Wdr47, Camsaps, and Katanin cooperate to generate ciliary central microtubules

The axonemal central pair (CP) are non-centrosomal microtubules critical for planar ciliary beat. How they form, however, is poorly understood. Here, we show that mammalian CP formation requires Wdr47, Camsaps, and microtubule-severing activity of Katanin. Katanin severs peripheral microtubules to produce central microtubule seeds in nascent cilia. Camsaps stabilize minus ends of the seeds to facilitate microtubule outgrowth, whereas Wdr47 concentrates Camsaps into the axonemal central lumen to properly position central microtubules. Wdr47 deficiency in mouse multicilia results in complete loss of CP, rotatory beat, and primary ciliary dyskinesia. Overexpression of Camsaps or their microtubule-binding regions induces central microtubules in Wdr47−/− ependymal cells but at the expense of low efficiency, abnormal numbers, and wrong location. Katanin levels and activity also impact the central microtubule number. We propose that Wdr47, Camsaps, and Katanin function together for the generation of non-centrosomal microtubule arrays in polarized subcellular compartments.

M icrotubules (MTs) are organized into different arrays as cytoskeletons of subcellular structures such as the spindle, neurites, and cilia and as tracks for MT-based molecular motors dynein and kinesin [1][2][3][4] . Although the centrosome functions as a major MT-organizing center (MTOC) in most animal cells, non-centrosomal MTs also widely exist and are organized into unique arrays in certain cell types or subcellular compartments [5][6][7] . For instance, non-centrosomal MTs are constructed into parallel arrays perpendicular to the tissue plane in epithelial cells by anchoring their minus ends to adapters at the apical membrane domain 8,9 . They also form unidirectional and bidirectional bundles respectively in axons and dendrites following neuronal polarization 1 . Recently, calmodulin-regulated spectrin-associated proteins (Camsaps) have been found to stabilize free MT minus ends through direct binding to facilitate the rapid growth of the MT plus ends [10][11][12][13][14] . They function in the organization of non-centrosomal MT arrays in epithelial and neuronal cells [15][16][17][18][19] .
The majority of motile cilia, e.g. those on the surface of mammalian trachea and ependyma or forming the tail of sperms, contain a "9 + 2"-type axoneme with a pair of non-centrosomal central (C1; C2) MTs surrounded by nine MT doublets extended from the basal body 20,21 . Such a central pair (CP) of MTs are positioned over the center of the transition zone, with their plus ends pointing to the ciliary tip [22][23][24] . They are interconnected by bridges constituted by proteins such as Spag16 and coated with distinct repetitive arrays of proteinous projections. The resultant supramolecular structure, the CP apparatus, contacts directly with radial spokes emanated from peripheral MT doublets to coordinate axonemal dynein activities [23][24][25] . Tracheal and ependymal cilia beat in a back-and-forth, or planar, manner. Their loss of CP or CP-associated proteins, such as the C1-associated Spag6 or C2-associated Hydin, results in an abnormal beat pattern and in humans contributes to primary ciliary dyskinesia (PCD) 3,21,[26][27][28][29][30] .
Centrosomal MTs are nucleated from and stabilized at their minus ends mainly by the γ-tubulin ring complex 2,31 . Usually, non-centrosomal MTs use MTs released from the centrosome as "seeds" 5,6 . Cilium, however, is a subcellular compartment gated by the transition zone 32,33 . While tubulin dimers are found to enter cilia through the intraflagellar transport (IFT) machinery, no evidence suggests that cytoplasmic MTs can pass the transition zone 34,35 . How the initial seeds of the central MTs are produced is thus an outstanding question. Furthermore, although we have previously reported that the mammalian CP formation requires Spef1-mediated MT stabilization in cilia 36 , how the number and the position of the central MTs are precisely controlled is still unknown.
In this work, we demonstrate that the WD40 repeat-containing Wdr47 (also called Nemitin) 15,37 , Camsaps, and the MT-severing enzyme Katanin 12,38-40 , co-operate to achieve the initial central MT formation in mammalian multicilia. Katanin produces central MT seeds by severing the peripheral MTs. Wdr47 concentrates Camsaps into the central lumen, where Camsaps stabilize the minus ends of the seeds to allow central MT elongation. As emerging evidence suggests that Wdr47, Camsaps, and Katanin also similarly impact neuronal polarization and axonal growth 15,16,18,41,42 , our findings indicate that the same pathway is used to properly generate and organize non-centrosomal MT arrays in polarized subcellular compartments such as the axon and motile cilium.

Results
Wdr47 is a CP-associated protein implicated in CP MT formation. Data mining in our cDNA array results 43 and subsequent immunoblotting indicated that Wdr47 (Fig. 1a) was upregulated during multiciliation of mouse tracheal epithelial cells (mTECs) cultured at an air-liquid interface (ALI) (Fig. 1b). Wdr47 was highly expressed in mouse tissues abundant in the 9 + 2 type of cilia, such as the lung, the testis, the ependyma, and the oviduct (Fig. 1c). Consistently, immunostaining of cultured mouse ependymal cells (mEPCs) revealed a ciliary-shaft localization of Wdr47 in multicilia (Fig. 1d). Interestingly, Wdr47 was highly concentrated in short cilia, especially at the ciliary tip. Its ciliary-tip intensities declined following ciliary elongation, accompanied with the emergence of Wdr47-positive puncta along and especially at the bottom of the ciliary shaft above ciliary transition zone (TZ), when Cep162, a protein at the bottom region of TZ 44 , was used as marker (Fig. 1d) 45 . By contrast, Wdr47 was not detected in the shaft of primary cilium (Supplementary Fig. 1).
Super-resolution microscopy revealed that the ciliary Wdr47 was distributed at the CP region when Hydin 23,28,29 was used as CP marker, with its bottommost distribution preceding that of Hydin (Fig. 1e). Interestingly, in short cilia decorated strongly with Wdr47, Hydin was sometimes weak or undetectable (Fig. 1e), suggesting that central MTs are either not yet formed or still in initial production stages. In such cilia, however, Wdr47 was still concentrated in the central lumen of axonemes in addition to its tip accumulation (Fig. 1e). When we pre-extracted the cells with Triton X-100 to remove soluble proteins prior to fixation, Wdr47 became undetectable in short cilia but still remained on CP in long cilia (Fig. 1e). Therefore, Wdr47 is a previously undocumented CP-associated protein preferentially enriched at the minus-end region of CP. More importantly, as CP MTs in Chlamydomonas are initially assembled at the ciliary distal region 46 , the strong enrichment of Wdr47 in short cilia lacking Hydin strongly implies a role in early stages of CP formation.
To clarify whether the rotatory pattern of the Wdr47 −/− multicilia was due to loss of the central MTs 26,36 , we examined the ciliary ultrastructure by transmission EM. While multicilia of Wdr47 +/+ or Wdr47 +/− mEPCs contained normal 9 + 2 axonemes, those of Wdr47 −/− mEPCs completely lacked both CP MTs (Fig. 1i). Sometimes electron-dense materials were observed in the central lumen of Wdr47 −/− cilia (Fig. 1i), similar to CP-free flagella of some Chlamydomonas mutants 47,48 . When ciliary cross-sections (n = 126) were classified into proximal ones, based on the presence of surrounding microvilli 49,50 , and distal ones, the luminal materials were observed in 71% of the distal sections and 23% of the proximal ones, possibly due to remnants of CP components. EM analyses indicated that Wdr47 −/− respiratory multicilia were also CP-less ( Supplementary Fig. 2). Wdr47 is thus essential to the CP MTs formation.
Wdr47 is critical for both ciliary localization of CP proteins and CP maintenance. We observed that the ciliary localizations of multiple CP proteins, including CP MTs-associated Spef1, C2 MTassociated Hydin, C1 MT-associated Spag6, and C1-C2 bridge-   Fig. 1 Wdr47 is an essential CP protein implicated in initial CP MT formation. a Diagram for mouse Wdr47. LisH, Lis1 homology; CTLH, carboxyl-terminal to LisH. b Upregulation of Wdr47 during multicilia formation. mTECs were cultured at an air-liquid interface (ALI) for the indicated days. Multicilia formation was indicated by the increased levels of Ift80, Bbs3, and acetylated tubulin (Ac-tub) from after ALI d3. Gapdh served as loading control (n = 2 biological replicates). c Wdr47 was abundant in mouse tissues containing motile cilia. The tissue lysates were prepared from 8-week-old mice (n = 2 biological replicates). d Wdr47 localized in multicilia as puncta. Cultured mEPCs were serum starved at day 0 to induce multiciliation and fixed at day 7, followed by immunostaining and confocal imaging. Ac-tub and Cep162 labeled the ciliary axoneme and the transition zone (TZ), respectively. Cilia pointed by arrows were magnified by 150% to show distribution changes of Wdr47 following the ciliary elongation. Arrowheads point to Wdr47 puncta at the base of cilia over the TZ. Diagrams are provided to aid comprehension (n = 5 biological replicates). e Wdr47 entered ciliary central lumen earlier than Hydin in short cilia and was enriched at the bottom region of CP (arrowheads) in long cilia. mEPCs at day 7 were fixed either directly (top panel) or after preextraction with Triton X-100 to remove soluble proteins (bottom panel), followed by immunostaining and three dimensional structured illumination microscopy (3D-SIM associated Spag16 (Fig. 2a and Supplementary Fig. 3) 23,36,51 , were abolished or markedly reduced in Wdr47 −/− mEPCs, whereas Rsph4a, a radial spoke component 49,52 , was not affected ( Fig. 2b and Supplementary Fig. 3). To further corroborate this, we purified multicilia from cultured Wdr47 +/+ and Wdr47 −/− mEPCs and performed label-free quantitative (LFQ) proteomic analysis 53 . In addition to Wdr47, 10 CP proteins were hit in the wild-type cilia. Their levels declined by at least 3.7-fold in the Wdr47 −/− cilia (Fig. 2c, d). By contrast, subunits of intraflagellar transport (IFT) complexes A and B 4 were either unchanged or only reduced by <2fold (Fig. 2c, d). Therefore, CP proteins require Wdr47 for their ciliary localization.
To further confirm the importance of Wdr47 in CP formation, we performed rescue experiments by expressing GFP-tagged Wdr47 or Centrin1 (negative control) into Wdr47 −/− mEPCs. GFP-Wdr47 entered the Wdr47 −/− multicilia and restored the ciliary localization of CP proteins ( Fig. 2e and Supplementary Fig. 4). By contrast, the CP proteins still displayed no or weak ciliary localization in the cells expressing Centrin1-GFP ( Fig. 2e and Supplementary Fig. 4). Live imaging revealed that multicilia in Wdr47 −/− mEPCs expressing Centrin1-GFP were still rotatory, whereas those in Wdr47 −/− mEPCs expressing GFP-Wdr47 displayed planar beat ( Fig. 2f and Supplementary Movie 2). These results further attribute the lack of CP formation in Wdr47 −/− multiciliated cells ( Fig. 1i and Supplementary Fig. 2) to the loss of Wdr47.
Next we investigated whether Wdr47 was also required for the stability of pre-existing CP. To do this, we tried to deplete Wdr47 after the CP formation by expressing Cre recombinase in cultured Wdr47 flox/flox mEPCs from day −4 or −1 and examined multicilia at day 5 or day 15 (Fig. 2g). The Cre expression at day −4 resulted in rotatory beat pattern and loss of ciliary Hydin even at day 5 ( Fig. 2h and Supplementary Fig. 5), suggesting that under the condition Wdr47 was depleted prior to the CP formation in the cells. By contrast, in mEPCs expressing Cre from day −1, Wdr47 was depleted after the CP formation because the incidence of rotatory beat increased dramatically from day 5 to day 15, accompanied with a similar extent of reduction in Hydinpositive multicilia ( Fig. 2h and Supplementary Fig. 5). Transmission EM confirmed a 3.5-fold increase of CP-less cilia from day 5 to day 15 (Fig. 2i). Therefore, the maintenance of CP also requires Wdr47.
Wdr47-deficient mice display PCD-like phenotypes hydrocephalus and sinusitis. Next, we investigated whether Wdr47 deficiency induced multicilia-related pathological disorders in mice. Wdr47-deficient mice die immediately after birth due to central nervous system defects 15 and were not suitable for the investigation. The Wdr47 Kof/Kof mice, however, are quite heterogeneous in viability: the majority of them die before P55 but the remaining mice live longer than P153 54 , probably due to variations in Wdr47 expression levels 15 . To maintain minimal levels of Wdr47 for mouse survival, we created Wdr47 Kof/− mice and found that they were viable after birth. Wdr47 Kof/− mice were considerably smaller in size than their wild-type littermates (Fig. 3a) and could only survive for <4 weeks. While Wdr47deficient neonatal mice 15 and Wdr47 Kof/Kof adult mice 54 only display slightly enlarged brain ventricles, the ventricles of Wdr47 Kof/− mice expanded markedly when examined at P14 (Fig. 3b), indicating hydrocephalus 21,28,55 . Moreover, massive amounts of mucus were observed bilaterally in the paranasal cavity of P14 Wdr47 Kof/− mice (Fig. 3b), indicating chronic sinusitis 55 . These phenotypes are hallmarks of primary ciliary dyskinesia (PCD) 21,26,56 .
To further clarify that the enlarged ventricles were not due to Wdr47 deficiency-induced neuronal death 54 , we removed the Kof cassette 15 and generated Wdr47 flox/flox ;GFAP-Cre conditional knockout (cKO) mice to specifically knockout Wdr47 in GFAPpositive glial cells (Fig. 3c) 57 , including EPC progenitors 58,59 . The cKO mice were born normally but developed dome-shaped heads within 3 weeks (Fig. 3d) and died at~4 weeks old. Coronal brain sections revealed dilated ventricles and magnetic resonance imaging further confirmed severe hydrocephalus of the mice (Fig. 3e). Scanning EM revealed that multicilia in individual ependymal cells tended to cluster together in Wdr47 flox/flox mice but became scattered in the cKO mice (Fig. 3f). Live imaging of brain slices revealed planar beat of the Wdr47 flox/flox ependymal cilia but rotatory beat of the cKO cilia (Fig. 3g). Wdr47 deficiency or insufficiency thus causes PCD.
Camsaps are CP-associated proteins colocalizing with Wdr47. How does Wdr47 impact the CP formation? As Wdr47 can be recruited to MTs by MT-binding proteins such as Map8 and Camsap3 15,37 , we speculated that it might function with certain MT-binding proteins for the nucleation or stabilization of central MT seeds. Interestingly, we noticed that all three mouse paralogues of Camsaps, Camsap1, Camsap2, and Camsap3, were readily identified from our purified wild-type multicilia samples by mass spectrometry (Fig. 4a). Camsap3 appeared to be the most abundant among the three Camsaps according to both their LFQ intensities and unique peptide counts (Fig. 4a). We have previously shown that all three Camsaps can associate with Wdr47. Moreover, Camsap1 and Camsap3 function as downstream effectors of Wdr47 in neuronal polarization 15 . Therefore, Camsaps might also function in multicilia with Wdr47.
Super-resolution imaging revealed that Camsaps were indeed CP-associated proteins with distribution patterns resembling those of Wdr47. In short cilia negative for Hydin, Camsaps were highly concentrated at the ciliary tip and also distributed in the axonemal central lumen (Fig. 4b). In long cilia they tended to concentrate in the bottom region of the central lumen complementary to Hydin (Fig. 4b). In addition, Camsap1 tended to display punctate distributions along the CP (Fig. 4b). Both endogenous and GFP-tagged Camsaps were distributed above TZ in ependymal cilia when Cep162 and Cep290 were used as TZ markers ( Fig. 4c) 44,60 . When mEPCs were pre-extracted with Triton X-100 to remove soluble proteins prior to fixation, the enrichment of Camsaps at the base of the central lumen became more obvious (Fig. 4c). Such a distribution pattern of Camsaps (Fig. 4b, c) resembles proteins localized at the CP minus end region, or the CP-foot 45 . Therefore, similar to Wdr47, Camsaps are preferentially enriched at the tip of nascent short cilia and the minus end region of central MTs in long cilia.
As our antibodies did not allow co-immunostaining with Wdr47, we examined GFP-Camsaps. In mEPCs expressing Centrin1-GFP, immunofluorescent signals of Wdr47 were independent of those of Centrin1-GFP (Fig. 4d), indicating no fluorescent bleed between the two channels. In contrast, GFP-Camsaps co-localized nicely with endogenous Wdr47 in multicilia (Fig. 4d), confirming their interplays with Wdr47 in ependymal cilia.
Although the levels of Camsaps in mEPCs were not affected by the Wdr47 deficiency (Fig. 4e), immunostaining confirmed that the ciliary localization of Camsaps was largely dependent on Wdr47 (Fig. 4f, g). Therefore, Camsaps could be downstream effectors of Wdr47 in the CP formation.

Overexpression of Camsaps induces central MT formation in
We have recently reported that overexpression of Camsap1 or −3 rescues the polarization defects of Wdr47-deficient neurons 15 . To understand whether excessive Camsaps could also functionally compensate for the Wdr47 Wdr47 is required for both ciliary localization of CP proteins and CP maintenance. a Diagram for detailed CP locations of the indicated proteins, referring to literature 23,36 . C1 central MT 1, C2 central MT 2. b Wdr47 deficiency markedly reduced the immunofluorescence of ciliary CP proteins. Wildtype or Wdr47 −/− mEPCs fixed at day 10 were immunostained, followed by confocal microscopy. Representative micrographs for Hydin and Spef1 and a summary are shown (n = 2 biological replicates). Ac-tub was used as ciliary marker. Rsph4a, a radial spoke protein, served as negative control. Refer to Supplementary Fig. 3 for the supporting images. c, d Wdr47 deficiency markedly reduced ciliary levels of CP proteins. Multicilia purified from day-10 wildtype or Wdr47 −/− mEPCs were used for label-free quantitative (LFQ) mass spectrometry (c) or immunoblotting (d). The relative ciliary intensities were from two biologically independent experiments, in which the subunits of intraflagellar transport (IFT) complexes served as negative controls. In the immunoblots (d), Ift52 served as loading control, whereas Ac-tub and Gapdh were respectively used as positive and negative controls to indicate the quality of the cilia samples (n = 2 biological replicates). e Expression of GFP-Wdr47 in Wdr47 −/− mEPCs restored ciliary localization of CP proteins. mEPC progenitors from Wdr47 −/− E18.5 embryos were infected with lentivirus at one day before serum starvation (day −1) to express GFP-Wdr47 or Centrin1-GFP (negative control). At day 10, the cells were fixed for confocal microscopy. Refer to Supplementary Fig. 4 for additional supporting images (n = 3 biological replicates). f Expression of GFP-Wdr47 in Wdr47 −/− mEPCs restored the planar beat of multicilia. Wdr47 −/− mEPCs infected as in e were live imaged at day 10 to monitor cilia motility (n = 4 biological replicates). Trajectories of four cilia during the first 56 ms of imaging are shown for each representative mEPC. The corresponding ciliary beat patterns and statistical results are also illustrated. Please also refer to Supplementary Movie 2. g Experimental scheme for CP maintenance assays. mEPC progenitors from Wdr47 flox/flox neonatal littermates were infected with adenovirus to express Cre and examined at day 5 and day 15. h, i The Cre expression from day −1 induced progressive CP disassembly. Multiciliated mEPCs treated as in g were live imaged for ciliary motility, followed by immunostaining for Hydin and transmission EM. At least 30 cells in the live imaging (h), 50 cells in the confocal imaging (h), and 50 ciliary cross-sections in the EM (i) were scored in each experiment and condition. Three (h) and two (i) biologically independent experiments were performed. Refer to Supplementary Fig. 5 for representative confocal images. Quantification results are presented as mean ± s.d. Twosided Student's t test: ns no significance; ***P < 0.001. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26058-5 ARTICLE NATURE COMMUNICATIONS | (2021) 12:5796 | https://doi.org/10.1038/s41467-021-26058-5 | www.nature.com/naturecommunications deficiency in mEPCs, we overexpressed GFP-tagged Camsaps or Centrin1 (negative control) in Wdr47 −/− mEPCs through lentiviral infection (Fig. 5a). Wdr47 −/− mEPCs expressing Centrin1-GFP still displayed rotatory multicilia, whereas the majority of the GFP-Camsap1-expressing cells displayed planar ciliary beat (Fig. 5b, c and Supplementary Movie 3). Different cilia, however, tended to beat towards different directions ( Fig. 5b and Supplementary Movie 3). The GFP-Camsap1 overexpression also increased the percentage of cells with mixed ciliary beat patterns by 3.1-fold (Fig. 5b, c). Although GFP-Camsap2 and GFP-Camsap3 were expressed in low levels ( Fig. 5a), percentages of mEPCs with planar and mixed beat patterns still increased by 6.5fold and 10.9-fold, respectively, as compared to the Centrin1-GFP-positive cells (Fig. 5b, c). When fixed cells were examined, GFP-Camsaps markedly rescued the ciliary Hydin localization as compared to Centrin1-GFP (Fig. 5d, e). These results suggest a partial rescue of CP by Camsaps overexpression.
Next, we examined the ciliary ultrastructure. As we were unable to recognize GFP-positive cells in EM, we extensively examined the GFP-Camsap1 samples due to their high viral infection efficiency (>80%), exogenous expression levels ( Fig. 5a), and rescue effects on the planar beat pattern and ciliary Hydin localization (Fig. 5b-e). 48.9% of their axonemal cross-sections displayed two central MTs; 3.5% displayed abnormal central MT numbers (1 or 3-to-6) (Fig. 5f, g). Occasionally we observed one dislocated central MT outside the central lumen (n = 2), (Fig. 5f) or extra doublet-like MTs (n = 3) (Supplementary Fig. 6a). Central MTs were observed to extend from the ciliary base to the tip in longitudinal sections (Fig. 5f). We also observed multiple short MTs at the tip (Fig. 5f), suggesting that the incidence of axonemes with extra central MTs (2.6%; Fig. 5g) is probably underestimated due to quantifications on cross-sections. By contrast, none of the axonemal cross-sections (total n = 307) from the Centrin1-GFP samples displayed visible central MT(s). We also examined a set of samples for GFP-Camsap2 and GFP-Camsap3. In all, 27% of axonemal cross-sections from the Camsap3 sample contained either one (1/173) or two (46/173) central MTs (Fig. 5g). All cross-sections from the Camsap2 sample (n = 239), however, were CP-less, possibly due to its low expression level and efficacy because EM does not allow choosing cells expressing relative high levels of GFP-Camsap2. b Wdr47 Kof/− mice displayed both hydrocephalus and chronic sinusitis (n = 3 biological replicates). Shown are representative HE-stained coronal sections of the brain and the paranasal cavity. Arrows and arrowheads indicate enlarged ventricles and accumulated mucus, respectively. c Strategies for generating Wdr47 flox/flox ;GFAP-Cre conditional knockout (cKO) mice to knockout Wdr47 in progenitors of mEPCs. Wdr47 Kof/+ mice were crossed with Flp mice to produce Wdr47 flox/+ mice, which were then crossed with GFAP-Cre mice to produce Wdr47 flox/+ ;GFAP-Cre mice. d The Wdr47 cKO mice displayed a dome-shaped skull (arrow) (n = 3 biological replicates). e The cKO mice displayed progressive hydrocephalus (arrows) as shown by coronal brain sections at P10 and transverse magnetic resonance images at P21. Experiment was performed once. f, g Multicilia in the Wdr47 flox/+ ;GFAP-Cre ependyma were disorganized under scanning EM and rotatory in live imaging, acquired at 140 frames per second (fps) using live brain slices of P10 mice. Scanning EM was performed twice. Live imaging was carried out once. Trajectories of four traceable cilia over time are shown on representative image sequences.
Taken together, we conclude that increasing the total levels of Camsaps partially compensate for Wdr47 deficiency in the central MT formation.
Wdr47 recruits Camsaps to the central lumen for proper CP formation. As the partial rescue effects of overexpressed Camsaps ( Fig. 5a-g) suggested a role of Wdr47 for efficient production and proper positioning of central MTs, we speculated that Wdr47 might function by increasing the regional ciliary concentration of Camsaps. We have previously shown that Wdr47 interacts with Camsaps through its N-terminal region (WdrN) but not the C-terminal WD40 repeats (WdrC) (Fig. 5h) 15 . When expressed in Wdr47 −/− mEPCs, GFP-Wdr47 entered multicilia and rescued the ciliary localizations of endogenous Camsap1 and Hydin, the CP marker ( Fig. 5i; also see Fig. 2e). GFP-WdrC, but not GFP-WdrN, strongly localized into multicilia (Fig. 5i). Neither construct, however, was able to function as the full-length to restore ciliary Camsap1 and Hydin (Fig. 5i) or the planar ciliary beat (Fig. 5j vs.  Fig. 2f). Wdr47 thus requires both its Camsap-interacting and cilialocalization regions for the ciliary enrichment of endogenous Camsaps and, consequently, proper CP formation.
We further examined the detailed localization of GFP-WdrC through super-resolution microscopy and found that GFP-WdrC was distributed in the ciliary central lumen (Fig. 5k). As Wdr47 −/− multicilia are CP-less (Fig. 1i), such a result suggests that this C-terminal region can target Wdr47 to the central lumen independently of central MTs. Taken together, we conclude that Wdr47 binds to Camsaps through its N-terminal region and targets them to the ciliary central lumen through its C-terminal region for efficient and proper CP formation.   Fig. 4 Camsaps colocalize with Wdr47 in multicilia. a Camsaps were detected in purified wild-type multicilia by mass spectrometry. LFQ intensities (relative to that of Camsap3) were from one of the two independent experiments described in Fig. 2c. Unique peptide counts from both experiments are listed. b Camsaps localized to ciliary tip and central lumen prior to Hydin (CP marker) in short cilia and preferentially to the bottom region of CP (arrowheads) in long cilia. Ac-tub labels axonemes. Dashed lines mark the base of ciliary shafts based on the immunostaining of Ac-tub. Representative 3D-SIM images of cilia were from mEPCs at day 7. n = 4 biological replicates. c Representative cilia showing that Camsaps-decorated CP base regions were above TZ. Cultured mEPCs were immunostained for Ac-tubulin, Cep162, and one of Casmsaps, followed by confocal microscopy (top panels). mEPCs infected with lentiviral particles to express GFP-tagged Camsaps were extracted with 0.5% Triton X-100 for 30 s prior to fixation to remove soluble protein fractions. The samples were then immunostained, followed 3D-SIM (bottom panels). Dashed lines mark the position of TZ based on the immunostaining of Cep162 (top panels) or Cep290 (bottom panels). n = 3 biological replicates. An illustration is provided to aid understanding. d Colocalization of ciliary Camsaps with Wdr47. Cultured mEPCs were infected with lentiviral particles at 1 day before serum starvation (day −1) to respectively express GFP-Camsaps and fixed at day 7, followed by immunostaining and confocal imaging. Arrows point to representative short cilia. Centrin1-GFP served as negative control. Arrowheads denote representative cilia that indicate different localization patterns of Wdr47 and Centrin1-GFP. n = 4 biological replicates. e-g Wdr47 deficiency impaired the ciliary localization of Camsaps. Cultured Wdr47 +/+ or Wdr47 −/− mEPCs at day 10 were lysed for immunoblotting (e) (n = 3 biological replicates) or fixed for immunostaining (f). The framed areas were magnified to show details. Relative intensities (g) were obtained by normalizing fluorescent intensities of Camsaps to those of Actub in corresponding cilia. At least 101 cilia were quantified in each condition from three biological independent experiments. The bars and errors represent mean and s.d., respectively. Two-sided Student's t test: ***P < 0.001. Camsaps promote the central MT formation by stabilizing the MT minus end. We have previously shown that Camsap3 can recruit Wdr47 to MT minus ends 15 , implying that Wdr47 might target Camsaps by binding to regions outside their MT-binding domains. To clarify this, we mapped the Wdr47-interacting region of Camsap1 by co-immunoprecipitation. Indeed, the linker region between the CH and the first coiled coil (CC) domains of Camsap1 (312-858 aa) strongly interacted with Wdr47, whereas the CC-containing region and the CKK domain involved in the MT association 10-13 were dispensable (Fig. 6a, b). Camsaps function in MT dynamics by stabilizing the minus end of pre-existing MTs 10,12,13 . It is known that, although the CKK domain alone is sufficient for binding to the MT minus end, a longer fragment containing the CC region displays markedly increased lattice-binding and minus end-stabilization ability [10][11][12][13] . To understand whether Camsaps used the similar mechanism to induce central MT formation, we overexpressed constructs containing the CKK domain and a longer CC-containing fragment (LC) as GFP fusion proteins in Wdr47 −/− mEPCs (Fig. 6c, d). Both types of constructs displayed significant rescue effects on ciliary beat patterns as compared to Centrin1-GFP ( Fig. 6e and Supplementary Movie 4), suggesting that the CKK domain alone is already functional. Consistently, all the constructs displayed ciliary localizations as well as an ability to rescue the ciliary localization of Hydin in the Wdr47 −/− mEPCs ( Fig. 6f and Supplementary Fig. 7). The LC constructs generally yielded higher percentages of ciliary Hydin-positive mEPCs than the CKK constructs ( Fig. 6f and Supplementary Fig. 7).
EM further confirmed that all the constructs of Camsaps induced central MT formation in the Wdr47 −/− multicilia (Fig. 6g, h). Notably, 70% of the axonemal cross-sections from the GFP-Ca1LC samples contained one or more central MTs and 21% contained >2 (up to 13) central MTs (Fig. 6g, h). Furthermore, 28 sections (8%) contained dislocated central MTs outside the central lumen (Fig. 6h). By contrast, 25% of the cross-sections from the GFP-Ca1CKK samples contained central MTs (Fig. 6g). Among them, three sections (2%) contained dislocated central MT(s). In addition, axonemes with an extra MT doublet were also observed for both constructs ( Supplementary Fig. 6b, c). We observed both long and short central MTs in longitudinal axonemal sections ( Supplementary  Fig. 6d). In comparison, 38% and 44% of the cross-sections respectively from the GFP-Ca2LC and GFP-Ca3LC samples contained one or more central MTs, with 3% and 18% of the total cross-sections containing >2 (up to 10) central MTs (Fig. 6g, h). The incidences of central MT-containing cross-sections from the GFP-Ca2CKK and GFP-Ca3CKK samples were 15% and 38%, respectively (Fig. 6g).
These results further strongly support a direct role of Camsaps in the CP formation by binding and stabilizing the minus ends of central MTs. The prominent rescue effects of GFP-Ca2LC and GFP-Ca2CKK (Fig. 6e-g), despite their low expression levels relative to the corresponding constructs of Camsap1 and Camsap3 (Fig. 6d), also confirm that Camsap2 is involved in the CP formation. Furthermore, as Camsaps are unable to nucleate MTs [10][11][12][13] , the formation of extra and even 13 central MTs indicates that the available central MT seeds in an axoneme can largely exceed two.
Central MT formation requires the MT-severing activity of Katanin. The seeds for non-centrosomal MTs usually come from severed MTs or MTs released from the γ-tubulin ring complex 5,6,12,13 . As the MT-severing enzyme Katanin directly binds to Camsap2 and −3 12,40 and protozoan Katanin is essential for the CP formation and has been shown to bind to and severing axonemal MT doublets 61-63 , we speculated that Katanin might function to generate the central MT seeds for mammalian multicilia. Immunostaining revealed that Katanin p60 displayed multiciliary localizations in mEPCs, abundant at the tip of short cilia (Fig. 7a). GFP-p60 also displayed similar ciliary localizations and co-localized with Camsaps at the ciliary tip, when Camsap1 was used as marker (Fig. 7b).
Next we explored whether the rescue effect of Camsaps in Wdr47 −/− mEPCs required Katanin activity by co-expressing RFP-p60 K257A or RFP-p60. We initially performed the experiments with GFP-Camsap1 because it displayed the strongest rescue effect (Fig. 5a-g). Moreover, as Camsap1 does not interact with Katanin 1 , its overexpression would not influence the effects of RPF-p60 and RFP-p60 K257A . Compared to Centrin1-RFP, coexpression of RFP-p60 K257A with GFP-Camsap1 induced a 3.3fold decrease in percentage of cells with planar ciliary beat and a 2.8-fold increase in that with rotatory ciliary beat ( Fig. 7g and Supplementary Movie 6). In comparison, co-expression of RFP-p60 did not significantly alter the rescue effects of GFP-Camsap1 in Wdr47-deficient mEPCs ( Fig. 7g and Supplementary Movie 6). We then examined GFP-Ca3LC, the Camsap3 deletion mutant displaying a remarkable rescue effect (Fig. 6d-h) and lacking Katanin-binding region (Fig. 6c) 40 . Compared to Centrin1-RFP, co-expression of RFP-p60 K257A also similarly compromised the rescue efficiency of GFP-Ca3LC (Fig. 7g). Therefore, Katanin activity is important for the Camsap overexpression-induced central MT formation.
Therefore, like protozoan Katanin 61-63 , mammalian Katanin also functions in the CP formation. As tubulin dimers can be transported into cilia through intraflagellar transport 35 , we modified the model by Sharma and colleagues 63 and propose that Katanin provides central MT seeds by severing the tip region of axonemal MT doublets (Fig. 7h).

Discussion
Our results indicate that Wdr47 ensures the efficient formation and accurate positioning of central MTs by concentrating Camsaps into cilia. Wdr47 is a prerequisite because CP completely failed to form upon Wdr47 deficiency (Figs. 1 and 2, and Supplementary Figs. 2  and 3). Ciliary Camsaps co-localized with Wdr47 mainly at the tip of nascent short cilia and the base of central lumen in long cilia and became markedly reduced in Wdr47-deficient mEPCs (Figs. 1d, e  and 4b-g). Although central MTs can form in Wdr47 −/− mEPC when the total levels of Camsaps were increased through Wdr47-/-C a 1 L C C a 2 L C C a 3 L C C a 1 C K K C a 2 C K K C a 3 C K K  overexpression (Figs. 5a-g and 6, and Supplementary Fig. 7), they emerged with low efficiency and abnormal numbers, and sometimes at a wrong location: comparing to a near 100% efficiency of the CP formation in wild-type ependymal and airway multiciliated cells (Fig. 1i and Supplementary Fig. 2) 28,30,65 , only <50% of ependymal ciliary cross-sections were rescued to two central MTs and at least 30% were still CP-less (Figs. 5f, g and 6g, h). Furthermore, up to 21% contained more than two central MTs, and dislocated singlet MT(s) outside the central lumen were also observed (Figs. 5f, g and 6g, h, and Supplementary Fig. 6d). As Camsaps do not induce MT nucleation [10][11][12][13] , the emergence of excessive numbers (up to 13 in our EM results) of central MTs (Figs. 5f, g and 6g, h) indicate that Camsaps function by stabilizing the minus ends of pre-existing MT seeds whose number can largely exceed two in a cilium. Our interaction and localization analyses (Figs. 5i, k and 6a, b) further suggest that Wdr47 binds to the N2 region of Camsaps through its N-terminal region and recruits them to the ciliary central lumen through its C-terminal region.
Therefore, the Wdr47-Camsap interaction allows Camsaps to function at low physiological concentrations to avoid production of extra central MTs. How a cilium manages to produce precisely two central MTs, however, still requires future quantitative studies.
Camsap1-3 appear to function redundantly and contribute collectively to the central MT formation. Different Camsaps share common and also display individual properties [10][11][12][13][14] . As mEPCs express all three proteins concomitantly (Fig. 4), the contribution of each Camsap may be affected by both expression level and efficacy relative to the others (Figs. 5 and 6). Although multiciliarelated defects have not been documented for Camsap1-or Camsap2-deficient mice 14,18 , during the revision process we noticed that a recent publication reports that Camsap3 localizes to the base of axonemes in murine nasal multiciliated cells and Camsap3 deficiency results in the CP loss in 77.4% axonemal cross-sections of the cells 66 . Consistently, Camsap3-deficient mice display severe nasal airway blockage 66 , similar to our Wdr47 Kof/mice (Fig. 3b). In sharp contrast to the severe hydrocephalus of  (Fig. 3c-e), Camsap3deficient mice display only mild or no hydrocephalus phenotypes 66 . Two subsequent publications report that Camsap3deficient mice have normal beating cilia in ependymal cells 67 and normal CP formation in oviduct multicilia 68 . These results also echo our proposed redundant and collective effects of Camsaps. Future systematic analyses using single, double, and triple gene knockout mice of Camsaps will be still required to determine detailed contributions of individual Camsaps and their collective effects on ciliary CP formation of different tissue cells.
Our results suggest that Katanin is involved in the production of central MT seeds. p60 and p80 subunits of Katanin are in fact the only two proteins known to be essential to the CP formation in protozoa 48,[61][62][63] . Chlamydomonas p80 specifically distributes in ciliary outer doublet compartment 61 , whereas overexpressed GFP-p60 has been shown to specifically bind to and sever outer doublets in Tetrahymena 63 . Protozoan Katanin has also been proposed to provide central MT seeds by severing MTs nucleated from intraflagellar γ-tubulin 62 or free tubulin dimers for CP assembly by severing outer doublets 63 . We found that both endogenous p60 and GFP-p60 display ciliary localization and also enrich at the tip of short cilia (Fig. 7a, b). Katanin activity is also important for Camsaps-mediated CP formation (Fig. 7e-g). As overexpression of GFP-p60 induced CP loss in a portion of ependymal axonemes (Fig. 7e, f), mammalian Katanin unlikely functions by providing tubulin dimers for the CP formation. On the other hand, no evidence to date suggests a localization of γtubulin in mammalian multicilia [69][70][71] . As the observations of up to 13 central MTs in our rescue experiments (Figs. 5f, g and 6g, h) provide solid evidence for the presence of excessive central MT seeds and nascent central MTs have been shown to emerge initially from the top region of Chlamydomonas flagella 46 , we modified the previous models and propose that mammalian Katanin severs the tip of axonemal outer MTs to generate central MT seeds for Camsaps to bind and stabilize at early stages of ciliogenesis (Fig. 7h). Consistently, we observed doublet-like MTs additional to the nine outer doublets upon the overexpression of Camsap1 or its deletion constructs (Supplementary Fig. 6a-c). Future studies will be required to verify whether Katanin p60/p80 enters multicilia by binding to Camsap2 and Camsap3 12,40 . In addition, two paralogues of Katanin p60, Katnal1 and Katnal2, also affect the growth and motility of multicilia 38,72,73 . Katnal1 has also been shown to sever MTs to build the dense MT arrays in Drosophila mechanosensory cilia 38,[72][73][74] . As the overexpress of p60 K257A did not completely abolish the CP formation in mEPCs (Fig. 7e, f), whether Katnal1 and Katnal2 have a role in metazoan central MT seeds production also needs to be clarified in the future.
Taken together, we propose a model that Wdr47, Camsaps, and Katanin function together to produce the CP of mammalian multicilia: Katanin severs the tip of outer MTs to generate central MT seeds in nascent short cilia, Camsaps stabilize the seeds by binding to their minus ends, and Wdr47 concentrates Camsaps and facilitates their targeting into ciliary central lumen so that the Camsaps-bound MT seeds eventually develop into CP following the elongation of cilia (Fig. 7h). In addition to the CP formation, emerging clues of their corporative actions are reported in neurons 15,16,18,41,42,75 and epithelial cells 17,19,76 . Other noncentrosomal MT arrays requiring Camsaps 5-7 could involve Katanin and Wdr47 as well. Both Katanin and Wdr47 are expected to bind between the CH domain and the first CC region of Camsap2 and -3 (Fig. 6a, b) 40 . Therefore, these three groups of proteins might similarly constitute a team for the formation of non-centrosomal MT arrays in polarized subcellular compartments. It will thus be interesting to clarify their detailed interplays and consequences of the interactions as well in other types of cells.

Methods
Plasmids. The full-length or partial cDNAs for mouse Wdr47 (NM_181400), mouse Camsap1 (XM_006497897), mouse Camsap2 (NM_001347109), and mouse Katanin p60 (NM_011835) were amplified by PCR from total cDNAs from mouse testis, brain or mTECs. The full-length of mouse Camsap3 (NM_027171) was amplified by PCR from the GFP-Nezha plasmid (kindly provided by Dr. Wenxiang Meng, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences) 77 . To express GFP-or RFP-fusion proteins, the cDNA fragments were constructed into the lentiviral expression vector, pLV-EGFP-C1 or pLV-RFP-C1 78   Pooled data (mean ± s.d.) were from three biologically independent experiments. Student's t-test, ns, no significance; **P < 0.01; ***P < 0.001. a, b Katanin p60 displayed multiciliary localizations. Wild-type mEPCs were fixed at day 7 either directly for visualizing endogenous p60 through immunostaining (a) or after infection with lentivirus at day −1 to express GFP-p60 (b). Ac-tub and Hydin served as ciliary and CP markers, respectively. Arrowheads and arrows point to representative short cilia and long cilia, respectively. n = 3 biological replicates. c Expression level of GFP-p60 and GFP-p60 K257A in mEPCs. Wild-type mEPCs were infected with lentivirus at day −1 and day 2 to express Centrin1-GFP, GFP-p60 or GFP-p60 K257A and harvested at day 10 for assays in c-f. Gapdh served as loading control in immunoblotting (c). Note that the expression of GFP-tagged p60 or p60 K257A had little influence on the levels of endogenous p60, Wdr47, and Camsap1, as compared to Centrin1-GFP. subcloned into the pcDNA3.1-NFLAG vector. The full-length cDNAs of mouse Wdr47 or mouse Camsap1 (1073-1382 aa) were PCR amplified and constructed into pET28a to express His-tagged antigens for antibody production and into pGEX4T-1 to express GST-fusion proteins for antibody purification. All the primers were list in the Supplementary Table 1. All the constructs were verified by sequencing.
Mice. Mice experiments were performed in accordance with the ethical guidelines of Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences, and approved by the Institutional Animal Care and Use Committee.
Wdr47 Kof/+ mice (Wdr47 tm1a(EUCOMM)Wtsi ) were purchased from Wellcome Trust Sanger Institute 79 . Wdr47 +/− and Wdr47 flox/+ mice were generated by crossing the Wdr47 Kof/+ mice with Ella-Cre or FLP mice (Model Animal Research Center of Nanjing University, China), respectively. To knockout Wdr47 in the GFAP-positive glial cells, Wdr47 flox/flox mice were crossed with GFAP-Cre mice (a gift from Dr. Leping Cheng, Institute of Neuroscience, Chinese Academy of Sciences). Primers used for genotyping are listed in Supplementary Table 1.
To express exogenous proteins, HEK 293T were transfected with polyethlenimine (PEI, Polysciences, 23966-2), respectively, for 48 h. Lentiviral production and infection were performed as described previously 78 . Cultured mEPCs were infected at day one before serum starvation (day −1) unless otherwise stated. For rescue experiments and expressing GFP-tagged Camsaps, and their deletion constructs in Wdr47 −/− ependymal cells, mEPCs derived from Wdr47 −/− mice were infected with lentivirus at day −1, day 2, and day 5 and assayed at day 10.
Purification of ependymal cilia and LFQ mass spectrometric analyses were carried out as described 53 . To purify ependymal cilia, two 75-cm 2 flasks of ependymal cells derived from 10 E18.5 Wdr47 +/+ or Wdr47 −/− embryos at the day 10 post serum starvation were harvested by centrifugation at 2000 rpm for 5 min at 4°C. The cells were resuspended in 2.4 ml of deciliation buffer (20 mM PIPES, 250 mM sucrose, 20 mM CaCl 2 , 0.05% Triton X-100, pH 5.5) and agitated vigorously for 10 min on a vortex mixer (G560OE, Scientific Industries). After centrifugation at 600 × g for 5 min at 4°C, the cilia-containing supernatant was collected by centrifugation at 20,000 × g for 30 min. The pelleted cilia were washed twice with 1 ml of PBS, followed by centrifugation at 20,000 × g for 15 min. The isolated cilia were lysed in 100 μl of lysis buffer [20 mM Tris-HCl (pH 7.5), 100 mM KCl, 0.1% NP-40, 1 mM EDTA, 10 mM Na 4 O 7 P 2 , and protease inhibitors] and boiled at 100°C for 10 min. 50 μl of the samples was used for Mass spectrometry analysis and another 50 μl was used for immunoblotting (Fig. 2d). The full scan blots are provided in the Source Data file.
Label-free quantitative mass spectrometry. In all, 50 μL of lysed wild type and Wdr47 −/− cilia samples were precipitated with acetone. The protein pellet was dried by using a Speedvac for 1−2 min. The pellet was subsequently dissolved in 8 M urea, 100 mM Tris-HCl, pH 8.5. TCEP (final concentration is 5 mM) (Thermo Scientific) and iodoacetamide (final concentration is 10 mM) (Sigma) for reduction and alkylation were added to the solution and incubated at room temperature for 30 min, respectively. The protein mixture was diluted four times and digested overnight with Trypsin at 1:50 (w/w) (Promega). The tryptic-digested peptide solution was desalted using a MonoSpin TM C18 column (GL Science, Tokyo, Japan) and dried with a SpeedVac.
The peptide mixture was analyzed by a home-made 30-cm-long pulled-tip analytical column (75 μm ID packed with ReproSil-Pur C18-AQ 1.9 μm resin, Dr. Maisch GmbH), the column was then placed in-line with an Easy-nLC 1200 nano HPLC (Thermo Scientific, San Jose, CA) for mass spectrometry analysis. The analytical column temperature was set at 55°C during the experiments. The mobile phase and elution gradient used for peptide separation were as follows: 0.1% formic acid in water as buffer A and 0.1% formic acid in 80% acetonitrile as buffer B, 0-1 min, 3-8% buffer B; 1-301 min, 8-25% buffer B; 301-339 min, 25-50% buffer B, 339-340 min, 50-100% buffer B, 340-360 min, 100% buffer B. The flow rate was set as 300 nl/min. Data-dependent tandem mass spectrometry (MS/MS) analysis was performed with a Q Exactive Orbitrap mass spectrometer (Thermo Scientific, San Jose, CA). A cycle of one full-scan MS spectrum (m/z 300-1800) was acquired followed by top 20 MS/MS events, sequentially generated on the first to the twentieth most intense ions selected from the full MS spectrum at a 28% normalized collision energy. Full scan resolution was set to 70,000 with automated gain control (AGC) target of 3e 6 . MS/MS scan resolution was set to 17,500 with isolation window of 1.8 m/z and AGC target of 1e 5 . The number of microscans was one for both MS and MS/MS scans and the maximum ion injection time was 50 and 100 ms, respectively. The dynamic exclusion settings used were as follows: charge exclusion, 1 and >8; exclude isotopes, on; and exclusion duration, 15 s. MS scan functions and LC solvent gradients were controlled by the Xcalibur data system (Thermo Scientific).
MS/MS data were processed using Maxquant software version V1.6.10.43. MS/ MS spectra were searched by the Andromeda search engine against the SwissProt Mouse database at a false discovery cutoff ≤1%. Data were searched at 20 ppm mass tolerances for precursor ions for mass calibration and six amino acids were required as the minimum peptide. LFQ intensity was used as relative quantification of protein.
Light microscopy. mEPCs grown on fibronectin-coated 29-mm glass-bottom dishes (Cellvis, D29-14-1.5-N) were fixed with 4% paraformaldehyde in PBS for 10 min at room temperature and permeabilization with 0.5% Triton X-100 for 15 min. In Figs. 1e and 4c, mEPCs were pre-extracted with 0.5% Triton X-100 for 30 s before fixation. GFP signals in mEPCs were visualized by immunostaining using anti-GFP antibody. All the antibodies used are listed in Supplementary  Table 2.
Confocal images were captured by using Leica TCS SP8 system with a ×63/1.40 oil immersion objective and Z-stack images were obtained with maximum intensity projections. 3D-SIM images were captured with Delta Vision OMX SR imaging system (GE Healthcare) equipped with a Plan Apo ×60/1.42 NA oil-immersion objective lens (Olympus). Serial Z-stack sectioning was performed at 125-nm intervals. Images were processed with SoftWoRx software.
Ciliary motilities were recorded at 140 fps (frames per second) by using an Andor Neo sCMOS camera on Olympus IX71 microscope with a ×63/1.40 oil immersion objective 43 . To track fluid flows driven by ciliary beat 82 , fluorescent beads (Fluoresbrite PolyFluor 570 microspheres, Polysciences, 24061-10) were added at 1:200 dilution to the culture medium and imaged at 20 fps for 30 s with a ×63/1.40 oil immersion objective on an Olympus IX81 equipped with Hamamatsu EMCCD camera. Images were processed with ImageJ and Adobe Photoshop CS6.
Electron microscopy. For transmission EM, mouse trachea or cultured mEPCs were fixed in 2.5% glutaraldehyde overnight at 4°C, washed with PBS, and treated with 1% OsO4 for 30 min at room temperature. The samples were dehydrated with graded ethanol series and embedded in Epon 812 resin. In total, 70-nm ultrathin sections were stained with 1% lead citrate and 2% uranyl acetate. Images were captured at 80 KV using a Tecnai G2 Spirit transmission electron microscope (FEI, Hillsboro, OR).
For scanning EM, the cortex of P10 mouse brains was fixed in 2.5% glutaraldehyde and 2% paraformaldehyde overnight at 4°C and dehydrated with a graded ethanol series. Critical point drying was carried out before metal shadowing. The images were taken with a FEI Quanta 250 scanning electron microscope.
Magnetic resonance imaging, tissue section and H&E staining. Mice were anesthetized with isoflurane. T2 weighted spin echo images of the head were acquired by using BioSpec 70/30 USR (Bruker). The imaging parameters were: slice thickness 0.5 mm; time of repetition (TR) 2777.2 ms; and time of echo (TE) 34 ms.
P10 mouse brains were dissected and sectioned into 250-μm-thick sagittal slices using a Leica VT 1000S vibratome. Images were captured with an Olympus SZX16 Stereo Microscope.
For H&E staining, P14 mouse brains and noses were paraffin-embedded and sectioned into 5-μm thick slices with an RM 2235 microtome (Leica). The sections were deparaffinized with xylene, rehydrated, and stained with hematoxylin for 5 min and eosin for 1 min. The tissue images were captured with an Olympus BX51 microscope.
Quantification and statistics. Microscopic and biochemical results were repeated at least twice. Quantification results are presented as mean ± s.d. unless otherwise stated. Two-sided Student's t test (GraphPad Prism software) was used to calculate P-values between unpaired samples. Differences were considered significant when P < 0.05. Only results from three or more independent experiments were applied to the t-tests.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Raw data of label-free quantitative (LFQ) proteomic analysis results (Fig. 2c) have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomex change.org) via the iProX partner repository 83 with the accession code (PXD028219). Source data are provided with this paper. Any remaining data that support the results of this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.