Mutation of CFAP57 causes primary ciliary dyskinesia by disrupting the asymmetric targeting of a subset of ciliary inner dynein arms

Primary ciliary dyskinesia (PCD) is characterized by chronic airway disease, male infertility, and randomization of the left/right body axis, and is caused by defects of motile cilia and sperm flagella. We screened a cohort of affected individuals that lack an obvious TEM structural phenotype for pathogenic variants using whole exome capture and next generation sequencing. The population sampling probability (PSAP) algorithm identified one subject with a homozygous nonsense variant [(c.1762C>T) p.(Arg588*) exon 11] in the uncharacterized CFAP57 gene. In normal human nasal epithelial cells, CFAP57 localizes throughout the ciliary axoneme. Analysis of cells from the PCD patient shows a loss of CFAP57, reduced beat frequency, and an alteration in the ciliary waveform. Knockdown of CFAP57 in human tracheobronchial epithelial cells (hTECs) recapitulates these findings. Phylogenetic analysis showed that CFAP57 is conserved in organisms that assemble motile cilia, and CFAP57 is allelic with the BOP2 gene identified previously in Chlamydomonas. Two independent, insertional fap57 Chlamydomonas mutant strains show reduced swimming velocity and altered waveforms. Tandem mass spectroscopy showed that CFAP57 is missing, and the “g” inner dyneins (DHC7 and DHC3) and the “d” inner dynein (DHC2) are reduced. Our data demonstrate that the FAP57 protein is required for the asymmetric assembly of inner dyneins on only a subset of the microtubule doublets, and this asymmetry is essential for the generation of an effective axonemal waveform. Together, our data identifies mutations in CFAP57 as a cause of PCD with a specific defect in the inner dynein arm assembly process. Significance Motile cilia are found throughout eukaryotic organisms and performs essential functions. Primary ciliary dyskinesia (PCD) is a rare disease that affects the function of motile cilia. By applying a novel population sampling probability algorithm (PSAP) that uses large population sequencing databases and pathogenicity prediction algorithms, we identified a variant in an uncharacterized gene, CFAP57. This is the first reported example of PCD caused by a mutation that affects only a subset of the inner dynein arms, which are needed to generate the waveform. CFAP57 identifies an address for specific dynein arms. These findings demonstrate the effectiveness of the PSAP algorithm, expand our understanding of the positioning of dynein arms, and identify mutations in CFAP57 as a cause of PCD.


Introduction
Motile cilia are complex organelles that project from the surface of cells and are essential for propelling fluids (e.g., in the airways, ventricles) or for providing cell locomotion (e.g., sperm, Chlamydomonas). The microtubule-based axoneme shows the classic structure of 9 doublet and 2 central pair microtubules, which is conserved throughout the eukaryotic lineage. Motility comes from the coordinated activity of inner and outer dynein arms (IDA and ODA, respectively) that are attached to the A tubule of the microtubule doublets. Defects in motile cilia cause primary ciliary dyskinesia (PCD), which is a genetically and phenotypically heterogeneous disorder that is characterized by chronic and debilitating respiratory disease, and is frequently accompanied by laterality defects (~50% of patients) due to abnormal left-right body asymmetry (1). The most common genetic lesions that cause PCD are those that affect components of the ODA, including DNAH5, DNAI1, and DNAH11 (2)(3)(4). Another group of PCD associated genes encode ODA-docking complex components (CCDC114 and CCDC151) (5)(6)(7), or the proteins that play a role in the cytoplasmic assembly of dynein (SPAG1, DNAAF1-3, HEATR2, and DYX1C1) (8)(9)(10)(11)(12)(13)(14). In addition, mutations in two genes, CCNO (15) and MCIDAS (16), cause a PCD-like phenotype by greatly reducing the number of motile cilia. Although genetic variants that cause PCD have been identified in over 40 genes (2,(17)(18)(19)(20)(21)(22)(23), there are individuals with confirmed clinical features of PCD but normal axonemal structure as determined by transmission electron microscopy, for whom the genetic basis of their disease is unknown.
In an ongoing effort, we have performed whole exome capture sequencing on more than 400 unrelated cases in order to identify genetic causes of PCD. Within this cohort, there were 99 unrelated cases with clinical features of PCD that include chronic otosino-pulmonary symptoms, and low levels of nasal nitric oxide who presented with no obvious axonemal defects by transmission electron microscopy (TEM). Using a new algorithm to analyze exome sequencing data (24), we identified an apparently homozygous stop-gain variant in ciliary and flagella associated protein 57 (CFAP57; MIM: 614259; NM_152498.3) in a patient with classical symptoms of PCD that included bronchiectasis, neonatal respiratory distress, otitis media, and sinusitis.
CFAP57 is allelic with the BOP2 gene identified previously in Chlamydomonas.
The BOP2 locus was first identified as a suppressor of the swimming defect of the pf10 mutant (25). Double mutant analysis, using IDA and ODA mutants, suggested that the mutation in the BOP2 locus affects only the IDAs. Electron tomography of the bop2 mutant showed that a subset of dynein arms were missing on only certain microtubule doublets (26). Specifically, electron tomography revealed reduced intensity on doublets 5, 6, and 8, while doublet 9 showed an intermediate loss of intensity (26).
To date, no pathogenic variants have been identified in proteins that affect only the IDA complexes in PCD patients. The IDAs play a crucial role in determining axonemal waveform and are more heterogeneous than the ODAs. For example, Chlamydomonas reinhardtii has six single headed IDA (a, b, c, d, e, and g) and one double headed IDA (I1/f) that span the 96 nm repeat, as determined by cyro-EM tomography using mutant strains affecting these proteins (27,28). In addition, there are three unique minor inner dynein arms that are only found in the proximal regions and replace the major IDAs; these include DHC3, DHC4, and DHC11 (29). Mutations in IDAs are difficult to detect by TEM cross sections since there are 7 dynein arms in 96 nm, and TEM sections are often 60 nm deep. Thus, TEM lacks the resolution necessary to detect changes in individual IDAs, as a single IDA would be obscured by the other IDAs in the section. Our data suggests that homozygosity of a pathogenic variant of CFAP57 causes PCD, likely by a failure to assemble a subset of inner dynein arms, and should be considered a candidate gene in cases of PCD with apparently normal axonemal structure by TEM and significantly reduced CBF.

Results
Whole exome capture and sequencing identifies a CFAP57 pathogenic variant in a PCD subject: We studied a 38 year old male from family UNC-1095 (Fig. 1a). The subject has situs solitus with a classical PCD phenotype that includes neonatal respiratory distress, otitis media, sinusitis, bronchiectasis, and a low rate of nasal nitric oxide production (40 nl/min; cut-off 77 nl/min) (2), but no axonemal defect was detected by TEM (Fig. 1b) (30), DNA from the subject, who had been previously screened for mutations in genes known to be associated with PCD, was sequenced using the IDT capture reagent (31). We analyzed the resulting genotype data using population sampling probability (PSAP), a statistical framework for assessing the significance of variants from n=1 cases of rare genetic disease (24) (Table S1). A new candidate gene, CFAP57, was identified and confirmed by direct Sanger sequencing ( Fig. 1c and 1d).
Analysis revealed a homozygous stop-gain variant [(c.1762C>T) p.(Arg588*)] in exon 11 of CFAP57, which is predicted to result in complete loss of function due to nonsensemediated decay of the primary CFAP57 transcripts. Two unaffected siblings were determined to be wild-type and a carrier of the pathogenic variant, consistent with an autosomal pattern of inheritance. CFAP57 is ~144 kD and contains 10 WD repeats along with 3 predicted coiled-coil domains. PSAP analysis identified the homozygous stop-gain in CFAP57 as the second most deleterious change in the subject's genome, following a homozygous missense change in CD2, a T-cell surface antigen. Mutations in the latter do not explain the subject's symptoms (Table S1).

Expression of CFAP57 in ciliated human airway epithelial cells: Normal primary
human tracheal epithelial cells (hTEC) cells were cultured at an air-liquid interface as previously described (32,33). Under these conditions, the cells first proliferate as an undifferentiated monolayer and then undergo ciliated cell differentiation. In hTEC cells, the expression of CFAP57 increased in parallel to the expression of FOXJ1, a key gene that drives ciliogenesis (Fig. 2a) and to the expression of DNAI1, a known ciliary specific gene (Fig. 2b). Similarly, the levels of CFAP57 protein increased as ciliated airway cells underwent differentiation, as determined by increased levels of FOXJ1 detected by immunoblot (Fig. 2c). This signal was strongly enriched in samples of detergent isolated ciliary axonemes (Fig. 2d). Immunofluorescent staining of CFAP57 in both intact hTEC cultures (Fig. 2e) and isolated cells demonstrate strong positive reactivity throughout the length of the ciliary axoneme (Fig 2f). These results demonstrate that CFAP57 is an axonemal protein, as suggested previously by proteomic analysis of both human and Chlamydomonas axonemes (34,35).
A pathogenic variant of CFAP57 causes defective ciliary beating: To characterize the effects of the genetic variant in CFAP57 on the function of motile cilia, we obtained human nasal epithelial (HNE) cells from the PCD subject and healthy controls. Cells were expanded in culture as conditionally reprogrammed cells (CRCs) (36), then allowed to differentiate using air-liquid interface cultures as previously described (22). Immunofluorescent staining of isolated ciliated cells from the control cultures showed CFAP57 reactivity along the entire length of the axoneme, while no positive staining was observed in cells from the PCD subject (Fig. 3a). Immunofluorescent intensities for the ODA protein DNAH5 and the radial spoke component RSPH1 were not different between PCD and control cells (Fig. S2), in agreement with the TEM analysis that showed no obvious axonemal structural defect (Fig. 1b). Although the fluorescent intensity of the IDA component, DNALI1, was not obviously different in cells obtained from the PCD subject compared to control cells, it appeared that there was a reduction in intensity at the tip ( Fig. 3b; Fig. S1). The lack of CFAP57 resulted in a significant reduction of ciliary beat frequency (CBF) in the PCD cells. n=21) was significantly reduced (p < 0.00001). Thus, the PCD cells exhibited shorter cilia, an altered waveform, and a reduced CBF.
CFAP57 silencing in hTEC results in reduced motile cilia motility: To confirm these results, the role of CFAP57 in motile cilia function was further defined by silencing expression using an RNAi approach in primary hTEC that were transduced with a control plasmid expressing a non-targeted shRNA sequence and a green fluorescent tag (37) or a CFAP57-specific shRNA plasmid together with a recombinant lentivirus that contains a cassette that confers puromycin resistance (33). In three biological replicates, CFAP57 expression was reduced by three out of the four CFAP57-specific shRNA sequences when compared to cells transduced with non-targeted shRNA sequences using RT-PCR (Fig 4a). Immunoblot analyses (Fig. 4b) confirmed the absence of the protein from total cell lysates (Fig. 4b), and immunofluorescent staining showed an absence of CFAP57 in ciliated cells. Silencing CFAP57 did not affect the degree of ciliogenesis in cultured airway cells (Fig.4). High-speed video microscopy analysis of ciliary motility (38) of the CFAP57-silenced cultures showed significantly reduced CBF when compared to control cells. (Fig. 4d). Analysis of ciliary beat using high speed video microscopy showed subtle changes in the ciliary waveform, which results in a slightly reduced curvature in CFAP57-silenced cells compared to control cells (Movies S3-S6).
CFAP57 is conserved in organisms with motile cilia: To evaluate the conservation of CFAP57 across species, we constructed a phylogenetic tree for CFAP57 (Fig. S3).
CFAP57 is found in most organisms with motile cilia, except in the cycads, gingkos, and the water fern, Marselia. As expected, CFAP57 is missing in organisms that lack motile cilia (flowering plants, nematodes, most fungi). The N-terminus of the protein is predicted to be composed of WD40 repeats that form beta-sheets and the C-terminus is predicted to be -helical (39).

Effect of CFAP57 mutations in Chlamydomonas:
Chlamydomonas reinhardtii is an important model for identifying ciliary genes and probing their functions (40). To further investigate the function of CFAP57, we carried out additional studies in Chlamydomonas. We obtained three insertional mutant strains (LMJ.RY0402.157050, LMJ.RY402.107706 and LMJ.RY0402.211005) from the CLiP collection, which is an indexed library of insertional mutations in Chlamydomonas (41,42). Insertions in two of the three strains were verified by PCR (Fig. S4). The insertion and drug resistance cosegregate with the swimming phenotype in 21 tetrads for LMJ.RY402.107706 and 7 tetrads for LMJ.RY0402.157050. LMJ.RY402.107706 carries an insert in exon 2 and LMJ.RY0402.157050 carries an insert in intron 7. We examined the mRNA for the two mutants. The transcript in LMJ.RY402.107706 could not be amplified across exons 2 and 3. In strain LMJ.RY402.107050, the transcript for exons 1-7 is present, but we were unable to amplify the remaining transcript (Fig. S4). These alleles failed to complement the bop2-1 allele in diploid strains and were tightly linked to the locus (n=89 tetrads).
Defective swimming and ciliary waveform in fap57 Chlamydomonas. We analyzed the swimming behavior of these mutants. Swimming velocity analysis shows that fap57-050 and fap57-706 swim significantly slower than wild-type cells (CC-125). fap57-050 was also slower than fap57-706 ( Fig. 5a; Table 1). Ciliary beat frequency extracted from the trajectory was unchanged compared to wild-type cells (Fig. 5b).
To characterize the waveform of mutants, both the fap57-050 and fap57-706 mutants were crossed with the uniciliate mutant uni1-2 (3). Fifty movies of fap57-050; uni1 and 51 movies of fap57-706; uni1 were compared to 153 movies of uni1-2 (wild-type samples previously described by Bottier et al. (44) (Fig. 5a). Body motion analysis shows that both mutants have a significantly slower body rotation rate compared to wild-type cells (Fig. 5c), which agrees with the reduced velocity. Both fap57 strains have a significantly reduced bend amplitude (Fig. S5b) and an increased average curvature compared to wild-type cells (Fig. S5c). The fap57-050 mutants also have an average curvature significantly smaller than fap57-706, which suggests that the fap57 mutant waveforms are less wavy, and straighter. Both the amplitude of power and torque are significantly reduced for the fap57 mutants compared to wild-type (Fig. S4d). Those results are compatible with a reduced swimming velocity as well as a reduced body net rotation. Both alleles displayed waveforms with a shorter stroke.
Protein composition of fap57 mutant cilia show a partial phenotype: Using isobaric tags (TMT) for tandem mass spec, we analyzed isolated cilia from four wild-type strains and two technical replicates of two different fap57 meiotic progeny. There was an average of 107 FAP57 peptides in wild-type cilia, but only 6 peptides in mutant cilia, which suggests a strong loss of function ( Table 2). Several of the inner dynein arms (DHC7, 3, 2) are reduced ( Table 2) compared to the other dynein heavy chains (Table   S3; Fig. S6). The translational elongation factor 1 alpha (EEF1A1) is also reduced in the fap57 mutant strains to the same extent as the inner dynein arms. This elongation factor is present in proteomic analyses of both Chlamydomonas and human axonemes (34,35,(45)(46)(47), which suggests it is not a cytoplasmic contaminant. It can be extracted with KCl and is found in the membrane matrix fraction (34). However, its role in the axoneme is unknown. Four additional proteins are significantly reduced but have few spectral counts ( Table 2). Cre12.g540050 and Cre10.g438500 have no orthologs outside of the green algae. Cre13.g562800 has a paralog in Chlamydomonas (Cre07.g313850) and both are likely orthologs of WDR49. Cre10.g438500/FAP264 is an ortholog of LRRC74.
In humans, WDR49 is expressed in the fallopian tubes and lung and FAM74A is highly expressed in the testes based on the GTEx project (48). They have no known functions.
To further examine these additional proteins, we tried a commercial antiserum to WDR49 (Table S5), and it failed to show specific staining in hTECs or Chlamydomonas.
We examined an insertional mutant strain from the CLiP collection (41) (Table S4).
Taken together, the above data demonstrates that CFAP57 plays an important role in the assembly of a specific subset of IDAs, and the proper positioning of these IDAs is important for the generation of normal waveforms. This is consistent with the phenotype observed in the PCD subject, in which absence of CFAP57 results in reduced CBF and altered waveform, in the absence of a structural defect detectable by TEM.

Discussion
Mutations in over 40 genes have been reported to cause PCD, and these account for the majority of disease cases (2,49). However, many cases of PCD (~30%) remain unsolved at the genetic level. The large number of causative genes are a direct reflection of the complexity of ciliary assembly and structure. Many of the known mutations occur in genes that encode structural proteins of the ODAs or preassembly factors for ODAs and IDAs. To date, no mutations have been reported that cause a defect that affect only the IDAs. We performed exome capture and sequencing on a patient with a clinical diagnosis of PCD who had no known genetic mutation, and no obvious structural changes visible by TEM. Using PSAP, a newly developed algorithm to analyze exome sequence together with large genome databases, we identified a homozygous stop-gain mutation in CFAP57. The PSAP algorithm uses large population sequencing databases and pathogenicity prediction algorithms to calculate the probability of sampling a particular genotype, or set of genotypes, observed in a single n=1 case. PSAP is most useful in rare disease studies where families are unavailable, and/or the cohort size is modest (dozens to hundreds). We currently use PSAP on cases of idiopathic nonobstructive azoospermia to identify new causative genes (50), and expect that PSAP will become a useful tool to identify causative genetic variants in other rare diseases. These may include diseases similar to azoospermia and PCD that are characterized by large effect mutations and extensive locus heterogeneity.
In cultured nasal epithelial cells from the PCD patient, we observed a complete absence of CFAP57 by immunofluorescence and the cells exhibited a significantly reduced CBF.
Knockdown of CFAP57 in hTEC cells using shRNA, confirmed by immunofluorescence and immunoblotting, also show a reduced CBF. In contrast, TEM analysis of nasal epithelial cells from the subject appear normal and immunostaining for the ODA protein DNAH5 was unchanged. However, immunostaining for DNALI1 suggested a defect in the localization along the axoneme, with an accumulation at the base and a reduction at the tip. The ortholog of DNAL11 in Chlamydomonas is p28/DII1 and it is found in three of the inner dynein arms, including the d dynein. Although the mechanisms that regulate basal CBF have been studied extensively, they remain unclear. These results suggest that a proper balance between ODAs and IDAs is also required to maintain normal CBF. Waveform analysis revealed subtle differences between the PCD or knockdown cells and controls that varied among the cells. The heterogeneous waveform observed between individual cilia in the PCD samples could be related to the maturity of individual cilium. It has been shown that the variation in ciliary waveform is associated with progression of ciliary length and the differentiation of ciliated cells (52).
To characterize the function of CFAP57 in more detail, we utilized the well-studied model organism Chlamydomonas. We show here that CFAP57 encodes a WDR protein that plays a role in the docking/assembly of a subset of IDAs. Although the ciliary axoneme is described as showing nine-fold symmetry, there are many structural asymmetries in the cilium (reviewed in Dutcher, in revision). The generation of waveforms requires the spatial and temporal regulation of dyneins, and this is likely to require structural asymmetries. Multiple approaches have helped to catalog the asymmetric structures and proteins in Chlamydomonas cilia. These asymmetries have been identified through analysis of mutant Chlamydomonas flagella by electron tomography and proteomics and are beginning to provide a wealth of information to use for understanding how asymmetric and symmetric waveforms are generated and propagated (27,28,45). If all of the ciliary dyneins were active at one time, the cilia would be in a rigor state, resulting in no net movement or bending. In order to generate an effective bend, dynein motor function must be tightly controlled both along the length of the cilium and around the circumference of the axoneme across a defined axis.
In elegant cryo-EM tomography studies, Lin and Nicastro unexpectedly found that most dyneins are in an active state confirmation with a smaller population of inactive dyneins (53). The locations of the inactive dyneins are asymmetric and bend-dependent. They propose a switch-inhibition mechanism in which the bend is generated by inhibiting, rather than activating, dyneins on one side of the cilium. The data suggest that the initiation of a bend starts with the inhibition of the inner dynein arms (a, d, and g) on specific doublet microtubules (DMT). DMT 2 to 4 is needed for the asymmetric waveform and DMT 7 to 9 for the symmetrical waveform that promotes a bend by the dyneins on the other side of the axoneme. CFAP57 is likely to play a key role in regulating the waveform.
The electron tomographic analysis of the Chlamydomonas bop2-1 mutant showed that CFAP57 is required for the assembly of dynein arms on only a subset of the nine doublet microtubules (26). A loss of IDAs on doublets 5, 6, and 8 was observed. In recent work using cryo-EM tomography, there is a loss of a subset of inner dynein arms on doublet microtubules 5-8 with a partial loss on doublets 1 and 9 (Lin et al.,   (Table 3).
In comparison to Chlamydomonas, less is known about the detailed structure of the IDA in human cilia. Based on the high level of conservation between species, it is likely that CFAP57 plays a similar role in human cilia as it does in Chlamydomonas. However, because the planar waveform of human cilia is inherently different from the waveforms of Chlamydomonas, the exact positioning and regulation of IDA activity is also likely different. Additional studies using advanced techniques to culture and study human respiratory cells are needed to address these questions.
In summary, our results show that a genetic variant in CFAP57 reduces CBF and alters waveform, likely by affecting the assembly of a subset of IDA, resulting in PCD. This is the first reported example of PCD caused by mutation of a protein that apparently affects only a subset of the IDAs, expanding our understanding of how dynein arms are positioned during cilia assembly. These findings demonstrate the usefulness of the PSAP algorithm and set a precedent to consider it in the evaluation of other cases of PCD with no obvious structural defects. Identifying the genetic basis of PCD and the functional defects is an important step toward developing personalized treatments for this rare disease.

Human subjects. The individuals included in this study provided informed consents and all protocols involving human studies were approved by the University of North
Carolina Medical School Institutional Review Board.
Human genetic analysis. A cohort of 99 PCD patients with no obvious EM phenotype was assembled for exome sequencing. Exome libraries were prepared using an IDT capture reagent. Genetic variants were discovered and genotyped using a validated analysis pipeline at the Washington University McDonnell Genome Institute as previously described (31). Each case was analyzed using the population sampling probability (PSAP) framework, a published statistical method for identifying pathogenic mutations from n=1 cases of rare disease (24). Segregation analysis was performed on the available DNA from family members (family UNC-1095). The primers used are listed in Table S6.

Airway epithelial cell cultures Human nasal epithelial (HNE) cells from the PCD subject (proband 2-II) and controls
were obtained as described (7). The nasal cells were expanded as conditionally reprogrammed cells (CRC) (8) and cultured as previously described (9).

Human tracheobronchial epithelial cells (hTEC) were obtained from non-smoking donors lacking respiratory pathologies provided by the Cystic Fibrosis Center Tissue
Procurement and Cell Culture Core (59), or were isolated from surgical excess of tracheobronchial segments of lungs donated for transplantation as previously described (11). These unidentified cells are exempt from regulation by HHS regulation 45 CFR Part 46. hTEC cells were expanded in vitro and allowed to differentiate using air-liquid interface (ALI) conditions on supported membranes (Transwell, Corning Inc., Corning, NY), as previously described (11,32,59). These protocols have been approved by the

Analysis of CFAP57 expression. RNA was isolated from cells using an RNeasy Mini
Kit (Qiagen) and RT-PCR was performed using specific primers (Table S6) as previously described (22). For qRT-PCR, RNA was reverse-transcribed using an Applied Bioscience High-Capacity Reverse Transcription Kit (Thermo Fisher Scientific). FluoroPure™ (Life Technologies). Slides were mounted using ProLong Diamond antifade mountant (Thermo Fisher). Images were acquired using a Zeiss -710 microscopy system or a Leica epifluorescent microscope (LAS X, Leica, Buffalo Grove, IL) Images were processed and the fluorescence intensity analyzed using FIJI (10) as previously described (11). Brightness and contrast were adjusted globally using Photoshop (Adobe Systems, San Jose, CA). Isotype matched control antibodies had no detectable staining under the conditions used. The antibodies used are listed in Table   S5.
Ciliary isolation, protein extraction and Immunoblot. Ciliary isolation and protein extraction was performed as previously described (35). Protein extraction from tissue or cells and immunoblot analysis was performed as previously described (35,61).

Ciliary beat frequency and waveform analysis. The ciliary beat frequency (CBF) in
HNE cultures (n=6) was measured as previously described (22,63). In hTEC cultures, CBF was measured in at least 5 fields obtained from each preparation. The cultures were maintained at 37 °C using a temperature controller and a stage heater block. Highspeed videos (120 frames/s) were recorded and processed with the Sisson-Ammons Video Analysis system (SAVA, Amons Engineering, Mt Morris, MI) as described (11,64). To analyze the ciliary waveform, cells were lifted off the supportive membranes and imaged directly on slides. High resolution videos of 6 cells in 3 different cultures were recorded as previously described (22). The videos were analyzed by an experienced scientist blinded to the genotype of the cells. Videos were replayed in slow motion. The ciliary length was measured and the waveform of the front and back cilia in 4 ciliated cells was traced manually. Deviation from linearity was determined by measuring the furthest ciliary displacement from the linear axis as defined by the end-effective and end-recovery position, as previously described (65).
Chlamydomonas axoneme isolation and mass spectroscopy. Chlamydomonas cilia were isolated using the dibucaine method (66). Cilia were demembranated with the addition of 1% Nonidet P-40 in HMDS-EGTA (10 mM HEPES, 5 mM MgSO4, 1  Chlamydomonas swimming analysis. Each video were analyzed using ImageJ (69) to create a binary file only displaying the cells. Each pixel had a spatial resolution of 310 x 310 nm and the temporal resolution between 2 consecutive time points was 1 ms.
Cells were tracked using the 2D/3D single-particle tracking tool of the MosaicSuite for Chlamydomonas kinematic analysis of cilium. The uniciliate mutant strain uni1-2 and the double mutants fap57-050; uni1 and fap57-706; uni1 were generated from meiotic crosses, as described by Dutcher (70) . The single mutant uni1 is considered the wild-type reference as previously published (43). Videos were analyzed using a custom-made program written in Matlab R2016a (The Mathworks, Natick, MA, USA) previously published (44). From each video, a sequence of 200 consecutive frames was stored in a 3D matrix of pixel intensity values. Each pixel had a spatial resolution of 169 x 169 nm and the temporal resolution between 2 consecutive time points was 0.5 ms.
Components of the forces exerted by cilia on the fluid are directly calculated from the Cartesian coordinates using parameters previously reported (44).

Statistical analyses. Group variation is described as mean ± standard error (SEM).
Statistical comparisons between groups were made using 1-way analysis of variance (ANOVA) with Tukey post-hoc analysis. Individual group differences were determined using a 2-tailed Student's t-test. A p value of 0.05 was considered to represent a significant difference. Non-parametric data are shown as the median and the 25 th and 75 th intraquartile ranges. Data were analyzed using Prism (GraphPad, La Jolla, CA).

Protein extraction and digestion
The samples were lysed in lysis buffer (8M Urea in 50mM HEPES pH 8.0) and sonicated briefly. Samples were reduced with 10 mM TCEP and alkylated with 25 mM iodoacetamide. Protein concentrations were determined by BCA protein assay (Thermo Scientific). Fifty μg of protein was taken from each sample for digestion. Sequencing grade protease Lys-C was added with 1:250 ratio and sample was digested overnight at room temperature with mixing. A second digestion was performed by diluting the sample with 50 mM HEPES to lower the urea concentration to 1M and trypsin was added with 1:100 ratio for a further 12 hour digestion at room temperature with mixing.
Digests were acidified with formic acid and subjected to Oasis HLB solid phase extraction column (Waters).

Tandem Mass Tag (TMT) Labeling
Digested peptides were labeled according to the TMT 10plex reagent kit instructions.
Briefly, TMT regents were brought to room temperature and dissolved in anhydrous acetonitrile. Peptides were labeled by the addition of each label to its respective digested sample. Labeling reactions were incubated without shaking for 1 h at room temperature. Reactions were terminated with the addition of hydroxylamine.
Subsequent labeled digests were combined into a new 2 mL microfuge tube, acidified with formic acid, subjected to Sep-Pak C18 solid phase extraction and dried down.

High pH Reverse Phase Fractionation
The dried peptide mixture was dissolved in 110 μL of mobile phase A (10 mM ammonium formate, pH 9.0). 100 μL of the sample was injected onto a 2.1 x 150 mm XSelect CSH C18 column (Waters) equilibrated with 3% mobile phase B (10 mM ammonium formate, 90% ACN). Peptides were separated using a gradient (71)  Peptides were resolved using 75 μm x 25 cm PepMap C18 column (Thermo Scientific).

Data Analysis
All MS/MS samples were analyzed using Proteome Discoverer 2.1 (Thermo Scientific).
The Sequest HT search engine in the Proteome Discover was set to search    (E and F) The localization of CFAP57 to the cilia was confirmed by immunofluorescence in (E) whole ALI culture and in (F) isolated ciliated cells.