We first examined a 48-nm repeat cryo-EM density map from native DMTs of Tetrahymena WT (CU428 strain) (Fig. 1A) (S. Kubo et al., 2023). We identified and modeled all visible αK40 loops of α-tubulins in the map, which are known to be almost 100% acetylated (Akella et al., 2010; Gaertig et al., 1995). Most Tetrahymena MIPs have been localized and identified in this map (S. Kubo et al., 2023). In contrast to the acetylated singlet microtubule structure (Eshun-Wilson et al., 2019), we observed many fully structured αK40 full (missing no more than two residues from residues 37 to 48) and partial structure loops (missing 3-5 residues) in DMT (Fig. 1C-G). In certain PFs, including A1, A3, A5, A11, and B1, all αK40 loops are structured. Interestingly, the structured αK40 loops are much less abundant in the B-tubule (Fig. 1G). There is a pattern in the distribution of αK40 loops: likely due to their flexibility, ordered αK40 loops are visible at positions where MIPs and tubulins interact but are difficult to resolve in places with little or no MIPs. (Fig. 1H, Table S1). Notably, in A1, where RIB72A and RIB72B are in contact with αK40 (Fig. 1C, D), all αK40 loops are fully structured, in agreement with the 8-nm alternating pattern of Rib72A and Rib72B (Fig. 1G). This pattern is consistent with previous observations of the αK40 loops in the green algae Chlamydomonas reinhardtii DMT (Khalifa et al., 2020; Ma et al., 2019). Therefore, the B-tubule has fewer ordered αK40 loops compared to the A-tubule, probably due to fewer MIPs in the B-tubule.

A superimposition of the conformations of all fully structured αK40 loops in the DMT revealed that they adopt multiple conformations (Fig. 2A). The exact structures of αK40 loops might adapt to both the inter-PF angle and interactions with the contacting MIPs. We did not observe any conformations where acetylated αK40 was directly involved in tubulin-tubulin lateral interactions (Fig. 2A, Fig. 1C-F).

We noticed that the αK40 loop is always structured when interacting with DM10 domains of the MIPs, suggesting that DM10 is a αK40 loop interacting domain. Superimposing all seven DM10 domains (three from RIB72A, three from RIB72B and one from CFAP67) (Fig. 2B) showed that while there are some variations in other parts of DM10 domain, the region interacting with the αK40 loop are conserved in its topology (Fig. 2B, C). Sequence alignment of the DM10 domains suggests that an aromatic residue (phenylalanine or tyrosine) is conserved and might be responsible to the interaction with the DM10 domain (Fig. 2C, D, E).

To determine whether the conformation of the αK40 loops is conserved between species, we compared the αK40 loops in the DMTs of Tetrahymena thermophila and Chlamydomonas reinhardtii (Fig. 2F, G, H). The αK40 loop conformations from B9 and B10 are similar between the two species, and both interact with the conserved CFAP52. These results suggest that the αK40 loop conformation is tuned by its interactions with adjacent MIPs. The interaction between MIPs and the αK40 loop implies that acetylation might play a role in DMT assembly or stability or both.

With our observation that acetylated αK40 in the DMT of wild type Tetrahymena cells adopts a fixed conformation when interacting with MIPs, we evaluated whether the absence of αK40 acetylation affects the MIPs and hence the overall DMT structure. We obtained the DMT structures from the Tetrahymena MEC17-KO and K40R mutants at 4.5 and 3.5 Å resolution, respectively. In MEC17-KO, an ortholog of the mammalian α-TAT1 α-tubulin K40 acetyltransferase, is knocked out to abolish detectable αK40 acetylation, while in the K40R mutant, lysine 40 on ATU1 (the single canonical α-tubulin isotype in Tetrahymena) is mutated to arginine to prevent acetylation at the αK40 position (Akella et al., 2010).

Our first observation is that all the MIPs look intact in both the MEC17-KO and K40R mutants (Fig. 3A). This observation is consistent with the phenotypes that the MEC17-KO and K40R cilia look similar to the WT cilia (Akella et al., 2010; Gaertig et al., 1995). Therefore, our cryo-EM analyses indicate that acetylation does not affect DMT and MIP assembly.

Next, we examined the tubulin structure by comparing the αK60 and αK40 loop structures at PFs A3 and B1 (Fig. 3B-I, Fig. S2). Previously, it was reported that αK40 acetylation led to a reduction in the distance of αK60 from the M-loop of the adjacent tubulin from 12 Å to 8 Å (Eshun-Wilson et al., 2019). Our comparison of (acetylated and non-acetylated) αK40 and αK60 loops did not reveal significant structural differences. However, this result does not rule out a change in αK40 and αK60 conformation, which is likely too small to be observed accurately at this resolution.

Since αK40 loops are suggested to be involved in lateral interactions (Cueva, J., & Huang, 2012; Eshun-Wilson et al., 2019), acetylation might affect inter-PF angles. We measured the inter-PF angles in the DMTs of the WT and the two acetylation-deficient mutants. We observed only minor changes in the inter-PF angles in the A-tubule (Fig. 4A). However, in the B-tubule, the changes in the inter-PF angles were more prominent (Fig. 4A-J). Most notably, the change in the inter-PF angle between PFs B7 and B8 and B9 and B10 ranges from 3 to 6 degrees (Fig. 4H-J, Table S2). These are significant changes in specific PF curvatures.

We also inspected the tubulin lattice by measuring the distances between adjacent tubulin dimers. There were small changes in the inter-dimer distances in PFs A1, A3, A5, A8, B3 and B7 between WT and mutants (Fig. S3, Table S3). This scale of changes in the tubulin lattice is significantly smaller (0.3 – 0.5 Å) than the known impacts of the nucleotide state (Zhang, Alushin, Brown, & Nogales, 2015) and missing MIPs (Ichikawa et al., 2019) (∼2 Å difference). This is similar to the observation of differences in interdimer distances in 96% acetylated and 99% de-acetylated microtubules using in vitro reconstitution.

Overall, our analysis shows that tubulin lattice changes, specifically inter-PF angles, can be used as an indication of microtubule stability when acetylation is lacking.

While we observed some structural differences between acetylated and non-acetylated DMT, our resolution does not allow us to see small changes in the αK40 loops. Therefore, we attempted to detect structural differences using molecular dynamic simulations.

We performed an all-atom molecular dynamic simulation for α-tubulin with acetylated and non-acetylated αK40 to determine whether acetylation changes the loop behavior (Fig 5A, B). To determine how acetylation of αK40 affects the structure of the αK40-loop, we used the K-means clustering method for the αK40-loop region based on the Cα position. To avoid bias in clustering, the structures obtained during the simulated trajectories with acetylated and non-acetylated αK40 were mixed, and 10 clusters were created (Fig. 5A). Interestingly, clusters comprise mostly acetylated or non-acetylated conformations but not mixed populations. This indicates that acetylated and non-acetylated αK40 loops adopt distinctively different conformations. In addition, the most populated cluster consists of only ∼ acetylated conformations and ∼50% of all acetylation conformations, while the non-acetylated αK40 loops form five clusters of less than 30% conformations. Therefore, the acetylated αK40 loop adopts more rigid conformations than non-acetylated αK40 loop. It appears that this finding is similar to that of a previous study (Eshun-Wilson et al., 2019) despite differences in molecular dynamic setups of porcine and ciliate tubulins and clustering methods.

Since we did not observe any structural changes in the MIPs of MEC17-KO and K40R mutants, non-acetylated αK40 does not seem to significantly weaken the binding of MIPs. Therefore, we further explored the degree to which the interaction between the αK40 loop and MIP is affected by acetylation using coarse-grain molecular dynamic simulation to compare the energy between tubulins and CFAP52 with the acetylated αK40 and the non-acetylated αK40 structures (Fig. 5C).

We found that both acetylated and non-acetylated αK40 have a stabilizing effect when binding to CFAP52. On the other hand, when calculating the mean and standard error of the energy between tubulins and CFAP52 from the entire trajectory, the energy of the acetylated αK40 case is only slightly lower than that of the non-acetylated αK40 case (Fig. 5D). Similarly, no difference between acetylated and non-acetylated αK40 in interactions with similar setup using RIB72A. Even when accounted for the periodic structures of DMT with many MIPs, the energy difference between acetylated and non-acetylated αK40 with MIPs is still not significant.

Our coarse-grain molecular dynamic simulations of MIPs and αK40 suggest that acetylation likely does not play a role in the interaction with MIPs.

Our analysis thus far has failed to reveal significant structural differences in the MIPs within the DMT structures of WT, K40R and MEC17-KO mutants. Therefore, we searched for more subtle changes in the axoneme composition using mass spectrometry. We analyzed the same DMT samples used for cryo-EM, which contain no membrane and matrix fractions, to see any proteomic changes due to the lack of acetylation. In addition, to eliminate the downstream effect of the lack of acetylation i.e. destabilization of DMT, we combined the mass spectrometry results in this study with the mass spectrometry studies of the RIB72A/B-KO and RIB72B-KO mutants (S. Kubo et al., 2023). The RIB72A/B-KO mutant lacks both RIB72A and RIB72B, which leads to a significant number of MIPs missing and slower swimming speed (Stoddard et al., 2018). Therefore, we can use RIB72B-KO and RIB72A/B-KO as controls to specifically look for the upstream effect due to the lack of acetylation.

We first performed a control of proteins not supposed to interact with αK40 by analyzing the levels of radial spoke proteins and found no significant differences (Fig. S4A). We then looked at the MIP abundance. Most MIPs showed no significant changes in their abundance (less than 1.5-fold changes) except for CFAP112, CFAP141 and RIB27 (Fig. 6A) in the K40R and MEC17-KO samples. However, we did not observe any differences in the periodicity of CFAP112, CFAP141 and RIB27 in the cryo-EM maps of the DMTs from WT, K40R and MEC17-KO mutants. As shown recently in the case of CFAP77A and CFAP77B, certain MIPs might not localize consistently along the length of the cilium (S. Kubo et al., 2023), and there might be some changes in the occupancy of those MIPs in a specific region that lead to changes in their abundance.

There were 11 proteins significantly elevated in both K40R and MEC17-KO mutants by fourfold compared to the wild type (Fig. 6B, Fig. S4B, C). One of these proteins was also elevated in the RIB72A/B-KO mutant vs. WT, and therefore, 10 proteins were specifically upregulated in both the K40R and MEC17-KO strains (Fig. 6B, G). Similarly, we found that three proteins were significantly reduced in both K40R and MEC17-KO but not in the RIB72A/B-KO knockout mutant (Fig. 6C, E). Among the proteins reduced or missing in the K40R and MEC17-KO mutants, there are protein phosphatase 2A-related proteins: PP2A regulatory subunit A (TTHERM_00766530, UniProt ID: I7MAR7), PP2C (TTHERM_00316330, UniProt ID: I7LW71), and PP2A (TTHERM_00355160,

UniProt ID: Q22Y55) (Fig. 6D). Furthermore, one kinase was downregulated in MEC17-KO and K40R cells (Fig. 6E) (TTHERM_00623090, UniProt ID: Q240X5). In Chlamydomonas, PP2A is present in DMT and required for normal ciliary motility (Elam et al., 2011). It was previously reported that the knockdown of PPP1R2 (protein phosphatase inhibitor 2) reduces αK40-acetylation in the primary cilium of human retinal epithelial cells (Wang & Brautigan, 2008). Inhibition of protein phosphatase (1 and 2A) with calyculin in PPP1R2 knockdown cells partially rescued the acetylation of ciliary microtubules. These results suggest that the lack of αK40 acetylation reduces PP2A activity in DMT. Therefore, we can infer that acetylation and phosphorylation interact in DMT.

Bean media (Williams, Wolfe, & Bleyman, 1980) was used to store Tetrahymena cells (WT (CU428), K40R, MEC17-KO), with 4 μL of the cell culture transferred into 40 mL of liquid SPP media (Gorovsky, Yao, Keevert, & Pleger, 1975). The cells were grown for 7 days at room temperature and then transferred into 50 mL of SPP liquid media overnight growth at 30°C with 150 RPM shaking in a Thermo Fisher MAXQ8000 incubator. Then, 50 mL of this overnight culture was added to 900 mL of liquid SPP media and grown at 30°C (MAXQ8000) at 150 RPM for two days or until the OD₆₀₀ was 0.7.

To harvest Tetrahymena cells, overnight culture was centrifuged at 700 × g for 10 minutes with slow deceleration at 22°C in an Avanti Centrifuge (Rotor JLA-8.1000). Ten milliliters of room temperature SPP medium supplemented with DTT was used to resuspend the cell pellet and then adjusted to a final volume of 24 mL, followed by transfer to an ice-cold 250 mL Erlenmeyer flask. Immediately, 1 mL of dibucaine (dissolved in distilled water at 25 mg/mL) was added to the flask and gently swirled for exactly 60 seconds in an ice water bath. To stop the reaction, 75 mL of ice-cold SPP media (supplemented with 1 mM EGTA) was immediately added to the Erlenmeyer flask and then split into two 50 mL conical tubes for centrifugation at 2000 × g for 10 minutes at 4°C with no deceleration (Sorvall ST 16R, Rotor 75003181). The supernatant that contained cilia was transferred to centrifuge tubes for the Beckman Coulter JA 25.50 rotor, approximately 30 mL per tube, for centrifugation at 17000 × g for 40 minutes at 4°C with slow deceleration (Avanti, Rotor JA25.50). The pellet was gently washed with Cilia Wash Buffer (50 mM HEPES at pH 7.4, 3 mM MgSO4, 0.1 mM EGTA, 1 mM DTT, 250 mM sucrose). and frozen with liquid nitrogen for storage in a −80°C freezer.

The cilia suspension was thawed on ice and then centrifuged at 16000 g and 4°C for 10 minutes in a microfuge in a refrigerated room (Eppendorf, Centrifuge 5415 D). The pellet was resuspended in 250 μl of ice-cold Cilia Final Buffer (50 mM HEPES at pH 7.4, 3 mM MgSO4, 0.1 mM EGTA, 1 mM DTT, 0.5% trehalose, 1 mM PMSF). To clean the cilia, the resuspended cilia were centrifuged at 16000 × g for 10 minutes at 4°C in a microfuge (Eppendorf, Centrifuge 5415 D), and the supernatant was removed. Then, the pellet was resuspended in 250 μl of Cilia Final Buffer [without trehalose but with 44.1 µl of 10% NP-40 alternative (Millipore Sigma, 492016)] added to a final concentration of 1.5% NP-40]. The sample was placed on ice to incubate for 30 minutes before the demembraned flagella supernatant was removed after centrifugation at 16,000 × g for 10 min at 4°C in a microfuge (Eppendorf, Centrifuge 5415 D). The pellet containing the axoneme was resuspended in 245 μl of Cilia Final Buffer (without trehalose), and then 2.5 μl of 100 mM ADP (to a final concentration of 1 mM ADP) was added for incubation at room temperature for 10 minutes. This was followed by adding 2.5 μl of 10 mM ATP (to a final concentration of 1 mM ATP) for incubation at room temperature for 10 minutes. Bradford reagent (Bio-Rad 5000201) was used to measure the total protein concentration, and the final protein concentration was adjusted to 3 mg/mL using Cilia Final Buffer (without trehalose).

The concentration of the DMT solution was adjusted to 3 mg/ml. Quantifoil R2/2 grids (Electron Microscopy Sciences, #Q225CR-06) were treated using 1 mL of chloroform overnight followed by negative glow discharge (30 seconds at 25 mAh). Then, 3.5 μl of axoneme sample was applied to treated grids inside the Vitrobot Mark IV (Thermo Fisher) at a blot force of 3, blot time of 5 seconds, and drain time of 0.5 seconds, followed by plunge freezing into liquid ethane.

Using a Titan Krios 300 kV FEG electron microscope (Thermo Fisher) with a K3 Summit direct electron detector (Gatan, Inc.) and the BioQuantum energy filter (Gatan, Inc.), movies of the axoneme were acquired at 64kx nominal magnification (calculated pixel size of 1.370 Å/pixel) using SerialEM (Mastronarde, 2005). A total dose of 45 electrons per Ų over 40 frames for the WT and MEC17-KO datasets. A total dose of 73 electrons per Ų per frame over 30 frames for the K40R dataset. The defocus range was between −1.0 and −3.0 μm at an interval of 0.25 μm.

Motion correction and dose-weighting of the movies were performed using MotionCor2 (Zhang et al., 2017) implemented in Relion 3 (Zivanov et al., 2018), and the contrast transfer function parameters were estimated using Gctf (Zhang, 2016). Micrographs with apparent drift, ice contamination, and poor contrast transfer function estimation were discarded (18384, 25610, 4283 micrographs for WT, K40R and MEC17-KO data, respectively). The filaments were picked manually using e2helixboxer (Tang et al., 2007).

An 8-nm periodicity was used to pick particles of 512 x 512 pixels, binned twice, and prealigned using a modified version of the Iterative Helical Real Space Reconstruction script (Egelman, 2007) in SPIDER (Frank et al., 1996) to work with nonhelical symmetry. The alignment parameters were then transferred to Frealign to align the particles for 6 iterations in Frealign (Grigorieff, 2007) and then converted into Relion 3.0. In Relion 3, iterative per-particle-defocus refinement and Bayesian polishing were performed for the 80 nm particles.

Particles were subtracted from their tubulin lattice signal and underwent 3D classification into two classes to obtain the 16-nm repeat particles. The 16-nm repeat particles were then subjected to 3D classification into 3 classes to obtain the 48-nm repeat particles. The 48-nm particles were then refined, resulting in resolutions of 3.5 and 4.3 Angstrom for K40R and MEC17-KO DMTs from 182387 and 39417 particles, respectively.

To improve the local resolution for each PF during modeling, we performed focused refinements by using masks to cover adjacent PF regions in the DMT. Next, the maps were enhanced by DeepEmhancer (Sanchez-Garcia et al., 2021) to improve visualization and interpretability.

The WT tubulin model 6U0H (Ichikawa et al., 2019) was first fitted into the higher-resolution K40R cryo-EM map and then locally modeled using Coot (Emsley et al, 2010) and real space refined in Phenix (Adams et al, 2010) for generation of the K40R tubulin model. The WT and MEC17-KO tubulin models were generated by fitting the K40R tubulin model in WT and MEC17-KO cryo-EM maps, respectively, followed by refinement in Coot (Emsley et al, 2010) and Phenix (Adams et al, 2010). The αK40 loop regions from WT, K40R, and MEC17-KO were locally modeled in focused refinement cryo-EM maps using Coot (Emsley et al, 2010) and Phenix (Adams et al, 2010). All the maps and model visualization were taken using ChimeraX (Goddard et al., 2018).

Based on the atomic structures of the four tubulin dimers, CFAP52 and IJ34, we performed coarse-grained molecular dynamic simulations. The purpose of this simulation was to check the effect of acetylated αK40 on the binding stability of CFAP52. In the coarse-grained model, each amino acid was represented as a single bead located at its Cα position. To observe their dynamics, we used the excluded volume effect, electrostatic interaction, and the energy function AICG2+ (Li, Terakawa, Wang, & Takada, 2012; Li, Wang, & Takada, 2014). In AICG2+, the reference structure was assumed to be the most stable conformation, and their parameters were modified from the reference. It is known that the intradimer interaction is much stronger than the interdimer interaction, and the interdimer interaction is much stronger than the intra-PF interaction, so we set the interdimer and PFs’ nonlocal native interaction force to 0.8 and 0.3 times the original value, respectively, while that of the intradimer was kept as the original value (1.0 times). Of note, three residues (PHE133, GLY308, and GLU401) of the B9-PF beta-tubulin at the plus end side were anchored in their position for convenience analysis. We performed the simulation 10 times with acetylated αK40 and non-acetylated αK40 structures using the CafeMol package version 2.1 (Kenzaki et al., 2011). Each molecular dynamic simulation took 3×10^7 molecular dynamic steps, and they were conducted by the underdamped Langevin dynamics at 300 K temperature. We set the friction coefficient to 0.02 (CafeMol unit), and default values were used for others.

Normally, in dealing with electrostatic interactions, LYS and ARG, GLU and ASP, and other amino acids were given a charge of +1, −1, and 0, respectively. However, in this simulation, it was necessary to evaluate the electrostatic interaction as accurately as possible, so we calculated the surface charge density from the all-atom structure and remapped the charge distribution using only Ca beads to reproduce the all-atom surface charge distribution. The technique is called RESPAC (Terakawa & Takada, 2014). We applied RESPAC to regions without missing data, such as the αK40-loop and E-hook. For the missing region (and so we performed loop modeling by MODELLER (Šali & Blundell, 1993), we treat their charge by default definition. In the coarse-grained model, each amino acid is treated as a single bead, so we simply assumed that if αK40 was acetylated, its charge was zero, while non-acetylated αK40 had a +1 charge.

The purpose of the all-atom molecular dynamic simulation was to check whether the acetylated αK40 loop takes fewer conformations than the non-acetylated αK40 loop. In the all-atom molecular dynamic simulation using GROMACS (Abraham et al., 2015; Pronk et al., 2013), we used the GROMOS54a7 force field for protein (Schmid et al., 2011) and SPC for solvent water (Jorgensen & Tirado-Rives, 2005). We added sodium and chloride ions to neutralize the system and to make the salt concentration approximately equal to 0.1 Energy minimization by the steepest descent minimization algorithm was followed by equilibration with NVT and NPT for 100 ps at 300 K. In the production run, we used NPT ensemble simulations with 1 atm and 300 K. The production run consisted of a 1 fs step for 180 ns. The α-tubulin at the plus end of B9-PF was used as a reference structure for the simulation. To model the structure of acetylated αK40, Vienna-PTM 2.0 was used (Margreitter, Petrov, & Zagrovic, 2013; Margreitter, Reif, & Oostenbrink, 2017; Petrov, Margreitter, Grandits, Oostenbrink, & Zagrovic, 2013). The force field used (GROMOS54a7) had already been set up for the acetylated lysine.

The inter-PF rotation angle can be defined by the lateral rotation angle between each subsequent PF pair. The rotation angles and Z-shift between PF pairs were measured using the “measure” command from ChimeraX (Pettersen et al., 2004). Data for WT, K40R, and MEC17-KO cells were compiled into GraphPad Prism 9 to perform ANOVA and plotting.

We docked in the atomic models of the α- and β-tubulins in the maps of WT, K40R, and MEC17-KO. The interdimer distance was measured between the N9 GTP of α-tubulin and that of the next α-tubulin in the same PF using the “distance” command from Chimera (Pettersen et al., 2004).

Samples prepared for cryo-EM were used for mass spectrometry analysis. Laemmli buffer at 4X (#1610747, Bio-Rad) was added to the microtubule fraction samples in Cilia Final Buffer buffer so that it was 1x, and 25-30 μg protein was loaded on the SDS_PAGE gel. Electrophoresis was performed, but the run was terminated before the proteins entered the separation gel. A band containing all proteins in the sample was then cut out from the gel and subjected to in-gel digestion. The obtained peptides (∼2 μg) were chromatographically separated on a Dionex Ultimate 3000 UHPLC. First, peptides were loaded onto a Thermo Acclaim Pepmap (Thermo, 75 μm ID × 2 cm with 3 μm C18 beads) precolumn and then onto an Acclaim Pepmap Easyspray (Thermo, 75 μm × 25 cm with 2 μm C18 beads) analytical column and separated with a flow rate of 200 nl/min with a gradient of 2-35% solvent (acetonitrile containing 0.1% formic acid) over 2 hours. Peptides of charge 2+ or higher were recorded using a Thermo Orbitrap Fusion mass spectrometer operating at 120,000 resolution (FWHM in MS1, 15,000 for MS/MS). The data were searched against the Tetrahymena thermophila protein dataset from UniProt (https://www.uniprot.org/).

Mass spectrometry data were analyzed by Scaffold_4.8.4 (Proteome Software Inc.). Proteins with mean values of exclusive unique peptide count of 2 or more in the WT mass spectrometry results were used for analysis. Raw mass spectrometry data were normalized by the total spectra. ANOVA statistical tests were applied to MEC17-KO, K40R and WT mass spectrometry results using biological triplicates. Proteins exhibiting a minimum of twofold increase/decrease and a statistical significance threshold (p < 0.05) in mutants compared to WT were identified as up- or downregulated.