 13.99 ( 0.29 pmol (n ) 5) was obtained. Fresh columns with Bt’s of 18 and 15 pmol were used for screening libraries A and B, respectively. We note that immobilization effectively stabilized the otherwise unstable biotin-enzyme. We have found that columns

Figure 4. Schematic of FAC effluent sampling strategy for both the GnT-V and control columns. The four insets represent LC/MS data for a strong ligand (m/z 795, [M + H]+) for successive fractions of FAC effluent, from library A. Blue traces represent XIC of the compound eluted from the GnT-V column, and red indicates the XIC from the blank column. The insets are referenced to idealized online FAC/MS chromatograms for the strong ligand (red, control column; blue, GnT-V column).

prepared in this fashion are stable for months at 4 °C, whereas the solution form of the biotin-enzyme was inactive after 12 h. Optimization of FAC-LC/MS on Library A. The mixture of 81 compounds represented by library A was assembled from smaller mixtures of known composition, to assess the capability of FAC-LC/MS methodology to minimize false positives and false negatives in the hit list returned from data analysis. Library A was infused at a concentration of 0.25 µM/compound through the GnT-V column assay, and nine fractions (A-I) were collected. In the FAC-LC/MS method, compound breakthroughs are represented as chromatographic peaks. The concept is shown in Figure 4, where the strongest ligand in library A elutes over fractions G and H. Including this new LC dimension in the FAC/ MS approach supports the detection of larger numbers of compounds at greater sensitivity, owing to the purification, preconcentration, and separation capability of the LC step. That is, we were able to detect all 81 compounds in library A in this fashion but not under conventional FAC/MS conditions even at 4 times the library concentration (data not shown). Operating at lower library concentrations reduces ligand competition and increases the Kd range accessible to the system. This in turn supports the estimation of dissociation constants from the FAC-LC/MS data, as will be shown below. Since compounds that bind nonspecifically in the assay would demonstrate breakthrough volumes greater than the void volume

of the assay system, and give a false indication of binding activity, the library was infused through a blank column and the effluent fractionated and processed via LC/MS, to provide a control for subsequent data analysis and hit identification. Figure 4 shows the analysis of the strongest ligand processed through the blank column, and the resulting peak intensity on a per-fraction basis. Notice that the breakthrough is observed over two fractions (G, H), suggesting a breakthrough volume of ∼35 µL. Figure 4 serves to highlight the goal of subsequent data analysis. For each FAC fraction in the GnT-V assay, extracted LC/MS ion chromatograms (XICs) across the entire mass range require comparison with the corresponding fraction from the blank run. Chromatographic peaks apparent in the blank but absent in the assay indicate the presence of a hit; the quality of the hit is in turn determined by the fraction where compound breakthrough occurs (quantification of hit quality will be discussed below). Corresponding fractions were compared and hits identified using the CompareLCMS algorithm in MS Processor, an approach involving application of the component detection algorithm (CODA),18 followed by the identification of mass chromatograms that differ between two highly similar samples according to a set of user-defined peak parameters. The CompareLCMS approach returns bin number/retention time pairs identifying these differ(18) Windig, W.; Smith, W. F.; Nichols, W. F. Anal. Chim. Acta 2001, 446, 467-476.

Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

6129

Table 1. CompareLCMS Parameters Used in Library Data Analysis Compare LCMS MCQ threshold ) 0.5 smoothing window width (scans) ) 5 noise level (counts) ) 20 peak width level (scans) ) 100 delta scan level (scans) ) 20 Ratio Level* ) 0.1, 5

peak picking minimum fwhm (scans) ) 5 S/N threshold ) 2.0 number of most abundant peaks picked ) 5

*A ratio level of 0.1 was used for library A and 5 for library B.

ences between corresponding fractions. CODA involves a prefiltering step based on the calculation of a similarity index (MCQ value) between a raw data set and its processed form, where higher index values approaching 1 are associated with higher confidence in chromatographic peak detection.18 MCQ values represent a composite score reflecting all features within a given XIC. Its primary function is to rank XICs according to “quality”, where it is implied that quality correlates with well-defined chromatographic peaks arising from sample compounds. XICs with MCQ values above a threshold are selected and passed forward for comparative analysis according to a ratio of intensities. Individual peaks within an XIC from a given fraction are selected according to thresholds (minimum intensity, maximum peak width) and matched with their corresponding peak in the second fraction according to a minimum acceptable retention time variability. The optimal settings used for the processing of libraries A and B are presented in Table 1. Overall, we determined that comparisons are most sensitive to the reproducibility of the LC system and method, given its relationship to the ratio level. Figure 5 shows that HILIC analyses of libraries A and B respectively were reproducible over 3 h, but drift imparted by the electrospray ion source prevented day-today comparisons. As a result, we collected LC/MS data for blank fractions immediately after corresponding assay fractions and employed a ratio level of 0.1 or greater in screening libraries, to identify chromatographic differences exceeding 10% (∼3-fold greater than the standard deviation in peak area). This procedure was successful in identifying 14 hits from library A, for a hit rate of 17% (Table 2). This table reflects the raw output of the comparative analysis, where each entry indicates an m/z value with at least one peak in its extracted ion chromatogram showing a significant difference versus the control. As expected from Figure 4, m/z 795 would be present in each column except fraction I, where complete breakthrough has occurred. Ligands were identified on the basis of exact masses of the hits, determined through visual inspection of the 14 XICs. The hits can be rank ordered according to the fraction corresponding to breakthrough, but it is difficult to bin the hits into approximate Kd ranges without additional information. The extreme nonlinearity of binding due to a multiplicity of ligands serves to accelerate the breakthrough of ligands.6 Even the nonhit compounds may possess a degree of aggregate binding to the target protein sufficient to accelerate the breakthrough of stronger ligands. However, it is possible to reliably measure the Kd range 6130 Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

Figure 5. Reproducibility in LC/MS analyses to support the comparison of assay vs control fractions for hit identification, using a rule-based comparator. XICs for m/z 745 (upper panel) from library A and m/z 1011 (lower panel) from library B. Table 2. Output of MS Processor for the Library Aa fractions

total

A

B

C

D

E

F

G

H

734 737 745 751 763 767 771 775 788 790 795 801 807 835 14

734 763 771 775 788 790 795 801 807

771 775 788 790 795 801 807

771 775 795 801

771 775 795

771 795

795

795

9

7

4

3

2

1

1

I

a Tabulated masses represent 1 u bins, the XICs of which contain at least one chromatographic peak representing a significant difference compared to the control. The strong ligand at m/z 795 appears in all columns as it elutes late from the GnT-V column.

for the observed hits under nonlinear conditions. All compounds in library A were present at e0.25 µM and each of the 67 nonhit compounds in the library can possess a Kd value no better than ∼36 µM. (Our minimal detectable breakthrough volume is 0.5 µL, based on the ability to detect a 10% difference in peak area between control and assay for a given compound. With a column capacity of 18 pmol, this implies a maximum detectable Kd of 36 µM, representing our hit level.) On the other end of the Kd range, the strongest observed ligand eluted in ∼35 µL. This single measurement cannot be converted into a Kd value, but with

Figure 6. Chemical structures of the four strongest ligands from library A.

reprocessing a dilution of the mixture via FAC-LC/MS, a simple linearization routine can be used to determine the strongest ligand’s Kd even under extreme nonlinear conditions (see Appendix for derivation). Following this approach, a Kd of 300 nM was measured for 1 (Figure 6), in reasonable agreement with the Kd determined via FAC/MS for the isolated compound (800 nM), considering the coarse sampling of the ligand breakthrough volume using the LC/MS method. The structures of all four significant ligands discovered in library A are presented in Figure 6. It is clear from these structures that an unsaturated, conjugated aglycon at the C4 position of mannose represents a relatively potent modification. One feature of the online FAC/MS experiments particularly useful in screening multiligand mixtures is the roll-up effect. Here, ligand displacement during a frontal experiment under nonlinear conditions leads to a transient overconcentration of a weaker ligand by a stronger ligand.19 This helps confirm that two ligands undergo specific interactions with the target protein. The resolution of our sampling strategy in FAC-LC/MS is sufficient to detect this event (Figure 7), which is represented by a chromatographic peak of higher amplitude in the assay than the corresponding blank run (fraction G, upper panel Figure 7). High-Throughput Screening of Enzyme Inhibitors from Library B. Drug discovery efforts have migrated toward the generation of “targeted chemical diversity”, where libraries based on structure-function data are assembled in an effort to reduce the size of compound libraries to be screened.20 This often involves 1000-10 000 compoundsslarge numbers, yet at least 1 order of magnitude lower in size than earlier practices. To test the ruggedness of FAC-LC/MS for this type of screening, we analyzed a targeted library of ∼1000 trisaccharide compounds, (19) Chan, N. W. C.; Lewis, D. F.; Hewko, S.; Hindsgaul, O.; Schriemer, D. C. Comb. Chem. High Throughput Screening 2002, 5, 395-406. (20) Langer, T.; Krovat, E. M. Curr. Opin. Drug Discovery Dev. 2003, 6, 370376.

Figure 7. XICs for m/z 771 eluted from GnT-V (dashed line) and control (solid line) columns in fractions G and H. The area of the chromatographic peak at 13 min is significantly higher in the assay run vs the control for fraction G but drops to control levels in subsequent fractions.

prepared as a single mixture (Figure 8) and doped with 15 compounds from library A (10 of which were determined to be hits through the screening of library A). All compounds in this library B were present at the same nominal concentration of 0.25 µM. A new GnT-V microcolumn was used in the FAC assay with an active capacity of 15.0 pmol. Five fractions were collected and processed as in the library A screen followed by application of the same CompareLCMS data analysis routine. The results are summarized in Table 3. Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

6131

Figure 8. Composition of library B, prepared as a single combinatorial mixture of nominally 1000 compounds.

Table 3. Summary of the Hits Found in a Screen of Library B rank order (1, weak; to 4, strong)

number of hits in library B

approximate Kd values (µM)a

1 2 3 4

42 6 3 1

g3 1.5-3 1.0-1.5 0.6

a The K ranges of the ligands are estimated in the table, based on d an assumption of infinite mixture dilution. Actual Kd values will be lower. See text for the Kd measurement of the single strongest ligand.

There were 52 hits discovered in library B, 42 of which eluted in the first fraction (fraction A). Masses corresponding to all ligands from the 15 doped compounds of library A were detected. Reanalysis of library B minus the 15 compounds confirmed the origin of these compounds as library A. Interestingly, 3 ligands 6132 Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

discovered in library B (with accurate masses of m/z 732.339, 786.405, and 805.382), arising from the “15”-compound supplement, are unintended synthetic products. All doped ligands eluted from the assay in the same order as in the screen of library A; the strongest ligand 1 observed in library A was also the strongest ligand in library B (molecular weight 794.384). Our data indicate that a large mixture of structurally similar compounds does not prohibit the detection and ranking of ligands, even though the column assay was fully saturated with ligand in the process (a competitive “indicator test”19 using OGMG confirmed the absence of residual binding capacity). The ligands did exhibit compressed breakthrough volumes in the library B screen compared to the library A screen, but this only effects detection of the weakest ligands. Strong ligand 1 eluted in 25 µL, as opposed to 35 µL in the screen of library A. Simple dilution and reprocessing would extend the resolution of hit ranking, as was the case in library B.

CONCLUSIONS Insertion of an LC purification step into the FAC/MS method for ligand discovery supports substantially improved screening rates and library coverage. Previous work has shown the online FAC/MS method to substantially underrepresent library composition leading to an unacceptably high false negative rate.19 Through fractionating the FAC effluent for LC/MS processing, we have removed the requirement of developing MS-compatible assay buffers and provided an extra dimension (LC retention time) for resolution of library compounds. This comes at the cost of a reduced resolution in hit determination, as the breakthrough volumes for each hit can only be estimated. The FAC effluent can be sampled more often to improve resolution; however, coarse fractionation will in most situations suffice for screening purposes. The insertion of an LC step provides an increase in sensitivity and the ability to detect lower concentrations of ligand; this is important when screening large mixtures as high total library concentrations can induce compound precipitation. Based on this work, screening rates exceeding 5000 compounds/day are easily achieved with a single LC/MS system. Hit identification was achieved through comparison with control data obtained through processing the libraries through a blank column. This serves to estimate nonspecific binding on a per-compound basis and helps to avoid misinterpreting a falsely high breakthrough volume as a strong hit. The application of CompareLCMS substantially reduced the burden of initial data analysis by presenting a reduced set of ion “bins” for visual assessment. Our doping experiment indicates the robustness of this approach for ligand detection. Overall the FAC-LC/MS method represents a useful approach to screening complex mixtures and should open up additional possibilities in combinatorial library preparation. Hit discovery is independent of compound concentration6 as long as a measurable signal is generated in the LC/MS analysis. This implies a wider range in synthetic yield can be tolerated; for example, split-and-pool syntheses can be evaluated by FAC-LC/MS to rapidly explore different areas of chemical space without optimization of reaction chemistry. Additionally, natural product extracts could be efficiently screened via this approach. ACKNOWLEDGMENT The authors thank Michael Pierce of the Complex Carbohydrate Research Centre, University of Georgia, for the generous

gift of GlcNAcT-V, and Lisa Prichard for assistance in the functional assays. Funding for this project was provided by the Alberta Ingenuity Centre for Carbohydrate Science (AICCS) and NSERC. APPENDIX The breakthrough volume for the strongest ligand in a multiligand FAC experiment may be represented by the following adaptation of the mass balance equation:5

Bt

(

Vn - V0 )

)

n - 1[A ] i 0

∑K

[An]0 + Kd,n 1 +

i)1

d,i

where Vn represents the breakthrough volume for the strongest ligand n, Bt the active capacity of the column, [Ai]0 the infusion concentration of ligand i, and Kd,i the dissociation constant for ligand i. Allowing for dilution alters this expression with a dilution factor, x:

Bt

(

Vn - V 0 )

x[An]0 + Kd,n 1 +

which can be linearized to

1 Vn - V0

)

Kd,n Bt

(

n - 1x[A

∑ i)1

[An]0 + Kd,n

+

Kd,i

)

n - 1[A ] i 0

∑K i)1

Bt

)

i]0

d,i

x

Thus, a plot of reciprocal breakthrough volume for the strongest ligand versus dilution factor allows determination of the ligand Kd from the y-intercept, without knowing the concentration of the ligand (with the assumption that Bt can be measured independently). Received for review June 24, 2005. Accepted July 25, 2005. AC051131R

Analytical Chemistry, Vol. 77, No. 19, October 1, 2005

6133