Structural basis of outer dynein arm intraflagellar transport by the transport adaptor protein ODA16 and the intraflagellar transport protein IFT46

Motile cilia are found on unicellular organisms such as the green alga Chlamydomonas reinhardtii, on sperm cells, and on cells that line the trachea and fallopian tubes in mammals. The motility of cilia relies on a number of large protein complexes including the force-generating outer dynein arms (ODAs). The transport of ODAs into cilia has been previously shown to require the transport adaptor ODA16, as well as the intraflagellar transport (IFT) protein IFT46, but the molecular mechanism by which ODAs are recognized and transported into motile cilia is still unclear. Here, we determined the high-resolution crystal structure of C. reinhardtii ODA16 (CrODA16) and mapped the binding to IFT46 and ODAs. The CrODA16 structure revealed a small 80-residue N-terminal domain and a C-terminal 8-bladed β-propeller domain that are both required for the association with the N-terminal 147 residues of IFT46. The dissociation constant of the IFT46-ODA16 complex was 200 nm, demonstrating that CrODA16 associates with the IFT complex with an affinity comparable with that of the individual IFT subunits. Furthermore, we show, using ODAs extracted from the axonemes of C. reinhardtii, that the C-terminal β-propeller but not the N-terminal domain of CrODA16 is required for the interaction with ODAs. These data allowed us to present an architectural model for ODA16-mediated IFT of ODAs.

Cilia are organelles that protrude from the surface of eukaryotic cells to serve a number of functions in cell motility, sensory reception, and developmental signaling. Motile cilia are the cellular machines that propel unicellular organisms through aqueous media (1). They are also vital for human health because they drive the mucociliary flow to remove inhaled pathogens from the airways, move egg cells through the fallopian tubes to reach the uterus, and drive the left-right asymmetry during development, which is required for the correct positioning of inner organs (2). It is thus not surprising that mutations causing absence or defects in motile cilia result in diseases and disorders categorized as primary cilia dyskinesia (3,4).
Primary and motile cilia adopt the same slender hair-like appearance and share a microtubule-based axoneme surrounded by a ciliary membrane (5). They do, however, differ in the presence of a number of motility complexes found only in the axoneme of motile cilia. These include outer dynein arms (ODAs), 3 which are large ATPase complexes that, together with several inner dynein arms, power the bending of cilia required for cilium-mediated motility (6). ODAs are MDasized multisubunit complexes that preassemble in the cytoplasm before trafficking into the cilium (7). Four ODAs are found within the 96-nm repeat of the axoneme, where they attach stably to the A tubule of the microtubule doublets and transiently interact with the B-tubule to cause bending of the axoneme resulting in motility (8 -12).
Several high-resolution structures have recently shed light onto the molecular mechanisms of dynein motors (13)(14)(15). Less is known about the molecular basis of how ODAs are recognized and transported into the axonemes of motile cilia. This trafficking process is, however, known to require the adaptor protein ODA16, which was discovered through the characterization of a C. reinhardtii oda16 mutant displaying reduced beat frequency and altered swimming behavior because of a partial loss of axonemal ODAs (16). ODA16 is a predicted WD protein conserved only in organisms with motile cilia (WDR69 in human, 62% sequence identity to the C. reinhardtii protein (16)) and is also required for proper motile cilium function in zebrafish (17).
A more detailed picture of the molecular function of ODA16 emerged when it was discovered that ODA16 acts in ODA transport through an interaction with the intraflagellar transport (IFT) machinery (18,19). IFT is an evolutionarily conserved process (20,21) that moves proteinacious particles bidirectionally inside cilia to deliver axonemal building blocks such as tubulin to the tip of cilia (22)(23)(24)(25)(26). IFT relies on a 22-subunit IFT complex that organizes into a 16-subunit IFT-B and a 6-subunit IFT-A complex (27)(28)(29). The 16-subunit IFT-B com-This work was supported by Novo Nordisk Foundation Grant NNF15OC0014164 and European Research Council Grant 310343. The authors declare that they have no conflicts of interest with the contents of this article. The atomic coordinates and structure factors (code 5MZH) have been deposited in the Protein Data Bank (http://wwpdb.org/). This article contains supplemental Figs. S1-S3. 1 These authors contributed equally to this work. 2 To whom correspondence should be addressed. E-mail: el@mbg.au.dk. plex can be further subdivided into a 10-subunit IFT-B1 subcomplex and a 6-subunit IFT-B2 subcomplex (30 -32), both of which can be reconstituted and purified separately (31,33). IFT46 is an IFT-B1 subunit that is necessary for IFT complex assembly through an interaction of its C-terminal domain with IFT52, another IFT-B1 subcomplex component (33)(34)(35). IFT46 was shown to interact directly with ODA16 (18) and to be required for the transport of ODAs into cilia (19). Whereas the ift46 null mutant is unable to assemble flagella, likely because of a failure in IFT complex formation, a suppressor strain that alleviates this "no cilia" phenotype of ift46 was isolated and shown to have full-length cilia that specifically lack ODAs. This ift46 suppressor mutant likely expresses the C-terminal domain of IFT46 required for IFT complex formation but lacks the N-terminal region of IFT46, suggesting that the ODA16 interacting domain is found within this region of IFT46 (19).
Here we purified recombinantly produced ODA16 and ODA16-IFT46 complexes of the biflagellate green alga C. reinhardtii and dissected the molecular basis of the IFT46-ODA16-ODA interaction underlying the ciliary trafficking of ODAs. We determined the high-resolution crystal structure of ODA16 and mapped the interaction with ODAs and IFT46 using biochemical and biophysical methodology. The results show that ODAs associate with the C-terminal ODA16 ␤-propeller, whereas IFT46-interaction requires the small 80-residue N-terminal domain of ODA16.

ODA16 binds directly to the N-terminal 147 residues of IFT46
Previously published yeast two-hybrid and pulldown analysis revealed an interaction between ODA16 and IFT46 (18). Furthermore, analysis of an IFT46 suppressor strain suggested that the N-terminal region of IFT46 is responsible for the IFT of ODAs (19). To analyze the ODA16-IFT46 association using direct protein-protein interaction assays, we purified recombinant ODA16 (Fig. 1, a and b) and GST-IFT46 (all proteins used in this study are of the green alga C. reinhardtii) and carried out pulldown experiments, which revealed a strong ODA16-IFT46 interaction (Fig. 1c). The ODA16-IFT46 complex appears stoichiometric at 200 mM NaCl, but the complex gradually falls apart when the salt concentration is increased to 1 M (Fig. 1c), suggesting at least a partly hydrophilic mode of association. Size exclusion chromatography (SEC) of IFT46, ODA16, or an equimolar mixture of the two proteins in 200 mM NaCl demonstrated the formation of a stable complex (Fig. 1d). Whereas ODA16 elutes from SEC close to the position of a globular protein, IFT46 elutes at a significantly smaller volume, suggesting an elongated shape. This is consistent with the predicted modular multidomain organization, as well as with the presence of a long flexible region (Figs. 1d and 2a and supplemental Fig. S1). To quantify the affinity of the ODA16-IFT46 interaction, we titrated purified untagged ODA16 with untagged IFT46 in isothermal titration calorimetry (ITC) experiments. The result shown in Fig. 1e demonstrates that ODA16-IFT46 forms a stoichiometric complex (n ϭ 0.96) with a dissociation constant (K d ) of 217 nM in agreement with the stable complex formation observed in SEC. The affinity of ODA16 for IFT46 is thus similar to that observed between IFT subunits, suggesting that ODA16 could be an integral subunit of the IFT-B complex during cilium formation (Refs. 31, 33, and 34 and data not shown).
To map the domain(s) of IFT46 required for association with ODA16, we carried out binding studies with truncated versions of IFT46. The C-terminal domain of IFT46 associates with IFT52, is required for the formation of the IFT-B1 complex (33)(34)(35), and rescues the more severe ciliogenesis phenotype observed in the ift46 C. reinhardtii mutant but not the lack of axonemal ODAs (19). In agreement with these data, IFT46 147-C (IFT46C) did not co-elute with ODA16 in SEC (Fig. 2b). Consistently, IFT46C did not show any detectable interaction with ODA16 in ITC experiments, suggesting that the association, if any, is at least 2 orders of magnitude weaker than for full-length (fl) IFT46 (Fig. 2c). To probe the interaction between the N-terminal IFT46 1-147 region (IFT46N) and ODA16, the two proteins were incubated and applied to SEC, which demonstrated a stable ODA16-IFT46N complex (Fig. 2d). To assess the interaction quantitatively, ODA16 was titrated with IFT46N in ITC, revealing a stoichiometric complex (n ϭ 1.08) with a K d of 247 nM (Fig. 2e). IFT46 (fl) and IFT46N thus interact with ODA16 with similar affinities (difference in K d is smaller than 1 standard deviation; see Figs. 1e and 2e). This indicates that the IFT46N construct contains the entire ODA16-binding region. We were unable to carry out a more detailed mapping of the IFT46N construct, because shorter IFT46 fragments tested were either insoluble or not expressed (data not shown). We conclude that residues 1-147 of IFT46 interact with nanomolar affinity with the ODA cargo-adaptor ODA16.

Crystal structure of ODA16
All protein constructs and complexes shown in Figs. 1 and 2 were screened for crystallization, which resulted in crystal hits for ODA16 but not IFT46 (fl or N) or ODA16-IFT46 complexes. X-ray diffraction data at 2.4 Å resolution together with the molecular replacement model of an eight-bladed ␤-propeller structure from the ribosomal assembly factor 4 (RSA4; PDB code 4wjs) were used to determine the three-dimensional structure of ODA16 ( Fig. 3 and Table 1). Full-length ODA16 was crystallized, and the resulting map had visible electron density for both the smaller ϳ80 residue N-terminal domain (ODA16N) and the ␤-propeller but not for the most ϳ25 C-terminal residues that are presumably disordered (Figs. 1a and 3a and supplemental Fig. S2). The N-terminal ODA16 domain folds into a three-stranded anti-parallel ␤-sheet followed by two ␣-helices that connect to the C-terminal ␤-propeller domain via a seven-residue linker region (Fig. 3a). This linker region together with residues from two loop regions (amino acids 13-15 and 33-36) of ODA16N interact with residues from loop regions of the N-terminal face of the ␤-propeller to position the two domains relative to each other via a rather small interface (Fig. 3b). ODA16N shows structural similarity to domains found in otherwise functionally unrelated proteins such as the Ser/Thr kinase WNK1 (PDB code 2lru) and the ribosomal proteins UL2 and UL12 (PDB codes 4w20 and 1vx1) (36,37). The protein structure in the PDB most similar to the ODA16 ␤-propeller is the ribosome biogenesis factor RSA4 (38), which interestingly also has a small but structurally divergent domain positioned on the N-terminal face of the ␤-propeller (Fig. 3c).
The relative positions of the two ODA16 domains create a pronounced interdomain cleft with dimensions of ϳ20 ϫ 20 ϫ 10 Å 3 (Fig. 4, a and b). This cleft is relatively well conserved, lined by a mixture of hydrophobic and hydrophilic residues and constitutes a potential protein-protein interaction site. A second highly conserved surface patch is found on the C-terminal face of the ␤-propeller (Fig. 4c). This conserved patch is mostly acidic in nature and overlaps with the protein interaction interface of structurally similar ␤-propellers found in proteins such as the Fbw7 subunit of a cyclin E ubiquitin ligase complex (Fig.  4, c and d) (39). Both the interdomain cleft and the C-terminal ␤-propeller surface of ODA16 constitute likely interaction interfaces for IFT46 and ODAs.

Both N-and C-terminal domains of ODA16 are required for ODA16-IFT46 complex formation
To assess whether IFT46 binds to the conserved acidic patch located on the C-terminal side of the ODA16 ␤-propeller, GST-IFT46 was used to pull down fl ODA16 or any of three different ODA16 constructs lacking either the N-terminal domain (residues 80 -445, ODA16⌬N), the C-terminal unstructured tail (residues 1-423, ODA16⌬C), or both (residues 80 -423, ODA16⌬N⌬C) (see schematics in Fig. 5a). The results of this experiment revealed that the N-terminal domain of ODA16 but not the C-terminal unstructured tail is required for interaction with IFT46 (Fig. 5b). Given that ODA16N is required for the interaction with IFT46, it appears unlikely that the C-terminal conserved patch of the ODA16 ␤-propeller, located more than 20 Å away from ODA16N, constitutes the interaction interface for IFT46. To test whether ODA16N is sufficient for the inter- action with IFT46, GST-tagged IFT46 was used in pulldown assays with ODA16N, which showed no detectable interaction (Fig. 5c). Both ODA16N and ODA16C domains are thus required for the association with IFT46. To test whether ODA16N and ODA16⌬⌵ have to be covalently attached to mediate IFT46 interaction, GST-IFT46 was incubated with a mixture of purified ODA16N and ODA16⌬⌵ in trans, which did also not result in any detectable interaction (Fig. 5c). Given that only few interdomain interactions dictate the relative position of ODA16N and the ODA16 ␤-propeller (Fig. 3b), the requirement of covalently attached ODA16N and ODA16⌬⌵ for the association with IFT46 suggests that the interdomain cleft may constitute the IFT46 binding site.
To gain structural insights into the ODA16-IFT46 complex, SEC combined with small-angle X-ray scattering (SEC-SAXS) was used to obtain solution scattering data for ODA16 alone as well as in complex with IFT46N (Table 2 and Fig. 6, a and b). The pair distance distribution functions in Fig. 6c show that the maximum distance between scattering points (D max ) increases from ϳ8 to ϳ13 nm when ODA16 associates with IFT46N, suggesting a complex with an elongated shape. Envelopes calculated from the SAXS data revealed that the ODA16-IFT46N complex has significant extra density extending along the axis of the N-and C-terminal ODA16 domains (Fig. 6, d and e). This result indicates that IFT46N binds close to the N-terminal ODA16 domain consistent with the biochemical data shown in Fig. 5. Based on the biochemical and structural data, we conclude that the N-terminal region of IFT46 associates with the N-terminal region of the ODA16 molecule in an interaction where both the N-and C-terminal domains of ODA16 are required for ODA16-IFT46 complex formation.

ODAs bind to the ␤-propeller of ODA16
To investigate how ODAs interact with ODA16, we first wanted to confirm that we can detect such an interaction in pulldown assays. Flagella were isolated from C. reinhardtii cultures and subsequently fractionated to obtain an axonemal extract enriched in ODAs. When a GST-IFT46-ODA16 complex was used as bait, several outer dynein-arm components were efficiently enriched, as confirmed by mass-spectrometric analysis (nano-LC-MS/MS), as well as immunoblotting with anti-IC2 antibodies (supplemental Fig. S3). Furthermore, subunits of the inner dynein arms, IFT particles, IFT motors, radial spokes, docking complex, tubulins, and flagellar-associated proteins were identified in the MS experiments (supplemental Fig. S3), indicating that the GST-tagged IFT46 construct assembles into larger IFT complexes together with factors present in the extract. To investigate the interaction between ODA16 and ODAs more directly, as well as to test for the contributions of ODA16 domains to this interaction, we used various His-tagged versions of ODA16 for pulldowns from axonemal flagellar extract, followed by immunoblotting for the IC2 subunit of ODAs. Although both the full-length and the isolated ␤-propeller constructs co-precipitated significant amounts of IC2, the N-terminal domain alone did not (Fig. 7a). This result shows that the ODA binding site resides in the ODA16 subunit and is not formed by association of ODA16 with IFT46, a possibility we could not rule out in the pulldown with GST-IFT46-ODA16 as bait (supplemental Fig. S3). Fur-thermore, whereas both the N terminus and the ␤-propeller of ODA16 are required for the interaction with IFT46 in the IFT complex, only the ␤-propeller is necessary for binding to an as-of-yet-unidentified component of ODAs. Given that this domain contains highly conserved surface residues, which in structurally related proteins are responsible for direct proteinprotein interactions (Fig. 4, c and d), it is tempting to speculate that ODA16-ODA binding occurs via a similar mechanism. Future experiments should be aimed at identifying the direct ODA16 interaction partner and determining the structural basis for their association.

Discussion
Here we provide structural insights into the IFT of outer dynein arms during flagellar assembly by dissecting interactions within the IFT46-ODA16-ODA protein complex. IFT46 is a core member of the IFT-B1 complex and is required for IFT complex formation (19,33,40) through the association of its C-terminal domain with IFT52 (Fig. 7b). The region containing the N-terminal 165 residues of IFT46 is mostly predicted to be disordered (41) (Figs. 2a and supplemental Fig. S1), is not required for IFT complex formation (19,33), and likely protrudes from the IFT complex. We now show that the 147 most N-terminal residues of IFT46 are sufficient for the high affinity interaction with ODA16 (K d of ϳ200 nM). The stability of the ODA16-IFT46 complex is similar to what we previously observed for purified IFT subcomplex (31,33,34), 4 suggesting that ODA16 could be a stoichiometric subunit of the IFT complex during flagellar assembly. Interestingly, both the small N-terminal domain and the C-terminal ␤-propeller, but not the very C-terminal 22 residues of ODA16 (only found in the C. reinhardtii ODA16 sequence), are required for the association with IFT46. The crystal structure of ODA16, together with SAXS reconstructions of the ODA16-IFT46N complex and biochemical data, suggests that the N-terminal region of IFT46 may associate with the pronounced cleft observed between the N-and C-terminal domains of ODA16 (summarized in Fig. 7b). The SAXS reconstruction of the ODA16-IFT46N complex suggests that it adopts an elongated conformation (Fig. 6c). Interestingly, although IFT46 is conserved in all organisms that assemble cilia via IFT, the N-terminal region of IFT46 is only well conserved in organisms with motile cilia (42), consistent with a function in ODA16-mediated IFT of ODAs. Whereas IFT46 binds to the N-terminal face of the ODA16 molecule, we show that ODAs only require the C-terminal ␤-propeller for association with ODA16 (Fig. 7a). We observe no direct association between IFT46 and ODAs (data not shown), suggesting that ODA16 is a true cargo adaptor sandwiched between IFT46 and the MDa-sized ODA complexes. However, the observation that some ODAs are still incorporated into the axoneme in the absence of ODA16 suggests that there might be alternative albeit less robust mechanisms of ciliary import of ODAs (16,18). Although it is difficult to imagine how larger quantities of the MDa-sized ODA complexes enter cilia by diffusion, additional weaker binding sites on IFT components other than IFT46 may result in the ciliary import of smaller quantities of ODAs. Another currently unanswered question is which component(s) of ODAs interact with IFT46. Our pulldown experiments with ODAs extracted from C. reinhardtii axonemes do not provide us with any insights into which ODA subunit may be responsible for the binding to ODA16. Also, although it appears likely that the conserved extended patch on the C-terminal face of the ODA16 ␤-propeller is the interaction surface for ODAs (Fig. 4c), this notion

Crystal structure of ODA16
remains to be proven. Another important unanswered question is how ODAs dissociate from the ODA16-IFT complex at the tip of flagella to allow incorporation into the growing axoneme. Possibly, the remodeling of IFT complexes at the tip of cilia weakens the ODA16-ODA interaction to release the ODA cargoes and allow for axonemal incorporation or such a switch is mediated by as yet unidentified post-translation modification of one or more key players in this process.
An important question in cilia biology is how large amounts of axonemal components are transported along cilia to their assembly site at the ciliary tip and how all these factors are bound by the large, but still limited, surface of the multisubunit IFT complex. Whereas it is conceivable that less abundant cargoes share individual binding sites, those that are required in large amounts during ciliary/flagellar assembly (e.g. tubulin dimers, axonemal motors, radial spokes, etc.) should have their own dedicated attachment points (42,43). We and others have previously shown that tubulin dimers are transported by association with a binding module formed by the N-terminal domains of IFT81 and IFT74 (Fig. 7b) (22,26). Another tubulinbinding domain was shown to be located at the N terminus of IFT54 (31), but its exact function during ciliogenesis in vivo remains to be determined. Accumulating evidence points to dedicated binding sites for inner dynein arm subunits in the IFT subunits IFT56 (44) and IFT57 (45), based on the analysis of the flagellar proteome from respective Chlamydomonas mutants (Fig. 7b). Biochemical studies will be necessary to confirm such interactions in vitro. Taken together, our understanding of the interactions between the IFT complex and abundant ciliary cargoes has increased significantly in the past few years, but future studies will be necessary to determine the underlying biochemical and structural basis of efficient IFT-mediated ciliogenesis.

Cloning, expression, and purification of CrODA16 constructs
The CrODA16 open reading frame was obtained by PCR amplification from a C. reinhardtii cDNA library. Sequences encoding the full-length protein or the various N-and C-terminal truncations were amplified with addition of a C-terminal hexahistidine tag, as well as a 5Ј BamHI and a 3Ј HindIII recog-  figure. b, pulldowns of various ODA16 constructs with GST-IFT46 shows that the N-terminal ODA16 domain, but not the ϳ25-residue unstructured tail of ODA16, is required for association with IFT46. c, neither ODA16N alone nor the N-and C-terminal domains in trans are significantly pulled down by GST-IFT46, demonstrating the requirement for covalent attachment of the two ODA16 domains for IFT46 association. C-term., C-terminal; N-term., N-terminal; MW, molecular mass. nition sequence, and cloned into MCS2 of the pFL vector (46) between BamHI and HindIII restriction sites. The resulting plasmid was transformed into DH10Bac competent cells, and recombinant baculovirus was produced in Sf21 cells exactly as described previously (31). The virus was then used to infect large volumes (typically 3-6 liters) of HighFive cells at a density of 10 6 cells/ml and incubated for 3 days at 26°C. The cells were harvested, resuspended in lysis buffer (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 10% glycerol, 5 mM ␤-mercaptoethanol, 1ϫ complete protease inhibitor mixture (Roche), 10 g/ml DNa-seI), and sonicated to open the cells (VS70T sonication tip, 40% amplitude, 1 s on and then 2 s off; total energy typically 8 -10 kJ). The lysate was clarified by centrifugation at 75,000 ϫ g for 45 min, and the supernatant was incubated with 3 ml of nickelnitrilotriacetic acid resin (Roche) for 2 h at 4°C. The resin was then collected and washed three times in lysis buffer, and the bound material was recovered by incubation with 35 ml of elution buffer (20 mM Tris, pH 7.5, 200 mM NaCl, 10% glycerol, 500 mM imidazole, pH 7.5). The obtained eluate was passed through a 5-ml ion-exchange column (Q-Sepharose HP), and the flowthrough was concentrated and loaded onto a HiLoad Super-dex200 SEC column in SEC buffer (10 mM HEPES-KOH, pH ODA16 ODA16-IFT46N Input (ODA fraction) 7.5, 200 mM NaCl). Finally, peak fractions were analyzed by SDS-PAGE, and fraction containing the protein at sufficient purity were pooled and concentrated and then either used immediately or snap-frozen in aliquots in liquid nitrogen for future use.

Cloning, expression, and purification of CrIFT46 constructs
CrIFT46 constructs were amplified and cloned into the pEC vector series for bacterial expression (with N-terminal His 6 / GST/TEV or His 6 /TEV tags) as described in previously (33,34). BL21(DE3) Escherichia coli cells were transformed with the resulting plasmids and grown (typically as 2 liter cultures) to an A 600 of 1.0 at 37°C. The temperature was then reduced to 18°C, and expression was achieved by the addition of 0.5 mM isopropyl ␤-D-thiogalactopyranoside and incubation in a shaker at 18°C overnight. The cells were harvested by centrifugation and resuspended in 2 pellet volumes of lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10% glycerol, 5 mM ␤-mercaptoethanol, 10 g/ml DNaseI, 1 mM PMSF). Cell lysis and protein purification were performed exactly as described for other IFT complexes (47). To obtain GST-tagged IFT46 for pulldowns, the elution from the nickel-nitrilotriacetic acid column was concentrated and subjected to SEC.

GST-affinity pulldowns
200 l of 10 M GST-tagged IFT46 protein was immobilized on 15 l of GSH-affinity resin in binding buffer (20 mM Tris-HCl, pH 7.5, 200 mM NaCl, 5% glycerol, 1 mM DTT). When necessary, resin was incubated only with buffer to check for nonspecific interaction of the prey (ODA16 full-length protein or truncation) with the beads. After 2 h of incubation at 4°C on a rotating wheel, the resin was collected by centrifugation (500 ϫ g, 4°C) and washed once with binding buffer. For each pulldown reaction, the prey was diluted to a concentration of 10 M in a total volume of 200 l with binding buffer, and an input sample was removed for SDS-PAGE analysis. The prey dilution was added to the washed resin, the volume was adjusted to a final volume of 500 l, and binding was allowed to occur at 4°C on a rotating wheel for 2 h. The resin was then collected by centrifugation and washed three times using binding buffer supplemented with higher concentrations of NaCl as required for this experiment. Finally, the bound material was eluted by incubation with 40 l of binding buffer supplemented with 20 M reduced glutathione. Input and elution samples were then analyzed by SDS-PAGE, and the gel was stained with Coomassie solution.

Interaction analysis using ITC
ITC experiments were done using a Microcal PEAQ-ITC calorimeter (Malvern). Both IFT46 and ODA16 proteins were in a buffer containing 10 mM HEPES, pH 7.5, 100 mM NaCl, and 5% glycerol. A volume of 200 l of ODA16 at a concentration of 20 M was titrated with different IFT46 protein constructs at a concentration of 200 M at 25°C. For each ITC experiment, a background curve consisting of the titration of the same IFT46 sample into buffer was subtracted to account for heat generated by dilution. Each experiment was carried out in triplicates. The program MicroCal PEAQ-ITC provided by Malvern was used to analyze all ITC data.

Crystallization of ODA16
Full-length CrODA16 at a concentration of 10 mg/ml in a buffer containing 10 mM HEPES, 500 mM NaCl, and 1 mM DTT was mixed in equal volume of precipitant solution containing 20% PEG 3350, 300 mM ammonium sulfate, and 100 mM Bis-Tris, pH 5.5, and crystallized using the sitting drop vapor diffusion technique. Crystals grew after 2-3 days at 18°C and were cryo-protected in mother liquor supplemented with 15% glycerol and subsequently flash-cooled in liquid nitrogen.

X-ray diffraction data collection and crystal structure determination
X-ray diffraction data were collected at the Swiss Light Source (Villigen, Switzerland) at the PXII beamline on a Pilatus 6M detector and indexed with the XDS package (48) before scaling with Aimless as part of the CCP4 package (49). The space group appeared to be P6 1 22 at the stage of data scaling, but molecular replacement and refinement revealed the true space group to be P6 1 with merohedral twinning (twin law h, ϪhϪk, Ϫl, twin fraction of 0.31 according to phenix.refine; see Table 1 for more twinning statistics). Molecular replacement using an eight-bladed ␤-propeller structure with the highest available sequence identity to ODA16 (RSA4, PDB code 4wjs) gave a clear solution in the program Phaser (50). However, because of the high 8-fold pseudosymmetry within the ␤-propeller, the molecular replacement solution had to be turned by "two blades" for successful structure solution. The N-terminal domain placed on top of the ␤-propeller was built manually in Coot (51) followed by refinement in PHENIX (52). Refinement in space group P6 1 including the twin law in PHENIX resulted in an R free 7.5% points lower than for refinement in the higher symmetry P6 1 22 space group. The data and refinement statistics are listed in Table 1.

Small-angle X-ray scattering of ODA16 and ODA16-IFT46N
SAXS experiments on ODA16 or the ODA16-IFT46N complex were performed at the BM29 Beamline (European Synchrotron Radiation Facility, Grenoble, France) using a Pilatus 1M detector using a similar protocol the one used previously (53). In brief, SAXS data were collected on proteins and protein complexes eluting directly from a Superdex 200 10/300 GL SEC column. Radiation damage, data merging, and buffer subtraction were performed on site and later verified manually using the program PRIMUS (54). All SAXS parameters such as maximum particle size (D max ) were extracted using GNOME (55) from the ATSAS package software (56). Theoretical SAXS curves were calculated from crystal structures using CRYSOL (57) and fitted to the experimental data. Ab initio models of ODA16 and ODA16-IFT46N were calculated using the beadmodeling program DAMMIN (58) to generate 10 independent reconstructions. SUPCOMB (59) and DAMAVER (60) were employed to generate the average representative models shown in Fig. 6e.

Flagellar isolation and fractionation
Chlamydomonas flagella were isolated from 8 liters of initial culture grown in the Tris acetate phosphate medium using the pH shock method (61). To obtain the outer dynein arm fraction, purified flagella were resuspended in HMDEK buffer (30 mM HEPES, 5 mM MgSO 4 , 1 mM DTT, 0.5 mM EGTA, 25 mM KCl), snap frozen in liquid nitrogen, and then allowed to thaw at room temperature (62). After three consecutive freeze-thaw cycles, flagella were pelleted by centrifugation at 10,000 rpm for 10 min. Supernatants was removed, and the pellets constituting the axonemes were resuspended in HMDE (30 mM HEPES, 5 mM MgSO 4 , 1 mM DTT, 0.5 mM EGTA) buffer containing 0.6 M NaCl. After 30 min of incubation on ice, the salt-extracted fraction enriched in flagellar ODAs were separated from axonemes by centrifugation at 13,000 rpm for 20 min and dialyzed overnight in HMDE buffer containing 30 mM NaCl.

Proteomics
For each pulldown reaction, flagellar outer dynein arm fraction was used as an input. SEC-purified GST-IFT46 in combination with untagged ODA16 and His-tagged ODA16 (fl), ODA16 (80 -423) , and ODA16  were used as baits for GST pulldown and Ni 2ϩ pulldown, respectively. Recombinant GST protein immobilized on GSH-Sepharose beads or Ni 2ϩ beads were utilized for negative control experiments. Bait proteins were immobilized on the respective beads by incubating 200 l of 10 M protein with a 20-l bed volume of beads. After incubation for 1 h, the beads were collected, washed thrice with HMDEK buffer, and incubated with Chlamydomonas ODA fraction for 4 h at 4°C on a rotating wheel. GSH or Ni 2ϩ resins along with co-precipitated protein complexes were separated by low-speed centrifugation at 400 ϫ g for 2 min. The beads were washed twice with cold 20 mM HEPES buffer. Co-precipitated proteins were eluted from GSH beads with GST elution buffer (10 mM Tris-HCl, pH 8, 25 mM reduced glutathione) and from Ni 2ϩ beads with HMDEK buffer with 300 mM immidazole. All experiments were performed in triplicate, and elution fractions corresponding to each of the reaction were loaded separately and stacked in a 10% SDS-PAGE gel. Total protein from each lane was digested according to the standard in-gel digestion protocol (63). Digested peptides were purified and concentrated on a C18 StageTip (64) and were analyzed using a Q Exactive HF mass spectrometer (ThermoFisher Scientific). An Easy nLC 1000 system (Thermo Scientific) was used to separate peptides via a 140-min gradient through a column (15-cm column length, 75-m inner diameter, packed with 1.9-m beads) maintained at a constant temperature of 45°C. Raw data were processed using MaxQuant computational platform (65), and statistical analysis was done using the Perseus framework.