MyD88 TIR domain higher-order assembly interactions revealed by microcrystal electron diffraction an

MyD88 and MAL are Toll-like receptor (TLR) adaptors that signal to induce pro-inflammatory cytokine production. We previously observed that the TIR domain of MAL (MALTIR) forms filaments in vitro and induces formation of crystalline higher-order assemblies of the MyD88 TIR domain (MyD88TIR). These crystals are too small for conventional X-ray crystallography, but are ideally suited to structure determination by microcrystal electron diffraction (MicroED) and serial femtosecond crystallography (SFX). Here, we present MicroED and SFX structures of the MyD88TIR assembly, which reveal a two-stranded higher-order assembly arrangement of TIR domains analogous to that seen previously for MALTIR. We demonstrate via mutagenesis that the MyD88TIR assembly interfaces are critical for TLR4 signaling in vivo, and we show that MAL promotes unidirectional assembly of MyD88TIR. Collectively, our studies provide structural and mechanistic insight into TLR signal transduction and allow a direct comparison of the MicroED and SFX techniques.

Toll-like receptors (TLRs) detect pathogens and endogenous danger-associated molecules, initiating innate immune responses that lead to the production of pro-inflammatory cytokines. Signaling by TLRs is initiated by dimerization of their cytoplasmic TIR (Toll/interleukin-1 receptor [IL-1R]) domains, followed by recruitment of the TIR-containing adaptor proteins, including MyD88 (myeloid differentiation primary response gene 88) and MAL (MyD88 adaptor-like/TIRAP)(Fig. 1)1. Combinatorial recruitment of these adaptors via TIR : TIR interactions orchestrates downstream signaling, leading to induction of the pro-inflammatory genes. In previous work, we showed that MAL TIR domains (MALTIR) spontaneously and reversibly form filaments in vitro. They also formed co-filaments with TLR4 TIR domains (TLR4TIR) and nucleated the assembly of MyD88TIR into crystalline arrays2. These results suggested signaling by cooperative assembly formation (SCAF), a mechanism prevalent in innate-immunity and cell-death pathways3,4, and we proposed a model for signal amplification, in which the TLR4, MAL and MyD88 TIR domains sequentially and cooperatively assemble into a higher-order TIR domain complex. This assembly then induces the formation of the Myddosome, involving the death domains of MyD88 and the protein kinases, IRAK2 and IRAK4, leading to proximity-based activation of these kinases (Fig. 1)5,6. The 7 Å cryogenic electron microscopy (cryo-EM) structure of the MALTIR filament revealed a hollow tube composed of 12 two-stranded protofilaments of TIR domains and mutational analyses revealed that protein interactions within these protofilaments are likely to represent higher-order TIR-domain interaction interfaces during in vivo signaling, although the structures formed within cells may be more limited in size7. However, the structural basis of how MyD88TIR and TLR4TIR domains self-assemble and interact with MALTIR remained uncharacterized.

Pathogen-associated molecular patterns (e.g., LPS) binding to the extracellular LRR domain of a TLR (e.g., TLR4) induces dimerization of its TIR domains, which leads to the recruitment of an adaptor TIR domain (e.g., MALTIR) to the extended surface created by the TLR4TIR dimer. Elongation of this trimer through recruiting additional adaptor’s TIR domains (e.g., MALTIR or MyD88TIR) into a higher-order complex leads to clustering of MyD88 DDs and subsequent recruitment of IRAKs through DD interactions. The initial TIR dimerization and trimerization steps are likely to be unfavourable and rate limiting, whereas subsequent monomer additions are more favourable, rapid and cooperative. LRR, leucine-rich repeat domain; LPS, lipopolysaccharide; TLR, Toll-like receptor; TIR, Toll/interleukin-1 receptor domain; DD, death domain.

Full size image

Here we set out to structurally characterize the MyD88TIR crystalline assemblies observed in our previous work2. As the crystals were too small for conventional X-ray crystallography, we employed the complementary techniques of microcrystal electron diffraction (MicroED) and serial femtosecond crystallography (SFX). MicroED8,9 enables structure determination of submicrometre-sized crystals. In MicroED data collection, the crystal is continuously rotated in a transmission electron microscope (TEM)10,11,12, analogous to the rotation method used in X-ray crystallography13, and to related three-dimensional electron diffraction methods in TEM14. MicroED can complement existing methods in structural biology such as conventional X-ray crystallography, where growing crystals of sufficient size and crystallinity is often the major barrier to structure determination15,16,17,18. Indeed, many failed crystallization trials have been shown to contain microcrystals19,20,21. Furthermore, small macromolecular crystals potentially have reduced defects22,23,24, and controlled perturbations to the sample, such as soaking and vitrification, may be applied rapidly and more uniformly24,25,26. MicroED has already enabled protein structure determination from microcrystals9,10,23,27,28,29,30, structure solution of a previously uncharacterized metalloenzyme31, structure determination of membrane proteins from microcrystals embedded in lipidic cubic phase32,33,34,35 and the visualization of ligand-binding interactions25,36. Furthermore, MicroED enables the study of biomolecules that naturally aggregate or assemble into microcrystals, facilitating structure determination of several short peptide fragments from thin prion protofibrils23,37,38,39. Such naturally occurring crystalline assemblies are of special interest, as they can reveal the interactions occurring in assemblies within cells, illustrating the underlying mechanisms guiding the assembly formation and providing relevant structural insights.

More or less in parallel to the development of MicroED, SFX has emerged as a powerful technique for structure determination and the study of protein dynamics of microcrystalline samples24,40,41,42,43. SFX exploits the femtosecond-scale duration of extremely brilliant X-ray free-electron laser (XFEL) pulses for the collection of high-quality diffraction data at room temperature, which occurs before the onset of structure-altering radiation damage44,45,46,47. In SFX, diffraction data are collected as single snapshots from randomly oriented microcrystals45,47. With the crystals delivered to the beam at room temperature and minimal sample handling, challenges associated with cryo-cooling and potential protein conformation restrictions are avoided48. SFX has facilitated structure determination from submicrometre crystals of radiation-sensitive proteins49,50,51 and membrane proteins such as G protein-coupled receptors52,53,54,55. SFX has also enabled time-resolved studies of light-sensitive proteins with unprecedented temporal resolution55,56,57, enabling the study of reactions initiated by ligand binding and exploiting the submicrometre crystal size for rapid reaction initiation49,50,51,54,56,57,58. In particular, SFX has advanced fibril studies, e.g., amyloids or microtubules, where the fibrous biomolecule assemblies may have partial or no crystallinity, approaching the regime of single-molecule imaging59,60.

Here we present MicroED and SFX structures of the MAL-induced MyD88TIR microcrystals at 3.0 Å and 2.3 Å resolution, respectively. Importantly, both structures show several distinct remodelled loop regions that adopt conformations that are different from previously determined monomeric X-ray and nuclear magnetic resonance (NMR) structures61,62. Crystal packing analysis revealed that the MAL-induced MyD88TIR crystals have a two-stranded higher-order assembly arrangement of TIR domains identical to that observed previously within spontaneously formed MALTIR filaments2, and mutagenesis studies demonstrated that the interfaces within these higher-order MyD88TIR assemblies are important for signaling. This identical architecture suggested a unidirectional templating mechanism for nucleation and assembly of the higher-order MyD88TIR oligomers, which we confirmed using crystal growth assays. Moreover, structural comparison of the MyD88TIRhigher-order assembly and monomeric MyD88TIR enabled us to understand the conformational changes that MyD88TIR monomers undergo upon joining the higher-order assembly. Collectively, our studies shed light on the hierarchical nature of the SCAF mechanism operating in TLR and IL-1R pathways.

Data acquisition

The MAL-induced MyD88TIR microcrystals were typically 100–200 nm in diameter, making them ideally suited to both MicroED (Fig. 2) and SFX.

a, b Electron micrograph of aggregated microcrystals, only showing poor-quality diffraction data. Scale bar, 1 µm. c, d Multiple microcrystals are overlapping, showing multiple lattices in their corresponding diffraction patterns, complicating data indexing. Scale bar, 1 µm. e, f Single hydrated microcrystal, showing high-quality diffraction data up to 3.0 Å resolution. Scale bar, 1 µm. The cyan rings on the micrographs indicate the 1.5 μm diameter parallel beam, defined by the selected area aperture, used for MicroED data collection. Electron diffraction patterns were collected with an angular increment of 0.68° per frame, at a dose rate of 0.12 e−/Å2 per frame. The data in a–f are representative of three EM grids prepared using 3 µl of a 1 : 50 MALTIR : MyD88TIR crystal solution.

Full size image

The microcrystals were deposited on Quantifoil EM grids and vitrified for screening and MicroED data acquisition (Supplementary Fig. 1). The microcrystals had a tendency to aggregate, forming large bundles that diffracted poorly (Fig. 2a, b). Furthermore, the bent and overlapping crystals complicated the data interpretation (Fig. 2c, d). Using a small parallel electron beam of 1.5 μm diameter, defined by the selected area aperture, only single thin hydrated microcrystals were selected for MicroED data collection (Fig. 2e, f). The MyD88TIR microcrystals diffracted to 3.0 Å resolution and provided high-quality electron diffraction data (Fig. 2f). Data from 18 crystals were integrated, scaled and merged (Table 1). The overall completeness is limited, owing to a preferred orientation of the MyD88TIRmicrocrystals on the grid and because of the limited tilt range of the goniometer.

The MyD88TIR microcrystals were studied in parallel using SFX. Initially, serial crystallography was attempted on a fixed target at the PETRAIII PII beamline, with a beam size of 2 × 2 µm. However, in this setup, data collection and analysis were complicated by the frequent bundling of microcrystals into larger aggregates (Supplementary Fig. 2). To reach higher resolution and overcome microcrystal aggregation, the sample was delivered as a stream of solvated microcrystals with a gas-dynamic virtual nozzle (GDVN) injector63,64 to a pulsed XFEL beam at the Linac Coherent Light Source (LCLS), SLAC National Acceleratory Laboratory65. By using a micro-focused beam (nominally 1 × 1 µm full width at half maximum (FWHM)), and optimizing crystal concentration (7.5 × 108 crystals/ml), high-quality diffraction patterns from individual crystals were collected. Overall, the SFX dataset comprised 4725 indexed patterns from 13,528 hits (35% indexing rate) out of 1,029,868 detector frames (average hit rate of 1.3%). The lattice parameters derived from the SFX data were found to be slightly larger than in the MicroED data collected under cryo-conditions (Table 1).

Structure solution, model building and refinement

The structure of MyD88TIR was initially solved using the MicroED data by molecular replacement, finding a well-contrasting unique solution in space group C2. The solution was found with a search model derived from a distantly related Toll-related receptor 2 (TRR2) TIR domain, sharing only 30% sequence identity with MyD88TIR (Fig. 3). The structure of MyD88TIR was iteratively built and refined using the MicroED data (Table 2) and, despite moderate completeness and resolution, the electrostatic potential map showed well-resolved features and enabled remodelling of the loop regions that differed from the previously determined monomeric crystal and solution structures61,62 (Figs. 3 and 4a). The higher-resolution SFX structure (2.3 Å) was first solved using the MicroED MyD88TIR model as a template for molecular replacement followed by iterative rebuilding and refinement using a different protocol compared to the MicroED structure (Table 2, SFXa). To enable a direct comparison between the MicroED and SFX models, we also solved, rebuilt and refined the SFX MyD88TIR structure using an identical protocol as described for the MicroED data (Table 2, SFXb). The SFXa map (Fig. 4b) showed well-resolved features, including water molecules that were not modelled in the MicroED structure. To check whether the MicroED and SFX maps were biased by the search model, simulated annealing (SA) composite omit maps were calculated, confirming the interpretation of our structural models (Supplementary Fig. 3). As the microcrystals contain a small proportion of MALTIRmolecules, there may be a contribution of this heterogeneity to the diffraction, but this is likely to have a negligible effect. Accordingly, there is no evidence of the presence of MALTIRmolecules in the electron density and electrostatic potential maps of the MAL-induced MyD88TIR crystals.

a Ribbon diagram (blue) of a monomer from the MyD88TIR higher-order assembly structure. Structural elements are labelled sequentially in TIR domains, with the BB-loop connecting strand βB with helix αB, according to the established nomenclature129. b–f Superposition of the MyD88TIRSFX structure (blue), with b the MyD88TIR MicroED structure (orange); c the monomeric MyD88TIRX-ray crystal structure (PDB ID 4EO7; magenta); d the monomeric MyD88TIR NMR solution structure (PDB ID 2Z5V; green); e the crystal structure of the TIR domain of the Toll-related receptor TRR2 from the lower metazoan Hydra vulgaris (PDB ID 4W8G; yellow); and f the MALTIR higher-order assembly cryo-EM structure (PDB ID 5UZB; red).

Full size image

Models and maps are presented of the remodelled BB loop (residues 186–204; top) and CD loop (residues 242–251; bottom) for the a MicroED and b SFXa structures. The carbon atoms in the MicroED and SFXa structures are shown in grey and green, respectively. Nitrogen, oxygen and sulfur atoms are shown in blue, red and yellow, respectively. The electrostatic scattering potential (MicroED) and electron density (SFX) 2mFo − DFc maps (blue isomesh) are contoured at 1.2σ, and the difference mFo − DFc maps (green and red isomesh for positive and negative density, respectively) are contoured at 2.8σ. No missing reflections were restored using weighted Fc values for map calculations.

Full size image

Structural comparison of MyD88TIR structures

The MicroED and SFXb MyD88TIR structures, which were built and refined using the same protocol, are almost identical, with a root mean square deviation (RMSD) of 0.4 Å for 138 Cα atoms. Minor differences in some side-chain conformations can be observed, which is most likely due to the flexibility of certain regions resulting in poorly defined electron density or as a result of the difference in the data collection temperature (Supplementary Fig. 4). The MyD88TIR SFXa structure was used for the comparison with other TIR domain structures and for the analyses of interaction interfaces within the crystal. The structure of MyD88TIRwithin the MAL-induced higher-order assembly exhibited conformational differences from the known NMR (RMSD of 2.4 Å for 107 Cα atoms)61 and X-ray (RMSD of 2.0 Å for 118 Cα atoms)62 structures of monomeric MyD88TIR. This is especially apparent in the region encompassing the BB loop and αB helix, and in the CD loop (Fig. 3). The conformational differences are likely due to participation of these regions in TIR : TIR interactions within the MAL-induced higher-order assemblies. Among the known TIR domain structures, the MALTIR filament structure (Fig. 3) and the TLR1, TLR2, TLR6 and IL-RACP crystal structures possess similar BB-loop and αB-helix conformations2.

MyD88TIR interaction interfaces in the microcrystal

Analysis of the crystal packing reveals MyD88TIR higher-order assemblies, each consisting of two offset parallel strands of TIR domains, with subunits in a head-to-tail arrangement forming each strand (Fig. 5a–c and Supplementary Tables 1–3). Formation of the MyD88TIRassemblies is mediated by two major types of asymmetric TIR domain interactions: one within each of the two strands (intrastrand interface) and one between the two strands (interstrand interface).

a Surface representation of the MyD88TIR microcrystal, consisting of two-stranded higher-order assemblies (black dotted lines). The two strands are shown in blue and magenta, respectively. bRibbon diagram of the MyD88TIR higher-order assembly. A yellow sphere indicates the N terminus of each TIR monomer and a red sphere indicates the C terminus of each TIR monomer. The two strands are shown in blue and green, and magenta and dark salmon, respectively. c Schematic diagram of the MyD88TIR microcrystals and the two types of asymmetric interactions within the higher-order assembly. BB surface consist of residues in BB loop; EE surface consist of residues in βD and βE strands, and the αE helix; BC surface consist of residues in αB and αC helices; whereas CD surface consist of residues in CD loop and the αD helical region. d, e Detailed interactions within the higher-order assembly d intrastrand interface and e interstrand interface. f Detailed interactions between the two-stranded higher-order assemblies, forming the sheet structure.

Full size image

Based on the SFX structure, the intrastrand interface involves opposite sides of the MyD88TIR domain, which together buries ~18.0–18.6% (1500 Å2) of the total surface area per subunit in the structure. It is composed of interactions between residues located in the BB loop of one subunit (BB surface) and the βD and βE strands and the αE helix on the next subunit (EE surface) (Fig. 5b–d and Supplementary Table 1). The highly conserved proline residue (P200 in MyD88) in the BB loop is buried in a shallow pocket between the βE strand and the αE helix consisting of residues I253, C274, L290 and A292. Hydrogen bonds (Supplementary Table 1) and a hydrophobic stacking interaction between the side chains of W284 and R196 stabilize the interface. The conformation of the BB loop is also stabilized by an internal salt-bridge between E183 and R196 (Fig. 5d).

The interstrand interface buries ~12.0–12.2% of the total surface area per subunit (991 Å2) and is composed of interactions between residues located on the αB and αC helices of one molecule (BC surface) and the CD loop and the αD helical region of the partner molecule (CD surface) (Fig. 5b, c, e and Supplementary Table 2). Several residues (W205, F235, K238, F239, L241, P245, I267 and F270) contribute hydrophobic interactions to this interface (Fig. 5e).

The interactions between the MyD88TIR two-stranded assemblies, which form a continuous sheet in the microcrystals, involve residues predominantly located in the αA helix and the CD and EE loops (Supplementary Table 3). The interface buries ~7–8% (570 Å2) of the total surface area per subunit and is less extensive than the intrastrand and interstrand interactions (Fig. 5f and Supplementary Table 3). These inter-assembly interactions are most likely analogous to non-biological crystal contacts in macromolecular crystals2,66.

Mutation of MyD88TIR intrastrand and interstrand residues perturbs assembly formation and signaling

We previously showed that alanine mutations of R196, D197, P200, W284 and R288 in the intrastrand interface, and K238, L241, S266 and R269 in the interstrand interface disrupted MAL-induced MyD88TIR microcrystal formation in solution2. To demonstrate the biological importance of the interaction interfaces, we tested the effect of interface residue mutations in a HEK293 TLR4 reporter cell line with an nuclear factor-κB (NF-κB)-driven mScarlet-I reporter and with endogenous MYD88 knocked out (Fig. 6a, b and Supplementary Fig. 5). Intrastrand mutations R196A, W284A, I253D and R288A abolished NF-κB activation by the TLR4 ligand lipopolysaccharide (LPS), whereas P200A in the BB loop substantially reduced activation (Fig. 6a and Supplementary Fig. 5d). In the interstrand interface, mutants K238A, L241A, F270A and F270E had little or no LPS response. An alanine mutation of F239 in this interface, which predominantly is involved in hydrophobic interactions with αB helix residues within the same subunit, only led to ~20% loss of activity. Mutants localized at the periphery of the interstrand interface had either intact signaling (P245H and R269A) or ~20% loss in activity (D234A). Mutant K282A, located at the interface forming the sheet structure that is considered not biologically important (Fig. 5f), also had intact signalling These signaling results agree very closely with analyses of the LPS-induced clustering of expressed MyD88 in cells (Fig. 6b and Supplementary Fig. 5e, f). The results are also consistent with our previous study on spontaneous and MAL-induced MyD88 clustering2, except that here, using a cell line deficient in endogenous MyD88, an effect of interstrand mutations can be clearly seen.

a, b Effects of MyD88 mutations on LPS-induced signaling and MyD88 clustering were tested in HEK293 cells expressing TLR4, MD2 and CD14, with MYD88 knocked out and stably transfected with an NF-κB-driven mScarlet-I fluorescent reporter. The cells were transfected with plasmids expressing wild-type or mutant V5-tagged MyD88, or empty vector, and then treated with (black bars) or without (grey bars) LPS (100 ng/mL) overnight, immunostained to detect MyD88-V5 and analysed by flow cytometry. Cells with very low expression of MyD88 were used for analysis to avoid spontaneous signaling (Supplementary Fig. 5b, c). The mean ± range from n = 2 independent experiments is shown. The death-domain mutation G80K, which has previously been shown to prevent MyD88 clustering130, and a TIR domain alone construct provided negative controls. a NF-κB activation measured by the geometric mean fluorescence intensity of the mScarlet-positive population relative to LPS-treated cells expressing wild-type MyD88. The dotted line indicates level of activation in cells with empty vector. b The percentage of cells with clustered MyD88 was determined based on the elevated height-to-area ratio of the MyD88 signal, which is observed when MyD88 clusters2 (Supplementary Fig. 5e). c Wild-type MyD88 and mutants were expressed in a cell-free system with an N-terminal GFP tag and the fluorescent samples were analysed by single-molecule spectroscopy on a home-made confocal microscope. To characterize the propensity of wild-type MyD88 and mutants to form higher-order assemblies, the average brightness values (equation (1)) of the proteins were calculated72. The results show that S209R, S244D, P245H and T281P mutants have higher propensity than wild-type MyD88 to oligomerize. The mean ± SEM of n= 3 or n = 2 (F270E and T281P) experiments using different lysate batches with two technical repeats per experiment is shown. The G80K mutant and the TIR domain were used as negative controls.

Full size image

Disease-related mutations and post-translational modification sites modulate assembly formation

Several MyD88 TIR domain missense mutations (V204F, S206C, I207T, S209R, S230N, M219T, L252P and T281P) sustain lymphoma cell survival due to constitutive NF-κB signaling67,68,69. Mapping of these residues onto the MyD88TIR assembly revealed that the S209R mutation is likely to directly impact interstrand interactions, whereas the T281P mutation may impact intrastrand interactions (Supplementary Fig. 6). L252 is buried and not directly involved in higher-order assembly interactions, but molecular dynamics simulations suggest that this mutation is likely to modulate the conformation of the CD loop70, which is critical for interstrand interactions in the MyD88 higher-order assembly. To directly test the hypothesis that these disease-related mutations increase MyD88 higher-order assembly formation, we analysed their effects on clustering in both cell-based and cell-free systems (Fig. 6a–c). Consistent with previous reports, expression of the S209R, L252P and T281P mutants in our reporter cell line showed increased basal NF-κB activation (Fig. 6a). L252P showed no further inducibility by LPS, whereas S209R and T281P were LPS responsive. All three mutants had increased basal clustering compared to wild-type (WT) MyD88, which was further increased by LPS for S209R and T281P (Fig. 6b). The aggregation propensity of these mutants was also evaluated by single-molecule spectroscopy, by measuring the brightness of the fluorescence time traces of cell-free expressed green fluorescent protein (GFP)-tagged proteins71 (Fig. 6c). The S209R and T281P mutants had increased aggregation propensity, forming larger particles than WT MyD88 (Supplementary Fig. 7). By contrast, the L252P mutant formed smaller particles than WT MyD88 (Supplementary Fig. 7), but the complexes were found in higher numbers and formed at lower protein concentrations, as previously reported72.

The MyD88 TIR domain has been reported to be phosphorylated on S242 (αC helix) and S244 (CD loop), with phosphomimetic mutations of these residues leading to opposite effects on NF-κB activation: the S244D mutation becomes hyperactive, whereas the S242D mutation has an inhibitory effect70,73. S242 forms a hydrogen bond with W205 in the MyD88 higher-order assembly and mutation of this residue to an aspartate is thus likely to destabilize the interstrand interface (Supplementary Fig. 6). S244 is not directly involved in higher-order assembly interactions, but similar to L252P, molecular dynamics simulations suggest that the S244D mutation causes a change in the CD loop conformation70. When the ability of MyD88 to cluster in HEK293 cells was tested (Fig. 6b), the S244D phosphomimetic mutation increased MyD88FL clustering, whereas S242D inhibited clustering, which is in perfect agreement with NF-κB activation by these mutants (Fig. 6a). Similar data were observed in the single-molecule assay, using cell-free expressed proteins (Fig. 6c). Overall, our new data strongly suggest that MAL-induced MyD88 TIR-domain clustering directly correlates with the level of NF-κB activation and therefore support the relevance of our structure as a model of MyD88 TIR domain association in vivo.

Comparison of MALTIR and MyD88TIR assemblies

To gain deeper insights into TIR-domain assembly formation, we compared the MyD88TIRmicrocrystal structure (Fig. 5) with our previously published cryo-EM structure of the MALTIR filament2. Both assemblies share a common overall architecture with head-to-tail intrastrand interactions mediated by the BB and EE surfaces, and interstrand interactions mediated by the BC and CD surfaces (Supplementary Fig. 8a). The conformations of the αE helix and the EE and CD loops are different in MyD88 compared to MAL (Fig. 3 and Supplementary Fig. 8a), resulting in an increase in the buried surface of both the intrastrand and interstrand MyD88TIR interactions (Supplementary Table 4).

The conformational differences in the αE helix and EE loop also lead to differences in the interface between the two-stranded higher-order assemblies (Supplementary Fig. 8b). In the MyD88TIR microcrystal, these interactions involve the αA helices and the CD and EE loops, whereas in the MALTIR cryo-EM structure the αA, αC and αD helices and the AA and EE loops contribute to these interactions. The differences in these interactions result in distinct packing of the two-stranded higher-order assemblies, MALTIR forming a tube consisting of 12 protofilaments, whereas MyD88TIR forms a continuous sheet (Supplementary Fig. 8c).

MALTIR nucleates MyD88TIR assembly formation unidirectionally

MALTIR nucleates the assembly of the MyD88TIR microcrystals2. The similar architecture observed in the MALTIR and MyD88TIR higher-order assemblies suggests a molecular-templating mechanism for nucleation and assembly, in which MALTIR serves as a platform to promote unidirectional assembly of MyD88TIR through intra- and interstrand interactions. To test this hypothesis, we captured MyD88TIR microcrystal growth using differential interference contrast (DIC) and fluorescence microscopy. Either MALTIR or GFP-MALTIRfusion proteins acted as nucleators of assembly formation and GFP-MALTIR nucleates the same type of MyD88TIR microcrystals as MALTIR (Supplementary Fig. 9). Short MALTIR-MyD88TIR crystal seeds were washed to remove MAL and then mixed with MyD88TIR. The results revealed that MyD88TIR assembly formation was unidirectional, with a substantial number of seeds observed with growth from one end only (Fig. 7a and Supplementary Movie 1). However, the tendency of MyD88 microcrystals to aggregate also presented a problem here, as the assemblies could also be seen growing in multiple directions from seed aggregates (Supplementary Fig. 10). GFP fluorescence is observed throughout the GFP-MALTIR:MyD88TIR crystal seeds, suggesting MALTIR can also incorporate within the MyD88TIR higher-order assembly, which is consistent with our previous report showing that MALTIR and MyD88TIR can form smaller heterogeneous complex structures when mixed at a 1 : 1 ratio2. As the concentration of GFP-MALTIR used for preparing the seeds (0.25–2 µM) is significantly lower than the critical concentration for MALTIR filament formation (30 µM)2, and the initial concentration of MyD88TIR is ~50–400× higher than GFP-MALTIR or MALTIR, the seeds must predominantly consist of MyD88TIR molecules, with a small fraction of MALTIR molecules localized at one end and also scattered throughout the seed. Furthermore, the MyD88TIR assemblies continue to grow after removal of GFP-MALTIR, demonstrating that MALTIR is only required for MyD88TIR assembly nucleation and not elongation.

a Time-lapse imaging of MyD88TIR microcrystal formation. Representative images of microcrystals growing from single GFP-MALTIR-MyD88TIR and MALTIR-MyD88TIR seeds are shown. The seeds were washed to remove MAL and then mixed with MyD88TIR. Data are representative of five independent experiments. Asterisks denote seeds with unidirectional growth. Scale bars: left panel 5 µm; middle and right panels 10 µm. b Ribbon diagrams of MyD88TIR (NMR solution structure of monomeric MyD88TIR (PDB ID 2Z5V) and higher-order assembly structure) and MALTIR (NMR solution structure of monomeric MALTIR (PDB ID 2NDH) and higher-order assembly cryo-EM structure (PDB 5UZB)), highlighting the rearrangement of the BB loop and αB helix (magenta in MyD88TIR and green in MALTIR) during the monomer-to-oligomer transition. c Two models of interstrand interactions, transitioning between MyD88TIR monomer (yellow) and MyD88TIR higher-order assembly (blue): (i) EE surface of MyD88TIR monomer docks onto BB surface of MyD88TIR higher-order assembly. This interaction does not require any conformational changes in the BB loop and αB helix to occur prior to binding. (ii) BB surface of MyD88TIR monomer docks onto EE surface of MyD88TIR higher-order assembly. This interaction requires significant conformational changes in the BB loop and αB helix to occur prior to binding and is therefore less favoured. d Model of MyD88TIR unidirectional assembly formation. The conformational changes in BB loop and αB helix required for the recruitment of new TIR domain subunits are induced by interstrand interactions. The higher-order assembly conformations of MALTIR and MyD88TIR, and the monomeric conformation of MyD88TIRare shown in orange, blue and yellow, respectively.

Full size image

To predict whether any of the inter- and intrastrand interface surfaces in MALTIR are preferred for the interaction with MyD88TIR, we calculated the predicted buried surface areas of possible MALTIR and MyD88TIR interactions. The calculations showed that the MALTIR BB surface–MyD88TIR EE surface interaction has the largest buried surface area (Supplementary Table 4). We also mapped the electrostatic potential on the surface of MALTIR and MyD88TIR, and found that the MALTIR BB surface and MyD88TIR EE surface are the only interaction interfaces that are highly charge complementary (Supplementary Fig. 11). Furthermore, molecular dynamics simulations on MALTIR : MyD88TIR complexes revealed that complexes involving the MAL BB and MyD88 EE surfaces are more stable than complexes involving the MAL EE and MyD88 BB surfaces (Supplementary Fig. 12). Consistent with these analyses, we have previously demonstrated that mutations in the MALTIR BB surface prevented full-length MAL-induced MyD88 clustering both in vitro and in cells (R121A, P125A and P125H)2.

We also compared the structures of MALTIR and MyD88TIR monomers with their respective structures within higher-order assemblies. This comparison revealed large conformational differences in the BB and BC surface regions (BB loop and αB helix), whereas the EE and CD surface regions adopt similar conformations (Fig. 7b). Models of the recruitment of monomeric MyD88TIR to a growing strand demonstrate that recruitment of new subunits to the assembly via their EE surfaces requires only minimal conformational changes prior to binding, whereas recruitment of new subunits to the assembly via their BB surfaces requires large rearrangements of both the BB loop and αB helix prior to binding, and would therefore be predicted to be less favourable (Fig. 7c). Overall, our structural analyses suggest that in the nucleation and elongation steps of MyD88TIR assembly formation, the EE surface of incoming MyD88TIR molecules dock onto the BB surface of MALTIR or MyD88TIR subunits. Interstrand interactions via BC and CD surfaces then trigger a rearrangement of the αB helix and BB loop in these newly incorporated TIR domain molecules, enabling them to interact with the EE surface of new incoming MyD88TIR subunits (Fig. 7d).

Over the last decade, crystallography has expanded in several different directions, both in terms of electron crystallography, through developments in MicroED8,9, and in terms of X-ray crystallography, through SFX45,46,47. Here we used MicroED and SFX to determine the structure of the MyD88 TIR domain from hydrated microcrystalline arrays at 3.0 Å and 2.3 Å resolution, respectively. Both of these techniques have their advantages and disadvantages. SFX utilizes high-intensity X-rays to generate high-resolution structures at room temperature and is able to use injector sample delivery systems to overcome crystal aggregation issues at the expense of high sample consumption (typically 0.3–12 mg of protein)74,75,76. By contrast, MicroED is able to minimize sample consumption (<1 μg protein) allowing for near-complete sampling of reciprocal space using the rotation method of vitrified microcrystals at cryogenic temperature using only a few or even just a single crystal. However, this can come at the expense of often having worse crystallographic quality metrics than is typically achieved in X-ray crystallography. Future advancements in this method, such as serial electron diffraction77,78, improved electron diffraction detectors, and accurate modelling of the electrostatic potential, taking into account the charged state of atoms and the potential distribution, are likely to improve map quality and provide information about charge interactions30,79. For SFX, developments in mix-and-inject experiments at XFELs using nano-focused X-ray beams80,81 alongside advancements in data analysis82 will provide future opportunities to conduct time-resolved studies of protein assembly formation. The eventual goal of structural biology at XFELs is to try and push the limits of signal-to-noise, to the point where it is possible to image single molecules in solution83.

In our investigation, only subtle differences were observed between the MicroED and SFX structures, which may be explained by the differences in the data resolution and completeness, flexibility of certain regions, and difference between cryogenic and room temperature data collection. To our knowledge, only one other group has reported a comparison of these two techniques on the same protein crystal system24. Their work showed a slight expansion of the unit cell in the SFX case, which was linked to differences in the data collection temperature. Our room-temperature SFX data also showed a slight increase in lattice parameters along the a-axis, when compared to the cryogenic MicroED data (Table 1), indicating the lattice change is related to the temperature difference between the two data sets.

SCAF, which involves assembly of higher-order oligomers for transmission of receptor activation information to cellular responses, is an emerging theme in signal transduction4and operates in several innate-immunity and cell-death pathways including inflammasome signaling84, RIG-I-like receptor85 and TLR pathways2,5. In this study, we found that the MAL-induced MyD88TIR crystalline assemblies contain a two-stranded head-to-tail arrangement of TIR domains, as previously observed for the TIR domains of the adaptor protein MAL2. Analysis of single amino-acid MyD88 mutations for their effect on cellular signaling support the biological relevance of the defined interfaces. Previous functional analyses have measured spontaneous signaling by MyD88 overexpressed in HEK293 cells70. Our analysis here has several advantages. First, we used cells with endogenous MYD88 knocked out, which gives a more stringent determination of the function of mutants. This improvement allowed us to demonstrate the importance of residues in the interstrand interactions, which were not apparent in our earlier study2. Second, through the use of flow cytometry, we can analyse single-cell responses and select only cells with MyD88 expressed at very low levels to avoid spontaneous signaling. This gives us the ability to observe the response of the mutants to LPS treatment in an intact signaling pathway, avoiding artefacts of overexpression. With this technique, we demonstrated that the R196A mutant is completely inactive, whereas prior work showed it promoted 56% of WT NF-κB activity in the presence of endogenous MyD88, despite having defective TIR domain interactions70. Consequently, we are confident in the biological relevance of the signaling assay reported here, which confirmed the importance of several critical residues in both the intra- and interstrand interfaces of the MyD88TIR assembly.

We provide evidence demonstrating that MALTIR serves as a platform to promote unidirectional assembly of MyD88TIR oligomers. One feature of unidirectional elongation is establishment of hierarchy in the higher-order oligomers, in which upstream molecules can nucleate the assembly formation of downstream molecules, but not vice versa, and appears to be a common feature in many innate-immunity pathways. For example, elongation of the BCL10 adaptor in the CARMA1–BCL10–MALT1 assembly is unidirectional, with growth at one end only as revealed by confocal imaging86, and structures of the RIG-I : MAVS CARD, the FADD : caspase-8 DED and the MyD88 : IRAK4 : IRAK2 DD assemblies revealed that the RIG-1, FADD and MyD88 oligomers recruit their downstream partners via only one CARD, DD and DED surface, respectively5,85,87.

Our data add support to a sequential and cooperative mechanism for TLR signal transduction, in which receptor and adaptor TIR domains assemble via the inter- and intrastrand interactions observed in the MyD88TIR and MALTIR higher-order assemblies, leading to formation of a TIR-domain signalosome. This would then promote clustering of MyD88 DDs to form the Myddosome, with recruitment and activation of IRAKs5. The Myddosome defined in vitro is a helical array of DD of MyD88-IRAK4-IRAK2 in a 6 : 4 : 4 arrangement5. In contrast to this mechanism suggesting stepwise recruitment of MyD88 proteins, it has recently been proposed that some MyD88 pre-exists in unstimulated cells in a free oligomeric complex via DD interactions, but cannot recruit IRAK4 due to the TIR domain blocking access to the IRAK4 binding surface6. Upon receptor activation, it is proposed that MyD88 TIR domains are recruited into the TLR4TIR-MALTIR signaling complex, releasing the autoinhibition and enabling recruitment of IRAKs to the pre-formed MyD88 oligomer. Further data are needed to validate either of these models, but there are a number of caveats regarding the possibility of pre-formed autoinhibited complexes. First, MyD88 DD surfaces involved in IRAK4 interactions are also required for the assembly of MyD88 DDs into a hexamer and binding of MyD88 TIR domains to these surfaces is likely to prevent DD oligomer formation altogether. Second, there is a sharp concentration dependence for oligomerization of both full-length MyD88 and MyD88 DD in vitro72 and the threshold for MyD88 clustering in cells is readily exceeded by overexpression. The spontaneous signaling seen with overexpression88 argues against MyD88 clusters being intrinsically inhibited for IRAK recruitment. At normal cellular concentrations, autoinhibition is likely to play a role in limiting self-association of MyD8872. Stepwise TIR domain-mediated recruitment into a TLR signalosome would then increase the local concentration of DD, leading to Myddosome assembly.

In conclusion, our study provides new insights into the architecture and assembly mechanism of TIR-domain signalosomes in TLR pathways, and at the same time allows for a comparison of the complementary techniques of MicroED and SFX. The detailed TIR : TIR interactions reported in this study may also provide templates for designing small-molecule mimics of the important interfaces to inhibit MyD88 higher-order assembly formation for potential therapeutic applications.