TLR4-MyD88 expression and significance in PD-relevant brain regions
We first explored the relevance of TLR4/TLR2-MyD88 signalling in human brain using publicly available cell-specific transcriptomic data . RNA sequencing of purified cell types derived from human cortex demonstrated that while TLR2 is expressed almost exclusively in microglia, TLR4 and MyD88 are expressed by human astrocytes and microglia. Furthermore, comparison with murine cortex showed that TLR4 expression in human glia is higher than in mouse (Supplementary Fig. 1)  and that while TLR4 is predominantly expressed by mouse microglia, in humans mature astrocytes also express high levels of this gene .
Given that cell type-specific RNA sequencing data in humans was derived from only one brain region (temporal cortex), we used publicly available data derived from post-mortem control human brain to further explore TLR4, TLR2 and MyD88 expression across the human brain, including within the substantia nigra and putamen. TLR4 and MyD88 expression was detectable across the human brain using data provided by the UK Brain Expression Consortium (UKBEC,  and Genotype-Tissue Expression Consortium (GTEx; ) (Fig. 1), which use independent sample sets and different platforms for measuring gene expression. The substantia nigra and putamen were amongst the brain regions expressing the highest levels of these genes according to both resources, though UKBEC data did not include information on the spinal cord. However, we noted some discrepancy in the pattern of MyD88 expression in particular between the UKBEC and GTEx resources, with GTEx reporting MyD88 expression as highest in the cerebellum, whereas UKBEC reported highest expression in substantia nigra. This discrepancy may be due to differences in the platforms used to measure expression or in the brain donation and collection procedures used by the brain banks from which samples were derived . In contrast, TLR2 expression was at the detection limits of microarrays  and had very low expression when measured using RNA sequencing .
Given the high expression of TLR4 and MyD88 in the substantia nigra, we investigated possible functional relationships between these genes and genes causally implicated in PD through genome-wide association studies (GWAS). Since the vast majority of risk loci for PD are non-coding with the potential to be associated to multiple genes, we used the versatile gene-based association study method (VEGAS)  to assign risk SNPs to genes, basing the analysis on GWAS data from the International Parkinson’s Disease Genomics Consortium . This approach incorporates information from all SNPs within a gene and accounts for linkage disequilibrium between SNPs using simulations from the multivariate normal distribution to produce gene-based (as opposed to SNP-based) association p values. Next, we applied gene co-expression analysis to transcriptomic data from control human brain, originating from UKBEC (see Supplementary Fig. 2 for a visual description of the pipeline) with the aim of grouping genes into biologically meaningful modules in an unsupervised manner. In this way, we hoped to identify novel gene relationships, including possible relationships between TLR4-MyD88 signalling and genes associated with complex PD. Finally, we tested all gene co-expression modules for evidence of enrichment of genes associated with PD (as determined using VEGAS) and results showed a significant enrichment within MyD88-containing modules in the putamen and medulla (putamen, enrichment p value = 2.20 × 10−3; medulla, enrichment p value = 2.92 × 10−2; Supplementary Table 1).
Interestingly, these modules were not only significantly enriched for PD-associated genes, but for microglial markers (putamen module p value = 3.84 × 10−9; medulla module p value 3.05 × 10−7) with some evidence for the more specific involvement of activated (Type 2) microglia in putamen (putamen module p value 1.01 × 10−4). To robustly asses the genetic evidence for the specific involvement of type 2 microglia in PD, we used stratified LD score regression to determine if functional marks for activated innate immune cells were enriched for genetic heritability. Given the absence of state-specific data for human microglia, we used expression-quantitative trait loci (eQTLs) generated from monocytes at rest and following activation to address this question [14, 39]. This approach demonstrated significant heritability enrichment for PD in eQTLs identified in lipopolysaccharide (Bonferroni-corrected p value = 1.05 × 10−2), but not interferon-γ-activated monocytes (Bonferroni-corrected p value = 0.20) or untreated monocytes (using data from Raj et al. Bonferroni-corrected p value = 0.39; using data from Fairfax et al. p value = 0.21). Given that monocyte activation by lipopolysaccharide occurs through TLR4-MyD88 signalling these findings suggested that TLR4-mediated activation of microglia/astrocytes generated a cell state of key importance in the development of PD. This finding is consistent with previous reports highlighting the involvement of the immune system in sporadic PD [5, 17, 38, 51].
Characterization of the oligomeric α-syn used in cell experiments
To determine the mechanism by which physiological concentrations of α-syn oligomers activate microglial and astrocytes, we performed experiments on cells in culture so that experiments could be performed under controlled conditions, using well-characterized solutions of small soluble oligomers. We generated synthetic aggregates of unlabeled α-syn by taking aliquots of α-syn solution directly from an aggregation reaction at the end of the lag phase. Using single-aggregate imaging with Thioflavin T (ThT), we counted the number of aggregates present to ensure that each reaction was reproducible .
Our previous work using single-molecule Förster Resonance Energy Transfer (sm-FRET) aggregating the same concentration of dye-labeled α-syn monomer showed that approximately 1% of the total solution would be oligomeric [7, 23]. We have confirmed these sm-FRET results using both SEC  and TEM . The small-soluble high-FRET oligomers that are formed under these conditions, by a slow conversion process from low-FRET oligomers, were found to be the most toxic species producing increased ROS when added to a neuron and astrocyte co-culture [7, 23]. Once formed, the oligomers are stable for several hours enabling us to dilute the stock aggregation mixture to produce solutions of known initial oligomer and monomer concentration. However, since aggregation was performed in a slightly different buffer to be compatible with cell culture, we repeated the oligomer characterization.
Experiments using sm-FRET [7, 23] showed that the % of oligomers in our preparation was 1.5 ± 0.5% (n = 3, SD) at 26.5 h of incubation and 1.7 ± 0.4% (n = 3, SD) after 50 h (26.5 h of incubation with shaking followed by 24.5 h of static incubation) see Supplementary Fig. 3. The derived FRET efficiency histograms were centered at E = 0.65 (Eq. 3), corresponding to high-FRET oligomer type, also in good agreement with previous measurements [7, 21, 23] and confirmed that they were smaller than 20 mers.
It is also possible that additional oligomers are formed during the incubation with cells for 24 h before buffer is exchanged. To test this, we performed experiments with unlabeled protein. Both monomer only and preformed oligomers were added to cells (Supplementary Fig. 4). α-syn monomer was only incubated with BV2 microglia cells at concentrations ranging between 10 pM–4000 nM and left for 24–120 h. The supernatant of the cells was then removed and analyzed by either ThT fluorescence or ELISA. Under conditions where α-syn monomers were incubated with the microglia cells at a concentration less than 1 nM, no significant (p = 0.8) increase in fluorescence was observed using bulk ThT assays and no increase in cytokine production was recorded with ELISA’s. 200 nM monomer left for 24 to 48 h also showed very minimal aggregation but oligomer formation was detectable for longer times, since an increase in TNF-α production and ThT signal was observed and aggregates could be detected by direct imaging. Similarly, 1000 nM monomer left for 24 h showed minimal aggregation but significant TNF-α production beyond this time compared to the media control (p = 0.006). We also added preformed oligomers, with monomer, to cells. Direct imaging of the number of aggregates in the solution before and after incubation with cells for 24 h, using ThT imaging, confirmed that there was only a detectable increase in oligomer number at 1000 nM monomer (approximately double) and no detectable increase at lower monomer concentrations. Overall, these experiments show that for incubations below 1000 nM monomer and 15 nM oligomers, there is no significant increase in oligomer concentration over the 24-h incubation.
To determine how long we could perform experiments, cell survivability experiments were performed. Cells were grown to a density of 0.4 × 106 cells/ml and α-syn oligomers in new cell media introduced. Supernatants were taken each day and fresh buffer added, after wells were washed in LPS-free PBS. Four concentrations of oligomeric α-syn were used to test the effect oligomer concentration had on survival. An approximate 90% survival (compared against hour 0) in cell population was observed up to the 96 h mark with 1000 nM oligomer or lower, see Supplementary Fig. 5.
Oligomeric α-syn-induced cytokine production is predominantly generated by TLR4-dependent MyD88 signalling
Initial experiments established that soluble oligomers, but not monomers or fibrils, induced an inflammatory response from the same initial preparation of α-syn monomer (Fig. 2). This confirmed that there was minimal or no LPS contamination in the α-syn preparation. Immortalised murine microglia cells (BV2) were incubated with monomeric, oligomeric or fibril forms of α-syn at varying monomer concentrations and left at 37 °C for 24 h. Levels of TNF-α, IL-1β and NO were then measured in the cell supernatant by ELISA (Fig. 2). As expected, the incubation of these cells with the TLR4 agonist LPS (at ng/ml concentrations) induces robust activation of this receptor.
The α-syn oligomers (70 µM total monomer containing approximately 100 nM oligomers) produced 750 pg/ml of TNF-α in comparison to the same concentration of the monomer or fibril solutions (40 and 25 pg/ml, respectively). At this oligomer concentration, considerable cell death occurred by 24 h (33% cell death compared to day 0, see Supplementary Fig. 5a). Increasing concentrations of fibril or monomer samples (Fig. 2c) failed to show a significant increase in TNF-α production confirming that only oligomers cause an inflammatory response, as previously reported [6, 8].
To determine which receptors were involved in the immune response, previously characterized immortalized murine bone marrow-derived macrophage cells were used derived from mice lacking either TLR4, TLR2 or MyD88 [22, 24, 46]. These cells were incubated with different concentrations of α-syn oligomers or monomers for 24 h. The cytokine levels present in the cell supernatant were then analyzed by ELISA (Fig. 3). As expected in wild-type cells, cytokine production was seen only with oligomers and not monomers.
In TLR4−/− cells, the inflammatory response to oligomers (TNF-α, Il-1β and NO) was significantly reduced compared to wild-type cells, ~ 10 to ~ 100 fold decrease. Likewise, in MyD88−/− cells and to a much lesser degree in TLR2−/− cells there was statistically significant decreased TNF-α, IL-1β and NO production compared to wild-type cells. Overall, these experiments show that TLR4 signalling through MyD88 is the dominant pathway by which α-syn oligomers are recognised by macrophages. TLR2 plays a more minor role in α-syn-induced inflammation. Since only α-syn oligomers cause signalling, the α-syn oligomer concentration used here is quoted for the rest of the paper.
Low concentrations of α-syn oligomers sensitizes inflammatory responses through TLR4
Oligomer concentrations in the CSF of patients suffering from PD are of the order of 1–10 pM. We, therefore, incubated BV2 microglial cells with lower concentrations of α-syn oligomers (1 nM–10 pM), together with α-syn monomer (100–1 nM), for a sustained period of time (up to 5 days; Fig. 4a) and measured supernatant cytokine production at 24, 48, 72 and 120 h. Lower oligomer concentrations no longer killed the BV2 cells. Significant (p = 0.009) induction of pro-inflammatory cytokines in response to α-syn oligomers, compared to unstimulated cells, was observed. Rat astrocytes produce a response to α-syn oligomers but not monomer or fibrils (Fig. 4b) and this response was detectable at low pM oligomer concentrations after 72 h of incubation (Fig. 4c, p = 0.009, compared against unstimulated cells). There was no significant cytokine production in TLR4 knockout macrophage cells (Supplementary Fig. 6) indicating that TLR4 is required for signalling. By 120 h α-syn oligomer stimulated cells started to die so longer incubations times were not feasible (Supplementary Fig. 5a). This α-syn sensitization response was effectively inhibited by the TLR4 antagonists RSLA or TAK-242 (Fig. 4d).
Assuming increased mRNA expression of TLR4 in human glia is indicative of higher protein expression then this would be expected to result in increased levels of TNF-α on exposure to oligomer concentrations. Experiments on cultured human astrocytes explicitly tested this prediction and confirmed that sensitization also occurred at low-oligomer concentrations (Supplementary Fig. 7) and was significantly reduced by TLR4 antagonists (Supplementary Fig. 8).
α-syn oligomer-induced inflammatory sensitization is reversible
To determine whether microglia could recover if α-syn oligomers were removed from the cells we measured TNF-α production over time, after exposure of BV2 microglia to α-syn oligomers for 72 h followed by oligomer removal (Fig. 5). The production of TNF-α decreases slowly with time, but is not completely restored to basal levels after 2 days. To determine whether TLR4 inhibitors could also reverse α-syn oligomer-induced sensitization, BV2 cells were incubated with α-syn oligomers for 24–72 h. Cells were washed and the media replaced with RSLA and fresh α-syn oligomers and a marked reduction in TNF-α production was seen in the following 48–72 h (Fig. 5b–d). Cell viability showed no significant change (Supplementary Fig. 5b). These data suggest the α-syn sensitization is reversible if α-syn oligomers are removed from the cells or if the cells are incubated with a TLR4 antagonist even if α-syn oligomers are still present in the media.
TLR4 antagonists protect neurons from oligomer-induced toxicity
α-syn oligomers can cause excessive ROS generation, and cell death in neurons. To determine the role of TLR4 signalling in ROS and cell death we used TLR4 antagonists. We also used an extracellular chaperone, clusterin, to bind the oligomers and block all oligomer-induced processes. Clusterin preferentially binds amyloid beta oligomers over monomers [33, 34] and been shown to have an effect on α-syn .
To test the effect of clusterin, RSLA and TAK242 on the production of ROS, we used dihydroethidium (DHE) dye which allowed us to measure the rate of oxidation of the dye by cellular superoxide production. Rat co-cultures of neurons and astrocytes were treated with either 1 μM TAK242 or 0.1 μg/ml RA-LPS for 30 min prior to application of oligomers to the culture. Clusterin was pre-incubated with α-syn oligomers before application to the cells. The production of ROS (mainly superoxide) after application of the α-syn oligomers was compared to the basal level (Fig. 6a). The results showed that α-syn oligomers produced significantly increased levels of ROS, approximately 4-fold as shown previously, and this overproduction was prevented in cells pre-treated with TLR4 inhibitors, RSLA and TAK242. ROS production was also reduced by pre-incubation with clusterin. Application of oligomers for 24 h induced cell death, as previously shown. Oligomer induced cell death in the rat co-culture was reduced to basal levels by both clusterin and the TLR4 antagonist RSLA (Fig. 6aiii). Interestingly, clusterin was less effective when used in experiments on BV2 cells and significantly higher clusterin concentrations were needed for a reduction in TNF-α to be observed (Supplementary Fig. 9). Clusterin needs to bind to the oligomers to prevent TLR4 signalling but only a few TLR4 receptors need to signal for cytokine production. Therefore, both the number of TLR4 receptors and the number of oligomers that are not bound or partially bound by clusterin will determine the extent of TLR4 signalling and resulting cell death. This result suggest that targeting TLR4 directly may be a more effective strategy than binding the oligomers, especially if TLR4 expression is high as is presumably the case with BV2 cells.
Next, we tested whether TLR4 inhibitors reduced the α-syn oligomer-induced cell toxicity in rodent neurons, astrocytes and co-cultures of neurons and astrocytes (Fig. 6ci–iii, ei–iii). Cell preparations were characterized for neuronal and astrocyte enrichment using the markers β-III Tubulin and GFAP (Fig. 6bi). Neuron-enriched cultures contained 90% ± 1.3% β-III Tubulin-positive cells and astrocyte enriched cultures contained 97% ± 0.1% GFAP-positive cells. Neuron-astrocyte co-cultures contained 43 ± 1.1% neurons and 50 ± 1.2% astrocytes (Fig. 6bii). Cells were incubated with α-syn oligomers with or without RSLA or TAK242 overnight, and cell death was assessed by automated counting of the number of dead cells (labelled with sytox green) and number of total cells (Hoechst) (Supplementary Fig. 10a). We also measured the amount of TNF-α production (Supplementary Fig. 11a). Neuronal cell death was significantly increased by α-syn oligomers (p = 7.673E-5) in the enriched neuronal cell culture, and this was not affected by TLR4 blockade (Fig. 6ci). Oligomers also induced very low level, but highly significant, toxicity to astrocytes (p = 6.968E − 8), and this was prevented by TLR4 antagonists (RSLA; p = 7.746E-8, TAK242; p = 8.103E − 8, Fig. 6cii). Notably, the oligomers induced significant cell death in the neuron-astrocyte co-culture (p = 4.493E-4), and this was reduced to basal levels by the presence of the TLR4 inhibitors (RSLA; p = 1.127E-4, TAK242; p = 0.001, Fig. 6ciii) and TNF-α levels reduced to basal levels (Supplementary Fig. 11a). This data suggested that astrocytes are able to drive cell death in rodent co-cultures through oligomer-induced TLR4 signaling.
As we have reported that α-syn oligomers drive human astrocytic production of TNF-α (Supplementary Figs. 7, 8) via TLR4-dependent pathways, we tested the effect of TLR4 inhibition on oligomer induced cell toxicity in a primary human model (Fig. 6d, e). We generated enriched human neuronal cultures (92 ± 2.1% β-III Tubulin-positive cells), enriched human astrocyte cultures (99 ± 1.2% GFAP) and a neuron-astrocyte co-culture (β-III Tubulin: 52 ± 2.8%, GFAP: 43 ± 3.2%, Fig. 6dii). Application of α-syn oligomers induced significant levels of cell death in both the enriched neuron (p = 0.046, Fig. 6ei) and the enriched astrocyte (p = 0.0288, Fig. 6eii) cultures. As noted with the rodent cells, the oligomer-induced neuronal death was not abolished by TLR4 blockade in the absence of astrocytes (Fig. 6ei). TLR4 antagonism did show a tendency to prevent oligomer-induced astrocytic cell death (Fig. 6eii). Importantly, in the human neuron-astrocyte co-culture experiment, oligomer-induced cell death was prevented by the application of TLR4 antagonists RSLA and TAK242 (p = 0.048 and p = 0.0333, respectively, Fig. 6iii, Supplementary Fig. 10b) and the TNF-α levels reduced to basal levels (Supplementary Fig. 11b).
We estimated the proportion of astrocytic and neuronal cell death in the co-culture experiment, by sampling the nuclear size of neurons versus astrocytes from the enriched cultures, and segmenting the sytox green-positive nuclei on the basis of size in the co-cultures (Supplementary Fig. 12). For both the rodent and human co-culture, the majority of the oligomer-induced cell death was neuronal rather than astrocytic (91 ± 2.6% in rodent and 68 ± 2.9% in human co-culture preparation).
Taken together, this data shows that oligomers are able to induce neuronal toxicity through two mechanisms: neurons alone are directly susceptible to oligomer-induced TLR4-independent toxicity. However, in the presence of astrocytes, the dominant mechanism of oligomer-induced neuronal cell death is TLR4-dependent oligomer-induced activation of astrocytes leading to the production of TNF-α and other cytokines by astrocytes.