An orally active small molecule fusion inhibitor of influenza virus

Abstract

Recent characterization of broadly neutralizing antibodies (bnAbs) against influenza virus identified the conserved hemagglutinin (HA) stem as a target for development of universal vaccines and therapeutics. Although several stem bnAbs are being evaluated in clinical trials, antibodies are generally unsuited for oral delivery. Guided by structural knowledge of the interactions and mechanism of anti-stem bnAb CR6261, we selected and optimized small molecules that mimic the bnAb functionality. Our lead compound neutralizes influenza A group 1 viruses by inhibiting HA-mediated fusion in vitro, protects mice against lethal and sublethal influenza challenge after oral administration, and effectively neutralizes virus infection in reconstituted 3D-cell culture of fully differentiated human bronchial epithelial cells. Co-crystal structures with H1 and H5 HAs reveal that the lead compound recapitulates the bnAb hotspot interactions.

Summary:

A small molecule that mimics the binding and functionality of a broadly neutralizing antibody as an effective fusion inhibitor of influenza virus.

The WHO estimates that annual influenza epidemics cause around 3 – 5 million cases of severe illness and up to 500,000 deaths worldwide (1, 2). Seasonal influenza vaccination still remains the best strategy to prevent infection, but the currently available vaccines offer very limited breadth of protection. The discovery of human broadly neutralizing antibodies (bnAbs) to influenza virus provides hope for development of broad-spectrum, universal vaccines (3–14). Because of the high level of conservation of their epitopes in the HA stem, these bnAbs neutralize a wide range of viruses within and across influenza virus subtypes. Their binding prevents the pH-induced conformational changes in HA that are required for viral fusion in the endosomal compartments of target cells in the respiratory tract (6–11, 13–15). Efforts have therefore been made to develop vaccination modalities aimed at directing the immune response to the HA stem through different vaccination regimens (16, 17), sequential vaccination with different chimeric HA constructs (18, 19), and administration of stem-based immunogens (20–24). In addition, several bnAbs themselves are being evaluated in clinical trials as passive immunotherapy (25). Another recent strategy to prevent influenza infection stems from development of a highly potent multidomain antibody with almost universal breadth against influenza A and B viruses that can be administered intransally in mice using adeno-associated virus-mediated gene delivery (26).

Therapeutic options to treat acute influenza infection also include antiviral drugs directed at blocking virus uncoating during cell entry (M2 proton channel inhibitors) and progeny release from infected cells (neuraminidase inhibitors) (27, 28). However, resistance to antiviral drugs is an emerging problem due to the high mutation rate in influenza viruses and their genetic reassembly possibilities (29). New antiviral drugs (30, 31) and combination therapies (32, 33), with alternative mechanisms of action against alternative viral targets are therefore urgently needed. Small molecule drugs, in contrast to antibodies, offer the advantage of oral bioavailability, high shelf stability and relatively low production costs.

Influenza A viruses have been classified into 18 hemagglutinin subtypes (H1-H18), which can be divided phylogenetically into two groups (1 and 2), and 11 neuraminidase subtypes (N1-N11). Antibody CR6261 broadly neutralizes most group 1 influenza A viruses (7, 9). Co-crystal structures of CR6261 in complex with H1 HA (7, 9), stimulated design of small protein ligands of about 10 kDa that target the conserved stem region. These small proteins mimic the antibody interactions with HA and inhibit influenza virus fusion (34–36). Co-crystal structures of bnAbs FI6v3 and CR9114 with HAs (6, 14) further enabled design of even smaller peptides as influenza fusion inhibitors (37) . However, neither small proteins nor peptides generally are orally bioavailable.

Development of small molecule ligands directed at antibody binding sites is challenging. Antibody epitopes, as for other protein-protein interfaces, are generally flat, large and undulating (~1,000 –2,000 Ų) (38), in stark contrast to the small concave pockets (typically in the 300–500 Ų range), which are common as targets for small molecule drugs (39). To mimic the function of a bnAb, a small molecule should be able bind to the antibody epitope and reproduce the key interactions that lead to fusion inhibition. We have therefore identified and optimized small molecules with such properties through application of a strategy that was guided by detailed knowledge of the binding mode and molecular mechanism of bnAb CR6261 (7, 15) and encouraged by successes in the design of small proteins and peptides to the HA stem (34, 35, 37).

High-throughput screening and optimization

To identify potent small molecules that mimic group 1 bnAb CR6261, in terms of breadth of binding (7, 9, 35), virus neutralization, and mechanism (Fig. 1A), we screened for compounds that selectively target the CR6261 epitope on HA. We applied the AlphaLISA (Amplified Luminescent Proximity Homogeneous Assay) technology in competition mode as our high-throughput screening (HTS) method (Fig. 1B). A diverse library of ~500,000 small molecule compounds was screened for displacing HB80.4, which is a CR6261-based computationally designed small protein with very similar binding mode and fusion inhibition profile (34, 35). HB80.4 was used instead of CR6261, as avidity effects leading to higher apparent affinity of the bivalent antibody would have resulted in a more stringent and thus less sensitive assay. This approach biased the screen towards compounds that act via the desired mechanism of action. About 9000 small molecules with weak to medium binding capacity were initially retrieved; binding of 300 compounds was confirmed through repeated testing and via the Truhit AlphaLISA counter assay that can identify false positive hits.

Fig. 1. Approach for small molecule discovery through mimicking the binding mode and functionality of broadly neutralizing group 1 influenza antibody CR6261.

(A) Binding mode, breadth of binding, and fusion inhibition profile of influenza HA stem targeting antibody CR6261. The left panel shows the binding mode of CR6261 to influenza HA with the HA trimer represented in a gray molecular surface with different shades for the different protomers, CR6261 in green with a molecular surface for the Fab interacting with the HA and a cartoon for the other Fab and Fc of the IgG, and the binding epitope on the HA in red. The middle panel illustrates the HA phylogenetic tree showing the relationship between the 18 HA subtypes of influenza A virus (group 1 and 2) and the two lineages of influenza B viruses. The breadth of binding of the CR6261 to multiple group 1 subtypes is shown in green, light grey indicates where binding was not tested, and black indicates no binding. The right panel shows the mechanism of action of CR6261 to block the pH-induced HA conformational changes. (B) Our HTS utilized the AlphaLISA technology to identify small molecules targeting the CR6261 epitope. The CR6261-mimicking designed protein HB80.4 was used in the binding competition assay. HB80.4 is represented in blue and the CR6261 epitope on the HA in pink. Small molecules are indicated as illustrative structural formulae.

Benzylpiperazines emerged as a major hit class with JNJ7918 being the most promising candidate with an IC₅₀ of 1.39 and 13.06 µM against H1N1 A/California/07/2009 (H1/Cal) and H5N1 A/Vietnam/1203/2004 (H5/Viet) HAs, respectively (Fig. 2). No competition with H1 HA head binding antibody 2D1 (40) was detected against H1/Cal, further validating the HA stem-specificity (Fig. 2A). Key chemical modifications to increase molecular interactions with the HA stem, i.e. introduction of a functional group at the benzylic position and modification of the phenyl with a propargyl moiety (Fig. 2B), greatly improved binding (~30–80 fold) to HAs from H1/Cal, H1N1 A/New Caledonia/20/1999 (H1/NCa), H1N1 A/Brisbane/59/2007 (H1/Bris), and H5/Viet, as exemplified by the second-generation compound JNJ6715 (Fig. 2C). Moreover, virus neutralization assays with a panel of H1 and H5 influenza strains, which represent human pandemic, seasonal and emerging viruses in influenza A group 1, demonstrated improved neutralization (~30–500 fold). Like bnAb CR6261, JNJ6715 showed heterosubtypic neutralization of group 1 viruses without measurable cytotoxicity in MDCK cells (Fig. 2D and table S1).

Fig. 2. Optimization of in vitro HA stem-binding and virus neutralization.

(A) Representative dose-dependent competitive binding for compound JNJ7918 that emerged from a HTS employing the AlphaLISA assay. This compound inhibited binding of the designed protein HB80.4 against the HA stem to H1N1 A/California/07/2009 and H5N1 A/Vietnam/1203/2004 HAs with IC₅₀’s of 2.3 µM and 21.4 µM (pIC₅₀ = 5.6 and 4.7) respectively, but not to the head binding antibody 2D1, when tested with H1N1 A/California/07/2009 HA. The x-axis depicts the log molar concentration of the compound and the y-axis depicts the normalized response, relative to the positive (no inhibitor, no HA) and negative (no inhibitor) control. (B) Chemical structures of JNJ7918 and JNJ6715 with key modifications to the latter highlighted in yellow circles. (C) Scatter plot depicting H1N1 A/California/07/2009 (H1/Cal), H1N1 A/New Caledonia/20/1999 (H1/NCa), H1N1 A/Brisbane/59/2007 (H1/Bris) and H5N1 A/Vietnam/1203/2004 (H5/Viet) binding, calculated from AlphaLISA assay as log (IC₅₀) versus virus neutralization as log (EC₅₀). Each dot represents a tested compound. The dotted line represents the lower limit of quantification (LLOQ). (D) Virus neutralization EC₅₀ values in µM for compounds JNJ7918 and JNJ6715 against the indicated virus strains: H1N1 (A/Brisbane/59/2007) (H1/Bris), H1 (A/California/07/2009) (H1/Cal), H1N1 (A/New Caledonia/20/1999) (H1/NCa), H1N1 (A/Puerto Rico/8/1934) (H1/PR8), H1N1 (A/Solomon Islands/3/2006) (H1/SI06), H5N1 (A/Hong Kong/156/1997) (H5/H97), H5 (A/Vietnam/1194/2004) (H5/Viet), H3N2 (A/Brisbane/10/2007) (H3/Bris), H7 NY (7:1 reassortant virus with the HA of A/New York/107/2003 (H7N7) and remaining segments from A/PR/8/34 rescued and propagated on PER.C6^® cells), and influenza B (B/Brisbane/60/2008) (B/Bris).

In vitro and in vivo pharmacokinetic profiling

Further in vitro profiling of JNJ6715 revealed poor kinetic aqueous solubility of 11.3 µM at pH 7.4 and poor metabolic stability of >346 mL/min/kg based on intrinsic clearance in mouse and human liver microsomes, making this compound unsuitable for in vivo testing (Fig. 3A). To enhance its drug-like properties, we replaced the methyl ester functionality at the benzylic position by a 2-methyl oxadiazole group, substituted the central phenyl by pyridine, and replaced the 6-methoxy benzothiazole by a 5-trifluoro-methoxy-benzoxazole group, generating JNJ8897. The compound was further optimized to generate JNJ4796 by replacing the 2-methyl oxadiazole group by 2-methyl tetrazole and the trifluoro-methoxy group by methyl amide (Fig. 3B). These modifications reduced the intrinsic clearance in mouse and human liver microsomes for both compounds while sustaining virus neutralization [(EC₅₀) for H1/Cal and H1/NCa of 0.064/0.076 µM for JNJ8897 and 0.066/0.038 µM for JNJ4796]. In particular, JNJ4796 demonstrated significantly reduced human and murine intrinsic clearance (Figs. 3, A and C) and increased aqueous solubility, which translated to a favorable in vivo pharmacokinetics profile with compound half-life (t_½) in mice of 2.4 hour after oral administration. The overall bioavailability for this compound was 30.0%, with a plasma concentration reaching 1152 ng/ml, equivalent to 2.1 µM, after oral administration at 10 mg/kg (Fig. 3A). Testing JNJ4796 in a diverse panel of pharmacologically relevant receptors, ion channels, and transporters showed that the compound does not substantially inhibit any of these potential off-targets (table S2).

Fig. 3. Lead compound JNJ4796 is orally bioavailable and effective in vivo.

(A) ADME (absorption, distribution, metabolism, and excretion) data of SM compounds: kinetic solubility at pH 4.0 and 7.4 in µM, CL_int (intrinsic clearance) in hLM/mLM (human/murine liver microsomes) in µl/min/mg protein, plasma protein binding (PPB) (% unbound) human/murine (hu/mu), mouse t_½ (h-hours, iv - intravenous, po - per os), AUC 0−∞ (area under the curve) in h*ng/ml, C_max (maximum concentration) po in ng/ml, t_max (time point for maximum concentration) po in hours and po bioavailability in % ( ND- not determined) (B) Chemical structures of JNJ6715, JNJ8897 and JNJ4796 with key chemical modifications compared to the previous generation highlighted in orange and red circles. (C) Scatter plot depicts CL_int (intrinsic clearance) of SM compounds in mouse and human liver microsomes in µl/min/mg versus their average log (EC₅₀) neutralization value against H1N1 A/California/07/2009 and H1N1 A/New Caledonia/20/1999 HAs. The individual black dots represent the analyzed SM compounds and the yellow, orange and red dots represent JNJ6715 (EC₅₀ = 0.14 µM), JNJ8897 (EC₅₀ = 0.063 µM) and JNJ4796 (EC₅₀ = 0.033 µM), respectively. The shaded area reflects the optimal area of high virus neutralization capacity and low intrinsic clearance. (D) Survival curves and weight loss of mice challenged intranasally (at day 0) with a lethal dose of the mouse-adapted H1N1 A/PR/8/1934 virus (25x MLD₅₀) after oral administration of the indicated compounds at day −1 to 5 (twice daily).

Oral efficacy against influenza infection

In line with its in vitro neutralizing activity and acceptable pharmacokinetics data, oral administration of JNJ4796 protected mice from lethal challenge of 25 times the median lethal dose (MLD) of H1N1 A/PR/8/1934 virus (Fig. 3D). Doses of 50 and 10 mg/kg of JNJ4796 twice daily, initiated one day before challenge and continuing for seven days, resulted in 100% survival at day 21 in comparison to the less potent compound JNJ8897 for which less than 50% survival was achieved (day 21 survival was 40% for 50 mg/kg, 25% for 10 mg/kg) (Fig. 3D, and tables S3 and S4). JNJ4796 only partly alleviated morbidity in this stringent H1N1 infection model as demonstrated by animal weight loss. Nevertheless, dose-dependent reversal of weight loss was observed at the study end for both 50 mg/kg and 10 mg/kg of JNJ4796 (Fig. 3D). Oral doses of JNJ4796 also resulted in dose-dependent efficacy after a sublethal viral challenge (MLD90), with twice daily administration of 15 and 5 mg/kg JNJ4796 giving rise to 100% survival and only moderate weight loss effects that were moreover restricted to the period directly after treatment (fig. S1, and tables S3 and S4).

To assess the potential practical utility of JNJ4796 in the context of human airway infection, the neutralization capacity of JNJ4796 was evaluated in a reconstituted 3D-cell culture of fully differentiated human bronchial epithelial cells (HBEC) derived from a pool of donors. This model system recapitulates several relevant characteristics of human airway target tissue, such as production of mucus, cilia-beating and local metabolic activity (41), which could potentially limit the compound’s efficacy. Incubation of an HBEC culture, infected with H1N1 A/PR/8/1934 virus, with the compound dramatically reduced the viral titers when assessed 96 hours post-infection (fig. S2).

Group-1 binding specificity and mechanism of JNJ4796

Like bnAb CR6261, JNJ4796 further demonstrated heterosubtypic group 1 HA binding and virus neutralization breadth without cytotoxicity (Fig. 4, A and B, fig. S3, and tables S1 and S5). CR6261 neutralizes group 1 influenza A viruses by inhibiting the low-pH induced HA conformational change, which triggers fusion of the viral and endosomal membranes and release of the viral genome into the host cell (42). Each HA monomer is comprised of two subunits (HA1 and HA2). CR6261 and JNJ4796 bind components of both HA1 and HA2 in the trimeric HA stem, thereby stabilizing the pre-fusion conformation of HA and preventing conformational rearrangements in the HA stem that lead to the post-fusion structure. This inhibition is demonstrated in two different experiments. First, in a conformational change inhibition assay (Fig. 4C), JNJ4796 dose-dependently prevents pH-induced transition of HA to the post-fusion conformation and subsequent loss of the HA1 subunit after reduction of the inter-chain HA disulfide. Secondly, the protease-susceptibility assay indicates that, like the stem-targeting bnAbs, compound JNJ4796 stabilizes the pre-fusion conformation and blocks the HA conformational change at low pH and the subsequent susceptibility to trypsin (43) (Fig. 4D).

Fig. 4. Group-1 binding specificity and mechanism of JNJ4796.

(A) Small molecule JNJ4796 binds multiple group 1 subtypes (green text) as displayed on the HA phylogenetic tree (see Fig. 1A). Grey indicates that binding against H8, bat H17 and bat H18 HAs subtypes was not tested and black indicates no binding. (B) Comparison of breadth of virus neutralization for JNJ4796 and CR6261 against representative influenza viruses: Influenza A H1N1 A/Brisbane/59/2007 (H1/Bris), H1N1 A/California/07/2009 (H1/Cal), H1N1 A/New Caledonia/20/1999 (H1/NCa), H1N1 A/Puerto Rico/8/1934 (H1/PR8), H1N1 A/Solomon Islands/3/2006 IVR-145 (H1/SI06), H5N1 A/Hong Kong/156/1997 (H5/H97), H5N1 A/Vietnam/1194/2004 (H5/Viet), H3N1 A/Brisbane/10/2007 (H3/Bris), H7 NY (7:1 reassortant virus with the HA of A/New York/107/2003 (H7N7) and A/PR/8/34 was rescued and propagated on PER.C6^® cells), and influenza B B/Brisbane/60/2008 (B/Bris). (C) Conformational change inhibition assay showing that binding of JNJ4796 to cleaved H1N1 A/Brisbane/59/2007 HA blocks the low pH-induced conformational change after lowering the pH to 5.25. The log of the molar concentration of JNJ4796 is plotted against the inhibition of conformational change normalized to the full inhibition by CR6261 (positive control). The vertical intersecting line represents the IC₅₀. (D) Small molecule JNJ4796 inhibits the low pH-induced conformational changes in H1/PR8 HA. In the trypsin susceptibility assay, exposure to low pH renders the HA (H1/PR8) as sensitive to trypsin digestion (lanes 7 vs 8), but small molecule JNJ4796 prevents its conversion to a trypsin-susceptible conformation (lanes 11 vs 12). The mechanism is similar to that of fusion-inhibiting CR6261 Fab (lanes 9 vs 10).

Crystallographic analysis of JNJ4796-HA complexes

To decipher the structural basis for its mechanism of action and broad group 1 specificity, crystal structures of JNJ4796 in complexes with H1N1 A/Solomon Islands/3/2006 (H1/SI06) and H5N1 A/Vietnam/1203/2004 (H5/Viet) HAs were determined at 2.72 Å and 2.32 Å resolution, respectively (Figs. 5 and 6, fig. S4, and table S6). JNJ4796 binds with a stoichiometry of three binding sites per trimer in a highly conserved hydrophobic groove at the HA1/HA2 interface in the HA stem (Fig. 5A). The JNJ4796 binding site comprises HA1 His¹⁸, Thr³¹⁸, β-strand residues His³⁸-Leu⁴², HA2 Thr⁴¹-lle⁵⁶ from helix-A, and Gly²⁰ and Trp²¹ from the N-terminal fusion peptide (HA2 residues are in italics throughout). The epitope recognized by the small molecule is similar to the epitopes of stem-targeting bnAbs CR6261, FI6v3 and CR9114 (Fig. 5, C and F) (6, 7, 14). JNJ4796 occupies the same hydrophobic groove as CR6261, where hydrophobic residues from the heavy-chain complementarity determining regions (HCDRs) and framework region 3 (HFR3), including HCDR2 signature residues Ile⁵³ and Phe⁵⁴, encoded by the germline gene VH1–69, insert into the binding site (Fig. 5C).

Fig. 5. Structural characterization of the JNJ4796 binding site on influenza HA.

(A) The crystal structure of JNJ4796 in complex with influenza HA from the group 1 A/Solomon Islands/3/2006 (H1/SI06) strain. JNJ4796 is shown in a ball-and-stick representation with one HA protomer of H1/SI06 rendered as a cartoon and the other two protomers in surface representation. HA1 is in light gray and HA2 in aquamarine. A zoomed in view of one of three JNJ4796 binding sites in the HA trimer is shown with the C/O/N atoms of JNJ4796 in yellow, red and blue, respectively. (B) 2Fo-Fc map (black color mesh) contoured at 1σ is displayed around the bound conformation of JNJ4796 in complex with H1/SI06 HA. (C) Overlay of the structure of JNJ4796 (yellow ball-and-sticks) in complex with H1/SI06 with the HA-interacting loop residues (green sticks) from CR6261 Fab of the HA-Fab complex (PDB 3GBN). JNJ4796 occupies the same conserved hydrophobic groove at the interface of HA1 and HA2 as residues from HFR3, HCDR1, HCDR2 and HCDR3 of Fab CR6261. (D) The molecular structure of JNJ4796 with A-E rings labeled. (E) CH-π interactions in the JNJ4796-H1/SI06 HA complex, with the interactions of each ring (labeled in Fig. 5D, depicted in black dotted lines and measured in Å). The centroid of the rings is shown in red spheres. (F) Comparison of the footprints of small molecule JNJ4796 and stem-targeting bnAbs on the HA. The JNJ4796 footprint on H1/SI06 HA is highlighted in red with the interacting residues labeled in white. Footprints of Fabs CR6261, FI6v3 and CR9114 in the complex with H1N1 A/Brevig mission/1/1918 (PDB 3GBN), H1N1 A/California/04/2009 (PDB 3ZTN), and H5N1 A/Vietnam/1203/2004 (PDB 4FQI), respectively, depicted in the green and with the corresponding interacting residues labeled in white.

Fig. 6. Structural basis for the group-specific binding of JNJ4796 on influenza HA.

(A & C) The binding mode of JNJ4796 on H1/SI06 and H5/Viet HAs. HAs are represented in gray and lavender molecular surfaces, and JNJ4796 in yellow and purple sticks, respectively. (B) Overlay of the JNJ4796 binding modes in H1/SI06 and H5/Viet. The V40Q mutation and conformations of the E-ring of JNJ4796 are highlighted in the red dotted ellipses. (D-F) Key differences in the binding site of JNJ4796 bound to H1/SI06 and H5/Viet HAs. (D) The average B-values of JNJ4796 and its individual rings A and E in the structures of JNJ4796-H1/SI06 and JNJ4796-H5/Viet HAs. (E) Non-covalent interactions of the B-ring of JNJ4796 with H5/Viet HA, where the interactions are represented in black dotted lines and measured in Å. The centroid of the ring is shown in the red sphere. The B-ring is represented in the purple ball-and-stick model and HA in light blue cartoon. (F) Overlay of the conformation of E-ring of JNJ4796 bound to H1/SI06 (yellow) and H5/Viet (purple) HAs, with nitrogen atoms of E-ring shown in blue. The E-ring shows an ~180° flip in the H1 vs H5 HAs binding site. (G-H) Superimposition of JNJ4796 (yellow) from its group-1 H1/SI06 complex onto a group-2 apo H3/HK68 (cyan) HA. H3/HK68 residues of interest are shown in cyan sticks with a transparent molecular surface. Potential steric clashes between A-E rings of JNJ4796 are shown for the glycosylated Asn³⁸ and His¹⁸ from HA1 and Trp²¹, Asn⁴⁹ and Leu⁵² from HA2 of H3/HK68 HA, respectively. (I) Zoomed in view of the conformation of residues that have potential steric clashes from group 2 H3/HK68 (cyan sticks) versus the corresponding residues from group 1 H1/SI06 (gray sticks).

JNJ4796 contacts this hydrophobic groove through both hydrophobic and polar interactions (Figs. 5E and fig. S5). The substituted benzoxazole moiety (A-ring; n.b. rings are depicted in Fig. 5D and indicated in bold throughout) of JNJ4796 occupies a small hydrophobic cavity formed by Val⁴⁰, Leu⁴² on HA1 and Val⁵², Asn⁵³ and IIe⁵⁶ on helix-A (Fig. 5, A and E). In addition, JNJ4796 makes a polar CH-π interaction with the Cγ1 CH of Val⁵². Such H-bond interactions can contribute ~1.0 kcal/mol to the binding energy (44, 45). The B-ring of JNJ4796 engages HA1 Thr³¹⁸ through a direct H-bond with its hydroxyl group and a CH-π interaction with Cγ2 CH from Thr³¹⁸. The edge of the B-ring also makes nonpolar contacts with IIe⁴⁸ and Thr⁴⁹ from helix-A. The C and D-rings of JNJ4796 make CH-π bonds with His¹⁸, His³⁸ from HA1 and Trp²¹ from HA2, whereas the E-ring makes a CH-π bond with Cδ1 CH from IIe⁴⁵. This triad of rings, C to E locks the conformation of Trp²¹, which is important in the fusion process (46, 47). Overall, JNJ4796 buries ~453 Ų (on H1) and ~472 Ų (on H5) of surface at the HA interface, completely enveloping the central hydrophobic groove as compared to more discontinuous coverage by stem-targeting bnAbs (CR6261, FI6v3 and CR9114) (Fig. 5F).

The structural basis for the ~25 fold decreased binding affinity of JNJ4796 to H5 HA compared to H1 HA was elucidated by comparing the crystal structures of JNJ4796 bound to H1/SI06 and H5/Viet HAs (Fig. 6, A to F). The A and E-rings of JNJ4796 have higher thermal mobility when bound to H5 as compared to H1 HA (Fig. 6D). The presence of Val⁴⁰ in H1 HA allows for a more stable interaction of the A-ring with the hydrophobic groove, whereas the corresponding Gln⁴⁰ in H5 with its longer and more flexible side chain does not interact as tightly, thereby increasing the dynamics of the bound A-ring that contributes to the higher thermal mobility of JNJ4796 (Fig. 6, A-D). The E-ring is relatively solvent exposed in both HA-JNJ4796 complexes, but is rotated ~180° in H5 compared to H1 HA (Fig. 5F). Another key difference pertains to the B-ring, where the pyridine moiety is displaced outward by ~ 0.4 Å in the H5- JNJ4796 complex, resulting in slightly weaker interactions compared to the H1- JNJ4796 complex (Figs. 5E and6E).

To rationalize the inability of JNJ4796 to bind with group-2 HAs, the apo structure of group-2 HA, H3N2 A/Hong Kong/1/1968 (H3/HK68) was compared with group 1 HA (H1/SI06) bound to JNJ4796. The key differences in the epitope of the small molecule on H1 HA with the corresponding region on H3 HA are the presence of a glycosylation site at HA1 Asn³⁸ in group 2 HAs, the orientations of His¹⁸ and Trp²¹ on HA1 and HA2 respectively, and substitution of Thr⁴⁹ and Val⁵² in helix-A of group-1 H1 to larger Asn⁴⁹ and Leu⁵² in group-2 H3 HAs (Fig. 6, G to I). The stem-targeting bnAbs FI6v3 and CR9114 are able to reorient the glycan and accommodate these helix-A differences to acquire pan-influenza reactivity, whereas bnAbs CR6261 and F10 are unable to do so and exhibit group-1 specificity. The different orientations of these antibodies on the HA surface also reflect their different breadths for group 1 and group 2 HAs (48). JNJ4796 probably also is unable to reorient the Asn³⁸ glycan and accommodate the helix-A mutations and, therefore, would experience steric clashes with group 2 H3 HA (Fig. 6, G to I) that could account for its group 1 specificity.

Conclusions

While antibodies are increasingly recognized as effective therapeutics, they may fail to achieve broad applicability in some settings due to inconvenience in administration and relatively high cost. Therefore, replacing antibodies with small molecules is desirable. Here we present proof-of-concept for antibody-guided, small molecule discovery. Starting from a well-characterized antibody with a desired activity profile, we selected and further improved a small molecule ‘antibody mimetic’ that recapitulates the antibody features in vitro and in vivo. We demonstrated the feasibility of targeting the conserved stem epitope on influenza HA with an orally bioavailable small molecule JNJ4796 with comparable high affinity and breadth of binding to bnAb CR6261. The compound mimics the key interactions observed in the antibody-HA co-crystal structures, inhibits the pH-sensitive conformational change of HA, neutralizes influenza viruses in vitro, and protects mice from lethal viral challenge. The success of this approach demonstrates the advantage of rigorously considering the targeted epitope-specific binding activity in close combination with the associated functional activity and mechanism of action, instead of focusing on potency alone. This strategy allows for generation of robust structure-activity relationships, thereby yielding a great level of control over the pharmacodynamic and pharmacokinetic properties of the selected ligand classes.

Material and Methods

Expression and purification of the hemagglutinin for binding and x-ray crystallography.

Hemagglutinin (HA) proteins for binding and crystallographic studies were expressed using a baculovirus system as described previously (49, 50). Briefly, each HA was fused with gp67 signal peptide at the N-terminus and to a BirA biotinylation site, thrombin cleavage site, trimerization domain, and His-tag at the C-terminus. The HAs were purified using metal affinity chromatography using Ni-NTA resin. For binding studies, each HA was biotinylated with BirA and purified by gel filtration chromatography. For crystallization studies, each HA was treated with trypsin (New England Biolabs, 5mU trypsin per mg HA, overnight at 4°C) to produce uniformly cleaved HA (HA1/HA2), and to remove the trimerization domain and His-tag. The cleaved material was purified by gel filtration.

Site-specific modification of the HA for conformational change inhibition assay.

HA constructs were designed with a Sortase A recognition motif (LPETG) between the trimerization domain and C-terminal His-tag to allow site-specific modification via Sortase A mediated transpeptidation (37). Expi293F cells (Thermo Fischer) were transfected with a pcDNA2004 mammalian expression plasmid encoding the modified HA protein, according to manufacturer’s instructions. The cell culture supernatant was harvested 7 days post-transfection. Soluble HA was purified from clarified supernatant via a three-step purification protocol: HisTrap Excel (GE Healthcare Life Sciences), HisTrap HP (GE Healthcare Life Sciences) and finally Superdex 200 Size Exclusion (GE Healthcare Life Sciences). Purified HA proteins were biotinylated using a peptide-based Sortase A recognition sequence, GGGGGK-Biotin (Pepscan). The labeling reaction was performed as described previously (51). Excess peptide and Sortase A were separated from the modified HA via Superdex 200 Size Exclusion (GE Healthcare Life Sciences). For the conformational change assay, HA0 was cleaved into HA1 and HA2 by incubation with trypsin (Trysin EDTA, Gibco) and the reaction was stopped by addition of Trypsin Inhibitor (Gibco).

Expression and purification of mini-protein HB80.4.

DNA encoding HB80.4-His was cloned into pET29b(+) (Novagen) and expressed using Rosetta (DE3) cells (Novagen) and auto-induction Magic Media (ThermoFisher). Bacteria were grown to OD600 ~0.6 at 37ᵒC and then at 25ᵒC overnight. Periplasmic extracts were obtained using BugBuster protein extraction reagent (EMD Millipore) according to the manufacturer’s specifications. HB80.4 was purified from periplasmic extract using HisTrap FF (GE Healthcare Life Sciences) followed by dialysis with PBS pH 7.4 (Gibco). Purity was determined by SDS-PAGE and was >95%.

Human antibodies and influenza viruses for assays.

Fully human IgG1 antibodies CR6261(7), CR9114 (6)and CH65 (52)were expressed and purified as described previously (8). Wild-type influenza viruses A/Brisbane/59/2007 (H1N1), A/California/07/2009 (H1N1), A/New Caledonia/20/1999 (H1N1), A/Puerto Rico/8/1934 (H1N1), A/Solomon Islands/3–2006 IVR-145 (H1N1), A/Hong Kong/156/1997 (H5N1), A/Vietnam/1194/2004 (H5N1), A/Brisbane/10/2007 (H3N2), H7 NY (7:1 reassortant virus with the HA of A/New York/107/2003 (H7N7) and the remaining segments from A/PR/8/34 and influenza B B/Brisbane/60/2008) were propagated and rescued in PER.C6® cells (53) by standard viral culture techniques.

AlphaLISA competition assay.

Small molecule (SM) compounds were dissolved at 5 mM in 100% DMSO and three-fold serially diluted in 100% DMSO nine times in 96-well, half-area microtiter plates. The compounds were further diluted 1:40 in assay buffer (PBS, 0.05% BSA, 0.05% Tween-20), and subsequently spun down for 15 minutes at 1000x g to separate any insoluble material. 10µl of the diluted compounds were incubated for 60 minutes with 10µl HA biotinylated with a Lightning Link kit (Innova Biosciences, 2.5 nM in assay buffer) in white 96-well, half-area untreated plates (PerkinElmer), after which 10 µl of His-labeled competitor protein HB80.4 (diluted in assay buffer) was added, followed by another 60-minute incubation, addition of 10µl of anti-His acceptor beads (PerkinElmer; 50 µg/ml in assay buffer), and a third 60-minute incubation. Finally, 10µl of streptavidin donor beads (PerkinElmer; 50 µg/ml in assay buffer) were added, followed by 60-minute incubation before the plates were read in a microplate reader at 615nm (Biotek Synergy Neo). Eight copies of samples containing no compound, with and without HA, were prepared for each plate and served as high and low controls, respectively. The sigmoidal inhibition curves were fitted with a robust four-parameter logistic model to derive IC₅₀ values. For the high throughput screen, only a single concentration of the SM compounds was tested (final concentration of 30 µM).

Truhit AlphaLISA counter assay.

Three-fold serially diluted SM compounds dissolved in 100% DMSO were screened using the TruHits kit (PerkinElmer) following the manufacturer’s instructions. The compounds were tested in the same dilutions as described in the AlphaLISA competition assay (assay buffer: PBS, 0.05% BSA, 0.05% Tween-20). Eight copies of samples containing no compound, with and without HA, were prepared for each plate and served as high and low controls, respectively.

Cell toxicity assay.

MDCK cells (ATCC CCL-34) were maintained in Dulbecco’s Modified Eagle Medium (DMEM, Gibco) supplemented with 10% fetal bovine serum and 2 mM L-glutamine at 37°C, 10% CO₂. On the day of the experiment, MDCK cells were seeded in 2x infection medium (DMEM, 1x L-glutamine, 6µg/ml trypsin-EDTA) at 25,000 cells/well (50µL) in white opaque 96-well plates (BD Falcon). The 96-well plates containing three-fold serially diluted compounds were further diluted 1:10 in incomplete medium (DMEM, 1x L-glutamine) and subsequently spun down (1000x g, 15 minutes). A volume of 12 µl was added to 48 µl incomplete medium in a fresh 96-well plate (96-well sterile cell culture plate, V-bottom, Greiner). Then, 50 µl of the compound solution was added to the MDCK plate followed by incubation for 96 hours at 37ᵒC, 10% CO₂. After incubation, cell viability was measured by adding 70 µl of ATPlite 1step reagent (PerkinElmer) to the wells. Luminescence was measured in a Biotek Synergy neo plate reader.

Virus neutralization assay (VNA).

Three-fold serial dilutions of the SM compounds were prepared as described above. On the day of the experiment, MDCK cells were seeded in 2x infection medium (DMEM, 1x L-glutamine, 6 µg/ml trypsin-EDTA) in white opaque 96-well plates at 25,000 cells/well. A 10 x pre-dilution of the SMs was prepared in incomplete medium (DMEM, 1x L-glutamine). CR6261 was diluted to 15 µM in incomplete medium and serially diluted 1:3 nine times. Subsequently, the SMs and CR6261 were spun down at 1000x g for 15 minutes. After spinning, 12µl supernatant was added to 48µl of the respective virus dilutions in incomplete medium in a 96-well plate. Two solvent controls, with and without virus, were setup for each sample. Then, 50 µl of the virus/SM mixture was added to the cells followed by incubation for 4 days at 37ᵒC under 10% CO₂. After incubation, 70 µl of ATPlite 1step (PerkinElmer) was added to the wells and the plate read for luminescence.

Conformational change inhibition assay (CCI).

Three-fold serial diluted SM compounds were prepared as described in the AlphaLISA competition assay above, and further diluted 1:100 in assay buffer (PBS, 1% BSA, 0.1% Tween-20). White, half-area, high-binding, 96-well plates (Perkin Elmer) were coated overnight with 50µl 0.5µg/ml streptavidin (Pierce) in PBS (Gibco). The plates were washed (150 µl PBS, 0.05% Tween-20) and blocked by exposing them to assay buffer for 1 hour. Subsequently, plates were incubated for 60 minutes with 50µl sortase biotinylated, trypsin-EDTA (Gibco) cleaved HA (0.1 µg/ml in assay buffer). After 60 minutes, the plates were washed and 50 µl of the diluted compounds were added to the plates for 60 minutes while shaking. Then 10 µl 1M acetate pH 5.25 was added to induce HA conformational changes, followed by 20 minutes incubation at room temperature on a shaking platform. The plates were washed followed by addition of 2.5mM DTT (diluted in PBS) to reduce any post-fusion HA and remove HA1. To detect the presence or absence of HA1, after 60 minutes incubation on a shaking platform, plates were washed followed by incubation with 0.5 µg/ml HA1 head-binding antibody CH65-HRP (labeled with Lightning Link HRP, Innova Biosciences) in assay buffer for 60 minutes. The plates were washed and 50 µl of POD substrate (Roche) was added followed by luminescence read out on a Biotek Synergy Neo plate reader. Total inhibition of the low-pH induced HA conformational change was achieved by using HA stem-binding bnAbs CR6261 or CR9114 (4 nM) as a positive control. Complete conformational change of the HA was achieved at low pH in presence of a non-binding IgG1 control (40 nM).

Kinetic solubility.

SM compounds were dissolved in DMSO to obtain 5 mM DMSO stock solutions. Stock solutions were diluted in duplicate into the required buffers (pH 4.0 and pH 7.4) to a final maximum concentration of 100 μM in 2% DMSO/buffer. Sample plates were gently shaken for 4 hours at room temperature. The samples were centrifuged and the supernatants were transferred to a new plate and diluted 1:1 with HCL/ACN for UPLC analysis. Reference and analyte samples were analyzed by UPLC/UV using a generic UPLC method. The measured solubilities are presented as mean values of duplicate determinations, with a maximum solubility threshold of 100 μM, and the lower limit of quantitation governed by the UV absorption properties of the compound.

Metabolic stability.

SM compounds (1 μM) were incubated with pooled human liver microsomes (BD Ultrapool; pooled male and female) and pooled mouse liver microsomes (male CD mice). Microsomes (final protein concentration 0.5 mg/ml), 0.1 M phosphate buffer pH 7.4 containing 1 mM MgCl₂ and SM compound (final substrate concentration 1 μM; final DMSO concentration 0.05 %) were pre-incubated at 37 °C prior to the addition of NADPH (final concentration 1 mM) to initiate the reaction. The final incubation volume was 500 μl. Three species-specific control compounds were included with each species. All incubations were performed singularly for each SM. At 6 time points (0, 5, 10, 20, 40 and 60 min), reactions were stopped by transferring 50 μl of the incubation mixture into methanol. SM compound concentrations were analyzed by LC- MS/MS, and the resulting data used to determine the half-life and intrinsic clearance of the compound in each species.

Plasma protein binding.

The free and bound fractions of the SM compound in mouse and human plasma was determined by rapid equilibrium dialysis (RED device, Thermo Fisher Scientific, Geel, Belgium). The RED device consists of a 48-well plate containing disposable inserts bisected by a semipermeable membrane creating two chambers. A 300 μl aliquot of plasma containing SM compound at 5 μM was placed one side and 500 μl of phosphate buffered saline (PBS) on the other. The plate was sealed and incubated at approximately 37 °C for 4.5 h. Samples were removed from both the plasma and buffer compartment and analyzed for the SM compound using a specific LC-MS/MS method to estimate free and bound concentrations.

In vivo pharmacokinetics.

The pharmacokinetic profiles of the SM compound was evaluated in fed male BALB/c mice (n = 3/group). Mice were i.v. injected with the SM compound at 2.5 mg/kg, formulated as a 0.25 mg/ml solution in 20% w/v hydroxypropyl-beta-cyclodextrin pH 6.5, and blood samples were collected from the saphenous vein at 0.08, 0.17, 0.5, 1, 2, 4, 7, and 24 hours into EDTA-containing microcentrifuge tubes. The compound was administered p.o. at 10 mg/kg, formulated as 1.33 mg/ml solution in 20% w/v hydroxypropyl-beta-cyclodextrin pH 6.5, and blood samples were collected from the saphenous vein at 0.5, 1, 2, 4, 7, 12 and 24 hours into EDTA-containing microcentrifuge tubes. The blood samples were immediately centrifuged at 4 °C and the plasma was stored at −20 °C. SM compound concentrations from the plasma samples were analyzed using LC-MS/MS. Individual plasma concentration-time profiles were subjected to a non-compartmental pharmacokinetic analysis (NCA) using Phoenix^™ WinNonlin version 6.3 (Certara, NJ, USA).

In vitro selectivity profiling.

The binding selectivity of the SM compounds at concentration of 10 µM was assessed in a panel of 52 radioactive ligand displacement assays (Eurofins Cerep SA) as per the providers’ validated protocols. Results were expressed as percent inhibition of control specific binding. Inhibition values higher than 50% were considered to represent significant effects.

Mouse influenza challenge.

Dosing formulations for the SM compounds were freshly prepared on the day before administration by dissolving the SM compound in cyclodextrin (40% hydroxypropyl-beta-cyclodextrin). The SM compound (7.5 ml/kg per animal) was administered twice daily per os (p.o.). Female BALB/cAnNCrl mice (Charles River, Sulzfeld, Germany) were intranasally infected with 2× 25 μL of 25× 50% mouse lethal dose (MLD₅₀) or mouse sublethal dose (MLD₉₀) of H1N1 A/PR/8/34 dissolved in sterile phosphate buffered saline (D-PBS). All experiments were in accordance with the general principles governing the use of animals in experiments of the European Communities (Directive 2010/63/EU) and Dutch legislation (The Revised Experiments on Animals Act, 2014). This included licensing of the project by the Central Committee on Animal Experimentation and approval of the study by the Animal Welfare Body.

Viral neutralization assay in HBEC cultures.

MucilAir human bronchial epithelial cells (HBEC) (pool of 14 donors) (Epithelix Sàrl) were delivered and maintained at an air-liquid interface according to the manufacturer’s instructions. Each MucilAir insert contained ~500,000 well-differentiated respiratory epithelial cells consisting of ciliated cells, mucus-producing goblet cells, and basal cells. Prior to the start of the experiment, inserts were washed once with PBS (with Ca⁺⁺ and Mg⁺⁺) to remove mucus and cell debris. Cells were infected with H1N1 A/Puerto Rico/8/1934 at an MOI of 0.1 and concomitantly treated with a JNJ4796 at different concentrations. Cells were treated both at the apical and basolateral side of the tissue culture. After 1 hour of incubation, virus and compounds administered to the apical compartment were removed to reconstitute the air-liquid interface, whereas cells remained exposed to JNJ4796 through the basolateral compartment. Sham-treated (PBS supplemented with DMSO at a final concentration identical to compound-treated samples) and mock-infected wells, were taken along as positive and negative controls, respectively. For each experimental condition, 4 biological replicates were included. 96 hours post-infection, apical washes (D-PBS, 200µl/insert) of the epithelium were used to determine the amount of released viral RNA by quantitative reverse transcriptase polymerase chain reaction (qRT-PCR).

Trypsin susceptibility (TS) assay.

In the TS assay, 5 μM H1/PR8 HA was pre-incubated separately with 25 μM of JNJ4796 and 10 μM CR6261 Fab for ~30 minutes at room temperature. Control reactions were incubated with 2% DMSO. The pH of each reaction was lowered using 1M sodium acetate buffer (pH 5.0). One reaction was retained at pH 7.4 to assess digestion at neutral pH. The reaction solutions were then thoroughly mixed and incubated for about 30 minutes at 37°C. After incubation, the reaction solutions were equilibrated at room temperature and the pH was neutralized by addition of 200 mM Tris buffer, pH 8.5. Trypsin-ultra^™ (NEB Inc.) was added at final ratio of 1:50 by mass and the samples were digested for about 40 minutes at 37°C. After incubation with trypsin, the reaction solutions were equilibrated at room temperature and quenched by addition of non-reducing SDS buffer and boiled for 2 minutes at 100°C. All samples were analyzed by 4–20% SDS-PAGE gel and imaged using BioRad ChemDoc^™ imaging system.

Surface plasmon resonance (SPR).

All SPR experiments were performed using a Biacore T200 instrument operating at 25°C. Biotinylated HA was covalently immobilized on a streptavidin-coated, carboxymethylated dextran sensor surface (SA chip, GE Healthcare). JNJ4796 was dissolved at 10mM in 100% DMSO and then diluted in the running buffer (20 mM PBS, 137 mM NaCl, 0.05% P-20 surfactant, pH 7.4 (GE Healthcare), supplemented with 2% DMSO). Binding constants were obtained from a series of injections of JNJ4796 from 0.1 nM to 1 µM with a flow rate of 30 µl/min. Data from single-cycle kinetics were analyzed using BIAevaluation software. Base lines were adjusted to zero for all curves, and injection start times were aligned. The reference sensorgrams were subtracted from the experimental sensorgrams to yield curves representing specific binding followed by background subtraction (i.e. double-referencing). Binding kinetics was evaluated using a 1:1 binding model (Langmuir) to obtain association rate constants (k_a) and dissociation rate constants (k_d). Binding affinity (K_D) was estimated from the concentration dependence of the observed steady-state responses.

Thermodynamic binding profile.

Biotinylated HA was purified using a spin-column to remove excess biotin and covalently immobilized on a streptavidin-coated, carboxymethylated dextran sensor surface (SA chip, GE Healthcare). JNJ4796 was dissolved at 10 mM in 100% DMSO and diluted in the running buffer (20 mM PBS, 137 mM NaCl, 0.05% P-20 surfactant, pH 7.4 (GE Healthcare), supplemented with 2% DMSO). Binding constants were obtained from a series of injections of JNJ4796 from 0.1 nM to 10 µM in 5 half-log serial dilutions, individually injected at the following temperatures: 10°C, 15°C, 25°C, 35°C, and 40°C, at a flow rate of 30 µl/min. Experiments were performed in duplicate (for H1/NCa) or triplicate (for H1/Bri) to ensure reproducibility. Data from single-cycle kinetics were analyzed using BIAevaluation software (GE Healthcare). Base lines were adjusted to zero for all curves, and injection start times were aligned. The reference sensorgrams were subtracted from the experimental sensorgrams to yield curves representing specific binding followed by background subtraction (i.e. double- referencing). Binding kinetics was evaluated using a 1:1 binding model (Langmuir) to obtain association rate constants (k_a) and dissociation rate constants (k_d). Binding affinity (K_D) was estimated from the concentration dependence of the steady-state responses observed. Changes in thermodynamic parameters enthalpy (ΔH) and entropy (ΔS) were calculated from the slope and intercept, respectively, of the temperature dependence of the dissociation constant using the van’t Hoff approximation: lnK_D = −ΔH ∕ R∙T + ΔS ∕ R, where R is the gas constant and T is the absolute temperature. The binding free energy, ΔG, was derived from the Gibbs-Helmholtz equation (ΔG = ΔH − T∙S).

Crystallization and structure determination of the JNJ4796-HA complexes.

Gel filtration fractions containing H1N1 (A/Solomon Islands/3/2006) (H1/SI06) and H5N1 A/Vietnam/1203/2004 (H5/Viet) HAs were concentrated to ~10 mg/mL in 20 mM Tris, pH 8.0, and 150 mM NaCl. Compound JNJ4796 at ~5 molar excess was incubated with the HAs for ~1 hour at room temperature and centrifuged at 14,000g for ~2–3 min. before setting up crystallization trials. Crystallization screens used the sitting drop vapor diffusion method with our automated CrystalMation robotic system (Rigaku) at TSRI. Within 3–7 days, diffraction quality crystals had formed in 0.2 M disodium hydrogen phosphate, 20% w/v PEG3350 at 20 C (for H1/SI06) and 0.2 M lithium sulfate, 20% w/v PEG3350 at 4 C (for H5/Viet). The resulting crystals were cryoprotected with 5–15% ethylene glycol, flash cooled, and stored in liquid nitrogen until data collection. Diffraction data were collected at 100 K on the Stanford Synchrotron Radiation Lightsource beamline 12–2 and processed with HKL-2000 (54). Initial phases were determined by molecular replacement using Phaser (55, 56) with HA models corresponding to PDB codes 1RU7 (for H1/SI06) and 4FQI (for H5/Viet). Refinement was carried out in Phenix (57), and alternated with manual rebuilding and adjustment in COOT (58). The final coordinates were validated using MolProbity (59). Data collection and refinement statistics are summarized in table S6.

Structural analyses.

Surface areas buried on the HA upon binding of JNJ4796 were calculated with the Protein Interfaces, Surfaces and Assemblies (PISA) server at the European Bioinformatics Institute (60). MacPyMol (DeLano Scientific) was used to render structure figures.

Statistics.

Inhibition potencies in AlphaLISA, viral neutralization and conformational change inhibition assays were determined by robust four-parameter logistic (4PL) regression routines in SPSS (IBM) or R (www.R-project.org). The inflection point (C value) of the curve is taken as the IC50 value. A transform-both-sides (TBS, square root) approach, robust regression techniques (including Huber’s M) and the optional inclusion of the low and high control values as anchor points were used to stabilize the variance, down weigh outliers, and to fit incomplete curves, respectively. For reasons of presentation, results and curves were scaled from 0 to 100% that represent the range between the robust averages of the low and high control values. For the H1N1 A/PR/8/1934 in vivo challenge studies, the differences in survival (as compared to vehicle control) were tested using a 2-sided Fisher’s exact test. The significance of the differences was adjusted for multiple testing by Holm-Bonferroni adjustment for the compounds at their highest dose (within the two studies in which the compounds were investigated), followed by a step-wise approach for lower doses. Statistical analysis of bodyweight was based on Area Under the Curve (AUC) analysis. For this analysis, the last observed bodyweight was carried forward if a mouse died during follow-up of the study. Briefly, the weight per mouse at day 0 was used as baseline and weight change was determined relative to baseline. The AUC was defined as the summation of the area above and below the baseline. Differences in net AUC’s were tested in a one-way Analysis of Variance (ANOVA) with the same test strategy and post hoc adjustment as mentioned for survival. Statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., USA) and SPSS version 20 (SPSS Inc., USA). Statistical significance level was set at α=0.05. For studies performed on HBEC cultures, results were statistically analyzed by Anova between the log titers and the (categorized) concentrations. The significance is based on the contrasts between the values of each of the three highest concentration with the aggregated results for the concentrations in the lower plateau.