Identification of C3b-binding Small Molecule Complement Inhibitors Using Cheminformatics

Abstract

The complement system is an elegantly regulated biochemical cascade formed by the collective molecular recognition properties and proteolytic activities of over two dozen membrane-bound or serum proteins. Complement plays diverse roles in human physiology which include acting as a sentry against invading microorganisms, priming of the adaptive immune response, and removal of immune complexes. However, dysregulation of complement can serve as a trigger for a wide range of human diseases which include autoimmune, inflammatory, and degenerative conditions. Despite several potential advantages of modulating complement with small molecule inhibitors, small molecule drugs are highly underrepresented in the current complement-directed therapeutics pipeline. In this study we have employed a cheminformatics drug discovery approach based on the extensive structural and functional knowledge available for the central proteolytic fragment of the cascade, C3b. Using parallel in silico screening methodologies we identified 45 small molecules which putatively bind C3b near ligand-guided functional hot-spots. Surface plasmon resonance experiments resulted in the validation of seven dose-dependent C3b-binding compounds. Competition-based biochemical assays demonstrated the ability of several C3b-binding compounds to interfere with binding of the original C3b ligand which guided their discovery. In vitro assays of complement function identified a single complement inhibitory compound, termed cmp-5, and mechanistic studies of the cmp-5 inhibitory mode revealed it acts at the level of C5 activation. This study has led to the identification of a promising new class of C3b-binding small molecule complement inhibitors, and to our knowledge, provides the first demonstration of cheminformatics-based complement-directed drug discovery.

1. Introduction

Human complement is best known as an intravascular system consisting of ~30 membrane-bound or serum proteins whose pattern and surface recognition properties give rise to potent antimicrobial effector functions. While complement’s role in host defense is well established, current views place complement at the nexus of several important physiological processes including homeostatic maintenance, priming of adaptive immune responses, and clearance of apoptotic debris and immune complexes (1). Many of the complement components present in serum are synthesized in the liver, however, nearly all human cell types are capable of producing complement proteins (2, 3). For instance, the major source of complement in immune-privileged sites, such as the brain, are a product of local biosynthesis (3, 4). Emerging evidence suggests that locally synthesized complement maintains distinct functions from systemic serum complement (3), and this has been exemplified by studies which demonstrate a crucial role for extrahepatic complement in the induction and modulation of T cells (5–7). Thus, in addition to acting as a sentinel against invading pathogens, complement is an extensive and diverse player in the broader context of human physiology.

As our understanding of the physiological functions for complement have improved, so too has our awareness of its role in pathological processes (8, 9). Inappropriate complement activation or dysregulation contributes significantly to an ever growing list of autoimmune, inflammatory, proteinuric, ischemia-reperfusion, and neurodegenerative diseases and conditions (8, 9). Although there has been a long-standing interest in the development of complement-directed therapeutics, the field has been undoubtedly energized over the past decade by the development of the anti-complement drug eculizumab (Alexion Pharmaceuticals). Eculizumab is currently approved by the US Food and Drug Administration for the treatment of paroxysmal nocturnal haemoglobinuria (PNH) (10) and atypical haemolytic uremic syndrome (aHUS) (11), and is in various stages of clinical development for nearly 20 separate indications (12). Despite the relative clinical success of eculizumab, the future outlook of complement-directed therapeutics is met with several challenges. For example, the estimated per patient per year cost of eculizumab, which is in excess of $350,000 USD for treatment of PNH, has been the subject of international scrutiny (12, 13). Furthermore, it is now clear that the specific nature of complement’s involvement in a particular pathology likely precludes a ‘one size fits all’ model for treatment of complement related diseases (12, 14). Together these factors have accelerated efforts to develop novel complement-directed drugs which specifically target and inhibit discrete steps within the cascade.

While examples of complement activation by extrinsic proteases are increasingly known (9, 15), complement is conventionally described as being triggered by three pathways (classical, lectin, or alternative) which are defined by their underlying modes of pattern recognition and/or activation mechanism. All pathways converge on the central molecule of the cascade, complement component C3, which is cleaved by surface assembled multi-subunit enzymes called convertases. Upon C3 cleavage, the anaphylatoxin C3a is released, while the opsonic fragment, C3b, forms a covalent bond to the activating surface via exposure of a previously protected thioester moiety. Surface-attached C3b serves as a non-catalytic scaffold for new C3 and C5 convertases and downstream cleavage products of C3b (i.e. iC3b and C3dg) promote clearance by directly interacting with complement receptors expressed on the surface of circulating immune cells (16).

Due to its central position in the cascade, C3 has been considered as an ideal drug target for certain complement-related diseases such as C3 glomerulopathies and age-related macular degeneration (AMD) (17). In this regard, a promising intervention point appears to be the targeting of C3b’s prominent role in C3 and C5 convertases (12, 17). C3b interacts with complement factor B (fB) and factor D (fD) to form the alternative pathway (AP) C3 convertase and at high C3b surface densities acts as a major component of both AP and classical pathway/lectin pathway (CP/LP) C5 convertases. Importantly, monoclonal antibodies against C3b which block the binding of factor B have been shown to effectively disrupt AP C3 convertase formation and show promising therapeutic potential in in vitro disease models of PNH and dense deposit disease (17–19).

Therapeutic intervention strategies involving the targeting of large protein-protein interactions like those formed by C3b (~175 kDa) in convertases (C3bBb = ~235 kDa; C3bBbC3b or C4b2a3b = ~410 kDa) has traditionally been thought to be restricted to the realm of biologics. However, recent developments in the area of low molecular weight complement inhibitors have challenged this view (20, 21). A prominent example is compstatin, which is a 13-residue cyclic peptide that binds to C3/C3b/C3c and prevents C3 cleavage by AP and CP C3 convertases (22). Compstatin analogues are a promising drug candidate for several complement-related diseases and to date have been evaluated in disease models of AMD, sepsis, PNH, haemodialysis-induced inflammation, periodontal disease, and transplantation (23). While compstatin is of synthetic origin, the feasibility of lower molecular weight complement inhibitors has also been provided by nature itself. The human bacterial pathogen, Staphylococcus aureus, secretes several small C3b-binding proteins which act as potent inhibitors of complement (24) and peptidic mimics (< 1.5 kDa) of one family of these immune evasion molecules, staphylococcal complement inhibitor (SCINs), have been reported (25).

Among the advantages of small molecule drugs compared to biologics/proteins are lower production cost, lack of immunogenicity, and greater formulation flexibility. Despite this, relatively few rule-of-five adherent compounds (i.e. < 500 Da; logP < 5; H-bond donors < 5; H-bond acceptors < 10) (26) have reached clinical trials for complement-related indications (12, 21). A particular challenge facing the development of drug-like compounds as complement-directed therapeutics arises from the potential difficulty of disrupting large protein-protein interactions with molecules no bigger than 500 Da. Unlike antibodies or other macromolecule-based drugs, where steric contributions greatly increase the likelihood of successfully disrupting the target protein-protein interaction, small molecules must be precisely targeted to critical interaction-mediating sites. Nonetheless, successful examples of small molecule compounds which modulate protein-protein interactions have steadily increased over recent years in fields such as oncology (27, 28).

Our work in the area of bacterial complement evasion proteins has led to the hypothesis that the diverse functional activities of C3b can be disrupted by targeting relatively small hot-spots on an otherwise large C3b surface. In this study, we have leveraged the substantial structure-function knowledge that exists for C3b and certain C3b ligands in combination with a directed cheminformatics approach. We report the discovery of several C3b-binding small molecules, including one known hereafter as “cmp-5”, which uniquely inhibits the formation of the membrane attack complex (MAC) by preventing cleavage of C5. To our knowledge cmp-5 and the cmp-5 analogues reported here are the first small molecules which bind specifically to C3b and have anti-complement activity. As such, these inhibitory compounds and the novel in silico screening methodologies used for their identification, have significant implications for the future development of complement-directed therapeutics.

2. Materials and Methods

2.1 Proteins and compounds

All compounds were commercially synthesized and obtained from ChemDiv (factor B library 1), ChemBridge (factor B library 2), or Enamine (compstatin and SCIN libraries). Each compound was resuspended to a final concentration of 10 mM in DMSO (Gold Biotechnologies) and stored at −20°C until just prior to use. Synthesis of compound “cmp-5” / N-(3-(5-methyl-1H-pyrazol-1-yl)phenyl)-2-(3-oxo-1,3-dihydroisobenzofuran-1-yl)acetamide was performed as follows: 2-(3-oxo-1,3-dihydroisobenzofuran-1-yl)acetic acid (0.12g, 0.62 mmol), TBTU (0.205g, 0.62 mmol) and Hunig’s base (350 μl, 1.86 mmol) were taken in DMF (5 mL) and stirred for 30 min. 3-(5-methyl-1H-pyrazol-1-yl)aniline (0.13g, 0.74 mmol) was added to the reaction and stirred for 14h. The reaction mixture was subsequently extracted with saturated bicarbonate and brine, dried and concentrated to get the crude mixture. The crude mixture was purified on silica using a gradient of ethyl acetate-hexane (20–45% (v/v) EA in 10 min) to obtain the title compound (0.06g, 27% yield). ¹H NMR (400 MHz, DMSO-d₆) δ ppm 2.27 (s, 3 H) 2.83 (dd, J=15.41, 8.56 Hz, 1 H) 3.19 (d, J=4.40 Hz, 0 H) 3.23 (d, J=4.40 Hz, 0 H) 6.02 (dd, J=8.44, 4.52 Hz, 1 H) 6.33 (d, J=2.20 Hz, 1 H) 7.35 – 7.52 (m, 3 H) 7.60 – 7.66 (m, 1 H) 7.72 – 7.77 (m, 1 H) 7.78 – 7.84 (m, 1 H) 7.87 (d, J=7.82 Hz, 1 H) 8.07 – 8.16 (m, 1 H) 8.30 (d, J=2.45 Hz, 1 H) 10.27 (s, 1 H). MS (m/z): 348.1 (M+H).

C3, C3b, C3c, C5, CVF, factor D and factor B (fB) were obtained as purified proteins (Complement Technologies). Compstatin peptide was obtained from Tocris/R&D Systems while the compstatin derivative Cp40 was a generous gift from Dr. John D. Lambris (University of Pennsylvania). For production of the thioredoxin-compstatin fusion protein (TRX-4W9A), a DNA sequence was designed to encode for Escherichia coli F11 thioredoxin (Genbank #: EDV64981.1), followed by a TEV recognition sequence and Gly-Ser linker (amino acid sequence: ENLYFQGGSGSGSG), followed by the amino acid sequence ICVWQDWGAHRCT which corresponds to the affinity-optimized compstatin peptide 4W9A (29–32). An E. coli codon-optimized DNA sequence, flanked with a 5′ BamH1 site, a 3′ Not1 site and a stop codon was synthesized commercially using IDT Technologies’ gBlock Gene Fragment service. The resulting DNA fragment was cloned into the pT7HMT vector, expressed in E. coli Oragami(DE3), and purified according to previously published protocols (33). An identical protocol was followed to generate a recombinant fB truncation mutant encoding for the three N-terminal complement control protein (CCP) domains of the Ba fragment (residues 35–220, UNIPROT#: P00751).

2.2 ChemVassa, molecular docking, and compound scoring using AutoDock Vina

The “Leads Now” set of 4.27 million ligands was downloaded as MOL2 files from the ZINC database (http://zinc.docking.org/subsets/leads-now). This subset of the ZINC library includes all “In Stock” compounds that have a molecular weight between 250–350, xLogP values less than or equal to 3.5, and no more than 7 rotatable bonds. The files were converted to PDBQT files using the prepare_ligand.py script from AutoDockTools1.5.6 (34). The protein target was prepared using Chain A of the crystal structure of C3b from the PDB file 2XWJ (35). All ligand and water molecules were removed and hydrogens were added using AutoDockTools1.5.6 before generating the PDBQT file. Binding boxes were selected for distinct sites: the fB CCP-3 domain (“fb-1/fb-2”) or the C3c/compstatin site (31). Coordinates for the center x,y,z and size [x,y,z in Å] of target boxes are; fb-1/fb-2: center [60, −30, −40], size [26, 28, 30] and C3c/compstatin: [115, −70, −15], [22, 30, 36]. Molecular docking was then performed using AutoDock Vina 1.1.2 (36) as described previously (37). Scores corresponding to six poses for each compound were collected for the full dataset. Compounds scoring 4.5 standard deviations from the average docking score were selected for small molecule comparison analysis and availability from selected vendors.

We used an alternative cheminformatics approach to screen for small molecules that are chemically similar to a region of S. aureus SCIN-B that has previously been shown to be critical for C3b-binding and complement inhibition (25, 38). A novel chemical information software tool, ChemVassa (39), was used to calculate a numeric fingerprint for residues originating from the second α-helix of SCIN-B. The fingerprint can be compared across molecules to calculate a bitwise score which looks at the position, charge, and atomic properties of each atom across two molecules. The difference is called the “G-score”. A G-score of zero represents an identical molecule, and the bitwise score has a high dynamic range across which similarity can be measured. These scores were then used to search the ZINC database subset “Clean Drug-like” consisting of 14.4 million compounds for small molecules with similar chemical information signatures. Results were triaged by establishing a cutoff two bits from the maximal score matched G-score across all ZINC hits (39). This stringent cutoff resulted in six hits which were then docked onto C3b and scored by theoretical binding energy using AutoDock Vina as described above.

2.3 Small molecule comparison analysis

The ChemMine Web Tools software package (40) was used to analyze and cluster small molecules by structural similarity. Hierarchical clustering using a single linkage method was performed against all compound libraries and a distance matrix-based heat map was generated.

2.4 Surface plasmon resonance

Direct binding of each compound to C3b was measured by surface plasmon resonance (SPR) using a Biacore T200 instrument (GE Healthcare) at 25°C. CMD-200 sensorchips (Xantec) were used to create C3b biosensors by standard amine coupling chemistry as follows. An equal volume mixture of 0.1M N-hydroxysuccinimide (NHS) and 0.4M ethyl(dimethylaminopropyl) carbodiimide (EDC) was injected for 7 min at 5 μl min⁻¹ over a single flowcell. Next, 50 μg ml⁻¹ C3b diluted in 10 mM sodium acetate (pH 5.0) was injected for 15 min and immediately followed by a 7 min injection of 1 M ethanolamine (pH 9.0) to quench remaining reactive groups. In total five separate amine coupled C3b surfaces were generated during the course of this study with the following immobilization levels: 16,400, 10,200, 16,000, 18,300, and 17,000 resonance units (RU). All experiments were performed in a running buffer of 20 mM HEPES (pH 7.3), 140 mM NaCl, 0.005% (v/v) Tween 20, and 5% (v/v) DMSO (HBST-DMSO) using a flowrate of 30 μl min⁻¹. Solvent correction curves were obtained at the beginning, end, and after every 50 injection cycles by injecting varying DMSO concentrations (4.0, 4.4, 4.6, 4.8, 5.0, 5.2, 5.4, and 5.9% (v/v)). A reference flowcell was created on each sensorchip by EDC/NHS activation followed by immediate ethanolamine quenching.

To evaluate C3b binding, each compound was diluted from 5 mM stock solution in neat DMSO to 250 μM final concentration in a buffer containing 1.05x HBST. Samples were injected over the C3b biosensor (18,300 RU surface) for 30 s followed by 15 s of dissociation phase and a subsequent wash step with a 50% (v/v) DMSO solution. Duplicate injections were performed for all samples and a buffer only injection was performed every 12 injection cycles. The experiment was concluded with duplicate control injections of the compstatin derivative Cp40 at 100 μM. The resulting sensorgrams were analyzed using Biacore T200 Evaluation Software v3.0 as follows. Solvent correction curves were generated and applied to all datasets and each sensorgram was double referenced by subtracting the nearest buffer blank injection. The signal just prior to injection stop of these corrected sensorgrams was treated as the C3b-binding response and was normalized to the molecular weight of each compound by expressing this value as RU/100 Da. Dose-dependency of the strongest binding compounds was then evaluated by injecting varying concentrations of each compound (31, 63, 125, 250, and 500 μM) and the resulting data was treated as described above. Depending on the compound identity, a minimum of three, and up to seven, independent variable concentration injection series were performed across the three to five C3b surface preparations. Steady-state affinity analysis was performed on each by allowing the maximal binding signal to be freely fit (R_max,_fit) or by constraining this value to the empirically derived control signal (R_max,Cp40) and a mean K_D was computed. The ratio of R_max,_fit/R_max,Cp40 was used to evaluate superstoichiometric binding behavior as described in the Results. Identical methods were used for the evaluation of cmp-5 binding to native C3 (25,900 RU) and C3c (12,690 RU) using duplicate variable concentration injections.

2.5 Complement inhibition assays

The ability of each compound to inhibit complement activation by the alternative pathway (AP) or classical pathway (CP) was performed as previously described (25, 41) with the following modifications. Each experiment began with overnight coating of ELISA plates (Costar EIA/RIA, Fisher Scientific) with either 25 μg ml⁻¹ Salmonella typhosa lipopolysaccharides (LPS) for the AP (Sigma Aldrich) or 3 μg ml⁻¹ human IgM (ImmunoReagents) for the CP in a buffer containing 100 mM Na₂CO₃/NaHCO₃ pH (9.6) followed by blocking with a buffer of PBS, 1% (w/v) bovine serum albumin, and 0.05% (v/v) Tween-20 for 1 h at 37°C. All volumes were 100 μl unless otherwise noted and all steps were preceded by washing three times with a buffer containing 50 mM Tris (pH 8.0), 150 mM NaCl, and 0.05% (v/v) Tween-20.

For AP experiments a buffer of 20 mM HEPES (pH 7.5), 0.1% (w/v) gelatin, 140 mM NaCl, 5 mM MgCl₂, and 5 mM EGTA was used while CP experimental buffer was composed of 20 mM HEPES (pH 7.3), 0.1% (w/v) gelatin, 140 mM NaCl, 2 mM CaCl₂, and 0.5 mM MgCl₂. A final concentration of 20% (v/v) (AP) or 1% (v/v) (CP) normal human complement serum (Innovative Research) was used in single compound concentration assays. Compounds were diluted from a 10 mM stock in neat DMSO to yield a 250 μM final compound concentration and 2.5% (v/v) DMSO concentration into serum containing AP or CP buffer. The serum/compound mixture was transferred to the washed ELISA plate and incubated at 37°C for 1 h. After washing, deposited complement components were detected using monoclonal antibodies (anti-C3b: WM1, Sigma; anti-C4b: C4-1, Cell Sciences; anti-MAC/C5b-9: aE11, Santa Cruz Biotechnology) at 1/300 dilutions and detected as previously described (25). All measurements were normalized by treating the mean value of vehicle control as 100% signal.

2.6 Label-enhanced SPR (LE-SPR)

For all LE-SPR experiments a C3b biosensor was created in a manner which uniformly captures C3b in a physiologically relevant orientation and has been described in detail elsewhere (25, 38, 42). Briefly, C3b was site-specifically biotinylated using a CVF-based C3 convertase and EZ-link Maleimide-PEG₂-biotin (Thermo Scientific) through a protocol identical to that described by Garcia et al. (38). Following purification, C3b-biotin was then captured on a flowcell of a Xantec CMD-200 sensor chip which had previously been amine coupled with NeutrAvidin Protein (Thermo Scientific). A total of eight C3b biosensors were created for use across all LE-SPR experiments. For TRX-4W9A a 700 RU surface was used for the conventional SPR, LE-SPR direct binding, and LE-SPR control competition experiments, while a 3,800 and 3,100 RU surface was used for compound competitions. For C5 experiments a 6,560 RU surface was used for conventional SPR experiments, a 5,160 RU surface was used for LE-SPR direct binding, and a 4,460 RU was used for LE-SPR competitions. For fB 6,940 and 5,330 RU surfaces were used for all experiments. A reference flowcell was created on each sensorchip as described above for use in conventional SPR experiments while LE-SPR experiments do not require use of reference surface.

Conventional SPR experiments were carried out to characterize the binding of native TRX-4W9A, C5, and fB to C3b. These experiments were performed in a running buffer of HBST with the exception of fB wherein HBST was supplemented with 5 mM Mg²⁺. All analytes were injected for 2 min followed by 3 min dissociation phase at a flow rate of 30 μl min⁻¹. Regeneration was not required for any of the complexes under study here. In each case, K_D’s were calculated using reference corrected sensorgrams from a two-fold variable concentration injection series which were fit to a kinetic 1:1 Langmuir binding model using Biacore T200 Evaluation Software.

LE-SPR experiments began with dialysis of each analyte (TRX-4W9A, C5, or fB) into phosphate buffered saline (pH 7.4) for 2 h at room temperature. Each sample was then incubated for 1 h in the dark with 5-fold molar excess of the dye reagent Episentec^™ NHS-B23 (Episentum) followed by dialysis into HBS at 4°C overnight. A two-fold variable concentration dilution series of each B23 labeled analyte was injected over C3b using the predefined method “LE-SPR Concentration Analysis/Inhibition in Solution Assay” provided by Episentum. This procedure uses a 10 μl min⁻¹ flow rate and 1 min association and dissociation times. In addition to collecting standard sensorgrams this procedure also records SPR dip curve data. An injection of unlabeled analyte (i.e. non-B23 labeled or “cold” analyte) matching the highest B23-labeled concentration in each injection series was also included. Calculation of absorption sensorgrams (termed “epigrams”) was carried out for each dataset using EpiGrammer^™ software by utilizing the sensorgram enhancement feature. A weighted sum of the Dip Width and Dip Position of the unlabeled analyte injection was calculated and applied to all injections. The arbitrary units of the epigram are enhanced resonance units (eRU), which are defined to correspond with approximately 1 RU (i.e. conventional SPR unit) of B23 labeling dye (43, 44). To obtain a K_D, the resulting epigrams were then fit to a kinetic 1:1 Langmuir binding model using BiaEvaluation Software v3.0.

LE-SPR-based competition assays were carried out as above with the following modifications. For all experiments involving small molecule compounds, the running buffer was supplemented with 5% (v/v) DMSO. Reactions were prepared as described above for conventional SPR direct binding assays involving small molecule compounds. A fixed concentration of B23-labeled analyte (TRX-4W9A and fB = 500 nM; C5 = 250 nM) was co-injected with each putative competitor molecule. A parallel competitor injection lacking the B23-analyte was used for calculating individual epigrams. The residual steady-state binding signal for each injection was calculated by averaging the eRU for 10 seconds prior to injection stop. B23-analyte injections lacking a competitor species or containing DMSO only vehicle control were treated as the maximal binding signal.

2.7 AlphaScreen Competition Assays

We developed an AlphaScreen bead based competition assay to evaluate the ability of fB-2 library compounds to compete with the Ba fragment of fB for C3b-binding. All reaction mixtures consisted of 250 nM recombinant HIS-Ba(1–3), 5 nM C3b-biotin (described above), 20 μg ml^−l nickel chelating AlphaScreen Acceptor beads, 20 μg ml^−l streptavidin AlphaScreen Donor beads, and a putative competitor molecule. A 25 μl reaction volume in a buffer containing 20 mM HEPES (pH 7.4), 140 mM NaCl, 0.1% (w/v) bovine serum albumin, and 0.05% (v/v) Tween-20 was used in all experiments. Assays involving small molecule compounds were carried out in a final DMSO concentration of 2.5% (v/v) and assays involving full-length fB were supplemented with 5 mM Mg²⁺. Competition experiments began by mixing HIS-Ba(1–3), a putative competitor, and C3b-biotin together followed by a 30 min incubation. At this time AlphaScreen Acceptor beads were mixed into each reaction and allowed to incubate for 15 min in the dark. Finally, AlphaScreen Donor beads were added and a final 15 min dark incubation was carried out. Reactions were transferred to a ½ area 96-well microplate (PerkinElmer) and read on an Enspire Alpha multimode plate reader (PerkinElmer). AlphaScreen signals were normalized to wells containing vehicle only control (2.5% (v/v) DMSO). Where applicable, K_D’s were calculated from fits obtained using four parameter variable slope nonlinear curve-fitting in GraphPad Prism5 software as previously described (25, 38, 45).

2.8 C5a Detection Assays

We probed CP ELISA reaction mixtures for C5a through a C5a capture ELISA. A monoclonal antibody which specifically recognizes a C5 activation-dependent neoepitope on C5a (46) (C17/5, BioLegend, San Diego, CA) was coated overnight at 1 μg ml⁻¹ in 100 μl volume on Costar EIA/RIA plates at room temperature. After blocking for 1 h at 37°C and washing, a 2-fold dilution of each CP ELISA reaction (at completion) was added to the plate and incubated for 1 h at room temperature while rocking. After washing, rabbit anti-human C5a polyclonal antibody (Complement Technologies) which had been previously purified (HiTrap Protein G, 1mL, GE Healthsciences) and concentrated to 1.0 mg ml⁻¹ was used to detect C5a at 1/200 dilution. Following a 1 h incubation at room temperature and washing, goat-anti-rabbit IgG-HRP (Thermo Scientific) was added at 1/5000 dilution for 1 h and developed using 1-step Ultra TMB ELISA Substrate Solution according to manufacturer’s instructions. Plates were read at 450 nm using a VersaMax microplate reader and all wells were normalized by treating the mean of the vehicle control value as 100% signal.

2.9 Statistical Analysis

All statistical analysis was performed using GraphPad Prism v5.04. Measures of statistical significance were assessed using an unpaired, two-tailed, t-test of each experimental series versus vehicle control where ^*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001, ^****p ≤ 0.0001. IC₅₀/EC₅₀ values were obtained by four-parameter variable slope nonlinear regression analysis where the top and bottom of the curves were constrained to 0 and 100.

3. Results

3.1 A cheminformatics approach to complement-directed drug discovery

We sought to identify small molecules which bind C3b near sites of functional importance under the premise that such compounds may possess complement inhibitory activity. We began by selectively docking the 4.27 million compound “Leads Now” subset of the ZINC database onto defined spaces of the C3b surface that overlap binding sites of two separate C3b ligands (fB or compstatin). In order to facilitate a SCIN-specific screen we used an alternative methodology, termed ChemVassa, which evaluates the structural and chemical features of the ligand (i.e. SCIN) rather than the target surface (i.e. C3b) (39). We have previously determined that SCIN residues originating from the second α-helix are critical for mediating C3b recognition (25, 38) and a chemical information signature for this region of SCIN-B was calculated by ChemVassa (39). This value was then used to query the 14.4 million compound “Clean Drug-like” ZINC database subset for small molecules with similar chemical information signatures and top hits were subsequently docked onto C3b using Autodock Vina.

Following each in silico screen, predicted C3b binding energies were calculated and used to rank each compound. Highly scored compounds were further analyzed for predicted aqueous solubility by evaluating the xLogP value obtained from the ZINC database (47), and a prioritized subset of compounds were obtained commercially for the purpose of hit validation. C3b-binding and complement inhibitory activity were assessed in vitro followed by an iteration of hit expansion using structure-based similarity queries of validated hit compounds against the purchasable subset of the ZINC database (22.7 million compounds). A schematic of the discovery process is presented in Fig. 1.

Figure 1. Cheminformatics workflow schematic.

Small molecules that putatively bind C3b near functionally important “hot-spots” were identified using an in silico docking approach. Target binding boxes were defined on C3b corresponding to a region of the Ba domain contact site of the C3b/fB interface (PDB: 2XWJ) (30) or the entirety of the C3c/compstatin interface (PDB: 2QKI) (31). Next, 4.27 x 10⁶ unique drug-like compounds originating from the ZINC database “Leads Now” subset were docked into each set of coordinates using AutoDock Vina (32). Results were ranked according to calculated binding energies and compounds exhibiting +4.5 S.D. from the average value were triaged. This compound subset was then assessed for favorable predicted physicochemical properties (i.e. xLogP/solubility), structural diversity, and commercial availability. An alternative ligand-based method (ChemVassa) was used to identify and dock small molecules with chemical and structural similarity to SCIN-B. Here, the C3c/SCIN-B crystal structure (PDB: 3T4A), guided by previously obtained mutagenesis-based C3b affinity data (25, 34, 35), was used to calculate a ChemVassa information signature. A prioritized set of compounds from each screen were obtained via commercial synthesis and evaluated for direct binding to C3b (SPR) and complement inhibitory activity (ELISA). Validated hits were queried against the entire set of purchasable compounds in the ZINC database and those exhibiting high similarity were rescreened following an identical workflow.

3.2 Selection of C3b target sites, ChemVassa implementation, docking of ZINC compounds using AutoDock Vina, and identification of prioritized compounds for in vitro testing

Our targeted in silico screening approach was initiated with the selection of specific C3b ligands on the basis of three primary criteria: i) the availability of significant structure-function knowledge for the ligand including a relevant and published co-crystal structure; ii) a relatively small C3b-binding ‘footprint’ or biochemically defined interaction hotspot for the ligand, and iii) the potential for small molecules with an overlapping binding site on C3b to act as a complement inhibitor. In addition to these considerations, the selected ligands (fB, compstatin, and SCIN-B) were also chosen for their potential to uncover mechanistically distinct small molecule inhibitors due to their different C3b-related activities. For example, compounds which bind C3b near the fB site could potentially disrupt AP proconvertase formation (i.e. C3bB) and thereby interfere with the central amplification step of the cascade (Fig. 2A, top panel). On the other hand, compounds that interact with C3b near compstatin or SCIN-B, hold potential to block complement via compstatin-like or SCIN-like inhibitory mechanisms (Fig. 2B and C, top panels).

Figure 2. In silico screening of small molecules which bind C3b near functional “hot-spots”.

(A) AP C3 proconvertases are formed upon interaction of surface-attached C3b (grey) with fB (cyan oval). Subsequent cleavage of fB by the serine protease fD (red) results in active AP convertases (C3bBb) and release of the Ba fragment (open cyan triangle). Compounds capable of disrupting the C3b/fB protein-protein interface hold potential as complement inhibitors. A site formed between the CCP3 domain of the Ba fragment and C3b was selected as the target region for the fB-based screen (PDB: 2XWJ). The C3b/C3c contact residues used to guide each screen have been highlighted in yellow (middle panels) while the final Autodock Vina binding box site selection is shown on C3b in red (bottom panels). (B) Compstatin (orange pentagon) is a synthetic cyclic peptide which potently inactivates complement by binding C3 and preventing its cleavage by AP or CP/LP (i.e. C4b2a) C3 convertases. The co-crystal structure of C3c/compstatin (PDB: 2QKI) was used to define binding box coordinates for compound docking. (C) SCINs (purple cylinder) are multifunctional C3b-binding complement evasion proteins expressed by S. aureus which act at the level of AP C3 convertases. To identify potential small molecule mimics of SCIN, ChemVassa was used to calculate a chemical information signature for the primary C3b interaction site formed by residues on the second α-helix of SCIN-B (PDB: 3T4A). Compounds in the ZINC database with closely matched information signatures were subsequently docked onto C3b using Autodock Vina.

Among the three selected C3b ligands, fB forms by far the largest interface, comprising nearly 3,100 Ų of total buried surface area (all interfaces were analyzed using the EBI PISA server (48) and the crystal structures denoted in Fig. 2). Despite this large contact area, the initial C3b/fB interaction occurs via a fast metal-independent binding event involving the Ba fragment of fB (49, 50). Importantly, a monoclonal antibody which targets an epitope on the Ba fragment is a potent inhibitor of the AP (51) and mutagenesis data suggests that C3b/Ba complex formation is driven by three relatively small interaction sites comprised of roughly 523, 111, and 181 Ų, respectively (35, 52). To improve the chance of identifying small molecules capable of impeding the C3b/fB interaction, we targeted a region on C3b which corresponds to a C3b/Ba interaction site formed by the CCP3 domain and centered on the C3b residues His-870, Asn-871, Pro-872, Ala-873, Lys-906, Gly-908, Leu-909, Gln-910, Glu-911, Glu-913, Arg-928, Ser-930 (UNIPROT#: P01024) (Fig. 2A, middle and lower panels). We performed a parallel docking screen which was guided by the entirety of the C3c/compstatin interface (31). Compstatin is a synthetic peptidic inhibitor of complement that prevents cleavage of C3 by C3 convertases. It binds C3/C3b/C3c at a 563 Ų site between the MG4 and MG5 domains of the β-chain (23, 53) (Fig 2B, middle and lower panels).

The S. aureus SCINs are relatively small (~10 kDa) secreted proteins characterized by a three-helix bundle fold which bind directly to C3b and stabilize an inactive form of the AP C3 convertase (38, 50, 54–56). SCIN activity is multi-faceted and in addition to disrupting downstream complement-dependent bacterial opsonization, SCINs also function to mask the recognition of C3b by phagocytic receptors (45, 57). The SCIN protein family includes SCIN-B, which forms a 610 Ų binding site on the C3b α-chain (38) (Fig. 2C, middle). Recently we have reported that certain SCIN peptidomimetics possess complement inhibitory activity in vitro (25), indicating that lower molecular weight inhibitors which bind the SCIN site on C3b have potential to interfere with complement activation. The close proximity of the selected fB binding box to the SCIN/C3b site presented a reasonably high potential for redundant hit identification. To ensure efficient use of computational resources, we thus chose to perform a parallel ligand-based cheminformatics screen to search for compounds with SCIN-like activity. This approach was guided by our previous demonstration that two residues on the second α-helix of SCIN-B (Arg-44 and Tyr-51) are critical for high-affinity C3b-binding (25, 38) (Fig. 2C, middle). ChemVassa (39) was used to identify small molecules with similar chemical and structural features to this key region of SCIN-B. The top ChemVassa hits were then subsequently docked onto C3b and theoretical binding energies were used as a secondary scoring metric (Fig. 2C, bottom).

The cheminformatics screening approach described above, which was designed to target relatively small, functionally critical surface areas on C3b, yielded 45 high priority compounds across the three screens. Compounds were organized into four libraries: fb-1 (10 compounds); fb-2 (21 compounds); scn (SCIN site: 2 compounds); and cmp (compstatin site: 13 compounds) (Table S1). Compounds originating from the fB screen were separated into two sub-libraries based on structural similarities discussed below. Each of the 45 compounds was obtained in low milligram quantities via commercial synthesis services for use in in vitro validation.

3.3 Hierarchical clustering analysis of all hit compounds reveals groups of common substructures

To determine if underlying structural relationships might be present amongst the compound libraries, we used hierarchical clustering analysis as implemented in the ChemMine software suite (40) (Fig. S1). This analysis revealed common substructures present within the respective libraries. For example, of the ten compounds that make up the fb-1 library, seven are structurally related by a central isothiazole moiety (Fig. S1, box A). A broader similarity based on different substructures can be detected within the fb-2 library (Fig. S1, box B) and within the cmp library (Fig. S1, box C). In contrast, both compounds from the scn library as well as several compounds from fb-1 and fb-2 are structurally dissimilar to all other library compounds. Thus, while a set of common moieties are represented within each individual library, the 45 compounds across the four libraries are characterized by appreciable structural diversity.

3.4 Measurements of direct C3b-binding by SPR for highly scored cheminformatics compounds

We adopted an SPR approach to determine C3b binding affinities for each compound. Compounds were initially injected at a single concentration (250 μM) over a sensorchip previously immobilized with human C3b. The molecular weight and solvent corrected responses observed for each compound are presented in Fig. 3A. An experimental maximal signal (R_{max =} 13 ± 0.27 RU/100 Da) was obtained by injecting a saturating concentration of the compstatin derivative Cp40 (100 μM) which binds C3b with high affinity (K_D = 0.5 nM) (58). In total, 14 compounds exhibited C3b-binding responses ≥ 45% of the mean control signal and a majority of these originated from the fb-2 library (11 of 14 compounds). Among these the strongest binding compounds were fb-2-5 and fb-2-10, whose median binding responses were nearly 90% of the control.

Figure 3. SPR measurements of direct C3b-binding.

SPR was used to measure direct interaction of small molecules with C3b. (A) Each compound was injected at 250 μM over a C3b biosensor in duplicate. Binding response was obtained as detailed in the Materials and Methods section and is expressed as RU/100 Da. Cp40 (100 μM) was injected to derive an experimental maximal signal for this surface (13 ± 0.27 RU/100 Da). Compounds fb-2-2, fb-2-13, and fb-2-19 exhibited abnormal sensorgram shape and were removed from further analysis. These compounds are denoted with an asterisk (*). (B) Representative sensorgrams and fit using free R_max from a variable concentration injection series of cmp-5 over C3b. A dissociation constant (K_D) was obtained by steady-state analysis using Biacore T200 Evaluation Software and K_D is reported as the mean ± S.D. from seven independent injection series performed across five separate C3b biosensors. LE = ligand efficiency computed as ΔG/N_{heavy atoms} where ΔG = −RT x ln (K_D,SPR).

Of the 45 compounds tested at a single concentration, 15 were further tested for their ability to bind C3b in a dose-dependent manner (fb-1: -6; fb-2: -4 -5, -6, -8, -9, -10, -11, -14, -16, -18, -20, and cmp: -3, -5, -11). We evaluated these data for the presence of superstoichiometric binding by freely fitting the maximal binding response (R_max,fit) and comparing this value to the experimentally-derived maximal binding response (R_max,Cp40). Of the compounds tested, 8 showed evidence of superstoichiometric binding behavior (i.e. R_max,fit ≫ R_max,Cp40), which is a strong indication of secondary and/or nonspecific binding under these assay conditions. While detection of nonspecific binding does not strictly preclude C3b-specific binding by these compounds, it greatly restricts the ability to obtain reliable C3b affinity information using an SPR-based direct binding approach. This situation included both fb-2-5 and fb-2-10, which gave the initial appearance of strong specific binding when only a single concentration was considered. However, seven compounds fit well to 1:1 equilibrium binding models. Steady-state dissociation constants (K_D) were calculated for cmp-5 (K_D = 450 μM) (Fig. 3B), which also had similar affinity for C3 and C3c (Fig. S2B, C). In addition to cmp-5, C3b affinities were obtained for compounds cmp-3 and fb-2: -4, -11, -14, -16, and -18 (ranging between K_D = 340 and 820 μM) (Fig. S2A).

Ligand efficiency (LE) represents the free energy of binding per non-hydrogen atom and is a useful metric for lead prioritization. LE is defined as ΔG/N_{non-hydrogen atoms} where ΔG =-RT x ln K_D. All seven compounds exhibited calculated LE’s within a narrow range (0.16 – 0.20), suggesting that they have similar potential for medicinal chemistry-driven lead optimization.

3.5 The effect of fB-2 compounds on the fB/C3b interaction

The experiments described above confirmed that several small molecules identified by our in silico screens are capable of binding to C3b. However, the docking results suggest that these C3b-binding compounds overlap the binding site of the natural C3b ligands used to guide the computational screening. In order to test this prediction directly, we designed two orthogonal competition-based assay systems to evaluate the effect of the strongest binding compound library (fB-2) on the fB/C3b interaction. The lowest energy docking pose for each fB-2 library compound overlaps the fB/C3b molecular interface near the Ba fragment binding site (Fig. 4A). To determine if fb-2 compounds interfere with Ba/C3b binding, we adapted a luminescent, proximity-based, microbead competition assay, which has been used in our lab previously for several other C3b ligands (25, 38, 45). For this purpose we expressed and purified a recombinant protein corresponding to the three N-terminal CCP domains of the fB Ba fragment (termed Ba(1–3) here). Optimization screens demonstrated that a fixed concentration of 250 nM recombinant HIS-tagged protein and 5 nM site-specifically biotinylated C3b when mixed with 20 μg ml⁻¹ nickel chelating AlphaScreen Acceptor and streptavidin Donor beads yielded a strong Alpha signal (1.8 x 10⁶ counts per second). To validate a competition assay format we used these conditions in the presence of a two-fold dilution series of either full-length fB (K_D = 340 nM) or untagged Ba(1–3) (K_D = 4,700 nM) (Fig. 4B). Next we assayed the effect of each fB-2 compound at 250 μM. Under these conditions, six compounds significantly reduced the Ba(1–3)/C3b-specific Alpha response (fB-2-4, 2-9, 2-10, 2-18, 2-20, 2-21) (Fig. 4C).

Figure 4. The effect of fB-2 compounds on the fB/C3b interaction.

(A) The docking positions of the fB-2 compound library are shown in the context of the fB/C3b crystal structure (PDB: 2XWJ). C3b is shown in surface representation (grey), while a cartoon rendering of fB is shown with the Ba fragment colored cyan and the Bb fragment colored green. Docking of the fB-2 library predicts that each compound binds to a deep cleft overlapping the fB Ba fragment binding site near the CCP-2/CCP-3 interface. (B) To evaluate the ability of fB-2 compounds to interfere with the Ba/C3b interaction, we developed an AlphaScreen luminescent microbead competition assay which uses a recombinant form of Ba encompassing CCP domains 1–3 in conjunction with site-specifically biotinylated C3b. To validate the competitive assay format a variable concentration series of full-length fB (K_D = 340 nM, dashed line) or untagged Ba(1–3) (K_D = 4,700 nM, solid line) was used. Control wells lacking competitor have been treated as 100% response. (C) The ability of each fB-2 compound (250 μM) to interfere with the Ba/C3b interaction was assessed in duplicate competition experiments. Compounds 2–4, 2–9, 2–10, 2–18, 2–20, and 2–21 exhibited significant competition at this concentration relative to wells containing DMSO vehicle control (^*p ≤ 0.05, ^**p ≤ 0.01) (D–G) To determine the effect of fB-2 library compounds on the full-length fB/C3b interaction, we developed an LE-SPR-based competition assay which utilizes site-specifically biotinylated C3b to generate a biosensor where C3b is oriented in a homogenous and physiologically relevant orientation. (D) Native fB using conventional SPR or (E) dye-labeled B23-fB using LE-SPR were first evaluated in a variable concentration injection series. Each experiment was performed in duplicate and K_D’s were derived using 1:1 Langmuir kinetic models of binding (fits overlaid as red lines). (F) A representative LE-SPR competition experiment where fixed 500 nM B23-fB was co-injected with selected fB-2 compounds (500 μM) over the C3b biosensor. Only compound fB-20 showed weak but statistically significant competition (^***p ≤ 0.001). Each competition experiment was performed in duplicate and the residual binding signal of each injection is reported in (G) as the percent (%) signal relative to DMSO vehicle control. Additional details for all experiments shown here are provided in the Materials and Methods section.

These data provide a measure of validation for the predicted C3b docking sites and serve as further evidence that several small molecules identified in our in silico screen are capable of binding C3b. However, we reasoned that small molecule complement inhibitory activity would only manifest if the interaction of full-length fB with C3b was disrupted. Efforts to develop an analogous AlphaScreen assay for the full-length fB/C3b interaction were unsuccessful. Thus, to address this question, we developed a Label-Enhanced SPR (LE-SPR)-based competition assay. LE-SPR allows for real-time interaction monitoring of analytes which have previously been labeled with a dye that absorbs strongly and has an anomalously high refractive index at the SPR measurement wavelength (43, 44). In the case of a small molecule competition assay, LE-SPR holds two significant advantages over conventional SPR. First, the LE-SPR signal is virtually independent of refractive index changes and thus the large bulk signal changes associated with small differences in DMSO concentrations between samples are eliminated (43, 44). Second, the presence of non-dye-labeled interacting analytes (such as our C3b-binding small molecules) do not contribute to the LE-SPR signal. These features allow for highly specific measurements of only the dye-labeled analyte and greatly simplified interpretation of co-injections involving a dye-labeled species in the presence of non-dye-labeled C3b-binding analytes/putative competitors.

To implement this approach, we first generated C3b biosensors using site-specifically biotinylated C3b whereby C3b is immobilized on the SPR chip surface in a uniformly physiological orientation (25, 38, 42, 45). Using conventional SPR, we measured the interaction of native fB with C3b (K_D = 480 ± 20 nM) (Fig. 4D). Next, fB was labeled using NHS-B23 dye reagent (Episentum) and LE-SPR experiments were performed to measure the direct binding of dye-labeled fB to C3b (K_D = 1,800 ± 810 nM) (Fig. 4E). Compounds were then selected from the fB-2 library that bound C3b in a dose-dependent manner (Fig. S2A), competed with the Ba/C3b interaction (Fig. 4C), or exhibited both activities, and co-injected at 500 μM concentration with a fixed concentration of B23-fB (500 nM) (Fig. 4F, G). Compared to DMSO vehicle control, only fB-2-20 exhibited a small but significant reduction in the B23-fB specific LE-SPR signal (Fig. 4G). In aggregate, these data indicate that we identified several C3b-binding small molecules which interact with C3b near the Ba binding site but that these compounds do not appear to substantially interfere with proconvertase formation.

3.6 Evaluation of complement inhibitory properties for all compound libraries

The C3b ligands involved in the originating screens all play a role in the AP and we reasoned that any effects by our compounds should be apparent when monitoring AP-mediated complement activation. Thus, we first evaluated each compound at 250 μM concentration in an ELISA-based AP-specific complement assay where terminal activation of complement was detected by MAC deposition. All compounds from the SCIN, fb-1, fb-2, and compstatin libraries failed to inhibit the AP at the concentration used, including those compounds which bound C3b in a dose-dependent manner, with a lone exception. Compound cmp-5 exhibited small but significant and reproducible inhibition of MAC deposition in this assay format (73 ± 4.6% relative to control response; n=5) (Fig. 5A).

Figure 5. Evaluation of complement inhibitory activities of all compounds.

(A) The ability of each compound (250 μM) to inhibit complement under conditions selective for the AP was evaluated in an ELISA where an anti-MAC antibody was used for detection. Cp40 (10 μM) was used as a positive control. Data is normalized to signal produced by a 2.5% (v/v) DMSO vehicle control. Data are represented as the mean ± S.D. of three independent experiments. Only cmp-5 exhibited significant inhibition at this concentration. (B) The compstatin library was evaluated using a CP-specific ELISA where MAC detection was used. Cp40 (10 μM) was used as a positive control. Only cmp-5 exhibited significant inhibition at this concentration (^****p ≤ 0.0001).

The inability of compounds which originated from the fB or SCIN screens to block complement in this assay is in agreement with the data obtained in the SPR and LE-SPR assays presented above. On the other hand, the finding that dose-dependent C3b-binding by compound cmp-5 could reduce AP-mediated MAC formation was intriguing. Therefore we refocused our efforts towards understanding the nature of its activity. In contrast to SCIN and fB, whose activity is restricted to the context of the AP, compstatin acts at the level of native C3 and interferes with C3 cleavage by not only AP C3 convertases (i.e. C3bBb) but also by CP C3 convertases (i.e. C4b2a). We performed additional screening on all the compstatin library compounds under conditions selective for CP activation. In accordance with the observations made for the AP experiments, we found that only cmp-5 was capable of inhibiting MAC deposition at 250 μM (59 ± 2.4% relative to control response; n=13) (Fig. 5B).

3.7 Evaluation of cmp-5 analogues as complement inhibitors

Compound cmp-5 (systematic name: N-(3-(5-methyl-1H-pyrazol-1-yl)phenyl)-2-(3-oxo-1,3-dihydroisobenzofuran-1-yl)acetamide) is characterized by a central aniline group linked to a left-hand side methyl pyrazol moiety and a right-hand side isobenzofuran ring (Fig. 6A). To identify structural analogues to cmp-5 that may also act as complement inhibitors, we queried the 22.7 million compound “purchasable” subset of the ZINC database for compounds with similar structure to cmp-5 and identified 2,541 unique compounds with at least 50% chemical structure similarity. Each of these compounds was subsequently docked onto the C3b/compstatin binding site by employing the AutoDock Vina methodology described above. Using cmp-5’s Vina binding energy (−8.0 kcal mol⁻¹) as a benchmark, we found 17 new compounds that had equal or improved predicted C3b-binding energies ranging from −8.0 to −8.6 kcal mol⁻¹. Of these compounds, 11 were obtained commercially for in vitro analysis (Fig. 6A). In total, six compounds from this cohort inhibited the CP ELISA when used at 250 μM (cmp5-2, cmp5-5, cmp5-6, cmp5-8, cmp5-14, and cmp5-15) (Fig. 6B).

Figure 6. Structure-based similarity search yields inhibitory cmp-5 analogues.

(A) Compound cmp-5 is characterized by a central aniline group linked to a pyrazol moiety and an isobenzofuran ring. The purchasable subset of the ZINC database was queried for compounds that possessed ≥ 50% structural similarity to cmp-5, which yielded a total of 2,514 compounds. This compound set was docked onto C3b using the coordinates for the compstatin binding box, scored, and ranked as outlined in Figures 1 & 2. This resulted in a total of 17 compounds with equal or more favorable predicted C3b-binding energies, of which 11 were obtained commercially. Structural similarity measurements relative to cmp-5 were obtained by calculating Tanimoto coefficients (italicized) (73, 74). (B) A CP ELISA assay was used to assess the complement inhibitory activity of each cmp-5 analogue. Several cmp-5 analogues significantly inhibit MAC formation (cmp5-2, -5, -6, -8, -14, and -15; ^****p ≤ 0.0001).

3.8 Compound cmp-5 and cmp-5 analogues compete with compstatin for C3b binding

An analysis of the lowest energy binding pose from the cmp-5 docking study indicates that cmp-5 would be expected to interfere with compstatin/C3b binding (Fig. 7A). To test this directly, we developed an appropriate LE-SPR-based competition assay. As a practical source of a compstatin reagent which could be readily labeled with NHS-B23 dye, we designed a compstatin fusion protein which consisted of E. coli thioredoxin fused to a linker and followed by the sequence of an early affinity-optimized compstatin sequence termed 4W9A (amino acid sequence: ICVWQDWGAHRCT). First, we characterized the C3b-binding affinity of the native TRX-4W9A construct by conventional SPR (K_D = 770 nM) (Fig. S3A) as well as that of B23-labeled TRX-4W9A by LE-SPR (K_D = 560 nM) (Fig. S3B). To validate the competition assay format we co-injected a fixed concentration of B23-TRX-4W9A (500 nM) with a variable concentration series of non-dye-labeled or “cold” TRX-4W9A. Fitting of these data to a kinetic competition model resulted in a calculated K_D = 590 nM for the native species and 790 nM for the dye-labeled species (Fig. S3C). Together these data indicate that native and dye-labeled TRX-4W9A have similar affinities for C3b and that LE-SPR can be used to monitor the specific competition of the dye labeled species by an unlabeled C3b-binding analyte.

Figure 7. Cmp-5 analogues interfere with C3b/compstatin binding.

(A) The lowest energy docking pose for cmp-5/C3b was superimposed onto the C3c/compstatin crystal structure (PDB: 2QKI). Cmp-5 (red) overlaps a significant region of the compstatin (cyan) binding site on C3b (grey surface) (note, C3c has been hidden for clarity). (B, C) To assess whether cmp-5 or cmp-5 analogues can compete with compstatin for C3b binding we performed LE-SPR-based competition experiments (see Fig. S3A-C for assay validation). A fixed concentration of B23-TRX-4W9A (500 nM) was co-injected with 500 μM of each compound or with vehicle control (5% (v/v) DMSO). As a positive control for competition, 65 μM of compstatin peptide (Tocris) was also co-injected. The presence of compounds cmp-5, -2, -3, -7, -14, and the compstatin peptide all result in the reduction of specific C3b-binding by B23-TRX-4W9A. Each compound was tested on a minimum of two independent C3b surfaces (^***p ≤ 0.001, ^****p ≤ 0.0001). Panel (B) depicts representative epigrams while panel (C) reports the residual binding of B23-TRX-4W9A in the presence of each competitor relative to DMSO controls.

Next we performed co-injections using cmp-5 and all cmp-5 analogues at 500 μM as competitors and measured the residual binding signal of the dye labeled TRX-4W9A species by LE-SPR (Fig. 7B, C). As an additional positive control for competition, we used the original version of the compstatin peptide obtained commercially from Tocris. While cmp-5 showed a modest ability to compete with TRX-4W9A, cmp5-2, -3, -7, and -14 all caused a strong reduction in the B23-TRX-4W9A-specific LE-SPR signal (Fig. 7C). These data strongly support the identity of the C3b-binding site predicted by the cmp-5/cmp-5 analogue docking experiments.

3.9 Compound cmp-5 prevents cleavage of complement component C5

In order to address questions related to the mechanism of cmp-5 inhibition we synthesized cmp-5 de novo to yield a large quantity of purified compound (~100 mg). We first examined compound solubility and found that cmp-5 was maximally soluble at concentrations near ~600 μM in the CP ELISA assay buffer (Fig. S2D, E). With this constraint in mind, we sought to profile the dose-dependency of cmp-5 in the CP ELISA assay by varying the concentration of cmp-5 (0 – 625μM) under a fixed serum concentration (1% (v/v)). Indeed, cmp-5 inhibits MAC deposition in this assay format in a dose-dependent manner, and although saturable inhibition could not be reached, a half-maximal inhibitory concentration (IC₅₀) of 530 μM could be calculated (Fig. 8A). Importantly, this value is in agreement with the measured affinity of cmp-5 for C3b (K_D = 450 μM, Fig. 3B).

Figure 8. Compound cmp-5 inhibits MAC deposition in a dose-dependent manner and prevents cleavage of C5.

(A) Dose-dependent inhibition of complement by cmp-5 was assessed by a CP ELISA assay where the deposition of C3b, C4b, and MAC in 1% (v/v) serum were monitored. While cmp-5 had no effect on C3b or C4b deposition at cmp-5 concentrations up to 625 μM, an IC₅₀ of 530 μM for cmp-5 was measured when MAC deposition was detected. Detection of (B) C3b or (C) C4b deposition using a CP ELISA in the presence of 250 μM cmp-5 analogues. (D) MAC deposition was measured in the presence of a fixed concentration of cmp-5 (500 μM) or 2.5% (v/v) DMSO vehicle control under varying serum concentrations (0–4% (v/v)). An EC₅₀ of 0.88% (v/v) serum (DMSO control) vs. 1.43% (v/v) serum (cmp-5) was calculated (MAC deposition). (E) A C5a capture ELISA was used to determine the relative amount of C5a produced in the corresponding CP ELISA reactions presented in panel B. An EC₅₀ of 1.55% (v/v) serum (DMSO control) vs. 2.04% (v/v) serum (cmp-5) was observed, indicating a correlation of lower MAC deposition with less C5 cleavage. Data are represented as the mean ± S.D. of three or four independent experiments. All fits were obtained by 4-parameter variable slope nonlinear regression analysis using GraphPad Prism v5.04. (F) An LE-SPR-based competition assay was used to assess whether compstatin or cmp-5 compounds can interfere with C5/C3b binding (see Fig. S3D–F for corresponding assay validation experiments and representative epigrams). A fixed concentration of B23-C5 (250 nM) was co-injected with either unlabeled C5 (250 nM) (i.e. “cold” C5), TRX-4W9A (12 μM), or compstatin peptide (65 μM) in a running buffer of HBS-T. Each of these analytes reduces the B23-C5 specific LE-SPR signal. The presence of 500 μM cmp-5, cmp5-2, and cmp5-14 did not affect the B23-C5-associated LE-SPR signal relative to DMSO vehicle control. Each competition experiment was performed between two and four times (^*p ≤ 0.05, ^***p ≤ 0.001, ^****p ≤ 0.0001). Injections of 250 nM B23-C5 were treated as the control signal for “cold” C5, TRX-4W9A, and compstatin. In contrast, each of the small molecule compounds has been normalized to injections of 250 nM B23-C5 in the presence of vehicle only control (5% (v/v) DMSO).

Surprisingly, we found that although cmp-5 prevented MAC deposition, it had no effect on C3b or C4b deposition (Fig. 8A). Furthermore, all cmp-5 analogues, including those that blocked MAC deposition (Fig. 6B), fail to block C3b or C4b deposition at the same concentration (Fig. 8B, C). This result is in contrast to what one would expect if cmp-5 was inhibiting complement via a compstatin-like mechanism. While compstatin binds C3b/C3c, its functional target is native C3 and it potently inhibits the deposition of C3b (and therefore MAC) but not C4b (Fig. 8B, C). To further explore this unexpected outcome, we repeated the CP ELISA using a fixed 500 μM cmp-5 concentration while varying serum concentration (0–4% (v/v)). Again we found that cmp-5 significantly inhibited MAC deposition in a serum-dependent manner relative to vehicle control, as judged by the half-maximal effective concentration of serum (EC_{50, DMSO} = 0.88% (v/v) vs. EC_{50, cmp-5} = 1.43% (v/v)) (Fig. 8D). We then probed these reactions for C5a using a capture ELISA which employs a monoclonal antibody that recognizes a C5 activation-dependent neoepitope on C5a, but not C5 (46). Relative to vehicle control, cmp-5 significantly inhibited the production of C5a (EC_{50, DMSO} = 1.55% (v/v) vs. EC_{50, cmp-5} = 2.04% (v/v)) (Fig. 8E). Together, these data strongly suggest that cmp-5 exerts its inhibitory effects by preventing cleavage of C5 into C5a and C5b, rather than by directly inhibiting formation of MAC per se.

Despite the overlapping binding site suggested by the C3b docking and compstatin/C3b competition assays (Fig. 7), there is an apparent divergence in complement inhibitory mechanism between cmp-5 and compstatin. While compstatin prevents C3 cleavage, cmp-5 inhibits the downstream conversion of C5. Upon activation of the complement cascade, C5 is cleaved by trimolecular enzymes called C5 convertases (AP: C3bBb3b or CP/LP: C4b2a3b) that form on surfaces with high surface densities of C3b. Because C3 to C3b activation is strongly inhibited in the presence of compstatin, C5 convertase formation is indirectly prevented. However, we wondered if compstatin, or molecules like cmp-5 which bind C3b near the compstatin binding site, may exert an effect on C5 recognition of C3b-dependent C5 convertases. To test this we developed an LE-SPR competition assay based on the interaction of C5 with physiologically oriented C3b sensorchips. First, C3b affinity measurements using conventional SPR and native C5 (K_D = 350 nM) or dye-labeled B23-C5 LE-SPR (K_D = 3,200 nM) were performed (Fig. S3D, E). Next, a fixed concentration of B23-C5 (250 nM) was co-injected with either 250 nM unlabeled C5, 12 μM TRX-4W9A, or 65 μM compstatin peptide. “Cold” C5 and TRX-4W9A reduced the specific binding of B23-C5 nearly 50% while a smaller competition effect was observed for the compstatin peptide (Fig. 8F and S3F). Although the B23-C5/C3b interaction was sensitive to vehicle control (5% (v/v) DMSO), a strong LE-SPR signal could nonetheless be measured for 250 nM B23-C5. On the basis of their ability to both compete with compstatin for C3b binding (Fig. 7B, C) and also inhibit MAC formation (Fig. 6B) compounds cmp-5, cmp5-2, and cmp5-14 were selected as putative C5/C3b competitors. However, these compounds failed to perturb the B23-C5/C3b interaction when used at 500 μM. Taken together, these data suggest that under these assay conditions that larger and/or higher affinity compounds which target C3b at the compstatin binding site (i.e. TRX-4W9A or compstatin peptide) interfere with C5/C3b binding, but that in the strictest sense cmp-5 does not.

Discussion

The field of complement-directed therapeutics has seen significant advancement over the past decade. Beyond the landmark FDA approval of eculizumab for the treatment of PNH and aHUS (10, 11), preclinical studies and clinical trials are now ongoing for dozens of new complement-targeted drugs (12). The breadth of complement-related pathologies which can be successfully targeted by pharmacological intervention is thus primed to include not only rare or orphaned diseases but also more common inflammatory diseases such as cardiovascular disease and arthritis (12). The momentum driving the development of novel complement-directed drugs will likely only be strengthened by recent discovery of causal links for complement in devastating neurodegenerative conditions, such as Alzheimer’s disease (59) and schizophrenia (60).

Despite an era of an increasingly diverse anti-complement “toolbox”, there remain relatively few examples of small molecule complement inhibitors (12). The idea that large protein-protein interactions, like those that form the basis of the complement system, are recalcitrant to targeting by rule-of-five compliant compounds continues to be pervasive. However, examples of both orthosteric and allosteric small molecule modulators of protein-protein interactions are prevalent. For instance, by 2014, twelve distinct small molecule drugs which inhibit protein-protein interactions of various oncogenic targets were in clinical trials or had already reached market (28).

Small molecule drug discovery provides opportunities for employing upstream computational screening methodologies. In addition to cost savings over traditional high-throughput biochemical screening approaches, in silico screening provides means to more comprehensively and selectively search the vast chemical space estimated to exist for drug-sized compounds (≥10⁶³ unique molecules) (27). An important prerequisite to this approach is the availability of high resolution structural models of the drug target. In this regard, the field of complement structural biology has been revolutionized in the past two decades by the solution of a considerable number of important crystal structures. This progress is exemplified by complement component C3, where in addition to the structure of native C3 (61), crystal structures of the C3 fragments C3b (53), C3c (61), and C3d (62) have all been reported. Furthermore, an impressive number of C3 co-crystal structures have been published, including C3b/fB & C3b/fB/fD (35), C3b/factor H(domains 1–4) (63), C3d/factor H (domains 19–20)/sialic acid (64), C3b/Bb/SCIN-A (50), C3c/SCIN-B (38), C3d/Efb-C (65), C3d/CR2 (66), C3c/CRIg & C3b/CRIg (67), and C3c/compstatin (31), to name only a few. Indeed, perhaps no other complement target is better positioned than C3 for the use of cheminformatics-based drug discovery.

In this study, we have combined virtual screening methodologies with structural knowledge of specific C3b ligands to enrich our in silico hits with small molecules which not only bind C3b but also have anti-complement activity. Of the 45 compounds which were selected for in vitro validation we confirmed seven bound C3b in a dose-dependent manner consistent with 1:1 stoichiometry. This relatively high apparent in vitro hit rate (~16%) exceeds even most fragment based drug discovery screens, and identifies C3b as a tractable target for small molecule ligands. This latter observation is supported by a recent published study where C3d-binding small molecules were developed as potential noninvasive in vivo diagnostic tools for complement-related conditions (68). However, similar to the compounds reported in that study, a majority of the confirmed C3b-binders found here failed to inhibit complement. This underscores the likely importance of pinpointing small molecule ligands to susceptible and functionally critical surfaces of C3b in order to achieve complement inhibitory activity. Cheminformatics may serve as an important means to this end, as we have demonstrated here that a number of these C3b-binding small molecules disrupt the interaction of the original C3b ligand which guided their discovery.

Ultimately, the in silico targeting of the compstatin binding site yielded a group of structurally related compounds which were capable of inhibiting C5 activation, and thereby blocking downstream MAC formation. High-affinity compstatin analogues, such as Cp40, potently inhibit complement by preventing C3 convertase-mediated cleavage of C3 regardless of the initiation pathway (23). Our rationale for targeting the compstatin binding site on C3b was the potential to identify small molecules with compstatin-like inhibitory activity. It was therefore initially surprising that cmp-5 and cmp-5 analogues had no apparent effect on the activation of C3 in our assays despite blocking MAC deposition (Fig. 8). This observation suggested that cmp-5 was acting downstream of C3 activation. We thus considered the possibility that cmp-5 was exerting its inhibitory effect at the level of C5 convertases, where C3b serves as an integral non-catalytic component. Indeed, we observed concomitant inhibition of C5a production and MAC deposition when cmp-5 was present (Fig. 8D, E). These data demonstrate that cmp-5 prevents cleavage of C5 rather than acting by another mechanism, such as the fluid phase consumption of complement seen for inhibitors like cobra venom factor (CVF) (69).

When our data are considered as a whole, they suggest that unlike compstatin, cmp-5 directly inhibits the activity of the C3b-containing C5 convertases thereby preventing cleavage of C5 into C5a and C5b. Models based on the C5/CVF crystal structure propose a two site point of contact for the non-catalytic subunits of C5 convertases (i.e. C3b or C4b) (70). One of these sites overlaps closely with the position of compstatin and cmp-5 on C3b, as judged by the co-crystal structure or the in silico docking studies, respectively (Fig. 7A). Consistent with this idea we found that compstatin interferes with C5/C3b binding, although cmp-5 did not have an effect in this assay (Fig. 8F). The underlying mechanisms which result in switching of the molecular arrangement and substrate specificities of C3 convertases to C5 convertases remain poorly understood. However, recent advances have been made in this area which have made use of high-affinity C5 targeted complement inhibitors (71, 72), and these new systems may better enable future mechanistic studies of cmp-5-based compounds. Thus, while important questions about the mechanism of cmp-5 inhibition remain to be addressed by future studies, it is feasible that cmp-5 could disrupt C5 convertase activity by binding this functionally critical region of C3b.

Blockade of complement at the level of C5 has taken center stage in the area of complement-directed therapeutics due to the continued success of the anti-C5 monoclonal antibody eculizumab. Three additional anti-C5 monoclonals, as well as an anti-C5a antibody had reached clinical trials by late 2015 (12). Non-antibody drugs including the C5-binding protein derived from a tick salivary protein (Coversin), a C5 blocking peptide, small molecule C5a receptor antagonists, and nucleic acid-based C5 targeting therapeutics have all reached clinical development stages (12). In contrast to targeting the substrate (i.e. C5), cmp-5 appears to inhibit C5 cleavage by interfering with the C5 convertase directly. Despite this distinction, there are several important limitations to the current set of cmp-5 and related compounds that preclude a complete mechanistic description of their inhibitory mode. For instance, although the effect of cmp-5 on C5 cleavage is clear, we cannot rule out that higher affinity and/or saturable versions of cmp-5 may also exhibit effects on C3 cleavage/deposition as occurs with compstatin. An interesting and speculative corollary to this emerges from our data, which suggests that compstatin derivatives may be potentially tuned to act at the level of C5 convertases. Among our future directions are the improvement of cmp-5’s potency (i.e. C3b-binding affinity and IC₅₀) and/or the physicochemical properties of the current compounds (i.e. aqueous solubility) through traditional medicinal chemistry approaches. Furthermore, cmp-5 and cmp-5 analogues contain chiral centers and docking of the (R)-isomer results in a more favorable binding energy relative the (S)-isomer (−8.0 vs. −7.3 kcal mol⁻¹). However, all compounds tested here were synthesized as purified racemic mixtures and it will be important in the future to experimentally test the cmp-5 (R)-/(S)-isomers independently in order to empirically determine what role stereochemistry may play in driving cmp-5/C3b affinity. True structure activity relationship studies will also be greatly bolstered by the confirmation of cmp-5’s (or cmp-5 analogues’) C3b-binding mode through x-ray crystallography studies which are ongoing in our lab. Nonetheless, to the best of our knowledge, cmp-5 and the cmp-5 analogues reported here are the first C3b-binding inhibitory small molecules to be reported and the first example of a small molecule inhibitor of C5 cleavage.

While a deluge of structural information and the central position of C3 have made it an ideal candidate for cheminformatics-based drug discovery, it is by no means the lone potential target within the complement cascade. Indeed, similar strategies to the one employed here may well be readily adapted to other complement targets such as C4 or C5, for which an abundance of structural and biochemical data are now available. Thus, in addition to discovering a promising new class of small molecule complement inhibitors, this study serves as an important proof of principal which has high potential impact on the future of novel complement-directed drug discovery.