Receptor residence time trumps drug-likeness and oral bioavailability in determining efficacy of complement C5a antagonists

Drug discovery and translation are normally based on optimizing efficacy by increasing receptor affinity, functional potency, drug-likeness (rule-of-five compliance) and oral bioavailability. Here we demonstrate that residence time of a compound on its receptor has an overriding influence on efficacy, exemplified for antagonists of inflammatory protein complement C5a that activates immune cells and promotes disease. Three equipotent antagonists (3D53, W54011, JJ47) of inflammatory responses to C5a (3nM) were compared for drug-likeness, receptor affinity and antagonist potency in human macrophages, and anti-inflammatory efficacy in rats. Only the least drug-like antagonist (3D53) maintained potency in cells against higher C5a concentrations and had a much longer duration of action (t1/2 ~ 20 h) than W54011 or JJ47 (t1/2 ~ 1–3 h) in inhibiting macrophage responses. The unusually long residence time of 3D53 on its receptor was mechanistically probed by molecular dynamics simulations, which revealed long-lasting interactions that trap the antagonist within the receptor. Despite negligible oral bioavailability, 3D53 was much more orally efficacious than W54011 or JJ47 in preventing repeated agonist insults to induce rat paw oedema over 24 h. Thus, residence time on a receptor can trump drug-likeness in determining efficacy, even oral efficacy, of pharmacological agents.

study demonstrates an important lesson in drug discovery and development, that ligand residence time on its receptor can trump rule-of-five considerations and be an overriding feature in dictating drug efficacy in vitro and in vivo, even oral efficacy for compounds with vastly inferior oral bioavailability. Our study highlights the need for more sophistication now in approaching drug discovery and development in order to successfully translate compounds to market.

Results
Comparative antagonism of C5aR. Comparative antagonist potencies and mechanisms under identical conditions were investigated here for the three chemical probes (3D53, W54011, JJ47) in human monocyte-derived macrophages (HMDM). In competitive radioligand-binding experiments using recombinant human 125 I-C5a, the binding affinities of 3D53 and W54011 for HMDM were comparable, and only slightly weaker for JJ47 ( Fig. 2A-C). The concentration-response curves for calcium mobilization induced by rhC5a were determined in the presence of escalating concentrations of each of the three antagonists ( Fig. 2D-F). A reduction of the maximal C5a responses was observed as the concentration of 3D53 increased, but there was no rightward shift of the curve typical of competitive or surmountable antagonism, consistent instead with insurmountable C5aR antagonism by 3D53 (Fig. 2D). By contrast, both W54011 and JJ47 were dependent on the C5a concentration. Both caused a rightward shift in concentration-response curves for C5a-induced calcium release in HMDM, without depressing the maximal responses, indicating that both compounds were surmountable antagonists (Fig. 2E,F). A Schild plot and pA 2 analysis revealed a slope of 0.5 for 3D53 (Fig. 2G), indicating it as a non-competitive rather than competitive antagonist 18 . On the other hand, the slopes for W54011 (Fig. 2H) and JJ47 (Fig. 2I) were both ~1, consistent with these compounds being competitive antagonists with C5a on HMDM and antagonist IC 50 values being dependent on the concentration of C5a used. The antagonist IC 50 value for 3D53 was independent of C5a concentration, maintaining potency against low (1 nM) and high (300 nM) concentrations of C5a (Fig. 3A). On the other hand, W54011 lost most of its antagonist activity against 100 nM C5a (Fig. 3B), with JJ47 being even less potent and losing its antagonist activity against just 10 nM C5a (Fig. 3C). Clearly, W54011 and JJ47 are only antagonists if measured against very low concentrations of C5a, and are unlikely to be very effective in vivo under pathophysiological conditions. Accordingly, for inflammatory diseases such as sepsis 19 or for chronic disease where concentrations of C5a can be as high as 10-100 nM, neither W54011 nor JJ47 would be expected to be very effective antagonists.
C5a-induced chemotaxis is inhibited most effectively by 3D53. Insurmountable antagonists with a long residence time (slow off-rate) from the receptor-binding site may not reach a true equilibrium between the agonist and antagonist on the receptor in the short timeframe of the calcium release assay (i.e. < 5 min). Since C5a is one of the most potent endogenous agents known to induce chemotaxis of immune cells, a more stringent and longer duration test of surmountability is a chemotaxis migration assay, which was performed to investigate the C5aR antagonist mechanism. The antagonist-treated HMDM were exposed to C5a for 16 h continuously. Antagonists with a short residence time will be ineffective in this assay, because once the antagonist has dissociated from C5aR, C5a immediately binds and induces migration. While 3D53 was able to block C5a-induced migration at a concentration of 100 nM, W54011 and JJ47 were not effective antagonists even at 1 μM concentrations ( Fig. 4A) over this 16 h timeframe. Since W54011 and JJ47 failed to block chemotaxis here, it seems unlikely that these antagonists would be functional in vivo in blocking chemotaxis.
Extended receptor residence time of 3D53. Residence time refers to the duration that a compound (ligand) is bound to its target and is usually determined by measuring 'on' and 'off ' kinetic rates. Drugs with long residence times potentially offer better selectivity, lower toxicity and a broader therapeutic window 20 . To determine the functional residence time of the three antagonists, washout cell experiments were performed in which HMDM were pre-treated with a saturating concentration of each antagonist (1 μM) for 1 h. HMDM were washed after 1 h to remove unbound antagonist, and then these antagonist-treated HMDM were subsequently challenged with 3 nM C5a at specified time points. From these temporal experiments W54011 and JJ47 were observed to display only short-term antagonist action, with t 1/2 = 1.2 h and 0.6 h, respectively (Fig. 4B). On the other hand, 3D53 maintained a much longer duration of action, t 1/2 = 18.2 h (Fig. 4C). These comparative data highlight the strikingly longer residence time of 3D53 as compared to W54011 and JJ47 on C5aR of HMDM.

Antagonism of C5a-induced inflammatory gene expression in human and rat macrophages.
Real-time PCR analyses were conducted to further check the effect of residence time in other assays. All three antagonists were found to block C5a-induced expression of the inflammatory genes TNF, IL1B, CCL3 and PTGS2 in HMDM at 1 h post-antagonism, while W54011 and JJ47 blocked C5a-induced CCL3 gene expression to a lesser extent (Fig. 5A). At 16 h post-antagonist treatment, 3D53 still blocked or significantly inhibited C5a-mediated expression of these human genes, whereas W54011 and JJ47 did not (Fig. 5B). Together, these findings are consistent with W54011 and JJ47 having much shorter durations of action due to much shorter residence times on C5aR of HMDM, being completely displaced from the receptor well within 16 h. Although the responses from human C5a on rat macrophages were not as potent across the same panel of genes, all three antagonists were able to block C5a-mediated responses at 1 h post-antagonism (Fig. 5C). This suggested the feasibility of comparing in vivo antagonist activity of 3D53, W54011 and JJ47 in an animal model of inflammation.   Molecular dynamics simulations of C5aR-3D53 complex. To investigate a possible mechanism accounting for the long residence time of 3D53 on the receptor C5aR, molecular dynamics (MD) simulations were performed for the C5aR-3D53 complex derived from a homology model of C5aR built from an antagonist-CXCR4 crystal structure 21 (PDB code: 3ODU). After 0-10 ns simulation (Fig. 6A), Arg6 of 3D53 faced extracellular loop 2 (ECL2) of C5aR making a short-lived contact with the Cys188 backbone carbonyl oxygen. After 15 ns, Arg6 moved closer to transmembrane helix TM7 and began to form hydrogen bonds with Asp282 ( Fig. 6B). This interaction remarkably closed the distance between the sidechains of these residues from 15 Å to 2 Å (Fig. 6B,D). During intermediate MD simulations, Arg6 of 3D53 sandwiched between TM2 and TM7 (TM2-TM7 dist 13 Å to 6.5 Å, Fig. 6D), forming stable hydrogen bonds with the backbone oxygen of Ile91 and the sidechain of Asp282 (Fig. 6C). This led to a stable salt-bridge throughout the later 85 ns and constituted the most stable protein-ligand contact observed ( Supplementary Fig. 1). Previous results supported an interaction between Arg6 of 3D53 and Asp282, since Asp282Ala mutation caused a 10-fold loss in affinity of 3D53 22 . In MD simulations, Arg175 and Glu199 also maintained hydrogen bond contact (Fig. 6B,C), suggesting stabilization of an inactive C5aR conformation. Trp5 of 3D53 was found to switch between aromatic pi-interactions and a H-bond interaction with Tyr290, the distance between the NH from the indole ring and the hydroxyl group of Tyr290 being monitored in Fig. 6D.
In early stages of the MD simulations, Trp5 was surrounded by Trp255 6.48 , Tyr258 6.51 and Tyr290 7.43 that formed a hydrophobic cage ( Supplementary Fig. 2). This observation is consistent with previous structure-activity relationships which found that an aromatic residue (Trp or Phe or Naphthylalanine) at position 5 of 3D53 and its analogues was essential for conferring antagonism and its analogues, whereas non-aromatic residues large or small conferred agonist activity 13 . Indeed MD simulations (20 ns-40 ns) reflected pi-stacking of the indole ring of Trp5 between TM6 (Tyr258) and TM7 (Tyr290) ( Supplementary Fig. 3), but after this timeframe Trp5 moved slightly outwards to form a H-bond with Tyr290 in another long-lasting interaction (50 ns-100 ns). Hence, Trp5 appears to play a dual role in conferring both antagonism and a long residence time for 3D53 on C5aR. The corresponding Tyr 7.43 residue in other GPCRs has also been proposed to be important for activation of GPCRs, including angiostensin-II type 1 receptor and M2 muscarinic receptor [23][24][25] . Hence, interaction of receptor residue Tyr290 7.43 with Trp5 of 3D53 may help stabilize C5aR in an inactive conformation. Accordingly, the RMSF plot ( Fig. 6E) indicated that the sidechain of the conserved "toggle switch" Trp255 6.48 of C5aR was lower than 1.0 Å and, together with the conserved NPxxY motif on TM7 (RMSF < 1.0 Å, shown in Fig. 6E), suggested that the C5aR-3D53 complex maintained an inactive conformation throughout the 100 ns simulations. It is well documented that during GPCR activation, TM3, TM5, TM6 adopt large conformational changes and play important roles in the activation of the receptor 26,27 . The lack of such TM conformational change in our MD simulations ( Supplementary Fig. 4) suggests that C5aR is locked in an inactive conformation by 3D53. This finding of a long-lasting locked conformation of the 3D53-C5aR complex supports previous mutagenesis results for key residues in C5aR 22 and provides novel insights into the possible antagonist mechanism accounting for insurmountability and long residence time of 3D53 on C5aR.
Inhibition of C5aR-PA-induced rat paw oedema. Exogenous recombinant C5a protein is rapidly metabolized in serum within seconds-minutes to C5a-des Arg by carboxypeptidases, which remove the C-terminal arginine that is important for high affinity binding to C5aR 28 . Consequently, C5adesArg has 10-1000 times lower agonist potency than C5a depending upon the function being measured and cell type examined 29 . The very rapid degradation of recombinant C5a protein precludes its use in vivo as an agonist. Since only 6-8 residues at the C-terminus of C5a are responsible for agonist activity, although the remainder contribute to high affinity binding, a hexapeptide derivative of the C-terminus (Ac-FKP-dChaCha-dR-OH, referred to as C5aR-PA) was instead used for inducing C5aR-mediated paw oedema in rats. This hexapeptide is a more stable agonist than C5a but with a comparable spectrum of functional C5aR-mediated responses 17,30 . This enabled comparison of the specific C5aR antagonists 3D53, W54011 and JJ47 for their effectiveness in inhibiting paw oedema induced by C5aR activation in vivo (Fig. 7).
C5aR-mediated rat paw oedema was induced by intraplantar administration of 350 μg C5aR-PA per hind paw ( Supplementary Fig. 5A). To inhibit C5aR-PA-induced paw swelling, 3D53 (5 or 10 mg/kg) was given by oral gavage prior to C5aR-PA injection, this reducing paw swelling in a sustained effect at 2 h post-injection of agonist ( Supplementary Fig. 5B). On the other hand, W54011 at 5 mg/kg per oral did not block C5aR-PA-induced paw swelling. Even at 10 mg/kg, W54011 did not attenuate paw swelling until 1 h post-agonist injection, and even this effect was quickly lost by 2 h post-injection ( Supplementary Fig. 5C). W54011 has been reported to block C5a-induced gerbil neutropenia at concentrations ranging from 3-30 mg/kg 14 (Fig. 7A). At 30 min post-injection, W54011 was more effective than 3D53 and JJ47 with 35% reduction in paw swelling (Fig. 7B). However, W54011 was no longer effective after 2 h. Interestingly, JJ47 only caused 15% reduction in paw swelling  Table 1) determined in male Wistar rats and measured under identical conditions gave the rank order W54011 (F ~ 74%) > JJ47 (F ~ 12%) > 3D53 (F ~ 2%), the same trend also being observed for maximal plasma concentration (C max ) and plasma half-life (t 1/2 ). This finding highlights the remarkable oral efficacy of 3D53 in spite of much lower oral bioavailability than the other two antagonists, its much longer receptor residence time overcoming substantial perceived liabilities associated with its inferior drug-likeness and making it more orally active and a superior antagonist of C5aR.

Discussion
Complement protein C5a plays important amplification roles in recruiting immune cells to sites of infection and in inducing release of inflammatory cytokines and other mediators. When C5a formation continues unabated or its receptor C5aR is not effectively regulated, it can lead to a diverse range of multiple pathological conditions, including anaphylactic shock, sepsis and arthritis 19 . An orally bioavailable small molecule antagonist of the by-product C5aR could avoid this complication and have therapeutic benefit for treating chronic inflammatory diseases. Three compounds 3D53, W54011 and JJ47 (Fig. 1) have been reported to be non-covalent antagonists of C5a interacting with C5aR at low nanomolar concentrations, albeit measured under different conditions on different cells. A detailed comparative study here has identified a major impediment in the development of small molecule C5aR antagonists appropriate for clinical trials and the importance of carefully considering often overlooked biological principles, in this case receptor residence time of a drug, in favour of more fashionable and better known rules for drug development.
First, a key point often overlooked is that IC 50 is a concentration-and mechanism-dependent term. IC 50 values for antagonists can only be compared against the same concentration of agonist on the same cell type under identical experimental conditions. Differences shown here for the three antagonists have very important implications (Figs 2 and 3). Cyclic peptide 3D53 was an insurmountable non-covalent binding antagonist of C5aR on HMDM with an IC 50 almost independent of C5a concentration between 0.1 nM to 300 nM, whereas W54011 and JJ47 were competitive and surmountable non-covalent binding antagonists of C5aR with IC 50 values that varied by three orders of magnitude against 0.1 nM to 300 nM C5a (Figs 2 and 3). Serum levels of C5a can be as high as 10-100 nM in patients with sepsis 19 and the local concentration of C5a is also expected to be high at sites of inflammation, although this remains to be documented. The results of this study question whether competitive or surmountable antagonists can even be effective therapeutic agents in the clinic for treating C5a-mediated diseases.
A second key point concerns antagonist residence time on its target receptor. Washout experiments used here (Figs 2-5) showed that the more drug-like compounds W54011 and JJ47 exerted their effects on macrophages for much shorter durations than 3D53 (ca. t 1/2 ≤ 2 h vs 20 h), irrespective of the functional readout measured (Ca 2+ release, inflammatory gene expression, chemotaxis). The striking differences in duration of action, being an order of magnitude longer for 3D53 than for W54011 and JJ47, reflects a much longer residence time on the receptor C5aR for 3D53. This increased residence time was linked via molecular dynamics simulations of the antagonist-receptor complex to a ligand-induced conformational change that results in the antagonist being uniquely trapped, relative to two other competitive antagonists, in the receptor between TM helices (Fig. 6). An insurmountable antagonist with long residence time such as 3D53 could be advantageous in systems with rapid and transient signalling 31,32 . Generation of C5a is localized at the membrane during inflammation and can be profoundly high for brief but repeated periods. A rapidly dissociating antagonist (W54011 or JJ47) would be less efficacious than a more slowly dissociating antagonist (3D53) that remains bound to C5aR even in the presence of high C5a concentrations. Thus 3D53 would require only once daily administration, whereas W54011 and JJ47 would be required multiple times a day and in much larger doses. The important advantages conferred by longer residence time are not widely appreciated and often overlooked in drug design and development. The importance of receptor residence time of a drug has begun to be better appreciated for kinases, with most marketed kinase inhibitors until recently having short residence times 33 . This has in part led to development of covalent kinase inhibitors 34 , and there has been a resurgence of interest in covalent drugs more generally 35 .
A third important point concerns a very misleading pharmacokinetic measurement, F% or oral bioavailability. Since drugs that are administered orally or via other routes have to survive degradative and metabolizing enzymes, cross membrane barriers, and be cleared only slowly from the circulation in order to exert systemic drug action, the degree of oral bioavailability is usually considered to be of paramount importance in drug development. Indeed rule-of-five compliance 1-3 in medicinal chemistry has completely changed the direction of drug discoverers and the chemical space in which they operate over the past two decades. F% is the fraction of orally administered drug in systemic circulation unchanged relative to its intravenously administered concentration defined as 100%. The higher F%, the lower the dose needed for therapeutic effect and this in turn reduces risks of side effects or off-target effects and toxicity. Drugs with low F% on the other hand can result in low oral efficacy, requiring higher doses, which in conjunction with high patient variability can lead to unpredictable drug responses. It is almost a paradigm in clinical settings that patient health must be correlated with circulating drug levels. However, this notion is fundamentally flawed if a drug has a fast on-rate, is cleared quickly from blood, and has a slow off-rate due to being bound to its receptor for extended durations. For example, 3D53 has a very low F% and high clearance rate versus W54011 and JJ47 (Fig. 7). Despite its inferior oral bioavailability, supposedly reflecting a much smaller amount of drug exerting its effect on its receptor, 3D53 is clearly the superior antagonist in vivo with a 10-fold longer duration of action in vivo and much greater potency in the presence of escalating concentrations of C5a (Fig. 7).
This study has demonstrated that receptor residence time of a non-covalent binding antagonist measured in vitro can relate to duration of action and degree of efficacy in vivo. Drug discovery has traditionally focused on optimizing drug affinity for a receptor, functional potency, receptor selectivity and drug-like chemical components in order to maximize pharmacokinetics and oral bioavailability. However, these properties do not necessarily maximize in vivo efficacy, as highlighted herein. Fast plasma clearance combined with low F% is commonly interpreted as resulting in low efficacy, a conclusion that can be incorrect as shown here. The affinity of a ligand for its receptor does not, per se, define the effectiveness and duration of biological action. Rather, it is the lifetime of the binary receptor-ligand complex that in part dictates the effect in the cellular and organismal context. If the drug is circulating in blood, unbound to its receptor, it is not exerting its effect and this needs to be considered more carefully during drug optimization. Antagonist action is certainly related to binding to its target, but the longer the receptor-antagonist complex is maintained intact, the longer the antagonist effect. This fundamental property of ligand residence time on the receptor is not a new concept 36,37 , but perhaps is under-appreciated. The recent resurgence of interest in covalent binding drugs 35 has not yet been matched by efforts to create non-covalent drugs with longer residence times.
The value of receptor residence time has been dramatically highlighted here for an extreme case of a molecule that violates physicochemical parameters traditionally used to characterize drug-likeness. Even though 3D53 has low oral bioavailability (Fig. 7D) it has a substantial oral activity and outperforms much more drug-like and more orally bioavailable antagonists of comparable in vitro potency when compared in vivo as an orally administered anti-inflammatory compound. As drug discovery researchers, developers and translators work through the tremendous challenges in taking effective and safer drugs to the market, perhaps one of the larger problems is a less realized one, that we may have lost sight of the importance of fundamental biological principles. Time is of the essence in drug development, and also for drug efficacy where residence time of a drug on its specific receptor can be critical in determining the duration and effectiveness of drug action.
Binding affinity assay. A scintillation proximity assay was used to measure the affinity of C5aR antagonists as described 39 . HMDM were incubated for 1 h with the specified concentration of C5aR antagonist, polyvinyltoluene scintillation beads were coated with wheat germ agglutinin (PerkinElmer) and 25 pM of [ 125 I]-C5a. Radioligand binding was then analyzed using a Microbeta scintillation counter (PerkinElmer). Intracellular calcium release assay. Intracellular calcium release was induced in HMDM as described 40 and measured using a FLIPR instrument (Molecular Devices). HMDM were seeded in multiple 96-well plates for each time point and allowed to adhere overnight. 3D53, W54011 and JJ47 were prepared in DMSO as 10 mM stock and dilutions were performed in HBSS. To investigate antagonist residence time on C5aR, HMDM were first pre-treated with each C5aR antagonist (1 μM) for 1 h at 37 °C. After 1 h, excess unbound antagonists were removed by washing with HBSS and the antagonist-treated HMDM were then challenged by stimulating with rhC5a (3 nM) after different incubation times (0, 0.5, 1, 2, 4, 6,8,16,22,36,48, 60 h) to compare the residence time of antagonist.
In vitro migration assay. HMDM were serum-deprived overnight prior treatment with C5aR antagonist (0.1 or 1 μM) for 1 h at 37 °C. Excess unbound antagonists were washed with serum-free medium and HMDM (0.1 × 10 6 cells/mL) then seeded into 5 μm Transwell inserts (Corning) creating a modified Boyden chamber. To initiate cell migration, serum-free medium with rhC5a (3 nM) was added to the lower chamber and incubated for 16 h at 37 °C. Migrated cells on the lower side of the membrane were stained with DAPI and counted. The chemotactic index was expressed as fold change versus control.
Gene expression analysis. RNA Fig. 6). A Ramachandran structural plot suggested that the homology model of the TM region of C5aR based on CXCR4 was both stereochemically and energetically reasonable ( Supplementary Fig. 7). The antagonist bound structure of CXCR4 serves a reasonable starting template. Initial sequence alignment was performed at the TM-coffee website. Further alignment was adjusted using Jalview program. Homology models were built with Modeller software with the selection of the final model gaining overall high score from both DOPE and GA341. The whole ECL2 was further refined using the loop refinement option (default setting) in Prime (version 3.1, Schrödinger, LLC, New York, NY, 2012) and the final model that incorporated the optimized ECL2 was minimized in Prime using the truncated-Newton energy minimization (OPLS_2005 force field with restrained helical backbone). The "structure assessment" module available at Swiss-Model portal (http://swissmodel.expasy.org/) was used for local model quality estimation (QMean Score) global model quality estimation (DFIRE energy) and stereochemistry check (Procheck) 45,46 . The NMR solution structure 13 of 3D53 was docked into the C5aR homology model using GOLD software with no constraints in order to find all possible poses that might agree with previous modeling and experimental results 22 . The top docked pose ( Supplementary Fig. 8) reflected a similar docking mode to previous modeling for C5aR and 3D53, where Arg6 interacted with Asp282 at the top of TM7 near to ECL3. Molecular dynamics simulations of C5aR-3D53 complex. Using the top docking pose, 100 ns MD simulations for C5aR-3D53 complex were investigated. The whole system consisting of the docked 3D53 in C5aR was first prepared with standard protocol in "Protein Preparation Wizard" within Maestro, version 9.5 (Schrödinger, LLC, New York, NY, 2013). Hydrogen atoms were added, and hydrogen bond assignment, tautomer, and protonation of amino acids at pH 7.4, were optimized. The prepared structure was then embedded in 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) membrane via membrane set up panel in Desmond Molecular Dynamics System v3.5 (D. E. Shaw Research, New York, NY, 2013). TIP3P water was used for system solvation. Neutralization of the system was implemented by addition of Cl-ions at physiological concentration of 0.15M. After solvating the whole system, a membrane protein relaxation protocol was generated and used for further simulation with Berendsen coupling. The all-atom optimized potential for liquid simulations (OPLS-2005) force field implemented as the default force field in Desmond was adopted for all molecules in the system. Prior to the MD production, two steps of minimization were applied. First step included minimization with restraints on solute heavy atoms with a force constant of 50 kcal/mol/Å 2 . Second step was the minimization without any restraints. The minimized system was further relaxed before the actual simulation. Six steps were included: 1) heating up system to 300 K with NVT ensemble with Berendsen coupling for 60 ps. 2) 200 ps equilibration NPT restrained on heavy atoms. 3) NPT equilibration of solvent and lipids for 100 ps. 4). NPT with protein heavy