Structural characterization of the RH1-LZI tandem of JIP3/4 highlights RH1 domains as a cytoskeletal motor-binding motif

JIP3 and JIP4 (JNK-interacting proteins 3 and 4) are adaptors for cargo recruitment by dynein/dynactin and kinesin1 motors. Both are dimers that are stabilised by two sections of leucine zipper coiled coils. The N-terminal Leucine Zipper I (LZI) belongs to a section that binds dynein-DLIC and kinesin1-KHC, whilst the medial Leucine Zipper II (LZII) binds dynactin-p150glued and kinesin1-KLC. Structural data is available for the LZII, but the LZI section is still uncharacterized. Here we characterize the N-terminal part of JIP3/4 which consists of an RH1 (RILP homology 1) domain followed by the LZI coiled coil using bioinformatical, biophysical and structural approaches. The RH1-LZI tandem of JIP3 associates as a high affinity homodimer exhibiting elongated alpha-helical fold. 3D homology modelling of the RH1-LZI tandem reveals that the kinesin1-KHC binding site mainly overlaps with the RH1 domain. A sequence comparison search indicates that only one other protein family has RH1 domains similar to those of JIP3/4, the RILP (Rab-interacting lysosomal protein) family which consists of adaptor proteins linking Rab GTPases to cytoskeletal motors. RILPL2 is recruited through its RH1 domain by the myosin 5a motor. Here, we showed that the RH1 domain of JIP3 also interacts with myosin 5 A in vitro, highlighting JIP3/4 as possible myosin 5a adaptors. Finally, we propose that JIP3/4 and RILP family members define a unique RH1/RH2-architecture adaptor superfamily linking cytoskeletal motors and Rab GTPases.

To search for remote homologs of the RH1 domain of the human JIP3 protein in the PDB 30 and Pfam 31 databases, as well as in the human proteome (Supplementary Table S1), we used the HHpred webserver 32,33 . In the PDB database, only mouse RILPL2 (PDB code 4KP3_D) was identified as a high confidence match (HHsearch probability > 95%). Despite the low sequence identity (around 20%), the high HHsearch probability (99.7%) as well as the consistent matching of secondary structures between the two proteins strongly support the structural homology. High confidence matches in the human proteome include only isoforms of the JIP3 and JIP4 proteins, as well as the RILP, RILPL1 (RILP-like 1) and RILPL2 (RILP-like 2) proteins: all display HHsearch probabilities >99%. Other matches are less confident (HHsearch probability <94%) and often align to shorter portions of the domain. Proteins matched for model species in InterPro family IPR019143 (corresponding to PF09744) 34 also show the same homologs. An HHpred search starting from the sequence of the RH1 domain of RILPL2 (PDB code 4KP3_D) led to the identification of the same homologs. This reveals that RH1 domains are found only in JIP3/4 and in RILP family proteins.
Overall sequence analysis of the RH1 domain between JIP3/4 and RILP family proteins reveals that profile-profile alignment (HHsearch) identified strong homology relationships among these proteins (HHsearch probabilities >99%), while classical sequence-sequence alignment analysis showed low sequence identities. Indeed, RH1 domain sequence identities among RILP family proteins are low (35-40% identity) and even lower between JIP3/4 and RILP family proteins (20-35% identity). Interestingly, the RH1 domain, like the PH domain for instance 35 , share structural homology, whereas they exhibit low sequence homology. Such low sequence homology constitutes a determinant for versatility in partner recognition.
Biophysical characterization of the N-terminal part of JIP3. In order to characterize the N-terminal part of JIP3 and JIP4, we conceived different fragments of both isoforms (Fig. 2a, Table 1 and Supplementary  Fig. S3a and Table S2). The longest fragment consists of the RH1 domain followed by the LZI coiled coil without the extreme N-terminus (RH1-LZI fragment), while two shorter fragments consist of the LZI coiled coil alone (LZI fragment) or the RH1 domain alone (RH1 fragment). RH1-LZI and LZI fragments of both JIP3 and JIP4 were soluble ( Supplementary Fig. S3b), while RH1 fragments were insoluble, revealing that in the absence of the LZI coiled coil, the RH1 domain is unstable.
Different biophysical approaches were used to characterize, in solution, soluble JIP3/4 fragments. We determined the oligomeric state of JIP3 and JIP4 fragments, using size exclusion chromatography coupled with Multiple Angle-Light Scattering (SEC-MALS). For JIP3, RH1-LZI and LZI fragments eluted in single peaks that contain species with a molecular mass of 39.2 ± 0.4 and 27.2 ± 0.5 kDa, respectively (Table 1 and Fig. 2b), indicating a dimeric form in solution for both fragments. Also, QELS measurements determine a hydrodynamic radius (Rh) of 4.1 ± 0.3 and 3.5 ± 0.2 nm, for RH1-LZI and LZI fragments respectively (Table 1). These Rh values indicate that the RH1-LZI fragment is more elongated than the LZI. For JIP4, SEC-MALS experiments showed that both RH1-LZI and LZI fragments elute in single, but broad peaks that contain species with a molecular mass of 57.8 ± 0.3 and 41.2 ± 0.4 kDa, respectively (Supplementary Table S2 and Fig. S3c). Despite these Mw values, which correspond to almost 3 times the mass of the monomer, the shape of the peak and the fact that the equivalent fragments of JIP3 have a dimeric state, suggest that there is a fast-reversible equilibrium between a dimeric and a higher oligomeric state in solution for both fragments of JIP4. Due to this oligomeric state heterogeneity, we subsequently focused only on JIP3 fragments characterization. Circular Dichroism (CD) experiments showed that RH1-LZI and LZI fragments of JIP3 are mainly alpha-helical with 81.1% and 74.4% helix content, respectively (Table 1). These helix content percentages correspond to 135 and 87 residues, respectively for the RH1-LZI and the LZI fragments which are consistent with the consensus secondary structure prediction (132 and 94 residues, respectively; Supplementary Fig. S2). Further, the ratio of ellipticities between 222 and 208 nm for RH1-LZI and LZI fragments is greater than 1 (Table 1) which is indicative of the presence of interacting α-helices 36 . Differential Scanning Calorimetry (DSC) experiments gave a melting temperature (Tm) of 49.2 °C and 35.4 °C for RH1-LZI and LZI fragments of JIP3, respectively (Table 1). Furthermore, while RH1 domain alone is unstable, DSC data reveals that its presence stabilizes the LZI coiled coil since a 14 °C difference is observed between the RH1-LZI and the LZI fragments. Thus, RH1-LZI fragment of JIP3, as well as the LZI alone, are stable dimers and adopt all α-helical elongated shapes.
Sedimentation velocity (SV) analytical ultracentrifugation coupled with fluorescence detection (FDS) were used to determine the stability of the self-association of the LZI coiled coil in presence or absence of the RH1 domain. SV-FDS experiments confirmed that the JIP3-RH1-LZI fragment is a stable dimer in solution with a sedimentation coefficient of 2.3 S and exhibits an elongated shape with a f/f0 of 1.9 (Fig. 2c), consistent with the f/f0 calculated from SEC-MALS data ( Table 1). Titration experiment then indicates that at low concentration (1.1 nM), the RH1-LZI fragment is still a stable dimer in solution with a sedimentation coefficient of 2.3 S (Fig. 2c). Therefore, we conclude that the Kd of the RH1-LZI dimer is sub-nanomolar. The JIP3-LZI fragment is also a dimer in solution with a sedimentation coefficient of 1.9 S and exhibits an elongated shape with a f/f0 of 1.8, consistent with the f/f0 calculated from SEC-MALS data ( Fig. 2d and Table 1). However, titration experiment indicates that the sedimentation coefficient decreases and settles down to 1.7 S at the lowest concentrations.
Similarly, the f/f0 decreases to 1.4 which no longer corresponds to an elongated shape. The Kd of the JIP3-LZI dimer is determined to 110 nM with a confidence interval from 65 nM to 185 nM (Fig. 2d). This data therefore indicates that the RH1-LZI fragment forms a more stable dimer than LZI does which is consistent with the Tm variation observed between these two fragments using DSC (Table 1 and Fig. 2d). Altogether, these biophysical characterizations reveal that the RH1 domain and the LZI coiled coil form a structural tandem since (i) in absence of the LZI coiled coil, the RH1 domain is unstable and (ii) in absence of the RH1 domain, the LZI coiled coil is less stable. for the LZI fragment the resulting sw isotherms (inset) from integration of c(s) are shown. In the isotherm plots, solid circles are the sw data from the dilution series, and the solid line is the best-fit isotherm. www.nature.com/scientificreports www.nature.com/scientificreports/ Homology modelling of the N-terminal part of JIP3. Next, we sought to obtain 3D information on the N-terminal part of JIP3/4. Because our crystallization attempts have been unsuccessful so far, we computed a model of the RH1-LZI fragment of human JIP3. On the one hand, the RH1 domain (residues ) of JIP3 was modelled based on the crystal structure of the RH1 domain of RILPL2 (PDB code 4KP3 37 ;), giving a very high confidence score (99.8%) using the Phyre2 web server 38 . The RH1 domain is a homodimer mediated by a four-helix bundle constituted by two anti-parallel helices from each protomer, with a length of 40 Å ( Fig. 3a; helices α1 and α2 consist of residues [26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43] and , respectively). Hydrophobic residues are found at the dimer interface: (i) Val24 from the linker between the extreme N-terminus and the α1 helix, (ii) Val29, Leu32, Ile36, Phe40, Leu43, Ile44, Tyr47 from α1 helix and (iii) Val51, Val52, Leu55, Met56, Val59, Val62, Leu63, Leu66, Val69 and Leu70 from α2 helix (Fig. 3b). Sequence comparison between our model of the RH1 domain of JIP3 and the crystal structure of that of RILPL2 shows that most hydrophobic residues at the interface of the dimer are conserved (Fig. 3c). Due to sequence differences between JIP3 and RILPL2, the modelling of the α1-α2 loop is unclear. In RILPL2, the loop is short consisting of 3 residues, while in JIP3/4 this segment is longer with 3 extra residues (Fig. 3c). Here, we modelled the α1-α2 loop of JIP3 with 6 residues, but we cannot exclude that some of them contribute to extend α1 and/or α2 helices, as suggested by secondary structure predictions in which differences are observed at this position among the various prediction methods ( Supplementary Fig. S2). In addition, at the beginning of the α2 helix of JIP3/4, a proline residue (well known as helix "breakers") is found (Pro57 in JIP3), whereas it is absent in RILPL2. Thus, this proline residue might either induce a kink into the α2 helix or shorten it. To resolve these two issues, the experimental 3D structure of the RH1 domain of JIP3/4 will be required. On the other hand, the LZI coiled coil (residues ) of JIP3 was modelled as a parallel dimeric coiled coil using CCbuilder 2.0 39 . The LZI consists of a straight coiled coil of 150 Å-length ( Fig. 3a) since no glycine or proline residues are observed in the sequence (Fig. 1b). It should be noted that while coiled coil prediction indicates the end of the LZI at residue 177, secondary structure predictions indicate that residues beyond 177 also adopt α-helical structure ( Supplementary Fig. S2). Therefore, residues 178 to 185 were modelled as a continuity of the coiled coil, but we cannot exclude that these residues exhibit a different α-helical structure. Finally, to model the RH1 domain relative to the LZI coiled coil, we took advantage of the known 3D structure of the RH1-cc tandem of RILPL2 (PDB code 4KP3 37 ). The sequences corresponding to the hinge between the RH1 domain and the LZI coiled coil is longer than that found in RILPL2 (Fig. 3c) and exhibits no glycine/proline residues. Secondary structure predictions of both JIP3 and JIP4 indicate that these residues adopt an α-helical structure ( Supplementary  Fig. S2). Thus, we anticipate that the hinge between the RH1 domain and the LZI coiled coil is structured as a α-helix and consequently that the RH1 domain is orientated straight in relation to the main axis of the LZI coiled coil. This model of the RH1-LZI tandem of JIP3 reaches a total length of 210 Å for a thickness of 20 Å (Fig. 3a). The predicted Rh and Cs of this model, calculated by the Hydropro program 40 , are 4.4 nm and 2.2 S, respectively, which are like those measured using SEC-MALS (Table 1) and SV-FDS (Fig. 2c). Thus, our structural model of the RH1-LZI tandem is concordant with our biophysical data.
Conformational properties of the RH1-LZI tandem of JIP3 from SAXS studies. To further assess the model of the N-terminal part of JIP3, we collected small angle X-ray scattering (SAXS) data on the RH1-LZI fragment, as well as on the LZI fragment alone for comparison ( Supplementary Fig. S4a). For the RH1-LZI fragment, the Guinier fit from q~0.0085 to 0.018 gives a Rg value of 63.4 ± 2.9 Å without indication of aggregation (Supplementary Fig. S4b and Table S3). The molecular mass estimated from SAXS data with the program SAXSMoW 41 is 42.3 kDa, which corresponds to a dimeric state of the RH1-LZI fragment of JIP3, as previously determined using SEC-MALS (Supplementary Table S3). The dimensionless Krakty plot 42 shows a bell-shaped curve indicating that the RH1-LZI fragment is a folded molecule, with a maximum of 4.1 at 7.4 qRg ( Supplementary Fig. S4c). For the LZI fragment, the maximum is 3.4 at 6.8 qRg (Supplementary Fig. S4c) indicating that the LZ1 fragment is shorter than the RH1-LZI fragment. For comparison, the maximum of a fully folded globular protein with one domain is close to 1.1 at 1.75 qRg. Thus, the shift from these values for the LZI and the RH1-LZI fragments of JIP3 is an indication of an elongated shape. Interestingly, the course of the dimensionless Krakty plots at high q indicates some flexibility of the RH1-LZI fragment ( Supplementary  Fig. S4c). The pair distribution function P(r) exhibits a typical shape of a long rod with a maximum around 17 Å and a Dmax of 231 Å ( Supplementary Fig. S4d), which is concordant with the thickness and the elongated shape of our model. For comparison, the pair distribution function P(r) of the LZI alone also exhibits a shape typical of a long rod with a maximum around 17 Å, but a Dmax of 198 Å ( Supplementary Fig. S4d). Lastly, the Rg determined from the distance distribution function is 66.1 ± 0.4 Å for the RH1-LZI fragment which is close to those calculated from the Guinier fit (Supplementary Table S3). For comparison, the predicted Rg for the RH1-LZI model of JIP3, calculated by the Hydropro program 40 is 64 Å which is quite similar to the experimental Rg determined using SAXS. Thus, SAXS analysis of the RH1-LZI fragment is concordant with our previous biophysical characterization. The model of the RH1-LZ1 tandem of JIP3 fits the SAXS data with a χ 2 of 2.108 ( Supplementary Fig. S4c). To further refine the model against the experimental data, we used the DADIMODO program which allows for the release of structural restraints at specific defined regions 43,44 . We released structural restraints at regions that we expected to exhibit some flexibility: (i) the N-terminus of the RH1 domain, (ii) the C-terminus of the LZI coiled coil and (iii) the hinge between the RH1 domain and the LZI coiled coil. Five independent refined models provide an improved fit against SAXS data with χ 2 ranging from 1.271 to 1.312 (Fig. 4a). The most important characteristic of the refined models is a movement of the RH1 domain relative to the LZI coiled coil indicating some flexibility at the hinge between the RH1 domain and the LZI coiled coil (Fig. 4b). In order to estimate this flexibility degree, the Rg distribution variability of the JIP3-RH1-LZ1 model was calculated using EOM 2.1 45 . The Rg distribution of 5 independent runs of EOM indicates a tendency to extended conformation (run 1 to 5, Rg = 64.4 Å) in solution compared to aleatory conformations (pools) (Fig. 4c), which agrees with the DADIMODO models (Rg = 64.6 to 65 Å). Thus, overall SAXS data indicates that the RH1-LZI tandem is mostly in an extended conformation in solution, and the hinge between the RH1 domain and the LZI coiled coil could be bent to some degree.

Does myo5a-GTD interact with the RH1 domain of JIP3/4?. RILPL2 which shares a RH1
domain-coiled coil (RH1-cc) tandem with JIP3/4, is recruited through its RH1 domain by the myosin 5a (myo5a) actin-based motor 46 . Therefore, we wondered if myo5a also interacts with JIP3/4 through its RH1 domain. The crystal structure of the RH1-cc tandem of RILPL2 bound to the globular tail domain (GTD) of myo5a (PDB code 4KP3 37 ) shows that myo5a-GTD makes interactions with both α1 and α2 helices from one protomer of the RH1 domain of RILPL2; the coiled coil (cc) of the tandem is not involved in the interaction (Fig. 6a). The interface can be described by two sets of interactions: (i) a hydrophobic patch of RILPL2-RH1, mainly constituted by residues (2019) 9:16036 | https://doi.org/10.1038/s41598-019-52537-3 www.nature.com/scientificreports www.nature.com/scientificreports/ Val31, Ser35, Gly39, Leu42 from α1 helix and Val59 and Met62 from α2 helix (Fig. 6b,c, black arrows) mediates interactions with Val1498 and Leu1502 from myo5a-GTD and (ii) the Phe56 from the α2 helix of RILPL2-RH1 (Fig. 6b,c, red arrow) inserts its bulky side chain into a hydrophobic pocket formed at the myo5a-GTD surface 37 . Structure superposition of the RH1 domain of JIP3 with that of RILPL2 reveals that the first set of interactions might be conserved between JIP3-RH1 and myo5a-GTD (Fig. 6b,c). Val31 and Val59 residues of RILPL2 are conserved in JIP3 (Val29 and Val60, respectively), Ser35, Leu42 and Met62 are similar (Ala33, Phe40 and Leu63, respectively in JIP3), but Gly39 is different (Tyr37 in JIP3). However, a reverse substitution (Tyr36 in RILPL2 to Gly34 in JIP3) is found one helix turn before which might complement this sequence difference. Concerning the second set of interactions, the critical Phe56 in RILPL2 is replaced by a proline residue in JIP3/4-RH1 (Fig. 6b,c). Interestingly, this position is also a proline residue in RILPL1 and RILP proteins (Fig. 6c). Wei and al. showed that substitution of the RILPL2-Phe56 to a proline residue disrupts the interaction with myo5a-GTD, while substitution of RILP-Pro55 in phenylalanine renders the RILP-RH1 mutant capable of Myo5-GTD binding, even with a higher binding affinity than the wild-type RILPL2-RH1 37 . Consequently, since this position is occupied by a proline residue, we assume that the JIP3/4-RH1 domain cannot interact with myo5a-GTD. However, due to the low sequence homology (23% sequence identity) between RH1 domains of RILPL2 and JIP3, we cannot exclude that complementary sequence substitutions may nevertheless allow myo5a-GTD interaction.
To our knowledge, no data has been reported on a direct interaction between JIP3/4 and myo5a, but both proteins are found on the same vesicles 47 making their interaction possible in a cellular context. To experimentally assess the ability of the RH1 domain of JIP3 to interact with myo5a-GTD, we performed in vitro binding experiments. Sedimentation velocity analytical ultracentrifugation (AUC-SV) experiments indicated that a mixture of myo5a-GTD over an excess of JIP3-RH1-LZI shows two species: one with a sedimentation coefficient (Cs) of 2.4 S corresponding to the excess of JIP3-RH1-LZI and a second one with a higher Cs of 3.9 S (Fig. 7a), while a reverse mixture also shows two species: one with a Cs of 3.0 S corresponding to the excess of myo5a-GTD and a second one with the same higher Cs at 3.9 S (Fig. 7a). We conclude that the species with a Cs = 3.9 S corresponds to a myo5a-JIP3 complex. To further characterize this interaction, microscale thermophoresis (MST) experiments were performed to confirm the interaction and further delineate the binding interface. In order to compare with the binding affinity measured between RILPL2-RH1-cc and myo5a-GTD 37 , we generated a shorter fragment of JIP3 which is equivalent to that of RILPL2. The RH1-lz3 fragment of JIP3 consists of the RH1 domain followed by the three first heptad repeats of the LZI coiled coil (residues , Supplementary Fig. S5a). SEC-MALS experiment showed that this shorter fragment is dimeric ( Supplementary Fig. S5b) revealing that truncation of the rest of the LZI coiled coils did not affect the overall structure of the fragment. MST data showed that myo5a-GTD binds to the labelled JIP3-RH1-lz3 with an affinity of 2.2 ± 0.5 µM (Fig. 7b) which is weaker than that of RILPL2-RH1-cc (Kd = 0.3 ± 0.05 µM, determined using ITC 37 ;). Although there is often a difference in Kd between different biophysical approaches, a 7-fold difference indicates a significant variation in binding affinity. Thus, our binding data reveals that myo5a-GTD binds to the RH1-lz3 region of JIP3, in vitro. However, differences in binding affinity and sequence, such as the presence of a proline residue in JIP3/4 instead of the critical phenylalanine residue found in RILPL2 37 , suggest that the mode of recognition varies in detail between RILPL2 and JIP3/4. Further experiments are now needed to confirm that JIP3/4 is a binding partner of myo5a in a cellular context.

Discussion
Our study reveals that the N-terminal part of JIP3/4 is not only the binding site for the kinesin1-KHC 7 and Dynein-DLIC 25 which are microtubule-based motors with opposite directionality, but also for myosin 5a, an actin-based motor. This highlights that the RH1-LZI tandem, like the LZII coiled coil located in the median part of the protein, are important and versatile protein-protein interaction modules for multiple cytoskeletal motors. Thus, JIP3/4 interact with kinesin1-KHC and kinesin1-KLC, independently at both the RH1-LZI and LZII binding sites, respectively 6,7,9 and with dynein/dynactin complex (via DLIC and p150 glued ) also, respectively at both binding sites 8,9,25 , while JIP3 interacts with myosin 5a at the RH1-LZI binding site in vitro (this study). To our knowledge, no adaptor has yet been described as interacting directly with both microtubule-and actin-based motors. Interestingly, kinesin1-KLC and dynactin-p150 glued compete for their interaction on the LZII binding site of JIP3, as well kinesin1-KHC and myosin 5a both interact with the RH1 domain and one to three first heptads of the coiled coil that might be the basis for further competition. Such motor competition for JIP3/4 interaction, and thus JIP3/4 recruitment are potentially determinants for cargo transport regulation. Furthermore, JIP3/4 are not simply structural adaptors required for site of attachment, but also allosteric regulators for motor activation, like in the case of kinesin1-KHC 7 . Altogether, these studies show that JIP3/4 are critical regulatory "hubs" allowing the coordination of cargo transport between the microtubule and actin cytoskeletons. www.nature.com/scientificreports www.nature.com/scientificreports/ In recent years, numerous structural data from X-ray crystallography or CryoEM reported cytoskeletal motors, such as kinesin1, dynein/dynactin and myosin5 motors [48][49][50] bound to different type of adaptors [48][49][50] . Two distinct modes of adaptor recognition shared by multiple cytoskeletal motors have emerged and were recently summarized 48 . Unstructured peptide motifs (10 to 25 residues) represent the first structural motor-binding motif, like the globular tail binding motifs (GTBM) found in various myosin5 adaptors (for example Melanophilin (MLPH), Spir-1/2 or Granuphilin) or W-acidic and Y-acidic motifs found in various kinesin1-KLC adaptors (for instance in SKIP and JIP1, respectively). Dimeric coiled coils represent the second structural motor-binding motif, like the LZII coiled coil of JIP3/4 that is recognized by kinesin1-KLC or coiled coils found in various dynein/dynactin adaptors (for instance in HOOK3, BICD2 and JIP3). Here, we highlight a third structural motor-binding motif which is also recognized by multiple cytoskeletal motors, namely the RH1 domain. Several studies reported such motor-adaptor modes of recognition: (1) the RH1 domain of JIP3 was identified as the binding site for the heavy chain of kinesin1 (KHC) (this study 7 ); (2) the RH1 domain of RILPL2 is the binding site of the GTD domain of myosin 5a 37,46 ; (3) the 50 first residues of RILP, which encompasses its RH1 domain were proposed as critical for the interaction with the p150 glued subunit of dynactin complex [51][52][53] ; and (4) the RH1 domain of JIP3 can interact in vitro with the GTD domain of myosin 5a (this study). In addition, dynein-DLIC interacts with both the N-terminal half of RILP encompassing the RH1-cc tandem 54,55 and the N-terminal part (aa. 1-240) of UNC-16/JIP3 encompassing the RH1-LZI tandem 25 . However, further experiments are required to determine if the RH1 domain of both RILP and UNC-16/JIP3 are critical for these interactions. Altogether, this identifies the RH1 domain as a versatile structural cytoskeletal motor-binding motif.
Finally, remote homology searches revealed that RH1 domains are found only in JIP3/4 and RILP families, defining one common characteristic of these motor adaptors. Interestingly, a second characteristic of these RH1 domain-containing proteins is the presence of a RH2 domain (RILP homology 2) (Fig. 8). The RH2 domains of JIP3 and RILPs were shown to recruit various Rab small GTPases 22,56 which are critical regulators of intracellular transport connecting membrane vesicle to cytoskeleton motors or adaptors 17 . Thus, JIP3/4 and RILP family members constitute a unique RH1/RH2-architecture adaptor superfamily that links various Rab GTPases to multiple cytoskeletal motors allowing specific membrane association to directional movement.

Materials and Methods
Sequence analysis. To perform primary sequence analysis, human JIP3 and JIP4, with accession numbers NP_055948.2 and NP_001123999.1 respectively, were used. Sequence alignment of human JIP3 and JIP4 was performed using ClustalW 57 . Coiled coil primary sequence characteristics were analysed using paircoil2 28 . The phylogenetic study was performed using full-length sequences of JIP3 and JIP4 homologs retrieved using 2 iterations of InterEvolAlign 58 : one iteration against the OMA 2011 'Entire genomes' database 59 , then another one against the RefSeq database 60 . Those full-length sequences were re-aligned using the MAFFT E-INS-i algorithm 61 . A phylogenetic tree was calculated using the SeaView graphical user interface 62 . First, a site selection was performed using Gblocks 63 with options for less stringent selection. Then, the tree was computed with the PhyML program (version 3.1) 64 using the LG substitution model (with optimized across-site rate variation using 4 substitution rate categories), empirical amino acid equilibrium frequencies and an optimized fraction of invariable sites. The final tree was visualized with Dendroscope 3 65 . The associated multiple sequence alignment was visualized with Jalview 66 .
The HHpred webserver 32,33 with 1 iteration of HHblits against the uniclust30_2018_08 for query profile generation and all other parameters set to their default values was used to search for homologs of the sequence of the JIP3_MOUSE N-terminal region (positions 1-98) in the PDB database (PDB_mmcif70_10_Apr), the Pfam database (Pfam-A_v32.0) and the human proteome (Homo_sapiens_04_Jul_2017). A search with the same parameters was also performed using the sequence of PDB 4KP3_D as a query, for comparison purposes.  Analytical ultracentrifugation. All sedimentation velocity experiments were performed on an analytical ultracentrifuge XLI (Beckman Coulter, Palo Alto, USA) with an An-50 Ti rotor at 20 °C. For the characterization of the self-association of the JIP3-Nter dimer, experiments were done with a fluorescence detection system (AVIV Biomedical, Lakewood, NJ). Two-channels 12 mm path-length Epon charcoal-filled centrepieces were used. JIP3 protein was labelled with Monolith NT.115 Protein Labelling Kit BLUE NHS (Nanotemper Technologies) following the manufactures' instructions. Titration experiments were performed with the RH1-LZI and LZI fragments at concentrations ranging from 460 nM to 1.1 nM and from 770 nM to 2.4 nM, respectively. For protein sample from 1.1 nM to 60 nM, only labelled protein was used; contrary to higher concentrated samples that corresponded to the mixture of labelled protein (60 nM) with unlabelled protein to reach the total concentration. 400 μL of JIP3 samples were centrifuged at 50,000 rpm (181,827 g). Sedimentation profiles were collected every 4 min and analysed as described [69][70][71] .
Homology modelling and structural analysis. We submitted the sequence of the N-terminal part (residues 1-187) of JIP3 to the web portal for protein modelling by threading using Phyre2 38 . With a confidence of 99.8%, Phyre2 identified the crystal structure of the RH1 domain of RILP-like protein 2 (RILPL2; PDB code 4KP3 37 ;), as the highest scoring template for the RH1 domain and the beginning of the LZI of JIP3. Phyre2 gave same results when the JIP4 N-terminal sequence is submitted. Although Phyre2 modelled the last part of the extreme N-terminus (residues 18-21), the linker (residues 72-77) and the first part of the LZI coiled coil (residues 78-98), we only conserved the RH1 domain of JIP3 (residues 22-76). Then, we modelled the LZI of JIP3 Small angle X-ray scattering (SAXS). SAXS measurements were done at the SWING beamline of the SOLEIL synchrotron at a working energy of 12.0 keV (λ = 1.033 nm) and detector to sample distance was 1.5 m. The samples were subjected to a size exclusion purification online with the HPLC system ( Supplementary Fig. S3) connected to a quartz capillary placed under vacuum cell using an Agilent Bio SEC. 3-300 column 76 . The samples were injected at 6 mg/ml with the elution buffer containing 20 mM Hepes pH 7.1, 300 mM NaCl, 0.5 mM TCEP. Data reduction, averaging of identical frames corresponding to the elution peak and buffer subtraction were performed with the SWING in-house software Foxtrot. The radius of gyration (Rg) and forward intensity at zero angle I(0) were derived by the Guinier approximation using the software PRIMUS 77 . The maximum dimension (Dmax) and the Rg were derived from the pair distribution function P(r), calculated with the software GNOM 78 . The molecular weight was calculated with SAXSMow2, extending the integration limit to a q m of 0.5 Å −1 41 .
In order to have a full model corresponding to the fragment produced for SAXS data comparison, residues 181-187, as well as Gly-Ser-His residues remaining after the thrombin cleavage were added manually to the RH1-LZ1 tandem model. Crysol was used to fit the theoretical scattering of this model to the experimental data 79 . To refine the model against the SAXS data, we used the DADIMODO server that allows to kept rigid domains, while other regions can be defined as flexible 43,44 . The RH1 domain (residues 26-66) and the LZI coiled coil (residues 78-176) were kept as rigid bodies, while the N-terminus (residues Gly-Ser-His- [22][23][24][25]), the linker and few residues before (residues [67][68][69][70][71][72][73][74][75][76][77], and the C terminus (residues 177-187) were allowed to move. Five independent refinements were done. The web interface of EOM 2.1 45 was used to generate 10,000 structures using the model of the RH1-LZ1 tandem of JIP3 as template. Only the flexible linker (residues 67-77) were considered as a flexible region. Five independent runs of 100 cycles were run to estimate the variability in the distribution of the Rg values.
Microscale Thermophoresis. JIP3-Nter fragments and mutant were labeled using the Protein Labeling Kit Blue-NHS (NanoTemper Technologies). The labeling reaction was performed according to the manufacturer's instructions in the supplied labeling buffer applying a concentration of 20 μM protein (molar dye:protein ratio ≈ 3:1) at room temperature for 30 min. The labeled proteins were stocked in MST buffer (50 mM TrisHCl pH 7.8, 150 mM NaCl, 10 mM MgCl 2 , 0.05% Tween20). For the binding assays, the labeled JIP3-Nter fragments were adjusted to 25 nM with the assay buffer (20 mM hepes pH 7.5, 150 mM NaCl, 2 mM MgCl 2 , 0.5 mM TCEP, 0.05% tween). Myo5a-GTD was dialyzed in the assay buffer and a series of 16 1:1 dilution was prepared. For the measurement, each myo5a-GTD dilution was mixed with one volume of labeled JIP3-Nter fragments and mutant. The samples were loaded into Monolith NT.115 Capillaries MO-K022 (NanoTemper Technologies). MST was measured using a Monolith NT.115 instrument (NanoTemper Technologies) at 25 °C. Instrument parameters were adjusted to 20% LED power and medium MST power. Data of three independent replicates was analysed using the signal from an MST-on time at 4-5 s (MO.Affinity Analysis software version 2.3, NanoTemper Technologies). Accession codes. SAXS data has been deposited into the SASBDB database (http://www.sasbdb.org/) with entry codes SASBDB: SASDGV4 (JIP3-RH1-LZI) and SASDGW4 (JIP3-LZI).