Native State Mass Spectrometry, Surface Plasmon Resonance, and X

Feb 16, 2016 - We embarked on a trimodal fragment screening campaign against CA II ..... Nanomate (Advion Biosciences) automated nanoESI interface...
0 downloads 0 Views 4MB Size
Article pubs.acs.org/jmc

Native State Mass Spectrometry, Surface Plasmon Resonance, and X‑ray Crystallography Correlate Strongly as a Fragment Screening Combination Lucy A. Woods,† Olan Dolezal,‡ Bin Ren,‡ John H. Ryan,‡ Thomas S. Peat,*,‡ and Sally-Ann Poulsen*,† †

Griffith University, Eskitis Institute for Drug Discovery, Brisbane, Queensland Australia CSIRO Biomedical Manufacturing Program, Melbourne, Victoria Australia



S Supporting Information *

ABSTRACT: Fragment-based drug discovery (FBDD) is contingent on the development of analytical methods to identify weak protein−fragment noncovalent interactions. Herein we have combined an underutilized fragment screening method, native state mass spectrometry, together with two proven and popular fragment screening methods, surface plasmon resonance and X-ray crystallography, in a fragment screening campaign against human carbonic anhydrase II (CA II). In an initial fragment screen against a 720-member fragment library (the “CSIRO Fragment Library”) seven CA II binding fragments, including a selection of nonclassical CA II binding chemotypes, were identified. A further 70 compounds that comprised the initial hit chemotypes were subsequently sourced from the full CSIRO compound collection and screened. The fragment results were extremely well correlated across the three methods. Our findings demonstrate that there is a tremendous opportunity to apply native state mass spectrometry as a complementary fragment screening method to accelerate drug discovery.



nonclassical natural product CA inhibitors.8,9 CA inhibitors that represent new chemical entities are needed for the drug discovery pipeline to deliver new and improved CA-based therapeutics of commercial value.10 Fragment-based drug discovery (FBDD) is a recent innovation to address the challenges associated with identifying novel chemical starting points with better developmental prospects for optimization in drug discovery research.11 “Better developmental prospects” equates to a compound that is more likely to be optimized into a drug that targets the “sweet spot” of potency, safety, stability, specificity, solubility, absorption, and other relevant pharmacokinetic parameters.12 Fragment screening permits an unbiased identification of new structural motifs that bind to a protein family. However, unlike high throughput screening (HTS) where hits are relatively potent (KD of nM to μM), fragments bind to their target with considerably weaker affinities (typical KDs of μM to mM); hence, the identification of fragment hits is less amenable to

INTRODUCTION Carbonic anhydrases (CA, EC 4.2.1.1) are zinc metalloenzymes that catalyze the reversible hydration of carbon dioxide to bicarbonate and a proton: CO2 + H2O ⇆ HCO3− + H+.1 Most reported small molecule CA inhibitors incorporate a primary sulfonamide functional group, with the sulfonamide anion, RSO2NH−, coordinating to the active site zinc and forming hydrogen bond interactions with active site residues in the immediate vicinity of the zinc. The HCO3−/CO2 equilibrium is critical for human health, and blocking the endogenous chemistry catalyzed by CA II has been the target of therapeutic intervention for several decades.2 The primary sulfamate (OSO2NH2), primary sulfamide (NHSO2NH2), and cyclic secondary sulfonamide (e.g., saccharin) chemotypes, with a structural resemblance to primary sulfonamides, also inhibit CA enzymes via coordination of their anionic form to the CA II active site zinc.3−7 Despite the importance and longevity of CA as a drug target, alternative classes of inhibitors to the classical primary sulfonamide class have remained scarce. In recent times natural products have been explored as a source of new CA inhibitors. Coumarins are the most notable example of © 2016 American Chemical Society

Received: December 15, 2015 Published: February 16, 2016 2192

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

Figure 1. Stage 1 screening of the 720-member CSIRO Fragment Library against human carbonic anhydrase II (CA II) using SPR. (A) Binding sensorgrams for fragments interacting with immobilized CA II. Fragments were injected sequentially at 100 μM over the CA II surface for 30 s and then allowed to dissociate in running buffer for a further 30 s. (B) Plots of fragment binding responses taken at t = 25−27 s after injection. 4Carboxybenzenesulfonamide (4-CBS), a positive control, was injected at a saturating concentration (20 μM) four times during each screen (blue squares). Sensorgrams are scaled based on the molecular weight ratio of the test compound to the compound with a molecular mass of 200 Da. The dashed line represents the threshold level used for initial selection of potential binders. Compounds showing undesirable SPR binding characteristics as previously described by Giannetti et al.33 are labeled with an asterisk (*); these compounds were not considered fragment hits.

classical biochemical screens.13 To detect weak binding, sensitive biophysical screening methods are used instead, together with high fragment concentrations. Fragment screening campaigns often involve a cascade of biophysical screens,11 and NMR, surface plasmon resonance (SPR), X-ray crystallography, and isothermal titration calorimetry (ITC) are the more commonly used fragment screening methods.14 Each of these popular screening methods has distinct advantages and limitations associated with sensitivity, throughput, fragment concentration, and quantity of protein and fragment consumed. Native state electrospray ionization mass spectrometry (ESIMS) has been used extensively to directly observe native state proteins and protein complexes. The method has a wide dynamic range and allows the direct detection of protein− ligand noncovalent complexes with KDs as weak as 1 mM.15 The method is rapid, consumes minimal protein and ligand, is label-free, and does not require sample immobilization, chromatography, or sample manipulation to separate ligands from protein. ESI-MS analysis necessitates the transfer of solution sample components to the gas phase as fully desolvated molecular ions. Once in the gas phase, a measurement of the mass-to-charge (m/z) ratio of the ions is determined. If a small molecule (such as a fragment) is added to a sample of the target protein and binds to the protein, then both the [unbound protein] m/z and the noncovalent complex of [protein + small molecule] m/z are observed as peaks in the ESI mass spectrum. The m/z difference (Δm/z) between the [unbound protein] peaks and [protein + small molecule] peaks when multiplied by the charge state z gives the molecular weight (MW) of the small molecule, i.e., MW = Δm/z × z. Hence, the observation of [protein + fragment] m/z and determination of Δm/z can be highly informative for fragment-

based screening applications. This analysis enables the classification of a fragment as a “binder” or “nonbinder”, and if a “binder”, it provides the molecular weight as an inbuilt identifier associated with fragment identity. Enthalpic interactions, dependent on electrostatic and polar interplay, survive transfer from solution to the gas phase and are even strengthened in the gas phase.16,17 In contrast, hydrophobic (entropy driven) interactions are weakened in the gas phase and may not survive transfer from solution to the gas phase.16 The screening of fragments using ESI-MS may thus favor the identification of fragment hits where enthalpy dominates protein binding. These hits may be preferred starting points for the subsequent medicinal chemistry optimization campaign, with entropy-governed optimization of protein−ligand interactions occurring downstream.18 The association between fragment thermodynamic profile, fragment selection, and fragment development prospects is however not straightforward.19 ESI-MS has been employed successfully for screening with related in situ approaches, including dynamic combinatorial chemistry and tethering.20−23 With these approaches, fragments are covalently linked in the presence of the protein target, and the resulting linked fragment combinations are identified owing to their unique masses. Nanospray ESI mass spectrometry (nanoESI-MS) operates at lower sample flow rates and smaller sample droplets than standard ESI-MS sources; this contributes increased signal sensitivity and enhanced retention of protein−small molecule noncovalent complexes during desolvation.24 A substantial reduction in protein consumption is an additional advantage of nanoESI-MS for fragment screening campaigns, as protein availability is often a limiting factor. In principle, this merger of the combined attributes offered by ESI-MS and nanoESI-MS 2193

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

Table 1. Correlation of Screening Results for Stage 1 Fragment Hit Chemotypes with SPR, ESI-MS, NanoESI-MS, and X-ray Crystallography (green tick  hit, red cross  not a hit, n/a fragment not tested)

Dose−response experiment performed at 25 °C with a five-point fragment concentration series range. bNano MS hit with ratio of unbound CA II:fragment bound CA II peak intensities in brackets.

a

are well suited to address the detection needs of fragment screening; however, in practice, there are just a few examples where mass spectrometry has been employed for screening of fragments.25−30 Mass spectrometry generally receives only a cursory mention (if any mention) in reviews that describe FBDD or fragment screening. We embarked on a trimodal fragment screening campaign against CA II using native state mass spectrometry together with two established fragment screening methods, SPR and Xray crystallography, to discover novel CA binding chemotypes with prospects for development as CA inhibitors. Our overarching intention was however to purposefully evaluate

native state mass spectrometry as a fragment screening method. Both a standard ESI-MS and a nanoESI-MS source with direct infusion of the test solutions were employed in this study. Sources were interfaced with a Bruker solariX XR 12.0 T Fourier transform ion cyclotron resonance mass spectrometer (FTICR MS) fitted with a ParaCell. This instrument configuration provides extremely high-resolution broadband acquisition, and isotopic resolution of CA II protein is routine. The screening attributes of the native state mass spectrometry method, including very good correlation of the screening hits with those hits identified using the proven fragment screening methods of SPR and X-ray crystallography, will be highlighted. 2194

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

Figure 2. NanoESI mass spectra of stage 1 hit fragments with CA II. The 9+ charge state is shown, with [protein + fragment]9+ peaks in red. (A) hCA II protein only. (B) 1, primary sulfonamide chemotype. (C) 2, cinnamic acid chemotype. (D) 3, benzoic acid chemotype. (E) 4, phenylacetic acid chemotype. (F) 5, tetrazole chemotype. (G) 6, tetrazole chemotype. (H) 7, 1,2,4-triazole chemotype.



RESULTS AND DISCUSSION The screening campaign targeting CA II for this study was conducted in two stages, with stage 1 screening against the CSIRO Fragment Library. This recently described 720-member fragment library has been shown to be useful across multiple target classes and, importantly, has been shown to be a suitable library for fragment screening by SPR.31,32 All fragments are within the molecular mass range of 100−350 Da. Initially, the screening of fragments against minimally biotinylated CA II was performed by SPR. CA II was immobilized on the sensor chip surface via biotin−streptavidin interactions and compounds screened at a single concentration of 100 μM at 15 °C (Figure 1). 4-Carboxybenzenesulfonamide (4-CBS), a control compound with CA II binding specificity, was used to monitor the protein stability on the chip surface during the screen. Seven fragment hits were identified in this initial SPR screen and rescreened against CA II in follow-up SPR dose−response experiments, performed at 25 °C with a five-point fragment concentration series range, to determine the binding affinity of hit fragments (Supporting Information Figure 1). The structures and estimated affinity values (KD) are shown (Table 1, columns 1 and 2). Fragment hit 1 is a primary sulfonamide.34 This chemotype is the classical CA inhibitor chemotype, and as expected, strong binding was observed with a KD = 1.35 μM. Fragment 1 itself is not a reported CA inhibitor; however, the simpler fragment pyrimidine-2-sulfonamide was reported as a CA inhibitor in 1950.35 The other six stage 1 fragment hits had weaker binding

than the primary sulfonamide fragment, with KD estimates in the range 194−1280 μM. The hits include a selection of carboxylic acid chemotypes, specifically a cinnamic acid chemotype (2),36,37 a benzoic acid chemotype (3),38 and a phenylacetic acid chemotype (4), and two substituted tetrazoles (structural isomers 6 and 5)39−41 and the 1,2,4-triazole (7).42 The carboxylate group is the most common zinc binding functional group, followed by the primary sulfonamide group, in ligands appearing in protein−ligand structures deposited in the Protein Data Bank.43 Interestingly, the benzoic acid fragment 3 is embedded within a tricyclic natural product core that we reported when first investigating natural products as a source of novel CA inhibitors.44,45 Similarly the phenylacetic acid chemotype of fragment 4 and the cinnamic acid chemotype of fragment 2 are embedded within natural products reported with CA inhibition,8,44,45 and their appearance as hits in this fragment screening campaign thus has strong grounding. The inhibition of CA II by the unsubstituted acidic heterocycles, tetrazole and 1,2,4-triazole, was reported by Lindskog and colleagues in 198546 when assessing a panel of simple organic compounds for CA II inhibition. 5-Substituted-1H-tetrazoles are a common bioisosteric replacement for a carboxylic acid moiety, and this motif appears within the structures of several US FDA approved drugs.47,48 The 5-substituted tetrazole fragment hits, 5 and 6, are, however, not known previously to bind to CA II. In summary, stage 1 hits identified by SPR include chemotypes that comprise the well-known zinc binding groups of primary sulfonamides and carboxylates, for which there is 2195

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

Figure 3. Difference density maps of stage 1 fragments in the CA II binding site. (A) 1, primary sulfonamide chemotype. (B) 2, cinnamic acid chemotype. (C) 3, benzoic acid chemotype. (D) 4, phenylacetic acid chemotype. (E) 5, tetrazole chemotype. (F) 6, tetrazole chemotype. Difference density (mFo-DFc) for all maps is shown at a 3σ contour level where the fragments were omitted from the model.

acid containing fragment, 4, is not as clean, and this fragment can be potentially modeled with or without an intervening water molecule; we have elected to present the model with a direct interaction to the zinc, Figure 3, entry D. X-ray crystallography confirmed the interaction of the N-1 nitrogen of both the tetrazole hits (5 and 6) with the CA II active site zinc, Figure 3, entries E and F. The 3-alkylthio-1,2,4-triazole fragment (7) was not observed by either soaking or cocrystallization methods. This is the only stage 1 fragment hit that did not provide a structure, and interestingly this fragment also was the weakest observed using nanoESI-MS, with the lowest intensity [protein + fragment] peak of the seven fragments. The screening results obtained across the three screening methods are in excellent agreement, Table 1. Notably, all seven hits detected by SPR were observable by nanoESI-MS (Table 1, Figure 2), while six of the seven soaked fragments led to CA II:fragment structures observed by X-ray crystallography (Table 1, Figure 3; see Supporting Information). The correlation of standard ESI-MS with SPR, nanoESI-MS, and X-ray crystallography was also very good (Table 1, column 3); however, the fragment set was incomplete at time of screening by ESI-MS (two of the seven fragments were not available for screening). Stage 2 screening against CA II was carried out with 70 compounds sourced from the CSIRO proprietary compound collection, comprising of over 50 000 unique, mainly lowmolecular-weight heterocyclic molecules. For stage 2, the screening was conducted independently and in parallel using ESI-MS, nanoESI-MS, and SPR. X-ray crystallography was then performed on selected stage 2 hit fragments where SAR was of interest. As was found for stage 1 screening, the overlap between SPR and MS hits was in very good agreement, Table 2.

precedent for CA inhibition with both synthetic compounds and natural products. The lesser known zinc binding groups of tetrazole and 1,2,4-triazole were also hits, and as simple unsubstituted heterocycles they have previously been reported in association with CA inhibition. To the best of our knowledge, however, substitution of these heterocycles has not been investigated further in the context of structure− activity relationships or inhibitor development targeting CA enzymes. Next, the stage 1 fragment hits were assessed by native state mass spectrometry (standard flow ESI-MS and nanoESI-MS) (Figure 2) and protein X-ray crystallography (Figure 3). Native state mass spectrometry spectra were acquired with test samples comprising CA II with 1 equiv of added fragment. Figure 2, entry A, is the nanoESI mass spectra for CA II, while Figure 2, entries B−H, are the nanoESI mass spectra for the seven fragment hits. The ratio of the intensity of the [protein + fragment] peaks (in red) relative to [unbound protein] peaks (in black) provides a qualitative indication of the fragment binding affinity, Table 1, Figure 2. The primary sulfonamide fragment (1) is almost completely bound to CA II, while the mass spectra of the other six fragments have both unbound protein and fragment-bound protein present, with relative peak intensities that are indicative of weaker binding for these chemotypes compared to the primary sulfonamide chemotype. X-ray crystallography shows that the sulfonamide moiety of 1 exhibits the canonical interactions with the zinc and active site residues (Figure 3, entry A). For carboxylic acid fragments 2 and 3, there is clear and clean electron density that shows that these acids bind to the zinc through a bridging water molecule that is often found bound to the zinc in CA structures, Figure 3, entries B and C. The electron density for the third carboxylic 2196

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

Table 2. Correlation of Screening Results for Stage 2 Fragment Hits with SPR, NanoESI-MS, and X-ray Crystallography (green tick  hit, red cross  not a hit, n.m. not measured)

2197

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

Table 2. continued

Dose−response experiments performed at 25 °C with a five-point fragment concentration series range. SPR hit threshold is KD ≤ 3000 μM. bMS hit threshold is 1:0.02 ratio of unbound CA II:fragment bound CA II peak intensities. cCompound showed undesirable SPR binding characteristics.33 a

The 70 stage 2 compounds were selected from the CSIRO compound collection (which contains some commercially available compounds), using substructure and ad hoc similarity searches based on the stage 1 fragment hit chemotypes. The compound collection database ActivityBase (8.0) (IDBS, Surrey, UK; http://www.idbs.com/) was interrogated to generate lists of structural analogues that had dry stock available for testing. A nonhit was classified as a fragment that was not detected in the fragment screen by MS or SPR. Overall 37 stage 2 compounds were classified as hits, and of these 24 were hits by both MS and SPR, Table 2, while 33 stage 2 compounds were nonhits (not detected by either MS or SPR). The correlation of screening results and SAR for hits and nonhits for the individual chemotypes is described next. Stage 2: Primary Sulfonamide Chemotype (stage 1 hit 1). Seven analogues of 1 were selected, and of these three were primary sulfonamides and four were secondary or tertiary sulfonamides. Secondary or tertiary sulfonamides, with few exceptions, are expected to be poor CA inhibitors,3,53 and consistent with this, the four secondary or tertiary sulfonamides were not detected in the fragment screen by MS or SPR. Two of the three primary sulfonamide compounds (8 and 9) were hits, and this was expected. The third primary sulfonamide was not a hit, this fragment is however sterically hindered around the sulfonamide functional group, and this may contribute to

the lack of binding to CA II owing to a disruption of the sulfonamide anion−zinc cation interaction.35 The primary sulfonamide hit fragments were not followed up with X-ray crystallography, as the interactions of the primary sulfonamide chemotype with CA II are well-known and are typically invariant.49 Stage 2: Cinnamic Acid Chemotype (stage 1 hit 2). A total of 36 analogues of 2 were available, and this was the largest set of analogues screened in stage 2. Of these, 29 compounds were in full agreement with SPR and MS methods (12 hits/17 nonhits), providing broad structure−activity relationships. Compounds 10 and 19 were also detected by X-ray crystallography while an X-ray crystal structure for 22 was attempted; however, no density was seen in the active site for this compound. For the remaining seven cinnamic acid analogues that were identified as a MS hit, a KD value for each was measured using SPR and was outside of the cutoff value to be classified as a reliable SPR hit. Of these, compound 26 was assessed by X-ray crystallography and no electron density for the fragment was observed. Stage 2: Benzoic Acid Chemotype (stage 1 hit 3). Eleven analogues of 3 were screened in the stage 2 campaign. Eight of these analogues have their carboxylate functionality either capped as an acyl ester (not able to form a free carboxylate) or are sterically hindered around the free 2198

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

conformer in the cocrystallized structure (Figure 4). This is due to the lack of strong electron density for the region of the molecule distal from the zinc, with the second carboxylic acid moiety (“cinnamic acid chemotype 2-like”) free to move, as it does not make any strong interactions with the CA II active site residues. Compound 33 was a hit by MS, while for SPR the binding was outside the hit threshold. This compound was observed by X-ray crystallography directly bound to the CA II active site zinc atom with 1.9 and 2.6 Å distances to the carboxylate oxygen atoms. There are two water molecules close by, at 2.6 and 2.8 Å distance to the carboxylate moiety; however, the compound does not make any other hydrogen bond or ionic interactions with the protein. Interestingly, the carboxylic acid moiety is almost 90° turned compared to the same moiety of stage 1 hit compound 4. We believe this is almost entirely due to the structural differences between the compounds that are distal to the carboxylic acid and zinc interaction. A second molecule of 33 has been modeled in at a crystal interface where the carboxylate makes two potential interactions, one with the Phe131 backbone nitrogen (2.8 Å) and one with the Asp72 side chain (2.5 Å). In this case the benzyl ring of 33 sits about 3.5 Å from the Cα of Gly235 of a crystallographically related molecule. Stage 2: Tetrazole Chemotype (stage 1 hits 5 and 6). Eight tetrazole analogues were selected in the stage 2 compound set. Of these, three compounds comprise a disubstituted tetrazole with a substituent on the tetrazole N-1 nitrogen atom. These compounds were not hits by either MS or SPR. This finding is consistent with the X-ray crystal structure for the stage 1 tetrazoles wherein there is a direct interaction between the N-1 atom and the CA II active site zinc, reinforcing the likely importance of the interaction for specific binding to CA II. Compound 36 is a 3-chlorophenoxy isomer, while 37 is a 3,4-dichlorophenoxy analogue of the stage 1 tetrazole hits, which have a 2-chloro- and a 4-chlorophenoxy substituent, respectively. Both 36 and 37 were hits by SPR, MS, and X-ray crystallography; this is consistent with maintenance of the same key interactions with CA II in the presence of these subtle structural variations in the chlorine substitution pattern. Compound 40 is also a hit compound by SPR and MS that was confirmed by X-ray crystallography. This tetrazole lacks the ether oxygen of the substituent of stage 1 tetrazole 6 but is otherwise identical; this finding suggests that the oxygen of the tetrazole substituent is not critical for the fragment interaction with CA II. Compound 38 has an N-1 hydroxyl group and is classified as a hit, albeit weak, using nanoESI-MS; however, this compound was above the SPR threshold for a hit (KD = 4100 μM), and a crystal structure was not attempted. Compound 39 is a SPR hit (KD = 2280 μM) and a weak hit by nanoESI-MS. This fragment comprises the less flexible 4-chlorophenyl substituent. X-ray crystallography was attempted; however, the compound was not observed. Stage 2: 1,2,4-Triazole Chemotype (stage 1 hit 7). There were no analogues comprising the 3-(2-phenylethanyl)thio-1,2,4-triazole substructure available in the full CSIRO compound collection. A comparison of the features of the popular fragment screening methods NMR, X-ray crystallography, SPR, and ITC and to a lesser extent MS have been described by others.14,57,58 There is however no definitive study that has systematically compared the attributes of ESI-MS, SPR, and X-ray crystallography methods when using an identical protein and fragment library combination. Herein we have combined native

carboxylate group. These eight compounds were not identified as hit fragments by MS or SPR, consistent with the loss of the carboxylate−zinc interaction. The remaining three benzoic acid chemotypes (29−31) consist of a free carboxylate. All were hits by MS, with 29 also a hit by SPR (KD = 2740 μM respectively). Compounds 31 and 30 were above our threshold for a SPR hit. Compound 30 is benzophenone-2-carboxylic acid, while 31 is a substituted benzophenone-2-carboxylic acid. Compound 30 was the only stage 2 benzoic acid analogue evaluated by X-ray crystallography; however, no fragment was detected. Stage 2: Phenylacetic Acid Chemotype (stage 1 hit 4). Six analogues of 4 were selected for stage 2 follow-up. Two compounds were not hits by either MS or SPR, while the remaining four compounds were hits by nanoESI-MS (32, 33, 34, and 35) with measurable KD values. Only compound 35 was within the cutoff to be considered a reliable SPR hit (KD = 826 μM). Compound 35 is an interesting compound from a structural standpoint. It may be considered an analogue of both the phenylacetic acid chemotype 4 and the cinnamic acid chemotype 2, having both functionalities present in a head-totail arrangement within its structure. Owing to the dual carboxylate chemotypes within this compound, the orientation of binding to CA II was of great interest. A protein X-ray crystal structure was obtained for 35 by both soaking and cocrystallization. In both structures, 35 is observed directly bound to the zinc and is oriented similarly to 4, with the phenylacetic acid carboxylate associated with the zinc, although offset by 1.9 Å (Figure 4). Compound 35 was modeled in two different conformers in the soaked structure and as a single

Figure 4. hCA II structure with bound 4 was superposed on to the hCA II structure with bound 35 using the SSM algorithm54 as implemented in Coot.55 The structure of 35 is colored with green carbons whereas the structure of 4 has cyan carbons with the zinc cation shown as a gray sphere; also shown are Thr199 and the three histidine residues which coordinate the zinc in the active site. The distance from the zinc to the closest oxygen in the carboxylic acid of either 4 or 35 is 1.8 to 2.0 Å, and the distance to the second oxygen atom of the acid moiety is 2.5 to 2.7 Å. The distance between the two carboxylic carbons of 4 and 35 is 1.9 Å. The figure was made with Pymol.56 2199

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry



Article

CONCLUSION Native state MS has been recognized as a rapid, sensitive, high throughput, and label-free method to directly investigate protein−ligand interactions for some time; however, studies using this approach as a screening method to identify relevant protein−fragment interactions in FBDD are few. Here we show the first fragment screening analysis of ESI-MS and nanoESIMS using a high resolution FTICR instrument in parallel with the most popular primary fragment screening method of SPR. The observation of [protein + fragment] m/z and determination of Δm/z was straightforward and enabled the classification of a fragment as a “binder” or “nonbinder”, and if a “binder”, it provided the molecular weight as an inbuilt identifier associated with fragment identity. Very high overlap of SPR with ESI-MS was observed, while follow-up X-ray crystallography also provided excellent correlation. Our findings support a place for the wider spread implementation of native state MS as a complementary primary fragment screening method, requiring only the combination of fragment and protein, without further manipulation such as sample immobilization or chromatography. The MS method thus may assist to prioritize fragment candidates for detailed kinetic or structural interrogation. The low fragment concentration needed with ESI-MS or nanoESI-MS (10 μM fragment), compared to more popular methods (300−500 μM of fragment), has an added advantage in that this widens the accessibility to fragments that may be otherwise discarded from fragment libraries owing to poor fragment solubility. In future screening campaigns we plan to assess the full CSIRO fragment library by ESI-MS.

state ESI-MS, together with two popular and established fragment screening methods, SPR and X-ray crystallography, in a fragment screening campaign to target CA II. In an initial fragment screen against a 720-member CSIRO Fragment Library, seven CA II binding fragments, including a selection of nonclassical CA II binding chemotypes, were identified by SPR. Follow-up nanoESI-MS was in full agreement (100% overlap of hits). Stage 2 screening was conducted on 70 compounds that comprised the initial stage 1 hit chemotypes within their structure. For stage 2, the screening was conducted independently and in parallel using ESI-MS, nanoESI-MS, and SPR. X-ray crystallography was then performed on selected stage 2 hit fragments where SAR was of interest. As for stage 1 screening, the overlap between SPR and MS hits in stage 2 were in very good agreement, with 58 of 70 compounds in agreement (83% agreement), Table 2. All compounds identified as hits by SPR were also identified as hits by MS. It is important to highlight that the level of agreement is dependent on the arbitrary selection of a threshold to discriminate a hit from a nonhit. For SPR, this threshold was set at KD ≤ 3000 μM, while for MS, a peak intensity ratio above 1:0.02 for unbound CA II:fragment bound CA II peaks was defined as a hit. Follow up X-ray crystallography provided good correlation, with a total of thirteen data sets containing electron density maps with compound density, including a positive data set with compound density from six of the seven fragments in stage 1 and seven of fourteen compounds in stage 2. Recently, Klebe and colleagues evaluated a 361-member fragment library using five biophysical screening methods (thermal shift assay, STD-NMR, reporter-displacement assay, nanoESI-MS, and microscale thermophoresis) in parallel against endothiapepsin.59 This previous study, like the current study, is inspired by a need to assess the overlap of fragment screening methods. Although they employed a suite of methods, not all library fragments were suited to each method owing to solubility, purity, or aggregation limitations; for example, only 206/361 fragments were assessed by STD-NMR. Their study also did not extend to SPR, one of the most common of the biophysical methods for FBDD.60,61 Unlike us, they report poor overlap among the different fragment screening approaches, with no individual fragment characterized as a hit by all methods. Native state ESI-MS was assessed, however, with a 5-fold excess of ligand (0.1 mM) to protein (20 μM), in contrast to our study, wherein an equal concentration of fragment and protein (10 μM of each) was analyzed. A high hit rate was reported for ESI-MS (66%), with many of these hits consisting of multiple binders. The selected solution conditions, with a 5-fold excess of fragment to protein, substantially increases the likelihood of observing nonspecificor colloidal aggregation−protein binding events in the gas phase, and this may lead to downstream difficulties associated with identifying a true hit from a false hit under the conditions.62,63 The ESI-MS and nanoESI-MS sample conditions of 1:1 fragment:protein employed in our study avoid the ambiguity that may arise by the formation of multiply bound complexes that occur at high fragment-to-protein ratios. The fragment-to-protein ratio used by fragment screening methods is worthy of further investigation by mass spectrometry to assess the impact of excess fragment concentrations on hit rates due to nonspecific and colloidal aggregate protein binding.



EXPERIMENTAL SECTION

Fragment Library and Fragment Analogues. The development of the CSIRO Fragment Library has been reported.31 The 720member compound library was stored in 96-deep-well plates at 10 mM concentration in DMSO within CSIRO Compound Library cold storage facility. Daughter plates were obtained from the initial master plates for both single point and dosage curve studies for the SPR experiments. Individual compounds were weighed out from stocks for the crystallography studies and mass spectrometry work. A number of fragment analogues were supplied by Sigma-Aldrich, including 9−18. The purity of the stage 1 and stage 2 compounds has been determined by UPLC with MS detection and NMR analyses. Purity of all compounds was ≥ 95%, except for 6 compounds that had purity levels between 80% and 95% (some are isomers which when combined would give higher overall purity). Expression and Purification of CA II for Biophysical Studies. The CA II protein was expressed and purified as previously described.64 Surface Plasmon Resonance Experiments. Minimal Biotinylation of CA II for SPR Studies. Minimal biotinylation of CA II was achieved by mixing 6.7 nmol of protein (1 equiv) with 5.0 nmol (0.75 equiv) of EZ-link sulfoNHS LC-LC-biotin (Thermofisher Scientific) in water (prepared fresh). The biotinylation reaction mixture was incubated on ice for 2 h. To remove unreacted biotin reagent, the biotinylation mixture was subjected to size exclusion chromatography on a Superdex 200 HR (10/300) column equilibrated in 1× PBS buffer pH 7.4. The minimally biotinylated CA II protein was stored in 50 μL aliquots at −80 °C. Immobilization of Biotinylated CA II on a Sensor Chip Surface. Streptavidin (Sigma) was simultaneously immobilized in all four channels of a CM5 sensor chip (GE HealthCare) docked in a Biacore T200 instrument as described previously.32 Approximately 10 500 response units (1 RU = 1 pg of protein/mm2) of streptavidin was immobilized with a minimal variation between the four channels on the same chip. Minimally biotinylated CA II protein (20 μg/mL) was 2200

DOI: 10.1021/acs.jmedchem.5b01940 J. Med. Chem. 2016, 59, 2192−2204

Journal of Medicinal Chemistry

Article

captured onto the freshly prepared streptavidin chip surfaces at 25 °C in 1× HBS-P+ as instrument running buffer. The CA II-biotin preparation was injected in a single channel at a constant flow-rate of 5 μL/min for 300 s, resulting in approximate immobilization levels of 12 500 RU. Streptavidin surface in flow-cell one was blocked with a double injection of D-biotin (10 μg/mL for 60 s at 5 μL/mL) and utilized as a reference channel. For negative protein target purposes (off-target control), minimally biotinylated human Hsp90 (N-terminal domain) protein was immobilized in one of the flow cells on the same chip. Fragment Library Screening and Hit Validation. For the stage 1 screening campaign, the 720-member CSIRO Fragment Library collection was screened against biotinylated CA II with SPR as a single-concentration screening experiment at 15 °C. Fragment stock solutions (10 mM in DMSO) were diluted stepwise to a final concentration of 0.1 mM, first diluting (1 in 33.3) to an intermediate concentration of 0.3 mM with 1.03× DMSO-free screening buffer (51.5 mM HEPES, pH 7.4, 152.5 mM NaCl, 0.0515% Tween-20 (v/ v)) and then a further 1 in 3 dilution in the fragment binding buffer (50 mM HEPES, pH 7.4, 150 mM NaCl, 0.05% Tween-20 (v/v), and 3% DMSO (v/v)). Two separate fragment screen runs were performed, each time screening 360 fragments. Fragment solutions (0.1 mM) were injected over the protein sensor surface at 60 μL/min for 30 s association and 30 s dissociation. To validate CA II protein stability on the chip surface, 4-carboxybenzenesulfonamide (4-CBS, control compound) was injected after every 120 fragment injections. To determine binding affinity of candidate hit fragments as well as the follow-up fragment analogues in the stage 2 screening campaign, SPR dose−response experiments were performed at 25 °C. Fresh 100 mM DMSO fragment solutions were diluted directly with the fragment binding buffer to a final concentration of 200 μM and then diluted 2fold to 12.5 μM aiming for a five-point concentration series range for the SPR dose−response experiment. Each compound was injected for 30 s association and 60 s dissociation. SPR Data Analysis and Hit Selection Principles. Scrubber 2 (www. biologic.com.au) and Microsoft Excel software packages were utilized for data processing and analysis. Using Scrubber software, SPR signals were referenced against the blank surface (streptavidin + D-biotin) and further corrected for DMSO refractive index change as previously described.65 A normalization scheme of Giannetti et al.33 was applied to the processed data on the basis of the maximal binding response (Rmax) that has been determined from experiments with the control compound, 4-CBS. For the initial ranking of the best hits, the KD values were estimated using eq 1 derived from the Langmuir adsorption isotherm, KD =

R max × C −C R eq

CA II was centrifuged at 14 000g for 10 min on a Heraeus Pico 21 benchtop centrifuge (Thermo Fisher). The filtrate was discarded, and the concentrate was reconstituted into 450 μL of 10 mM NH4OAc pH 7.0 buffer. This washing phase was repeated five times, to ensure that any residual salts from the microsolute were kept at a minimal level. The concentrate was recovered by centrifugation at 1000g for 1 min. NH4OAc pH 7.0 buffer (10 mM) was added to the final concentrate to give a CA II protein concentration of 25 μM. The protein concentration was checked using A280 with an extinction coefficient of 5.4 × 104 M−1 cm−1.67 For ESI analyses, samples were infused into an ESI Apollo II source (Bruker) with a flow rate of 120 μL h−1. Positive ion ESI was used with a capillary voltage of 3.5 kV, a drying gas flow of 2.0 L min−1, and a nitrogen nebulizing gas pressure of 3.0 psi. Mass calibration was performed by an infusion of perfluorohexanoic acid cluster ions. Data were acquired over the range m/z 300− 10 000 with the quadrupole set at 600 m/z. Instrument parameters were optimized to maximize signal intensity while ensuring that conditions remained gentle and that the protein retained in a nativelike state. In summary, a capillary voltage of −3500 V, a drying temperature of 160 °C, a skimmer 1 voltage of 60 V, and a time-offlight of 2.3 ms was used. For nanoESI analyses, sample was infused using a Triversa Nanomate (Advion Biosciences) automated nanoESI interface. Positive ion ESI was used with a capillary voltage of 1.2 kV and a nitrogen nebulizing gas pressure of 0.4 psi. Mass calibration was performed by an infusion of perfluorohexanoic acid cluster ions, and the raw data was processed using DataAnalysis 4.2. Data were acquired over the range m/z 300−10 000 with the quadrupole set at 600 m/z. Instrument parameters were optimized to maximize signal intensity while ensuring that conditions remained gentle and that the protein remained in a native-like state. In summary, a drying temperature of 100 °C, a skimmer 1 voltage of 15 V, and a time-of-flight of 2.3 ms was used. Test compound dry stocks were dissolved in DMSO to give 5 mM stock solutions. For standard ESI analysis, 0.25 μL of each stock solution was added to a 50 μL aliquot of CA II (25 μM) in 10 mM NH4OAc, giving a final protein:ligand ratio of 1:1 with 0.5% DMSO present. Samples were incubated for 2 min and were then infused into the MS using the syringe pump mounted to the solariX XR. Experiments were repeated after 15 min to see if additional incubation time resulted in additional ligand binding. For nanoESI analyses, test sample was similarly prepared with