Structural basis of TIR-domain-assembly formation in MAL- and MyD88-dependent TLR4 signaling.

Protein production and purification

The cDNAs of MAL FL (residues 1–221, N-terminal c-Myc-tag), MAL TIR (residues 79–221, N-terminal c-Myc-tag), TLR3 TIR (residues 750–904), and TRAM TIR (residues 70–235) were amplified by PCR and cloned into the pMCSG7 expression vector 47 by ligation-independent cloning. The resulting constructs encode an N-terminal His 6 -tag and were verified by sequencing. MAL FL , MAL TIR , TLR3 TIR , TRAM TIR , MyD88 TIR (residues 155–296 in pET28b, C-terminal His 6 -tag 48 ), and TLR4 TIR (residues 671–820, pET28b, N-terminal His 6 -tag) were produced in Escherichia coli BL21 (DE3) cells using auto-induction media. Cells were grown at 303–310 K until the mid-exponential phase (OD 600nm of 0.6–0.8) was reached. The temperature was then reduced to 288–293 K and the cultures were grown for approximately 16 h before harvesting. The cells were lysed in 50 mM HEPES (pH 7–8), 500 mM NaCl and 1 mM DTT using sonication and the resulting supernatant was applied onto a 5 ml HisTrap FF column (GE Healthcare). Bound protein was eluted using a linear gradient of imidazole from 30 to 250 mM and the fractions containing the protein of interest were pooled, concentrated and applied onto a Superdex 75 HiLoad 26/60 gel-filtration column (GE Healthcare) pre-equilibrated with 10 mM HEPES pH 7.5 and 150 mM NaCl. The peak fractions were pooled, concentrated to a final concentration of 1–10 mg/ml and stored in aliquots at 193 K. Alanine mutations in MAL TIR were generated using PCR-based gene-synthesis 49 , 50 , followed by cloning into the pMCSG7 vector. Similar to wild-type MAL TIR , the alanine mutants were produced in E. coli BL21 (DE3) cells using auto-induction media, but they were purified in batch mode using Ni-NTA resin and an elution buffer containing 50 mM HEPES pH 7.5, 150 mM NaCl and 250 mM imidazole. Alanine mutations of MyD88 TIR were generated using traditional PCR and produced in E. coli BL21 (DE3) cells using auto-induction media. They were purified using the same method as described for wild-type MyD88 TIR 48 .

Assays for monitoring

TIR-domain assembly formation in solution Assembly formation was analyzed at 25–30°C in 10 mM HEPES pH 7.4, 150 mM NaCl by either monitoring the change in turbidity (absorbance at 350 nm) in a SpectraMax 250 96-well spectrophotometer (Molecular Devices), or by analyzing the soluble and insoluble fractions of the samples by SDS-PAGE. Negative-stain EM Four µl of sample from the turbidity or precipitation assays was applied to glow-discharged carbon-coated Formvar grids, adsorbed onto the grid surface for 2 min, followed by negative staining with 1% w/v uranyl acetate for 2 min, and then blotted and air-dried for 10–15 min. The samples were imaged on a JEOL JSM-1011 transmission electron microscope (TEM) operating at an accelerating voltage of 80 kV, or a FEI Tecnai 12 TEM operating at an accelerating voltage of 120 kV.

Cryo-EM of the MAL TIR filament Grids containing vitrified MAL

TIR filaments were imaged using an FEI Titan Krios TEM operating at 300 kV equipped with a 4k × 4k Falcon II direct detection device with a backthinned CMOS chip. Images were collected using a sampling of 1.05 Å/pixel. A total of 816 images were acquired on a Titan Krios operating at 300 kV, and the contrast transfer function (CTF) was estimated using CTFFIND3 51 . Images with a poor CTF were rejected, and 446 images were used for further processing. The defocus values ranged from 0.5 to 3.0 µm. The phases in the micrographs were corrected by multiplying the images by the theoretical CTF (a Wiener filter in the limit of a very poor signal-to-noise ratio). Long filaments were cut from the images using e2helixboxer within the EMAN2 suite of programs 52 . Overlapping boxes 384 pixels long, each shifted by 22 px (~ 1.5 times the axial rise), were cut from the long filaments, yielding 54,169 segments. The SPIDER software package 53 was used for most subsequent operations, including the IHRSR 54 implementation. The final volume, which had been multiplied by the CTF twice (once by the microscope, and once computationally when the phases were corrected), was modified in amplitudes by multiplying by the sum of the squared CTFs (a Wiener filter in the limit of a high signal-to-noise ratio). The volume was sharpened by applying a negative B-factor of 250 Å 2 . Model building and refinement The MAL NMR ensemble (PDB ID 2NDH), and the MAL crystal structures (PDB ID 2Y92 and 3UB2) modeled with either an AB loop 18 or BB loop (based on the TRAM NMR structure, PDB ID 2M1W) were initially docked into the density map using Chimera’s fit into density tool 55 . The central parallel β-sheet of the crystal structure and NMR ensemble showed good agreement with the experimental data, but the conformations of some of the surrounding helices and loop regions were visibly different. While no density could be observed for the AB-loop region, continuous density was clearly visible in the region of the BB loop. However, the fit of the BB loop to the density was poor in both the NMR ensemble and the crystal structure modeled with a BB loop, suggesting that this region adopts a different conformation in the filament structure. To obtain a model of the MAL TIR that fits better with the density map, two approaches were used in parallel: direct interactive modeling using interactive molecular dynamics flexible fitting (iMDFF) in VMD 56 , 57 , and Rosetta density-guided rebuilding 58 as described below. The top models from both approaches were then combined and refined using RosettaCM 59 , and the top RosettaCM model was subjected to a final refinement with iMDFF in VMD. iMDFF/VMD: The MAL crystal structure (PDB ID 2Y92) modeled with a BB loop (based on the TRAM NMR structure, PDB ID 2M1W) was initially docked into the density map using Chimera’s fit into density tool. iMDFF was carried out on an octameric fragment containing four chains from each of two adjacent proto-filaments – the minimal fragment that captures all protein-protein interfaces. Short contiguous segments of 10–20 residues were mobilized one at a time at 100 K under implicit solvent conditions, together with surrounding contacting residues. These were remodeled by interacting with the running simulation using a haptic interface, guided by fit to the density, favorability of side-chain interactions, and appearance of the Ramachandran plot. At regular intervals, symmetry was regularly re-imposed by first propagating modifications made to one dimer throughout the octamer, followed by ~50 ps equilibration and minimization to convergence with strong symmetry restraints on all heavy atoms. Model building in Rosetta. A crystal structure template of the TIR domain of MAL (PDB ID 3UB2) was initially docked into the density map using Chimera’s fit into density tool. Comparative modelling was carried out in RosettaCM using a sequence alignment generated with hhsearch . Comparative modelling was carried out in the full symmetric complex (a symmetry definition file was generated using make_symmdef_file.pl ) and the experimental density was used in model building and refinement. From this starting point, 1000 independent rebuilding trajectories were run; the best was selected from the top 100 models by Rosetta energy, followed by the top 5 models by fit to data (using the integrated Fourier-shell correlation [FSC] between 10 Å and 5.5 Å). These five models were then subjected to iterative local refinement following a previously published protocol 58 ; from each of the 5 starting models, 100 models were generated, and the top 5 were chosen using the previously described criteria. Model recombination in Rosetta: Following visual inspection, both iMDFF models and Rosetta models had local features that seemed to better explain the experimental data in specific regions, likely due to each method exploring different parts of conformational space. We wanted to build a model that combined features from both models. Therefore, we again used RosettaCM to build a final model, using as input both the MDFF model and the 5 Rosetta models described previously. Given multiple input structures, RosettaCM combines segments of both, attempting to minimize the total energy of the system. As before, this was done in the complete symmetric complex, with a score term assessing agreement to experimental data enabled within Rosetta. In this final model recombination, another 1000 models were generated, the top 100 were selected by Rosetta energy, and the top model was selected by integrated high-resolution FSC.

Cryo-EM of the MyD88 TIR assemblies

For cryo-EM of the MyD88 TIR domain assemblies, 4 µl of protein sample was transferred onto a lacey carbon grid, in an FEI Vitrobot Mark 3, with the chamber set to 25°C and 100% humidity, and plunged into liquid ethane. Frozen/vitrified samples were viewed on a Tecnai F30 microscope operating at 300 kV, and imaged at 59,000x magnification with a Direct Electron LC1100 4k x 4k camera (Direct Electron, San Diego, USA), using the low-dose mode of SerialEM image acquisition software ( http://bio3d.colorado.edu/SerialEM/ ). Micrographs of the MyD88 TIR assemblies were treated as 2D crystals, and auto-indexing (with manual refinement), unbending (to correct for lattice distortions), and corrections for the contrast transfer function were performed with the program 2dx 60 .

Structural analysis and modeling

Structural analyses were performed using Chimera 55 , PyMOL (The PyMOL Molecular Graphics System, Version 1.7.4 Schrödinger, LLC), DALI 61 , PISA 62 , and PDBeFold 63 . Homology models were made using Modeller 64 , multiple sequence alignments were prepared using MUSCLE 65 , and alignments were formatted using ESPript 66 .

Constructs for mammalian expression and cell culture

Full-length human MAL and MyD88, the TIR domains of human MAL (residues 79–221) and MyD88 (residues 155–296), and the N-terminal region of MAL (residues 1–79) were cloned into the pEF6/V5-His-TOPO vector. The MAL constructs encoded an N-terminal c-Myc-tag, while the MyD88 constructs encoded a C-terminal V5-His-tag. Mutants of MAL and MyD88 were generated by traditional PCR. All constructs were sequence-verified. The human MAL FL template encoded a naturally occurring single-nucleotide polymorphism (G164A) resulting in a S55N amino-acid change. This amino acid change has previously been established to have no effect on MAL-induced NF-αB activation or MAL localization 67 . HEK293 cells were obtained from the ATCC, were tested for mycoplasma, and maintained as previously described 68 .

Microscopy

HEK293 cells were plated on poly-L-lysine coated coverslips, transfected and immuno-stained as described previously 68 . Myc-tagged MAL was detected using mouse monoclonal (9B11) anti-Myc antibody (Cell Signaling Technology) at a 1:1000 dilution overnight at 4°C followed by goat anti-mouse–Alexa Fluor-555 (Life Technologies) at a 1:1000 dilution for 1 h at room temperature. V5-tagged MyD88 was detected using rabbit monoclonal (D3H8Q) anti-V5 antibody (Cell Signaling Technology) at a 1:500 dilution overnight at 4°C followed by goat anti-rabbit–Alexa Fluor-488 (Life Technologies) at a 1:1000 dilution for 1 h at room temperature. Immuno-stained cells were viewed on either a Zeiss Axioplan 2 microscope and captured using an Axiocam 506 colour/greyscale digital imaging camera with Zeiss Zen software or on a Zeiss LSM 710 laser-scanning confocal microscope also using the Zeiss Zen software. Images from the LSM 710 were merged using ImageJ. Brightness and contrast of all images were adjusted using Adobe Photoshop CS6 software.

Evaluation of MyD88 clustering in HEK293 cells by flow cytometry

Assays were performed using similar conditions to those used to evaluate ASC speck formation 17 . To assess the ability of wild-type or mutant MyD88 to form clusters, HEK293 cells (200,000) were plated in 12-well plates and grown overnight. Transfections for each well contained empty vector alone or 200 ng of plasmid expressing V5-tagged wild-type or mutant MyD88. To assess MAL-induced clustering of Myd88, 200 ng of plasmid expressing Myc-tagged MAL was co-transfected with the V5-tagged MyD88-expressing plasmid. Empty vector was added so that the final amount of DNA in each sample was 800 ng. DNA-Lipofectamine complexes were added to cells followed by centrifugation of the plates at 700 x g for 10 min. Cells were harvested after ~16 h and fixed for 15 min with 75% ethanol. Cells were immuno-stained using the mouse monoclonal (9B11) anti-Myc (Cell Signaling Technology) at a 1:2000 dilution and the rabbit monoclonal (D3H8Q) anti-V5 antibody (Cell Signaling Technology) overnight followed by goat anti-rabbit–Alexa Fluor-488 (Life Technologies) and goat anti-mouse Alexa Fluor-647 (Life Technologies) at a 1:10,000 dilution for 60 min. The immuno-stained cells were run on a BD Accuri C6 flow cytometerV5-positive or Myc- and V5-double positive cells were gated and used for further analysis of speck formation using height-to-area analysis. The percentage of speck-forming cells was determined in a window with a broad range of expression levels or low MyD88 (V5) expression as indicated. The data were analyzed using the FlowJo software.

Cell-free expression and seeding of polymerization of MAL

FL and MyD88 FL MAL FL , MAL TIR , MyD88 FL and MyD88 TIR were expressed in vitro using a cell-free protein production system based on Leishmania tarentolae 69 . The proteins were tagged with GFP at the N-terminus and single-molecule confocal spectroscopy was performed on a home-made microscope to measure protein aggregation in the cell-free extracts 70 . Protein expression levels were measured based on the average fluorescence, and protein oligomers and filaments were detected as large, very intense bursts of fluorescence. Two-color seeding experiments were performed using the same detection principle.

Preparation of Leishmania tarentolae cell-free extracts Leishmania tarentolae cell-free extracts

(LTE) were produced as previously described 69 , 71 , 72 . Briefly, Leishmania tarentolae Parrot strain was obtained as LEXSY host P10 from Jena Bioscience GmbH, Jena, Germany and cultured in TBGG medium containing 0.2% v/v penicillin/streptomycin (Life Technologies) and 0.05% w/v hemin (MP Biomedical). Cells were harvested by centrifugation at 2500 x g , washed twice by re-suspension in 45 mM HEPES, pH 7.6, containing 250 mM sucrose, 100 mM potassium acetate and 3 mM magnesium acetate and re-suspended to 0.25 g cells/g suspension. Cells were placed in a cell disruption vessel (Parr Instruments, USA) and incubated under 7000 kPa nitrogen for 45 minutes, then lysed by rapid release of pressure. The lysate was clarified by sequential centrifugation at 10 000 x g and 30 000 x g and anti-splice leader DNA leader oligonucleotide was added to 10 µM. The lysate was then desalted into 45 mM HEPES, pH 7.6, containing 100 mM potassium acetate and 3 mM magnesium acetate, supplemented with a coupled translation/transcription feeding solution and snap-frozen until required.

Gateway plasmids for cell-free protein expression

The proteins were cloned into the following cell-free expression Gateway destination vectors 73 : N-terminal GFP-tagged (pCellFree_G03), N-terminal mCherry-myc-tagged (pCellFree_G07) or C-terminal mCherry-cMyc-tagged (pCellFree_G08). The open reading frames (ORFs) corresponding to full-length MAL and full-length MyD88 were obtained from the human ORFeome at the Diamantina Institute, UQ 74 , the TIR-domain constructs for both proteins were sub-cloned from the bacterial vectors described above. Transfer of ORFs between vectors was carried out using Gateway PCR cloning protocol, based on insert amplification with primers to attB1 and attB2 sites (forward primer: GGGGACAAGTTTGTACAAAAAAGCAGGCTT (nnn)18–25, reverse primer: (nnnn)18–25AACCCAGCTTTCTTGTACAAAGTGGTCCCC) 75 .

Single-molecule fluorescence spectroscopy

Single-molecule spectroscopy was performed as described previously 71 , 76 , directly in the cell-free expression mixtures, without any purification steps. Labelling of the proteins for fluorescence measurements was simplified by the use of genetically encoded fluorophores (GFP and mCherry). The principle of the single-molecule fluorescence technique is the following: on a confocal microscope, two overlapping lasers exciting GFP and mCherry fluorophores are used to create a small detection volume in which GFP and mCherry fluorescence emitted by proteins is recorded separately on two single photon-counting detectors. As proteins are freely diffusing, they constantly enter and exit the detection volume of the microscope, creating fluctuations in the fluorescence intensity collected. The amplitude and frequency of these fluctuations are quantified to characterize the oligomerization and aggregation of proteins 70 . In Supplementary Fig. 2 , MAL and MyD88 proteins were expressed in LTE by adding the template DNA to 10 µL lysate to a final concentration of 20 nM. The reaction was kept at 27°C for 2.5 h, then the mixture was diluted 10 times in 25 mM HEPES, 50 mM NaCl. A volume of 20 µL of each sample was placed into a custom-made 192-well silicone plate with a 70 × 80 mm glass coverslip (ProSciTech). Plates were analyzed at room temperature on a Zeiss Axio Observer microscope with a custom-built data-acquisition setup. A 488 nm laser is focused in solution using a 40x/1.2 NA water immersion objective (Zeiss). Fluorescence was collected and separated using a 565 nm dichroic mirror and was passed through a 525/20 nm band-pass filter. Single-molecule fluorescence spectroscopy: seeding experiments In Supplementary Fig. 2 , mCherry-tagged “seeds” of MyD88 FL and MAL FL were purified using N-terminally tagged MyD88 FL and C-terminally tagged MAL FL . Proteins were expressed in LTE by addition of the respective template DNAs in 10 µL lysate at a final concentration of 40 nM. The expression was allowed to be carried out for 3.5 h at 27°C. To produce the “seeds”, the samples were then spun down at 13,000 x g for 5 min. 80% of the supernatant was discarded and the solution was sonicated for 1 min in a water bath. In the meantime, GFP-tagged MyD88 FL or MAL FL were expressed for 2.5 h, from 2 nM DNA template, and diluted 10 times before being placed under the microscope. Two lasers (488 nm and 561 nm) were focused in solution using a 40x/1.2 NA water immersion objective (Zeiss). Fluorescence was collected and separated using a 565 nm dichroic mirror; signal from GFP was passed through a 525/20 nm band-pass filter, while fluorescence from mCherry was filtered by a 580 nm long-pass filter. The fluorescence of the two channels was recorded simultaneously in 1 ms time bins. Fluorescence time traces were recorded for 75 s before 1 µL of “seeds” was introduced into the mixture and the measurements were carried out for an additional 300 s.

Data availability The 3D cryo-EM density map of MAL

TIR has been deposited in the Electron Microscopy Data Bank (EMDB) under accession numbers EMD-8625. Atomic coordinates for MAL TIR have been deposited in the Protein Data Bank (PDB) under accession number 5UZB.