Nde1 promotes Lis1 binding to full-length autoinhibited human dynein 1

Abstract

Cytoplasmic dynein 1 (dynein) is the primary motor responsible for the retrograde transport of intracellular cargoes along microtubules. Activation of dynein requires the opening its autoinhibited Phi conformation, a process driven by Lis1 and Nde1/Ndel1. Using biochemical reconstitution and cryo-electron microscopy, we demonstrate that Nde1 enhances Lis1 binding to autoinhibited dynein and facilitates Phi opening. We identify a key intermediate in this activation pathway where a single Lis1 dimer binds between Phi-like (Phi^L) motor rings. In this ‘Phi^L–Lis1’ complex, Lis1 interacts with one motor domain through canonical sites at the AAA+ (adenosine triphosphatases associated with diverse cellular activities) ring and stalk, and with AAA5, AAA6 and linker regions of the other motor domain. Mutagenesis and motility assays confirm the critical role of the Phi^L–Lis1 interface in dynein activation. This intermediate forms rapidly in the presence of Nde1, although Nde1 is not part of Phi^L–Lis1. These findings provide key insights into how Nde1 promotes Lis1-mediated Phi opening.

Cytoplasmic dynein is a motor protein responsible for transporting nearly all intracellular cargoes toward the minus ends of microtubules (MTs). Its cargoes include organelles, proteins, RNA, viruses and vesicles¹. Additionally, dynein has key roles in other cellular processes such as mitosis and organelle positioning^1,2. Mutations affecting dynein or its regulatory proteins have been linked to neurodevelopmental and neurodegenerative diseases, including spinal muscular atrophy, amyotrophic lateral sclerosis, Huntington disease, lissencephaly and microcephaly^3-9.

The 1.4-MDa dynein complex comprises paired copies of six distinct subunits. The largest subunit is the dynein heavy chain (DHC), which contains the N-terminal tail domain and C-terminal motor domain. The tail facilitates dimerization of DHCs, recruits the two copies of intermediate chain (DIC) and light intermediate chain (DLIC) and interacts with dynactin, cargo adaptors and other regulatory proteins^10-13. The N terminus of DIC (DIC-N) recruits three pairs of light chains, Robl, LC8 and Tctex^1,14,15. DIC segments binding to LC8 and Tctex, known as the IC–LC tower, have a crucial role in assembling the dynein–dynactin–adaptor (DDA) complex^11,12,16-19. The motor domain belongs to AAA+ (adenosine triphosphatases (ATPases) associated with diverse cellular activities) protein families and consists of six AAA subdomains (AAA1–AAA6)^20,21. AAA1 is the primary subdomain that hydrolyzes ATP to power dynein motility along MTs. The dynein motor domain attaches to MTs through the coiled-coil stalk and the small globular MT-binding domain and connects to the tail through the linker domain located at the surface of the AAA+ ring^21-24.

Dynein alternates between two key conformational states: autoinhibited Phi and open conformations^14,25. Phi dynein adopts a compact structure that limits its interactions with MTs and dynactin, which serves to minimize unnecessary ATP hydrolysis when the motor is not engaged in active transport^14,26. Opening of Phi enables dynein to assemble with dynactin and the cargo adaptor, allowing processive movement along MTs^10,14.

Activation of dynein motility is also facilitated by dynein regulators Nde1/Ndel1 and Lis1 (refs. 3,15,27-34) but the underlying mechanism promoting Phi to open transition is unclear. Lis1, the first gene identified in a neuronal migration disease, lissencephaly, has a crucial role in dynein-related function^3,11,35-43. Lis1 possesses an N-terminal LisH domain that facilitates dimerization and a C-terminal WD40 β-propeller domain that binds to Nde1/Ndel1, dynein and other proteins³. Both domains are important at different stages of dynein activation^44-46. The LisH domain interacts with dynactin p150, thus promoting the recruitment of dynactin and adaptors at the later stage of dynein activation¹¹. WD40 domains of Lis1 can directly bind to the dynein motor domain at the AAA3–AAA4 sites (site–ring) and coiled-coil stalk (site–stalk)^11,42,47.

Functional studies proposed that Lis1 facilitates the formation of the DDA complex by favoring the release of Phi dynein and stabilizing open dynein^38-41,48. Lis1 binding is thought to be incompatible with Phi dynein as a function of the steric clash when docking a Lis1 β-propeller to the motor domains of Phi dynein^3,39,41,48. However, there is no direct biochemical or structural evidence of whether Lis1 can open Phi dynein. It also remains unclear whether Lis1 and dynein form intermediate states before the opening of Phi dynein. A recent structural study on tail-truncated yeast dynein motor domains reported a Chi conformation, in which two Lis1 dimers are wedged between two AAA+ rings⁴¹, suggesting that two Lis1 dimers between the motor domains are required to crack open Phi dynein. Furthermore, it is well known that the AAA+ ring undergoes substantial conformational changes under different nucleotide-binding states, which can potentially regulate dynein–Lis1 binding. A more recent study demonstrated that specific nucleotide ‘codes’ at the three nucleotide-binding sites (AAA1, AAA3 and AAA4) govern the stoichiometry of dynein–Lis1 interactions by tuning their binding affinity at two distinct locations⁴⁹. However, the intermediate structure for full-length human dynein alone bound to Lis1 is still lacking and whether dynein forms a Chi conformation at the initial state of its activation remains unclear.

Nde1 and its paralog Ndel1 (refs. 12,27) are critical for all dynein-mediated functions in cell division^32,50, cargo trafficking^51,52 and neuronal migration^53,54. Nde1/Ndel1 is composed of the N-terminal coiled-coil domain and the C-terminal unstructured region^17,55,56. The coiled-coil domain interacts with DIC-N, which overlaps with the DIC-binding site of the p150 subunit of dynactin^17-19. It also interacts with the WD40 domains of Lis1 and its binding site on the WD40 domain overlaps with the Lis1-binding site of dynein^17,30. The multiprotein interaction modes of Nde1/Ndel1, along with its overlapping binding sites with other proteins, make the functional interpretation of how Nde1/Ndel1 contributes to the activation of dynein elusive. It has been proposed that Nde1 tethers Lis1 to dynein^12,30,57-59 and promotes Lis1-mediated activation of dynein^11,30 (Fig. 1a). According to this model, Nde1 recruits Lis1 to Phi dynein. When Phi dynein transiently adopts an ‘open’ conformation, locally enriched Lis1 binds the open dynein and prevents it from transitioning back to Phi, thus favoring the assembly of the DDA complex^11,38,39,48. Consistent with this model, overexpression of Lis1 can rescue the deletion of Nde1/Ndel1 in cells^59,60 and Nde1/Ndel1-mediated recruitment of Lis1 to dynein enhances DDA assembly in vitro³⁰. An alternative model suggests that Ndel1 negatively regulates dynein activation by competing with p150 for DIC-N binding and by sequestering Lis1 away from dynein¹⁷. Consequently, how Lis1 and Nde1/Ndel1 form a complex with dynein and promote the opening of Phi dynein is not well understood.

Fig. 1 ∣. Nde1 promotes Lis1 binding to Phi dynein and synergistically releases dynein autoinhibition.

a, Schematic of dynein activation. Nde1 tethers Lis1 to dynein and promotes Lis1-mediated formation of the active DDA complex. Intermediate states between the association of Lis1–Nde1 and the opening of the Phi conformation are unknown. b, MP shows that Lis1 alone slowly binds to dynein over a time course of 30 min, whereas Nde1 promotes more efficient binding of Lis1 to dynein within 1 min. Dynein, Lis1 and Nde1 were mixed at a 1:2:2 ratio. Only one Lis1 dimer is tethered to one dynein dimer. c, MP shows that Lis1 efficiently binds to open dynein and Nde1 does not further promote Lis1 binding to open dynein compared to the Lis1-alone condition. d, A representative image of WT dynein motors captured using negative-stain EM. Without Lis1 and Nde1, dynein is distributed almost equally between Phi and open conformations (n = 6 biological replicates). e, Relative percentage of Phi dynein in the presence and absence of Lis1 and Nde1 (mean ± s.d.); n = 6 biological replicates for D:L = 1:1 and 1:2; n = 3 for D:N = 1:1; n = 4 for D:N = 1:2; n = 4 for D:L:N = 1:1:1 and 1:2:2. P values were calculated from a two-tailed t-test. The control (Ctrl) represents Phi percentage under dynein alone from Extended Data Fig. 3 and is normalized to 100%. In b,c, solid curves represent a fit to multiple Gaussians to predict the average mass and percentage of each population.

To understand how Nde1/Ndel1 and Lis1 prime dynein for the DDA assembly, we reconstituted and visualized their effects on full-length human dynein. We show that Nde1 promotes dynein–Lis1 complex formation and, together with Lis1, greatly biases the equilibrium from Phi to the open conformation. Cryo-electron microscopy (cryo-EM) reveals a key intermediate, the Phi-like (Phi^L)–Lis1 complex, where a Lis1 dimer binds between the two motor domains of Phi dynein. Lis1 engages canonical and newly identified interaction sites on dynein, which are essential for Nde1-mediated dynein activation, as shown by mutagenesis and single-molecule assays. These findings establish the critical role of this Phi^L–Lis1 and shed light on how Nde1 and Lis1 perform critical duties in the dynein activation pathway.

Results

Nde1 promotes Lis1 binding and rescue of Phi dynein

To determine whether Nde1 promotes Lis1 binding to dynein, we performed mass photometry (MP) assays to quantify the Lis1 and Nde1 binding to full-length human dynein under different conditions. In the absence of Nde1, Lis1 alone exhibited an increased binding to dynein in a time-dependent manner, with the formation of a 37% 1:1 dynein:Lis1 (DL) complex within 60 min. The inclusion of Nde1 greatly enhanced the dynein–Lis1 binding (69% of complex formation within a minute), suggesting that Nde1 accelerates this binding event, consistent with Nde1/Ndel1-mediated tethering of Lis1 to dynein^12,30,58 (Fig. 1b). To further test whether Nde1 is needed to recruit Lis1 to the Phi or open conformation, we performed these assays using the Phi mutant of dynein¹⁴ that only forms the open conformation, referred to as open dynein. Lis1 was readily bound to open dynein and the addition of Nde1 did not enhance Lis1 binding to dynein (Fig. 1c), suggesting that Nde1 facilitates Lis1 binding to Phi dynein. These results are also consistent with the previous observation that the Nde1 addition does not further enhance Lis1-mediated activation of DDA complexes assembled with the Phi dynein mutant³⁰. Interestingly, the formation of the dynein–Lis1–Nde1 tripartite complex was not observed (Fig. 1b,c), regardless of different nucleotide conditions (Extended Data Fig. 1a-e). Although previous single-molecule imaging^12,30 and pulldown assays^17,30 reported that Nde1 interacts with DIC-N, we also did not detect the dynein–Nde1 complex (Fig. 1b and Extended Data Fig. 1), suggesting that Nde1 rapidly dissociates from the complex after handing off Lis1 to dynein¹².

To determine whether Nde1-mediated Lis1 recruitment to dynein shifts the equilibrium between Phi and open conformations, we used negative-stain EM imaging¹⁴ to quantify the ratio of Phi dynein under four conditions: without Lis1 or Nde1, with Lis1 alone, with Nde1 alone and with both Lis1 and Nde1. In the absence of Lis1 and Nde1, we found that ~50% of motors formed the Phi conformation (Fig. 1d and Extended Data Figs. 2a-d and 3a-c). We then incubated dynein with Lis1 and Nde1 and found that Lis1 or Nde1 alone does not change the Phi ratio (Fig. 1e and Extended Data Fig. 3a-c). However, the Phi ratio decreased to 44% when we incubated dynein with both Lis1 and Nde1 at a ratio of 1:2:2 (Fig. 1e and Extended Data Fig. 3c). Collectively, our results demonstrate that Nde1 specifically promotes Lis1 binding to Phi dynein and facilitates opening of this autoinhibited conformation.

The Phi^L–Lis1 structure

Our MP results suggest that there is a rate-limiting step during Lis1 binding to Phi dynein, which can be greatly accelerated by the addition of Nde1 (Fig. 1b). We used cryo-EM to capture the potential intermediate states to reveal the structural basis of this process (Extended Data Figs. 4a-d and 5a-c). We focused on the particles adopting the autoinhibited dynein conformation (Fig. 2a,b, and Extended Data Figs. 4a-d and 5a,b). In the presence of Lis1, we unexpectedly observed a structure in which a Lis1 dimer is wedged between the two stacked motor rings of Phi dynein (Supplementary Table 1). The dynein in this complex shows a compact conformation, similar to but not the same as the Phi structure¹⁴. We referred to this complex as Phi^L–Lis1 (Fig. 2a,c and Supplementary Video 1). Despite excess Lis1, nearly half of Phi dynein does not bind to Lis1 (42.7% Phi versus 57.3% Phi^L–Lis1) (Fig. 2a). Notably, we did not observe the Chi particles (two Lis1s bound to dynein), suggesting that the Chi conformation may be specific to yeast dynein or may form when truncated dynein containing only the motor domains is used instead of a full-length motor⁴¹.

Fig. 2 ∣. Structure of the Phi^L–Lis1 complex.

a, Top: full-length human dynein was incubated with Lis1 at a molar ratio of 1:2 on ice for 90 min before being flash-frozen. Bottom: Cryo-EM maps of Phi and Phi^L–Lis1 complex structures in front view and percentages of particles with these two conformations (open dynein excluded). b, Top: dynein was incubated with Lis1 and Nde1 at a molar ratio of 1:2:2 on ice for 5 min before being flash-frozen. Bottom: Structure of Phi^L–Lis1 obtained under this condition in front view and side view. All particles were classified into the Phi^L–Lis1 conformation and the Phi conformation could not be detected. c, Model of the Phi^L–Lis1 motor domains in three views. One Lis1 dimer is clamped between MD-A and MD-B of Phi^L–Lis1 on the front side, whereas the back side of the motor domains is vacant, with no Lis1 bound. The top view reveals that the linker regions (purple) of MD-A and MD-B interact with each other.

In the Phi^L–Lis1 structure, Lis1 localizes to the same side as the IC–LC tower of dynein (front side) (Fig. 2a,b), possibly because it has a different affinity to bind front and back sides of an asymmetric full-length dynein dimer. Previous results suggested that Lis1 can affect dynein’s mechanochemical cycle and nucleotide state^11,61,62. The high-resolution structure of the motor domain enabled us to identify and compare the nucleotide states of the motor domains in Phi and Phi^L–Lis1 structures. Specifically, we found that AAA1 pockets of both motor domains exhibit the same adenosine diphosphate (ADP)-Mg²⁺ density (Extended Data Fig. 6a,b), accompanied by a flexible sensor I loop, indicating the intermediate state of phosphate release (Extended Data Fig. 6c,d). Similarly, AAA3 pockets show clear ADP binding (Extended Data Fig. 6a,b), indicating that Lis1 binding does not influence the nucleotide states of the motor domains of autoinhibited dynein.

We next premixed equimolar amounts of Nde1 and Lis1 and then added dynein to achieve a final 1:2:2 ratio of dynein, Lis1 and Nde1 (Fig. 2b). Strikingly, we observed only Phi^L–Lis1 particles and no remaining Phi dynein in the presence of Nde1 (Fig. 2b, Extended Data Fig. 5a,b, Supplementary Table 1 and Supplementary Video 1). This is consistent with the MP analysis showing that Nde1 instantly promotes Lis1 binding to dynein (Fig. 1b) and forming Phi^L–Lis1. These results suggest that Phi^L–Lis1 formation is a rate-limiting intermediate state before dynein opening. Importantly, no discernible Nde1 density was observed in any of the Phi^L–Lis1 three-dimensional classes, indicating that Nde1 is not retained in this complex.

We also analyzed individual motor domains of open dynein and did not observe an increase in the propensity of the dynein–Lis1 complex compared to open dynein alone in the presence of Nde1 (Extended Data Fig. 5c). Consistent with the MP of open dynein¹⁴ (Fig. 1c), these results show that Nde1 promotes Lis1 binding to Phi dynein to form Phi^L–Lis1, whereas Lis1 does not require Nde1 to bind open dynein.

Lis1 induces a rotation in the Phi to enable its binding

Consistent with previous reports^39,41,48, our structural analysis reveals a severe steric clash between Lis1 and a motor domain of Phi dynein (Fig. 3a and Supplementary Video 1) and this needs to be relieved to accommodate Lis1’s binding in Phi^L–Lis1 (Fig. 3b). Comparing Phi to Phi^L–Lis1 reveals a relative rotation between the two motor domains, which results in the groove on the front side of dynein becoming larger than the corresponding groove on the back side (Fig. 3c,d). The enlarged groove on the front side allows Lis1 to fit between Phi^L motor domains. However, docking of Lis1 to the back side with a smaller groove shows severe clashes, explaining why there is only one Lis1 present in Phi^L–Lis1(Figs. 3e and 2a,b). Our MP results also support the 1:1 binding mode observed in Phi^L–Lis1 (Fig. 1b). In addition, the rotation between the motor domains in Phi^L–Lis1 causes a slight anticlockwise twist in the neck region (Fig. 3d), which may promote partial unwinding of the tail and lower the energy barrier to switch to the opening of the Phi conformation.

Fig. 3 ∣. Lis1 induces a transition in the Phi conformation to accommodate its binding.

a, Docking of a Lis1 monomer between dynein motor rings shows a clash with the rigid canonical Phi MD-A. b, A Lis1 dimer bound to dynein motors on the front side is compatible with the coordinated Phi^L structure. c, Top: the superimposition of the Phi and Phi^L–Lis1 motor domains. Phi and Phi^L–Lis1 are colored gray and rainbow, respectively. Bottom: comparison of the motor domains between Phi^L–Lis1 and canonical Phi. Lis1 is hidden for clarity. Vectors present the interatomic distance of pairwise Cα atoms between the Phi to Phi^L–Lis1 structures. Lis1 binding induces a slight rotation of the dynein motor domains from the Lis1-bound (front) side toward the back, resulting in a larger groove at the front side and a smaller groove at the back. d, Overlay of the motor domain maps of Phi and Phi^L–Lis1 with MD-A aligned at a lower contour level. The rotation of Phi^L–Lis1 motor domain induces rotation in the neck region, showing a slight unwinding trend, which contributes to the formation of a smaller groove on the back side. e, Docked Lis1 on the back side of Phi^L–Lis1 causes a notable clash with MD-B.

Interactions identified in Phi^L–Lis1

To define the structural basis of Lis1 binding to Phi dynein, we analyzed the Phi^L–Lis1 interfaces, which are further examined functionally in the next section (Fig. 4). In Phi^L–Lis1, WD40 domains of Lis1 interact with both motor domains of dynein (MD-A and MD-B). While Lis1 binds to MD-B through its canonical interaction sites at the AAA3, AAA4 and AAA5 regions^42,47,63, we observed previously uncharacterized interaction sites of Lis1 with AAA5 and AAA6 and linker regions of MD-A (Fig. 4a and Supplementary Video 1). Sequence alignment of Lis1 homologs shows that the MD-A and Lis1 interface is highly conserved among metazoans but less conserved in yeast (Supplementary Fig. 1), suggesting different regulatory roles of Lis1 between higher and lower eukaryotes. Structural comparison indicates that the two motor domains in Phi^L–Lis1 adopt an almost identical conformation. The root-mean-squared displacement (r.m.s.d.) of Cα atoms of the two motor domains was 0.5 Å (Fig. 4b). Additionally, MD-B binding to Lis1_ring and Lis1_stalk shows no notable difference from the structure of Lis1 bound to the human dynein motor domain⁴² (Cα r.m.s.d. = 0.9 Å) (Fig. 4c), suggesting that Lis1 binding does not induce structural changes within an individual motor domain.

Fig. 4 ∣. Interactions identified between Lis1_ring and MD-A in Phi^L–Lis1.

a, Cryo-EM density map highlighting the subdomains involved in the interface between Lis1 and the dynein motor. Lis1 and subdomains of MD-A and MD-B are colored separately. Interfaces between MD-A and Lis1_ring are marked by a dashed rectangle and enlarged in d–f. b, Comparison of MD-A (gray) and MD-B (rainbow) of Phi^L–Lis1. c, Comparison of MD-B (rainbow) with the Lis1-bound structure of the motor domain of human dynein (PDB 8FDT, gray). d–f, Representative interactions located at the AAA5–Lis1_ring (d), AAA6–Lis1_ring (e) and linker–Lis1_ring (f) interface are shown in stick representation, colored according to their respective subdomains. g, MP profiles illustrate the interaction of Nde1 (N) with WT Lis1 (L) and Lis1 mutants (Lis1^linker: residues (300–304)>A and M329A; Lis1^AAA6: N203K, D205K and Q222A; Lis1^AAA5: D192K, Y225A and C226D; Lis1^All: D192K, D205K, E300A and K303A). Nde1 interacts with one (NL) or two (NL₂) Lis1 dimers. h, MP shows the binding of Lis1 mutants to WT dynein with or without Nde1. Dynein, Lis1 and Nde1 were incubated for 2 min at a 1:2:2 ratio. i, Representative kymographs show the motility of WT DDR complexes with or without Nde1 and Lis1. j, Run frequency of WT DDR with or without Nde1 and Lis1 (mean ± s.d.; n = 30 MTs for each condition from two independent experiments). Results were normalized to the –Lis1, –Nde1 condition. P values were calculated from a two-tailed t-test. In g,h, solid curves represent a fit to multiple Gaussians to predict the average mass and percentage of each population.

The WD40 domain of Lis1 interacts with MD-A at regions distributed across AAA5, AAA6 and the linker regions (Fig. 4d-f). The AAA5–Lis1_ring WD40 interface is mainly composed of residues Q3636, S3613, T3612, D3616, D3617, A3618 and K3621 of dynein MD-A and Y225, C226, G190, H191, M172 and D192 of the Lis1 WD40 domain (Fig. 4d and Extended Data Fig. 7a). Notably, the side chain of Y225 of Lis1 forms a polar interaction with the side chain of Q3636. Additionally, the side chain of D3616 and the main chain of D3617 form hydrogen bonds with the main chain of C226. The side chain of A3618 forms hydrophobic interaction with the main chain of G190, while M172 of Lis1 forms hydrophobic interactions with the aliphatic portion of K3621. H191 engages in polar interaction with the terminal amino group of K3621.

Within the AAA6 interaction region, the interface is characterized by polar interactions involving N203, D205 and Q222 of the Lis1 WD40 domain and K4089, N4085, Q4117 and S4115 of MD-A. The side chain of K4089 forms a salt bridge with the side chain of D205 and establishes a hydrogen bond with the oxygen atom in the main chain of N203. Q222 and D205 of Lis1 also form polar interactions with S4115 and Q4117 and N4085 residues of dynein (Fig. 4e and Extended Data Fig. 7a).

We also observed interactions of the WD40 domain of Lis1 and the linker of MD-A. Specifically, the M329 side chain of Lis1 engages in hydrophobic interactions with V1563 and P1562 of the dynein linker. A loop (residues 298–309) on the surface of the Lis1 WD40 domain is located at the interface between dynein MD-A and Lis1_ring. Given their spatial proximity, residues 300–304 of this loop may also contribute to linker–Lis1_ring interactions. However, these side chains are not visible because of poor local density, likely caused by structural flexibility in this loop (Fig. 4f and Extended Data Fig. 7b).

Additionally, Phi dynein exhibits interactions between MD-A and MD-B interface similar to the previously reported structure¹⁴, including those at the linker–linker, linker–AAA4, AAA4–AAA4, AAA5–AAA5 and stalk–stalk regions (Fig. 5a,c). In contrast, Phi^L–Lis1 forms different interactions between MD-A and MD-B (Fig. 5b,d), involving the linker–linker, linker–AAA4, AAA3–linker, AAA5–AAA5 and stalk–stalk interfaces. Detailed views of the interface reveal a network of potential interactions (Fig. 5d), suggesting that Phi^L–Lis1 adopts a stable conformation.

Fig. 5 ∣. Comparison of the interface between MD-A and MD-B in Phi and Phi^L–Lis1.

a, Representative residues located on MD-A that are involved in the MD-A and MD-B interface of Phi, showing interactions at the linker–linker, linker–AAA4, AAA4–linker, AAA4–AAA4, AAA5–AAA5 and stalk–stalk interfaces. b, Representative residues located on MD-A of Phi^L–Lis1, involved in the MD-A and Lis1_ring interface (including linker–Lis1_ring, AAA6–Lis1_ring and AAA5–Lis1_ring), and the MD-A and MD-B interface (including linker–linker, AAA3–linker, linker–AAA4, AAA5–AAA5 and stalk–stalk interfaces). Residues in a,b are displayed in sphere representation. c, Detailed view of MD-A and MD-B interface of Phi, showing key residues at the linker–linker, linker–AAA4, AAA4–AAA4 and AAA5–AAA5 interfaces. d, Detailed view of the MD-A and MD-B interface in Phi^L–Lis1, showing key residues involved in interactions at the linker–linker, linker–AAA4, AAA3–linker and AAA5–AAA5 interfaces.

Phi^L–Lis1 regulates Nde1-dependent dynein activation

To evaluate whether these interaction sites we detected between Lis1 and dynein MD-A in Phi^L–Lis1 are critical for the activation of dynein motility, we introduced sets of mutations targeting Lis1 interactions with AAA5, AAA6 and the linker of dynein. Key residues of Lis1 at each interface (Lis1^AAA5, Lis1^AAA6 and Lis1^linker) and all three interfaces (Lis1^All) were subjected to alanine substitutions or charge reversals (Supplementary Table 2). Similar to wild-type Lis1 (WT Lis1), these Lis1 mutants formed homodimers, interacted with Nde1 (Fig. 4g) and bound to open dynein (Extended Data Fig. 8a). Lis1^linker exhibited more efficient binding to dynein in the presence of Nde1. In comparison, Lis1^AAA6 and Lis1^AAA5 had a substantially lower increase in dynein binding by Nde1 (Fig. 4h), indicating that these interactions contribute to the formation of the Phi^L–Lis1 complex. Mutations impacting all three interfaces fully diminished the Nde1-dependent increase in dynein binding of Lis1 (Fig. 4h). Notably, Lis1^AAA6 mutants resulted in a mass population corresponding to dynein bound to one Nde1 and two Lis1s (Fig. 4h). This population decreased while the population corresponding to dynein bound to single Lis1 increased over the incubation time (Extended Data Fig. 8b,c). Therefore, the complex with one dynein, one Nde1 and two Lis1s may serve as an intermediate state for the formation of Phi^L–Lis1.

To assess the impact of these mutations on dynein motility, we monitored the motility of single complexes assembled with WT dynein, dynactin and the BicDR1 adaptor (DDR) on surface-immobilized MTs in the presence and absence of Lis1 and Nde1. Consistent with our previous observations³⁰, WT Lis1 enhanced the run frequency of DDR about 2.5-fold and Nde1 and Lis1 together increased the run frequency ninefold (Fig. 4i,j and Supplementary Video 2). In the absence of Nde1, Lis1 mutants activated dynein motility at similar levels to WT Lis1. In the presence of Nde1, Lis1^linker triggered activation of DDR motility similarly to WT Lis1 (Fig. 4i,j). In comparison, Nde1 exhibited a modest increase in dynein run frequency together with Lis1^AAA5 and failed to enhance dynein motility with Lis1^AAA6 and Lis1^All (Fig. 4i,j). These results demonstrate that disrupting the interactions of Lis1 with AAA5 and AAA6 sites of dynein in the Phi^L–Lis1 conformation reduces the Nde1-mediated activation of dynein by Lis1, highlighting the importance of the Phi^L–Lis1 complex in the dynein activation pathway. We also generated different combinations of Lis1 mutants, such as deletion of the loop that binds to MD-A linker alone (Δ300–304) (Extended Data Fig. 8d), combined with mutations impacting AAA5-binding and AAA6-binding sites. Motility assays with these mutants also showed that Lis1 interactions with AAA5 and AAA6 are crucial for the formation of Phi^L–Lis1, while Lis1 binding to the linker has a less critical role in Lis1 binding to dynein and activation of dynein motility (Extended Data Fig. 9a-e).

Discussion

In this study, we investigated the structure and mechanism of how Lis1 and Nde1 rescue dynein from autoinhibition before the assembly of active dynein transport machinery. Using negative-stain EM imaging, we directly showed that Nde1 and Lis1 synergistically promote the opening of Phi dynein, whereas neither Nde1 nor Lis1 alone was sufficient to induce notable Phi opening (Fig. 1b,e). Lis1 can readily bind to open dynein (Fig. 1c) and facilitate the assembly of the DDA complex in the absence of Nde1 (ref. 30). Although Lis1 can also bind to Phi dynein and open this autoinhibited conformation on its own⁴⁹, we show that Nde1 substantially promotes docking of Lis1 to Phi dynein and switching of Phi to the open conformation. Nde1 likely functions as an assembly factor that transiently interacts with and dissociates from dynein after handing off Lis1, promoting dynein–Lis1 complex formation without being retained in the final assembly. Although previous studies have shown that dynein or DIC-N forms a stable complex with Nde1/Ndel1 (refs. 12,17-19,30,58,59,64), we did not detect it in our MP assay. This is likely attributed to differences in dynein constructs, Nde1 versus Ndel1 or assay conditions. Because Nde1 has an overlapping binding site with the p150 subunit of dynactin on DIC-N¹², dissociation of Nde1 from DIC-N after tethering Lis1 to dynein may enable efficient recruitment of dynactin to dynein–Lis1 complexes.

Our structural and functional studies of dynein–Lis1 complexes revealed a key intermediate step on Lis1–Nde1-mediated opening of dynein. Although a dynein dimer contains Lis1-binding sites on each motor domain, Phi dynein recruits a single Lis1. Using cryo-EM imaging, we revealed a Phi^L–Lis1 structure in which the two AAA+ rings of Phi dynein rotate slightly backward to accommodate binding of a single Lis1 dimer to the front side. This rotational motion reduces the spacing between the AAA+ rings and prevents binding of another Lis1 to the back side. Preferential binding of Lis1 to the front side may be because of the asymmetric orientation of the dynein complex in the Phi confirmation. In addition, Nde1 may tether Lis1 to Phi dynein by interacting with DIC-N^12,18,64, leading to formation of the transient dynein–Lis1-Nde1 only on the IC–LC tower side.

The Phi^L–Lis1 complex is fundamentally distinct from the previously reported Chi conformation of yeast dynein monomers⁴¹. The Chi conformation is stabilized by two Lis1 dimers, one on each side, and adopts a more open and extended conformation compared to the Phi^L–Lis1 motor domains (Extended Data Fig. 10a-c). In comparison, we used full-length human dynein and could not detect Chi particles even when we used this conformation as a reference during cryo-EM image processing. It is possible that isolated dynein motor domains can recruit two Lis1s and form a more extended Chi confirmation, whereas structural constraints imposed by the tail of full-length dynein restrict the relative movement of the Phi motor domains. As a result, Lis1 binding to the front side of Phi^L dynein reduces the spacing on the back side, thereby preventing the formation of Chi. Most interaction sites located at the Lis1_ring surface of Chi–Lis1 are also present at that of Phi^L–Lis1 (Supplementary Fig. 1) but Phi^L–Lis1 exhibits additional pairwise interactions between Lis1 and dynein (Fig. 4d-f, Extended Data Fig. 7 and Supplementary Fig. 1). Consistent with the Phi^L–Lis1 conformation, we did not detect complexes with one dynein and two Lis1s in MP, suggesting that Phi^L–Lis1 rather than Chi is the stable intermediate of full-length human dynein.

The mutagenesis of the interactions between Lis1 and the AAA5 and AAA6 subdomains of MD-A in Phi^L–Lis1 disrupts Nde1’s ability to promote Lis1-mediated dynein activation, confirming that the Phi^L–Lis1 is a key intermediate in the dynein activation pathway. However, these mutations did not disrupt the mutant Lis1’s ability to increase the run frequency of dynein severalfold on its own. This is because WT dynein can be in open or Phi conformations with near equal probability in our assay conditions. Our model predicts that mutant Lis1 can still bind and enhance DDA assembly of open dynein without Nde1, whereas efficient activation of Phi dynein requires both Nde1 and Lis1. Lis1 cannot stimulate the motility of the Phi^L mutants of dynein synergistically with Nde1 because these mutants are deficient in formation of the critical Phi^L–Lis1 intermediate. In addition, a small proportion of complexes containing one dynein, two Lis1 dimers and one Nde1 dimer (DL2N) was detected under the Lis1^AAA6 condition by MP analysis. This DL₂N complex could contribute to the reduced run frequency, as the failure to release Nde1 may compete with the binding of p150 of dynactin to DIC-N, thereby hindering the proper assembly of the DDA complex^12,27,30.

On the basis of our results and previous observations, we propose a mechanism underlying dynein activation by Lis1 and Nde1. In the absence of Nde1, Lis1 alone can bind to dynein motor domains¹¹, inducing a conformational change from canonical Phi to Phi^L. However, the efficiency of this process is low and Lis1 cannot open Phi^L dynein on its own (Figs. 1e and 2a). In the presence of Nde1, Lis1–Nde1 is initially recruited to DIC-N positioned at the front side of Phi dynein, facilitating more efficient binding of Lis1 to the front side of Phi. The local enhancement of Lis1 near the Phi motor by Nde1 facilitates more efficient binding of Lis1 to Phi dynein (Fig. 6, (i) and (ii)). Nde1 and Lis1 form a transient Phi^L–Lis1–Nde1 complex (Fig. 6, (i)). Nde1 dissociates spontaneously, leading to Phi^L–Lis1 formation (Fig. 6, (ii)). Lis1 binding induces a slight backward rotation of the two motor rings in Phi^L–Lis1 (Fig. 3c), suggesting an intermediate state before opening of Phi dynein. Additionally, the rotation of the dynein motor domains relative to each other induces a slight twist in the neck region, which may cause an unwinding trend in the tail and contribute to dynein opening (Fig. 3d). Subsequently, Phi^L–Lis1 transitions to open dynein–Lis1 with the assistance of Nde1 (Fig. 6, (iii)). The binding of Lis1 to dynein facilitates DDA assembly and activates dynein motility by recruiting the p150 subunit of dynactin to dynein through its LisH domain¹¹ (Fig. 6, (iv)). Future studies are required to understand how Nde1 hands off Lis1 to dynein and why it rapidly dissociates from dynein. In addition to its tethering role and facilitating the formation of Phi^L–Lis1, it remains to be determined whether Nde1 has roles in helping Lis1 convert Phi^L–Lis1 to open dynein.

Fig. 6 ∣. Model for the initial stage of dynein activation by Nde1 and Lis1.

i, Nde1 tethers Lis1 to Phi dynein and forms a transient Phi^L–Lis1–Nde1 complex. ii, Nde1 dissociates spontaneously from dynein, leading to the formation of Phi^L–Lis1. Without Nde1, Lis1 binding to Phi dynein and formation of Phi^L–Lis1 becomes a less efficient and slower process. The binding of Lis1 induces a slight backward rotation of the two motor rings and a twist in the neck region, which may partially unwind the tail and contribute to dynein opening. iii, Phi^L–Lis1 transitions to open dynein–Lis1 assisted by Nde1. iv, Lis1 further promotes DDA assembly by interacting with p150 and DIC-N through its LisH domain. The dashed line indicates the possibility that Lis1 in the Phi^L–Lis1 complex promotes dynactin recruitment and DDA assembly without switching to the open conformation.

Methods

Cloning and expression

The plasmid encoding full-length human dynein⁶⁵ was generously provided by A. Carter (His-ZZ-TEV-SNAPf DHC1_IC2C_LIC2_Tctex1_Robl1_LC8, Addgene plasmid 111903). The His-ZZ-TEV-SNAPf tag is fused to the N terminus of the DHC. Human Lis1 and Nde1 (N-terminal residues 1–190, which functions similarly to full-length Nde1 in the single-molecule motility assay³⁰) and the mouse BIDCR1 gene were each cloned individually into the pOmniBac backbone. The constructs featured a ZZ–TEV tag at the N terminus and a SNAPf tag at the C terminus. Lis1 mutants containing point mutations were generated by PCR using mutagenic primers designed to introduce specific substitutions. The mutations were verified by Oxford Nanopore full-plasmid sequencing. The constructs used in this study are listed in Supplementary Table 2. The primers used for mutant construction are provided in Supplementary Table 4.

These proteins were all expressed in insect Sf9 cells using established protocols^14,23,30 with slight modifications. Bacmid DNA isolated from the DH10MultiBac competent cells (Geneva Biotech) were transfected into the in Sf9 insect cells with the Cellfectin II (Gibco) reagent. Baculoviruses used for large-scale expression were amplified in Sf9 cells through two rounds of amplification. Protein expression in Sf9 cells was induced by infecting the cells with P2 virus at a cell density of 2.5 million cells per ml. For dynein expression, 28 ml of P2 virus was added to a 1.4-L culture of Sf9 cells. For Lis1, Nde1 and BicDR1, 7 ml of P2 virus was used to infect 0.7 L of Sf9 cells. Cells were harvested after 75 h by centrifugation at 1,000g for 15 min at 4 °C. The cell pellets were flash-frozen in liquid nitrogen and stored at −80 °C.

Protein purification

Full-length human dynein was purified from Sf9 cells at 4 °C. Cell pellets from a 1.4-L cell culture were resuspended in 100 ml of lysis buffer (50 mM HEPES pH 7.2, 100 mM NaCl, 1 mM DTT, 0.1 mM ATP and 10% glycerol), supplemented with two tablets of cOmplete EDTA-free protease inhibitor (Roche) and 2 mM PMSF. The suspension was homogenized using a Dounce homogenizer with a tight pestle for 15–25 strokes, followed by clarification through centrifugation at 65,000g with a Ti70 rotor (Beckman) for 1 h at 4 °C. The supernatant was then incubated with 3 ml of IgG Sepharose 6 fast flow resin (Cytiva) for 3–4 h on a roller at 4 °C. The resin was washed sequentially with 200 ml of lysis buffer and 200 ml of TEV buffer (50 mM Tris–HCl pH 7.4, 150 mM potassium acetate, 2 mM magnesium acetate, 1 mM EGTA, 10% glycerol, 0.1 mM ATP and 1 mM DTT). Afterward, the resin was incubated with TEV buffer supplemented with 400 ug TEV protease overnight at 4 °C. The flowthrough was collected and concentrated with a Amicon concentrator (100-kDa molecular weight cutoff), then loaded into a TSKgel G4000 column (Tosoh) preequilibrated with the GF150 buffer (25 mM HEPES pH 7.2, 150 mM KCl, 1 mM MgCl₂, 5 mM DTT and 0.1 mM ATP). Peak fractions were collected and concentrated to 2–3 mg ml⁻¹ for cryo-EM grid preparation. The quality of the sample was evaluated with SDS–PAGE and negative-stain EM.

The purification of Lis1, Nde1 and BicDR1 from a 0.7-L cell culture followed a similar protocol to that of dynein, with a few modifications. Specifically, 50 ml of lysis buffer was used to resuspend the cell pellets, and ATP was omitted from the GF150 buffer. Superose 6 column (Cytiva) was used for size-exclusion chromatography. The concentrated proteins were aliquoted, flash-frozen in liquid nitrogen and stored at −80 °C. The quality of the proteins was assessed using SDS–PAGE gels.

Dynactin was isolated from pig brains through a series of purification steps, including an SP Sepharose fast flow column and MonoQ ion-exchange chromatography (Cytiva), followed by size-exclusion chromatography using a TSKgel G4000SWXL column (Tosoh), as described in a previous protocol⁶⁶. Four pig brains were rinsed in cold PBS, trimmed of large blood vessels and the brainstem and then snap-frozen in liquid nitrogen. Brain tissue was ground to powder under liquid nitrogen and resuspended in 2 L of lysis buffer (HB: 35 mM PIPES pH 7.2, 5 mM MgSO₄, 1 mM EGTA, 0.5 mM EDTA, 1 mM DTT and 0.1 mM ATP) supplemented with 1.6 mM PMSF and three Roche protease inhibitor tablets. The lysate was clarified by centrifugation at 19,000g for 45 min using a type 19 rotor at 4 °C, followed by a second spin at 45,000g for 45 min with Ti45 rotor at 10 °C. The supernatant was filtered through a glass fiber prefilter and loaded onto an SP Sepharose fast flow column equilibrated with buffer A (HB). After washing, dynactin was eluted using a gradient of buffer B (HB with 1 M KCl) and peak fractions were pooled and filtered. The sample was then loaded onto a MonoQ column equilibrated with HB and eluted with a three-step salt gradient (15%, 35% and 100% buffer B). Dynactin-containing fractions were concentrated and further purified by size-exclusion chromatography (TSKgel G4000SWXL column) in size-exclusion chromatography buffer (25 mM HEPES pH 7.2, 150 mM KCl, 1 mM MgCl₂, 5 mM DTT and 0.1 mM ATP). Peak fractions were verified for purity by negative staining and SDS–PAGE, then collected, concentrated to 1–4 mg ml⁻¹, adjusted to 10% glycerol and snap-frozen.

MT reconstitution

MTs were reconstituted using porcine tubulin, which was either purchased from Cytoskeleton or purified in house in MT buffer (25 mM MES, 70 mM NaCl, 1 mM MgCl₂, 1 mM EGTA and 1 mM DTT, pH 6.5). The tubulin was concentrated to 10 mg ml⁻¹ at 4 °C, then flash-frozen and stored at −80 °C. To polymerize the MTs, the tubulin was diluted to 5 mg ml⁻¹ in MT buffer supplemented with 3 mM GTP. The tubulin mixture was incubated on ice for 5 min and then transferred to a 37 °C incubator for 1 h. After incubation, the MTs were pelleted at 20,000g for 8 min at room temperature and resuspended in MT buffer supplemented with 5 µM paclitaxel before being stored at room temperature.

MP assays

High-precision coverslips (Azer Scientific) were cleaned by alternating washes with isopropanol and water three times in a bath sonicator and then air-dried. The gasket was cleaned similarly, without sonication and air-dried before being placed onto a clean coverslip. A total of 19.5 μl of filtered MP buffer (30 mM HEPES pH 7.4, 5 mM MgSO₄, 1 mM EGTA and 10% glycerol) was added to a well for autofocus. The protein mixture was prepared with 200 nM dynein, 400 nM Lis1 and 400 nM Nde1. This was diluted 40-fold in the MP buffer by applying 0.5 μl to the well, resulting in final concentrations of 5 nM dynein, 10 nM Lis1 and 10 nM Nde1. Protein contrast data were collected using a TwoMP MP instrument (Refeyn 2) with two technical replicates. The instrument was calibrated with a standard mix of conalbumin, aldolase and thyroglobulin. MP profiles were analyzed by fitting to multiple Gaussian peaks, with the mean, s.d. and percentages calculated using DiscoverMP software (Refeyn). The data for parameters of a multi-Gaussian fit of MP measurements is summarized in Supplementary Table 3.

Negative-stain EM and data quantification

Freshly purified dynein was diluted in GF150 buffer to a final concentration of 14.3 nM in the presence of 0.1 mM ATP and subsequently evaluated using negative-stain EM. A 4-μl aliquot of the sample was applied to glow-discharged carbon film grids (EM Sciences) and stained with 2% uranyl acetate. The grids were then imaged using a 120-kV Talos L120C EM instrument. Micrographs were manually acquired at a magnification of ×45,000. More than 40 micrographs were collected per experiment for each condition to obtain sufficient particles for statistical analysis.

Samples for statistical analysis of the Phi ratio were prepared with the following molar ratios: dynein and Lis1(dimer) at 1:0, 1:1 and 1:2; dynein and Nde1(dimer) at 1:0, 1:1 and 1:2; dynein, Lis1 and Nde1 at 1:0:0, 1:1:1 and 1:2:2. The mixtures were incubated on ice for 90 min and subsequently subjected to negative staining. Each experimental group was accompanied by its respective control (dynein alone). The assay was repeated independently three or more times, using different batches of freshly purified protein.

Micrographs were processed using cryoSPARC, including blob picker, micrograph extraction and two-dimensional (2D) classification. Micrographs from each experimental group were merged and particle picking was performed using templates of Phi and open dynein. The particle diameter was set to 750 Å and the distance cutoff for dynein particles was 400 Å to optimize particle selection. All particles were extracted, followed by three rounds of 2D classification. Phi and open dynein particles were identified from the 2D classification and traced back to the corresponding micrographs for each condition, where Phi and open particles and single motor particles were quantified. The single motor parties were counted as open dynein, with this particle number halved. To calculate the normalized fraction of Phi, the total number of Phi and open dynein particles was determined for micrographs under a given condition. The ratio of Phi particles in the dynein-alone condition (n(Phi)/n(Phi + open)) was defined as proportion A, serving as the control for each group. The ratio of Phi particles in each experimental conditions, excluding the dynein-alone control, was defined as proportion B (n(Phi)/n(Phi + open)). The normalized fraction of Phi was calculated as B/A, with the control group defined as 100%. Graphs were generated using GraphPad Prism.

Cryo-EM sample preparation

For the dynein, Lis1 and Nde1 sample, Lis1 and Nde1 were incubated at a 1:1 molar ratio for 30 min on ice. Freshly purified dynein, at a concentration of 2 mg ml⁻¹, was then added to the Lis1–Nde1 complex at a 1:2:2 molar ratio and incubated on ice for 5 min, with 5 mM ATP added immediately before freezing. For the dynein and Lis1 complex, dynein (2 mg ml⁻¹) was incubated with Lis1 for 90 minutes on ice and 5 mM ATP was added just before vitrification.

For vitrification, 3.5 μl of the prepared sample was applied to glow-discharged Quantifoil holey carbon grids (R2/1, 300 mesh gold), which were treated for 45 s at 25 mA using a GloQube glow discharge system (Quorum Technologies). The grids were blotted for 2.5 to 4.5 s at 4 °C and 100% humidity and then vitrified by plunging into liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific).

Cryo-EM data collection

Data were collected at the Yale ScienceHill cryo-EM facility using a Glacios microscope (Thermo Fisher Scientific) operated at 200 keV and equipped with a K3 detector. Data collection was facilitated by SerialEM software, targeting a defocus range of −1.5 to −2.7 μm. Four exposures per hole were recorded as videos, comprising 40 frames each, with a total electron dose of 40 e⁻ per Ų. A total of 7,128 movies were collected for the dynein–Lis1–Nde1 condition, while 16,558 movies were collected for the dynein–Lis1 condition.

Cryo-EM data processing

Cryo-EM movies were preprocessed using cryoSPARC Live, which included patch motion correction and patch contrast transfer function (CTF) estimation. The processing workflows are illustrated in Extended Data Figs. 4 and 5. The statistics are summarized in Supplementary Table 1.

For the dynein–Lis1 condition dataset, particles were picked using the blob picker, extracted with a box size of 512 pixels and downscaled to 128 pixels with a pixel size of 3.456 Å. In total, 7,254,515 particles were extracted. The iterative 2D classification was performed to filter the particles, resulting in the selection of 204,005 high-quality particles for ab initio reconstruction. Initial maps for the dynein single motor domain and the Phi^L–Lis1 motor domains were identified. The map of the Phi^L–Lis1 motor domains was subsequently used for heterogeneous refinement of all original particles. The original particles were divided into four subsets, each subjected to heterogeneous refinement (four classes). Three rounds of heterogeneous refinement were performed, updating the reference each time, ultimately identifying the Phi and Phi^L–Lis1 motor domains. The Phi and Phi^L–Lis1 motor domains were merged separately and extracted from micrographs using a box size of 512 pixels, which was then binned to 384 pixels, resulting in a pixel size of 1.1573 Å. Two rounds of heterogeneous refinement were conducted to exclude junk particles. High-quality subsets were selected for homogeneous refinement, followed by two rounds of CTF refinement and local refinement. The Phi^L–Lis1 motor domains achieved a resolution of 2.86 Å and the Phi motor domains reached a resolution of 2.71 Å, exhibiting C₂ symmetry.

To reconstitute the tails of the Phi^L–Lis1 and Phi motor domains, the particles were recentered at the tail, then extracted from micrographs using a box size of 512 pixels and binned to 256 pixels, yielding a pixel size of 1.736 Å per pixel. Following this, heterogeneous refinement was applied to filter the particles and high-quality subsets were selected for homogeneous refinement. The overall tail resolutions reached 4.21 Å for the Phi tail and 4.05 Å for the Phi^L–Lis1 tail. Four masks were devised to cover the tail segments, which were divided into the NDD, left, right and neck regions. Maps of local refinement using these masks were integrated, referring to the consensus map of the tail. The full-length map was assembled by stitching together the tail and motor domains in ChimeraX, corresponding to Phi and Phi^L–Lis1, respectively.

For the Lis1–Nde1–dynein condition dataset, the process closely mirrored that of the dynein–Lis1 condition described above. The blob picker identified 2,393,279 particles. Following extraction from the micrographs and iterative 2D classification, 204,005 particles were selected for initial map generation through ab initio reconstruction. An initial map for the Phi^L–Lis1 motor domains was obtained, which was then subjected to iterative heterogeneous refinement using all particles with the Phi^L–Lis1 initial map as a reference. However, the Phi motor domains did not appear in the heterogeneous refinement, even when the verified Phi map from this study was used as a reference. Subsequently, the Phi^L–Lis1 domain map was reextracted from the micrographs and underwent two rounds each of heterogeneous refinement, CTF refinement and local refinement, achieving a resolution of 2.88 Å. The consensus map of the tail reached a resolution of 6.22 Å after recentering and reextracting the tail region. Masks and local refinement were used to enhance the local resolution of the tail. The full-length Phi^L–Lis1 structure was reconstituted by integrating the composite tail map and motor map in ChimeraX.

Model building and refinement

For model building, previously reported structures from the Protein Data Bank (PDB 9BLY (ref. 23), PDB 9BLZ (ref. 23) and PDB 8FDT (ref. 42)) were used as the initial models for the full-length and motor domains of Phi and Phi^L–Lis1. The individual domains, including the tail, single motor and Lis1 dimer were extracted from PDB 9BLY, PDB 9BLZ and PDB 8FDT and rigid-body fitting into the cryo-EM maps was performed using UCSF ChimeraX. The models were then manually constructed in Coot^67,68 and followed by real-space refinement in PHENIX⁶⁹. The quality of the refined models was assessed using the MolProbity module integrated into PHENIX, with the statistics reported in Supplementary Table 1.

Single-molecule motility assay

Fluorescent imaging was conducted using a custom-built, multicolor objective-type total internal reflection fluorescence (TIRF) microscope with a Nikon Ti-E microscope body. It was equipped with a ×100 magnification (1.49 numerical aperture) apochromatic oil-immersion objective (Nikon) and a Perfect Focus System. Fluorescence signals were captured by an electron-multiplied charge-coupled device camera (Andor, Ixon EM+, 512 × 512 pixels), with an effective pixel size of 160 nm after magnification. Probes such as Alexa488–GFP–mNeon-Green, LD555 and LD655 were excited by 488-nm, 561-nm and 633-nm laser beams (Coherent), coupled to a single-mode fiber (Oz Optics), and their emissions were filtered using 525/40-nm, 585/40-nm and 697/75-nm bandpass filters (Semrock), respectively. The entire system was controlled using MicroManager 1.4 software.

Biotin–PEG-treated flow chambers were treated with 5 mg ml⁻¹ streptavidin for 2 min, followed by washing with MB buffer (30 mM HEPES pH 7.0, 5 mM MgSO₄, 1 mM EGTA, 1 mg ml⁻¹ casein, 0.5% pluronic acid, 0.5 mM DTT and 1 μM taxol). Biotinylated MTs were then added to the chamber for 2 min and washed again with MB buffer. Proteins were prepared by diluting them to the desired concentrations in MB buffer. For DDRNL complex assembly, a mixture of 10 nM dynein, 150 nM dynactin, 50 nM BicDR1, 200 nM Lis1 and 10 nM Nde1 was incubated on ice for 15 min, then diluted tenfold into imaging buffer (MB buffer containing 0.1 mg ml⁻¹ glucose oxidase, 0.02 mg ml⁻¹ catalase, 0.8% d-glucose and 2 mM ATP) and introduced to the flow chamber. Motility was observed and recorded for 40 s.

Data analysis for single-molecule motility assay

Single-molecule motility of the DDR complex was captured for 200 frames per imaging area and analyzed as kymographs made in FIJI. Run frequency was determined by counting the number of processive BicDR1 molecules on each MT and then dividing this number by the MT length and the total data collection time, using a custom MATLAB script. The P values for the two-tailed Student’s t-test were determined in Excel.

Extended Data

Extended Data Fig. 1 ∣. MP analysis of nucleotide conditions on Nde1 and Lis1 binding to WT dynein.

MP shows that under apo buffer (a), 0.1 mM ATP (b), ADP (c), ATP.vi (d), and AMPPNP (e) conditions, Nde1 promotes Lis1 binding to dynein, forming a 1:1 dynein-Lis1 (DL) complex. The nucleotide condition does not affect Nde1’s ability to tether Lis1 to dynein. Importantly, the formation of dynein-Nde1, dynein-Lis1-Nde1, and 1:2 dynein-Lis1 complexes was not observed. In the Lis1-alone condition, no significant DL complex was formed immediately.

Extended Data Fig. 2 ∣. Workflow for negative-stain EM data processing.

a, Representative micrographs for dynein alone (42 micrographs), dynein-Lis1 at 1:1 (41 micrographs), and dynein-Lis1 at 1:2 (47 micrographs) molar ratios from batch #1. b, Particle picking from representative micrographs in each dataset using a template matching approach based on Phi and open dynein models (particle diameter: 750 Å, distance cutoff: 400 Å). c, Three rounds of 2D classification were performed after extracting all particles (box size: 960 Å × 960 Å), yielding class averages of Phi dynein, open dynein motors, single motors, and junk particles. d, Final classified 2D averages showing Phi dynein (43.7%), two-motor open dynein (43.3%), and single-motor open dynein (13.0%). The particle numbers for each group were counted, and single motors were considered as open dynein by dividing the total number of particles by two.

Extended Data Fig. 3 ∣. Quantification of Phi-dynein percentage in the presence or absence of Lis1 and Nde1.

Dynein (D) was incubated with Lis1 (L), Nde1 (N), or both proteins for 90 minutes at the indicated molar ratios of protein dimers. The percentage of Phi-dynein was quantified under each condition, D + L (a), D + N (b), D + L + N (c). Colors represent independent biological replicates. n = 6 biological replicates for D:L = 1:1 and D:L = 1:2; n = 3 biological replicates for D:N = 1:1 and n = 4 biological replicates for D:N = 1:2; n = 4 biological replicates for D:L:N = 1:1:1 and D:L:N = 1:2:2. The control (Ctrl) corresponds to the Phi-dynein percentage measured in the dynein-alone condition.

Extended Data Fig. 4 ∣. Cryo-EM data processing for the dynein-Lis1 dataset.

a, A representative cryo-EM micrograph and the flowchart of cryo-EM data processing. b, Fourier Shell Correlation (FSC) curves showing the final resolution estimates for the motor domains of the Phi (2.71 Å) and Phi^L-Lis1(2.86 Å) datasets. Orientation distribution of Phi (c) and Phi^L-Lis1(d).

Extended Data Fig. 5 ∣. Cryo-EM data processing for the dynein-Lis1-Nde1 dataset.

a, A representative cryo-EM micrograph and the flowchart of cryo-EM data processing. b, FSC curve showing the final resolution estimate for the motor domains of the Phi^L-Lis1 (2.88 Å) dataset. c, Comparison of Nde1’s effect on Lis1 binding to the open dynein motor. Cryo-EM particle numbers for open dynein-Lis1 and open dynein alone were quantified from the dynein-Lis1 and dynein-Lis1-Nde1 datasets. In the dynein-Lis1 dataset, 493,173 particles correspond to single dynein motors unbound to Lis1, and 624,475 particles correspond to dynein motors bound to Lis1 (including Lis1 dimers and monomers). In the dynein-Lis1-Nde1 dataset, 162,229 particles correspond to dynein motors unbound to Lis1, and 170,995 particles correspond to dynein motors bound to Lis1 (including Lis1 dimers and monomers), indicating that Nde1 does not promote Lis1 binding to the open dynein motor.

Extended Data Fig. 6 ∣. Comparison of local resolution, nucleotide binding in AAA1, AAA3, AAA4, and sensor-I loop conformation in MD-A of the Phi and Phi^L-Lis1.

Local resolution, and nucleotide binding states in MD-A at AAA1, AAA3, and AAA4 of the Phi (a) and Phi^L-Lis1 (b). MD-A and -B share the same nucleotide binding in AAA1, AAA3, and AAA4 across both the Phi and Phi^L-Lis1. The sensor-I loop adopts almost the same conformation in MD-A (or -B) of both Phi (c) and Phi^L-Lis1 (d), indicating that Lis1 binding does not affect phosphate release. The color scheme is the same as Fig. 4.

Extended Data Fig. 7 ∣. Density quality at the dynein MD-A and Lis1 interface in Phi^L-Lis1.

a, Well-defined density at the AAA5-Lis1_ring and AAA6-Lis1_ring regions, showing tight and stable interactions. The color scheme for the motor domains is consistent with Fig. 4. b, Flexible density at the linker-Lis1ring interface, indicating dynamic interactions in this region.

Extended Data Fig. 8 ∣. Binding characterization of Lis1 mutants and structural prediction of Lis1^Δ300–304.

a, MP profiles show binding of Lis1 mutants to open dynein. Open dynein and Lis1 mutants were mixed at a 1:2 ratio and incubated for 2 min prior to measurements. b, MP profiles show binding of the Lis1^AAA6 mutant to dynein with Nde1 at different incubation times. Dynein, Lis1^AAA6, and Nde1 were mixed at 1:2:2 ratio (D: dynein only, DL: one dynein and one Lis1, DL₂N: one dynein, two Lis1s, and one Nde1) and incubated for 2, 6, and 15 minutes before the measurements. c, The percentages of mass populations detected in b. d, Overlay of AlphaFold3-predicted Lis1^Δ300-304 and WT Lis1 bound to Phi^L dynein motor domains. Lis1^Δ300-304 is shown in violet, while WT Lis1 is colored green and sky blue. Dynein motors domains are displayed in grey and white. The structure overlay indicates that deletion of residues 300–304 in Lis1 surface loop does not induce notable conformational change of the rigid elements of Lis1^Δ300-304.

Extended Data Fig. 9 ∣. Characterization of additional mutants of Lis1 at the Phi^L-Lis1 interface.

a, MP profiles illustrate the interaction of Nde1 (N) with Lis1 mutants. Nde1 interacts with one (NL) or two (NL₂) Lis1 dimers. b, MP of the binding of Lis1 mutants to open dynein. Dynein and Lis1 were incubated for 2 minutes at 1:2 ratio (D: dynein only, DL: one dynein and one Lis1). c. MP shows the binding of Lis1 mutants to dynein with or without Nde1. Dynein, Lis1 and Nde1 were incubated for 2 minutes at 1:2:2 ratio (DL₂N: one dynein, two Lis1s, and one Nde1). d, Representative kymographs show the motility of WT DDR complexes with or without Nde1 and Lis1. e, Run frequency of WT DDR with or without Nde1 and Lis1 (mean ± s.d.; n = 30 MTs for each condition; statistics from two independent experiments). Results were normalized to the -Lis1, -Nde1 condition. P values are calculated from a two-tailed t-test.

Extended Data Fig. 10 ∣. Comparison of yeast Chi-Lis1 and human Phi^L-Lis1 motor domains.

a, The structure of yeast Chi-Lis1 (PDB:8DZZ)⁴¹, showing two tail-truncated yeast dynein motor domains (grey) bound to two Lis1 dimers (colored, Chi-Lis1 1:2). b, Residues of MD-A that interact with Lis1_ring are located in AAA6-Lis1_ring and AAA5-Lis1_ring regions and highlighted with dashed rectangle. Representative residues of MD-A involved in the canonical Lis1_ring binding sites are located in Lis1_ring-AAA3, Lis1_ring-AAA4, Lis1_ring-AAA5 and Lis1_ring-stalk region. Interactions between MD-A and MD-B are in stalk-stalk region. Residues are displayed in sphere mode and are colored according to the subdomains in Fig. 4. c, Superimposition of the human Phi^L-Lis1 and yeast Chi-Lis1 structures, showing that Chi-Lis1 adopts a more expanded conformation, with larger grooves on both the front and back sides compared to the more compact Phi^L-Lis1 structure. Lis1 is hidden for clarity. Vectors represent interatomic distances of pairwise Cα atoms between the Phi^L-Lis1 and Chi-Lis1 structures.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41589-025-01981-6.

Data availability

Cryo-EM density maps and models were deposited to the EM Data Bank and PDB under the following accession numbers: dynein and Lis1 condition, PDB 9E12 and EMD-47381 for full-length Phi, PDB 9E10 and EMD-47379 for the motor domains of Phi, PDB 9E13 and EMD-47382 for full-length Phi^L–Lis1 and PDB 9E11 and EMD-47380 for motor domains of Phi^L–Lis1; in the dynein, Lis1 and Nde1 condition, PDB 9E14 and EMD-47383 for full-length Phi^L–Lis1 and PDB 9E0Z and EMD-47378 for motor domains of Phi^L–Lis1. Source data are provided with this paper.