Tubb3 expression levels are sensitive to neuronal activity changes and determine microtubule growth and kinesin-mediated transport

Microtubules are dynamic polymers of α/β-tubulin. They regulate cell structure, cell division, cell migration, and intracellular transport. However, functional contributions of individual tubulin isotypes are incompletely understood. The neuron-specific β-tubulin Tubb3 displays highest expression around early postnatal periods characterized by exuberant synaptogenesis. Although Tubb3 mutations are associated with neuronal disease, including abnormal inhibitory transmission and seizure activity in patients, molecular consequences of altered Tubb3 levels are largely unknown. Likewise, it is unclear whether neuronal activity triggers Tubb3 expression changes in neurons. In this study, we initially asked whether chemical protocols to induce long-term potentiation (cLTP) affect microtubule growth and the expression of individual tubulin isotypes. We found that growing microtubules and Tubb3 expression are sensitive to changes in neuronal activity and asked for consequences of Tubb3 downregulation in neurons. Our data revealed that reduced Tubb3 levels accelerated microtubule growth in axons and dendrites. Remarkably, Tubb3 knockdown induced a specific upregulation of Tubb4 gene expression, without changing other tubulin isotypes. We further found that Tubb3 downregulation reduces tubulin polyglutamylation, increases KIF5C motility and boosts the transport of its synaptic cargo N-Cadherin, which is known to regulate synaptogenesis and long-term potentiation. Due to the large number of tubulin isotypes, we developed and applied a computational model based on a Monte Carlo simulation to understand consequences of tubulin expression changes in silico. Together, our data suggest a feedback mechanism with neuronal activity regulating tubulin expression and consequently microtubule dynamics underlying the delivery of synaptic cargoes. Supplementary Information The online version contains supplementary material available at 10.1007/s00018-022-04607-5.


Introduction
The microtubule (MT) cytoskeleton represents a dynamic structure mediating a variety of cellular and subcellular functions in neurons and other cell types [1]. In addition to the regulation of cell division, cell migration and neurite outgrowth, microtubules regulate intracellular transport of organelles, vesicles, mRNAs, and signaling molecules across the cell [2]. They consist of α/β-tubulin heterodimers that form protofilaments, which dynamically assemble to form a hollow tube. Due to the nature of the dimers, representing the microtubule building blocks, microtubules contain a plus-and a minus-end. They bind a number of accessory proteins at their plus-end (+TIPs) that include EB3 [3]. The dynamic behavior of microtubule ends is characterized by the process of dynamic instability, which comprises growth, catastrophe, shrinkage, and rescue driven by GTP-hydrolysis [3]. Consistent with MTs mediating diverse functions several genes encode for alpha-and beta-tubulin. At least 8 different alpha-tubulins (Tubas) and eight different betatubulins (Tubbs) are expressed in the mouse. In addition, microtubules undergo posttranslational modification, such as de-tyrosination, acetylation or polyglutamylation, and are stabilized through microtubule-associated proteins (MAPs) 1 3 575 Page 2 of 21 [4][5][6]. Due to the many possible combinations to form α/βtubulin heterodimers and their posttranslational modification patterns, individual microtubules are thought to have specific identities and functions [5,[7][8][9][10][11][12]. However, the complex roles of tubulin isotype functions, in particular in neurons that represent polar and excitable cell types, are just beginning to be understood. It is for instance unknown whether the expression of individual tubulin genes is regulated by neuronal activity, or vice versa, whether changes in the expression of tubulin isotypes regulates neuron-specific functions, such as synaptic plasticity, the ability of synapses to change in strength.
Plastic adaptions of neuronal transmission contribute to learning and memory in the central nervous system. For instance, long-term potentiation (LTP) induced through high-frequency synaptic stimulation [13,14] or chemical induction [15,16] triggers a long-lasting increase in synaptic strength. LTP also induces several pre-and postsynaptic mechanisms, such as the delivery and insertion of N-Cadherin and AMPA receptors into postsynaptic sites [17][18][19]. Whereas microtubule-dependent transport is critical for the activity-dependent delivery and rearrangement of synaptic factors [2,20,21], functional roles of individual tubulin isotypes, in particular the neuron-specific tubulins, have remained elusive.
Polymorphisms in tubulin genes of human patients lead to several neurological disorders, known as tubulinopathies [22][23][24]. They include Tubb3, a major tubulin isotype in brain, which is exclusively expressed in neurons and influences the dynamics of microtubules [8,25,26]. Knockdown of Tubb3 gene expression or specific missense mutations were shown to influence cortical development in mice [27,28]. It was further reported that Tubb3 expression is upregulated in human and rat epileptic tissue [29], linking Tubb3 to neuronal activity. In Tubb3 knockout mice, other Tubb genes are upregulated to compensate for the overall level of total beta-tubulin. However, specific Tubb3 functions cannot be replaced by other isotypes [30], indicating essential roles of Tubb3 in this context. Whether Tubb3 mediates unique functions with respect to the kinesin-mediated transport of plasma membrane proteins that regulate neuronal synapses, is presently unknown. Kinesin family 5 (KIF5) motor proteins, also known as kinesin I motors, travel along microtubules to deliver a variety of subcellular cargoes across the cell [2]. In neurons, KIF5 transports vesicles containing AMPA-type glutamate receptors (AMPARs) and/or the cell adhesion molecule N-Cadherin toward synaptic sites [31,32]. The extracellular domain of postsynaptic N-cadherins binds in trans to the extracellular domain of presynaptic N-cadherins, thereby spanning the synaptic cleft. N-Cadherin regulates cytoskeletal functions in association with catenins and affects voltage-dependent calcium influx [33]. It is sensitive to neuronal activity changes that induce mechanisms of synaptic plasticity [19,34,35]. In particular, LTP promotes the formation of N-Cadherin clusters in stimulated spines, thereby stabilizing spine structure [34]. In general, the delivery of neuronal cargoes is highly regulated by the combinatorial use of cargo adapters [2]. For instance, N-Cadherin transport through KIF5 requires the glutamate receptor-associated protein GRIP1 [32]. Other modes to regulate transport include tubulin posttranslational modifications, such as polyglutamylation. Previous studies in neurons revealed that increased polyglutamylation decreases the efficiency of neuronal transport along axons [36] or toward the synapse [37].
Here we asked whether changes in neuronal activity alter microtubule growth and/or the expression of individual tubulins. Vice versa, we aimed to understand whether the downregulation of a specific tubulin isotype alters microtubule dynamics and microtubule-dependent transport in neurons. We used the motor protein KIF5 and the LTP-sensitive cell adhesion molecule N-Cadherin (a KIF5 cargo) to monitor potential transport changes, following Tubb3 knockdown. Our data reveal that Tubb3 expression is sensitive to neuronal activity changes, whereas Tubb3 downregulation accelerates microtubule growth and microtubule-dependent transport of motors and cargoes.

Cell culture
Neuroblastoma N2a, human embryonic kidney 293 (HEK) and COS-7 cells were cultured in DMEM supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin. Cells were split every two to three days using 5% trypsin. To culture primary neurons, mouse embryos at stage E15 to E17 were used. Hippocampi were dissected in ice cold HBSS (Gibco, Thermo Scientific, #14170-088). To dissociate the cells, the hippocampi were incubated in 0.05% trypsin supplemented with EDTA at 37 °C for 5 min. The reaction was stopped with prewarmed DMEM supplemented with 10% fetal bovine serum. Cells were triturated with two fire polished Pasteur pipettes of different diameter in prewarmed HBSS. To count cell numbers, a Neubauer chamber was used. For immunofluorescent staining, 80,000 cells were seeded per 12 mm glass coverslip (24 well plate), coated with poly-l-lysine in 1 ml Neurobasal Plus Medium (Gibco, Thermo Scientific, # A35829-01). For time lapse imaging, 50,000 cells were seeded into the center of a 25 mm glass coverslip (6 well plate), coated with poly-l-lysine in 3 ml Neurobasal Plus Medium.

Transfection
To transfect N2a cells, the ScreenFect A-plus Transfection Reagent (InCella, #S-6000-1) was used according to the manual. HEK cells were transfected using the calcium phosphate method. Primary hippocampal neurons were transfected using the calcium phosphate method or Lipofectamine, respectively. For transfection of a 12 mm coverslip (24 well plate) with the calcium phosphate method, 2 µg DNA were mixed with water to a volume of 18.75 µl and 6.25 µl CaCl 2 (1 M) were added. This solution was mixed dropwise with 25 µl 2 × HBS (280 mM NaCl; 10 mM KCl; 1.5 mM Na 2 HPO 4 ; 12 mM dextrose; 50 mM HEPES) and incubated for 10 min at room temperature (RT). For a 25 mm coverslip (6 well plate), the double amounts were used, as described. Then, two thirds of the medium were removed and the transfection mix was added, followed by an incubation period between 45 min and 2 h at 37 °C and 5% CO 2 . Afterwards, the medium containing the transfection mix was removed and the cells washed twice in HEPES buffer (10 mM HEPES; 135 mM NaCl; 5 mM KCl; 2 mM CaCl 2 ; 2 mM MgCl 2 ; 15 mM glucose; pH 7.4). The remaining conditioned medium was then returned to the cells. For HEK cells, the transfection mix was added to the medium and incubated overnight, then, cells were lysed or fixed. For transfecting a 25 mm coverslip (6 well plate) with Lipofectamine, 2 µg DNA was mixed with 23 µl Optimem (Opti-MEM I(1x), gibco, #31985-070). In parallel, 2 µl Lipofectamine (Lipofectamine 2000, Invitrogen, #11668-019) were mixed with 23 µl Optimem. Both solutions were mixed and incubated for 20 min. Two thirds of the neuronal medium were removed from the well and the transfection mix was added for 2 h at 37 °C and 5% CO 2 . Afterwards, the medium containing the transfection mix was removed and cells were washed twice with HEPES buffer (10 mM HEPES; 135 mM NaCl; 5 mM KCl; 2 mM CaCl 2 ; 2 mM MgCl 2 ; 15 mM glucose; pH 7.4). Finally, the remaining Table 1 Energy parameter settings after refinement 100 microtubule growth velocity values were calculated for each of the filtered energy parameter sets, using control or Tubb3 knockdown concentrations, respectively. The velocity standard error of the mean (SEM), representing 100 calculations was 0.002 µm/s. Energy parameter sets that generated velocities in this range (scrambled control: 0.167 ± 0.002 µm/s; mi-RNA4-mediated Tubb3 knockdown: 0.201 ± 0.002 µm/s) were selected. Parameter sets that coincided for control or knockdown conditions were filtered. This procedure resulted in four individual parameter settings that fulfilled the requirements

Time-lapse video microscopy and image processing
Time lapse imaging was conducted with a spinning disc confocal microscope (Nikon ECLIPSE Ti) at 37 °C and 5% CO 2 using a CCD camera (Hamamatsu, EM-CCD, Digital Camera C9100). Axons and dendrites were identified by morphological characteristics. Videos were acquired with a 100 × objective (NA 1.45) and an image acquisition rate of 1 s over 3 min with the Visiview software (Visitron Systems). Fixed samples were imaged with a confocal microscope from Olympus (Olympus Fluoview FV1000) using a 60 × objective (NA 1.35) and the Fluoview software (Olympus). Analysis of the videos was conducted with an ImageJ macro.

Microtubule growth model
To simulate microtubule growth velocity, a two-dimensional computational Monte Carlo model was developed that incorporates three different tubulin dimer types as building blocks. Each rectangle represented one alpha-and beta-tubulin dimer, which formed the protofilaments marked by numbers 1 to 13 (Fig. 5a, right). Periodic boundaries mirrored protofilament 1 next to protofilament 13 and vice versa, to generate a continued grid. Bonds between dimers along one protofilament were characterized by longitudinal bond energies. Bonds between dimers of neighboring protofilaments were characterized by lateral bond energies. Protofilament 1 and 13 were shifted by 1.5 dimers to represent the MT seam. Accordingly, each dimer had 0, 0.5, 1, 1.5 or 2 lateral neighbors. Since we focused on microtubule growth velocities, the effects of catastrophe, shrinkage, rescue and GTP hydrolysis were neglected. During the growing phase of a microtubule, dimers stochastically bound to or dissociated from the protofilament tips, respectively. If more dimers were bound than were dissociated, the microtubule increased in length. Binding and unbinding were characterized by the bimolecular on-rate constant k on (µM −1 s −1 ) and the unimolecular off-rate constant k off (s −1 ). They were related by the equilibrium constant K (µM −1 ): The bimolecular on-rate constant was multiplied with the tubulin concentration to give a pseudo first order on-rate constant k on (s −1 ). The standard Gibbs free energy change was described by: with R being the universal gas constant and T the absolute temperature in kelvin. It was assumed, that each dimer that binds to a protofilament, forms one longitudinal bond changing the free energy by ∆G long . The energy changes for dimer immobilization and conformational change were included into ∆G long to simplify the parameter setting. For each lateral neighbor, the energy ∆G lat was added, leading to a change in the free energy upon dimer binding of: with x being the number of lateral neighbors. Therefore, with a given k on and combining Eqs. 1, 2 and 3, k off was calculated as:

Simulation procedure
To perform the simulation, the following possible events were considered. Each protofilament either bound a type one, type two or type three dimer, reflecting individual combinations of tubulin isotypes. To consider sterical hindrance of existing lateral neighbors at the potential binding site, an on-rate penalty of 2 (k on /2) was introduced for one lateral neighbor and a penalty of 10 (k on /10) for two lateral neighbors [40]. In addition, each dimer of the grid was able to dissociate. As shown in Eq. (4), the number of lateral neighbors determined the quantity of lateral binding energies and, therefore influenced k off . If a dimer dissociated from a position inside the grid, all upper dimers of this protofilament also dissociated and all lateral energies of these dimers were added. To enable dissociation from the beginning of the simulation, protofilaments were given a start-length of 10 dimers, each. Dimer types were chosen within a Monte Carlo simulation randomly to yield a uniform distribution of dimers. In the second step, the execution time t (s) for each event was calculated according to the literature [41].
with i being the index of the possible event, N being a uniformly distributed random number between 0 and 1, and k being the rate constant (s −1 ) of the event.
In a third step, the event with the shortest execution time was implemented and its time was added to the total amount of time. To model velocities, as observed experimentally, steps one to three were repeated 5,000 times. Simulation parameters were k on , ∆G long (1), ∆G long (2), ∆G long (3), ∆G lat (1), ∆G lat (2), ∆G lat (3) as well as the tubulin dimer concentrations C1, C2 and C3 (parameters and final values are summarized in Fig. 6b). The final output was the time of growth and the length of the protofilaments, from which the growth velocity of the microtubule tip could be calculated by: velocity = mean length of protofilaments total elapsed time .

Parameter determination
Tubulin heterodimers consist of one α-tubulin and one β-tubulin. In the present study, we focused on β-tubulin isotypes, therefore α-tubulin isotypes were not differentiated.
To determine the parameters, the model was adjusted to match experimental EB3 imaging data. Accordingly, dimer type 1 was defined to represent Tubb3-containing dimers and dimer type 2 to represent Tubb4-containing dimers. Since free energy changes are additive, dimer type 3 was defined to represent all other β-tubulin isotype-containing dimers (Fig. 5a). With the aim to predict longitudinal and lateral energies of Tubb3 and Tubb4 (∆G long (Tubb3), ∆G lat (Tubb3), ∆G long (Tubb4), ∆G lat (Tubb4)), published values for the onrate constant k on = 30 (µM −1 s −1 ) and the total tubulin concentration C = 7 µM were adapted from a study by Castle and colleagues [42]. Longitudinal and lateral energy values (∆G long (other Tubbs) = − 7 RT and ∆G lat (other Tubbs) = − 4 RT) for type 3 dimers were estimated based on references [40][41][42][43]. From RNA data in the literature [7,44], it was estimated that Tubb3 represents 14%, Tubb4 28% and the remaining Tubbs 58% of soluble tubulin in the neuronal cytoplasm. Using these values and a tubulin concentration of 7 µM, individual tubulin dimer concentrations were calculated for control conditions (C(Tubb3) = 0.98 µM, C(Tubb4) = 1.96 µM and C(other Tubbs) = 4.06 µM). For Tubb3 knockdown conditions, individual dimer concentrations were determined using our experimental results from this study. Following Tubb3 knockdown, Tubb3 expression levels were set to 43%, representing an average value of different experiments in this study and Tubb4 expression levels were set to 133%. Based on this, molarities of individual Tubbs were calculated (C(Tubb3) = 0.42 µM, C(Tubb4) = 2.61 µM and C(other Tubbs) = 3.97 µM). To determine longitudinal and lateral energies of Tubb3-and Tubb4-containing dimers, the program was executed with longitudinal energies ranging from − 3 RT to − 9 RT in 0.5 RT steps and lateral energies ranging from -1 RT to the respective longitudinal value. This procedure was justified, since longitudinal energies have to be smaller than lateral energies to polymerize stable MTs [45]. For each energy parameter set (∆G long (Tubb3), ∆G lat (Tubb3), ∆G long (Tubb4), ∆G lat (Tubb4)), microtubule growth velocities were calculated under control or Tubb3 knockdown conditions, respectively. Since plotting 5-dimensional data were not feasible, results were collected and processed using Excel (Microsoft), as described below. To display the results, energy values for Tubb3 and Tubb4 were plotted separately in contour graphs ( Fig. 5b-e). To plot Tubb3 energies, Tubb4 energy values were fixed to final result values (∆G long (Tubb4) = − 7.5 RT, ∆G lat (Tubb4) = − 6 RT). Likewise, the same procedure was applied for Tubb4 plots (∆G long (Tubb3) = − 5 RT, ∆G lat (Tubb3) = − 3 RT). Gray contour lines were displayed to mark experimental velocities (Control: 0.167 µm/s; Tubb3 KD: 0.201 µm/s). From the original results parameter sets that generated velocities in the range of: scrambled control: 0.167 ± 0.006 µm/s; mi-RNA4-mediated Tubb3 knockdown: 0.201 ± 0.008 µm/s, were selected. Parameter sets that coincided for control and knockdown conditions were filtered, resulting in 88 parameter sets that fulfilled the requirements (Table S1). To narrow their number, the program was executed with these parameter sets, calculating 100 velocities under control and Tubb3 knockdown concentrations, each. Since the velocity standard error of mean (SEM) for 100 calculations turned out to be 0.002 µm/s, energy parameter sets generating similar velocities (scrambled control: 0.167 ± 0.002 µm/s; mi-RNA4-mediated Tubb3 knockdown: 0.201 ± 0.002 µm/s) were selected. This procedure resulted in four parameter sets that fulfilled the requirements (Table 1).

Increase in microtubule growth and tubulin expression following cLTP induction
Neurons are excitable cells that adapt intracellular transport processes in response to activity changes [17,20,46]. They are characterized by a dynamic microtubule cytoskeleton consisting of different tubulin isotypes [3,8,47,48], however it is barely understood whether activity changes induce changes in tubulin expression and whether For graphical representation of the distribution, data are binned with a bin size of 0.05 a.u. Statistics: One way ANOVA followed by Student's t-test or Mann-Whitney Rank Sum Test. *p < 0.05, **p < 0.01 ◂ a differential use of specific tubulin isotypes alters microtubule and/or transport function. In an initial experiment, we applied established chemical protocols to induce longterm potentiation (cLTP) to trigger a long-lasting increase in synaptic strength in cultured hippocampal neurons [38,39,49]. Using a fusion protein of the microtubule + TIP factor EB3 (EB3-Tomato), known to label growing microtubules [50,51], we found that increased synaptic strength leads to a significant increase in microtubule growth velocities, as compared to control conditions (Fig. 1a, b and Figure  S1a, b). Following induction of cLTP in acute hippocampal slices over different time periods (10-30 min), we further observed significantly increased expression levels for total alpha-and beta-tubulin in the range of about 20%, detected by western blotting with pan-Tuba or pan-Tubb-specific antibodies, respectively. Significantly increased gene expression was detectable at 20 min following cLTP induction and declined afterwards (Fig. 1c-f). In a subsequent isotypespecific analysis, we confirmed the upregulation of tubulins, in particular of Tubb1 and Tubb3, which displayed significantly increased expression levels at 20 and 30 min after cLTP induction (Fig. 1g-j). In contrast, the tubulin isotypes Tubb2, Tubb4 and Tubb5 remained unaltered under these conditions ( Figure S2a-f), indicating that a long-lasting increase in synaptic activity [49] does not generally affect tubulin expression. Likewise, within this early phase of cLTP, which is known to be independent of the synthesis of neurotransmitter receptors [52], but is based on receptor rearrangement from endocytic reserve pools [17,53], the expression of the AMPA receptor subunit GluA2 or the postsynaptic protein Homer remained unaltered ( Figure S2g-l). Together, we conclude that individual tubulin expression levels and microtubule growth rates are sensitive to neuronal activity changes.

Neuronal microtubules grow faster and longer following Tubb3 downregulation
Since Tubb3 represents a neuron-specific tubulin isotype [23], we aimed to downregulate Tubb3 gene expression to assess functional consequences. Initially, we tested five different Tubb3 mi-RNA knockdown constructs (miRNA-1 to 5), compared to a scrambled control (miRNA-scr), following transfection in neuroblastoma N2a cells. Western blot analysis revealed that four individual constructs reduced GFP-Tubb3 significantly, leading to remaining protein levels in the range of 17 to 42% (Fig. 2a, b). Out of these candidates, we chose miRNA-2 and miRNA-4 to further test their knockdown capacity in mouse cultured hippocampal neurons. Based on the much lower transfection efficiency of neurons, immunocytochemistry turned out to be a suitable assay to assess Tubb3 protein levels in the cell soma (Fig. 2c). Quantitative analysis of fluorescent intensities in multiple regions of interest (small rectangles) confirmed that miRNA-2 and miRNA-4 significantly reduce endogenous Tubb3, leading to Tubb3 intensities of 56% or 45% compared to scrambled control levels, respectively (Fig. 2c, d,  Fig. S3). Likewise, reduced Tubb3 expression was found in neuronal axons that were detected by Ankyrin G immunostaining of their initial segment (Fig. 2e, f). A neuronal time course experiment revealed that Tubb3 gene expression gradually decreased to about 50% from day three onwards (Fig. 2g). These levels equal those of heterozygous human patients, carrying Tubb3 loss-of-function mutations [23]. For further analysis, we therefore focused on construct miRNA-4 (hereafter referred to as Tubb3 KD) at three days after transfection to investigate functional consequences of reduced Tubb3 levels.
First, we combined reduced Tubb3 gene expression (Tubb3 KD) with EB3-imaging in hippocampal neurons. Under control conditions (Control), we observed microtubule growth velocities with a median value of 0.164 µm/s in neuronal dendrites. Upon reduction of Tubb3 levels (Tubb3 KD), microtubules grew significantly faster (p < 0.001) with Fig. 2 Knockdown of Tubb3 gene expression. a Western blot analysis of GFP-Tubb3 expression in N2a cells, following cotransfection of miRNA constructs (miRNA-scr for scrambled control and miRNA-1 to 5) to assess their Tubb3 knockdown capacity. Gamma-Adaptin served as loading control. Control conditions were equally treated with transfection reagent, but without DNA. b Quantification of a (N = 3 experiments). Statistics: one-way ANOVA, followed by Dunnett's Method for multiple comparisons versus control group. mean ± SEM. c Cultured hippocampal neurons transfected with miRNA-scr, miRNA-2, or miRNA-4, respectively. Scale bar, 50 µm. Somata in boxed regions are enlarged in the middle (green: control, red: knockdown). Tubb3 intensities were measured at DIV12 inside the soma, indicated by the small colored boxes. DAPI staining of the nucleus (not shown) was used to verify that Tubb3 intensity measurements were performed in the cytoplasm. Scale bar, 5 µm.  (Fig. 3a, b). They also grew about 46% longer on average (Control: 4.69 µm; Tubb3 KD: 6.85 µm), whereas their growth duration remained equal ( Figure S4a, b). Likewise, in axons, microtubules grew faster (p < 0.01), when Tubb3 was reduced to about half of control levels (Control: 0.167 µm/s; Tubb3 KD: 0.190 µm/s) (Figs. 3c, d and S4c, d). To exclude off-target effects, increased EB3 growth velocities were confirmed following Tubb3 downregulation through an independent miRNA sequence ( Figure S5). These data show, that the knockdown of Tubb3 modulates microtubule growth and suggest that the concentration of individual tubulin isotypes matters with respect to the dynamics of the growing microtubule cytoskeleton.

Establishment of a computational model to predict microtubule growth velocities
Due to the large number of possible tubulin isotypes in mammalian neurons and the limited access to isotypespecific antibodies, we developed a computational model to predict microtubule growth in silico. A two-dimensional approach was applied, which incorporates three different tubulin dimers (types 1-3) in a flattened protofilament sheet, consisting of alpha-/beta-tubulin dimers (colored rectangles, Fig. 5a, right). In the model, dimer type 1 represents Tubb3-containing dimers, dimer type 2 represents Tubb4-containing dimers and dimer type 3 the remaining Tubb isotype-containing dimers. To reduce complexity, we neglected catastrophe, shrinkage, rescue and GTP hydrolysis. Model parameters, constants as well as the simulation procedure are summarized in the methods section ( Fig. 5a-f). To refine the system, we used 400 runs of the program, calculating microtubule growth velocities under control and Tubb3 knockdown conditions, respectively. This led to an accuracy in the standard error of the mean (SEM) in the range of 0.001 µm/s. Under these conditions, we found that the parameter setting (∆G long (Tubb3) = − 5RT, ∆G lat (Tubb3) = − 3RT, ∆G long (Tubb4) = − 7.5RT and ∆G lat (Tubb4) = − 6RT) mimicked our experimental EB3 imaging results (Fig. 6a, b; compare with Fig. 3c, d).
After determination of suitable energy parameters (Fig. 6b) simulating the EB3 imaging experiment (Fig. 3d), we used these settings to assess the number of individual dimer types incorporated into growing microtubules. The program was applied with 100 runs per condition and the actual numbers of dimer types were normalized to the total amount of dimers. Under control conditions, 4.3% turned out to be Tubb3-containing dimers (Type 1), 41.4% Tubb4containing dimers (Type 2) and 54.3% dimers containing other Tubbs (Type 3) (Fig. 6c, control conditions). In contrast, under conditions modeling Tubb3 knockdown, we obtained a 50% reduction in Tubb3-containing dimers (2.1%, Type 1), as expected (Fig. 6c, Tubb3 dimers/Tubb3 KD, compare with Fig. 2d). In parallel, Tubb4-containing dimers (Type 2) turned out to be increased to 48.3% (Fig. 6c, Tubb4 dimers/Tubb3 KD, compare with Fig. 4c, d), while dimers-containing other Tubbs (Type 3) accounted for 49.7% (Fig. 6c, other Tubb dimers/Tubb3 KD). These results confirm former experimental data suggesting that reduced Tubb3 levels can be compensated by increased Tubb4.
Together, the computational model was able to mimic our experimental results (Figs. 2, 3, 4). We therefore conclude that it is suitable to predict further neuronal microtubule growth values in silico.

Modeling microtubule growth in silico
Next, we aimed to simulate other combinations of Tubb expression levels with respect to their capacity in regulating microtubule growth velocity. Initially, we modeled a scenario of increased Tubb3 expression, referred to as scenario I (high Tubb3, low Tubb4, equal other Tubbs). 400 runs of the program predicted that scenario I will lead to a significant decrease in the growth velocities of microtubules, compared to control conditions (control: 0.168 ± 0.001 µm/s; scenario I: 0.092 ± 0.001 µm/s) (Fig. 6d, left). Likewise, scenario II (equal Tubb3, low Tubb4, more other Tubbs, 400 runs) predicts a significant decrease in microtubule In the flat two-dimensional protofilament sheet (right), α/β-tubulin heterodimers are simplified as single-colored rectangles which mark dimer types 1, 2 or 3. Dimer binding forms 13 protofilaments that assemble into a hollow tube, the microtubule. Periodic boundaries mirror protofilament 1 next to protofilament 13 and vice versa, so that a continues grid is generated. Longitudinal and lateral bond energies are depicted as vertical and horizontal arrows, respectively. Dimers a, b, c, and d differ in the number of lateral neighbors. Illustration of 1.5 lateral neighbors is not shown. b-e Contour graphs of longitudinal and lateral energy values of Tubb3 and Tubb4 dimers. Velocity values are color coded as indicated to the right. Gray contour lines mark experimentally measured velocities (control: 0.167 µm/s; Tubb3 knockdown: 0.201 µm/s). In stable microtubules, longitudinal energies are smaller than lateral energies [39], therefore no velocity values were calculated for the parameter sets at the bottom right (white areas). b Tubb3 energy values under control conditions. c Tubb4 energy values under control conditions. d Tubb3 energy values under Tubb3 knockdown condition. e Tubb4 energy values under Tubb3 knockdown condition. f Scheme of the program code ◂ growth velocities (control: 0.168 ± 0.001 µm/s; scenario II: 0.125 ± 0.001 µm/s) (Fig. 6d). Consistent with this view, rising Tubb4 levels in turn are predicted to speed up microtubule growth in scenario III (equal Tubb3, high Tubb4, equal other Tubbs; control: 0.168 ± 0.001 µm/s; scenario III: 0.234 ± 0.002 µm/s) (Fig. 6d). Interestingly, this seems to be different for Tubb3. While low Tubb3 speeds up microtubule growth, increased Tubb3 expression is predicted to also accelerate microtubule growth (high Tubb3, equal Tubb4, equal other Tubbs; control: 0.168 ± 0.001 µm/s; scenario IV: 0.181 ± 0.002 µm/s) (Fig. 6d).
To experimentally validate this in silico prediction, we aimed to test whether the latter result (Fig. 6d, scenario IV) would be reproduceable in a neuronal context. Control transfections using a mammalian Tubb3 expression construct induced high Tubb3 levels in HEK293 cells that lack endogenous Tubb3 expression (47) (Fig. 7a). Immunostaining with a Tubb3-specific antibody confirmed that heterologously expressed Tubb3 colocalizes with the microtubule cytoskeleton (Fig. 7b), suggesting that this approach is suitable to increase Tubb3 levels within microtubules. Finally, EB3 imaging following exogenous Tubb3 expression in hippocampal neurons, significantly increased microtubule growth velocities in axons (Fig. 7c, d), as predicted by computational modeling (compare with Fig. 6d, scenario IV).
In summary, we conclude that the expression levels of individual tubulin isotypes translate into individual functions with respect to the dynamics of microtubules. Due to the large number of alpha-and beta-tubulins and the many dimer combinations, computational modeling may be suitable to identify critical players.

Tubb3 levels affect polyglutamylation and kinesin-mediated cargo transport along microtubules
Since dynamic microtubules represent the tracks along which motors move, we asked whether the transport of motors and cargoes might be affected following Tubb3 knockdown. Cargo transport in neurons is highly sensitive to changes in tubulin posttranslational modifications (PTMs) [5,54,55]. Actually, increased polyglutamylation levels were shown to reduce the efficiency of KIF5-mediated neuronal transport, whereas decreased polyglutamylation levels increase axonal mitochondria transport [36,37,56]. In contrast, it is barely understood whether the concentration of specific tubulin isotypes contributes to changes in neuronal transport parameters.
Remarkably, we found that tubulin polyglutamylation levels in neuronal axons were significantly reduced to 57% of control values following Tubb3 knockdown (Fig. 8a, b). In subsequent time-lapse imaging experiments using living hippocampal neurons, we therefore analyzed the motility of the kinesin I-type motor protein KIF5C, which is highly mobile in axonal compartments [2]. A fluorescent fusion protein KIF5C-tomato-pex26, connected to a peroxisomal targeting signal [37], was used for neuronal live cell imaging. Under scrambled control conditions (Control), we observed a median value of KIF5C velocities of 1.31 µm/s (Fig. 8c, d). In contrast, the reduction of Tubb3 gene expression levels (Tubb3 KD) significantly increased the speed of movement to 1.47 µm/s (p = 0.018). Whereas the run length of the motor remained equal (Control: 24.0 µm; Tubb3 KD: 23.9 µm; p = 0.838), the duration of movement turned out to be significantly reduced (Control: 29.0 s; Tubb3 KD: 22.0 s; p = 0.017). In a further approach, we also performed timelapse imaging with a fluorescent N-Cadherin fusion protein (mRFP-N-Cad) in axons [32]. N-Cadherin represents a physiological cargo of KIF5C motors in neurons, which head toward the synapse underlying the regulation of synaptogenesis and LTP [32,34]. Similarly, as seen for KIF5C connected to peroxisomes, the velocity of axonal N-Cadherin movement significantly increased in neurons expressing reduced levels of the tubulin Tubb3 (median values, Control: 1.56 µm/s; Tubb3 KD: 2.18 µm/s; p = 0.028) (Fig. 8e,  f). Again, we observed that the run length of the motor remained equal (Control: 31.9 µm; Tubb3 KD: 27.9 µm; p = 0.595), whereas the duration of movement turned out to be significantly reduced (Control: 31.0 s; Tubb3 KD: 18.0 s; p < 0.001). We therefore conclude that the neuronal concentration of the tubulin isotype Tubb3 is a relevant parameter

Discussion
In this study, we report that microtubule growth and Tubb3 expression levels are sensitive to neuronal activity changes (Fig. 1). Our data further show that differential Tubb3 is incorporated into filamentous microtubules. Scale bar, 5 µm. c Live cell imaging in axons using DIV 12 cultured hippocampal neurons cotransfected with EB3-Tomato and control or Tubb3 constructs. Image acquisition: 1 frame/s over 3 min. Scale bar, 20 µm. EB3 comets in axonal segments in colored boxes of overview images (top panel) are visualized as kymographs (bottom panel). d Quantification of c, depicting median values (N = 3 experiments). ***p < 0.001. Each data point represents one EB3 comet; control (n = 75), exogenous Tubb3 (n = 82). For graphical representation of the distribution, data are binned with a bin size of 0.015 µm/s. Statistics: Mann-Whitney Rank Sum Test gene expression levels of individual tubulin isotypes translate into specific microtubule functions. We highlight the neuron-specific tubulin Tubb3 as an example to show that decreased Tubb3 levels alter microtubule growth parameters (Fig. 3) and microtubule-dependent transport of motor-cargo complexes (Fig. 8). Our study also reveals that decreased levels of a specific tubulin isotype (Tubb3) are partially compensated through Tubb4 upregulation (Fig. 4). Finally, we provide a computational Monte Carlo model that allows to simulate whether and how other tubulin isotypes affect microtubule growth velocities (Figs. 5 and 6).
It is presently incompletely understood how neuronal activity changes translate into changes of the microtubule cytoskeleton. A previous study identified that pharmacological activation of AMPARs or blockade of glycine receptors (GlyRs) alter tubulin polyglutamylation [54], however to our knowledge, cLTP-induced activity changes have not been studied in the context of microtubule growth or tubulin isotype expression. The observation that cLTP protocols induce increased levels of specific (Tubb1, Tubb3), but not all tubulin isotypes suggests for signaling cascades that mediate synapse-to-nucleus communication to target the expression of individual tubulin genes [57]. Tetraethylammonium (TEA)-induced cLTP has been shown to activate ERK-signaling in neurons, inducing phosphorylation of the nuclear transcription factor CREB [58]. Follow-up studies that search for signaling pathways connecting neuronal activity with tubulin expression could help to unravel the cooperation between both systems.
Previous in vitro microtubule dynamics assays had revealed that isolated microtubules display different growth rates, depending on the nature of individual tubulin isotypes. In fact, the growth velocity of Tuba1a/Tubb3 microtubules increased with decreasing amounts of Tubb3 [8]. Our data from hippocampal neurons (Fig. 3) confirm these in vitro results in a cellular context. Furthermore, it was reported that microtubules assemble faster in the presence of tau, as compared to the presence of MAP2 [10], suggesting that in addition to tubulin isotypes also microtubule-associated proteins contribute to their polymerization rates.
Tubb3 knockdown in the present study led to a specific upregulation of Tubb4, but no other Tubb isotypes. These results differ from Tubb3 knockout mice, which showed a general increase in other Tubbs in the range of 10-20% [30]. We conclude that these differences in the regulation of gene expression stem from the fact that our study did not manipulate the Tubb3 gene but applied RNA silencing, which might be more physiological. Differences between gene knockout and knockdown studies have been reported in other cases and may be due to the position of the targeting vector or to the genetic background [59]. With respect to Tubb4, it should be noted that Tubb4 antibodies might have some cross-reactivities [7], which makes it difficult to fully exclude that other Tubbs remain unaffected. Likewise, the Tubb2 antibody detects Tubb2a and Tubb2b, wherefore some compensatory changes might have been missed. Although these technical limitations exist, modeling predicted a Tubb4 upregulation following Tubb3 knockdown (Fig. 6c), suggesting that the computational and experimental data in this study support each other.
Due to the number of tubulin genes and the large variety of possible dimer formations, our computational model can help to predict microtubule growth rates. This is not at least helpful because many antibodies against tubulins are not specific and show cross-reactivity [7]. Former microtubule models were based on a single dimer type, thereby neglecting different isotypes [40][41][42][43]. In contrast, the computational model in the present study incorporates three individual types of dimers, considering changes in isotype conformation. Since energy parameters were based on experimental EB3 data obtained from living cells, the model also considers the modulatory effects of microtubule binding proteins and posttranslational modifications in regulating microtubule growth velocities. Although it simplifies certain aspects of natural microtubules, it has been proven suitable to mimic experimental data (compare Figs. 3c, d, 6a). With that it provides a starting-point for future upgrades that include hydrolysis, catastrophe or shrinkage.
In general, the physiological connection between neuronal activity and the microtubule cytoskeleton is highly relevant, since activity changes not only alter cytoskeletal parameters, but microtubule-based transport in turn delivers mRNAs and proteins toward synapses to regulate synaptic transmission and/or plasticity. Increased microtubule growth rates (Figs. 1b, 3) might be beneficial for microtubules that invade into dendritic spines, following synaptic NMDA receptor activation [60,61]. Since LTP triggers the KIF5mediated delivery of AMPARs and N-Cadherin toward synapses, these molecules represent critical factors to monitor activity-dependent transport. Our data suggest that low Tubb3 levels accelerate their delivery (Fig. 8). Interestingly, exogenous Tubb3 expression in hippocampal neurons also led to increased MT growth rates in axons. Just as Tubb3 downregulation is compensated through increased Tubb4 expression, it is possible that increased Tubb3 expression would also affect other isotype levels. Remarkably, tubulin polyglutamylation levels, which are also known to regulate neuronal transport, are significantly decreased following Tubb3 knockdown (Fig. 8a, b), accompanied by increased transport rates. Consistent with these results, previous studies reported that increased polyglutamylation levels decrease the efficiency of neuronal transport whereas reduced polyglutamylation boosts axonal transport of mitochondria [36,37,56].
In summary, our study reveals that the concentration of Tubb3 matters, with respect to activity-dependent Fig. 8 Tubulin polyglutamylation levels, KIF5C motor protein mobility, and N-Cadherin transport following Tubb3 knockdown. a Cultured hippocampal neurons transfected with control or Tubb3 KD constructs, respectively. Scale bar, 50 µm. Axons were identified by an Ankyrin-G staining of the axon initial segment (AnkG). Axonal regions in boxed regions are enlarged in the middle (green: control, red: knockdown). Polyglutamylation intensities were measured at DIV12 indicated by the small colored boxes. Scale bar, 5