Inhibition of the C1s Protease and the Classical Complement Pathway by 6-(4-phenylpiperazin-1-yl)pyridine-3-carboximidamide and Chemical Analogs

Abstract

The classical pathway (CP) is a potent mechanism for initiating complement activity and is a driver of pathology in many complement-mediated diseases. The CP is initiated via activation of complement component C1, which consists of the pattern recognition molecule C1q bound to a tetrameric assembly of proteases C1r and C1s. Enzymatically active C1s provides the catalytic basis for cleavage of the downstream CP components, C4 and C2, and is therefore an attractive target for therapeutic intervention in CP-driven diseases. Although an anti-C1s monoclonal antibody has been FDA-approved, identifying small-molecule C1s inhibitors remains a priority. Here we describe 6-(4-phenylpiperazin-1-yl)pyridine-3-carboximidamide (A1) as a selective, competitive inhibitor of C1s. A1 was identified through a virtual screen for small molecules that interact with the C1s substrate recognition site. Subsequent functional studies revealed that A1 dose-dependently inhibits CP activation by heparin-induced immune complexes, CP-driven lysis of antibody-sensitized sheep erythrocytes, CP activation in a pathway-specific ELISA, and cleavage of C2 by C1s. Biochemical experiments demonstrated that A1 binds directly to C1s with a K_d of ~9.8 μM and competitively inhibits its activity with a K_i of ~5.8 μM. A 1.8 Å resolution crystal structure revealed the physical basis for C1s inhibition by A1 and provided information on the structure-activity relationship of the A1 scaffold, which was supported by evaluating a panel of A1 analogs. Together, our work identifies A1 as a new class of small-molecule C1s inhibitor and lays the foundation for development of increasingly potent and selective A1 analogs for both research and therapeutic purposes.

INTRODUCTION

The complement system is an ancient proteolytic cascade that serves as a first line of defense against microbial pathogens (1–4). Complement activation proceeds through three canonical pathways, which are each defined by a specific initiating event. Classical pathway (CP) activation is controlled by the first component of complement, C1, and initiates upon C1 recognition of immune complexes. Likewise, the lectin pathway (LP) relies on recognition of the molecular patterns found on foreign carbohydrates by specific lectins, including mannose-binding lectin (MBL). In contrast, the alternative pathway (AP) is spontaneously activated by a ‘tick-over’ mechanism that does not involve pattern recognition, but instead relies on host regulators such as factor H (FH) for controlled activation on pathogen surfaces. Indeed, tight control of key steps in the complement system is afforded by endogenous proteins known as regulators of complement activation (RCAs) (5) .

Loss of finely-tuned control of complement activation can cause or exacerbate a wide range of human autoimmune and inflammatory diseases (4, 6). An improved understanding of the role of complement in certain diseases has stimulated significant interest in the development of specific complement-directed therapies (4, 6). To date, these efforts have facilitated the entry of nearly half a dozen different complement-directed therapeutics into the clinic (4, 6). Despite the encouraging efficacy and safety profiles of these ‘first-in-clinic’ drugs, it is recognized that no ‘magic bullet’ is likely to exist for complement-related diseases because the involvement of complement is often complex and disease specific (4, 6). Furthermore, complement-directed therapeutic development has disproportionately targeted the distal reactions of the pathway involving C5 cleavage. While less explored, targeting of the complement initiating complexes provides an upstream therapeutic strategy that may be desirable in conditions driven by pathway-specific complement activation (4, 6). For example, the CP in autoantibody-mediated complement activation has been implicated in several diseases, including cold agglutin disease (CAD) (7) and heparin-induced thrombocytopenia (HIT) (8). Furthermore, it has been shown that downstream inhibition of C3 or C5 may not always be effective during strong classical pathway activation (9, 10). Thus, development of specific inhibitors of the CP initiating complex (i.e., C1) represents a promising therapeutic approach for these diseases.

C1 is a multiprotein complex consisting of the pattern recognition molecule C1q bound to a heterotetrametric arrangement of two specialized serine proteases known as C1r and C1s (i.e., C1qC1r₂C1s₂) (1, 2, 4). Like many serine proteases found in the bloodstream, C1r and C1s circulate as inactive zymogens and must undergo proteolytic cleavage to manifest their catalytic activities. Binding of the C1 complex via C1q induces a conformational change in C1r that results in autoactivation. Activated C1r then cleaves C1s, converting it from its zymogen into an active enzyme. C1s then proteolytically cleaves C4 and C2 and the complement pathway proceeds through C3 activation and amplification to eventual formation of the lytic membrane attack complex (i.e., C5b-9) (1, 2, 4). Therefore, the C1 complex provides several potential intervention points at the level of pattern recognition (C1q) or proteolytic cleavage (C1r or C1s).

The clinical development of C1 inhibitors has gained momentum in recent years. A major milestone came in 2022 when the anti-C1s monoclonal antibody (Enjaymo/sutimlimab; Sanofi) became the first FDA-approved CP-specific drug for the treatment of patients with CAD. The anti-C1q monoclonal antibody, ANX005 (Annexon Biosciences), is currently in phase 2 clinical trials for Huntington’s disease (HD) and amyotrophic lateral sclerosis (ALS), and phase 3 clinical trials for Guillain-Barré syndrome (GBS). In contrast to these antibody-based drug modalities, the pace of development for small-molecule C1 inhibitors has lagged. While blocking the large protein-protein interactions involved in C1q/immune complex recognition may be difficult to achieve with a small molecule, C1r and C1s are more tractable targets for small molecule drug design. Indeed, proteases are generally regarded as one of the most druggable classes of enzymes (11). While selectivity remains a challenge for protease inhibitors, the successful development of highly selective small-molecule inhibitors of complement proteases factor B (FB) (12) and factor D (FD) (13) has been achieved. In that regard, C1s-targeted small-molecule inhibitors have been previously described (14–17), while a recent article reported the identification of orally-available and brain-penetrable C1s inhibitor (18). However, there remains no FDA-approved small-molecule inhibitor of C1s, nor has any such molecule been adapted for use as a tool in complement research.

We previously developed a cheminformatics approach for the discovery of small-molecule complement inhibitors targeting the AP (19). The underlying premise of that study was that existing structural information could be harnessed for in silico identification of small-molecule ligands that not only bind to complement proteins but also block their function. Although our initial work provided proof-of-concept that this strategy could be deployed against the central component of the complement alternative pathway, C3b, we hypothesized that the proteolytic machinery within C1 is also a tractable target for structure-guided drug discovery. In this study, we adapted this overall approach to utilize the structure-based machine-learning drug discovery platform, AtomNet (20), to identify C1s-binding small-molecules. Following in silico lead identification, we used functional, biochemical, structural, and medicinal chemistry approaches to characterize a selective small-molecule inhibitor of C1s. This compound represents a possible new lead for the future development of CP blocking drugs and research tools.

MATERIALS AND METHODS

Proteins –

Samples of endogenously expressed C1r and C1s that were purified from human serum were purchased from Complement Technologies. Samples of activated thrombin and Factor Xa that were purified from serum were obtained from Enzyme Research Laboratories. Samples of neutrophil elastase that were purified from human sputum were obtained from Elastin Products Corp. Samples of cathepsin-G that were purified from human neutrophils were obtained from Innovative Research.

Plasmids for prokaryotic expression of the 2SP region from human C1s, murine C1s, and human C1r were obtained from GenScript, Inc. Gene fragments encoding the CCP2 and SP regions of various proteases were synthesized and subcloned in the expression vector pT7HMT (21). An identical procedure was used to prepare each protease, with the exception that the corresponding expression vector was used. Briefly, the expression plasmids were transformed into E. coli strain BL21(DE3) and transformants were cultured for recombinant protein expression according to the general methods described previously (21). The induced cells were harvested by centrifugation and extracted with denaturant; following this, the extract was clarified by high-speed centrifugation prior to purification of the recombinant protein by IMAC performed under denaturing conditions (21). After elution, the recombinant protein was reduced with 1 mM TCEP for 30 min at room temperature prior to dilution into 10 volumes of 50 mM Tris-HCl (pH 8.3), 500 mM L-arginine, 5 mM EDTA, 3 mM reduced glutathione, and 1 mM oxidized glutathione; following an overnight incubation at 4 °C, the sample was buffer-exchanged into PBS (pH 7.4) and concentrated thereafter by tangential flow filtration. Soluble, monomeric protein was isolated by gel filtration chromatography on a Superdex S200 26/60 column that had been previously equilibrated in HBS (pH 7.4). Column fractions containing the target protein (as judged by SDS-PAGE) were pooled and concentrated. When necessary, the target protein was digested overnight with recombinant TEV protease at a 1:100 molar ratio to remove the fusion tag; upon verification of cleavage (as judged by SDS-PAGE), the target protein was separated from the free fusion tag, TEV protease, and other contaminants by IMAC where the unbound fraction was collected. The purified protein was stored temporarily at 4 °C until needed, or in 50% (v/v) glycerol at −20 °C for long-term storage.

A plasmid for expression of the C1s-2SP region of human C1s in HEK293(t) cells was also obtained from GenScript, Inc. A gene fragment encoding the native C1s signal peptide, followed by the CCP2 and SP domains of C1s, and a C-terminal octahistidine tag was synthesized and subcloned into pcDNA3.1+. Following preparation of a large quantity of plasmid, HEK293T cells were transiently transfected according to the general procedures outlined by Longo et al. (22). Once approximately 1 L of conditioned culture medium was collected, the sample was concentrated by tangential flow filtration to a volume of ~200 mL, exchanged into native binding buffer (20 mM tris (pH 8.0), 500 mM NaCl, 20 mM imidazole), and applied to a 5 mL HisTrap HP column (Cytiva). Following extensive washing with native binding buffer, the bound proteins were eluted with a linear gradient to 500 mM imidazole. Column fractions containing eC1s-2SP (as judged by SDS-PAGE) were determined to be of sufficient purity without further chromatographic procedures, and were pooled, concentrated, and stored at 4 °C until further use.

Both C1r-2SP and C1s-2SP expressed and purified from E. coli displayed catalytic activity without further modifications, as determined by their ability to cleave the model substrate ZGR (MP Biomedicals) in chromogenic assays (please see Protease Activity and Inhibition Assays). For C1r-2SP, this feature was expected since C1r is capable of self-activation. For C1s-2SP, this feature was surprising because C1s normally requires activation by C1r to exhibit catalytic activity. Although the reason for activation of C1s-2SP is not known, it is possible that either self-activation during refolding or prolonged incubation with a contaminating protease during the days-long purification and refolding procedure could be responsible. Nevertheless, the comparable catalytic properties between C1s purified from human serum and C1s-2SP strongly suggest that this recombinant protein replicates essential features of the native molecule. In the case of eC1s-2SP, the recombinant protein did not exhibit catalytic activity following purification. Therefore, purified eC1s-2SP was activated by incubation with a 1:200 molar ratio of C1r enzyme for 24 h at 37 °C. Following verification of cleavage (as judged by SDS-PAGE), the activated C1s-2SP was stored at 4 °C until further use.

Surface Plasmon Resonance –

Direct binding of small molecules to C1s-2SP was assessed by surface plasmon resonance using a T200 Biacore instrument (Cytiva) and the general methods described in an earlier publication (19). C1s-2SP (50 μg/mL in 10 mM acetate buffer (pH 4.0)) was used to modify flow cells 2, 3, and 4 of an HC1500M sensor chip (Xantec Bioanalytics, GmbH) that had been previously activated with EDC and NHS according to published procedures (19). A reference surface was prepared on flow cell 1 by EDC/NHS activation, followed by immediate quenching with ethanolamine. All experiments were performed in HBST-DMSO running buffer (HBS with 0.005% (v/v) Tween-20 and 5% (v/v) DMSO) using a flow rate of 30 μL/min. Solvent correction curves were obtained for DMSO concentrations ranging from 4.0% to 5.9% (v/v) as previously described (19). Correction curves were obtained after every 10 compounds for initial screening, while only a single correction curve series was collected at the beginning of each dose-response cohort, as described below.

An initial screen for binding to C1s-2SP was performed at a single concentration of each compound (i.e., 100 μM) using C1s-2SP surface densities of 37,488, 39,074, and 38,653 RU on flow cells 2, 3, and 4, respectively. Compounds that gave visual evidence of precipitation in the HBST-DMSO running buffer were excluded from further consideration. However, those compounds that exhibited a binding response greater than 25% of the theoretical maximum signal expected for saturation of the average C1s-2SP surface used (i.e., 38,405 RU) with the average molecular mass of the compounds in the library (i.e., 435.9 Da) were carried forward. Assuming equimolar binding of these compounds to C1s-2SP (i.e., 36,966 Da), this threshold value was 113 RU (i.e., [(38,405*(435.9/36,966))*0.25] = 113.2 RU).

Dose-dependent binding to immobilized C1s-2SP was investigated for the remaining compounds in cohorts of 2–3 at a time. Solutions of each compound were prepared at 1.56, 3.13, 6.25, 12.5, 50, 75, and 100 μM in HBST-DMSO and injected in order of increasing concentration over the sensor chip used for initial screening in reference subtraction mode. The association phase was 30 sec, while the dissociation phase was 30 sec. Dose-dependent binding to immobilized eC1s-2SP that had been enzymatically activated by C1r was also investigated for compound A1. In this case, triplicate injections over a surface with 7,116 RU of eC1s-2SP were performed using the same concentration series (absent the 75 μM sample) with an association phase of 90 sec and a dissociation phase of 150 sec. The resulting sensorgrams were analyzed using Biacore T200 Evaluation Software (Cytiva) using the steady-state affinity model and fit to a single-site binding curve using GraphPad Prism. For determination of apparent K_d values, the fitted number derived from each experimental flow cell was averaged and the standard deviation was calculated.

Protease Activity and Inhibition Assays –

The ability of A1 to inhibit the activity of C1s, C1r, and various other bloodstream proteases was assessed at 25 °C through microplate-format chromogenic assays employing substrates appropriate to each protease. In the cases of C1s, C1r, thrombin, and factor Xa,, Z-Gly-Arg-thiobenzyl ester (ZGR; ε₄₁₂ = 13.6 mM⁻¹ cm⁻¹) was used as primary substrate, while DTNB was included in the assay mixture as a fixed-concentration indicator. For the initial screening efforts, assays measuring the initial velocity of the C1s reaction (10 nM enzyme) in the presence of K_m concentrations of ZGR (i.e., ~0.16 mM) were followed in 96-well microplate format. The final concentration of each small molecule was 100 μM, while DMSO (5% (v/v)) was included in the reaction buffer to promote small-molecule solubility.

C1s.

The inhibition constant of A1 was determined using a classical 4×5 assay. The initial velocities of the C1s reaction (10 nM enzyme) were determined at four ZGR concentrations (~ 0.3 – 1.8 x K_m) using five inhibitor concentrations (0 - ~3.8 x K_i) per substrate concentration. The buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.005% (v/v) Tween-20, 5% (v/v) DMSO, and 0.44 mM DTNB. The data (n = 3) were globally fit to a competitive inhibition model using GraphPad Prism.

C1s-2SP.

The inhibition constant of A1 was determined using a classical 4×5 assay. The initial velocities of the C1s-2SP reaction (2.4 nM enzyme) were determined at four ZGR concentrations (~ 0.2 – 1.2 x K_m) using five inhibitor concentrations (0 - ~2.6 x K_i) per substrate concentration. The buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.005% (v/v) Tween-20, 5% (v/v) DMSO and 0.44 mM DTNB. The data (n = 3) were globally fit to a competitive inhibition model using GraphPad Prism.

Mouse C1s-2SP.

The inhibition constant of A1 was determined using a classical 4×5 assay. The initial velocities of the mouse C1s-2SP reaction (2.5 nM enzyme) were determined at four ZGR concentrations (~ 2.3 – 9.4 x K_m) using five inhibitor concentrations (0 - ~8.4 x K_i) per substrate concentration. The buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.005% (v/v) Tween-20, 5% (v/v) DMSO and 0.44 mM DTNB. The data (n = 3) were globally fit to a competitive inhibition model using GraphPad Prism.

C1r.

The inhibition constant of A1 was determined using a classical 4×5 assay. The initial velocities of the C1r reaction (10 nM enzyme) were determined at four ZGR concentrations (~ 0.1 – 0.7 x K_m) using five inhibitor concentrations (0 - ~0.8 x K_i) per substrate concentration. The buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.005% (v/v) Tween-20, 10% (v/v) DMSO and 0.44 mM DTNB. The data (n = 3) were globally fit to a competitive inhibition model using GraphPad Prism.

Thrombin.

The inhibition constant for A1 was determined using a classical 4×5 assay. The initial velocities of the thrombin reaction (2.7 nM enzyme) were determined at four ZGR concentrations (~ K_m – 9 x K_m) using five inhibitor concentrations (0 - ~3.4 x K_i) per substrate concentration. The buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.005% (v/v) Tween-20, 5% (v/v) DMSO with 0.59 mM DTNB as an indicator. The data (n = 3) were fit globally to a competitive inhibition model using GraphPad Prism.

Factor Xa.

Enzyme activity (3.5 nM enzyme) was determined at inhibitor concentrations that ranged from 0 to 226 μM. The concentration of ZGR was fixed at 42 μM. Buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.025% (v/v) Tween-20, 5% (v/v) DMSO with 0.59 mM DTNB as an indicator.

Neutrophil Elastase.

Enzyme activity (6.9 nM enzyme) was determined at inhibitor concentrations that ranged from 0 to 29.4 μM. The concentration of the substrate (methoxysuccinyl-Ala-Ala-Pro-Val-p-nitroanalide; ε₄₀₀ = 12.3 mM⁻¹ cm⁻¹) was fixed at 52 μM. Buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, and 0.05% (v/v) Tween-20.

Cathepsin G.

Enzyme activity (9.5 nM enzyme) was determined at inhibitor concentrations that ranged from 0 to 29.4 μM. The concentration of the substrate (succinyl-Ala-Ala-Pro-Phe-p-nitroanalide; ε₄₀₀ = 12.3 mM⁻¹ cm⁻¹) was fixed at 0.92 mM. Buffer conditions consisted of 50 mM HEPES (pH 7.4), 140 mM NaCl, 0.025% (v/v) Tween-20, and 20.7% (v/v) DMSO.

ELISAs for Complement Activity –

The ability of A1 to inhibit the CP was assessed through a pathway-specific ELISA using the general methods previously described (23–26). A dilution series of A1 (100 – 3.1 μM) was incubated with 2% (v/v) AS (Innovative Research) as a source of complement components in HBS++ buffer (10 mM HEPES (pH 7.3), 140 mM NaCl, 0.15 mM CaCl₂, 0.5 mM MgCl₂) plus 0.1% (w/v) gelatin. Reactions were added to the wells of ELISA plates that had been previously coated with human IgM (MP Biomedicals). Following a 1 h incubation at 37 °C, the wells were washed with TBS-T and probed with a mouse monoclonal antibody against C4b (Quidel). The wells were then washed and processed for detection using an HRP-conjugated anti-mouse IgG secondary according to standard procedures. Data were normalized to both an uninhibited reaction as a positive control and a serum-free reaction as a negative control. Triplicate data (n = 3) were fitted by non-linear regression to a four-parameter dose-response model using GraphPad Prism. The ability of A1 to inhibit the LP was assessed at a single 100 μM concentration using an identical protocol as for the CP with the following exceptions: IgM was replaced with 20 μg ml⁻¹ mannan from Saccharomyces cerevisiae (Sigma) and reactions were incubated with 2% (v/v) C1q-depleted serum (Complement Technologies).

Activation of the Classical Pathway by Heparin-Induced Immune Complexes –

The inhibitory effects of A1 on CP activation were examined using methods previously described (8, 9, 27). Freshly drawn whole blood was obtained from a healthy donor following written consent using an institutional review board-approved protocol (Duke IRB#: Pro00010740); human subjects were enrolled in accordance with the Declaration of Helsinki. Acid-citrate-dextrose (ACD) anticoagulated whole blood was prepared by mixing 1.4 mL of ACD solution (0.75% (w/v) citric acid, 2.20% (w/v) sodium citrate, 2.45% (w/v) dextrose) with 8.6 mL of whole blood. Samples were then pre-incubated either with buffer, a dilution series of A1 (1000 – 10 μM), 100 μg/mL Borrelia miyamotoi FbpA-C (BmA-C; positive control (28)) or with 100 μg/mL B. miyamotoi FbpB-C (BmB-C; negative control (28)) for 30 min at room temperature prior to addition of a monoclonal anti-PF4/heparin antibody (KKO, 50 μg/mL) and antigen (platelet factor 4, 25μg/mL and heparin 1U/mL) to form immune complexes in situ. Following a 30 min incubation at 37 °C, the whole blood mixture was separated by centrifugation to allow plasma collection. After separation, plasma was incubated with 10 mM EDTA to quench further complement activation, and immunoassays were used to detect C5a by a sandwich ELISA. Anti-human C5a/C5a(desArg) antibody (clone C17/5; Biolegend) was used for capture, while biotinylated anti-human C5a/C5a(desArg)/C5 Antibody (clone G25/2; Biolegend) was used for detection according to standard procedures.

Lysis of Erythrocytes by the Classical Pathway –

The ability of A1 to inhibit the CP was assessed via a hemolytic assay using the general methods previously described (23, 24, 29). A dilution series of A1 (100 – 3.1 μM) was incubated with antibody-opsonized sheep red blood cells (Complement Technologies) that had been freshly washed and 2% (v/v) AS (Innovative Research) as a source of complement components in HBS++ buffer plus 0.1% (w/v) gelatin. Following a 1 h incubation at 37 °C, the reaction was terminated by low-speed centrifugation. The absorbance of the supernatant was read at 412 nm using a multi-mode plate reader (Perkin Elmer). All data were normalized to an uninhibited control representing 100% lysis under the same reaction conditions. By contrast, background hemolysis from a negative control reaction consisting of buffer and red blood cells alone was subtracted from each reading. Triplicate data (n = 3) were fitted by non-linear regression to a four-parameter dose-response model using GraphPad Prism.

Cleavage of C2 by C1s –

The ability of A1 to inhibit cleavage of C2 by activated C1s was investigated using the general methods described previously (30). A dilution series of A1 and C1s (6.25 nM enzyme) was prepared in HBS++ buffer prior to adding 1.25 μL of C2 (0.5 mg/mL). The reaction was allowed to proceed at 37 °C for 1 hr prior to the addition of SDS-PAGE sample buffer. The reaction contents were separated by SDS-PAGE and visualized by Coomassie staining. Following staining, gels were imaged using a ChemiDocTM XRS+ (Bio-Rad) and processed for quantitation. Bands corresponding to C2b were corrected for total C2, including both cleavage products, and normalized to 100% cleavage relative to a control reaction consisting of C1s and C2 alone. Triplicate data (n = 3) were fitted by non-linear regression to a four-parameter dose-response model using GraphPad Prism.

X-ray Crystallography –

The structure of eC1s-2SP inhibited by A1 was obtained by soaking preformed crystals of eC1s-2SP with saturating concentrations of A1. First, crystals of eC1s-2SP were obtained by vapor diffusion of hanging drops prepared by mixing 1 μL of enzymatically-activated eC1s-2SP (5 mg/ml protein in 10 mM HEPES (pH 7.5), 50 mM NaCl) with 1 μL of precipitant solution (0.1 M Tris-HCl (pH 8.1), 0.2 M sodium acetate, 24% (w/v) PEG-4000) and incubating over 500 μL of precipitant solution at 20 °C. Plate-shaped crystals appeared in 2–3 days and grew to their final size in ~10–14 days. Droplets containing several single crystals were prepared for further use by adding 0.5 μL of an A1 stock solution (10 mM in ddH₂O) and incubating for 4–5 days prior harvest. Individual crystals were cryopreserved by a brief incubation in a buffer consisting of precipitant solution adjusted to 35% (w/v) PEG-4000 supplemented with 0.5 mM A1, followed by flash cooling in liquid N₂.

Monochromatic X-ray diffraction data extending to 1.8 Å limiting resolution were collected using beamline 5.0.1 of the Advanced Light Source at Lawrence Berkeley National Laboratory. Individual reflections were indexed, integrated, and scaled using HKL-2000 (31). The structure was solved by molecular replacement using PHASER (32) as implemented in the PHENIX software suite (33) and the structure of C1s-2SP (PDB entry 1ELV) as a search model (34). Following placement of the search model, an initial round of bulk solvent scaling, positional and B-factor refinement was performed using PHENIX.REFINE to calculate electron density maps (33). Thereafter, the A1 inhibitor was placed into the Fo-Fc map using PHENIX.LIGANDFIT (33). Iterative rounds of manual rebuilding using COOT (35) and refinement using PHENIX.REFINE were used to generate the final model. The final model consists of residues 358–434, 438–495, 500–598, 609–683 of human C1s (UniProtKB Accession Number P09871; https://www.uniprot.org/) separated by chain breaks due to either weak electron density or cleavage by C1r during the activation step, an oligosaccharide chain linked to Asn-406, a single copy of A1, and 335 ordered solvent molecules. Residual positive signal in the Fo-Fc electron density map was noticed near the sidechain of His-507; although this is consistent with phosphorylation of His-507, this modification was not modeled due to lack of supporting biochemical evidence or direct influence on A1 binding to eC1s-2SP. Additional information on X-ray diffraction data collection statistics and properties of the final model can be found in Table I. The final model and structure factors have been deposited in the RCSB Protein Data Bank with accession code 8TYP.

Table I.

ZINC Identifier and Chemical Formula for Compounds Within the Screening Library.

Position (1–6)	ZINC Identifier (1–6)a	Chemical Formula (1–6)	Position (7–12)	ZINC Identifier (7–12)	Chemical Formula (7–12)
A1	Z1343705767	C16H22Cl3N5	A7	Z95977728	C16H24BrN3O4S2
B1	Z1547281920	C13H15BrClN3O5S	B7	Z98335759	C21H22BrN5O4
C1	Z1336478616	C15H16N4O6S	C7	Z92636828	C21H20N4O4S2
D1	Z1326622806	C16H19N3O5S2	D7	Z295775598	C27H23N3O4S
E1	Z1212586468	C19H23FN4O4	E7	Z97494994	C23H23N3O5S2
F1	Z1286499503	C19H20N4O4S	F7	Z52343178	C25H23ClN4O4
G1	Z2099659602	C17H17N3O5S	G7	Z241516392	C14H22BrN3O4S2
H1	Z1090026012	C20H21IN4O2	H7	Z302575308	C26H20N4O3
A2	Z747119728	C15H22N4O3S	A8	Z295893766	C25H20N4O3S
B2	Z1135059766	C19H19N5O	B8	Z288695936	C19H19N3O5S
C2	Z972227838	C19H29N3O4S2	C8	Z225729548	C22H23ClN4O5S
D2	Z972260898	C17H22N4O4S2	D8	Z229770814	C20H17N3O5S2
E2	Z1739034031	C21H27N3O6S	E8	Z288956844	C15H16BrN3O5S
F2	Z1723943273	C21H33Cl2N7O2S	F8	Z32238113	C22H27N3O5S2
G2	Z1342136099	C17H21ClN4O5S	G8	Z108929330	C21H18N2O6S2
H2	Z440825006	C20H16ClFN2O5	H8	Z440639390	C18H19BrN2O5S2
A3	Z229280140	C23H19N7O3	A9	Z217940620	C21H15N3O4S2
B3	Z277956262	C21H21N5O6S	B9	Z29203941	C23H30N4O4
C3	Z238335716	C20H19N3O5S	C9	Z354559680	C24H16ClN5O2
D3	Z97002623	C22H21N5O4S2	D9	Z199746612	C19H23N3O4S2
E3	Z97372909	C24H22N4O4S	E9	Z212824170	C19H20N4O5S3
F3	Z101750316	C21H17ClN6O3S	F9	Z298227040	C16H10FN3O3S
G3	Z73960764	C21H22N4O4S	G9	Z192922634	C24H21N3O4S
H3	Z16859011	C24H21N3O4	H9	Z20236524	C25H20N4O3S2
A4	Z32325302	C22H21ClFN3O4S	A10	Z982918580	C15H18ClN3O5S3
B4	Z1181745231	C15H24IN3O	B10	Z992209426	C17H18N6O6S
C3	Z1997612270	C20H25ClN4O4	C10	Z19801969	C21H14N2O7S2
D4	Z1090012156	C17H21IN4	D10	Z1090019778	C23H34IN3O3
E4	Z1665447790	C15H10ClN3O6S2	E10	Z1020907282	C22H16N4O3S
F4	Z2526713116	C13H10F3N3	F10	Z202166220	C18H13N3O4S
G4	Z1926019888	C25H19N7O2	G10	Z66805418	C22H23N3O6S
H4	Z19159028	C18H13Cl2N3O5	H10	Z125623316	C19H24N4O4S2
A5	Z290228598	C21H19ClFN3O4	A11	Z27433306	C25H22N6O3S
B5	Z167918846	C20H15N3O5S2	B11	Z164709632	C18H22ClN5O4S2
C5	Z29403071	C21H23N3O5S3	C11	Z15563572	C21H22N2O5S
D5	Z97573819	C16H20N4O4S	D11	Z56990900	C21H25N3O6S2
E5	Z33532218	C17H17BrFN3O5S	E11	Z979350814	C22H29N3O4S2
F5	Z96789636	C25H26N4O4S	F11	Z16719846	C25H25N3O6S
G5	Z115240708	C21H22N4O5S2	G11	Z74521982	C19H15N3O4
H5	Z30002505	C22H20N4O6S	H11	Z152164986	C12H7BrN2O3S2
A6	Z62781625	C23H22N4O3S2	A12	Z19958452	C25H18ClN3O5
B6	Z117814650	C19H21N3O3S2	B12	Z18482885	C20H13F2NO7
C6	Z223649888	C18H16ClN3O6S	C12	Z57011283	C24H20N2O5
D6	Z234895193	C15H16ClN3O	D12	Z20063692	C22H27N3O6
E6	Z134706552	C20H17N3O4S	E12	Z26191186	C21H16ClN3O4S2
F6	Z30931184	C16H18BrN3O5S2	F12	Z27545504	C15H25N3O4S2
G6	Z165724264	C16H21N3O6S2	G12b	-	DMSO ( negative control)
H6	Z112422746	C22H23N3O6S2	H12b	-	DMSO ( negative control)

^aIdentifier and chemical formula taken from ZINC database (https://zinc.docking.org/) and supplied by Enamine (https://enaminestore.com).

^bPositions G12 and H12 contained DMSO alone and were used as negative controls in the validation assays.

Molecular Dynamics Simulations –

The crystal structure of eC1s-2SP in complex with A1 was used to perform molecular dynamics simulations. An all atom CHARMM molecular force field was used to parameterize the protein molecule (36–40). Topologies and parameters for A1 were obtained from the CGENFF webserver after adding explicit hydrogens to the compound via OpenBabel 2.1 (41) and converting the crystal structure coordinates of A1 to the mol2 format. The system was then solvated in a cubic box of dimensions of 101.1 Å, followed by the addition of 32,225 TIP3P waters with a cutoff of 10 Å set as the boundary of the periodic box. The solvated complex system was next charge neutralized by adding 11 Na⁺ ions in exchange for an equal number of TIP3P waters resulting in a total of 101,459 atoms for the system. Bond lengths were constrained by the LINCS algorithm (42) followed by the application of Particle Mesh Ewald electrostatics (43). Heavy atom positional restraints were applied to compound A1 via the “genrestr” program of Gromacs v2021.3 (44). The system was then subjected to an energy minimization protocol by the steepest descent method until the forces on the system converged to a minimum of ~1000 steps.

An isochoric-isothermal NVT ensemble (constant number of particles, volume, and temperature) was utilized to equilibrate the system to a final temperature of 300K for 100 ps using a Berendsen thermostat (45). Further equilibration of the system to maintain a constant pressure of 1 bar was achieved in an NPT ensemble (constant no. of particles, pressure, and temperature) following the Parinello-Rahman (46) protocol for 100 ps. All production MD runs for the system were performed in this NPT ensemble in triplicate for a period of 100 ns with coordinates being saved at 10 ps intervals. Programs “gmx trjconv”, “gmx rms”, and “gmx distance” were used to convert trajectories to different file formats (i.e., PDB), to calculate RMSDs for different group(s) of atoms, and to calculate pairwise atomic distances. Average RMSD values were calculated after each simulation equilibrated for 1 ns. Further analyses were performed using scripts written in Fortran 90, Python and C-shell.

Small-Molecule Compounds –

Compound A1 and the other compounds in the screening library were obtained commercially from Enamine and were used for all experimental validations described here. A1 and its analogs were also prepared in-house using the following general synthetic methods, with further details provided in the Supplementary Information file. Unless otherwise noted, all materials were obtained from commercial vendors and used as is without further purification. All reactions were monitored by using thin-layer chromatography (TLC) on silica gel coated glass plates or by high-pressure liquid chromatography (HPLC: Ascentis express peptide ES C-18 column, OD 3cm X 4.6cm; 6min separation time, flow rate 1 mL/min; mobile phase gradient = 95/5 → 5/95 CH₃CN-H₂O). Reverse-phase chromatography was performed on a Teledyne-Isco instrument using a Gold 50g C18 reverse-phase column with a mobile phase gradient of 100–0% → 0–100% acetonitrile-H₂O (0.1% TFA). ¹H NMR spectra were recorded on a Bruker Avance II nuclear magnetic resonance spectrometer (400 MHz) with chemical shifts reported in δ ppm relative to DMSO-d₆ (s = singlet, d = doublet, t = triplet, m = multiplet, dd = doublet of doublets, etc.). ¹³C NMR and ¹⁹F NMR spectra were obtained on the same instrument at 100 MHz and 376 MHz, respectively. Final compound purities were determined by HPLC using the conditions described above.

RESULTS

Structure-Based in silico Screening for Small-Molecule Inhibitors of C1s

The general workflow for our study was as follows (Fig. 1). We began by utilizing a machine learning approach for in silico drug discovery developed by AtomWise (20). The deep neural network AtomNet® was employed to predict small-molecule compounds that bind to the active site of C1s. A total of 94 compounds were selected for procurement and carried forward for bench validation in single concentration assays (i.e., 100 μM compound) using three orthogonal approaches that included: i) C1s-binding assays using surface plasmon resonance (SPR), ii) C1s enzyme inhibition assays, and iii) serum-based CP activation ELISAs. From an initial 94 compounds, a single compound met our hit criteria in all three assays: 6-(4-phenylpiperazin-1-yl)pyridine-3-carboximidamide, which we refer to hereafter as compound “A1”. We proceeded with quantitative characterization of A1 activity in five independent dose-response assays. Finally, we developed a structure-activity relationship (SAR) for A1 by solving a 1.8 Å crystal structure of C1s inhibited by A1, followed by subsequent synthesis and evaluation of A1 analogs.

Figure 1. Screening Workflow for the Identification of Novel Small-Molecule C1s Inhibitors.

AtomNet, a deep learning neural network developed by Atomwise, was used to screen a molecular library of several million compounds for small molecules that bind near the C1s active site. The crystal structure of C1s-2SP was used as the basis for the screen (PDB: 1ELV). Top scoring compounds were filtered to select for favorable physicochemical properties, including solubility, and a selection of 94 distinct compounds were obtained commercially for testing. Each of these 94 compounds was tested for direct binding to C1s by SPR, inhibition of C1s in purified enzyme assays, and inhibition of the CP in serum-based assays. Based on its performance across all three assay formats, compound A1 (i.e., 6-(4-phenylpiperazin-1-yl)pyridine-3-carboximidamide) was selected for further characterization. The dose-dependent binding and inhibitory properties of A1 were evaluated in five independent assays of complement function and a 1.8 Å crystal structure of a C1s fragment inhibited by A1 was obtained. SAR was then developed by synthesis and characterization of a limited series of A1 analogs.

Virtual Screening Identifies Putative C1s-Binding Small Molecule Inhibitors

The S1 site of C1s is critical for substrate recognition (i.e., C2 and C4). This functional importance arises due to S1-residue interactions with a specific substrate residue – conventionally called P1 – that lies immediately N-terminal to the scissile peptide bond (47, 48). Therefore, we hypothesized that any small molecules that bind to the S1 site of C1s would inhibit its catalytic activity. To test this idea, the previously solved 1.7 Å crystal structure of an activated form of truncated human C1s composed of the complement control protein domain 2 (CCP2) and serine protease (SP) domains (hereafter C1s-2SP) (PDB: 1ELV; (34)) was used with AtomNet to screen for compounds predicted to bind within the S1 site of C1s. Top scoring compounds from the virtual screen were clustered and filtered for favorable predicted physicochemical properties related to favorable aqueous solubility, a lack of toxic functional groups, improved blood-brain barrier penetration, and a predicted lack of pan-assay inhibitor (PAINS) behavior. The top 94 compounds were obtained commercially, dissolved in neat dimethyl sulfoxide (DMSO) to 10 mM final concentration, and added to a 96-well sample block to facilitate further characterization (Table I).

Empirical Validation of In Silico Hit Compounds

We then sought empirical validation of the hit compounds identified via virtual screening. All 94 compounds were first tested for their ability to directly interact with C1s-2SP at a single high concentration (i.e., 100 μM) using a biosensor surface that had been modified with a recombinant form of human C1s-2SP. A compound was initially considered positive for binding if it met the following criteria: (i) it exhibited clearly discernable association/dissociation phases in the sensorgram, and (ii) its background- and solvent-corrected response (i.e., resonance units (RU)) at the injection stop was greater than one standard deviation above the mean. A total of only six compounds met these stringent criteria, denoted here by their position in the 96-well plate: A1, D4, D10, F2, H1, and H9 (Fig. 2A). However, we noticed that approximately one-third of the screening compounds exhibited 25% of the theoretical maximum response for the average compound in this library (MW = 435.9 Da) if 1:1 stoichiometry was assumed and the immobilization level of C1s-2SP was considered (please see Materials & Methods). Therefore, we rescreened each of these thirty molecules, along with six non-binding compounds as negative controls, against replicate C1s-2SP biosensor surfaces to test for dose-dependent binding behavior (Fig. S1). The same six compounds identified as hits in the single concentration screen again showed the greatest binding responses across all the concentrations examined.

Figure 2. Evaluation of C1s Binding and Inhibition by Putative Hit Compounds.

(A) The pool of 94 candidate hit molecules was screened initially for solubility in HBS-T + 5% (v/v) DMSO. Those that exhibited signs of precipitation were excluded (compounds E3-E12, grey open circles). All other compounds were assayed at 100 μM for direct binding to C1s-2SP that had been previously immobilized on an SPR sensor chip. Six compounds exhibited binding responses greater than the mean value + S.D., as judged by the background- and solvent-corrected resonance units (RU) immediately prior to injection stop (A1, purple; D4, teal; D10, pink; F2 grey; H1, brown; H9, cyan). (B) The ability of C1s-2SP to cleave the substrate ZGR was measured in the presence of 100 μM of each compound. The initial reaction velocities from individual trials were plotted and compared to the mean. Four molecules diminished C1s activity to a rate of less than one standard deviation below the mean ((A1, purple; C7, blue; E3, blue-green; and E6, orange). (C) The activity of the CP was measured in triplicate by ELISA in the presence of 100 μM compound. Microtiter plates were coated with human IgM, washed, and incubated at 37 ˚C with a 2% (v/v) dilution of pooled, complement preserved human serum in a Ca²⁺ and Mg²⁺-supplemented buffer that contained 100 μM of each compound. Following a series of washes, CP activity in the presence of each compound was quantified by detection with an anti-C4b antibody coupled to colorimetric development with an HRP-conjugated secondary antibody. Mean values are presented for each compound. Only two molecules diminished CP activity to less than two standard deviations below the mean (A1, purple and B5, light brown). Since A1 was the lone compound that met each of these screening criteria, it was carried forth for further analyses. The outcome of screening assays shown in panels A-C are data representative of two replicates.

In addition to the SPR studies, we also tested whether any of the 94 compounds identified in the virtual screen were capable of inhibiting C1s in two different functional assays. The first assay monitored spectrophotometrically the ability of C1s-2SP to cleave the synthetic substrate, Z-Gly-Arg-Thiobenzyl ester (ZGR), followed by reaction of this product with the thiol indicator, 5,5′-dithiobis-(2-nitrobenzoic acid) (DTNB). Initial reaction velocities for C1s-2SP were compared for trials conducted either in the absence or presence a single, fixed concentration of inhibitor (i.e., 100 μM). We found that C1s-2SP activity was reduced to levels lower than one standard deviation below the mean by four compounds: A1, C7, E3, and E6 (Fig. 2B). In a second functional assay, we measured the ability of each compound to block C1s cleavage of C4 in the context of serum-mediated CP activation. Using IgM-coated microtiter wells as a specific initiator of the CP, we measured C4b deposition in the presence of a fixed concentration of compound (i.e., 100 μM) using an anti-C4b ELISA. Only two compounds reduced C4b deposition to less than two standard deviations below the mean: A1 and B5 (Fig. 2C). Of these, compound A1 alone resulted in essentially complete inhibition of C4b deposition at the concentration examined.

Compound A1 is a Dose-dependent Inhibitor of the Classical Pathway

Our initial validation assays identified compound A1 as the only molecule in our hit library capable of binding to C1s and inhibiting its enzyme activity in both purified and serological settings. Therefore, we obtained a resupply of the purified compound (~100 mg) from a commercial source and confirmed its identity using ¹H, ¹³C, and ¹H-¹H NMR spectroscopies as well as MALDI-TOF mass spectrometry (Data Not Shown) as a prelude to more detailed analyses. Although each of our original compounds was dissolved in neat DMSO, the A1 obtained commercially was provided as a trihydrochloride salt; this feature, along with a favorable calculated partition coefficient (cLogP) of 1.67 suggested that the compound would be water soluble. Indeed, we found that aqueous solutions of this compound could be prepared to concentrations of at least 10 mM. Thus, we performed all subsequent characterization of compound A1 in the absence of DMSO. Aside from its solubility, A1 had other favorable physicochemical properties including a relatively low molecular weight (281.36 Da), 2 hydrogen bond donors, 3 hydrogen bond acceptors, 3 rotatable bonds, and a total polar surface area of 70 Ų. Based upon these features, A1 met all Lipinski’s criteria for Rule of Five compliance (49, 50).

To test whether compound A1 exhibited dose-dependent and saturable inhibition of C1s, and to define its mechanism of inhibition, we carried out a series of functional assays in systems of decreasing overall complexity. We first examined whether A1 could block activation of the CP in whole blood triggered by formation of immune complexes consisting of platelet factor 4 (PF4)/heparin and an anti-PF4/heparin monoclonal antibody (8, 27). Using a dilution series of A1 that ranged from 1,000 – 10 μM, we found that increasing concentrations of A1 resulted in progressively greater inhibition of C5a generation (Fig. 3A). The greatest change in the extent of CP inhibition occurred between A1 concentrations of 50 and 250 μM, at which point a statistically significant decrease in C5a generation was observed when compared to buffer control. Although A1 resulted in only ~75% inhibition at the highest concentration examined, these experiments showed clear evidence of dose-dependent inhibition of the CP in whole blood.

Figure 3. Compound A1 is a Dose-Dependent Inhibitor of the Classical Pathway.

A1 was tested for the ability to dose-dependently inhibit CP activity in a series of functional assays. (A) ACD-anticoagulated whole blood was incubated with either a dilution series of A1 ranging from 1,000 – 10 μM, BmA-C (positive control), BmA-C (negative control), or buffer alone, and combined with samples of an anti-PF4/heparin monoclonal antibody and PF4/heparin antigen to form immune complexes. Each reaction was incubated at 37 °C for 30 min, and the extent of CP activity was assessed by isolating the plasma fraction and processing its contents for anti-C5a ELISA. The outcomes of triplicated measurements are shown with the error bars representing the standard deviation. Statistical significance was assessed by unpaired t-test of the relevant experimental groups versus buffer alone: *, p < 0.05, ***, p < 0.005. (B) A fixed dilution of complement preserved AS was mixed with a dilution series of A1 ranging from 100 – 3.1 μM. Following addition of the serum mixture to antibody-sensitized sheep erythrocytes, each reaction was incubated at 37 ˚C for 1 h, and the extent CP-mediated hemolysis was measured spectrophotometrically. The outcomes of triplicate measurements are shown as the mean (filled circles) with error bars representing the standard deviation. The change in hemolysis (%) versus A1 concentration was fit to a dose-response curve (solid line). The apparent IC₅₀ for A1 in this assay is 16 μM. (C) A fixed dilution of complement preserved AS was mixed with a two-fold dilution series of A1 ranging from 100 – 3.1 μM. Following addition of the serum mixture to IgM-coated microtiter wells, each reaction was incubated at 37 ˚C for 1 h, and the activity of the CP was measured by ELISA using levels of C4b deposition as a readout. The outcome of triplicate measurements is shown as the mean (filled circles) with error bars representing the standard deviation. The change in complement activity versus concentration was fit to a dose-response curve (solid line). The apparent IC₅₀ for A1 in this assay is 22 μM. (D-E) The ability of A1 to inhibit C1s cleavage of C2 was measured using an electrophoresis-based assay that monitored the appearance of the cleavage product C2b. A two-fold dilution series of A1 ranging from 100 – 1.56 μM was prepared and added to reactions containing 6.25 nM C1s and 670 nM C2. Following incubation at 37 ˚C for 1 h, the reaction contents were separated by SDS-PAGE. The results of densitometry analysis on triplicate measurements are shown in panel D, while a representative gel is shown in panel E. The experiments shown in panels A-E were performed a minimum of three times.

Next, we tested A1 in two independent serum-based CP assays. The first assay utilized a fixed dilution of human serum incubated with antibody-sensitized sheep erythrocytes to assess hemolysis spectrophotometrically. Using a dilution series of A1 that covered 100 – 3.1 μM, we observed saturable, dose-dependent inhibition of CP activity (Fig. 3B). The IC₅₀ value for A1 in this experiment was 16 μM, which was consistent with the potency observed in the previous whole blood study. We then repeated the CP-specific ELISA for C4b deposition (described above in the wider screening of all compounds) in the presence of dilution series of A1. Using a concentration range that spanned 100 – 3.1 μM, we likewise observed saturable, dose-dependent inhibition of the CP (Fig. 3C). The IC₅₀ value of A1 in this experiment was 22 μM, which again corresponded well to our earlier experiments.

While the CP ELISA provided insight into cleavage of C4 by C1s, cleavage of the other C1s substrate, C2, was not readily measured in that assay format. To test whether A1 could inhibit proteolysis of C2 by C1s in a dose-dependent manner, we performed a gel electrophoresis-based assay to monitor C1s cleavage of C2 in the presence of a dilution series of A1. Purified C1s enzyme was incubated with C2 substrate and a two-fold dilution series of A1 ranging from 100 – 1.56 μM and the products were separated by electrophoresis. We observed dose-dependent inhibition of C2 cleavage by C1s, as judged by densitometry of band intensities corresponding to the substrate, C2, and the cleavage product, C2b (Fig. 3D, E). Following fitting both band series to a dose-response curve, the IC₅₀ values for each was ~85 μM. Collectively, these results demonstrate that compound A1 is a dose-dependent inhibitor of CP activity in both serum and whole blood via its ability to block C1s cleavage of C4 and C2.

Compound A1 is a Competitive and Selective Inhibitor of C1s

To gain further insight into the mode of C1s inhibition, we determined an apparent inhibition constant (K_i) for A1 against both native C1s isolated from human serum and a recombinant C1s-2SP fragment, as well as recombinant mouse C1s-2SP. In these experiments, we carried out replicate measurements of enzyme activity across a range of substrate and inhibitor concentrations. We determined initial reaction rates for each substrate and inhibitor condition, which allowed derivation of apparent K_i values and other kinetic parameters using non-linear curve fitting. We found that compound A1 behaved as a competitive inhibitor of native human C1s, C1s-2SP, and mouse C1s-2SP with K_i values of 5.8, 8.5, and 24.6 μM, respectively (Fig. 4A-C). We noted that the K_i parameters for A1 inhibition of the human C1s-2SP fragment corresponded well to its apparent K_d of 9.8 ± 1.1 μM, as determined by SPR (Fig. S2A, B). This would be expected for a purely competitive inhibitor of enzyme activity. These observations were also consistent with what might be expected for a small molecule that binds within the S1 site, which is known to play an important role in substrate binding to the C1s enzyme. Interestingly, and in contrast to its effects on C1s, we found that compound A1 was a substantially weaker inhibitor of C1r with a K_i of 842 μM (Fig. 4D). This ~145-fold selectivity was surprising since C1s and C1r are closely related to one another and share ~40% sequence identity. This result also strongly suggests that the ability of A1 to block the CP arises from competitive and selective inhibition of C1s, rather than by broad-based inhibition of other proteases, including C1r.

Figure 4. Compound A1 is a Competitive Inhibitor of C1s.

The influence of increasing concentrations of A1 on the activity of C1s was examined by steady-state kinetics. Note that symbols identifying the plots corresponding to various concentrations of A1 are inset, with values in micromolar. (A) Apparent turnover number of C1s purified from human serum versus substrate concentration in the presence of five different concentrations of A1. Data were collected in triplicate. Fitting these data to a competitive inhibition model yields a K_i of 5.8 μM. (B) Analogous experiment to that in Panel A, except for using recombinant C1s-2SP expressed and purified from E. coli. Fitting these data to a competitive inhibition model yields a K_i of 8.5 μM. (C) Analogous experiment to that in Panel A, except for using recombinant mouse C1s-2SP expressed and purified from E. coli. Fitting these data to a competitive inhibition model yields a K_i of 24.6 μM. (D) Analogous experiment to that in Panel A, except for using C1r purified from human serum. Fitting these data to a competitive inhibition model yields a K_i of 842 μM. The experiments shown in panels A-D were performed a minimum of three times.

Although we found that A1 was selective in its inhibition of C1s when compared to C1r, there are numerous other serine proteases found in the bloodstream that may be subject to off-target effects of A1. To gain initial insight into this issue, we examined if A1 could block the LP. We incubated C1q-depleted serum with 100 μM A1 and monitored LP activation on ELISA plates previously coated with mannan. We found that A1 had relatively weak activity against the LP compared to the CP under these conditions (Fig. S2C). Next, we tested whether A1 had any effect on the activities of several other non-complement bloodstream serine proteases. Whereas we found that A1 was a competitive inhibitor of thrombin (Fig. S2D), its K_i value for thrombin of 51.2 μM was nearly an order of magnitude weaker than that for native C1s. Importantly, A1 did not inhibit Factor Xa (Fig. S2E), nor the two most abundant neutrophil serine proteases Neutrophil Elastase (Fig. S2F) or Cathepsin-G (Data Not Shown). These data, in conjunction with those for C1r, demonstrate that A1 is a selective serine protease inhibitor. Combined with its ability to inhibit murine C1s, these features support A1 as a potential lead compound for further development of C1s-targeted complement inhibitors and research tools.

Insights into the Structure-Activity Relationship for Inhibition of C1s by A1.

A1 was identified using the structure of unbound C1s as template for virtual screening. However, we sought empirical structural information on C1s inhibited by A1 to establish a structure-activity relationship and to lay a foundation for future medicinal chemistry efforts. We succeeded in growing crystals of a recombinant version of human C1s-2SP secreted from HEK293T cells (hereafter eC1s-2SP) that diffracted to 1.4 Å resolution in their apo-form (Data Not Shown). Thereafter, we incubated pre-formed crystals of eC1s-2SP with saturating concentrations of A1 and collected X-ray diffraction data to 1.8 Å limiting resolution (Table II). Following structure solution, we observed unambiguous electron density corresponding to A1 within the S1 site that is adjacent to the C1s catalytic site (Fig. 5A). We then modeled the inhibitor at full occupancy and refined the structure to an R_free of 19.6% (Fig. 5B, C & Table II). We found that the pyridine-3-carboximidamide group of A1 lay deep within the S1 substrate binding pocket where it engaged in apparent hydrogen bonds with both main chain- and side chain-derived groups from multiple residues of the protease, including Asp-626, Ser-627, Lys-629, Pro-657, and Tyr-662. By contrast, the terminal phenyl group of A1 occupied a position near several voids and channels that could be targeted during future hit expansion efforts (Fig. 5B, C).

Table II.

X-ray Diffraction Data, Structure Solution, and Refinement Statistics^a

	eC1s-2SP / A1
PDB Accession Code	8TYP
Data Collection
Space Group	P21
Wavelength (Å)	1.000
Cell Dimensions
a, b, c (Å)	41.71, 79.82, 60.44
α, β, γ (°)	90.00, 95.67, 90.00
Resolution (Å)	50.00–1.80
Wilson B-Factor (Å2)	24.0
Completeness (%)	97.1 (80.6)
I / σI	10.6 (1.0)
Rpim	0.065 (0.537)
CC1/2	0.973 (0.464)
Redundancy	6.1 (3.5)
Refinement
Resolution (Å)	48.03–1.80
Number of Reflections	35,339
Rwork / Rfree (%)	18.5 / 19.6
Atoms Modeled	2,795
Protein	2,390
Carbohydrate	49
Ligand (A1)	21
Water	335
Average B-Factors (Å2)	30.4
Protein	29.1
Carbohydrate	46.2
Ligand (A1)	26.9
Water	37.9
Ramachandran Plot
Favored/Allowed (%)	96.69 / 1.99
R.M.S. Deviations
Bond Lengths (Å)	0.007
Bond Angles (°)	0.91

^aValues in parentheses are for the highest-resolution shell.

Figure 5. Crystal Structure of Human eC1s-2SP Inhibited by A1 at 1.8 Å Resolution.

Preformed crystals of eC1s-2SP were incubated with saturating concentrations of A1 prior to crystal harvest. Following collection of X-ray diffraction data to 1.8 Å limiting resolution, the structure was solved by molecular replacement and refined to an R_free value of 19.6%. (A) Representation of the eC1s-2SP active site following initial structure solution and refinement, but prior to modeling A1. The 2Fo-Fc electron density map contoured at 1.3σ is shown as a blue cage, while the Fo-Fc electron density map contoured at 3.0σ is shown as a green cage. C1s-2SP is drawn in ball-and-stick convention with its carbon atoms colored orange. Residues corresponding to the catalytic triad appear with their carbon atoms colored silver. (B) Representation of the final model for eC1s-2SP inhibited by A1. eC1s-2SP is shown as an orange molecular surface, with the exception that atoms from catalytic triad residues are colored silver (carbon), red (oxygen), and blue (nitrogen). The A1 molecule is drawn in ball-and-stick convention with its carbon atoms in yellow. The 2Fo-Fc electron density map corresponding to A1 and contoured at 1.3σ is shown as a blue cage. Positions of key residues are denoted by text. (C) Schematic representation of the A1-binding site of eC1s-2SP in the final model. The A1 molecule and its corresponding electron density map are drawn identically to panel B, while key residues from eC1s-2SP are drawn in ball-and-stick convention. Catalytic triad residues appear with their carbon atoms colored silver. Likely hydrogen bonds between the enzyme and A1 (green), or with the catalytic triad (blue) are drawn as dashed lines.

To gain additional insight into the binding mode of A1, we used the crystal structure of eC1s-2SP inhibited by A1 to carry out 100 ns molecular dynamics (MD) simulations in triplicate. A movie of a representative simulation is available in the Supplemental Material online (Supplemental Movie 1). Root mean square deviations (RMSD) were monitored across the time course of each simulation to assess the stability of eC1s-2SP (C_α RMSD) and A1 (Ligand RMSD) relative to the conformation observed in the crystal structure. The backbone eC1s-2SP RMSD (i.e., C_α RMSD) equilibrated by ~1 ns and ranged between average values of 2.07 – 2.21 Å (SD = 0.23 – 0.38 Å) across the three independent simulations (Fig. 6A). By contrast, the A1 RMSD values ranged between 3.18 – 4.18 Å (SD = 0.69 – 1.18 Å) across the three simulations (Fig. 6B). Notably, we found that the terminal phenyl moiety is relatively dynamic in these simulations when compared to the other A1 ring structures.

Figure 6. Molecular Dynamics Analysis of eC1s-2SP Inhibited by A1.

The final crystallographic model of eC1–2SP inhibited by A1 was used as a starting point for three independent molecular dynamics simulations lasting 100 ns each (please see Materials & Methods). A movie showing the progress of a representative simulation is available in the Supplementary Material online. (A) RMSD of the eC1s-2SP Cα atoms across the duration of the simulation. A legend is inset. (B) RMSD of the A1 heavy atoms across the duration of the simulation. A legend is inset.

With empirical structural information in place, we set out to develop an initial SAR model for A1. To do this, we synthesized a limited panel of analogs based upon the A1 scaffold. A1 and various analogs were synthesized with minimal steps and in generally in good yields (Fig. S3). Using this scheme, we prepared the parent molecule along with seven different analogs that represented simple substitutions at positions within the phenyl ring (Fig. 7). Following chemical characterization of each analog, we determined K_i values against human C1s-2SP in the same enzyme activity assay used previously. Although none of these analogs had improved potency when compared to the parental A1 compound, several had comparable potency that allowed us to discern some SAR information. For example, substitutions at the para position of the ring (i.e., -R3) were deleterious as replacement of the hydrogen with even small groups like fluorine or chlorine resulted in noteworthy loss of activity (i.e > ~10-fold). By contrast, substitution at the ortho position of the ring (i.e., -R1) was better tolerated, since replacement of the hydrogen with chlorine resulted in only a ~5-fold loss in activity when compared to A1. Most importantly, we found that substitutions at the meta position of the ring (i.e., -R2) appeared to be preferred. Replacement of the hydrogen with fluorine (K_i = 19.5 μM), chlorine (K_i = 32.6 μM), or methyl (K_i = 22.6 μM) yielded compounds of comparable potency to A1 itself. We noted that each of these observations was consistent with our crystal structure of A1 bound to eC1s-2SP, which showed obvious voids projecting to either side of phenyl ring (Fig. 5B). Although the panel of analogs we prepared was limited in both its extent and chemical diversity, we believe this existing SAR information alongside the co-crystal structure suggests a productive trajectory for the development of improved compounds based on the existing A1 scaffold.

Figure 7. Structure-Activity Relationships of a Panel of A1 Analogs.

A synthetic scheme for A1 and various analogs was devised and used prepare new compounds (please see Materials & Methods). The effects of various substitutions within the terminal phenyl ring of the A1 molecule (top panel) were evaluated in a C1s enzyme activity assay. The results were used to generate information on the structure-activity relationship of the A1 scaffold (bottom panel) in conjunction with the crystallographic information presented earlier (please see Fig. 5). K_i determinations are reported as the mean of three independent replicates.

DISCUSSION

The field of complement targeted therapeutics has undergone rapid growth in the last decade. Building upon more than fifteen years of success with C5 inhibitor therapy for diseases such as paroxysmal nocturnal hemoglobinuria (PNH) and atypical hemolytic uremic syndrome (aHUS), a rich pipeline of candidates targeting other complement components has recently culminated in noteworthy approvals of new complement targeted drugs for various conditions (4, 6). Among these is the first truly CP-specific drug in the form of the C1s-inhibitory monoclonal antibody, sutimlimab, for treatment of CAD. However, since the CP activity requires three different proteases (i.e., C1r, C1s, and C2), it seems that new CP-specific drugs could be developed from selective small-molecule protease inhibitors. Here we used in silico screening and functional validation to identify a small molecule, A1, that binds within the S1 substrate-recognition site of complement protease C1s ( Figs. 1–2, 5-6 & Figs. S1-S2). Using studies on recombinant forms of C1s as well as enzyme purified from human serum, we found that A1 acts as a competitive inhibitor of C1s activity with potencies in the 1–10 μM range (Fig. 4). Consistent with these properties, we determined that A1 dose-dependently blocks CP activity in a pathway-specific ELISA, a hemolytic assay performed under CP-specific conditions, and a whole-blood assay that monitored CP activation by ultra-large immune complexes (Fig. 3). Finally, we used observations from X-ray crystallography, molecular dynamics simulations, and characterization of a panel of A1 analogs to derive SAR information on the A1 scaffold (Figs. 5–7). Although small-molecule inhibitors of C1s have been previously described (14–18), the breadth of structural and functional information we present here, along with the water-soluble nature of A1 and its small size (M.W. = 282.2 Da), make it an attractive starting point for the design and further development of C1s-targeted inhibitors.

A major consideration for developing protease inhibitors as drugs concerns their selectivity (11). In that regard, we found that A1 inhibits C1s with low micromolar potency (K_i = 5.8 μM), but that its effects on C1r were much weaker (K_i = 832 μM) (Fig. 4). Since C1s and C1r are present in identical stoichiometry in C1 (i.e., C1qC1r₂C1s₂) (1, 2, 4) and are found in essentially equal concentrations in human serum (51), this discrepancy in potencies argues that CP inhibition by A1 arises largely through its effects on C1s (Fig. 3). We also found that A1 inhibits thrombin (K_i = 51.2 μM), but at a potency intermediate to that of C1s and C1r (Fig. 4). Although this suggested that A1 might display somewhat promiscuous inhibition, we found that A1 only weakly inhibited the LP, and we failed to detect inhibition of Factor Xa, neutrophil elastase, or cathepsin-G (Fig. S2). This stands in contrast to other mechanism-based (e.g. diisopropyl fluorophosphate) or mixed mode (e.g. nefamostat mesylate) inhibitors, which exhibit broad-spectrum activity against many of the serine proteases found in the bloodstream. We suspect the lack of broad-spectrum inhibition by A1 arises from its interactions with the S1 substrate recognition site and areas adjacent (Fig. 5), which differ between these serine proteases, as opposed to the catalytic triad, which they all have in common. Collectively, our observations suggest that A1 has a promising selectivity profile for an early-stage inhibitor.

Although our results show that A1 is a dose-dependent inhibitor of the CP and of C1s activity, further work aimed at optimizing the potency and selectivity of next generation A1 analogs will be necessary. We expect this process to be greatly enabled by the 1.8 Å crystal structure of eC1s-2SP inhibited by A1 (Fig. 5), but especially when interpreted through the lens of supporting functional data. The need for both structural and functional data is illustrated by our studies on the pyridine-3-carboximidamide group of A1. The co-crystal structure revealed formation of six potential hydrogen bonds between residues lining the S1 pocket and the pyridine-3-carboximidamide moiety. We therefore reasoned that pyridine-3-carboximidamide itself might be an inhibitor of C1s. However, we failed to detect such inhibition, even at concentrations approaching ~375 μM (Data Not Shown). While surprising, this result is consistent with an earlier mechanistic investigation on inhibition of bovine pancreatic trypsin by nefamostat mesylate (52). In that case, the enzyme de-acylation product, 4-guanidinobenzoic acid, failed to inhibit trypsin at concentrations up to 200 μM (52). Considered together, these observations seem to suggest that simply binding to the S1 pocket is not sufficient for inhibition. Perhaps this also explains why other C1s binders (e.g. D4, D10, F2, H1, and H9) we identified failed to inhibit C1s enzyme activity as well ( Figs. 1–2, Fig. S1, & Table I).

When considering the distal region of A1, the co-crystal structure suggests that it should be possible to substitute onto the phenyl group. How this might be approached for the purpose of gaining potency and selectivity can be inferred from our characterization of several A1 analogs (Fig. 7). In those experiments, we found that substitution with even small halogens at the para position of the phenyl ring was not tolerated. This is consistent with the co-crystal structure, which showed that this portion of the A1 phenyl packs closely against Pro-520 and Glu-521 of C1s. By contrast, simple substitutions at the meta position (e.g. -F, -Cl, and -CH₃) yielded analogs of comparable potency to A1 itself. This is also consistent with the co-crystal structure, which showed minimal potential for steric clashes with C1s for either the ortho or meta positions of the phenyl ring. We suspect it will be possible to use the co-crystal structure to design novel compounds that not only take advantage of nearby functional groups of C1s, but also avoid unfavorable interactions. This should culminate in the gains in potency and selectivity necessary for further development of A1 analogs as therapeutic candidates.

A final aspect of our identification of A1 that should not be overlooked is its ability to inhibit murine C1s in addition to the human enzyme (Fig. 4). Many potent, synthetic complement inhibitors, such as Compstatins (53, 54) and certain modified DNA aptamers (55), as well as other pathogen-derived evasion proteins, such as SCINs (56, 57), exhibit strict host-species specificity in their action, and do not bind to the corresponding mouse complement components. Consequently, these inhibitors cannot be used to establish therapeutic proof-of-concept in widely deployed mouse models of complement-driven inflammatory diseases. This presents a significant barrier to the field, especially considering the powerful genetic tools and reagent systems otherwise available for use in experimental mice. In this regard, A1 could prove quite valuable as an experimental tool for the complement research community. Since A1 inhibits mouse C1s (K_i = 24.6 μM) with comparable potency to human C1s (K_i = 8.5 μM), it should be feasible to translate insights gained from inhibition studies involving A1 between mouse models and human diseases. As progressively potent and selective analogs of A1 are developed for human C1s, we envision that these compounds would either gain similar potency against mouse C1s or need relatively minor modifications to do so. Ultimately, such molecules would further our understanding of the role of complement and the CP in initiation and propagation of inflammatory diseases.

KEY POINTS.

*A1 was identified as candidate C1s binder through an AI-based virtual screen

*A1 inhibits C1s dose-dependently with low micromolar potency in functional assays

*Structure of C1s inhibited by A1 suggests strategies for next generation A1 analogs

ABBREVIATIONS

A1

6-(4-phenylpiperazin-1-yl)pyridine-3-carboximidamide

ACD

acid citrate dextrose

AI

artificial intelligence

AP

alternative pathway of complement

AS

active human serum

BSA

bovine serum albumin

CAD

cold agglutinin disease

CCP

complement control protein module

CP

classical pathway of complement

C1

complement component C1

C1r

complement component C1r

C1r-2SP

recombinant C1r fragment containing the second CCP domain and the SP domain

C1s

complement component C1s

C1s-2SP

recombinant C1s fragment containing the second CCP domain and the SP domain

eC1s-2SP

C1s-2SP expressed and secreted from HEK293(t) cells

C3

complement component C3

C3b

complement component C3b

DMSO

dimethylsulfoxide

DTNB

5,5′-dithio-bis-(2-nitrobenzoic acid)

EDC

1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide

FB

complement factor B

FD

complement factor D

FH

complement factor H

HPLC

high-performance liquid chromatography

HBS

HEPES buffered saline

HBS-T

HEPES buffered saline with tween-20

HBS++

HEPES buffered saline with calcium and magnesium chloride

HIS

heat-inactivated human serum

HIT

heparin-induced thrombocytopenia

HSA

human serum albumin

IMAC

immobilized metal ion affinity chromatography

LP

lectin pathway of complement

MD

molecular dynamics

NHS

N-hydroxy-succinimide

NMR

nuclear magnetic resonance spectroscopy

PBS

phosphate buffered saline

RU

resonance unit

SAR

structure-activity relationship

SCR

short consensus repeat domain

SPR

surface plasmon resonance

TBS-T

tris buffered saline with tween-20

TCC

terminal complement complex

TCEP

tris(2-carboxyethyl)phosphine

ZGR

Z-Gly-Arg thiobenzyl ester