Structure of LRRK2 in Parkinson's disease and model for microtubule interaction.

Leucine-rich repeat kinase 2 (LRRK2) is the most commonly mutated gene in familial Parkinson's disease1 and is also linked to its idiopathic form2. LRRK2 has been proposed to function in membrane trafficking3 and colocalizes with microtubules4. Despite the fundamental importance of LRRK2 for understanding and treating Parkinson's disease, structural information on the enzyme is limited. Here we report the structure of the catalytic half of LRRK2, and an atomic model of microtubule-associated LRRK2 built using a reported cryo-electron tomography in situ structure5. We propose that the conformation of the LRRK2 kinase domain regulates its interactions with microtubules, with a closed conformation favouring oligomerization on microtubules. We show that the catalytic half of LRRK2 is sufficient for filament formation and blocks the motility of the microtubule-based motors kinesin 1 and cytoplasmic dynein 1 in vitro. Kinase inhibitors that stabilize an open conformation relieve this interference and reduce the formation of LRRK2 filaments in cells, whereas inhibitors that stabilize a closed conformation do not. Our findings suggest that LRRK2 can act as a roadblock for microtubule-based motors and have implications for the design of therapeutic LRRK2 kinase inhibitors.

LRRK2 is a large (288 kDa) multi-domain protein. Its amino-terminal half is comprised of repetitive protein interaction motifs (Armadillo, Ankyrin, and Leucine-Rich Repeats) and its carboxy-terminal catalytic half contains a Ras-like GTPase (Ras-of-Complex, or RoC domain), a kinase domain, and two other domains (C-terminal Of Roc, or COR, and WD40) (Fig. 1a). High-resolution structural data on LRRK2 is limited to bacterial homologs[6,7] or isolated domains[8,9], whereas low resolution full-length protein structures have been obtained using negative stain[10] and cryo-electron microscopy (Cryo-EM)[11]. A recent study of LRRK2 bound to microtubules in cells using cryo-electron tomography (Cryo-ET) and subtomogram analysis led to a 14Å structure and proposed model of the catalytic half of LRRK2[5].

Figure 1 ∣

Cryo-EM structure of LRRK2RCKW.

a, Schematic of the construct used in this study. The N-terminal half of LRRK2, absent from our construct, is shown in dim colors. The same color-coding of domains is used throughout the paper. The five major familial Parkinson’s Disease mutations and a Crohn’s Disease-linked mutation are indicated. b, c, 3.5Å cryo-EM map (b) and local resolution (c) of the LRRK2RCKW trimer, with one monomer highlighted. d, e, 3.8Å cryo-EM map (d) and local resolution (e) of a LRRK2RCKW monomer with improved resolution for the RoC and COR-A domains. f, Ribbon diagram of the atomic model of LRRK2RCKW. g, 8.1Å cryo-EM map of monomeric LRRK2RCKW with the model in (f) docked in. h, Location of the Parkinson’s and Crohn’s Disease mutations listed in (a). i, j, Interface between the C-terminal helix and the kinase domain in LRRK2RCKW with residues involved in electrostatic and hydrophobic interactions indicated.

LRRK2’s interaction with microtubules is linked to disease as four of the five major PD-causing mutations (Fig. 1a)[1] enhance microtubule association of LRRK2[12]. Further, Rab GTPases, which mark membranous cargos that move along microtubules, are physiological LRRK2 kinase substrates[13,14]. These cargos, and others implicated in PD pathology[3], are transported by the microtubule-based motors dynein and kinesin. Here, we set out to determine a high-resolution structure of LRRK2’s catalytic half using cryo-EM, and to understand how it interacts with microtubules and how this impacts microtubule-based motor movement.

Structure of the catalytic half of LRRK2

High-resolution studies on human LRRK2 have been limited by the lack of efficient expression systems that yield stable protein. We tested many constructs (Extended Data Fig. 1a), leading to the identification of one consisting of the carboxy-terminal half of wild-type (WT) LRRK2 (amino acids 1,327 to 2,527), which resulted in robust insect cell expression and well-behaved protein (Extended Data Fig. 1b, c). This construct comprises the RoC, COR, kinase and WD40 domains of LRRK2 (Fig. 1a), which we refer to as LRRK2RCKW. The COR domain was previously defined as consisting of two subdomains, COR-A and COR-B[6].

Extended Data Figure 1 ∣

Optimization of LRRK2 constructs and cryo-EM analysis of a LRRK2RCKW trimer.

a, We systematically scanned domain boundaries (amino acid numbers of boundaries noted above domain names) to generate LRRK2 constructs that expressed well in baculovirus-infected insect cells and yielded stable and soluble protein. These attempts included full-length LRRK2, the kinase domain alone or with the WD40 domain, and other isolated domains. In this approach, only the GTPase domain on its own expressed well. Next, we gradually shortened LRRK2 from its amino-terminus. Red asterisks indicate constructs that were soluble. b, After identifying domain boundaries yielding constructs that expressed soluble protein, additional fine tuning of boundaries was performed. A Coomassie stained SDS-PAGE gel shows systematic N-terminal truncations at the RoC domain resulting in the identification of a construct with the highest expression levels: amino acids 1327 to 2527 (red asterisk, “LRRK2RCKW” here). c, A Coomassie stained SDS-PAGE gel of purified LRRK2RCKW after elution from an S200 gel filtration column. As predicted by its primary structure, LRRK2RCKW runs at ~140 kDa. d, Electron micrograph of LRRK2RCKW. e, 2D class averages of the LRRK2RCKW trimer. f, 2D/3D classification scheme used to obtain the 3.5Å structure of the LRRK2RCKW trimer. g, h, Fourier Shell Correlations (from Cryosparc) (d) and Euler angle distribution (e) for the LRRK2RCKW trimer.

We determined a 3.5Å structure of LRRK2RCKW in the presence of GDP using cryo-EM (Fig. 1b, c and Extended Data Figs. 1 and 2). On our grids, we observed a mixture of monomers, dimers and head-to-tail trimers; we used the trimer to solve the structure (Fig. 1b and Extended Data Figs. 1 and 2). Although critical for reaching high resolution, the trimer is likely specific to the cryo-EM grid preparation as LRRK2RCKW is predominantly monomeric, with a smaller percentage of dimers, in solution (Extended Data Fig. 2n) and we only observed the trimer when preparing grids with high concentrations of LRRK2RCKW (see Methods). Due to flexibility, the RoC and COR-A domains were lower resolution in our structure (Fig. 1b, c). We improved this part of the map using signal subtraction and focused 3D classification and refinement (Fig. 1d, e, Extended Data Figs. 2a-d). The final model was generated using the signal subtracted maps of the RoC and COR-A domains, and then combined with the COR-B, kinase and WD40 domains from the trimer map (Fig. 1f, Extended Data Fig. 2e-m and Supplementary Video 1). Our model fits well into an 8.1Å reconstruction we obtained of a LRRK2RCKW monomer (Fig. 1g, and see Extended Data Fig. 6), indicating that trimer formation does not cause major structural changes in the protein.

Extended Data Figure 2 ∣

Cryo-EM analysis of a signal-subtracted LRRK2RCKW trimer and map-to-model fit.

a, Processing strategy used to obtain a 3.8Å structure of LRRK2RCKW generated from a signal-subtracted trimer where only one monomer contains the RoC and COR-A domains. This structure improved the resolution of the RoC and COR-A domains relative to the full trimer (Extended Data Fig. 1). b-d, 2D class averages (b), Fourier Shell Correlations (from Relion) (c), and Euler angle distribution (from Relion) (d) for the 3.8Å-resolution signal-subtracted LRRK2RCKW structure. e, Close-ups (f-l) of different parts of the final model fit into the map. f, Section of the WD40 domain. g, C-terminal helix and its interface with the kinase domain. h, Active site of the kinase. Residues in the DYG motif are labeled. G2019, the site of a major PD-associated mutation (G2019S) and the last residue of the activation loop seen in our structure, is highlighted by a black rounded square. i, Interface between COR-B and the αC helix of the N-lobe of the kinase domain. j, Interface between the RoC and COR-B domains. R1441 and Y1699, two residues mutated in PD, are labeled. k, l, Two different views of the RoC and COR-A domains with GDP-Mg2+ modeled into the density. Side chains were omitted in these two panels, corresponding to the lowest-resolution parts of the map. m, Map-to-model FSC plots for the top-ranked LRRK2RCKW models, with (left) or without GDP-Mg2+ (right) in the RoC domain. The 0.143 FSC values are reported in Extended Data Table 1. n, Size Exclusion Chromatography-Multiple Angle Light Scattering (SEC-MALS) analysis of LRRK2RCKW under the conditions used for cryo-EM (Fig. 1). The table shows the calculated molecular weights (MW) of LRRK2RCKW according to SEC standards (“SEC”) and MALS.

Extended Data Figure 6 ∣

Cryo-EM analysis of a monomer and WD40- and COR-mediated dimers of LRRK2RCKW in the absence of inhibitor (“Apo”) and dimerization of LRRK2RCKW outside the filaments.

a, Data processing strategy for obtaining cryo-EM structures of a monomer and WD40- and COR-mediated dimers of LRRK2RCKW in the absence of inhibitor. The models used during the processing of the dimers (see Methods) are those shown in Extended Data Fig. 5d, along with an additional linear trimer (see Methods) used for particle sorting. The models used for processing of the monomer (see Methods) were the same dimer models as in Extended Data Fig. 5d (used for particle sorting) in addition to a monomer model generated from our LRRK2RCKW model (used for refinement). b, Two-dimensional (2D) class averages of WD40- and COR-mediated LRRK2RCKW dimers obtained in the absence of inhibitors (“Apo”) or in the presence of either Ponatinib or MLi-2. The same molecular models of the two dimers shown in Fig. 3 are shown on the left but in orientations similar to those represented by the 2D class averages shown here. For each class average, a projection from the corresponding model in the best-matching orientation is shown to its left. c, Two copies of the LRRK2RCKW structure were aligned to the RoC-COR domains of the LRR-RoC-COR structure from C. tepidum’s Roco protein (PDB: 6HLU) to replicate the interface observed in the bacterial homolog in the context of the human protein. This panel shows a comparison between the dimer modeled based on the C. tepidum LRR-RoC-COR structure and the dimer observed for LRRK2RCKW in this work. While the bacterial structure shows a dimerization interface that involves the GTPase (RoC), LRRK2RCKW interacts exclusively through its COR-A and -B domains, with the RoC domains located away from this interface. The two arrangements are shown schematically in cartoon form below the structures.

LRRK2RCKW adopts an overall J-shape, with the WD40, kinase and COR-B domains arranged along one axis, and COR-A and RoC turning around back towards the kinase. This brings the COR-A and tightly associated RoC domain into close proximity to the kinase’s C-lobe (Fig. 1f, and Supplementary Video 1). This arrangement likely underpins the crosstalk between LRRK2’s kinase and GTPase[15,16]. Part of the FERM domain in the FAK-FERM complex approaches the FAK C-lobe in a similar way[17] (Extended Data Fig. 3a, b). The RoC, COR-A and COR-B domains are arranged as seen in crystal structures of LRRK2 bacterial homologs[6,7,18]. The N-lobe of the kinase domain, in particular its αC helix, forms an extensive interaction with the COR-B domain, with COR-B occupying a location similar to Cyclin A in CDK2-Cyclin A[19] (Extended Data Fig. 3a, c).

Extended Data Figure 3 ∣

Comparisons between LRRK2 and other kinases and modelling of the Leucine-Rich Repeat (LRR) into LRRK2RCKW.

a, View of the LRRK2RCKW atomic model with COR-A, COR-B and kinase domains colored. The N- and C-lobes of the kinase are labeled, as is the αC helix in the N-lobe. b, c, The FAK-FERM (PDB: 2J0J)[17] (b) and CDK2-Cyclin A (PDB: 2CCH)[19] (c) complexes, shown in the same orientation as the kinase in (a). The αC helix of CDK2 is also labeled. d, Same view as in (a) with only the kinase domain and the C-terminal helix colored. e, Rotated view of LRRK2’s kinase domain with the C-terminal helix facing the viewer. f, g, CDKL3 (PDB: 3ZDU) (f) and RIPK2 (PDB: 4C8B)[32] (g) shown in the same orientation as LRRK2’s kinase in (e), with alpha helices with the same general location as LRRK2’s C-terminal helix colored in green. h, KSR2-MEK1 complex (PDB: 2Y4I), with the kinase oriented as in (e) (left) and after removing KSR2 for clarity (right). The alpha helix associated with the kinase is shown in green. i, HCK (PDB: 2HCK) in complex with its SH2 and SH3 domains with the kinase oriented as in (e) (left), and after removal of the SH2 and SH3 domains for clarity (right). A remaining alpha helix from the SH2 domain is shown in yellow. j, Front view of LRRK2’s kinase with the C-Spine and R-Spine residues colored in grey and white, respectively. k, Close-up of the DYG motif and neighboring R-Spine residues. A putative hydrogen bond between Y2018 and the backbone carbonyl of I1933 is shown (O-O distance: 2.7Å). This interaction provides a structural explanation for the hyperactivation of the kinase resulting from a Y2018F mutation[38], which would release the activation loop.

l, Crystal structure of the LRR-RoC-COR(A/B) domains from C. tepidum Roco (PDB: 6HLU)[7]. m, Homology model for human LRR-RoC-COR(A/B) based on the C. tepidum Roco structure (from SWISS-MODEL). n, Chimeric model combining LRRK2RCKW and the homology model for the LRR domain from (m) obtained by aligning their RoC-COR(A/B) domains. o, p, Two views of the hybrid LRRK2 model. q, Close-up showing the proximity between the active site of the kinase (with the side chains of its DYG motif shown) and the S1292 autophosphorylation site on the LRR. The close-up also highlights the proximity between N2081, a residue implicated in Crohn’s Disease, and the LRR.

The kinase in our LRRK2RCKW structure is in an open, inactive conformation. Its activation loop contains the site of two familial PD mutations (G2019S and I2020T) and is disordered beyond G2019 (Fig. 1h, Extended Data Fig. 2h, and Supplementary Video 2). R1441 and Y1699 are the sites of three other familial PD mutations and are located at the RoC-COR-B interface (Fig. 1h, Extended Data Fig. 2j, and Supplementary Video 2). Given that the kinase and GTPase interact with each other via the COR-A domain, it is possible that these mutations, located at the interface between the GTPase and COR-B, alter the conformational landscape of LRRK2 in response to ligands and/or regulatory signals and thus affect the cross-talk between LRRK2’s catalytic domains.

A unique feature of LRRK2 is a 28-amino acid α-helix located at its extreme C-terminus, following the WD40 domain (Fig. 1i, j, and Supplementary Video 3). While a number of other kinases have α-helices in the same general location, none form interactions as extensive as those observed in LRRK2 (Fig. 1i, j, Extended Data Fig. 3d-i). Deletion of this helix resulted in an insoluble protein (Extended Data Fig. 1a, b). A residue near its end (T2524) is a known phosphorylation site for LRRK2[20]. Given the close proximity between T2524 and the N-lobe of the kinase domain, as well as the adjacent COR-B domain, we hypothesize that phosphorylation of this residue may play a role in regulating the kinase. Since the last two residues of the C-terminal helix are disordered in our structure, as is a neighboring loop in COR-B, it is possible that conditions exist where these regions become ordered and turn the C-terminal helix into a scaffolding element that connects COR-B, the kinase and the WD40 domains.

We modeled the Leucine-Rich Repeats (LRR) into LRRK2RCKW by using a crystal structure of the LRR, RoC, and COR domains of C. tepidum’s Roco protein[7] (Extended Data Fig. 3l-p). In our model, the LRR wraps around the N-lobe of the kinase and approaches the C-lobe, placing the known S1292 autophosphorylation site in the LRR close to the kinase’s active site, and the Crohn’s disease-associated residue N2081[21], located in the kinase’s C-lobe, next to the LRR (Extended Data Fig. 3q), suggesting the functional relevance of this predicted interface.

Model of microtubule-bound filaments

Recently, a 14Å structure of microtubule-associated filaments of full-length LRRK2 (carrying the filament-promoting I2020T mutation[12]) was determined using in situ cryo-ET and subtomogram analysis[5] (Fig. 2a). The LRRK2 filaments formed on microtubules are right-handed[5]. Because microtubules are left-handed and no strong density connected the LRRK2 filament to the microtubule surface[5], it is unknown if LRRK2’s microtubule interaction is direct. To address this, we combined purified microtubules and LRRK2RCKW, either WT or I2020T, and imaged them by cryo-EM. Both WT and I2020T LRRK2RCKW bound to microtubules, and diffraction patterns calculated from the images revealed layer lines consistent with the formation of ordered filaments (Fig. 2b). Thus, the interaction between LRRK2 and microtubules is direct and the catalytic C-terminal half of LRRK2 is sufficient for the formation of microtubule-associated filaments. The layer line patterns of WT and I2020T LRRK2RCKW are different, with the I2020T diffraction pattern having an additional layer line of lower frequency, indicating longer-range order in the filaments (Fig. 2b). This is consistent with the observation that the I2020T mutation promotes microtubule association by LRRK2 in cells[12]. Understanding the structural basis for this effect will require high-resolution structures of the filaments formed by WT and I2020T LRRK2.

Figure 2 ∣

Modeling the microtubule-associated LRRK2 filaments.

a, 14Å cryo-ET map of a segment of microtubule-associated LRRK2 filament in cells. The microtubule is shown in blue and the LRRK2 filament in grey. b, Microtubule-associated LRRK2RCKW filaments reconstituted in vitro from purified components. (Top) Single cryo-EM images of a naked microtubule (left), and WT (center) and I2020T (right) LRRK2RCKW filaments. (Bottom) Diffraction patterns (power spectra) calculated from the images above. White and hollow arrowheads indicate the layer lines corresponding to the microtubule and LRRK2RCKW, respectively. Scale bar: 20nm c, Fitting of the LRRK2RCKW structure, which has its kinase in an open conformation, into the cryo-ET map. d, Atomic model of the LRRK2RCKW filaments from (c). The white circle highlights the filament interface mediated by interactions between COR domains, where clashes are found. e, Superposition of the LRRK2RCKW structure (colored by domains) and a model of LRRK2RCKW with its kinase in a closed conformation in blue. The dashed blue arrow indicates the closing of the kinase. f, Fitting of the closed-kinase model of LRRK2RCKW into the cryo-ET map. g, Atomic model of the closed-kinase LRRK2RCKW filaments (g) with a white circle highlighting the same interface as in (d). h, i, Cartoon representation of the two filament models, highlighting the clashes observed with open-kinase LRRK2RCKW (h) and resolved with the closed-kinase model (i). 82% of clashes were resolved using the closed-kinase LRRK2RCKW model (see Methods for details).

Previously, integrative modeling was used to build a model into the in situ structure of microtubule-associated LRRK2[5]. This modeling indicated that the well-resolved Cryo-ET density closest to the microtubule was comprised of the RoC, COR, Kinase and WD40 domains and gave orientation ensembles for each domain[5] that are in good agreement with our high-resolution structure of LRRK2RCKW (Extended Data Fig. 4a). Here, we built an atomic model of the microtubule-bound LRRK2 filaments by combining our 3.5Å structure of LRRK2RCKW with the 14Å in situ structure of microtubule-associated LRRK2 (Extended Data Fig. 4b-f). This revealed that the LRRK2RCKW structure is sufficient to account for the density seen in the in situ structure (Fig. 2c), in agreement with our ability to reconstitute microtubule-associated LRRK2RCKW filaments in vitro (Fig. 2b), and with the earlier modeling[5].

Extended Data Figure 4 ∣

Comparison between LRRK2RCKW and integrative models built into cryo-ET maps of LRRK2 filaments in cells and docking of LRRK2RCKW into those maps.

a, Root-mean-square deviation (RMSD) between the atomic model of LRRK2RCKW and each of the 1,167 integrative models generated by Watanabe, Buschauer, Bohning and colleagues[5]. RMSDs were calculated in Chimera[45] using 100% residue similarity and with pruning iterations turned off. RMSD values are grouped into 53 clusters of related models (see[5] for details), with the mean and standard deviation shown whenever the cluster contains two or more models. Integrative models that gave the lowest, median and highest RMSD values are shown. The models are colored according to the per-residue RMSD with the atomic model of LRRK2RCKW. b, The WD40s in the crystal structure of a dimer of LRRK2’s WD40 (PDB: 6DLP)[9] were replaced with the WD40s from our cryo-EM structure of LRRK2RCKW. c, The resulting dimer was fitted into the 14Å cryo-ET map of cellular microtubule-associated LRRK2 filaments[5]. d, Two views of the same fitting shown in (c), displayed with a higher threshold for the map to highlight the fitting of the WD40 β-propellers into the density. The white arrows point towards the holes at the center of the β-propellers densities. e, Four copies of LRRK2RCKW were docked into the cryo-ET map by aligning their WD40 domains to the docked WD40 dimer. f, Model containing the four aligned LRRK2RCKW. g-j, Modeling of the kinase-closed form of LRRK2RCKW. g, h, The structure of ITK bound to an inhibitor (PDB: 3QGY)[46], which is in a closed conformation, was aligned to LRRK2RCKW using only the C-lobes of the two kinases. i, The N-terminal portion of LRRK2RCKW, comprising RoC, COR-A, COR-B and the N-lobe of the kinase, was aligned to ITK using only the N-lobes of the kinases. RoC, COR-A and COR-B were moved as a rigid body in this alignment. j, Kinase-closed model of LRRK2RCKW.

Although our LRRK2RCKW structure fits the overall shape of the cryo-ET map, there were significant clashes at the COR domain interfaces (Fig. 2d). Since the kinase in our LRRK2RCKW structure is in an open conformation, we hypothesized that filament formation might require LRRK2’s kinase to be in a closed conformation. To test this, we modeled a kinase-closed LRRK2RCKW (Fig. 2e and Extended Data Fig. 4g-j) and used it to rebuild the LRRK2 filament. The kinase-closed LRRK2RCKW model resolved more than 80% of the backbone clashes we had observed with our kinase-open LRRK2RCKW structure (Fig. 2c, d, f, g). A closed conformation for the kinase was also proposed by the integrative modeling[5]. Given this data, we hypothesize that the conformation of LRRK2 controls its ability to oligomerize on microtubules, with a closed kinase promoting oligomerization and an open (inactive) one disfavoring it (Fig. 2h, i).

The LRRK2 filaments in our kinase-closed model are formed by two homotypic interactions: one is mediated by the WD40 domain, and the other by the COR-A and COR-B domains (Fig. 3a-d). Similar interfaces were reported based on the cryo-ET structure[5]. We also solved structures of LRRK2RCKW dimers, using the same grids that yielded the 3.5Å structure of LRRK2RCKW. We obtained structures of both the WD40/WD40- and COR/COR-mediated dimers, indicating that both interfaces mediate dimerization in the absence of microtubules (Fig. 3e, f and Extended Data Figs. 5 and 6). The interface in the COR-mediated dimer of LRRK2RCKW differs from that reported for the C. tepidum Roco protein[6,7]; while the GTPase domains interact directly in the dimer of the bacterial protein[7], they are not involved in the dimerization interface we observed for LRRK2 (Extended Data Fig. 6c).

Figure 3 ∣

LRRK2RCKW forms WD40- and COR-mediated dimers outside the filaments.

a-d, The filament model shown in Fig. 2j, k is shown here in grey, with either a WD40-mediated (a), or COR-mediated (c) LRRK2RCKW dimer highlighted with domain colors. The corresponding molecular models are shown next to the cartoons (b, d). e, f, Cryo-EM reconstructions of LRRK2RCKW dimers obtained in the absence of inhibitor (“Apo”), or in the presence of MLi-2. For each reconstruction, two orientations of the map are shown: down the two-fold axis at the dimerization interface (left), which matches the orientation of the models shown in (b, d), and perpendicular to it (right). The top row shows the cryo-EM map and the bottom row a transparent version of it with a model docked in. g, Molecular models of the WD40-mediated and COR-mediated LRRK2RCKW dimers obtained in the presence of MLi-2 (e, f) were aligned in alternating order. This panel shows the resulting right-handed helix. h, The helix has dimensions compatible with the diameter of a 12-protofilament microtubule (EMD-5192)[44], which was the species used to obtain the cryo-ET map shown in Fig. 2a[5], and has its RoC domains pointing towards the microtubule surface.

Extended Data Figure 5 ∣

Ab initio models for cryo-EM of LRRK2RCKW dimers and cryo-EM analysis of WD40- and COR-mediated dimers of LRRK2RCKW in the presence of the inhibitor MLi-2.

a, An initial dataset was collected from a sample of LRRK2RCKW incubated in the presence of the kinase inhibitor MLi-2 and dimers were selected. b, Representative two-dimensional class averages used for ab initio model building. c, Ab initio models with the structure of LRRK2RCKW docked in. d, Volumes generated form the molecular models in (b), filtered to 30Å resolution. e, Projections of the volumes in (d) shown in the same order as their corresponding 2D class averages in (b). f, Data processing strategy for obtaining cryo-EM structures of WD40- and COR-mediated dimers of LRRK2RCKW in the presence of the inhibitor MLi-2. The models used during this processing (see Methods) are those shown in (d) along with an additional linear trimer (see Methods) used for particle sorting.

We built an independent model of a closed-kinase LRRK2RCKW by splitting our 3.5Å structure in half at the junction between the N- and C-lobes of the kinase, and fitting the fragments into our cryo-EM map of a WD40/WD40 dimer obtained in the presence of a LRRK2-specific Type I kinase inhibitor, MLi-2[22,23], which is predicted to stabilize the kinase closed conformation (Extended Data Fig. 7a-c). We then docked this closed-kinase model (Extended Data Fig. 7c) into the cryo-EM maps of WD40- and COR-mediated dimers obtained in the presence of MLi-2 to generate molecular models of both dimers (Extended Data Fig. 7d, e). We aligned these models to build a polymer in silico. This resulted in a right-handed helix with the same general geometric properties seen in the cellular LRRK2 filaments, indicating that those properties are largely encoded in the structure of LRRK2RCKW itself (Fig. 3g, h, and Extended Data Fig. 7f). Docking the same two halves of LRRK2RCKW into the cryo-EM map of a monomer we obtained in the absence of inhibitors or ATP led to a structure similar to our 3.5Å structure obtained from trimers, further confirming that trimer formation does not alter the conformation of LRRK2RCKW (Extended Data Fig. 7g, h).

Extended Data Figure 7 ∣

Properties of the microtubule-associated LRRK2RCKW filaments.

a, b, The LRRK2RCKW structure solved in this work (a) was split at the junction between the N- and C-lobes of the kinase domain (L1949-A1950) (b). c, Docking of the two halves of LRRK2RCKW into a cryo-EM map of a LRRK2RCKW dimer solved in the presence of MLi-2. The dimer map is the same one shown in Fig. 3 and Extended Data Figs. 10 and 11. d, The model obtained in (c) was docked into cryo-EM maps of either WD40- or COR-mediated dimers obtained in the presence of MLi-2. e, Molecular models resulting from the docking in (d). f, Aligning, in alternating order, copies of the dimer models generated in (d, e) results in a right-handed filament with dimensions compatible with those of a microtubule, and its RoC domains pointing inwards (see Fig. 3g-i for more details). g, Docking of the two halves of LRRK2RCKW into a cryo-EM map of a LRRK2RCKW monomer solved in the absence of inhibitor (“Apo”). The map is the one shown in Fig. 1h and Extended Data Fig. 6. h, Three-way comparison of LRRK2RCKW (with domain colors) and the models resulting from the dockings into the MLi-2 WD40-mediated dimer map (c) (dark blue) and “Apo” monomer map (g) (light blue). The three structures were aligned using the C-lobes of their kinases and the WD40 domain. The superposition illustrates that the docking into the “Apo” map results in a structure very similar to that obtained from the trimer (Fig. 1) and that the presence of MLi-2 leads to a closing of the kinase. i, Molecular model of the microtubule-associated LRRK2RCKW filament obtained by docking a fragment of a microtubule structure (PDB: 6O2S) into the corresponding density in the sub-tomogram average (Fig. 2a). j, Same view as in (i) with the models shown as surface representations colored by their Coulomb potential. k, l, “Peeling off” of the structure shown in (j), with the LRRK2RCKW filament seen from the perspective of the microtubule surface (k) and the microtubule surface seen from the perspective of the LRRK2RCKW filament (l). Note: the acidic C-terminal tubulin tails are not ordered in the microtubule structure and thus are not included in the surface charge distributions. The Coulomb potential coloring scale is shown on the right.

These data, along with the apparent lack of any residue-specific interactions between LRRK2 and the microtubule lattice, suggest that the microtubule may provide a surface for LRRK2 to oligomerize on using interfaces that exist in solution, thus explaining the symmetry mismatch between the microtubule and the LRRK2 filament. Consistent with this, the surface charge of the microtubule facing the LRRK2RCKW filament is acidic, while there are basic patches on the LRRK2RCKW filament facing the microtubule (Extended Data Fig. 7i-l). The unstructured C-terminal tails of α- and β-tubulin, which were not included in the surface charge calculations, are also acidic.

LRRK2RCKW inhibits kinesin and dynein

To test our hypothesis that the conformation of LRRK2’s kinase domain regulates its interaction with microtubules, we needed a sensitive assay to measure the association of LRRK2RCKW with microtubules and a means to control the conformation of its kinase. We monitored microtubule association by measuring the effect of LRRK2RCKW on microtubule-based motor motility. We used a truncated human kinesin-1, Kif5B (“kinesin”)[24], which moves towards the microtubule plus end, and the activated human cytoplasmic dynein-1/ dynactin/ ninein-like complex (“dynein”)[25], which moves in the opposite direction. Using single-molecule in vitro motility assays (Fig. 4a), we found that low nanomolar concentrations of LRRK2RCKW inhibited both kinesin and dynein movement, with near complete inhibition at 25 nM LRRK2RCKW (Fig. 4b, c and Extended Data Fig. 8a, b). We hypothesized that LRRK2RCKW was acting as a roadblock for the motors. In agreement with this, the distance that single kinesins moved (run length) was reduced (Fig. 4d), while their velocity remained relatively constant (Fig. 4e). We obtained similar results with dynein (Fig. 4f). Other microtubule-associated proteins, such as MAP2 and Tau, also inhibit kinesin, but not dynein[26,27], likely due to dynein’s ability to sidestep on the microtubule[28-30]. LRRK2’s unusual ability to inhibit dynein may be a consequence of it forming oligomers that cannot be overcome by sidestepping.

Figure 4 ∣

LRRK2RCKW inhibits the motility of kinesin and dynein.

a, Schematic of the single-molecule motility assay. b, c, The percentage (mean ± s.d.) of motile events per microtubule as a function of LRRK2RCKW concentration for kinesin (b) and dynein (c). ****p < 0.0001 (Kruskal-Wallis test with Dunn’s posthoc for multiple comparisons for (b)) and ****p < 0.0001 (Mann Whitney test) for (c)). d, Cumulative frequency distribution of kinesin run lengths as a function of LRRK2RCKW concentration. Mean decay constants (tau) are shown. The 12.5 nM and 25 nM, but not 6.25 nM, conditions were significantly different (p < 0.0001) than the 0 nM condition (one-way ANOVA with Dunnett’s test for multiple comparisons using error generated from a bootstrapping analysis). e, Velocity of kinesin as a function of LRRK2RCKW concentration. Data are mean ± s.d. ****p < 0.0001 (one-way ANOVA with Dunn’s posthoc for multiple comparisons). f, Cumulative frequency distribution of dynein run lengths as a function of LRRK2RCKW concentration. Mean decay constants (tau) are shown. Data was resampled with bootstrapping analysis and was significant. p < 0.0001 (unpaired t-test with Welch’s correction using error generated from a bootstrapping analysis).

Extended Data Figure 8 ∣

Inhibition of motor motility by wild-type and I2020T LRRK2RCKW.

a, Example kymographs showing that increasing concentrations of LRRK2RCKW reduce kinesin runs. b, Example kymographs showing that 25 nM LRRK2RCKW reduces dynein runs. c, Representative kymographs of kinesin motility in the presence or absence of WT and I2020T LRRK2RCKW. d, The percentage of motile kinesin events per microtubule in the absence of LRRK2 or in the presence of 25 nM WT or I2020T LRRK2RCKW. Data are mean ± s.d. (n = 12 microtubules per condition quantified from two independent experiments). There is a significant difference between 0 nM and both 25 nM RCKW conditions (p < 0.0001), but no significant (ns) difference between the inhibitory effects of WT LRRK2RCKW versus I2020T LRRK2RCKW as calculated using the Kruskal-Wallis test with Dunn’s posthoc for multiple comparisons (compared to no LRRK2RCKW).

We also tested the inhibition of kinesin by I2020T LRRK2RCKW, which promotes filament formation when overexpressed in cells[12]. I2020T LRRK2RCKW inhibited kinesin to a similar extent as WT LRRK2RCKW (Extended Data Fig. 8c, d). Given that in vitro reconstituted filaments of I2020T LRRK2RCKW show longer range order compared with WT LRRK2RCKW (Fig. 2b), it is possible that the high sensitivity of the single-molecule motility assays does not allow us to distinguish differences in oligomer length and/or stability between WT and I2020T LRRK2RCKW.

Type II inhibitors rescue the motors

Our hypothesis predicts that the closed conformation of LRRK2’s kinase domain will favor LRRK2’s oligomerization on microtubules. Conversely, it predicts that conditions that stabilize the kinase in an open conformation will prevent oligomerization and thus decrease microtubule binding by LRRK2RCKW, resulting in relief of LRRK2RCKW-dependent inhibition of kinesin and dynein motility. To test these predictions, we searched for a Type II kinase inhibitor that binds tightly to LRRK2 with structural evidence that it stabilizes an open kinase conformation. We selected Ponatinib, which has a Ki for LRRK2 of 31 nM[31], and crystal structures show it bound to RIPK2[32] and IRAK4 (PDB: 6EG9) in open conformations (Extended Data Fig. 9a). Ponatinib inhibited LRRK2RCKW’s ability to phosphorylate Rab8a[13] (Extended Data Fig. 9f).

Extended Data Figure 9 ∣

Type II kinase inhibitors rescue kinesin and dynein motility.

a-e, Ponatinib is a Type II, “DFG out” inhibitor. a, Superposition of the structures of Ponatinib-bound RIPK2 (PDB: 4C8B)[32] and IRAK4 (PDB: 6EG9). Ponatinib is shown in yellow, and the DYG motif residues are shown in white. b, c For comparison, the structures of (a) Roco4 bound to LRRK2-IN-1 (PDB: 4YZM)[35], a LRRK2-specific Type I, “DFG in” inhibitor, and (b) a model of Mitogen-activated kinase 1 (MAPK1) bound to MLi-2 (PDB: 5U6I)[22], another LRRK2-specific Type I, “DFG in” inhibitor are shown. The inhibitor and DFG residues are colored as in (a). d, The structures in (a-c), as well as the kinase from LRRK2RCKW are shown superimposed. The color arrowheads point to the N-lobe’s β-sheet to highlight the difference in conformation between kinases bound to the two different types of inhibitors. Note that LRRK2RCKW’s kinase is even more open than the two Ponatinib-bound kinases. e, Rotated view of (d), now highlighting the position of the N-lobe’s αC helix. An additional alpha helix in the N-lobe of MAPK1 was removed from this view for clarity. f, The kinase inhibitors MLi-2 (1 μM), LRRK2-IN-1 (1 μM), Ponatinib (10 μM) and GZD-824 (10 μM) all inhibit LRRK2RCKW’s kinase activity in vitro compared to a DMSO control. A western blot using a phospho-specific antibody to Rab8a at the indicated time points is shown. g, A dose response curve showing the percentage of motile kinesin events per microtubule as a function of Ponatinib concentration with LRRK2RCKW (25 nM) or without LRRK2RCKW. Data are mean ± s.d. (from left to right: n = 12, 18, 16, 14, and 9 microtubules quantified from one experiment). ****p < 0.0001 (Kruskal-Wallis test with Dunn’s posthoc for multiple comparisons, compared to DMSO without LRRK2RCKW). h, Dose response curve of run lengths from data in (g) represented as a cumulative frequency distribution. From top to bottom: n = 654, 173, 584, 293, and 129 motile kinesin events. Mean decay constants (tau) ± confidence interval (CI) are (from top to bottom) 2.736 ± 0.113, 1.291 ± 0.181, 2.542 ± 0.124, 2.285 ± 0.134, and 1.653 ± 0.17. i, Representative kymographs of kinesin and dynein with DMSO or Type II inhibitors with or without LRRK2RCKW. j, The Type II kinase inhibitors Ponatinib and GZD-824 rescue kinesin run length, represented as a cumulative frequency distribution of run lengths with LRRK2RCKW (25 nM) or without LRRK2RCKW. From top to bottom: n = 893, 355, 507, 499, 524, and 529 runs from two independent experiments. Mean decay constants (tau) ± 95% CI are (from top to bottom) 2.070 ± 0.058, 0.8466 ± 0.091, 1.938 ± 0.065, 2.075 ± 0.07, 1.898 ±, 0.065, and 1.718 ± 0.064. Data were resampled with bootstrapping analysis and statistical significance was established using a one-way ANOVA with Dunnett’s test for multiple comparisons. DMSO run lengths were significantly different (p < 0.0001) between conditions (0 vs. 25 nM RCKW). Ponatinib (0 vs. 25 nM RCKW) and GZD-824 (0 vs. 25 nM LRRK2) were not significant. k, Same as in (j) but with dynein. From top to bottom: n = 659, 28, 289, 306, 254, and 339 runs from two independent experiments. Mean decay constants (tau) ± 95% confidence intervals; microns are 4.980 ± 0.147, 0.846 ± 0.415, 4.686 ± 0.142, 4.445 ± 0.172, 3.156 ± 0.09, 3.432 ± 0.188 (from top to bottom). Statistical significance as in (j) and run lengths were significantly different (p < 0.0001) between DMSO conditions (0 vs. 25 nM RCKW), and not significant for Ponatinib or GZD0824 conditions. The DMSO conditions are reproduced from Fig. 4f for comparison. l, Expression levels of GFP-LRRK2 (I2020T) in 293T cells treated with either DMSO or GZD-824 (5 μM). An Immunoblot with anti-GFP (LRRK2) and anti-GADPH (loading control), which is a representative image from three replicates, is shown. m, Quantification of GFP-LRRK2 (I2020T) expression levels from western blots similar to (l). Data are mean ± s.d. (n = 3 per condition). GZD-824 is not significantly different from the DMSO-treated control (Mann-Whitney test). n, 293T cells immunostained for tubulin showing that the microtubule architecture is not affected by GZD-824 or MLi-2 compared to DMSO treatment.

As our hypothesis predicted, Ponatinib rescued kinesin motility in a dose-dependent manner at concentrations of LRRK2RCKW (25 nM) that had resulted in almost complete inhibition of kinesin (Fig. 5a, Extended Data Fig. 9g-j). We observed similar effects with GZD-824, a chemically-related Type II kinase inhibitor[33] (Fig. 5a and Extended Data Fig. 9i, j). Our hypothesis also predicted that kinase inhibitors that stabilize the closed form of the kinase would not rescue the motors and might enhance the inhibitory effect of LRRK2RCKW by increasing its ability to form filaments on microtubules. Indeed, MLi-2[22,23] and another LRRK2-specific Type I inhibitor, LRRK2-IN-1[34], which are expected[22,35] to stabilize a closed conformation of the kinase (Extended Data Fig. 9b-e), further enhanced the inhibitory activity of LRRK2RCKW on kinesin (Fig. 5a). Dynein motility was also rescued by Ponatinib and GZD-824, but not by MLi-2 or LRRK2-IN-1 (Fig. 5b, Extended Data Fig. 9i, k). Like Ponatinib, GZD-824, MLi-2 and LRRK2-IN-1 inhibited phosphorylation of Rab8a by LRRK2RCKW (Extended Data Fig. 9f).

Figure 5 ∣

Type II, but not Type I, kinase inhibitors rescue kinesin and dynein motility and reduce LRRK2 filament formation in cells.

a, b, Effects of different kinase inhibitors on LRRK2RCKW’s inhibition of kinesin (a) and dynein (b) motility. Data shown is the percentage of motile events per microtubule (MT) as a function of LRRK2RCKW concentration in the absence (DMSO) or presence of the indicated inhibitors (Ponatinib and GZD-824: 10 μM; MLi-2 and LRRK2-IN-1: 1 μM). Data are mean ± s.d. ***p < 0.001 and ****p < 0.0001 (Kruskal-Wallis test with Dunn’s posthoc for multiple comparisons within drug only). c, Treatment with MLi-2 (500 nM) for 2 hrs increases WT GFP-LRRK2 filament formation in 293T cells. Data are mean ± s.d. ****p=0.0002 (Mann Whitney test). d, Treatment with GZD-824 (5 μM) for 30 mins decreases GFP-LRRK2 (I2020T) filament formation in 293 cells. Data are mean ± s.d. *p=0.0133 and **p=0.0012 (Kruskal-Wallis with Dunn’s posthoc test for multiple comparisons). e, Schematic representation of our hypothesis. LRRK2’s kinase can be in an open or closed conformation. The different species we observed are represented in the rounded rectangles, but only monomers are shown on the microtubule for simplicity. Our model proposes that the kinase-closed form of LRRK2 favors oligomerization on microtubules.

GZD-824 reduces filaments in cells

In cells, LRRK2 forms filaments that colocalize with a subset of microtubules and are sensitive to the microtubule depolymerizing drug nocodazole[12]. This association is enhanced by the PD-linked mutations R1441C, R1441G, Y1699C and I2020T[12,36] and by Type I kinase inhibitors[34,37]. We tested our kinase conformation hypothesis in human 293T cells by determining if Type I and Type II kinase inhibitors had opposite effects on the formation of microtubule-associated LRRK2 filaments in cells. Consistent with previous findings[37,38], the Type I inhibitor MLi-2 enhanced LRRK2’s microtubule association (Fig. 5c), suggesting that the closed conformation of the kinase favors binding to microtubules in cells. In contrast, we found that the Type II inhibitor GZD-824 reduced the filament-forming ability of overexpressed LRRK2 (I2020T) (Fig. 5d). This reduction in LRRK2 filament formation was not due to changes in LRRK2 protein expression levels or the overall architecture of the microtubule cytoskeleton (Extended Data Fig. 9l-n).

Conclusions

Here we reported the 3.5Å structure of the catalytic half of LRRK2 and used it, in combination with a 14Å cryo-ET structure of microtubule-associated LRRK2 filaments[5], to build an atomic model of these filaments. This modeling led us to hypothesize that the conformation of LRRK2’s kinase controls its association with microtubules (Fig. 5e). Cryo-EM structures of LRRK2RCKW dimers we obtained in the absence of microtubules showed that the same interfaces are involved in both dimerization and filament formation, and aligning them in silico resulted in a right-handed filament with similar properties to LRRK2 filaments observed in cells, suggesting that the ability of LRRK2 to form filaments is a property inherent to LRRK2, and specifically the RCKW domains. We propose that both the surface charge and geometric complementarity between the microtubule and LRRK2 promote the formation of LRRK2 filaments. It remains to be determined whether LRRK2RCKW monomers or dimers are the minimal filament forming unit.

We tested our model that the conformation of LRRK2’s kinase regulates microtubule association using kinase inhibitors expected to stabilize either the open (Type II) or closed (Type I) conformations of the kinase. In support of our model, Type II inhibitors relieved the LRRK2RCKW-dependent inhibition of kinesin and dynein and reduced LRRK2 filament formation in cells, while Type I inhibitors failed to rescue motor motility and enhanced filament formation in cells. Importantly, our single-molecule motility assays showed that low nanomolar concentrations of LRRK2RCKW negatively impact both kinesin and dynein. At these low concentrations it is likely that LRRK2 would not form the long, highly ordered filaments and microtubule bundles observed in cells overexpressing the protein; instead, we hypothesize that at endogenous expression levels in cells LRRK2 forms short oligomers on microtubules.

What is the physiological role of non-pathogenic microtubule-associated LRRK2? Our data show that LRRK2 acts as a roadblock for microtubule-based motors in vitro. In cells, dynein and kinesin bind directly or indirectly to many Rab-marked cargos[39-43]. Our data also show that the microtubule-associated form of LRRK2 has its kinase in a closed (and potentially active) conformation. Given this, microtubule-associated LRRK2 stalling of kinesin or dynein could increase the likelihood that LRRK2 phosphorylates cargo-associated Rabs[13], modulating effector binding[13] and resulting in changes in cargo dynamics. In support of this, the four PD mutations that enhance LRRK2 microtubule binding[12] also show higher levels of Rab phosphorylation in cells than the G2019S mutant[13,14], whose microtubule binding is not enhanced over wild-type LRRK2[12].

Our data have important implications for the design of LRRK2 kinase inhibitors for therapeutic purposes. They predict that inhibitors that favor the closed conformation of the kinase will promote LRRK2 filament formation, thus blocking microtubule-based trafficking, while inhibitors that favor an open conformation of the kinase will not. These results should be taken into account to enhance therapeutically beneficial effects of LRRK2 kinase inhibition and to avoid potential unintended side effects.

Extended Data Table 1 ∣

Cryo-EM data collection and model refinement statistics.

The model refinement statistics are reported for four different types of models, two including GDP-Mg2+ in the RoC domain and two excluding it. In each case, we report statistics for two types of model: “Monomer w/interfaces” consists of a LRRK2RCKW monomer plus fragments from the neighboring monomers in the C3 trimer that were used during model building and refinement; “Top 10 Monomers” are the top-10 results from Rosetta Relax with the neighboring fragments removed after processing in Rosetta. PDB accession numbers for the models and the EMD code of the maps used for model-building and refinement are indicated. EMD-21250 contains both the C3 map of the LRRK2RCKW trimer used to build the COR-B, kinase and WD40 domains and the signal-subtracted monomer used to build the RoC and COR-A domains. The final models reported here were refined into the signal-subtracted monomer map. (See Methods for details.)

	Trimer	Monomer(Apo)	Dimers(Apo)	Dimers(Mli-2)
	Dataset	Dataset	Dataset	Dataset
Data Collection
Number of grids used	1	10	3	2
Microscope	Titan Krios	Talos Arctica	Talos Arctica	Talos Arctica
Camera	Gatan K2	Gatan K2	Gatan K2	Gatan K2
Camera Mode (Counting/Super-Res.)	C	C/SR	C/SR	C
Voltage (kV)	300	200	200	200
Magnification	130,000	36,000	36,000	36,000
Pixel size (Å/pixel)	1.07	1.16/0.58	1.16/0.58	1.16
Dose rate (e−/Å2 second)	6.65	4.2-10	4.6-7.8	5.5
Exposure time (s)	8	6-12	7-11	9-10
Total dose (e−/Å2)	53.2	50.4-60	50.6-54.6	49.5-55
Number of frames	40	30-60	35-55	45-50
Defocus range (μm)	1-1.8	1-2	1-2	1-2
Micrographs collected (no.)	3,824	11,354	5,303	4,139
Initial particle number	836,956	1,184,330	431,387	706,807
Final particle number	70,954*/105,787[#]	15,999	63,435[a]/21,497[b]	72,249[a]/36,727[b]
Final Resolution (0.143	3.5*/3.8*	8.1	13.4[a]/9.5[b]	10.2[a]/9.0[b]
FSC threshold) (Å)
LRRK2RCKW Models	(without GDP-Mg2+)		(with GDP-Mg2+)
	Monomer w/interfaces	Top 10 Monomers**	Monomer w/interfaces	Top 10 Monomers**
	PDB: 6VP8EMD-21250	PDB: 6VP7EMD-21250	PDB: 6VP6EMD-21250	PDB: 6VNOEMD-21250
Model Refinement
Map-to-model resolution (0.143 FSC threshold) (Å)	3.98	3.98	3.98	3.98
Map B-factor (Å2)	−136	−136	−136	−136
Model Composition
Non-hydrogen atoms	13,161	8,534	13,161	8,534
Protein Residues	1,668	1,084	1,668	1,084
Ligands	0	0	1	1
B-factors (Å2)
Protein	42.75	49.76	45.49	50.4
Ligand	-	-	193.78	193.78
R.M.S. Deviations
Bond length (Å)	0.013	0.013	0.090	0.112
Bond angle (°)	1.374	1.384	2.238	2.596
Validation
Molprobity score	1.37	1.48	1.42	1.51
Clash score	1.69	2.35	2.03	2.61
Poor rotamers (%)	0,00	0.00	0.00	0.00
Ramachandran
Favored (%)	92.90	92.58	93.02	92.57
Allowed (%)	5.99	6.36	5.99	6.32
Disfavored (%)	1.11	1.06	0.99	1.11

C3 reconstruction.

signal-subtracted monomer.

WD40-mediated dimer.

COR-mediated dimer.

Numbers represent the average of the values for all 10 models.