RNA and Imidazoquinolines are sensed by distinct TLR7/8 ectodomain sites resulting in functionally disparate signaling events

Abstract

Toll-like receptors (TLR) 7 and 8 are pattern recognition receptors controlling antiviral host defense or autoimmune diseases. Apart from foreign and host RNA, synthetic RNA oligoribonucleotides (ORN) or small molecules of the imidazoquinoline family activate TLR7 and 8 and are being developed as therapeutic agonists. The structure-function relationships for RNA ORN and imidazoquinoline sensing and consequent downstream signaling by human TLR7 and TLR8 are unknown. Proteome- and genome-wide analyses in primary human monocyte-derived dendritic cells here showed that TLR8 sensing of RNA ORN vs. imidazoquinoline translates to ligand-specific differential phosphorylation and transcriptional events. Additionally, TLR7 and 8 ectodomains were found to discriminate between RNA ORN and imidazoquinolines by overlapping and non-overlapping recognition sites to which murine loss-of-function mutations and human naturally occurring hyporesponsive polymorphisms map. Our data suggest TLR7 and TLR8 can signal in two different ‘modes’ depending on the class of ligand. Considering RNA ORN and imidazoquinolines have been regarded as functionally interchangeable, our study highlights important functional incongruities whose understanding will be important for developing TLR7 or 8 therapeutics with desirable effector and safety profiles for in vivo application.

Introduction

The detection of invading microorganisms by vertebrates involves Toll-like receptors (TLRs), a family of pattern recognition receptors (PRR) (1). TLRs recognize diverse structural classes of microbe-associated molecules. For example, TLR7 and TLR8 sense bacterial and viral RNA, and TLR9 detects bacterial and viral nucleic acids containing CpG motifs, respectively. TLR7 and 8 detect ligands present in endosomes by their extracellular domain (ECD), whilst their cytoplasmic Toll/IL-1-receptor (TIR) domains relay an intracellular signal via the adaptor MyD88, IL-1 receptor-associated kinases (IRAKs) and TNF receptor-associated factor (TRAF) 6 (1-3). Depending on the cell type, this ultimately leads to the activation of mitogen-activated protein kinases (MAPKs), NF-κB and/or interferon regulatory factor (IRF) transcription factors (1). For instance, in human plasmacytoid dendritic cells (pDC) TLR7 signaling induces cell maturation, differentiation and type I IFN production via IRF5 and 7. In B lymphocytes, TLR7 activation triggers IL-10 and IL-6 production, antigen-specific stimulation and enhanced immunoglobulin G class switch DNA recombination required for humoral immune responses (1). Human monocytes and polymorphonuclear leukocytes (PMN) exclusively express TLR8 and respond to its activation with TNF production and/or degranulation (4). The molecular basis for these disparate TLR7 or TLR8-mediated signaling events remains elusive. The majority of results regarding TLR7 or TLR8 have been generated not using natural RNA but rather synthetic G/U-rich RNA oligoribonucleotides (ORN; MW ~6000, 60-80 Å length), or low molecular weight molecules of the imidazoquinoline family (5). Both types of ligands are able to trigger TLR7 and/or TLR8 and, due to their ability to induce type I IFN, are considered antiviral and antitumor therapeutic molecules. For example, imiquimod (also known as R837, trade-name Aldara; MW 240 Da, 9 Å), a human and murine TLR7 agonist, has been licensed for the treatment of genital warts and malignant tumors of the skin (6). Another imidazoquinoline, R848 (also known as Resiquimod; MW 314 Da, 9 Å), is a dual TLR7 and TLR8 agonist. Imidazoquinolines are being actively being pursued for therapeutic use (5). Both classes of TLR7 and/or TLR8 ligands, namely RNA ORN and imidazoquinolines, have been considered functionally congruent and interchangeable since so far no significant differences in the outcome of TLR7 or TLR8 activation by RNA ORN vs. R848 have been reported. Thus R848 has frequently been used as mimic for viral RNA detection via TLR7 or TLR8 in various functional studies (e.g. Ref. (7)).

Recognition of both RNA ORN and imidazoquinolines involves the TLR7 or 8 ECD which consist of ~25 tandem leucine-rich repeat (LRR) motifs forming a horseshoe-shaped solenoid (3). Some TLRs, including TLR7 and TLR8, contain ‘irregular’ LRRs with amino acid insertions thought to protrude from the LRR backbone (3). In TLR8 and 9, these insertions were recently shown to play a role in receptor function (8-10). Nucleic acid sensing by TLR3 and TLR9 involves N- and C-terminal charged binding patches (9, 11, 12). On the sequence level, TLR7 and TLR8 are most closely related to TLR9 and TLR3; hence an RNA recognition mechanism similar to TLR9 and TLR3 appears likely. The small size of imidazoquinolines has led to the proposal that their recognition by TLR7 and TLR8 may be altogether different to that of RNA ORN and involve insertion of a single imidazoquinoline molecule into a small binding interface (13). However, until recently, no systematic studies addressing the recognition principles for RNA ORN and imidazoquinolines by TLR7/8 have been conducted.

Given the therapeutic importance of RNA ORN- and imidazoquinolines as TLR7 or 8 agonists, we sought to determine functional outcomes of TLR8 activation by RNA ORN vs. R848 in terms of gene transcription and phospho-proteomics in primary human monocyte-derived DCs (MoDCs). These potent antigen-presenting cells play an important role as sentinels of infection (14) and as antiviral or cancer vaccines (15). We observed striking differences in transcription and phosphorylation patterns in MoDC stimulated with RNA ORN vs. R848. These were further analyzed in terms of ligand recognition of both classes of ligands combining 3D structural modeling, model-guided mutagenesis and the study of naturally occurring murine and human TLR7 mutations/variants, respectively.

Methods

Reagents and cells

Chemicals and cell culture reagents were from Sigma, unless otherwise stated. HEK293T cells from A. Dalpke (University of Heidelberg, Germany) were cultured as described (10). Monocyte-derived Dendritic Cells (MoDC) with >97% purity (CD11c+ DC-SIGN+) were prepared (16) from fresh buffy coats (UK national blood service, informed consent obtained) by density gradient centrifugation, CD14 magnetic bead purification (Miltenyi) and culture with IL-4 and GM-CSF (Peprotech) for 5 days. MoDC were then stimulated with RNA40 (pre-complexed using LyoVec, Invivogen) and R848 (Invivogen or Axxora). For experiments in HEK293T cells RNA40 ORN (IBA) were complexed with DOTAP (N-[1-(2,3-Dioleoyloxy)]-N,N,N-trimethylammonium propane methyl sulfate; Roche). The following antibodies were used: anti-HA (Sigma and Cell Signaling), anti-V5 (Invitrogen), anti-AU1 (Bethyl), anti-β-tubulin (Sigma), anti-rabbit-HRP (Vector Laboratories) and anti-mouse-HRP (Promega). Anti-phospho-antibodies were from Cell Signaling Technologies, p-ERK (3179), p-NF-κB (3039), p-MEK (9154), STAT3 (4904), and STAT6 (9362).

Plasmids and site-directed mutagenesis

Untagged hTLR7 and hTLR8 expression plasmids were from A. Dalpke (University of Heidelberg, Germany) and used to generate hTLR7/8-HA, hTLR7-YFP and hTLR7/8-Renilla, further information on request. Rab7a-CFP and ER-CFP were kind gifts from H. Huseby (Institute of Cancer Research and Molecular Medicine, The Norwegian University of Science and Technology) and H. Stanmark (Department of Biochemistry, The Norwegian Radium Hospital, Oslo, Norway), respectively. Point mutations introduced according to the QuikChange II XL Kit (Stratagene) were verified by DNA sequencing and back-cloned into the original backbones to avoid unwanted mutations. A list of all mutants, and PCR and mutagenesis primer sequences on request.

Reporter gene experiments

HEK293T cells in 24-well plates (7.5 × 10ˆ4 cells/well) were transfected using CaPO₄ (13) with a firefly luciferase reporter (Stratagene), Renilla luciferase control reporter (pRL-Tk), and pC1-EGFP (BD Clontech) and 24h later stimulated with 12 μg/ml DOTAP-complexed RNA40 or indicated concentrations of R848 (usually 0.75 μg/ml for TLR7 and 2.5 μg/ml for TLR8) for 18 h. Luciferase activities were determined using the Dual Luciferase Reporter Assay System (Promega) on a Fluostar Optima Instrument (BMG Labtech). Mean values of triplicates (+/- SD) are shown throughout.

Microarray analysis

For Microarray analysis cells were prepared as above and stimulated 4 hours, lysed and mRNA extracted using a microRNeasy (Qiagen) kit. RNA quality was verified by Bioanalyser 2100 (Agilent). Samples from five separate donors were submitted for microarray analysis (Agilent; G3 gene-expression chip). TIFF image files were processed using Agilent’s Feature Extraction software (v10.7.3.1) using protocol GE2_107_Sep09. Data was imported into GeneSpring and Log(2) ratios calculated as Cy3(sample)/Cy5(reference). ANOVA was used to compare the three groups (RNA40, R848 and control) for differentially expressed genes and, at the same time, a fold change analysis conducted. The Benjamini-Hochberg multiple testing correction was applied and a corrected p-value of <0.01 taken to be significant. All analysis was carried out in GeneSpring GX v11.5.

Quantitative PCR

RNA was extracted as above, DNA removed by incubation with DNAse I (Ambion) and reverse transcribed using High Capacity RNA-to-cDNA Kit (ABI). qPCR was carried out on a 7500 Fast real time PCR (ABI), using 50 ng of template, TaqMan gene expression master mix (ABI), and the following TaqMan probes: INF-b1 (Hs00277188_s1), IL-6 (Hs00985639_m1), DDX58 (Hs00204833_m1), IL8 (Hs00174103_m1), CXCL1 (Hs00236937_m1), IL10 (Hs00961622_m1), HHEX (Hs00242160_m1. All gene expression value were normalized against GAPDH (Hs99999905_m1) and 2ˆ-ΔC_T values plotted, each dot representing one donor and means indicated by horizontal bars. Alternatively, fold ratios were calculated over media for each ligand. In each experiment a minimum of seven donors were analyzed. LyoVec twice the amount used to complex RNA40 (condition ‘RNA40’) was used as an additional control in Fig. 1B and was non-significantly different to ‘Media’ for all targets. In Fig. 1C, differences in fold ratios (RNA-R848) over ‘media’ were plotted for RNA40 vs. R848. Donors with a difference <0 are shown as red indicating higher induction by R848, donors with a difference >0 as blue indicating higher induction by RNA40.

Phosphoproteomic screen

For protein studies, unless otherwise stated, 5×10ˆ6 MoDC from each biological replicate were stimulated with 5 μg of either RNA40, R848 or Lyovec control in 1 ml of media for 10 min at 37 °C. The stimulation was terminated by addition of ice cold HBS PI (20mM HEPES, 150mM NaCl, 1% Phosphatase inhibitor cocktail I (Sigma Aldrich), pH 7.4). Cells were washed twice in ice cold HBS PI and then lysed using the CHAPS lysis buffer from the QIAGEN Phosphoprotein purification kit with additional Phosphatase inhibitors as above. Cells were lysed for 40 minutes at 4 °C, cell debris removed by 30 min centrifugation at 16,000 g and the equal quantities (as quantified by Bradford assay)of protein run over a phosphoprotein purification column (QIAGEN). Eluates showed comparable protein amounts as assessed by Bradford assay. For immunoblot analysis using phospho-specific Abs, equal amounts were loaded for SDS-PAGE. For MS analysis, phospho-enriched samples from three biological replicates were cleaned by methanol:chloroform precipitation, and trypsinized. Tryptic digests were desalted and subjected to LC-MS/MS analysis using a Thermo LTQ Orbitrap Velos. Samples were randomized and analyzed in triplicates using Gas Phase Fractionation. Peptides were detected and quantified with Progenesis LC-MS software (version 3.1.4003.30577) using default settings (no de-convolution/de-isotoping, 200 most intense MSMS peaks). A merged peak list generated by Progenesis LC-MS was searched against the IPI human database (v.3.80, 86719 entries) using Mascot (http://www.matrixscience.com/) v2.3.01. Significantly regulated proteins were identified by LIMMA analysis conducted in the R software package.

Confocal cell imaging

HEK293 cells seeded to eight well chamber slides (Lab-Tek, Nunc) and transfected with FuGENE HD (Promega) for 16 h cells and then media exchanged to imaging media (Gibco). Live cell imaging was conducted by using an inverted laser scanning microscope (LSM 780, Zeiss, Germany), equipped with an Axio Observer Z1. Images were sequentially digitally acquired and further analyzed by using the Axiovision software. Further settings on request.

RNA40 pull-down assay and Immunoblot

HEK293T cells were transfected as before with 10 μg indicated plasmids or empty vector (pcDNA3.1[+]) and 1 μg pC1-EGFP. Forty-eight hours later, cells were scraped in lysis buffer containing protease and phosphatase inhibitor mixtures (Roche). Protein concentrations were adjusted and cleared by centrifugation. For the RNA40 pull-down assay 800 μl cell lysates were incubated with 15 μg 3’-Biotin-RNA40 (IBA) at 4°C and 2.5 ng/ml recombinant V5 control protein for 2 h. Subsequently 30 μl Streptavidin-agarose beads (Pierce) were added for 2.5 hours. Beads were washed 3 times in lysis buffer, boiled and equal volumes analyzed on 3–8% Tris-Acetate gradient gels (Invitrogen). Blocked membranes were probed using anti-HA, anti-AU1, or anti–β-tubulin Abs (Sigma-Aldrich) and anti-mouse or anti-rabbit HRP conjugates (Promega and Vector laboratories, respectively). For anti-phospho immunoblot, analysis was conducted using PVDF membrane and antibodies as decribed above. For confirmation of phospho-screen data western blots were run on phospho-enriched samples. Time course experiments were carried out on whole cell lysates, generated using the same lysis buffer. Visualization was carried out using ECL reagents (Pierce). Blots in Fig. 3E, F were quantified using ImageJ relative to WT TLR7 or TLR8.

RNA40 pull-down assay combined with Renilla measurement

HEK293 cells were prepared as described for the Ligand-precipitation by using the passive lysis buffer (Promega) for lysis and all washing steps but instead of TLR7-HA TLR7 fused to Renilla luciferase was transfected. The washed RNA40-biotin-coated Streptavidin beads remained in 50 μl of lysis buffer and were measured for bound luciferase activity (Promega) in a white microplate, in parallel with 50 μl reference whole cell lysates (raw luciferase activity)using a FLUOstar OPTIMA luminescence plate reader (BMG Labtech). Binding was calculated as the bound:raw ratio for each transfection.

Sequence and structural files, homology modeling, comparison of model and TLR8 crystal structure

At the time of initiating this study, structural information on the hTLR7 (NP_057646) and hTLR8 (NP_619542) ECD was not available. Therefore, 3D models were generated by homology modeling (15) based on the related TLR3 ECD structures 2a0z and 1ziw as recently done successfully for TLR9 (12, 13). GROMACS molecular dynamics, quality analysis tools (ANOLEA, VERIFY_3D and ERRAT), N-glycan analysis (GlyProt server), visualization/analysis tools (SwissPBD Viewer and PyMol) and software for the computation of surface charges were employed as described in Ref. 15 and citations therein. All mutations were selected on the basis of these models. For TLR8, a crystal structure of the ECD was recently published (17). Although the curvature of the TLR8 ECD model and this crystal structure differ (see below), the spatial organization of sensing sites is highly congruent. For example, site 2 residues D543, F568 and H566 are surface-exposed and in close proximity to one another and the R848 ligand. The differences in curvature can be explained by the use of molecular dynamics to sterically and energetically optimize the TLR8 ECD model, leading to a more open, ‘relaxed’ conformation, compared to the TLR8 ECD crystal structure which did not undergo molecular dynamics simulation. Similar differences were observed for TLR3 (12). We consider both crystal structure and molecular dynamics-optimized models to reflect the conformational space assumed by TLR ECD. Thus spatial distances for site 1-site 2 distances may be approximations varying in distance around 60-70 Å in both model and crystal, which matches the estimated dimension of the agonist molecules investigated here.

Polymorphism information, analysis and genotyping

A list of reported SNPs in the human TLR7 gene (Gene ID: 51284), was obtained from NCBI dbSNP at www.ncbi.nlm.nih.gov (as of 2009). rs numbers were as follows and the number of submissions or individual dbSNP submitter IDs are given in ( ): TLR7 Q11L rs179008 (multiple submissions), V222D rs55907843 (5 submitters), A448V rs5743781 (multiple submissions), N576D rs34501186 (ss43702473), F580S rs35160120 (ss43684376), Q599H rs36076482 (ss43784597), M603I rs55835602 (ss86353189, ss76868495), S610C rs36110053 (ss43920081), S620T rs34729893 (ss43674426), R627I rs34014664 (ss43741106), L634* rs34557368 (ss43945191) and R636T rs35337229 (ss43637315). Genotyping in different study collectives was carried out as described in Ref. (10).

Mice experiments

N-ethyl-N-nitrosourea mutagenesis, macrophage isolation, TLR stimulation and TNF measurement by an L-929 cell cytotoxicity bioassay based on propidium iodide absorption (18). Tlr7^rsq1 TLR7^T68I, Ref. (19)) and Tlr7^rsq2 (TLR7^N182Y, this study) mice were generated on a pure C57BL/6J background, and maintained as a homozygous/hemizygous stock. C57BL/6J mice were obtained from The Scripps Research Institute breeding colony. All animal procedures were in accordance with institutional animal care guidelines.

Data analysis and statistical testing

Experimental data were analyzed using MS Excel 2007 or GraphPad Prism 5. For luciferase assays, p-values were determined using the Student t-test by comparing to the respective wild-type condition. For Microarray and proteomics data significance testing ANOVA was used, for qPCR validation the Wilcoxon matched pair signed rank sum test. * denotes p < 0.05, ** p < 0.01

Results

R848 and RNA ORN differ in the transcriptional events triggered via the same receptor, TLR8

Both synthetic RNA ORN and imidazoquinolines have thus far been considered functionally interchangeable as in pDC R848 and RNA ORN both induce IFNα via TLR7, or TNF via TLR8 in monocytes (1). Given the fundamental differences in size and shape between the two classes of ligands we sought to compare cellular responses elicited through the same receptor (here TLR8) in primary human immune cells. Primary human primary monocyte-derived DCs (MoDC) express TLR8, but neither contain TLR7 mRNA (Fig. 1A) nor respond to Imiquimod, a strict TLR7 agonist (Fig. 1B). Comparison of the prototypical imidazoquinoline R848 and the 20mer synthetic RNA ORN RNA40 (20) in an initial microarray analysis of stimulated MoDC revealed that R848 significantly regulated 1140 genes with a bias towards a pro-inflammatory and pro-apoptotic response, compared to 240 for RNA40 biased towards antiviral response genes (see Fig. 1C, 1D and Table S1). Only 185 genes were regulated by both (Fig. 1C). Validation of selected genes by qPCR analysis (Fig. 2) at equimolar ligand concentrations further corroborated the differences between RNA40 and R848. In terms of relative mRNA levels (Fig. 2A, right graphs), CXCL1, IL6 and IL8 were strongly up-regulated by R848 but significantly less by RNA40. IL10 and HHEX were strongly regulated by R848 but hardly by RNA40. Conversely, RNA40 induced IFNb1 and DDX58 more strongly than R848 (Fig. 2B, right graphs). This was strikingly illustrated by a donor-by-donor analysis of mRNA fold changes (Fig. 2A, B, left graphs). Blockage of IFNb1, IL6, IL8 and IL15RA gene induction by bafilomycin, an inhibitor of the endosomal ATPase, ruled out involvement of cytosolic RNA sensors or cell surface TLR but confirmed the role of an endosomal TLR, i.e. TLR8 (data not shown). Summarized clearly by the plotted fold differences of gene induction (Fig. 2C), the gene profiles induced by RNA40 vs. R848 differ despite acting via the same receptor (TLR8) and at equimolar ligand concentrations.

Figure 1. In MoDC R848 and RNA ORN trigger different transcriptional profiles exclusively via TLR8.

(A) qPCR of basal expression levels of different PRR in MoDC (left) and PBMC (right). TLR7 is highly expressed in PMBC but not in MoDC. (B) qPCR analysis of cytokine induction upon stimulation with Imiquimod, a strict TLR7 agonist, in comparison to RNA40 shows unresponsiveness to TLR7 stimulation in purified MoDC but not PBMC. In (A) and (B) relative expression is shown as 2ˆ(-ΔC_T) values where each dot represents a different biological replicate. Means +/-SEM are also shown. Overview (C) and specific list of genes (D) significantly (p<0.01) regulated in a microarray analysis of MoDCs stimulated with RNA40 or R848 (see Methods and Table S1). Heat maps show antiviral, inflammatory and apoptotic gene sets identified by pathway analysis (DAVID). RNA40 and R848 elicit an antiviral signature, but RNA40 lacks the strong pro-inflammatory and pro-apoptotic signature induced by R848.

Figure 2. R848 and RNA ORN disparately induce specific immune-related genes.

qPCR analyses in individual donors (n=7) show that whereas (A) CXCL1, IL6, IL8, IL10, HHEX are more strongly regulated by R848 (red), (B) IFNb1 and DDX58 are more strongly regulated by RNA40 (blue) at equimolar concentrations. Relative expression is shown as 2ˆ(-ΔC_T) values (left graphs) or fold changes over ‘Media’ (right graphs). As HHEX is downregulated (<1), fold changes are inverted (1/mRNA fold). Each dot represents one donor. (C) Differences in fold ratios plotted against each other. Donors showing a positive difference (more strongly RNA40-regulated) are shown in blue, those with negative difference (R848-regulated) in red throughout. Differences were tested by Wilcoxon matched-pairs signed rank test.

In primary human monocyte DC R848 and RNA trigger different phosphorylation events

We hypothesized that different transcriptional outcomes might be due to differences in upstream events, e.g. i) phosphorylation of signaling mediators, ii) ligand trafficking and/or ii) recognition via the receptor ECD. To address the first possibility, we conducted a global phospho-proteomics analysis in MoDC to identify proteins whose phosphorylation status changed most significantly within 10 minutes of R848 or RNA40 treatment. Using three biological replicates, each run in triplicate allowed the use of statistics (see Methods). Under stringent correction for multiple comparisons, disparate phosphorylation events were identified between the two ligands. Whereas R848 caused differential phosphorylation of 32 proteins, only 6 were significantly modified for RNA40 (Table S2). Even among the three proteins regulated by both stimuli, the classical TLR-associated signaling proteins ERK1/2, MEK1/2 and STAT3, R848 stimulation augmented their phosphorylation, whereas RNA40 caused de-phosphorylation (Fig. 3A). These proteomics results were confirmed by immunoblot analysis of the phospho-enriched lysates fraction of stimulated MoDCs (Fig. 3B). Here, STAT3 showed R848-induced phosphorylation but no de-phosphorylation in response to RNA40. A time course analysis of ERK1/2 and MEK1/2 phosphorylation confirmed that RNA40 induced dephosphorylation of both ERK1/2 and MEK1/2 compared to control between 10 min and 1 hour following stimulation, whereas both were strongly phosphorylated by R848 up to 30 min post stimulation (Fig. 3C). These findings were further validated by phospho-flow cytometry (AL, unpublished data). As a control, NF-κB was phosphorylated by both stimuli, albeit with a slightly delayed time course for RNA40 (Fig. 3C). Thus, our data suggest differences in protein signaling between RNA40 and R848 that may contribute to the observed transcriptional differences.

Figure 3. R848 and RNA induce differential downstream signaling factor phosphorylation.

(A) Box plots of MoDC quantitative mass spectrometry data for MEK1/2, ERK1/2 and STAT3, which were enriched (phosphorylated) by R848 treatment but depleted (dephosphorylated) by RNA40 relative to Lyovec control (see also Table S2) in phospho-enriched lysate fractions. Both ligands were used at 1 μg/ml for optimal pathway stimulation. ‘No change’ (fold=1.0) marked by dashed line. (B) Densitometric analysis (above) of immunoblots (below) for phospho-MEK1/2, ERK1/2 and STAT3 done on phospho-enriched lysate fractions. Equal total protein amounts were used for phospho-enrichment and subsequently loaded (as assessed by Bradford assay). Two donors were analyzed for each protein. (C) Immunoblot analysis of a time course stimulation of MoDCs using phospho-MEK, -ERK and -NF-κB (p65, control) antibodies and actin as a loading control. Blots are run on whole cell lysates. One representative out of two independent experiments shown.

Differences between RNA40 and R848 extend to recognition by the TLR7 and TLR8 ECD

Disparate phosphorylation and transcription events may be due to differences in uptake kinetics, subcellular localization or actual differences in ligand sensing at the receptor level. We focused here on the latter as this has not been addressed in detail (see also Discussion) This was explored for RNA40 and R848 responses via TLR8 but in order to investigate if any differences were exclusive for TLR8 or might extend to the closely related TLR7, the latter receptor was also included in the analysis. Based on three-dimensional TLR7 and 8 ECD homology models (Ref. (12), see also Methods), charged or proline residues in TLR7 and TLR8 that had positions similar to site 1 and site 2 in TLR3 or TLR9 (9, 10, 12) on the glycan-free sides of both ECDs (greatest dimension ~135 Å) were preferentially selected for mutagenesis (see Fig. 4A, B). These TLR7 or TLR8 mutants were then assayed for their ability to activate NF-κB in HEK293T cells, a well-characterized model system for such structure-function analyses. TLR7 mutation at position R186E (site 1a in Fig. 4A), H304E (site 1b), R553, D555 or Y579 (site 2) resulted in attenuated RNA40 NF-κB activation (Fig. 4C). TLR7 L116, a residue integral to LRR stability, served as a negative control. For TLR8, mutation at position R53, K185 (site 1 in Fig. 4B), D543, Y567 and F568 (site 2) significantly reduced NF-κB activation (Fig. 4D). Immunoblot of HA-tagged constructs, confirmed similar expression levels for all mutants and WT TLR7 or TLR8 (Fig. 4E, F). Live cell microscopy of YFP-tagged TLR7 constructs confirmed localization to intracellular membranes like the endoplasmic reticulum identical to WT TLR7 (data not shown). Thus, in both TLR7 and 8 residues mapping to N-terminal (red/purple, site 1a/b) as well as the central part of their ECDs (blue, site 2) are involved in sensing of RNA40, and their distance (~60-70 Å) fits the anticipated dimensions of an RNA40 molecule (~60-80 Å, see Fig. 4A, B). To contrast RNA ORN with R848 recognition we investigated whether any of the point mutations rendering TLR7 or TLR8 defective in RNA40 sensing would also impact R848 recognition. Fig. 5A, B show that single mutations in both the N-terminal (site 1) as well as the central (site 2) RNA ORN sensing site of TLR7 and TLR8 resulted in drastically reduced R848-triggered NF-κB activation, which was unexpected for a small ligand. Similar results were obtained for the imidazoquinolines CL264 and Imiquimod (not shown). Thus, despite their small size, imidazoquinolines appear to require distal ECD sites for full receptor activation. Given that H304E, R553E and D555A in TLR7, and K185E, D543A/K, Y567A/F and F568S in TLR8, failed to fully activate NF-κB in response to both RNA40 and R848, it can also be concluded that imidazoquinoline and RNA ORN sensing share overall structure-function principles. However, when we compared the levels of NF-κB activation for both ligands within the same experiment relative to their respective WT (Fig. 5C, D), most notably, TLR7 R186E and Y579A, and TLR8 R53E responded normally to R848 but not RNA40, whilst the response of TLR8 H566E, and to a lesser extent TLR7 K328E, was greater for RNA40 than R848 (mutations highlighted in green in Fig.4A, B). Titration experiments showed this was not dependent on ligand concentration (not shown). Thus despite overlapping sensing areas, recognition of either ligand appears to involve several unique residues not required for the detection of the other ligand. Additional experiments demonstrated that the effect of mutagenesis was due to abolished recognition rather abolished binding (at least for RNA40): pull-down of HA- (Fig. 6A-D, S1A,B) or Renilla-tagged (Fig. 6E, S1C) TLR7 and/or TLR8 receptors by biotinylated RNA40 confirmed normal binding (i.e. similar quantities of precipitated and lysate TLR7- and TLR8-HA/-Renilla levels) for constructs containing single (Fig. 6A,B) or multiple mutations (Fig. 6C-F), excluding the possibility that N- and C-terminal sites cooperate in binding as for TLR3 (11). Since mutation at a single site was sufficient to completely abolish RNA40 responsiveness but not binding (even if combined with other mutations), we conclude that the spatially distal (distance ~60 Å) sites identified here are essential for ligand recognition but not ligand binding in TLR7 and TLR8, at least in terms of RNA recognition.

Figure 4. Structure-guided analysis identifies residues in the human TLR7 and 8 ECD essential for receptor function.

3D models of the imidazoquinoline imiquimod, RNA, human TLR7 (A) and TLR8 (B) ECD (grey) with mutated residues shown as spheres as indicated. Residues of functional importance for the sensing of both RNA ORN and R848 are colored red (site 1/1a), magenta (site 1b) or blue (site 2). Residues in green show differential importance for RNA ORN or R848 sensing. N-glycans are in orange. All structures to scale. (C and D) Signaling activity of TLR7 and TLR8 mutants in N- or C-terminal patches is compromised. HEK293T cells were transfected with pcDNA, WT or mutant constructs of TLR7 (C) or TLR8 (D), stimulated with 12 μg/ml RNA40 or DOTAP only (unstimulated) and NF-κB-activation measured by dual luciferase assay (triplicate means +SD). (E and F) TLR7 or TLR8 mutants express to similar levels as WT. HEK293T cells transfected with WT or mutant TLR7-HA (E) or TLR8-HA (F) constructs or an empty vector were analyzed by anti-HA and anti-tubulin immunoblot (loading control). Bands were quantified and intensities relative to the WT signal for each blot calculated. One representative out of three independent experiments shown each.

Figure 5. R848 follows overall RNA ORN recognition principles but certain TLR7 and TLR8 residues discriminate between the two structural different ligands.

(A and B) (A and B) HEK293T cells were transfected with pcDNA, WT or mutant constructs of TLR7 (A) or TLR8 (B), stimulated with RNA vs. R848 and analyzed for NF-κB-activation by dual luciferase assay. One representative out of three independent experiments shown (triplicate means +SD). (C and D) HEK293T cells were transfected as before and stimulated side-by-side with R848 (black, 0.75 μg/ml for TLR7 and 2.5 μg/ml for TLR8) and RNA40 (hashed, 12 μg/ml) and the NF-κB-activation data measured by dual luciferase assay determined and normalized to the level of WT (as 100%). As evident R186E and Y579A in TLR7, and R53E and H566E in TLR8 show significantly different activation between R848 and RNA40 treatment. One representative out of three independent experiments shown each (relative triplicate means + SD).

Figure 6. RNA binding is not compromised in TLR7 and TLR8 loss-of-function mutants.

TLR7 and TLR8 single point mutants (A, B) or mutants harboring multiple mutations (C, D, E) are precipitated in similar ratios compared to WT (A – D) Receptor pull-down using 3’-biotinylated RNA40 reveals intact binding for loss-of-function mutants of TLR7 and TLR8. HEK293T cells transfected as indicated with pcDNA, WT or mutant constructs of TLR7- (A, C) or TLR8-HA (B, D). Cells were lysed, incubated with 3’-biotinylated RNA40 and complexes precipitated with Streptavidin-beads (SA). Isolated complexes and untreated lysates were analyzed by anti-HA immunoblot. In C and D a biotinylated V5-tagged control protein was added to lysates prior to RNA pull-down to verify equal bead carryover. (E) RNA pull-down of Renilla-tagged TLR7 WT and mutant receptors. HEK293T cells transfected with TLR7-Renilla or Renilla-Renilla fusion constructs were lysed. The sample was split and one aliquot used for immediate raw Renilla activity measurement (triplicates). From the remainder, TLR7 or TLR8 were precipitated with biotin-labeled RNA40. Biotin complexes were captured with streptavidin beads and washed beads then measured for bound Renilla luciferase activity (triplicates ). Plots represent the relative light units (RLU) of bound vs. raw Renilla for each samples (see Fig. S1C,D), normalized to WT as 100%. One single experiment shown. (F) Overview of multiple point mutants used.

Functionally important human and murine TLR7 alleles map to the identified sensing sites

We next sought to determine if N- and C-terminal recognition sites were of general physiological importance in vivo. Therefore, reported loss-of-function mutations in murine Tlr7 and single-nucleotide polymorphisms in human TLR7 or TLR8 that mapped to these sites were functionally tested. R848-unresponsive mouse strains generated during N-ethyl-N-nitrosourea-mutagenesis (18) include the rsq2 strain (MGI: 3811335). Indeed rsq2 peritoneal macrophages failed to respond to R848 (Fig. 7A,B) like the ‘functionally null’ Tlr7 strain rsq1 (mutation T68I; Ref. (19)). rsq2 mice contained a N182Y mutation in Tlr7 which maps to the proposed N-terminal sensing site 1 (Fig. 7C). The equivalent mutation in human TLR7, N182Y, was also hyporesponsive to R848 in HEK293T cells (Fig. 7D). Thus, mutation in the N-terminal recognition site of human and murine TLR7 appear to lead to R848 hyporesponsiveness in vivo. Additionally, dbSNP, a database of reported human genetic variants, yielded several reported non-synonymous TLR7 variants mapping to the central part of the ECD, i.e. site 2 (Fig. 7C) but with an unknown frequency (see Methods). Genotyping in collectives of UK (n=502), Kenyan (n=369) and Indian origin (n=393; (10)) failed to identify homo- or heterozygous carriers (see Discussion). Consequently, cells from allele carriers were unavailable for functional analysis but experiments in HEK293T cells (Fig. 7E, F) showed that M603I, S610C, N576D and R627I resulted in >50% or complete loss of function despite normal expression levels. N576D and R627I also showed a dominant-negative effect over WT TLR7 (not shown). Since TLR7 is X-chromosomal, these findings suggest that hemizygous male carriers of site 2 variant alleles may have reduced TLR7 function and even in heterozygous female carriers the function of the WT allele may be impaired. Collectively, these data imply that the N- and C-terminal sensing sites in at least TLR7 bear physiological relevance.

Figure 7. Functionally important murine and human TLR7 alleles map to N- and C-terminal ectodomain sensing sites.

Peritoneal macrophages from C57BL/6J (filled circles), rsq1 (open circles) and rsq2 (Tlr7^N182Y, grey diamonds) mutant mice were stimulated with different TLR ligands (A), or a dose titration of R848 (B). TNF secretion was measured by an L-929 cytotoxicity bioassay based on propidium iodide absorption. One single experiment shown each. (C) TLR7 ECD 3D model (grey) with mutations corresponding to naturally occurring TLR7 mutations/SNPs. Mutations leading to altered receptor function are shown in red or blue, the rsq1 mutation reported earlier (19) in orange. (D) The rsq2-equivalent mutation in human TLR7, N182Y, leads to abrogated receptor responsiveness in transfected HEK293T cells stimulated with R848 and analyzed for NF-κB-activation by dual luciferase assay. One representative out of two independent experiments shown (triplicate means +SD). (E) The naturally occurring TLR7 variants N576D, M603I, S610C and R627I (blue) lead to reduced NF-κB-activation, despite similar expression levels (F) compared to WT. HEK293T cells were transfected with pcDNA, WT or TLR7 SNP-equivalent mutants, stimulated with R848 and analyzed for NF-κB-activation by dual luciferase assay (E) or lysates probed using anti-HA and anti-Tubulin (loading control) Abs (F). One representative out of three independent experiments shown (triplicate means +SD).

Discussion

Here we investigated similarities and differences between RNA40 vs. R848 downstream signaling and receptor recognition using both primary cells and model systems. Several novel findings warrant further discussion:

The first unexpected discovery was that in primary human MoDC both types of TLR8 ligands, RNA ORN and R848, at equimolar concentrations elicited different transcriptional events via the same receptor (cf. Fig. 1,2). These may result from differential ligand uptake kinetics, subcellular localization or actual differences in ligand sensing at the receptor level. For example, for both TLR4 and TLR9 it was shown that the subcellular location from which signaling is initiated strongly influenced the balance between NF-κB- vs. IRF-dependent signaling for the same ligand (e.g. LPS and CpG, respectively) (21-24). Thus it is conceivable that, if RNA ORN and IMQ concentrate in or transit through different endosomal compartments disparately, this could lead to differences in the relative strength of downstream signals and thus phosphorylation and transcriptional differences. Unfortunately, according to our knowledge no direct comparisons have been published for RNA ORN and IMQ. However, since IMQ and RNA ORN can functionally counter-influence each other (25, 26), we consider it highly likely that the uptake and trafficking profiles of IMQ and RNA ORN overlap significantly in a time frame relevant for receptor recognition and therefore that the differences observed here are likely to result from disparate recognition at the receptor ECD level. Disparate ligand recognition was reported for TLR3 and TLR4 which, depending on subtle ligand differences (e.g. length of dsRNA or phosphorylation status of Lipid Iva moiety, respectively), activate downstream signaling disparately (27, 28). Our ECD mutagenesis data provide good evidence that this applies to TLR7 and TLR8 and that both receptors,like TLR3 and TLR4, are able to operate in more than one signaling ‘mode’ in a particular cell type, depending on the class of ligand (RNA ORN or imidazoquinolines). Different signaling ‘modes’ may thus be a general feature of TLR signaling and can be mediated either directly by the receptor ECD (nucleic acid sensors TLR3, TLR7, TLR8) or via co-receptors (MD-2 with TLR4, TLR1/6 with TLR2). Mechanistically, two classes of ligands inducing different ECD conformations may result in different TIR-TIR and/or Myddosome post-receptor complexes (29) and subsequent downstream mediator phosphorylation or de-phosphorylation events (cf. Table S2) – evidenced here for ERK1/2 or MEK1/2 (cf. Fig. 3) – details of which will now be interesting to address in future studies. For the regulation of TNF in monocytes or IFNα in pDC different signaling ‘modes’ may be of no consequence. However, regarding other genes, e.g. CXCL1, HHEX, IL6, IL8, IL10 and IFNb1, and in certain cell types, e.g. MoDC (this study and Ref. (30)) or neutrophils (4), RNA ORN- and R848-induced gene transcription is interchangeable. We suspect similar results for TLR7-mediated R848 vs. RNA ORN signaling may emerge in pDC or B cells (which exclusively express TLR7) if a wider panel of activated genes is assessed. Our study thus cautions the extrapolation from one class of TLR7/8 ligand to the other, but rather suggests that both types of ligands need to be assessed separately. Therefore, studies in which R848 has been used as a surrogate for RNA to ‘simulate’ the response of cells to ‘viral infection’ may have to be re-evaluated. On the other hand, the insights reported here suggest that use of either RNA ORN or imidazoquinolines could be suitable for generating MoDC-based cellular vaccines with particular effector profiles (14). Further studies in this direction appear mandatory to anticipate beneficial effects and potentially harmful off-target effects linked to the use of RNA ORN and imidazoquinolines (5).

Secondly, our data provide evidence that both TLR7 and TLR8 employ spatially distal N-terminal and central ECD sensing sites for ligand sensing. For TLR7 and 8 the N-terminal site 1 involves LRRs 2-5 and their insertions – with contributions from residues in LRR 10/11 at least for TLR7 (site 1b) - and the central site 2 LRRs 14-18 (cf. Fig. 4A,B). Regarding both site 1 and 2, our data are in good agreement with earlier bioinformatics and experimental approaches (8, 13, 31) and a recent TLR8 ECD crystal structure (17) which shows two sites of R848 recognition (see also Methods). For the central TLR7 ECD sensing site, several reported hypo-sensitive TLR7 alleles described here indicate a functional importance (cf. Fig. 7C,E), albeit indirectly, as the frequency and functional phenotype of these variants will require further verification (see also below). Our mutagenesis data and the R848 hyposensitivity of rsq2 mice (mutated N-terminally at position 182; cf. Fig. 7A-C) implicate the TLR7/8 ECD N-termini in ligand sensing. It has been suggested that the reported requirement for endopeptidase cleavage for human TLR7 and TLR9 require to gain functionality (discussed in (3)) argued for the N-termini of endosomal TLRs to be non-essential for ligand recognition. However, recent biochemical (17) and cell biological evidence (32-34) clearly demonstrates that despite ECD cleavage, the N- and C-terminal fragments remain associated after backbone cleavage in a position between the two sensing sites proposed here. In good agreement with our data, both parts were also found to be strictly necessary for function. In HEK293T cells (this study and (9)) and human neutrophils (35), TLR7-9 cleavage is not observed and thus appears redundant here. Collectively, therefore, our data strongly support a direct functional and physiologically relevant role of the TLR7/8 N-termini in the process of ligand sensing. TLR7 and TLR8 thus follow a general principle of nucleic acid sensing by TLRs proposed by us earlier (9), namely that ligand recognition involves N- and C-terminal sensing sites. In terms of RNA recognition this appears plausible as the distance between N-terminal and central site (~60 Å, cf. Fig. 4A, B) also fits the minimum TLR7 stimulatory RNA sequence of 19 nt (corresponding to 77 Å) reported in Ref. (36). Interestingly, whereas site 2 residues are fully conserved between TLR7 and TLR8, N-terminal recognition residues are less conserved (not shown) which may explain differences in ligand preference amongst different imidazoquinolines. Of note, the sites identified here do not appear critical for ligand binding. Earlier studies showed a general promiscuity of TLR7 for both RNA and DNA (37) and in an ORN pull-down assay we found TLR7 binding was stronger for activating vs. non-activating ORN, but not altogether abrogated (19), indicative of a basal level of (non-specific) binding for nucleic acids. Similarly to TLR9 (10, 38) RNA ORN binding by TLR7 and TLR8 thus appears to be relatively non-specific with ligand recognition being a separate event dependent on the presence of activation determinants (e.g. G/U-richness). Such non-specific binding employed by TLRs 7-9 may be useful for sampling the endosomal nucleic acid content for low-abundance viral or bacterial RNA or DNA molecules.

The finding that not only sensing of an elongated RNA ORN ligand but also small molecular imidazoquinolines by TLR7 and TLR8 was influenced by two spatially distant sites (distance between site 1a and site 2 of ~60 Å) was unexpected as R848 in its longest dimension only spans ~9 Å (cf. Fig. 4A, B). Previously, it was proposed that the imidazoquinoline mechanism of receptor activation was the insertion of one imidazoquinoline at a single defined ECD site (13, 17). Our functional data argue for R848 and other imidazoquinolines to require multiple or extended contact sites with the receptor. This is confirmed by the TLR8 crystal structure which shows two imidazoquinolines ‘wedged’ by two TLR8 ECD (17). Based on the structural resemblance of imidazoquinolines to RNA nucleobases, imidazoquinoline sensing could also involve a multimeric, aggregated and/or sequentially arrayed ‘RNA-like’ R848 structure that could span spatially distal areas of the ECD. Interestingly, concentration-dependent, >100-fold accumulation of imidazoquinolines in acidified compartments was reported (8, 25) and is evidenced in molecular dynamics simulations (not shown). Imidazoquinolines and nucleic acids were also shown to interact (25, 39), so that aggregation may be stabilized by natural RNA or DNA fragments originating from cellular debris through base-pairing or intercalation (Ref. (26)). Whichever precise stoichiometry and molecular mechanism applies to this novel observation regarding imidazoquinoline sensing by TLR7 and TLR8, it is of fundamental interest and warrants future experimental investigation, not least as aggregation/sequential arrangement would influence the pharmacological properties of imidazoquinolines.

Our genotyping data argue that for wider therapeutic use of TLR7/8 agonists, hypofunctional alleles are unlikely to pose a problem as the TLR7 alleles studied here emerged with estimated frequencies of <1% in the several ethnic groups investigated. A TLR7 and TLR8 evolutionary genetics study (40) recently proposed low non-synonymous allele frequencies in TLR7 and TLR8 may reflect purifying selection, based on the assumption that viral infection has profoundly influenced human allelic inheritance. This was also applied to loss-of-function variants in other critical TLR pathway genes like TIRAP, MYD88 or TLR9 (10, 41-44). Although two children deficient in UNC93B1, an ER protein required for efficient TLR3, 7, 8 and 9 trafficking and thus functionality, appeared only susceptible to herpes simplex infection, other reports imply TLR7/8 in surveillance against HIV and endogenous retroviruses (e.g. Refs. (45) and (46)) and in autoimmune diseases like SLE (47) so that TLR7/8 functional alleles may have an impact here. Larger genotyping studies in such patient collectives may yield carriers of TLR7 loss-of-function alleles and open the way for further clinical and ex vivo studies in primary cells. This could shed more light on the physiological importance of the proposed sensing sites and also clarify the general role of TLR7/8 in human immunity.

In conclusion, our data reveal fundamental principles of RNA and imidazoquinoline recognition by human TLR7 and TLR8 and hint to a model of activation shared by other nucleic acid-detecting TLRs and involving multiple signaling ‘modes’. This structure-function framework will be vital to dissect further in order to better understand the role of these receptors in antiviral immunity or clinical autoimmunity. Considering the use of TLR7 or TLR8 ligands in vivo, our study suggests each class (ORN vs. imidazoquinolines) may lead to a distinct combination of cell type-specific favorable and unwanted cellular effects. Structure-guided approaches using chemically modified RNA ORN and/or imidazoquinolines may yield TLR7 and TLR8 agonists which elicit clinically desirable receptor outcomes. Proteome and genome wide profiling of downstream signaling events as employed here may be required for sufficient resolution of ligand-specific differences.

Abbreviations

DC

dendritic cell

ECD

extracellular domain

HA

hemagglutinin

IFN

interferon

IL

Interleukin

IRAK

IL-1 receptor-associated kinase

IRF

IFN regulatory factor

LRR

leucine-rich repeat

MAPK

Mitogen-activated protein kinase

MoDC

Monocyte-derived DC

MyD88

Myeloid differentiation 88

ORN

oligoribonucleotides

pDC

plasmacytoid DC

TLR

Toll-like receptor

TNF

tumor necrosis factor

TRAF

TNF receptor-associated factor