To characterize and dissect the functional properties of the catalytic domains of LRRK2, in particular the kinase domain, we used a multi-scale approach that begins with testing real time filament formation in live cells to assessing the consequences of molecular dynamics simulations of PD mutations in the kinase domain. Of primary importance was to characterize the biochemical properties of LRRK2RCKW following deletion of the catalytically inert N-terminal lid. To next confirm that LRRK2RCKW was a well-folded protein we used hydrogen deuterium exchange mass spectrometry (HDX-MS) and then to capture the allosteric features of the kinase domain we mapped by HDX-MS the conformational changes in LRRK2RCKW that result from adding MLi-2.
Capturing filament formation in real time.
We and others previously showed that treatment with a highly specific LRRK2 inhibitor (MLi‑2) induces filament formation of wt LRRK2 and the G2019S mutant when these proteins are transiently expressed in mammalian cells (17, 36). To capture the dynamics of such redistribution we performed time-lapse imaging of YFP-tagged G2019S and wt LRRK2. As shown in Figure 1 and Supplementary videos, under normal conditions G2019S LRRK2 is mostly diffuse in the cytosol; however, 15-30 min following MLi-2 treatment the protein begins to concentrate first in ‘satellite’ structures diffuse throughout the cells. It then polymerizes to form intricate thicker filaments by 2.0-2.5 h after treatment. Although wt LRRK2 follows a similar redistribution upon treatment with MLi-2, in general it takes longer, approximately 30 min‑1 h, before the first structures are observed (Supplementary movie). In both cases this effect is readily reversible: after washout of MLi-2 for 2 h, the proteins gradually diffuse back into the cytosol. To verify that this protein rearrangement was truly dependent on the specific MLi-2 inhibitor, we performed time-lapse imaging using a type 2 inhibitor, rebastinib (Figure 1B). Although rebastinib stabilizes a Kinase-WD40 construct of LRRK2, based on a thermal shift assay (Figure S1), it did not induce changes in the localization of G2019S proteins even after 8 h treatment, confirming the prediction of Deniston et al. (2020) (13).
Figure 1. MLi-2, but not Rebastinib, affects the localization of the kinase hyperactive LRRK2-G2019S mutant. A) Time-lapse imaging of HEK293T cells transiently expressing YFP-LRRK2-G2019S: confocal images (YFP fluorescence signal, maximum intensity projections) were acquired every 11 min. Representative images show the typical diffuse cellular localization of the proteins (t=0 h) prior to treatment with 100 nM MLi-2; following MLi-2 addition, proteins relocalize to form cytoplasmic filamentous structures (yellow arrows, +MLi-2, t=2.5 h). After washout of the inhibitor, the proteins gradually dissociate from the filaments into the cytosol (washout, t=2-3 h). B) Time-lapse imaging of HEK293T cells transiently expressing YFP-LRRK2-G2019S before (t=0 h) and after treatment with 100 nM Rebastinib. No changes in the localization of the proteins are observed. Scale bar, 20 microns.
LRRK2RCKW variants spontaneously form filaments around microtubules in an MLi-2 independent manner
In our filament formation assay, flag-tagged variants of the LRRK2RCKW construct were overexpressed and cells were analyzed after fixation via antibody staining in a confocal laser-scanning microscope. The majority of the transfected cells, regardless of the mutation, displayed constitutive filament formation (Figure 2). Most striking, in contrast to full length (fl) LRRK2, is that wt and G2019S are no longer dependent on MLi‑2 for docking onto MTs. This supports the hypothesis that the inert N-terminal scaffolding domains are not required for the filaments to form but instead are essential for protecting or shielding the catalytic domains to prevent them from docking onto MTs. In this way they promote the cytosolic distribution of LRRK2 prior to activation, which is likely further facilitated by phospho-dependent interactions with specific 14-3-3 proteins (22, 27). The N-terminal domains are also important for docking to Rab proteins such as Rab29, which are thought to initiate activation of LRRK2 (29). This would be a physiological mechanism where multiple biological functions are embedded in the catalytically inert N-terminal domains, this includes activation and/or localization by heterologous proteins as well as inhibition of the catalytic domains. We hypothesize that most of the PD mutations circumvent or “hijack” this normal process.
Of the mutants tested only LRRK2RCKW D2017A, a kinase dead mutant, showed strongly reduced docking onto MTs, and this is consistent with our earlier findings showing that the fl D2017A mutant did not dock onto MTs even in the presence of MLi2 (Figure 2). We confirmed here that MLi-2 did not have an additive effect on the percentage of cells showing LRRK2RCKW filaments and did not induce binding of the D2017A mutant (Figure 2). We conclude that the high affinity binding of MLi‑2 to the kinase domain is sufficient to unleash the N-terminal protective lid that normally shields the catalytic domains and promotes localization in the cytosol. We also show that simply removing the N-terminal lid is in most cases sufficient to promote docking onto MTs. The exception is the D2017A mutant, which cannot bind well under any conditions either because it lacks the ability to undergo a subsequent essential auto-phosphorylation step or, most likely, because it is locked into an open conformation similar to what we saw with rebastinib. We next asked whether LRRK2RCKW retained its full kinase catalytic activity even though the regulatory machinery embedded in the N-terminal domains is removed.
Figure 2. Filament formation of LRRK2RCKW is independent of MLi-2 treatment but reduced for LRRK2RCKW D2017A. A) Schematic domain organization of LRRK2 full length protein (blue box) and the LRRK2RCKW construct (red box). B) Plasmids, encoding LRRK2RCKW variants, were transfected into HEK293T cells for LRRK2RCKW overexpression. Transfected cells were then analyzed for the spatial distribution of LRRK2RCKW by immunostaining. All tested LRRK2RCKW variants displayed a high likelihood (80-90%) to form filaments inside the HEK293T cells except for LRRK2RCKW D2017A (20-30%). Interestingly, in contrast to LRRK2 full length the percentage of cells showing filament formation was independent of MLi-2 treatment or a specific LRRK2RCKW mutation. Scale bar, 30 microns.
Protein kinase activity is conserved and, in some cases, amplified in the LRRK2RCKW proteins.
To assess kinase activity, we used both LRRKtide, a small synthetic peptide, and Rab8a as substrates for the LRRK2RCKW proteins. In addition to wt LRRK2RCKW we measured the kinase activities of two ROC:COR domain mutations (R1441C and Y1699C) and four mutations in the kinase domain, more precisely in the DYGψ motif (D2017A, Y2018F, G2019S, I2020T). R1441 and Y1699 are located in the ROC and COR domains, respectively, and, based on homology models, are predicted to be part of the ROC:COR domain interface (11, 51, 52). Importantly, we found that wt LRRK2RCKW has kinase activity that is comparable to full-length LRRK2 (36) although in the absence of the N-terminal scaffolding domains the activity is no longer dependent on Rab activation. Using LRRKtide as a substrate we found that the pathogenic mutation R1441C slightly increased the kinase activity while Y1699C had only a minor effect on LRRKtide phosphorylation (Figure 3). In contrast, when we used a physiological substrate, Rab8a, Y1699C led to an enhanced pT72 phosphorylation in vitro, comparable to the phosphorylation by Y2018F, whereas R1441C behaved like wt (Figure 3). The fact that kinase activity is dependent on substrate may account for some of the confusion in the literature about the activity of various LRRK2 mutants but suggests that some of the mutations may change substrate specificity. If these residues are indeed at a domain interface, as predicted, they could also introduce a conformational change that would result in the unleashing of the N-terminal scaffolding domains and/or promote dimerization which is associated with membrane localization and activation of LRRK2 (29, 53, 54).
The strongest effects on kinase activity for LRRK2RCKW were observed for mutations embedded within the activation segment of the kinase domain, specifically in the DYGψ motif where ψ is typically conserved as a hydrophobic residue. Most other kinases have a DFGψ motif, and Y2018 was predicted earlier, based on activation when the Tyr is replaced with Phe, to serve as a brake that keeps LRRK2 in an inactive state (36). We measured the effect of mutating each of these residues on kinase activity. The D2017A (DYG) mutant was not able to phosphorylate either LRRKtide or Rab8a (Figure 3) which is consistent with other kinases, since this residue is part of the regulatory triad and is crucial for the correct coordination of the Mg2+-ions and the γ-phosphate of ATP in the kinase active site cleft (55). Reintroducing the classical DFG motif to LRRK2RCKW (DYG in LRRK2) increases the kinase activity for LRRKtide by a factor of 3-4, whereas Rab8a phosphorylation was only enhanced by a factor of 1.7 (Figure 3). LRRKtide phosphorylation by LRRK2RCKW G2019S was comparable to LRRK2RCKW Y2018F. When comparing Rab8a pT72 phosphorylation, G2019S showed two times higher Rab8a pT72 phosphorylation than Y2018F. The other tested pathogenic DYGψ mutation, I2020T, displayed a reduced phosphorylation of LRRKtide as well as Rab8a (Figure 3). This is also in accordance with our LRRK2 full-length data of I2020T, albeit full-length I2020T Rab8a phosphorylation was comparable to wt. The results for the I2020T mutation in LRRK2 full length and LRRK2RCKW demonstrate that LRRK2 pathogenicity is not driven solely by increased kinase activity but also by changed substrate preferences such as serine/threonine specificity as well as changes in subcellular localization. We later show that the dynamic properties are also altered by these mutations.
Figure 3. The LRRK2RCKW variants Y2018F and G2019S enhance LRRKtide and Rab8a phosphorylation. (A) A LRRKtide based kinase assay for LRRK2RCKW variants revealed that although it is a deletion construct it preserves LRRK2 full length kinase activity. Additionally, the DYGψ mutants tested here also resemble the results of their full-length counterparts. Interestingly, also the pathogenic mutations R1441C and Y1699C which are situated in the ROC:COR region of the LRRK2RCKW construct display a mild increase in kinase activity compared to LRRK2RCKW wt. Experiments for each mutant were performed at least in duplicates of duplicates for two independent LRRK2 expressions. Each dot represents the mean of a duplicate, while the dotted line represents the mean of the measured wt activity. To determine significant differences between LRRK2RCKW wt and mutant activity a one-way ANOVA test based on the Dunnett’s multiple comparisons test was performed. Hereby one asterisks (*) indicate a P value between 0.05 and 0.01, two asterisks (**) a P value between 0.01 and 0.001 and four asterisks (****) a P value below 0.0001. (B) When testing Rab8a as a substrate for the LRRK2RCKW construct employing Western blotting against pT72 and the His-tag of His-Rab8a, we revealed increased phosphorylation of Rab8a by LRRK2RCKW Y2018F, G2019S and Y1699C. MLi-2 was shown to efficiently block phosphorylation of Rab8a which was also found for the kinase dead mutant D2017A. Quantification was performed for two independent Western Blots using two independent protein preparations. For each quantification, the pT72 signals were referenced to the signal for the His-tag of 6xHis-Rab8a and then normalized to the resulting wt signal. The dotted line therefore represents 100% of the wt signal.
Mapping the conformational changes induced by MLi-2 using Hydrogen Deuterium Exchange Mass Spectrometry (HDX-MS).
To define the global conformational changes induced in LRRK2RCKW as a consequence of MLi-2 binding we used HDX-MS, which allows us to determine the solvent exposed regions of the protein over a time course of 5 minutes. The exchange data was mapped onto models of the ROC:COR and kinase domains and onto the solved structure of the human WD40 domain. Although this is a large protein (1200 residues), we obtained excellent coverage (>96 %), and the solvent exposed regions are consistent with the predicted folding of all four domains (Figure S2). While we focus here primarily on the kinase domain, the graph summarizing the overall solvent accessibility of the entire protein shows not only that the four domains are well-folded but also identifies several regions that are highly solvent exposed. Of particular note is the activation loop of the kinase domain as well as the segment that lies between the COR-B domain and the kinase domain and the segment that joins the GTPase domain to the COR-A domain. The HDX-MS data suggests that these regions having high deuterium uptake are highly flexible or unfolded. Conversely, there are also regions on the surface of each domain that are highly protected from solvent, implying that these are domain-domain interfacial surfaces (Figure S2). It is important to appreciate that the HDX-MS profile is obtained independent of a solved structure and can thus serve as validation of a predicted model. Overall LRRK2RCKW is a well-folded protein that is consistent with a complex topological model with inter-domain interactions.
Under apo conditions the N-lobe of the kinase domain is more shielded from solvent than the C-lobe (Figure 4A). The aC-b4 loop, for example, is almost completely shielded from solvent. This is somewhat unusual in that the N-lobe in the absence of nucleotide tends to be rather dynamic for many protein kinases. The ordered and stable structure of the LRRK2 N-lobe is predicted to be due to constraints imposed by the other domains. This is analogous to the way that cyclin binding orders the N-lobe of CDK2 in contrast to the isolated kinase domain (56). Most kinase structures represent just an isolated kinase domain so one cannot appreciate how other domains contribute to stabilization and, in turn, regulation of the N-Lobe. Our HDX-MS results also help to explain why it has not been possible so far to express the kinase domain independent of the rest of the protein. In the apo protein the activation loop of the kinase domain in the C-lobe has the highest deuterium uptake suggesting it is highly disordered and exposed to solvent (Figure 4 and S2).
Figure 4. The deuterium uptake of the LRRK2RCKW kinase domain. (A) The relative deuterium uptake after 2 min of deuterium exposure of the LRRK2RCKW kinase domain is shown in a color-coded homology model. Grey color indicates no deuterium uptake information. The N-lobe of the kinase mostly shows blue to green colors indicating low deuterium uptake. On the other hand, the αD, activation loop, the end of αF to αH have higher deuterium uptake suggesting a more dynamic, solvent accessible C-lobe. (B) Representative peptides that have almost no deuterium uptake are mapped on the kinase domain. Insets show uptake for the apo kinase (black) and the MLi-2 bound state (red).
To gain insight into the allosteric impact of inhibitor binding we next looked at the conformational changes in LRRK2RCKW following treatment with MLi-2. The overall changes, captured in the graph in Figure 5, show that there is subtle, albeit important, protection in regions that extend into the GTPase and COR-A:COR-B domains, but by far the largest changes are concentrated in the kinase domain and in the linker that precedes the kinase domain. There are no regions that show enhanced solvent accessibility in the presence of MLi-2. We focus here on the conformational changes that are localized to our kinase domain model. These changes lie not only in the N-lobe and the active site cleft where the inhibitor is directly docked but also in the C-lobe in regions that lie far from the active site cleft.
Figure 5. Binding of MLi-2 reduces the deuterium uptake of LRRK2RCKW. (A) The relative deuterium exchange for each peptide detected from the N to C terminus of LRRK2 in apo-kinase (Black) and MLi-2 bound (red) conditions at 2 min. The arrows indicate the regions of LRRK2 that have less deuterium uptake when bound to MLi-2 (B) The deuterium uptake of selected peptides is plotted and mapped on the kinase model. The uptake is reduced in the Glycine rich loop, the αC helix, the activation loop, the DYG motif, the YRD motif and the hinge region in the presence of MLi-2.
Looking more carefully at the protected regions, we saw that the binding of MLi-2 reduces the H-D exchange in the ATP-binding site, the activation loop, the αC helix and the hinge region (Figure 5). These regions that would be predicted to contact the inhibitor (57, 58) all show significantly reduced deuterium uptake. Peptides, for example, in the hinge region (aa 1948-1958), including the aD helix, experienced a large increase in protection upon MLi-2 binding (50% vs. 20%). The peptide covering the catalytic loop (aa 2013-2022) including the YRD motif also experienced protection (30% to <10%). The Glycine-rich loop (aa 1884-1893) is also highly protected. Most importantly we see that the peptide containing the DYGI motif (aa 2013-2022) is almost completely shielded as a consequence of MLi-2 binding; the deuterium exchange dropped from 70% to less than 10% suggesting that this region, highly solvent accessible in the absence of ligand, becomes almost completely protected by the coordination of the inhibitor (Figures 5 and 6). This is quite consistent with the prediction that the kinase domain assumes a compact and closed conformation in the presence of MLi-2. The C-terminus of this peptide contains the beginning of the activation loop, which now appears to be well-folded and shielded from solvent in contrast to the apo structure. Interestingly, the uptake spectra of the two peptides covering the activation loop show an EX1 bimodal distribution, which indicates two different conformations in solution (59). One of these peptides (aa 2028-2056) is shown in Figures 5 and 6. Most other peptides show a single peak indicative of the more typical EX2 exchange kinetics. In addition, MLi-2 treatment also induced slow-exchange in the DYG loop even though it is highly protected. The peptide that covers the N-terminus of the aC helix (aa 1915-1921) shows significant slow exchange even in the absence of MLi-2 that most likely continues beyond 5 min. Although the exchange is quenched in the presence of MLi-2, the slow exchange still persists.
The protection of the ATP-binding site and the hinge region by MLi-2 are consistent with other inhibitor bound homolog kinase structures (57, 58) although our data also reflects the dynamic change that the binding of MLi-2 has not only on the kinase domain but also on LRRK2RCKW overall. Essentially any region in LRRK2RCKW that interfaces with the kinase domain will sense binding of nucleotide or an inhibitor. This includes also the LRR domain, not included in our construct, but which is predicted to lie over the kinase domain (13, 51, 60) and would be displaced by the high affinity binding of a kinase inhibitor. HDX-MS shows that changes in conformation and dynamics of the kinase domain are felt through long-distances in LRRK2RCKW, as flexible regions throughout the protein exhibit increased protection upon MLi-2 binding (Figure 5A and S2).
Figure 6. HDX-MS for DYG, activation loop, and aC helix peptides reveal slow dynamics. In the DYG peptide (2013-2022) the apo state (black) plateaus within 2 min. The MLi-2 bound state (red) continues to slowly exchange at least up to 5 minutes, suggesting that with MLi-2 this region undergoes a slow dynamic process. The apo state of the activation loop peptide (2028-2056) again plateaus within 2 min while the MLi-2 bound state gradually increases after 2 min. From the spectral plot, the uptake of the activation loop peptide in the MLi-2 state exhibits bimodal behavior. One process has slow deuterium uptake (protected); and the other process has fast uptake (solvent exposed)- similar to the single process observed in the apo state. For the aC peptide (1915-1921), the deuterium increases without reaching a plateau over 5 minutes for both states.
GaMD simulations indicate that the LRRK2 kinase domain mutations Y2018F, G2019S and I2020T attenuate flexibility of the activation segment of the kinase core
GaMD simulations were performed on the activated kinase domain of LRRK2 (1865-2135) to investigate changes in the conformational landscape that are caused by the D2017A, Y2018F, G2019S, and I2020T mutations. During all 10 replicate accelerated simulations the wt kinase favors an open and inactive active cleft conformation as measured by the relative position of the N- and C-lobes and the αC-helix (Figure 7). The fully closed and active conformation, in which the N- and C-lobes are brought together in concert with an inward positioning of the αC helix to assemble the active site, is infrequently sampled by the wt kinase. In contrast, Y2018F, G2019S, and I2020T are all capable of accessing a closed and active conformation, while D2017A samples a much more open and inactive conformation (Figure 7a). The degree of stabilization of the closed conformation roughly correlates with the observed changes in MT association: D2017A<wt<G2019S<Y2018F <I2020T; it does not correlate with activity. The I2020T mutant is trapped in a mostly closed state without extensive open-to-close transitions and αC in-to-out motion compared to wt, Y2018F and G2019S. This loss of breathing dynamics may partially explain the reduced kinase activity of I2020T, where substrate/product kinetics may be impacted. Likewise, Y2018F and G2019S both populate a wide range of open and also closed-active conformations likely contributing to their increased kinase activity. The ability of all of the activating mutants to populate a closed conformation may play a role in their altered MT association compared to wt, where Y2018F and I2020T spontaneously form filaments and G2019S forms filaments faster than wt upon treatment by the type-I inhibitor MLi-2.
The hydroxyl moiety of Y2018 in the DYG motif forms persistent hydrogen bonds between the backbone of both I1933 in the αC-β4 loop and I2015 (Figure 7b). This interaction stabilizes the tyrosine side chain in an orientation that restricts the αC helix from assembling the active site due to steric clash with L1924. This simulation agrees well with a recent Cryo-EM structure of an inactive conformation of LRRK2RCKW, (13). These authors identify the same hydrogen bond between Y2018 and the shell residue I1933. Our simulations provide strong independent evidence that Y2018 in wt LRRK2 is a key stabilizer of the inactive kinase conformation and may also act as a sensor of the αC-β4 loop conformation, a conserved hotspot for kinase allosteric modulation (61). Absence of the OH hydrogen bonds in the Y2018F mutation leads to greater Y2018F side chain dynamics and packing with L1924 that resembles an active kinase configuration (or a properly formed R-spine). An assembled R-spine is the hallmark for an active kinase. Stabilization of the wt Y2018 side chain leads to a ‘frustrated’ DYG backbone free-energy landscape by pulling the motif out of ideal f/y space, which is likely important for the kinase's role as a switch (Figure S3). A consequence of freeing the side chain of Y2018F is the convergence of the DYG dihedral angles into a canonically active kinase f/y region (62)(Figure S3,S4).
The I2020T mutation introduces a hydrogen bond between the OH group of the threonine to the backbone of the catalytic YRD motif (Y1992) (Figure 7c). The interaction with the YRD motif both stabilizes the catalytic loop and also leads to a closed kinase active site (Figure 7a,c). The DYG motif is also stabilized in an active conformation, similarly to Y2018F, as measured by its ensemble DYG dihedral angles (Figure S4). The I2020T equilibrium is shifted to the closed conformation and activity may be reduced because the mechanism for opening is impaired. Finally, G2019S introduces a hydrogen bond with the sidechain of E1920 in the αC helix, which in turn forms a highly conserved salt-bridge with K1906 of β3 (Figure 7d). The influence of the G2019S mutation on the interaction between αC and β3 and the DYG loop favors the closed and active kinase conformation. The G2019S DYG motif is stabilized in an active conformation as described by its dihedrals (Figure S4).
Figure 7. Mutations in the DYGy loop alter kinase dynamics. (A) Kinase conformational free-energy landscape, represented by ‘open-close’: the distance from the top of the active site (K1906/β3 sheet) to the bottom of the active site (D1994/YRD motif), and ‘αC in-out’: the distance between K1906 and the αC helix (E1920). The white dashed line shows the closed-active kinase conformation. The active state is infrequently sampled by the wt kinase, whereas the DYGy mutants more readily access the closed-active conformation. However, the kinase dead D2017A mutant is destabilized to a more open conformation relative to wt. (B) In wt Y2018 (black panel) is locked in an inactive orientation by hydrogen bonds with I2015 and I1933. Y2018F (red panel) packs with L1924 of the αC helix and releases the DYG loop from an inactive state helping to assemble the active site. Y2018F breaks the interaction leading to increased side-chain dynamics, measured by the distance between the 2018 ζ-carbon and the backbone of I2015 (wt:black, mutant:red). (C) I2020T makes a hydrogen bond with the backbone of Y1992 in the YRD motif, coupling the DYG and catalytic loops, which results in decreased backbone dynamics. The mutation brings the DYG and YRD motifs together, measured as the distance from the Cβ of 2020 and the backbone of Y1992 (wt:black, mutant:red). (D) G2019S bridges the DYG loop to the αC helix and β3 sheet, through E1920 and K1906. This stabilizes the DYG loop, shown by root-mean-square deviation (wt:black, mutant:red), and promotes the closed kinase conformation.