High-throughput identification of G protein-coupled receptor modulators through affinity mass spectrometry screening

High-throughput identification of GPCR modulators through affinity MS screening.


Introduction
The superfamily of G-protein-coupled receptors (GPCRs) is the largest class of cell surface receptors and they play a central role in a variety of pathophysiological conditions. 1 GPCRs are recognized as an important family of therapeutic targets upon which an estimated 30-40% of marketed drugs act. 2 While much effort has gone into identifying novel ligands that can modulate the activity of a GPCR target with high efficacy and selectivity, conventional techniques for GPCR drug discovery remain subject to several critical limitations. For example, receptor functional assays, which measure GPCR downstream signaling effectors, 3 are inadequate for identifying allosteric or biased signaling modulators and oen generate hits unsuitable for subsequent optimization. 4 Radioligand binding assays, which assess receptor-ligand interactions on cell surfaces, are increasingly restricted due to high production costs and hazards to human health. 3 Alternative receptor binding assays using uorescently labeled probes require careful compound design and optimization because of the impact of uorophore attachment on ligand affinity and efficacy. 5,6 Finally, while surface plasma resonance and NMR have recently been employed in the identication of GPCR ligands, 7-9 they require highly puried and stabilized receptors, which are not feasible for a number of targets and their current throughput is not amenable to large-scale compound library screening.
Affinity mass spectrometry (MS) has emerged as a powerful approach for analyzing protein-ligand interaction and it plays a vital role in early-phase drug discovery. [10][11][12] In a typical affinity MS-based workow, the ligand-bound protein complexes are rst separated from unbound compounds by ultraltration or size exclusion chromatography. Then the ligands dissociated from the complexes are identied by LC-MS/MS analysis. 10,11,13,14 Similar to other biophysical approaches, affinity MS has been widely applied to ligand identication for puried protein targets from compound libraries.
Affinity MS-based assays have been developed for screening chemical ligands towards different soluble protein targets, especially enzymes and kinases of therapeutic values. 10,12,13,[15][16][17] However, the application of affinity MS techniques to ligand discovery for membrane receptors is substantially hampered due to the difficulty of obtaining membrane proteins of sufficient purity, activity and stability. Whitehurst et al. rst showcased the adaptation of affinity MS to screening ligands towards the membrane receptor CXCR4 that belongs to the GPCR family. 18 To search for an optimal form of the receptor for screening purposes, the authors laboriously compared different epitope tags and detergents to nd the best conditions for expression and purication of the receptor. They argued that sufficient yield and purity of the receptor is essential for successful usage of this screening approach. 18 However, it is widely recognized that many transmembrane receptors are unstable when isolated away from the cell membrane. Thus, biophysical techniques that can only analyze puried proteins such as isothermal titration calorimetry (ITC), surface plasmon resonance (SPR) as well as affinity MS approach are not amenable to many receptors that are attractive drug targets.
Here we developed a novel affinity MS technique that enables ligand screening towards wild-type active receptors embedded in the cell membranes. Most signicantly, the challenging and laborious receptor purication step is eliminated in our work-ow. We implemented this new approach to achieve highthroughput, label-free and unbiased ligand screening towards two GPCR targets, which resulted in the discovery of unreported orthosteric ligands and allosteric modulators for specic GPCRs.

Results and discussion
We rst applied our methodology using the human 5-hydroxytryptamine 2C receptor (5-HT 2C R), an anti-psychotic drug target for treating depression, schizophrenia and other mental disorders. 19,20 The membrane fractions from insect cells expressing 5-HT 2C R were directly incubated with a cocktail of compounds while the protein concentration was kept in large excess over any compound (see ESI †). Cell membranes were separated from the compound solution by ltration. Compounds associated with the receptor-expressing membranes were released aer washing and subjected to liquid chromatography coupled to high-resolution mass spectrometry (LC-MS) analysis (Fig. 1a). Insect cell membranes expressing rhodopsin were prepared in the same manner and used as a negative control (Fig. S1 †). To assess the specic association of a given compound with the receptor, we calculated a binding index (BI) dened as the ratio of MS response of the compound detected in the target versus the control incubations (Fig. 1b). This BI parameter allowed us to compare the affinities of different compounds bound to the GPCR target and discern compounds non-specically interacting with the cell membranes.
To validate our method, we incubated the 5-HT 2C Rexpressing membranes with a mixture of 5 known ligands to the receptor and 45 unrelated compounds. Given all positive ligands with K i 0.3-29 nM showing BI > 2 ( Fig. S2 †) as well as our previous experience in selecting ligands to soluble protein targets with affinity mass spectrometry, [15][16][17] we dened a BI threshold of 2 for distinguishing putative ligands to the receptor from non-binders. Next, we employed the affinity MS assay to screen a collection of 4333 small molecules against 5-HT 2C R. This library was divided into 9 cocktails (each containing 480 or 493 compounds) and each was separately incubated with the target-expressing and control membranes in quadruplicate. Representative LC-MS chromatograms are shown in Fig. 1c. Multivariate analysis 17 of the LC-MS data from reference, target and control revealed that the composition of compounds associated with 5-HT 2C R was substantially deviated from that in the mix-1 reference and the control (Fig. 1d).
Using stringent data processing criteria for compound annotation and hit selection (see ESI †), we identied a total of 23 initial hits from screening the 9 cocktails (Fig. 2, full data set in Table S1 †). Twelve were well characterized antagonists for 5-HT 2C R or a closely-related subfamily member 5-HT 2A R with a mean BI of 2.16 to 9.71 (full data set in Table S2 †). Most of these known ligands showed <10 nM binding affinity to the receptor as determined by the radioligand competition assay. Results from a thermal shi assay using puried 5-HT 2C R indicated that 11 known ligands could signicantly enhance the receptor's thermostability thereby validating their direct interactions with the receptor. Notably, 98 compounds in our collection that were reported to be 5-HT 2C R ligands had BI < 2 from the primary screening. While their IC 50 or K i values from the ChEMBL database were all above 10 nM, 85% had IC 50 or K i > 100 nM towards 5-HT 2C R. Therefore, the stringent BI cut-off dened for hit selection ensured identication of highest affinity ligands with a minimal false-positive rate. To verify the 11 initial hits with no documented binding to the 5-HT 2 receptor family, a simple mixture was created by pooling and incubating them with the receptor-expressing membranes. Four hits were veried in this secondary affinity MS assay (Fig. 3a). Those that were invalidated could have resulted from compound misidentication or altered binding properties in the original complex mixtures. The unknown ligand 3943 had a thermostabilizing effect on the puried receptor (Table S3 †) and it further exhibited moderate antagonist activity in the calcium mobilization assay as it inhibited the 5-HT induced activation of 5-HT 2C R with IC 50 of 1.01 mM (Fig. 3b). Notably, this compound displayed even stronger antagonism against 5-HT 2A R (IC 50 ¼ 0.12 mM) and 5-HT 2B R (IC 50 ¼ 0.51 mM) (Fig. S3a †). The radioligand competition assay veried potent antagonist binding of 3943 to all three 5-HT 2 receptor subtypes in the nanomolar range, with stronger affinity to 5-HT 2A/2B R than 5-HT 2C R (Fig. 3c, S3b †). It was not surprising to identify a new antagonist against three 5-HT 2 receptor subfamily members using our approach because of the high sequence homology and very similar ligand binding pockets among them. A molecular docking study demonstrated key interactions between the cyproheptadine scaffold in 3943 and conserved residues in the transmembrane helices III, V, VI and VII of 5-HT 2A/2B/2C receptors ( Fig. 3d and e).
Next we applied our affinity MS assay to a more challenging target, the human glucagon-like peptide-1 receptor (GLP-1R). GLP-1R is a class B GPCR that mediates the action of peptide hormone GLP-1 and exerts important functions in glucose homeostasis. 21,22 Although small molecule modulators of GLP-1R are expected to serve as critical chemical tools for investigating ligand-directed biased signaling, very few non-peptidic GLP-1R agonists have been published. 21 Before embarking on a real library screening, we rst optimized our approach using two negative allosteric modulators (NAMs) of the receptor, PF06372222 and NNC0640. Both NAMs were readily identied from a 50-compound mixture with the mean BI of 16.66 for PF06372222 and 3.31 for NNC0640, indicating their signicant association with the receptor in the membrane fraction ( Fig. S4 †).
We then employed the affinity MS approach to screen the previous nine compound cocktails using insect cell membranes expressing the GLP-1R transmembrane domain (TMD). A total of 29 putative ligands were obtained from the primary screening ( Fig. 4, full date set in Table S4 †), and none of them have been previously linked with GLP-1R.
In the secondary affinity MS screening assay, 18 were conrmed to bind the receptor (Fig. 5a). Importantly, in this step we also prepared cell membranes expressing another GLP-1R TMD construct with stabilizing mutations commonly exploited in the GPCR crystallography. 23,24 Interestingly, all ligands identied in our primary screening abrogated their binding to the thermostabilized receptor (Fig. 5a). This result highlighted the unique advantage of using cell membranes expressing wild-type receptors in ligand screening given that puried receptors with mutations could be locked in an inactive  conformation and fail to engage ligands that only interact with the active receptor. Subsequent radioligand binding assay revealed that four ligands (901, 3286, 4170, and 4279) augmented peptide binding to the receptor with EC 50 in the low micromolar range (Fig. 5b, Table S5 †). When conducting another binding assay using radiolabeled exendin-4  , which is a GLP-1R antagonist targeting the extracellular domain (ECD), we observed no obvious alteration of the binding potency between exendin-4  and GLP-1R by any of the four ligands ( Fig. S5a †). Moreover, increasing the amount of NAM PF06372222 in the presence of the four ligands did not compete off any ligand in the affinity MS assay (Fig. S5b †).
These results collectively imply that the four novel ligands allosterically modulate the binding of GLP-1 to the orthosteric pocket of GLP-1R and that their targeting sites are likely to be different from the NAM binding pocket. 25 Concordantly, all four ligands substantially promoted intracellular cAMP accumulation in the presence of GLP-1 with EC 50 in the low micromolar range (Fig. 5c, Table S5 †). By contrast, none of them stimulated cAMP production in cells expressing glucagon receptor (GCGR), a close homolog of GLP-1R, demonstrating the high selectivity of the four ligands for GLP-1R (Fig. S5c †). Therefore, we conclude that the four ligands identied in this study are all positive allosteric modulators (PAMs) of GLP-1R. For compound 3286, we performed molecular docking based on a previous model of the GLP-1/GLP-1R complex 26 to reveal a possible binding mode of this PAM (Fig. 5d and e). While the precise targeting sites and activation mechanism of the four PAMs warrant further investigation, we are excited by the discovery of a new class of ligands for this important therapeutic target and the efficiency of our approach in identication of novel GPCR ligands.
A notable limitation of our approach lies in the detectability issue for certain compounds by our LC-MS analytical platform. In our study, the percentage of detectable compounds in each library cocktail by electrospray MS in the positive mode varied between 66-86% (Fig. S6 †). Thus, the receptor binding capability of the undetectable compounds remains unclear. This could be partially addressed using MS instruments of increased sensitivity and operating in both positive and negative modes, or by optimizing the LC system for effective separation of abnormal compounds. Nevertheless, our current LC-MS method is generally applicable to the analysis of most druglike small molecules.

Conclusions
In summary, we developed an efficient and convenient affinity MS approach for high-throughput identication of GPCR ligands. Our new approach demonstrates ve major advantages over the conventional receptor functional assays or ligand binding assays: (1) both the receptor and compounds are labelfree; (2) the receptor target, with minimal sequence modication, is embedded in the cell membranes to retain its native  conformation during ligand interaction; (3) measurement of direct receptor-ligand binding in an unbiased manner facilitates identication of allosteric modulators targeting uncharacterized binding sites; (4) quantitative comparison of ligand response with a GPCR control enables identication of specic ligands with medium affinity while maintaining a low falsepositive rate; and (5) protein purication commonly required in affinity-based ligand screens is unnecessary, thus reducing experimental cost and eliminating purication-inherent drawbacks. Our screening technology, benchmarked on two GPCR targets, can be readily adapted to other membrane receptors. Its integration with complementary biophysical and functional assays could expedite the discovery of new GPCR modulators with therapeutic potential. Apart from direct application to drug discovery towards membrane receptors, our approach is expected to aid in delineating the membrane protein-small molecule interaction network within the cell.

Preparation of cell membranes over-expressing 5HT 2C -R or GLP1-R TMD
The 5-HT 2C R construct was produced comprising residues 40-393 with residues from intracellular loop 3 (245-301) replaced by thermo-stabilized E. coli apocytochrome b 562 RIL (BRIL). It was expressed with an N-terminal FLAG tag and a C-terminal 10Â His tag. The GLP-1R TMD construct comprised residues 128-431 with residues from intracellular loop 2 (258-263) replaced by rubredoxin and it was expressed with an N-terminal FLAG tag followed by BRIL and a C-terminal 10Â His tag. The rhodopsin construct comprised residues 2-331 with residues from intracellular loop 3 (235-241) replaced by BRIL and it was expressed with an N-terminal FLAG tag and a C-terminal 10Â His tag. All three proteins were expressed using the Bac-to-Bac Baculovirus Expression System (Invitrogen) in Spodoptera frugiperda (Sf9) cells. Cells were infected at a density of 2-3 Â 10 6 cells per mL with baculovirus at a multiplicity of infection (MOI) of 5. Cultures were grown at 27 C and harvested 48 h aer infection. Cell pellets were lysed by repeated washing and ultracentrifugation in the hypotonic buffer of 10 mM HEPES (pH 7.5), 10 mM MgCl 2 , 20 mM KCl, and the high osmotic buffer of 10 mM HEPES (pH 7.5), 1.0 M NaCl, 10 mM MgCl 2 , 20 mM KCl, both with EDTA-free protease inhibitor cocktail tablets (Roche). The washed membranes were re-suspended in the hypotonic buffer with 30% glycerol and stored at À80 C for further usage. Cell membrane proteins were extracted using 1% SDS in 0.1% NaOH and total protein concentration was measured with the BCA quantication kit (TIANGEN, China). The amount of 5-HT 2C R and GLP-1R present in the membrane extract was determined using ELISA with anti-His antibody (Genscript, China).

Compound library preparation
The library of 4333 small molecule members was purchased from Topscience (Shanghai, China) in DMSO stock. It was divided into nine cocktails (mix-1 to mix-9) by pooling different sub-fractions of the library. Eight cocktails (mix-1 to mix-8) comprised 480 compounds and the last one (mix-9) comprised 493 compounds. No compounds were overlapped between different cocktails. All cocktails were stored at À20 C.

GPCR ligand identication with affinity mass spectrometry
Sample preparation for screening and hit validation. In the primary screening, the membrane fraction expressing the receptor (5HT 2C R or GLP-1R TMD) was incubated with a compound cocktail in a buffer of 10 mM HEPES (pH 7.5), 10 mM MgCl 2 and 20 mM KCl in a total volume of 200 mL for 1 h at 25 C. During incubation, each compound concentration was at 50 nM and the receptor concentration was estimated to be 200-250 nM. The membrane fraction was separated from the incubation solution by rapid vacuum ltration through Multi-ScreenHTS FB Filter Plate (Millipore) using a MultiScreenHTS vacuum Manifold (Millipore). Aer washing the membrane fraction six times with ice-cold 150 mM ammonium acetate (pH 7.5), methanol was added to the membranes (100 mL, 4 times) and the ltrate, containing compounds initially associated with the membranes, was collected. We then used the Ostro™ 96well plate (Waters) to deplete phospholipids co-eluted with the compounds of interest. The eluted samples were evaporated by a speed vacuum and reconstituted in 50% methanol before LC-MS analysis. The control sample was prepared by using rhodopsin-expressing membranes in incubation with the same amount of total membrane proteins as the target-expressing membranes. Each pair of target and control samples was prepared in quadruplicate.
In secondary screening, initial hits were pooled to make a simple mixture, which was incubated with target and control membranes separately under the same conditions. In the 5HT 2C R experiment, 11 unknown ligands were pooled whereas in the GLP-1R experiment, all 29 initial hits were pooled. Compound concentration in incubation was still 50 nM and the receptor concentration was increased to 300 nM. In the method validation study, both target-expressing and control membranes were incubated with a 50-compound mixture containing specic known ligands and unrelated compounds with the same receptor and compound concentrations as mentioned above. Each pair of target and control samples was prepared in quadruplicate.
LC-MS data processing and hit selection. First, compounds in the reference were identied by extracting selected ion chromatograms (EICs) using Peakview 2.2 (AB SCIEX) based on accurate mass measurement (<10 ppm deviation) and isotope envelop matching (<20% difference from the theoretical envelop). H + and Na + adducts were considered for compound detection. Then compounds in target and control samples were identied by meeting the above criteria plus retention time (RT) matching with corresponding peaks in the reference (<0.2 min shi). Ambiguous peaks of isomeric compounds in the cocktail were distinguished by acquiring MS/MS spectra or injecting individual standards for RT differentiation. For each compound condently identied in target and control samples, its BI was calculated by dividing the EIC intensity of the compound detected in the target sample by that in the control. Given that target and negative control proteins were expressed at similar levels and their concentrations during ligand incubation were close to each other (<20% difference), we did not modify the BI ratios with the protein concentration ratios. Initial hits were selected based on a mean BI > 2 and p < 0.05 from four replicates. Hits were validated in the second-round MS affinity assay based on a mean BI > 2 and p < 0.01 from four replicates. Pvalues were determined by a two-tailed t-test of BI values against unity. Putative ligands were searched in ChEMBL, DrugBank, Binding DB and SciFinder databases to nd out whether they are known ligands to the receptor target.
Purication of 5-HT 2C R protein for the thermal shi assay The receptor protein was extracted from the previously puried membranes by adding n-dodecyl-D-maltopyranoside (DDM, Affymetrix) and cholesteryl hemisuccinate (CHS, Sigma) to the membrane suspension to a nal concentration of 1.0% (w/v) and 0.2% (w/v), respectively, in buffer of 50 mM HEPES (pH 7.5) and 150 mM NaCl, while stirring continuously at 4 C for 2 h. The supernatant was isolated by centrifugation at 160 000g for 30 min followed by overnight incubation in TALON IMAC resin (Clontech) at 4 C. The resin was washed with ten column volumes of wash buffer 1 (50 mM HEPES, pH 7.5, 800 mM NaCl, 0.1% (w/v) DDM, 0.02% (w/v) CHS, 20 mM imidazole, 10% (v/v) glycerol) and followed by ve column volumes of wash buffer 2 (50 mM HEPES, pH 7.5, 150 mM NaCl, 0.05% (w/v) DDM, 0.01% (w/v) CHS, 10% (v/v) glycerol). Proteins were eluted in 5 column volumes of wash buffer 2 with 250 mM imidazole. Protein purity and monodispersity were assessed by SDS-PAGE and analytical size-exclusion chromatography.
The thermal shi assay using a CPM uorescent dye (Sigma) was performed as described in the literature. 27 Briey, puried receptor protein pre-mixed with the CPM dye was incubated with a given compound at 200 mM at 4 C for 1 h. The protein sample was heated step-wise on a Rotor-Gene Thermo-optical Analyzer (QIAGEN Gmbh) from 25 C to 95 C. Upon temperature rise, the proteins unfolded and cysteine residues were exposed to form adducts with CPM, which were detected by uorescence at 387/463 nm using an EnVision multilabel plate reader (PerkinElmer). From the normalized thermal stability curve, melting temperatures (T m ) were obtained by tting the curve with a Boltzmann sigmoidal function using GraphPad Prism. Comparison of T m for the apo receptor and receptor incubated with a compound gave rise to T m shi (DT m ) for the compound.

Calcium mobilization assay
HEK293T cells stably transfected with 5-HT 2A/2B/2C receptor were seeded in 384-well plates at a density of 15 000 cells per well in DMEM containing 1% dialyzed FBS 8 h before assaying. Aer removing the medium, cells were incubated (20 mL per well) for 1 h at 37 C with Fluo-4 Direct dye (Invitrogen) and reconstituted in a FLIPR buffer (1Â HBSS, 2.5 mM probenecid, and 20 mM HEPES, pH 7.4). Aer the dye loaded, cells were placed in a FLIPR TETRA uorescence imaging plate reader (Molecular Devices). Drug dilutions, prepared at 3Â nal concentration in FLIPR buffer and aliquoted into 384-well plates, were also added to the FLIPR TETRA. The uidics module and plate reader of the FLIPR TETRA were programmed to read baseline uorescence for 10 s (1 read per s), then to add 10 mL of drug per well and to read for 6 min (1 read per s). Fluorescence in each well was normalized to the average of the rst 10 reads (i.e., baseline uorescence). Then the maximumfold increase, which occurred within 60 s aer drug addition over baseline uorescence elicited by vehicle or a test compound, was determined. In the test of potential positive allosteric modulators and antagonists, 5-HT at EC 20 (0.1 nM) and at EC 80 (3 nM) were added to the medium, respectively, for receptor activation.

cAMP accumulation assay
The human GLP-1R and glucagon receptor (GCGR) cDNAs were cloned into pcDNA3.1/V5-His-TOPO (Invitrogen). Aer 24 h transfection, stably expressing CHO-K1 cells were selected on 750 mg mL À1 G418 (Roche) for 2 weeks to obtain the clone with the highest expression and potent cell activity. Stable cells were seeded in 6-well plates for overnight culture before inoculation into 384-well plates (8000 cells per well) for the assay. Accumulation of cAMP was measured using the LANCE cAMP kit (PerkinElmer) according to the manufacturer's instructions. Briey, cells were incubated for 30 min in assay buffer (DMEM, 1 mM 3-isobutyl-1-methylxanthine) with varying concentrations of each compound (2.3 nM to 50 mM) in the presence of GLP-1 or glucagon (0.02 nM) at 37 C. Compound treatment was quenched by adding lysis buffer containing LANCE reagents. Plates were then incubated for 60 min at room temperature and time-resolved FRET signals were measured at 620 nm and 650 nm by an EnVision multilabel plate reader (PerkinElmer).
In the GLP-1R and GCGR binding assays, stable cell lines were seeded onto 96-well poly-D-lysine treated cell culture plates (PerkinElmer) at a density of 3 Â 10 4 cells per well. Cells were harvested aer 24 h post seeding, washed twice and incubated with blocking buffer (F12 supplemented with 33 mM HEPES, pH 7.4, and 0.1% bovine serum albumin (BSA)) for 2 h at 37 C. They were then washed twice with PBS and incubated in binding buffer (PBS supplemented with 10% BSA, pH 7.4) with a test ligand at room temperature for 3 h. Cells were treated with each ligand at varying concentrations (7.6 nM to 16.7 mM) in the presence of 125 I-GLP-1 (40 pM) or 125 I-exendin-4 (9-39) (40 pM). Cells were then washed three times with ice-cold PBS and lysed by 50 mL lysis buffer (PBS supplemented with 20 mM Tris-HCl, 1% Triton X-100, pH 7.4). The plates were subsequently counted for radioactivity (counts per minute, CPM) in a scintillation counter (MicroBeta2TM Plate Counter, PerkinElmer) using a scintillation cocktail (OptiPhase SuperMix, PerkinElmer).

Molecular modeling and docking
Binding mode prediction was performed using graphical user interface Maestro 10.4 in Schrödinger Suite 2015-4. For 5-HT 2C R, a homology model of the receptor was built based on the crystal structure of 5-HT 2B R 28 (PDB: 4IB4) using the Advanced Homology Modeling tool. For GLP-1R, a previously published model 26 was used and the allosteric binding site was identied with the SiteMap tool. Ligand 3D structures were generated using LigPrep 3.6. Molecular docking was carried out using Induced Fit Docking with the extra precision docking score and allowing optimization of residues within 5.0Å.

Conflicts of interest
There are no conicts to declare.