First, it was necessary to look at the self-assembly of full-length MyD88. Because both the TIR domain and the DD have the ability to self-assemble, the contribution of both domains in the behaviour of the full-length protein was of interest. Full-length MyD88 is difficult to express and purify recombinantly in E. coli, presumably due to its polymerisation propensity. Here we expressed the proteins in vitro, at controlled low concentrations, and studied their self-association in undisturbed samples using single-molecule counting techniques. More precisely, we used an in vitro translation system derived from Leishmania tarentolae [22] (Leishmania tarentolae extract, LTE). This eukaryotic system enables the rapid production of proteins (typically within 2 h) and the analysis of protein-protein interactions in a system orthogonal to the human proteome. By controlling the concentration of DNA priming the expression system, we can tune the final expression levels of proteins and co-express proteins at controlled ratios. We have tested this combination on a variety of biological systems and have demonstrated that the flexibility of cell-free protein expression is a great asset to study protein self-assembly [20, 23,24,25,26].
To determine the aggregation and oligomerisation propensity of proteins, we have developed diverse “counting” methods based on single-molecule fluorescence techniques. In single-molecule fluorescence spectroscopy, rare protein complexes can be easily detected in a background of monomers and their size can be evaluated by simply counting the number of fluorophores present in each complex. As we demonstrated recently in our study of the prion-like behaviour of ASC, these counting methods are well suited to study the heterogeneous processes of protein oligomerisation and polymerisation [20]. To visualise MyD88 and its mutants, the proteins were expressed as fusions with genetically encoded GFP or mCherry fluorophores and could be measured directly upon expression without further labelling, purification or enrichment steps. The levels of fluorescence obtained provided a direct readout of protein expression levels, after careful calibration with GFP/mCherry protein controls.
First, constructs containing the TIR domain alone (residues 159–296), death domain alone (residues 1–117, to include crucial part of the intermediate domain (ID)) and full-length MyD88, each fused to GFP at the N-terminus were expressed in LTE. After expression, samples were directly measured on a confocal microscope. A 488-nm laser was focused into the sample, creating a small focal volume, through which proteins could freely diffuse due to Brownian motion. In the range of concentrations used, multiple fluorophores are always present in the focal volume and as proteins constantly exchange in the detection volume, we ultimately interrogate a large number of proteins. Fluctuations of fluorescence intensity were recorded using high-speed single-photon counters; typical fluorescence time-traces obtained are shown in Fig. 1a.
At low concentrations, both the TIR and death domains are required for efficient polymerisation of MyD88
The fluorescence time-traces obtained for the full-length and separate domains of MyD88 exhibit very distinct characteristics. For TIR domain alone (Fig. 1a, in green), the fluctuations of intensity around the average value are limited (± 500 photons/ms); however, for the death domain (Fig. 1a, in blue), small bursts of intensities (> 1500 photons/ms above background) can be detected. These fluorescence peaks correspond to entries of single protein complexes, increasing the local number of proteins for a brief period of time. The amplitude and duration of the deviations from average are linked to the number of proteins co-diffusing in a single complex and the physical size of the diffusion complex. For full-length MyD88, we observe extremely bright and long-diffusing bursts of fluorescence, as shown in red.
The simplest analysis to quantify the presence of protein complexes is to calculate the average brightness of the diffusing species [20, 27, 28]. The brightness parameter B is calculated from the measured intensity values (I) as
$$ B=\frac{{\left(\mathrm{Standard}\ \mathrm{deviation}\ (I)\right)}^2}{\mathrm{average}\ (I)} $$
(1)
The main advantage of this B parameter is that it is independent of protein concentrations—in the absence of self-association, the B values should be constant as a function of expression levels, and increases of B values report on the formation of complexes. Variation of the final levels of protein expression was achieved by using serial dilutions of the priming DNA template in our cell-free expression system, with lower DNA concentrations resulting in lower protein concentrations. For each experiment, fluorescence over time was recorded and the brightness parameter was calculated and plotted as a function of protein concentration. As shown in Fig. 1b, all three constructs display self-association compared to the GFP control. TIR alone forms relatively small oligomers, and the concentration-dependence shows that TIR domains self-assemble at approximately 50 nM. The DD construct forms larger assemblies by itself and the concentration dependence shows a sharp increase into self-assembly at 60 nM. The full-length protein displays much higher brightness values, with the concentration for self-assembly into polymers at approximately 120 nM.
The average of the brightness values measured above threshold (> 150 nM) can be normalised by the monomeric sfGFP control. As shown in Fig. 2a, the average diffusing complexes formed by TIR and DD are small (approximately four and eight times brighter than monomeric GFP), while the full-length MyD88 has a greatly increased B value (100-fold larger). As the samples are heterogeneous, the precise number of proteins in the assemblies cannot be inferred directly using the average brightness value; however, it indicates that the TIR, DD and full-length (FL) MyD88 all form oligomers, with the full-length protein forming much larger species.
To characterise these assemblies in more detail, we performed fluctuation correlation spectroscopy (FCS) experiments (Fig. 2b), which report on the physical size of the diffusing particles. In the case of heterogenous samples with different brightnesses, the larger, brighter species contribute more (to the square of the oligomer size) to the autocorrelation function (G(τ)) than less bright, monomeric species. Here, a model of a single diffusing species fit the monomeric GFP, TIR, DD and FL data well. Fits were not improved by the addition of further components to the model. The average hydrodynamic radii obtained are plotted in Fig. 2c and normalised to sfGFP. Qualitatively, this data agree with the previous results, with the TIR, DD and FL protein complexes progressively increasing in average size. For the full-length protein, the FCS data show the presence of large species, probably higher order polymers. The average diffusion time is around 100-fold slower (i.e. hydrodynamic radius 100-fold larger) than that of sfGFP alone, suggesting that the higher order species are composed of > 100 monomeric units. Note that the monomeric unit of GFP-fused FL MyD88 would have a radius ~ 1.4× larger than sfGFP.
Comparisons of the brightness values and FCS data obtained confirm that in our hands, the TIR domain species are low-order oligomers (Fig. 2a-c), at concentrations > 50 nM (Fig. 1b). The absence of large events is consistent with the findings by Ve et al. that MyD88 TIR alone does not spontaneously polymerise [12]. Similarly, our data obtained on the separate death domain demonstrates that it too consistently forms small oligomers, with a hydrodynamic radius eight to ten times larger than monomeric GFP. This data complements previous myddosome data that shows assemblies of six to eight MyD88 DDs [29]. At our concentrations (< 300 nM), only full-length MyD88 is able to form very large assemblies, with a hydrodynamic radius 100-fold larger than sfGFP. Interestingly, the two domains seem to cooperate in the formation of higher order structures, and the full-length assemblies are much larger than the sum of the two individual domain oligomers. This cooperativity also appears to delay the self-assembly of the full-length protein, as the self-assembly transition occurs at higher protein concentrations (120 nM vs 50 nM). Overall, our in vitro results are consistent with the recent single molecule fluorescence microscopy studies by Latty et al. [29] demonstrating the formation of both smaller (approximately six MyD88 complexes) and “super” myddosomes at the cell surface. Our single-molecule traces show clearly the presence of anomalously bright events with > 100 proteins co-diffusing simultaneously.
MyD88 aggregation is a concentration-dependent, self-templated polymerisation event
Figure 1b shows that the aggregation of full-length MyD88 is a concentration-dependent process and reveals a sharp transition in behaviour at around 120 nM.
To confirm that MyD88 filaments can template the conversion from a soluble, monomeric species to a fibrillar form, we used a two-colour seeding assay (Fig. 3a). Briefly, full-length MyD88 tagged with mCherry was expressed at a concentration at which filaments readily form (~ 250 nM). Filaments were enriched by gentle spinning and sonication before being added to solutions containing MyD88 tagged with GFP expressed across a range of concentrations, as previously described. Self-templating was then investigated by two-colour single-particle coincidence spectroscopy. For these experiments, two lasers (488 nm and 546 nm) are focused on the same focal volume, allowing both mCherry and GFP-tagged proteins to be detected at the same time. A typical fluorescence time-trace demonstrating two colour-coincidence experiments with MyD88 is shown in Fig. 3a. The initial expression of GFP-tagged MyD88 is at subcritical concentrations. In the absence of mCherry-tagged MyD88 sonicated filaments (“seeds”), the GFP trace shows little fluctuation confirming that MyD88 is monomeric at this concentration. The mCherry-tagged MyD88 seeds were then added to the mixture and were detected in the mCherry channel. If GFP is recruited to the mCherry seeds, the presence of coincident bursts of fluorescence in both channels results. Indeed, within 20 s, large bursts of fluorescence in the GFP channel were seen and found predominantly to coincide with the presence of mCherry peaks, indicating that MyD88-GFP is recruited to the MyD88 mCherry seeds (Fig. 3b). Moreover, with time, the events detected in the GFP channel became brighter than those in the Cherry channel, indicating that the GFP-tagged MyD88 grow off the mCherry-tagged MyD88 seeds.
Figure 3c shows that seeding of MyD88 polymerisation occurs over a large range of concentrations. This is particularly obvious within the range of sub-threshold concentrations, where polymerisation does not normally take place on the timescale of our experiment. Ultimately, this allowed us to define the critical concentration for the polymerisation of FL MyD88. Below this critical concentration (10 nM), full-length MyD88 does not polymerise, even in the presence of seeds. In the supercritical zone (> 120 nM), full-length MyD88 can spontaneously polymerise, but addition of seeds increases the effect and the plateau in brightness values is reached earlier. A large metastable zone (10–120 nM) exists, where the tendency of MyD88 to polymerise on its own is low in the timescale of our experiment, but can be catalysed by the presence of the polymeric seeds.
Biologically, the existence of this metastable zone is important, as it shows that rapid amplification of MyD88 signalling can be achieved through seeding. The “in vitro” seeding is the introduction of MyD88 filaments; however, in vivo, seeding can be triggered by upstream proteins such as the recruitment of MyD88 through Mal nucleation. The depth of the metastable zone is also important: if this zone is too narrow, the system would respond too fast, initiating the highly effective pro-inflammatory innate immune response prematurely. A large metastable zone is therefore more physiologically desirable [30].
Disease-associated point mutations abrogate MyD88s ability to optimally polymerise
Having established that full-length MyD88 can undergo an active polymerisation process, we then investigated whether pathological point mutations could affect this protein polymerisation propensity. Hence, the L93P, R196C and L252P point mutations were individually introduced into the GFP-tagged full-length MyD88. Once again, expression of tagged MyD88 by the cell-free translation system was used and fluorescence time-traces were measured and plotted as distributions of fluorescence intensities.
In Fig. 4a, typical fluorescence time-traces obtained when all proteins were expressed at 150 nM concentrations reveal a different profile for the mutants, compared to the wild-type (WT) protein, with a loss of the brighter objects for all mutants. This is confirmed by the FCS data that demonstrate a reduction in sizes of the larger protein species, compared to the WT protein (Fig. 4b). Brightness profiles of the full-length MyD88 mutants were compared to those obtained for the isolated domains (Fig. 4c, d, e). The full-length MyD88 L93P mutant polymerisation profile roughly mimics that of a MyD88 TIR domain alone, while the full-length R196C and L252P mutant polymerisation profiles show a behaviour in between those of MyD88 DD alone and full-length MyD88. This suggests that the two point mutants have a higher tendency to oligomerise than the isolated DD, but they do not support the formation of the higher order assemblies observed with the WT protein. Overall, it appears that the point mutations decrease the capacity of the domains to contribute to polymerisation, possibly through impairing homotypic protein-protein interactions (PPIs).
L252P mutants form stable oligomers at a 40-fold lower concentration than wild-type MyD88
We then examined the behaviour of the mutants as a function of protein expression, taking advantage of the control one can exert with the cell-free translation system. Figure 5a shows the differences in polymerisation profiles exhibited by the mutants at the same low 3 nM concentrations. This contrasts with the profiles in Fig. 4, obtained at 150 nM. At this low concentration, we do not detect the presence of large objects for WT MyD88 or for any of the mutants and the traces obtained for WT, L93P and R196C MyD88 suggest the presence of mainly monomeric species. In contrast, L252P still seems able to oligomerise, as indicated by the presence of fluorescence bursts. To confirm this unexpected effect, the oligomerisation thresholds of the mutant MyD88 proteins were measured and analysed by plotting the B parameter as a function of protein concentration (Fig. 5b). In the case of R196C and L93P, the B values never reach those of the wild-type protein, indicating that the pathological point mutants alone cannot propagate polymerisation, no matter what concentration of protein is achieved (within the range of our experiment). The L252P mutant also never formed the large aggregates that are observed with WT MyD88 when expressed within our system. Strikingly though, at very low concentrations, where WT MyD88 and the other disease-associated point mutants exist only as monomers, the L252P mutant still forms stable low-order oligomers (Fig. 5b). The threshold for oligomerisation is extremely low (approximately 2 nM), in the subcritical zone of WT MyD88. Interestingly, this threshold concentration correlates with the concentration above which WT MyD88 can be caused to polymerise by seeding (Fig. 3c), suggesting that the L252P oligomers could be acting as an activated form of MyD88 [7].
The presence of L252P oligomers had been postulated previously based on from computational model studies [31], which predicted the existence of these oligomers at levels that are physiologically present in inactivated cells, i.e. without expression upregulation upon receptor-ligand binding and activation. This fits well with our observations and our data confirm the existence of these extremely stable low-order oligomers of MyD88.
Mutations within the same domain can lead to contrasting protein properties
Our data also show a drastic difference in the behaviour between L252P and R196C mutants, even though both residues are located within the same TIR domain. Differences in oligomerisation pattern could potentially explain the differences in the related pathologies, with the L252P protein producing stable oligomerisation at low concentrations leading to cancer, while the R196C protein producing a lack of oligomerisation/polymerisation propensity, leading to a dampening of the innate immune response to bacterial infection. However, these are not the only differences uncovered between these disease-associated mutants. The L252P mutation is a dominant mutation, whereas L93P and R196C are both recessive mutations. Because primary immunodeficiency only affects homozygous or compound heterozygous carriers of the point mutations L93P and R196C, we hypothesised that polymeric propagation could be rescued by the presence of the wild-type protein. To test this, GFP-tagged mutants and mCherry-tagged WT MyD88 were co-expressed in LTE and subjected to our brightness assay. The brightness parameters for the mutants obtained through single or co-expression could then be compared (Fig. 6a). In the case of L93P and R196C, the GFP brightness value is significantly larger upon co-expression, indicating that higher order polymers of the mutant MyD88, when expressed with WT FL mCherry MyD88, were forming. Indeed, examination of the fluorescence time-traces reveals the presence of coincident peaks (Fig. 6b–d), showing that WT MyD88 can recruit the mutants into its polymers. The overall degree of polymerisation is still lower than in the case of the wild-type protein alone, but the ability of the system to form large objects may be sufficient to restore normal signalling.
In contrast, the brightness of L252P upon co-expression is unchanged (Fig. 6a), indicating that the mutant species oligomerises regardless of whether the wild-type protein is present. Furthermore, few coincident peaks were detected (Fig. 6d), showing that WT MyD88 does not recruit this mutant into its polymers as readily as L93P and R196C (Fig. 6b, c). The differential incorporation of the mutants into the wild-type polymers correlates well with what is observed at the physiological level. Heterozygous patients carrying the L93P or R196C mutations do not suffer from the recurrent bacterial infections. The wild-type protein polymerisation, as well as the incorporation of the mutants into the polymerising wild-type protein, albeit at suboptimal levels (Fig. 6a), may be sufficient to propagate signalling efficiently. Impairment of polymerisation and subsequent signalling is only observed in the absence of wild-type MyD88, as would be the case for homozygous and compound heterozygous carriers (i.e. both alleles of the gene harbour mutations such as L93P and R196C) [4, 5]. In the case of L252P, a distinct population of finite-sized oligomers seems to always exist, irrespective of the presence of WT MyD88 (Fig. 6a, d). This would correlate with the fact that both the heterozygous and homozygous patients suffer from associated cancers [32].
L252P can seed WT MyD88 and recruit IRAK4
To test whether small oligomers of L252P could serve as seeding events for WT MyD88, we again used our seeding assay. Here, WT Myd88 full-length tagged with mCherry was expressed as a monomeric protein (Fig. 7a, grey traces). Upon addition of separately expressed GFP-L252P, peaks are detected in the red channel (Fig. 7a, black traces) indicating that WT MyD88 is now self-associating. As shown in Fig. 7b, only L252P, and not L93P or R196C, is able to induce an increase in WT MyD88 brightness and therefore induce MyD88 polymerisation.
An important question about the L252P oligomers is whether they are signalling competent. In vivo and cell data support this hypothesis. Studies have demonstrated that IRAK4 inhibition promoted killing of ABC DLBCL lines harbouring MyD88 L252P, by down-modulating survival signals, including NF-κB [33]. The established link between the L252P mutation and the occurrence of cancer allows us to hypothesise that the stable oligomers formed by this mutant may be all that is required for constitutive signalling.
In our system, we tested the ability of the L252P mutant to recruit IRAK4, as a proxy for its ability to signal. GFP-tagged WT MyD88 and mutants was added to a solution of His-tagged IRAK4. IRAK4 was fluorescently labelled during synthesis in the LTE system by addition of bodypi-lysines. GFP-nanotrap presenting sepharose beads were used to immuno-precipitate the GFP-tagged MyD88 constructs. The beads-bound fraction were then treated at 95 °C to release the proteins from the GFP-nanotraps. This treatment results in unfolding of GFP and loss of fluorescence of the MyD88 constructs, but does not affect the fluorescence of the bodypi-labelled IRAK4. Therefore, the amount of the IRAK4 that has been co-immunoprecipitated can be detected easily on a SDS-page gel scanned for fluorescence. This experiment shows that L252P is capable of recruiting IRAK4 to the same extent as the WT MyD88. On the contrary, L93P and R196C have a reduced capacity to recruit IRAK4 compared to WT MyD88. This supports the idea that L252P could act as an activator of MyD88. Further validation in vivo will be required to fully characterise the mechanisms leading to enhanced NFkB signalling observed in previous studies [34].