Inner lumen proteins stabilize doublet microtubules in cilia/flagella

Motile cilia are microtubule-based organelles that play important roles in most eukaryotes. Although it is known that microtubules in cilia are sufficiently stable to withstand their beating motion, it remains unknown how they are stabilized while serving as tracks for axonemal dynein and intraflagellar transport. To address this question, we identified a new class of microtubule-associated proteins, named FAP45 and FAP52, in Chlamydomonas. These proteins are conserved among eukaryotes with motile cilia. Using cryo-electron tomography (cryo-ET) and high-speed atomic force microscopy (HS-AFM), we established that lack of these proteins leads to a loss of inner protrusions in B-tubules and less stable microtubules. These inner protrusions are located near the inner junctions of doublet microtubules and lack of FAP45, FAP52, and FAP20 results in detachment of the B-tubule from the A-tubule, as well as flagellar shortening. These results demonstrated that FAP45 and FAP52 bind to the inside of microtubules and stabilize ciliary axonemes.


Introduction
Cilia and flagella are microtubule-based organelles that operate as both antennae and propellers in eukaryotic cells. The structures and related genes of cilia and flagella are well conserved among eukaryotes. Cilia are classified as either non-motile or motile. Non-motile cilia, or "primary cilia", function as antennae and are involved in signal transduction through the hedgehog, Wnt, and Ca 2+ signaling 1 . Motile cilia and flagella beat at 20-60 Hz 2, 3 and drive cellular motility and fluid flow. Since the cells in most tissues are ciliated, ciliary dysfunction leads to various types of human diseases termed ciliopathies, which include hydrocephalus, situs inversus, retinal degeneration, and nephronophthisis 4 . Furthermore, recent studies revealed that the assembly of cilia is significantly decreased in several types of tumors 5,6 , suggesting a close relationship between cilia and tumorigenesis.
The axoneme is the core structure of cilia and flagella and is highly conserved from protists to vertebrates. The axoneme is composed of a central pair of microtubules cylindrically surrounded by nine doublet microtubules (DMTs). This arrangement is often referred to as the 9+2 structure (Fig. 1a). In DMTs, 10 protofilaments of the B-tubule are attached to 13 protofilaments of the A-tubule at the inner and outer junction (Fig. 1a). In motile cilia and flagella, structures essential for motility, such as axonemal dyneins, radial spokes, and the nexin-dynein regulatory complex (N-DRC), are arranged on DMTs with a 96-nm repeating unit [7][8][9] . Axonemal dyneins bound on A-tubules slide on the neighboring B-tubules and this sliding propagates along the axoneme, generating a bending force. The activity of dyneins is regulated by the interaction between the radial spokes and the central pair of singlet MTs [10][11][12] .

Recent advances in cryo-ET techniques have dramatically revealed structural details of
DMTs, yet a significant question remains-What stabilizes DMTs? Cytoplasmic MTs frequently bend with various degrees of curvature 13,14 and this bending occasionally induces MT breakage 15 . Furthermore, a recent study demonstrated that cytoplasmic MTs were damaged even after bending only several times 16 . Therefore, to ensure a high beat frequency, the DMTs of motile cilia and flagella should be structurally robust compared with cytoplasmic MTs. However, the mechanism that provides DMTs with such robustness remains to be clarified.
Nicastro and colleagues have provided insights into the structural basis of DMTs using cryo-ET of Chlamydomonas flagella. They reported periodic high densities on the inner surfaces of A-tubules and B-tubules, which they named microtubule inner proteins (MIPs, Fig.   1a) 7,17 . MIPs have also been observed in the axonemes of higher organisms 7,[17][18][19][20] , implying that these inner structures are essential for the integrity of DMTs in motile cilia and flagella.
In this paper, we explored the mechanisms stabilizing DMTs using Chlamydomonas mutants and various techniques, including cryo-ET and HS-AFM. We found that two uncharacterized flagellar-associated proteins (FAP), FAP45 and FAP52, are essential for the stability of B-tubules. Lack of both FAP45 and FAP52 leads to the loss of MIPs in B-tubules, resulting in disruption of the MT walls. Furthermore, the B-tubule wall detached from the A-tubule at the inner junction, and flagellar shortening was observed when both FAP52 and FAP20 were absent. These results indicate that the B-tubule is reinforced by microtubule inner proteins and its stability is essential for maintaining the structure of the axoneme.

Identification of proteins essential for DMT stabilization in flagella
To identify proteins that stabilize DMTs, we searched the Chlamydomonas flagellar proteome database 21 using the assumption that these proteins are (1) abundant in the axoneme fraction, (2) tightly bound on the axoneme even in the presence of high salt, and (3) highly conserved among ciliated organisms. The proteins that satisfy these criteria are listed in Supplementary   Table S1. We focused on FAP45, an uncharacterized protein whose peptides were most frequently found in the proteomic analysis (Table S1, "total unique peptide" and "Axo" columns). FAP45 is a coiled-coil protein composed of 501 amino acids, has a predicted molecular mass of a ∼ 59 kDa, and is conserved among organisms with motile cilia. The human ortholog of FAP45 is coiled-coil domain-containing protein 19 (CCDC19), also known as NESG1. The transcript of this protein is enriched in the nasopharyngeal epithelium and trachea 22 . However, the functions of FAP45/CCDC19 are totally unclear and thus we focused on FAP45.
We first investigated the partner that interacts with FAP45 on the axoneme using chemical crosslinking. We treated isolated wild type axonemes with 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC, zero-length crosslinker). The crosslinked products were solubilized and immunoprecipitated with a polyclonal anti-FAP45 antibody. Tubulin was identified as the major crosslinked partner of FAP45 using western blot ( Fig. 1c, open arrowhead; data not shown), suggesting a direct interaction between FAP45 and tubulin. In addition, a mass spectrometric analysis revealed that the ∼ 130 kDa product is composed of FAP45 and FAP52 proteins, probably in a 1:1 ratio ( Fig. 1c and

FAP52 mutants
We used Chlamydomonas mutants lacking FAP45 and FAP52 to investigate their function.
The mutants were isolated from ∼ 10,000 clones in an insertionally-mutagenized Chlamydomonas library 25 . In these mutant strains, the ~1.8 kbp aphVII fragment was inserted into the 5'UTR of the FAP45 gene or the first exon of the FAP52 gene ( Supplementary Fig.   S1b). These strains did not express detectable FAP45 or FAP52 protein as analyzed by a western blot of axonemal fractions (Fig. 1b, Supplementary Fig. S1c). On the other hand, fap45 and fap52 axonemes retained the known major axonemal components, such as outer arm dyneins, inner arm dyneins, radial spokes, and the dynein regulatory complex (N-DRC) ( Fig. 1b). Despite FAP45 and FAP52 being crosslinked by a zero-length crosslinker, they localized on the axoneme independent of each other (Fig. 1b). Consistent with the mass spectroscopic analysis, the ~130 kDa crosslinked product of FAP45 and FAP52 was not detected in the crosslinked fap45 or fap52 axoneme ( Fig. 1c and d, filled arrowheads), suggesting that FAP45 and FAP52 are neighbors on the axoneme.
Next, we examined the motility phenotype of fap45 and fap52. The swimming velocity and beat frequency of fap45 cells were slightly reduced (Fig. 1e, Movie 1), whereas the phenotype of fap52 was quite similar to that of wild type (Fig. 1f, Movie 1). Intriguingly, the double mutant fap45 and fap52 (fap45fap52) swam significantly slower than wild type, with a low beat frequency (Fig. 1g, Movie 1), even though there was no decrease in the major known axonemal components essential for flagellar motility (Fig. 1b).

FAP45 and FAP52 are luminal proteins in B-tubules
We biochemically localized FAP45 and FAP52 in DMTs by fractionating the axonemal proteins using increasing concentrations of sarkosyl 26 ( Supplementary Fig. S2a). FAP45 began to be extracted from the pellet at 0.2% sarkosyl, half of the FAP45 was extracted at 0.3%, and the protein was completely extracted from the pellet at 0.7% ( Supplementary Fig.   S2b). FAP52 started to be solubilized at 0.3% sarkosyl and almost all the protein was solubilized at 0.7%, although a small amount of protein still remained in the pellet. These behaviors correlate well with solubilization of the B-tubule.
To distinguish whether FAP45 and FAP52 are located outside or inside the B-tubule, we prepared axonemes containing biotinylated FAPs and tested whether or not the proteins are accessible to streptavidin. These axonemes were purified from rescue strains expressing FAP proteins whose N-or C-terminus was fused to biotin carboxyl carrier protein tag (BCCP tag, Fig. S2b and c) 27,28 . However, essentially no signal of biotinylation was detected on those axonemes ( Supplementary Fig. S2d), suggesting that streptavidin access to BCCP tag was prevented, presumably by the B-tubule wall. Therefore, we treated axonemes with 0.15% sarkosyl, which partially broke the B-tubule wall and, as expected, resulted in detection of the tagged proteins ( Supplementary Fig. S2e). Taken together with the results that the loss of FAP45 and FAP52 did not affect the incorporation of other axonemal proteins bound on the outer surface of the DMT (Fig. 1b), these data indicate that FAP45 and FAP52 are luminal proteins enclosed by the B-tubule.

Lack of FAP45 and/or FAP52 causes structural defects in B-tubules
To reveal the defect caused by the loss of FAP45 and FAP52, we observed the mutant axonemes by cryo-ET. Interestingly, the mutants had structural defects inside of the B-tubule  To investigate the reasons why the double mutant fap45fap52 cells swim significantly slower than single mutants, we observed the fap45fap52 axoneme using cryo-ET. In this case, we could not apply 3D sub-tomographic averaging because many of the axonemes were frayed and the B-tubules were partially missing ( Fig. 3a and b). Since cryo-ET is not suitable for observing the shapes of individual DMTs due to the missing wedge effects, thin sections of the fap45fap52 axoneme were observed by conventional EM and revealed that the B-tubules were missing in ∼ 33.8% of the outer DMTs (Fig. 3c). All of the protofilaments in the B-tubules were completely missing in some DMTs, whereas DMTs remaining several protofilaments in the B-tubules were also observed. These data suggest that the loss of both FAP45 and FAP52 decreases structural stability between the B-tubule protofilaments. Of note, we also found that the density of dynein e, a species of inner arm dyneins was decreased in fap45 and fap52 ( Supplementary Fig. S3a-c). The density of dynein e is smaller than that of other inner arm dyneins even in wild type ( Supplementary Fig. S3a), implying that binding of dynein e to the B-tubule is flexible compared to that of other dyneins. Therefore, it is possible that the lack of FAP45 or FAP52 slightly changes the shape of B-tubules and affects the interaction between dynein e and B-tubules.

Direct observation of B-tubule depolymerization by HS-AFM
The above "static" DMT structures suggested that the B-tubules of fap45fap52 are less stable than those of wild type. We directly observed the stability of the B-tubule using HS-AFM 30 , in which a small cantilever tip (tip diameter We first observed in vitro polymerized MTs in the absence of a stabilizing agent such as taxol. It was previously reported 25 that MTs depolymerize spontaneously within two or three frames (1.0 to 1.5 s), suggesting that depolymerization requires less than 50 nm/s. Actual depolymerization could be faster 26

MIP3a is important for B-tubule anchoring to the A-tubule at the inner junction of DMT
Since MIP3a is located near the inner junction between the A-and B-tubules and appeared to attach to the A-tubule 28 , we investigated whether FAP52 is involved in stabilizing the junction.
We previously reported that FAP20 is a component of the inner junction and a null mutant of FAP20 (fap20) showed an abnormal swimming phenotype, although the structure of the DMTs appeared to be normal 31 . Based on the hypothesis that MIP3a anchors the B-tubule to the A-tubule together with FAP20, we constructed the double and triple mutants fap20fap45, fap20fap52, and fap20fap45fap52. Most fap20fap45 cells were trembling and their motility appeared slightly worse than that of fap20 cells (Movie 8). In contrast, fap20fap52 and fap20fap45fap52 cells were essentially paralyzed (Movie 8). Furthermore, the flagellar lengths of fap20fap52 and fap20fap45fap52 deviated greatly from wild type, fap20, and fap20fap45 (Fig. 5c), with a far higher ratio of short flagella. To identify the cause of these defects, we observed the axonemes of the mutants by thin section TEM. The axonemes and DMTs of fap20fap45 appeared similar to those of fap20 ( Fig. 5c and d). Interestingly, several doublet B-tubules in fap20fap52 and fap20fap45fap52 were detached from the A-tubules (Fig.   5c arrowheads; Fig. 5d) whereas no detachment of B-tubules from A-tubules was observed in fap20fap45 axonemes. Given that such defects were rarely observed in fap20 axonemes 31 , these results demonstrate that MIP3a is required for anchoring the B-tubule to the A-tubule at the inner junction together with FAP20, and the junction is important for stabilizing the axonemal structure.

Discussion
Ciliary/flagellar MTs are remarkably stable compared with cytoplasmic MTs. Of the "9+2" MTs, DMTs are directly stressed by the axonemal dynein-generated force and constantly bent and straightened in Chlamydomonas flagella or trachea cilia, yet their structures remain intact.
The mechanism that stabilizes axonemal MTs has not been clarified. Here, we identified FAP45 and FAP52 as proteins that stabilize DMTs by binding to the inner lumen of the B-tubule. This is the first report to identify factors essential for the stabilization of ciliary/flagellar MTs.
Based on our studies and existing data, we propose a schematic model of FAP45 and FAP52 (Fig. 6b). Our cryo-ET observations show that MIP3a and MIP3c are composed of FAP52 and FAP45, respectively (Fig. 6a). FAP52/MIP3a anchors the protofilaments A13 and B10, whereas FAP45/MIP3c is bound inside of B7, B8, and B9 protofilaments. Besides, biochemical crosslinking shows that FAP45 and FAP52 directly interact with each other.
Thus, these two proteins bundle B7-B10 protofilaments to reinforce the B-tubule. Consistent with this model, the B-tubules of fap45 were more vulnerable to physical stress than those of fap52 in the HS-AFM observation (Fig. 4d, Movie 6). The MIP3a structure was observed also in vertebrates 7, 18 , but MIP3c was not clearly described probably due to the limitation of resolution. However, the amino acid sequence of FAP45 is highly conserved from protists to mammals ( Supplementary Fig. S4), suggesting that the functions of FAP45 are conserved also in higher organisms as in Chlamydomonas.
Our data also suggest that DMTs are stabilized by "fail-safe" mechanisms. The single mutants of FAP45 and FAP52 did not show significant decreases in swimming speed and only the double mutant fap45fap52 showed slow swimming and defects of the B-tubules. Although the B-tubules of fap45fap52 were more easily depolymerized than wild type B-tubules, the speed of depolymerization was slower than that of in vitro polymerized pure MTs. This suggests that B-tubules are also stabilized by other mechanisms, such as tubulin acetylation 32 ( Supplementary Fig. S5) or fMIPs that bind inside the B-tubule along its length 29 .
The inner junctions of DMTs are also stabilized by "fail-safe" mechanisms involving multiple proteins. We previously demonstrated that FAP20 is a constituent of the inner juction 31 . In addition to the "true" inner junction, our data clarified that MIP3a also connects B-tubules to A-tubules ( Fig. 5c and d). Furthermore, some inner junctions appeared to be attached in fap20fap52 and fap20fap45fap52 mutants (Fig. 5c). In the triple mutant, additional proteins, such as tektin and PACRG, may contribute to stabilizing the inner junction, given our previous observation that these two proteins were partially retained in fap20 31 .

Strains, culture conditions, and isolation of fap45 and fap52 mutants
The Chlamydomonas strains used in this study are listed in Supplementary Table S3. The double and triple mutants were constructed by standard methods 3

. All cells were cultured on
Tris-acetate-phosphate (TAP) plates with 1.5% agar or in TAP medium. fap45 and fap52 mutants were isolated from a library of mutants generated by aphVIII gene insertion 25 . The mutants were backcrossed with wild type several times before use.

Motility assay using the CLONA system
To assay the motilities of Chlamydomonas cells in TAP medium or TAP with ficoll, videos were recorded using a high-speed camera (EX-F1; Casio) attached on a dark-field light microscope (BX51; Olympus) at 600 fps. The videos were analyzed using the CLONA system 34 and the results were plotted using R Project software.

Antibodies
The antibodies used in this study are described in Supplementary Table S4. Anti-FAP45 and anti-FAP52 antibodies were raised against full-length FAP45 and FAP52 proteins, respectively. The full-length cDNA of FAP45 or FAP52 was cloned between the EcoRI and BamHI sites of pMAL-c2x (New England Biolabs). In both constructs, a 6×His tag was inserted into the HindIII site of pMAL-c2x to enable purification in two steps. For affinity purification, each cDNA was also cloned between the NdeI and EcoRI sites of pColdI (Takara). Expression of the recombinant proteins was induced in Escherichia coli BL21(DE3) with 0.3 mM IPTG, and almost all of the expressed protein was solubilized from each construct. MBP-FAP45-6×His and MBP-FAP52-6×His were purified with amylose resin (New England Biolabs) and then further purified with Ni-NTA agarose (Qiagen). These purified proteins were used as antigens. Each antibody was affinity purified using polyvinylidene difluoride membranes blotted with FAP45-6×His or FAP52-6×His.

Immunofluorescence microscopy
Nucleoflagellar apparatus (NFA) was prepared as previously described 36 . After fixation with 2% formaldehyde for 10 min at room temperature, NFAs were treated with cold methanol (-20°C). The samples were immunostained as previously described 37 . Images were taken with a CCD camera (ORCA-R2; Hamamatsu Photonics) linked to a fluorescence microscope (IX70; Olympus).

Generation of BCCP-tagged strains
Fragments from the start codon to immediately before the stop codon in the FAP45 and FAP52 genes were amplified by genomic PCR and inserted into pIC2 plasmids 38 . In the FAP52 construct, a 3×HA tag was inserted into the C terminus of FAP52. Each construct was linearized and transformed into fap45 or fap52 cells by electroporation.
Streptavidin-Alexa546 staining of the axonemes was performed as previously described 31 .

Thin-section TEM
Samples for thin-section TEM were prepared as previously described 39

Image acquisition
Grids were transferred into the JEM-3100FEF using a high-tilt liquid nitrogen cryo-transfer holder (914; Gatan Inc.). Tilt series images were recorded at -180°C using a K2 summit direct detector (Gatan) and the serialEM 40 . The angular range was ±60° with 2.0° increments. The total electron dose was limited to 100 e -/Å 2 and the nominal magnification was 6,000x. An in-column energy filter was used with a slit width of 20 eV and a pixel size of 7.1 Å.

Image processing
Image processing for subtomogram averaging of DMT was carried out as previously described 27,28 . Tilt series images were aligned and back-projected to reconstruct 3D tomograms using IMOD 41 . Alignment and averaging of the subtomograms were performed using custom Ruby-Helix scripts 42 and PEET software 7 to average the 96-nm repeats of DMTs. UCSF Chimera was used for isosurface renderings 43 .
The AFM images were recorded using a home-built high-speed atomic force microscope [44][45][46] at frame rates of 1-2 fps. All observations were performed at room temperature.

Movie 1
Swimming of wild type, fap45, fap52, and fap45fap52 in TAP medium.

Movie2
Swimming of confluent cultured fap45fap52 in TAP medium.

Movie 3
Swimming of wild type and fap45fap52 in TAP with 7.5% ficoll.

Movie 4
HS-AFM movie of wild type DMT on a mica surface.

Movie 5
HS-AFM movie of fap45fap52 DMT on a mica surface.

Movie 6
HS-AFM movie of fap45 DMT on a mica surface.

Movie 7
HS-AFM movie of fap52 DMT on a mica surface.