Structural analysis reveals TLR7 dynamics underlying antagonism

Toll-like receptor 7 (TLR7) recognizes both microbial and endogenous RNAs and nucleosides. Aberrant activation of TLR7 has been implicated in several autoimmune diseases including systemic lupus erythematosus (SLE). Here, by modifying potent TLR7 agonists, we develop a series of TLR7-specific antagonists as promising therapeutic agents for SLE. These compounds protect mice against lethal autoimmunity. Combining crystallography and cryo-electron microscopy, we identify the open conformation of the receptor and reveal the structural equilibrium between open and closed conformations that underlies TLR7 antagonism, as well as the detailed mechanism by which TLR7-specific antagonists bind to their binding pocket in TLR7. Our work provides small-molecule TLR7-specific antagonists and suggests the TLR7-targeting strategy for treating autoimmune diseases.

T LR7 recognizes pathogenic single-stranded RNAs (ssRNAs) and plays a crucial role in the innate immune response to viral infections, such as human immunodeficiency virus 1 and influenza virus 1,2 . Our previous structural studies of TLR7 elucidated the homo-dimeric activated form of TLR7 3 . The activated TLR7 dimer contains two distinct ligandbinding sites. Small-molecule agonists, including imiquimod, guanosine and 2',3'-cGMP bind to site 1 in the dimerization interface and directly induce the formation of the activated TLR7 dimer, while uridine-containing ssRNAs bind to site 2 and allosterically enhance the binding affinity of site 1 ligands for TLR7 3,4 .
TLR7 is strictly localized to the endosomal compartment to avoid undesirable endogenous RNA-mediated activation 5,6 . In fact, excessive activation of TLR7 is considered to be involved in the pathogenesis of several autoimmune diseases such as systemic lupus erythematosus (SLE) [7][8][9][10] . In contrast, the participation in SLE of other members of the TLR7 family, TLR8 and TLR9, remains unclear or controversial 11,12 . These findings suggested that selective targeting of TLR7 would be useful for the treatment of SLE. Several TLR7 inhibitors, including oligonucleotides and antimalarial agents, have been reported 13,14 . Antimalarial agents, represented by hydroxychloroquine (HCQ), inhibit TLR7 and also TLR9 via indirect mechanisms that do not involve direct binding to TLRs 14 . To date, potent and selective TLR7 inhibitors have not been discovered.
We present here the development of potent TLR7-specific antagonists via the modification of TLR7 agonist. Using cryoelectron microscopy (cryo-EM) combined with X-ray crystallography, we reveal that the structural equilibrium between open and closed conformations that underlies TLR7 antagonism. In addition, high-resolution cryo-EM structure enables us unambiguously to visualize an antagonist binding sites inside the openform structure of TLR7 and the recognition mechanism of TLR7 antagonists at the atomic level. Finally, we examine the therapeutic efficacy of Cpd-7 in lupus-prone NZB/W F1 mice, which shows that oral administration of this promising compound can protect mice against deadly autoimmunity.

Results
Design of TLR7 antagonists based on its agonist. To identify TLR7-specific antagonists, we assumed that a TLR7-specific agonist could serve as a scaffold for development of TLR7-specific antagonists. We have previously reported a series of 8-oxoadenine derivatives  represented by Cpd-1, Cpd-2, and Cpd-3 as specific TLR7 agonists [15][16][17][18] . Therefore, we used these compounds as a starting point for further development (Fig. 1a). We confirmed the binding mode of 8-OAs by crystallography (Supplementary Table 1). These compounds bound to TLR7 site 1 (Fig. 1b) and the overall structures of the complex were similar to those of known TLR7/agonist complexes 4 Fig. 1a-c). The structure-activity relationships (SAR) of the 8-OAs indicated that both 6-amino and 8-oxo groups were important in TLR7 agonism 19 (Fig. 1a), but the 8-oxo group was too far to interact with TLR7 protein, resulting in a space in which other substituents might be introduced (Fig. 1b). We hypothesized that the substitution of the 8-oxo group in 8-OAs would convert an agonist into an antagonist. We synthesized Cpd-5 with an 8-pyridyl group, showing low solubility (0.0003 mg/ml at pH 7.4 buffer) and modest TLR7 inhibitory potency (IC 50 = 5.0 μM) comparable to HCQ without causing any inhibition of either TLR8 or TLR9 and activation of TLR7 (Figs. 1a, 1c and Supplementary Fig. 2). We further optimized the C2 side chain and 8-pyridyl moiety of Cpd-5, and also introduced an amine unit into the para-position of the 9-phenyl group in order to enhance solubility and potency with reference to our SAR of 8-OAs 18 (Fig. 1a). After these optimizations, Cpd-6 displayed potent TLR7 inhibitory activity (IC 50 = 25 nM) with a high selectivity for TLR7 over TLR8 and TLR9 ( Fig. 1c and Supplementary Fig. 2). We also synthesized a 6-methyl derivative, Cpd-7, based on the structure of Cpd-6. This modification increased the inhibitory activity ( Fig. 1d and Supplementary Fig. 2; IC 50 = 15 nM). Both Cpd-6 and Cpd-7 showed good solubility (>0.15 mg/ml at pH 7.4 buffer). The dose-response relationship between the inhibitors and R848 showed that both Cpd-6 and Cpd-7 acted as antagonists of TLR7 ( Fig. 1d and Supplementary Fig. 3).

(Supplementary
We also conducted the inhibition assays in primary human peripheral blood mononuclear cells (PBMCs) in which plasmacytoid DCs, but not other immune cells, produce interferon-α (IFN-α) in TLR7 dependent manner 20 . The result showed that both Cpd-6 and Cpd-7 completely blocked R848-induced interferon-α (IFN-α) production at 2.5-μM concentration (Fig. 1e), while neither Cpd-6 nor Cpd-7 blocked the TLR2dependent TNF-α production from human PBMCs at the same concentration ( Supplementary Fig. 4), suggesting that these inhibitors are not cytotoxic and specifically inhibit TLR7 response in human primary immune cells. These results demonstrate that TLR7 antagonist activity of Cpd-6 and Cpd-7 can be confirmed both in NF-κB/SEAP reporter gene assay and the assay using PBMC.
Structural analysis showed TLR7 dynamics upon antagonist binding. To understand the mechanism of action of our developed TLR7-specific antagonists, we conducted biochemical and structural studies using the extracellular domain of TLR7 from monkey. ITC analysis confirmed the direct binding of Cpd-6 and Cpd-7 to TLR7 and that the binding depends on an acidic pH (Fig. 2a). The critical pH near 5.0 coincides well with the pH inside mature endo-lysosomes where TLR7 resides. We next conducted a crosslinking assay to monitor ligand-induced TLR7 dimerization and performed its quantitative analysis (Fig. 2b). The potent agonist Cpd-3 induced almost complete dimerization under both neutral and acidic conditions. TLR7 in the absence of ligands remained mostly monomeric, but the extent of dimerization in the presence of Cpd-6 and Cpd-7 increased only in acidic conditions, demonstrating that direct binding of Cpd-6 and Cpd-7 to TLR7 induced dimerization in an acidic condition.
To our surprise, crystallographic study of the TLR7/Cpd-6 complex revealed that this complex adopted an activated dimeric structure and Cpd-6 bound to site 1 with a recognition mode very similar to that of Cpd-1 (Fig. 2c, Supplementary Fig. 1d and Supplementary Table 1). The C8 substitution group in Cpd-6 showed weak electron densities, indicating the high flexibility of this part (Fig. 2c). The static crystal structure showed one aspect of the binding mode, but its inhibitory mechanism remained elusive. We thus decided to take advantage of electron microscopy to capture the possible conformational dynamics associated with TLR7 antagonism. Crosslinked dimeric TLR7 proteins prepared with or without the addition of ligands were used for EM analysis (Supplementary Figs. 5 and 6). Although cryo-EM analyses of crosslinked TLR7/Cpd-3, TLR7/Cpd-6 and TLR7/Cpd-7 complexes suffered from a severe preferred orientation problem, leading to anisotropic resolution, the resulting 3D reconstitution clearly showed different conformations of TLR7 induced by different ligands (Supplementary Figs. 7-9 and  Supplementary Table 2). The TLR7/Cpd-3 complex exclusively adopted a closed-form conformation which is in good agreement with the crystal structure (Fig. 2d, Supplementary Figs. 1 and 7). Surprisingly, the 3D-reconstruction of TLR7/Cpd-6 yielded two major types of TLR7 dimer ( Fig. 2e and Supplementary Fig. 8): one was the closed form which was also observed in the crystal structure, and the other was the open form, identified by this study, in which two TLR7 promoters were separated from each other at the bottom of their ring structures. In contrast, Cpd-7 induced only the open form of TLR7, which was well fitted to that observed with the TLR7/Cpd-6 complex ( Fig. 2f and Supplementary Fig. 9). Negative-staining electron microscope analyses also showed similar results ( Supplementary Fig. 10).   The binding mode of Cpd-1 to TLR7 site 1 in its activated form is shown on the left. TLR7 is shown in ribbon-and-stick representations. Two TLR7 protomers in the dimeric structure are painted salmon and aquamarine, respectively. Dashed lines indicate hydrogen bonds. On the right is a schematic view of Cpd-1 binding to site 1. The space inside the dimerization interface and the key derivation part of Cpd-1 are indicated by a red ellipse. c Dose-response curves for the inhibition of human TLR7 activity by Cpd-5, Cpd-6, Cpd-7 and HCQ (positive control) measured by reporter gene assay with R848 (200 nM) as stimulating agent. Data are representative tests with three replicates. d Dose-response relationship between Cpd-7 and R848 measured by human TLR7 reporter gene assay. TLR7-expressing cells were incubated with Cpd-7 (0-100 μM), together with 0-1000 μM of R848. Data are representative tests with three replicates. e TLR7 inhibition assay in primary human PBMCs. PBMCs were pretreated by DMSO, Cpd-6 or Cpd-7 for 3 h before 20 h stimulation by DMSO (none) or R848. Secreted IFN-α levels quantified by ELISA were shown. Dot plots of the data with mean (n = 3 per group) are shown. . We speculated that the inducing of some heterogeneity to the recombinant proteins would help to improve the particle orientation problem. Actually, the distribution of particle orientation was much improved by the use of TLR7 proteins with partially uncleaved purification tags (Supplementary Fig. 12), leading to a high quality cryo-EM map (Supplementary Fig. 13). This TLR7/Cpd-7 complex is also covalently stabilized using glutaraldehyde crosslinker, but no densities of tags and glutaraldehyde were observed. We were able to build a complete model of a C2-symmetric open form of the TLR7 dimer with two molecules of Cpd-7 unambiguously assigned within the protein dimerization interface ( Supplementary Fig. 13). The open-and closed-form structures have totally different dimerization interfaces ( Supplementary Fig. 14). The distances between the C-termini of two protomers were about 69 Å and 36 Å in the open and closed forms, respectively (Fig. 3c). The larger spatial separation of C-termini in the open form conformation prevents the two intracellular TIR domains from dimerizing and activating TLR7, exhibiting an inhibitory effect. A defined antagonist binding site is spatially close to the site 1 in the closed form (Fig. 3d), but the binding mechanism differs considerably. TLR7 utilize the common platform for both agonist and antagonist binding on one side of the interface, while the other side is completely different due to a conformational Based on our structural analyses and the results of biomedical and in silico studies, we can summarize the mechanism of TLR7 ligand action (Fig. 3g). Cpd-1, Cpd-2, Cpd-3 and other known TLR7 agonists induce a closed form structure which activates TLR7. The antagonist Cpd-6 induces TLR7 conformational dynamics between the open and closed forms, and inhibits TLR7 activation. The more potent antagonist Cpd-7 predominantly or exclusively induces an open form structure in TLR7 and displays stronger inhibition than Cpd-6. Efficacy of Cpd-7 in treating lupus-prone NZB/W F1 mice. Finally, we evaluated the efficacy of Cpd-7 in treating lupus-prone NZB/W F1 mice (Fig. 4a). We verified the pharmacokinetic profile of Cpd-7 in NZB/W F1 mice following oral administration of a single 100 mg kg −1 dose, in addition to its inhibitory activity against mouse TLR7 and TLR9 ( Fig. 4a and Supplementary  Fig. 16). These results confirmed that Cpd-7 is an orally available TLR7-selective inhibitor (IC 50 = 0.192 μM). Preliminary pharmacokinetic simulations predicted that Cpd-7 would maintain an unbound concentration higher than the IC 50 values for 20 h with twice daily dosing at 100 mg kg −1 . Next, we assessed the effect of therapeutic interventions. Cpd-7 was administered orally once or twice daily at 100 mg kg −1 for four weeks to female NZB/W F1 mice with a urinary albumin/creatinine ratio over 300 mg g −1 , which marked the onset of glomerulonephritis. Treatment with Cpd-7 resulted in a significant decrease in proteinuria in a dosedependent manner compared to vehicle control, and prevented death in all mice that received either dose (P < 0.05) (Fig. 4b, c). Histological analysis of the kidneys confirmed that both treatment groups exhibited no glomerular lesions, whereas vehicle control displayed glomerulonephritis accompanied by dilated tubules, regenerating tubules, hyaline casts and perivascular infiltration of mononuclear cells. (Fig. 4d). Moreover, proliferation of the mesangium, an increase in glomerular cellularity, capsular adhesions and/or glomerular sclerosis were observed in the affected glomeruli. Consequently, glomerular lesion scores in the administration group were significantly reduced in a dosedependent manner compared to those in the vehicle control ( Fig. 4e). Although no toxicological evaluation was performed in the efficacy study, no clear toxic signs in the body weights or clinical signs were observed in both treatment groups.

Discussion
To date, various small-molecule TLR7 agonists have been synthesized and many of them are under clinical trials for treating viral diseases such as hepatitis B and influenza and several types of cancer 21 . In contrast, few small-molecule TLR7specific antagonist has been reported in literature. Since blocking of TLR7 is considered as a promising strategy for treating autoimmune diseases including SLE and psoriasis [22][23][24] , there is an urgent need for the development of TLR7-specific antagonist. In this study, starting with a TLR7-specific agonist, we successfully developed TLR7-specific antagonists. Cpd-6 and Cpd-7 with IC 50 at nano-molar concentrations showed much higher inhibitory activity compared to the indirect TLR inhibitor HCQ. These two compounds also displayed high TLR7 selectivity over both TLR8 and TLR9 which can be considered as TLR7-specific antagonists. Surprisingly, the crystal structure of TLR7/Cpd-6 complex adopted the closed-form conformation and Cpd-6 was observed inside the agonist binding site, while this complex formed both open-and closed-form conformations with comparable particle numbers in cryo-EM analysis. Meanwhile, the stronger antagonist, Cpd-7, induced a dominant open-form conformation in cryo-EM analysis. We speculate that the closed form of TLR7 induced by antagonist is not sufficient for subsequent signal transduction due to the rapid structural equilibrium.
Superimposing of the TLR7 open-form conformation onto the inactive TLR8 dimer structure also revealed that the two TLR7 protomers were apart from each other more than the TLR8 protomers, demonstrating that these open-form conformations exhibit an inhibited form. Both structures showed the similar interface area with~1300 Å 2 ( Supplementary Fig. 17) 25 .
To solve the structure, we used the protein with partially cleaved tag, and introduced the cross-linking. No densities of tags and glutaraldehyde were observed, probably because the conformation of the tag and crosslinked residues would not be rigid, only part of the proteins contained the tag, and/or all the residues would not be crosslinked. The overall conformation of the partially tag-cleaved TLR7 protein A tag is the same as the tagcleaved TLR7. Moreover, the TLR7 structures of the protomer are mostly the same both in X-ray crystallography (tag-cleaved, no crosslinking) and cryo-EM (partially-cleaved, cross-linking). Thus, introduction of the crosslinking and the protein A tag has no serious effect on conformations of TLR7.
The agonist-antagonist conversion was achieved through the introduction of a bulky substituent in the C8 position in 8-OAs (Fig. 1a). The effect of this modification can be interpreted in two different ways, leading to a shift in equilibrium toward the open conformation: in one scenario the bulky substituent destabilizes the activated dimer (Fig. 1b), while in the other the formation of extensive contacts with the antagonist binding pocket stabilizes the open conformation (Fig. 3e). Cpd-6 induces both open and closed forms of TLR7 (Fig. 2e), therefore antagonism of Cpd-6 would result from a rapid equilibrium between the two forms (Fig. 3g). Because the 6-amino group in 8-OAs is critical for TLR7 activation by interacting with D555* (Figs. 1b and 2c), as observed in all the agonist-bound TLR7 structures, the substitution of the 6-amino group (in Cpd-6) with a 6-methyl group (in Cpd-7) subverts the binding of Cpd-7 to the closed-form conformation, thus shifting the equilibrium toward the open conformation and thereby conferring a more potent inhibitory activity on Cpd-7.
Finally, we showed that oral administration of Cpd-7 was effective in the treatment of severe autoimmunity in a mouse model, and potentially should be useful in treating SLE. Our work provides a comprehensive analysis of the mechanism of action of TLR7 antagonists and a potential for TLR7-targeted SLE therapy and would be an incentive for future research and development.
Expression and purification of TLR7. The recombinant TLR7 proteins expressed in Drosophila S2 cells (Thermo Fisher Scientific) using Monkey (Macaca mulatta) TLR7 extracellular domain expression construct with N167Q, N399Q, N488Q and N799Q mutations and cleavable Z-loop and C-terminal protein A tag were prepared and were stored in a buffer containing 10 mM Tris-HCl pH 7.5 and 150 mM NaCl at concentrations of 10-15 mg ml −1 . Briefly, the purification steps included IgG affinity purification, the Z-loop and the protein A tag cleavage, and gelfiltration chromatography. Glycan-trimmed TLR7 proteins, expressed in the presence of 1.5 mg l −1 kifunensine (Glycosyn) and treated by Endo Hf (New England Biolabs) were used for crystallization. Glycan-untrimmed proteins were used in ITC analysis, crosslinking assay, negative-staining EM and cryo-EM (200 kV microscopes) analyses. Glycan-trimmed TLR7 proteins with partially-uncleaved protein A tag were used for cryo-EM analysis using Tatan Krios ( Supplementary  Fig. 11).
Diffraction data sets were collected with a wavelength of 1.0000 Å on beamlines PF BL5A or PF-AR NE3A (Ibaraki, Japan) under cryogenic conditions at 100 K. Crystals were soaked into cryoprotectant solutions containing 15-20% glycerol, and were flash cooled in the cryostream or liquid nitrogen. The diffraction datasets were processed with XDS 27 . Phases were determined by molecular replacement method using the program MOLREP 28 . Coordinates of monkey TLR7 in complex with R848 (PDB ID: 5GMH [10.2210/pdb5GMH/pdb]) were used as the search model. Models were refined with stepwise cycles of manual model building using the COOT program 29 and restrained refinement using REFMAC5 program 30 and phenix.refine program 31 . The qualities of the final structures were validated with the PDB validation server (http://wwpdb-validation.wwpdb.org/). Coordinates and the structure factor of all data were deposited in the Protein Data Bank. The statistics of the crystallographic data collection and refinement, and deposited PDB IDs were summarized in Table S1. Structure representations were generated in Pymol 32 .
ITC analysis. ITC experiments (Fig. 2a) were conducted at 298 K in a final buffer condition of~30 mM citrate-NaOH pH 4.7 or pH 5.7 (Hampton Research), 150 mM NaCl by using a MicroCal iTC200 (Malvern Panalytical). Solutions containing 1 mM Cpd-6 or Cpd-7 were titrated into solutions containing 80 µM TLR7 proteins, using a titration sequence of a single 0.4 µl injection followed by 18 injections, 2 µl each, with 180 s intervals between injections. OriginLab software (OriginLab) was used to analyze the raw ITC data. Thermodynamic parameters were extracted from curve fitting analysis with a single-site binding model.
Glutaraldehyde crosslinking assay. For glutaraldehyde crosslinking assay (Fig. 2b), proteins were buffer exchanged into a buffer containing 10 mM HEPES-NaOH, pH 8.0, 150 mM NaCl and were concentrated to~13 mg/ml using Amicon Ultra devices (Merck). Protein concentrations were adjusted to 25 µM using buffers containing 10 mM HEPES-NaOH, pH 7.0, 150 mM NaCl (for neutral condition) or 50 mM citrate-NaOH pH 5.0, 150 mM NaCl (for acidic condition). Cpd-3, Cpd-6 or Cpd-7 were added to final concentrations of 100 or 1000 µM. The reactions were started by adding glutaraldehyde (Wako) to a final concentration of 5 mM (for neutral condition) or 20 mM (for acidic condition), incubated at ambient temperature for 25 min and quenched by adding Tris-HCl pH 7.5 to a final concentration of 100 mM. Crosslinked samples were analyzed by non-reducing Crosslinked sample preparation for EM analyses. For preparation of crosslinked TLR7/Cpd-3 complex for negative-staining EM (Supplementary Fig. 10a) and cryo-EM analyses (Fig. 2d), 25 µM TLR7, 1 mM Cpd-3 and 20 mM glutaraldehyde were mixed in 50 mM citrate-NaOH pH 5.0, 150 mM NaCl buffer and reacted for 30 minutes at ambient temperature. The reaction was quenched by adding 100 mM glycine-HCl pH 5.0. The crosslinked sample was further purified using a repacked Superdex 200 Increase column (GE Healthcare) in 10 mM citrate-NaOH, pH 5.0, 150 mM NaCl buffer ( Supplementary Fig. 5). The dimer fractions were pooled and concentrated to~0.6 mg/ml using Amicon Ultra devices (Merck). Samples were flash cooled in liquid nitrogen and stored at −70°C for further use.
For preparation of crosslinked TLR7 (no ligand), TLR7/Cpd-6 and TLR7/Cpd-7 complexes (Supplementary Fig. 5) for negative-staining EM ( Supplementary  Fig. 10b-d) and cryo-EM analyses (Fig. 2e, f), 25 µM TLR7 in the absence or presence of 2 mM Cpd-6 or Cpd-7 and 20 mM glutaraldehyde were mixed in 50 mM citrate-NaOH pH 5.0, 150 mM NaCl buffer and reacted for 30 min at ambient temperature. The reactions were quenched by adding 100 mM glycine-HCl pH 5.0. The crosslinked samples were further purified using a Superdex 200 Increase column (GE Healthcare) in 10 mM citrate-NaOH, pH 5.0, 150 mM NaCl buffer (Supplementary Fig. 6). The dimer fractions were pooled and concentrated to~2 mg/ml using Amicon Ultra devices (Merck). Samples were flash cooled in liquid nitrogen and stored at −70°C for further use.
For preparation of crosslinked TLR7/Cpd-7 complex for cryo-EM analysis using Titan Krios (Fig. 3), 30 µM TLR7 with partially-uncleaved protein A tag ( Supplementary Fig. 10a, pooled #1), 2 mM Cpd-7 and 22.5 mM glutaraldehyde were mixed in 50 mM citrate-NaOH pH 4.7, 150 mM NaCl buffer and reacted for 30 minutes at ambient temperature. The reactions were quenched by adding 100 mM glycine-HCl pH 5.0. The crosslinked sample was further purified using a Superdex 200 Increase column (GE Healthcare) in 10 mM citrate-NaOH, pH 4.8, 100 mM NaCl buffer ( Supplementary Fig. 11d). The dimer fractions were pooled and concentrated to~1.4 mg/ml using Amicon Ultra devices (Merck). Samples were flash-cooled in liquid nitrogen and stored at −70°C for further use.
Negative-staining EM data acquisition and image processing. The crosslinked TLR7/Cpd-3, TLR7 (no ligand), TLR7/Cpd-6 and TLR7/Cpd-7 complexes samples were diluted to 100 nM using 20 mM citrate-NaOH, pH 5.0, 150 mM NaCl buffer. A 3-µl aliquot of samples was adsorbed for 30 s onto glow discharged holey grids coated with thin carbon (ELS-C10 STEM Cu100P, Okenshoji). After blotted off the excess sample, the grid was stained with 3 µl 2% (w/v) uranyl acetate for three times, and finally air-dried. The stained grids were analyzed using a JEM-2010F microscope (JEOL) at the University of Tokyo running at 200 kV equipped with a 4 K×4 K CCD camera. Data collection was operated at 60,000× magnification and −0.5~−1.5-µm defocus range. For each sample, 40~60 CCD images with pixel size of 1.5Å were collected. CTF parameters estimation, automated particle picking, and 2 cycles of 2D classification were performed using cisTEM software 33 ( Supplementary Fig. 10).
Cryo-EM data acquisition. The crosslinked TLR7/Cpd-3, TLR7/Cpd-6, TLR7/ Cpd-7 complexes samples for 200 kV microscope experiments (Fig. 2d-f) were diluted to 0.6-0.7 mg/ml using 10 mM citrate-NaOH pH 4.8, 100 mM NaCl buffer. The crosslinked TLR7/Cpd-7 complex with partially-uncleaved protein A tag sample for 300 kV microscope experiments (Fig. 3) was diluted to~0.8 mg/ml using 10 mM citrate-NaOH pH 4.8, 100 mM NaCl, 1 mM Cpd-7 buffer. A 3-µl aliquot of samples was applied to a glow-discharged Quantifoil grid (R1.2/R1.3 300 mesh, copper or copper/rhodium), blotted for 4-5 s in 100% humidity at 4-6°C and plunged into liquid ethane using a Vitrobot MkIV (Thermo Fisher Scientific). The Cryo-EM analysis was initially performed by using a Talos Arctica microscope (Thermo Fisher Scientific) at 200 kV with a Falcon3EC direct electron detector at the Cryo-EM facility in KEK (Ibaraki, Japan). Data collections of TLR7/Cpd-3 complex (Fig. 2d and Supplementary Fig. 7) and TLR7/Cpd-6 complex ( Fig. 2e and Supplementary Fig. 8) were performed by using a Glacios microscope (Thermo Fisher Scientific) at 200 kV with a Falcon3EC direct electron detector in the counting mode at RIKEN RSC Cryo-EM facility (Hyogo, Japan). Movie stacks were acquired using 150,000× magnification with an accumulated dose of 50.0 electrons per Å 2 over 40 frames. The pixel size was 0.67 Å. The data were automatically acquired by the beam-image shift method using the EPU software. Data collection of TLR7/Cpd-7 complex (Fig. 2f, Supplementary Fig. 9) was performed by using a Talos Arctica microscope (Thermo Fisher Scientific) at 200 kV equipped with a Gatan Quantum-LS Imaging Filter (GIF) and a Gatan K2 Summit direct electron detector in the electron counting mode at the Cryo-EM facility in the University of Tokyo (Tokyo, Japan). Movie stacks were acquired using 130,000× magnification with an accumulated dose of 50.0 electrons per Å 2 over 40 frames. The pixel size was 1.03 Å. The data were automatically acquired by the beam-image shift method using the SerialEM software 34 . The high resolution data collection of TLR7/Cpd-7 complex with partially-uncleaved protein A tag sample ( Supplementary Fig. 12) was performed by using a Titan Krios G3i microscope (Thermo Fisher Scientific) at 300 kV equipped with a Gatan Quantum-LS Imaging Filter (GIF) and a Gatan K3 direct electron detector in the electron counting mode at the Cryo-EM facility in the University of Tokyo (Tokyo, Japan). Movie stacks were acquired using 105,000× magnification with an accumulated dose of 60.0 electrons per Å 2 over 60 frames. The pixel size was 0.83 Å. The data were automatically acquired by the beam-image shift method using the SerialEM software 34 .
Cryo-EM data processing and model building. Cryo-EM data processing were performed using RELION 3.0 35 . Raw movies stacks were motion-corrected using MotionCor2 36 or RELION's own implementation 35 . The CTF parameters were determined using the CTFFIND4 program 37 . Data processing workflows for TLR7/ Cpd-3 complex (940 movie stacks from Glacios), TLR7/Cpd-6 complex (2,988 movie stacks from Glacios), TLR7/Cpd-7 complex (2,547 movie stacks from Talos Arctica), TLR7/Cpd-7 complex with partially-uncleaved protein A tag (4,464 movie stacks from Titan Krios) are summarized in Supplementary Figs. 6, 7, 8, and 11, respectively. The final resolution was estimated by gold-standard Fourier shell correlation (FSC) between the two independently refined half maps (FSC = 0.143). Because the first three data suffered from severe particle orientation preference which led to anisotropic resolution, we did not build structural models against these maps. In the TLR7/Cpd-6 complex dataset analysis, two distinct forms of TLR7 dimer were gained after the first round of 3D classification and each of them was refined separately (Supplementary Fig. 8). In the Titan Krios data processing, a total of three rounds of 3D classification generated a particle package containing 301,212 particles, which yielded an overall resolution of 2.8 Å ( Supplementary  Fig. 12). For model building against the EM map, coordinates of two TLR7 monomers (PDB ID: 5GMH [10.2210/pdb5GMH/pdb]) were docked into the final map using Chimera 38 . The model was then manually refined using the COOT program 29 . Real-space refinement in Phenix program 39 was performed after manual refinement.
The cryo-EM maps have been deposited in the Electron Microscopy Data Bank. The atomic coordinates of the TLR7/Cpd-7 complex structure have been deposited in the Protein Data Bank. The statistics of EM data processing and refinement were summarized in Supplementary Table 2. The local resolution maps were produced with the ResMap program 40 . Structure representations were generated in the Chimera 38 , ChimeraX 41 , Pymol 32 and COOT 29 programs.
Binding energy calculations. To calculate binding energies of Cpd-1, Cpd-2, Cpd-3, Cpd-6 and Cpd-7 to TLR7, the complex structures (TLR7Cpd-6 complex (closed form, crystal structure) and TLR7/Cpd-7 complex (open form, cryo-EM structure)) were processed using the protein preparation wizard 42 in Maestro 43 . Structures of compounds were modeled from the complex structures. Then, all complex structures were minimized using the OPLS3e force field 44 . The binding energies of Cpd-1, Cpd-2, Cpd-3, Cpd-6 and Cpd-7 to open and closed forms of TLR7 were calculated using the MM-GBSA method 45 with the VSGB solvation model 46 and the OPLS3e force field 44 .
Mice. Female NZBWF1/Slc (NZBW F1) mice over 25 weeks of age (Japan SLC Inc., Japan) were used in this study. All mice were maintained under specific pathogen-free conditions. All experiments were approved by the Institutional Animal Care and Use Committee at Sumitomo Dainippon Pharma.
Pharmacokinetic profiling. Cpd-7 was orally administered to 35-week-old female NZB/W F1 mice at 100 mg kg −1 dissolved in 0.5 w/v% Methyl Cellulose 400 Solution (WAKO, Japan). Blood was collected over 24 h from the isofluraneanesthetized mice via the jugular vein. The plasma samples were separated by centrifugation. Cpd-7 was extracted from the plasma samples by protein precipitation with methanol and measured using LC/MS/MS (HPLC: SHIMADZU CORPORATION, Japan; MS: AB Sciex Pte. Ltd., Japan). The lower limit of quantification was set at 5 ng/mL.
Animal study. Cpd-7 was dissolved in 0.5 w/v% Methyl Cellulose 400 Solution (WAKO) at 10 mg/ml. NZBW F1 mice were orally treated with Cpd-7 for 4 weeks. At the initiation of treatment, mice were randomly divided according to bodyweight and proteinuria level (urinary albumin/creatinine ratio over 300 mg g −1 ) and were orally once or twice daily. In treatment, mice were divided into 3 groups (n = 6 per group): vehicle control group (Methyl Cellulose 400 Solution, 10 µl g −1 body weight of mice), Cpd-7 once treated group (Q.D.: 100 mg kg −1 , day), and Cpd-7 twice treated group (B.I.D.: 100 mg kg −1 , twice day), and the bodyweight and proteinuria level were monitored weekly. During the treatment, nearly dead individuals were euthanized, then sera and kidneys were collected. In this case, the last known values for urinary protein was carried forward. No clear toxic signs in the body weights or clinical signs were observed in both treatment groups.
Quantification of urine albumin. Urine protein was quantified by automatic analyzer (BioMajesty ZERO(JCA-ZS050), JEOL Ltd, Japan) with Mouse Urinary Albumin Assay Kit (S-type) (AKRAL-021S, Shibayagi Co.,Ltd., Japan). Urine creatinine levels were measured in the same samples. The urine albumin excretion rate is expressed as the ratio of albumin to creatinine. Time course of proteinuria were analyzed by Dunnett's multiple-range test.
Histopathological analysis. The both kidneys were fixed in 10% neutral-buffered formalin, embedded in paraffin, sectioned, stained with hematoxylin and eosin (HE) and periodic acid-Schiff (PAS), and examined by light microscopy. Glomerulonephritis was scored on a 0-4 scale, based on the intensity and extent of histopathological changes 47 . A grade of 0 was given to kidney without glomerular lesions; a grade of 1 was given when there was minimal thickening of the mesangium; a grade of 2 was given when there was a noticeable increase in both mesangial and glomerular capillary cellularity; a grade of 3 was given when the preceding conditions were observed, along with superimposed inflammatory exudates and capsular adhesions; and a grade of 4 was given when the obliteration of the glomerular architecture included 70% of glomeruli. These classifications were used to grade 20 glomeruli within one area, and the mean glomerular histopathological score was calculated for each mouse. Data were analyzed by Tukey's test.
Human TLR7 reporter gene agonistic activity assays. Compounds were dissolved in dimethyl sulfoxide. Before stimulation with compound, Human TLR7 reporter cell line (Novus Biological) was seeded in 96-well microtiter plate (AGC techno glass) at 5 × 10 4 cells/well in DMEM (GIBCO) with 10% fetal bovine serum (GIBCO), and 10 ug/ml Blastcidin S HCl (GIBCO) and incubated at 37°C in 5% CO 2 . The plates were incubated over-night and then stimulated with various concentrations of compound and R848. The plates were then incubated an additional 20 ± 1 h at 37°C in 5% CO 2 and then the secreted embryonic alkaline phosphatase (SEAP) activity was measured as activation of human TLR7. The SEAP activity was evaluated as follows: pNPP (Invitrogen, USA) was added to the incubated sample (50 μl/well); after 15 min, 4 M sodium hydroxide solution (nacalai tesque, Japan) was added thereto (50 μl/well) to quench the reaction; and the absorbance of each sample was measured at 405 nm with microplate reader SpectraMax M4 (Molecular Devices, USA). The 50 % effective concentration (EC 50 value) of each compound was calculated based on 100 % of the SEAP activity wherein the test sample comprises no compound. Analyses of dose-response relationship. Compounds were dissolved in dimethyl sulfoxide. Human TLR7 reporter cell line (Novus Biological) were seeded in 96well microtiter plate (AGC techno glass) at 5 × 10 4 cells/well in DMEM (GIBCO) with 10% fetal bovine serum (GIBCO), and 10 μg ml −1 Blastcidin S HCl (GIBCO) and incubated at 37°C in 5% CO 2 . The plates were incubated over-night and then addition with various concentrations of compound. After 30 min, R848 (human TLR7/8 ligand) that was diluted with the medium was added, wherein final concentration was adjusted to 0.01, 0.03, 0.1, 0.3, 1, 3, 10, 30, 100, 300 and 1000 μM. The plates were then incubated an additional 20 ± 1 h at 37°C in 5% CO 2 , and then the SEAP activity was measured as activation of human TLR7. The SEAP activity was evaluated as follows: pNPP (Invitrogen) was added to the incubated sample (50 μl/well); after 15 min, 4 mol l −1 sodium hydroxide solution (nacalai tesque) was added thereto (50 μl/well) to quench the reaction; and the absorbance of each sample was measured at 405 nm with microplate reader SpectraMax M4 (Molecular Devices).
Mouse TLR reporter gene assays. Compounds were dissolved in dimethyl sulfoxide. The day before transfection, mouse TLR7 genestably-expressing HE293 cell line (293XL-mTLR7 cell, InvivoGen, USA) and mouse TLR9 genestablyexpressing HEK293 cell line (293-mTLR9 cell, InvivoGen) were seeded in collagencoated 6-well plate (AGC techno glass) at 3 × 10 5 cells/well in DMEM (GIBCO) with 10% fetal bovine serum (GIBCO), and 10 μg ml −1 Blastcidin S HCl (GIBCO) and incubated at 37°C in 5% CO 2 . The plates were incubated over-night and then cotransfected with 1 μg (mouse TLR7) or 0.3 μg (mouse TLR9) of pNF-kB-Luciferase reporter in a 10:1 ratio with 4 μl of FuGene6 transfection reagent (Promega corporation, USA) following the manufacturer's instructions. The plates were incubated 24 h. The cells re-seeded into 96-well plate (AGC techno glass) at 2 × 10 4 cells/well in DMEM (GIBCO) with 10% fetal bovine serum (GIBCO), and 10 μg ml −1 Blastcidin S HCl (GIBCO) and then inhibition with various concentrations of compound. HCQ (mouse TLR7/9, Acros Organics, United Kingdom) was used as positive control. After 30 min, R848 (mouse TLR7/8 ligand) or CpG1826 (mouse TLR9 ligand, Hokkaido System Science Co., Ltd.) added to each well, wherein each final concentration were adjusted to 200 nM R848 (mouse TLR7) or 100 nM CpG1826 (mouse TLR9). The plates were then incubated an additional 6 h at 37°C in 5% CO 2 , and then the luciferase activity was measured as activation of mouse TLR7 and 9. The luciferase activity was evaluated as follows: Bright-GloTM Luciferase Assay System (Promega) was added to the incubated sample (100 μl/well); after 2 min, the luminescence intensity of each sample was measured with a luminometer (Envision). The IC 50 value of each sample compound was calculated based on 100 % of the luciferase activity wherein the test sample comprises no sample compound.
Solubility measurements. Compound was dissolved in dimethyl sulfoxide at 10 mM. A 1.75% disodium hydrogen phosphate aqueous solution and a 5.53% citric acid aqueous solution were mixed while monitoring with a pH meter to prepare an isotonic buffer solution having a pH of 7.4. The dimethyl sulfoxide solution and the isotonic buffer solution of pH 7.4 was mixed and shaken at 110 rpm for 90 min at 25°C by a constant temperature incubator shaker. After standing for 16-20 h, the mixture was centrifuged to precipitate insoluble matter, and the supernatant was collected. Separately, standard solutions of the compound were prepared by dissolving the compound in a mixture of acetonitrile and water (1:1). The analytical sample and the standard solutions were measured by UHPLC-UV (Waters, USA). The degree of solubility was calculated by comparing the peak area of the analytical sample with standard solutions.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials and Source Data. Additional data related to this paper may be requested from the authors.