To investigate the role of tubulin deacetylation in cilia assembly and maintenance, we generated mutants in two HDACs which contain Zn²⁺ binding domains: Hdac6 and the related Hdac10 (both class 2b), and in one NADH-dependent deacetylase: Sirt2. Mutations were generated using TALEN and CRISPR/Cas9 mutagenesis in zebrafish (Danio rerio) (Zu et al., 2013); Heler et al., 2014; Sander and Joung, 2014). By targeting hdac6 with a TALEN, we obtained 3 mutant alleles containing deletions of 2, 8, and 14 bp. The 14 bp deletion allele (Figures 1A,B) was used for further analysis. This results in a frameshift and a premature stop codon at amino acid position 95 of 1081 (Figures 1B,C). Targeting the hdac10 gene with CRISPR resulted in one mutant allele harboring an 11 bp deletion. This mutation also causes a frameshift and the appearance of a premature stop codon at amino acid position 147 of 676 (Figures 1B,C). Lastly, we used CRISPR to generate several Sirt2 mutant alleles, including a 2 bp deletion, and a 2 bp deletion combined with a 24 bp insertion. The second allele, which results in a frameshift and a premature stop codon at amino acid position 46 of 379 (Figures 1B,C), was used in further studies. Although all three mutations are predicted to delete significant portions of the protein and conserved domains, homozygotes for these mutations do not display any obvious external phenotype, are viable, and fertile (Figure 1A). No overt defects were observed in homozygous progeny of homozygous in-crosses, excluding the possibility of maternal contribution of wild type mRNA masking any role in early development. To investigate whether these mutations lead to loss of function, we analyzed the levels of the relevant mRNA in 5 dpf larvae. This confirmed that hdac10 and sirt2 mutations lead to significant reductions in mRNA levels, indicating nonsense-mediated decay (Figure 1E and Supplementary Figure 1). However, the hdac6 mutation significantly elevated the level of its mRNA in single mutants, and this pattern was maintained in hdac6; hdac10 and hdac6; sirt2 double mutants, and hdac6; hdac10; sirt2 triple mutant (hereafter named the Triple mutant) (Figure 1E and Supplementary Figure 1). Hdac6 might therefore negatively regulate its own expression at the transcriptional level. We did not detect cross-compensatory effects between these genes at the expression level.

Since Hdac6 and Hdac10 possess highly similar catalytic domains (although, the second domain of Hdac10 is described as inactive (de Ruijter et al., 2003; Tong et al., 2002), we hypothesized that they may act redundantly. To test this, we generated hdac6;hdac10 double mutants. Similar to single mutants, these animals are viable and fertile and do not exhibit an overt external phenotype (Figure 1D). Since Hdac6 and Sirt2 were found to interact physically and catalyze tubulin deacetylation in vitro (North et al., 2003; Nahhas et al., 2007), we also generated hdac6;sirt2 double mutants. Again, no overt phenotypes were visible, and double mutant fish were viable and fertile (Figure 1D). We showed that this was also the case for hdac6; hdac10; sirt2 triple mutants (Figure 1D and Supplementary Figure 1). Therefore, the lack of a strong visible phenotype in single, double or triple mutants is not due to functional compensation involving these genes.

To investigate whether loss of function alleles in hdac6, hdac10, and sirt2 affect levels of tubulin acetylation in vivo we examined protein extracts from the eyes of single, double and triple mutant adults by Western Blotting. Each mutation appears to result in a higher level of tubulin acetylation when normalized to overall tubulin level (Figures 2A,B). The effect is enhanced and becomes statistically significant in the double mutants of hdac6 and sirt2. The mutation in hdac10 has only minimal or no effect on tubulin acetylation. Tubulin acetylation in triple mutant eye tissue does not differ from double mutants for hdac6 and sirt2 (Figures 2A,B). This finding is further supported by immunostaining of zebrafish eye cryosections (Figures 3E,F) and immunostaining of wholemount embryos showing elevated levels of acetylated tubulin within embryonic tissues (Supplementary Figure 2).

Tubulin acetylation is associated with long lived microtubules such as axonemal microtubules (Janke and Montagnac, 2017). We reasoned that disrupting tubulin deacetylating enzymes may affect cilia morphology and function, as inhibition of Hdac6 was shown to stabilize cilia from regulated resorption (Luxton and Gundersen, 2007; Pugacheva et al., 2007; de Diego et al., 2014; Ran et al., 2015). To evaluate acetylation of ciliary microtubules, we immunostained triple mutant embryos using anti-acetyl-α-tubulin (Lys40) (D20G3), a well-characterized and highly specific antibody that recognizes acetylated tubulin and is commonly used in cilia visualization (Skoge and Ziegler, 2016; Viau et al., 2018; Sheu et al., 2019; Zhang et al., 2019). This analysis revealed increased acetylation of cytoplasmic microtubules in triple mutants compared to wild type (Figure 3A and Supplementary Figure 2). Surprisingly, we observed that hair cell kinocilia in cristae and macule of the inner ear in triple mutants are barely immunoreactive at 3 days post fertilization (dpf) (Figures 3A,G). This phenotype is observed as early as 1 dpf, when macule cilia are formed, and is also visible at 5 dpf (Figures 3B,C and Supplementary Figure 3). This phenotype is not limited to cilia in the ear and was observed in kidney, skin, and brain, at 1 dpf (Supplementary Figure 3). To evaluate this phenotype further, we investigated the impact of single and double knockouts on tubulin acetylation in cilia in the ear. Hdac6 mutation decreased tubulin acetylation in kinocilia of cristae hair cells at 5 dpf by 20%, whereas mutations in either hdac10 or sirt2 did not show any significant effect. The combined mutation of hdac6 and sirt2 resulted in a decrease of ciliary tubulin acetylation by 60% compared to the control. Hdac10 mutation did not affect ciliary tubulin acetylation even when combined with hdac6, or hdac6, and sirt2 (Figure 3H). As reliable antibodies against zebrafish hdac6 are not commercially available to confirm the role of Hdac6 in the observed phenotype we utilized a chemical Hdac6 inhibitor, CAY-10603 (Figure 4 and Supplementary Figure 3) (Bitler et al., 2017). This drug was chosen over a more commonly used broad-range HDAC inhibitor trichostatin A (TSA) to avoid interference by inhibition of HDAC1, which severely affects zebrafish eye and inner ear development (Harrison et al., 2011; Finnin et al., 1999; Yamaguchi, 2005; He et al., 2016). We observed a tubulin hypoacetylation phenotype in kinocilia of cristae of wild-type zebrafish embryos treated with 5 μM of CAY-10603 from 6 to 72 hpf. Tubulin acetylation in cilia was decreased by 35% and was not significantly different from the hdac6 single mutant (Figure 4). Only treatment at 6 h post fertilization resulted in the hypoacetylated phenotype in crista kinocilia, and fish treated at 24 and 48 hpf do not display this phenotype (Supplementary Figure 3). This is interesting as cilia in the ear are formed around 18 hpf, which would suggest that the observed hypoacetylation of axoneme occurs during ciliogenesis rather than being the result of later modifications (Whitfield, 2020).

To determine whether this phenotype is consistently observed in cilia of other tissues, we investigated kinocilia in inner ear macule, the connecting cilium of photoreceptor cells in the eye, and cilia in the nasal duct Tubulin acetylation in the kinocilia of inner ear macule was reduced in hdac6, hdac10, sirt2 single mutants, and hdac6; hdac10 double mutants. In the hdac6; sirt2 double mutant and the triple mutant tubulin acetylation in cilia was reduced by 40%, compared with the control (Figure 3J). Tubulin acetylation in connecting cilia of photoreceptor cells of triple mutant retinae showed a 20% reduction compared to control (Figures 3E,F,L). Together these analyses show a consistent effect of loss of function of hdac6 and sirt2 on cilia tubulin acetylation in vivo. However, it appears that the magnitude of this effect differs in different cells, being strongest in cristae, intermediate in macule, and weakest in the specialized connecting cilium found in photoreceptors, whereas acetylation of cytoplasmic tubulin is uniformly increased (Figures 3A,E and Supplementary Figure 2). Finally, olfactory cilia in the nose of triple mutant embryos displayed normal levels of tubulin acetylation in cilia (Figures 3D,I).

To investigate possible abnormalities in cilia morphology, and the potential impact on other tubulin modifications that may modulate ciliary stability, embryos were stained with the following antibodies: GT335, against poly-glutamylated tubulin, and TAP952, against mono-glycylated tubulin. These experiments showed that hdac6, sirt2, or hdac10 knockout does not affect ciliary axoneme length or shape in cristae or macule (Figure 3K). Our analysis indicates that lower levels of tubulin acetylation in the axoneme of triple mutants is correlated with increased levels of tubulin mono-glycylation, but not tubulin glutamylation, in ear kinocilia (Figures 5A–D) and in the connecting cilium of photoreceptor cells (Figures 3E,F,M).

We also considered the possibility that increased tubulin acetylation leads to reduced expression of alpha-tubulin acetyltransferase1 gene (ATAT-1). A lower level of acetylated ciliary tubulin, we hypothesized, could be a response to elevated acetylation levels inside the cell body leading to decreased expression levels of ATAT-1 via negative feedback. To test this, we compared expression levels between wild-type and triple mutant embryos in 5 dpf larvae, using qPCR. This analysis shows that the level of ATAT-1 expression is not changed significantly in response to mutation in hdac6, hdac10, and sirt2 genes (Figure 6).

Kinocilia in the inner-ear are responsible for detecting both static and dynamic balance. Static balance is the response to linear motion and gravity, which is detected by the utricle and saccule located within the vestibule (Forcioli-Conti et al., 2016). Dynamic balance involves the response to rotational motion such as turning. This is detected by cristae cilia in the semi-circular canals in response to movement of the gel-like cupula, in which the kinocilia are embedded and are fully functional as early as 3 weeks of age (Akella et al., 2010; Baxendale and Whitfield, 2016). Cilia damage in the inner ear and ciliopathies such as Usher syndrome are shown to affect the vestibular system (Renga, 2019; Whatley et al., 2020). Growing evidence links acetylation of tubulin to rigidity and resistance to mechanical stress (Szyk et al., 2014; Portran et al., 2017, Eshun-Wilson et al., 2019). Therefore, we hypothesized that cilia containing hypoacetylated tubulin, in the triple mutants, may be more vulnerable to accumulative mechanical damage. To evaluate this hypothesis, we induced mechanical stress on cilia in the ear, by placing tubes containing 5 dpf larvae (in E3 solution) on a vertical rotor at 30 rpm, for 30 min. Similar to the effect of a roller coaster in humans, movement of the otolith, physically coupled to hair cell kinocilia in the macule, should exert force, due to gravity and centrifugal forces, which disorientates the larvae. To quantify the ability of zebrafish larvae to recover from mechanical stress, we measured their recovery immediately after being removed from the rotor and placed in a 10cm dish. After 30 min of rotation, wild type fish commenced swimming within 1 min, whereas the majority of the triple mutant fish remained immotile for 20 min (Figure 7 and Supplementary Figure 4). Without introduction of stimulus, wild type and triple mutant swimming patterns did not visibly differ (Figure 7). These data show that the triple mutant was significantly more susceptible to balance loss after stimulus. To investigate possible cilia damage, we visualized ciliary tubulin using the anti-mono-glycylated tubulin antibody (TAP952) (Figure 7). The staining did not reveal any visible damage or breaks in ciliary axonemes. Furthermore, no difference in the number of cilia was observed (Figure 7).

Hdac6 alleles were obtained using TALEN (Zu et al., 2013) (target site TTCACTCGCTTTCACTGCC TALEN binding sequences: GTATATGTAGACGCC and TCTGGGATGCCAGGT deleted 14 bp GCTTTCACTGCCTC name SH400). hdac10 (target site GGGAGCAACTCTGCAGCTGG Oligo1:TAGGGAGCAA CTCTGCAGCTGG Oligo2: AAACCCAGCTGCAGAGTTGCTC deleted 11 bp CTGGTGGACAG name SH401) and sirt2 (target site AGTCTTGGATGAGTTAACCC Oligo1: CACTT GGCCTTAGTCCTGGA Oligo2 GCACAGCAGCAAAGTGT TCA deleted 2 bp AA then inserted 24 bp GGATGAGT TGTCTTGGATGAGTTG name SH626) were created with CRISPR/Cas9 (Heler et al., 2014; Sander and Joung, 2014). Double and triple mutants were obtained by crossing parental strains and identified by DNA sequencing. Zebrafish were maintained in accordance with the UK Home Office regulations and UK Animals (Scientific Procedures Act 1986).

Sectioning and immunohistochemistry were performed using previously described protocols (Malicki et al., 2011). Following fixation, transverse cryosections, 20 μm thick, were taken through the retina or the ear and processed for staining with the following antibodies: anti-acetylated a-tubulin (rabbit) (D20G3; 1:1000; Cell signaling) Acetyl-α-Tubulin (Lys40) (mouse) (6-11B-1), anti-polyglutaminated tubulin GT335 (1:500; Adipogen) anti-monoglycylated tubulin TAP952 (1:500; Millipore) (Alexa Fluor 488- and 564-conjugated secondary antibodies were used for all staining procedures (1:1000) as well as DAPI for counterstaining. Images of cryosections were collected at the Wolfson Light Microscopy Facility using an Olympus FV1000 confocal microscope and a 60x immersion objective. Images of immunostained wholemount embryos were collected using the Olympus FV1000 confocal microscope with a 40× dipping lens.

Eyes of 2.5-year-old fish were dissected and subsequently snap frozen in liquid nitrogen. Eyes were homogenized in RIPA buffer (Catalog No.: 89900 Thermo Fisher Scientific) and centrifuged supernatant was transferred to Eppendorf tubes and used for the Western Blotting. Proteins were resolved by SDS–PAGE, transferred to nitrocellulose membrane (Amersham GEHealthcare), and incubated with antibodies C4 (MAB1501R Sigma-Aldrich) for actin and α-Tubulin (DM1A) (Mouse mAb #3873) for tubulin and Acetyl-α-Tubulin (Lys40) (6-11B-1) for acetylated tubulin. For detection an Amersham ECL Western Blotting Detection Kit was used as described in manual. To visualize blots a ChemiDoc MP Imaging System was used and signals were measured using ImageJ.

RNA was extracted from ten 5 dpf embryos using the Monarch Total RNA Miniprep Kit (New England Biolabs Inc., United Kingdom) following the manufacturer’s instructions. Extracted RNA was quantified with a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, United States). Reverse transcription of cDNA was performed using a High-Capacity cDNA Reverse Transcription Kit (Catalog No.: 4368814, Applied Biosystems, United Kingdom) following the manufacturer’s instructions. cDNA was kept at 4°C for immediate use or −20°C for long-time storage. cDNA was kept at 4°C for immediate use or −20°C for long term storage. qPCR reactions were carried out to quantify RNA expression of genes of interest using SYBR Green. Beta-actin was used as a housekeeping reference gene control.

Primers used: hdac10 F5′-TGTCCTATGAAGCGCTCAGG-3′ R5′-TTTGACCGCCTCCAGGTACT-3′, hdac6 F5′-GGAGATG AGCAATGAACTGCAA-3′ R5′-GACTGGCATCCCAGAGTGA A-3′, sirt2 F5′-TTTGCGCAGTCTTTTCTCGC-3′ R5′-CCCAG CTCCAACCATACAGA-3′, atat-1 F5′-TCCCTTACGACCTG AATGCG-3′ R5′-GATCTGGTCTCCCATGCGCT-3′, tubulin F5′-ATGCTCGTGGTCACTACACT-3′ R5′-GGAAACCCTGG AGACCTGTG-3′. Data Analysis for qPCR was conducted using Microsoft Excel 2010 (Microsoft Corporation, United States) and GraphPad Prism 8 (GraphPad Software, United States).

The threshold level of fluorescent intensity was set to 0.04 to obtain Ct values across all experiments. Average Ct values from three technical repeats were firstly calculated. ΔCt values were obtained by subtracting Ct values of beta-actin from Ct values of the target gene, from of the same biological sample. The fold change was calculated with the formula 2^–ΔCt, and the standard deviation was automatically calculated and shown as error bars.

For balance tests, three 5 dpf zebrafish larvae were transferred to Eppendorf tubes and exposed to the centrifugal force of approximately 0.0005 N (r = 0.125 m v = 0.5 m/s m = 0.00025 g), using a Stuart^TM Rotator Disk for 30 min (40 rpm). Subsequently, fish were transferred to a 12-well plate and their behavior was recorded (Supplementary Figure 5). Swimming distances were normalized to Wild type.

Embryos were treated with 5 μM CAY10603 (cat. no: B2819-5, Cambridge Bioscience) for 1, 2, or 3 days, with the solution changed daily.

Data from cryosections and whole-mount immunostaining were analyzed with Fiji (ImageJ) software. Schematic illustration of protein domains was created with biocuckoo software (Liu et al., 2015). Sequencing data was analyzed with SnapGene software.