Alternative Splicing Regulates Kv3.1 Polarized Targeting to Adjust Maximal Spiking Frequency*

Background: Ion channels play critical roles in converting synaptic inputs into digital outputs encoded by action potentials. Results: Kv3.1 axonal targeting, regulated by its C-terminal splice domain, enhances the maximal spiking frequency. Conclusion: Channel biophysical properties and localization are both important for Kv3.1-mediated fast spiking in neurons. Significance: This is the first report showing an important function of polarized targeting of Kv channels. Synaptic inputs received at dendrites are converted into digital outputs encoded by action potentials generated at the axon initial segment in most neurons. Here, we report that alternative splicing regulates polarized targeting of Kv3.1 voltage-gated potassium (Kv) channels to adjust the input-output relationship. The spiking frequency of cultured hippocampal neurons correlated with the level of endogenous Kv3 channels. Expression of axonal Kv3.1b, the longer form of Kv3.1 splice variants, effectively converted slow-spiking young neurons to fast-spiking ones; this was not the case for Kv1.2 or Kv4.2 channel constructs. Despite having identical biophysical properties as Kv3.1b, dendritic Kv3.1a was significantly less effective at increasing the maximal firing frequency. This suggests a possible role of channel targeting in regulating spiking frequency. Mutagenesis studies suggest the electrostatic repulsion between the Kv3.1b N/C termini, created by its C-terminal splice domain, unmasks the Kv3.1b axonal targeting motif. Kv3.1b axonal targeting increased the maximal spiking frequency in response to prolonged depolarization. This finding was further supported by the results of local application of channel blockers and computer simulations. Taken together, our studies have demonstrated that alternative splicing controls neuronal firing rates by regulating the polarized targeting of Kv3.1 channels.

The ability of a neuron to integrate synaptic inputs and generate proper action potential (AP) 2 -encoded outputs relies on the concerted actions of ligand-and voltage-gated ion channels that are present on its plasma membrane. Most neurons have multiple dendrites and one long axon, important for the neuro-nal input and output functions, respectively. Various types of Kv channels, with distinct biophysical and pharmacological properties, are differentially localized in dendrites and the axon. In mammalian brains, Kv1 (Shaker) channels are predominantly distributed along axons, whereas Kv2 (Shab) and Kv4 (Shal) channels are mainly localized in dendritic regions (1)(2)(3)(4)(5)(6)(7)(8)(9)(10). Kv3 (Shaw) channels display complex targeting patterns. Depending on the isoform, alternative splicing, and neuronal types, some Kv3 channels localize in axons and some in dendrites (11)(12)(13)(14)(15)(16)(17)(18)(19). Progress has been made in identifying molecular mechanisms underlying polarized targeting of different Kv channels (16, 20 -25). However, the exact function of the regulation of polarized targeting of Kv channels remains a mystery.
Excitatory and inhibitory synaptic potentials generated in dendrites and soma are summed and converted into one or a number of APs at the AIS of most neurons, where voltage-gated sodium (NaV) channels are highly concentrated. The AIS is also critical for gating the entry of axonal proteins (26,27). The spiking frequency of each neuron must properly reflect the strength of synaptic potentials. Regulation of the input-output relationship, which is emerging as a new form of plasticity of intrinsic excitability, is still poorly understood. The widest range of spiking frequencies is usually attributable to the presence of Kv3 channels, because of their unique biophysical properties, high activation threshold (about Ϫ20 mV), and rapid deactivation kinetics (28,29). However, it remains unknown whether Kv3 channel expression is sufficient for fast spiking, because not all Kv3-expressing neurons spike rapidly (11). This raises an intriguing question: how does the polarized targeting of Kv3 channels regulate the spiking frequency?
In this study, we use hippocampal neurons as a model system (30 -32) to examine how Kv3-polarized targeting affects the neuronal input-output relationship. Our results show that besides their essential role in increasing AP firing frequency, the expression of axonal Kv3.1b channels is sufficient to convert a slow-spiking young neuron to a fast-spiking one. Interestingly, the increase of an outward K ϩ current actually enhances neuronal excitability reflected by increased AP firing frequency. Moreover, our mutagenesis studies reveal novel mechanistic insights into axonal targeting of Kv3.1b. Our stud-Membrane potentials of isolated cells were normally held at Ϫ80 mV. Voltage pulses from Ϫ60 to ϩ60 mV with 200 -250-ms duration and 10-mV increments were applied.
Conductance-voltage relationships (G-V curves) for Kv channel constructs were: G ϭ I/(V m Ϫ V rev ), V rev ϭ Ϫ95 mV, normalized to the maximal conductance. Curves were fitted with Boltzmann function, G/G max ϭ 1/(1 ϩ exp[Ϫ (V Ϫ V 1/2 )/k]), where G max is the maximal conductance, V is the membrane potential, V 1/2 is the potential at which the value of the relative conductance is 0.5, and k is the slope factor. SigmaPlot 10.0 (Systat Software, Inc., Chicago, IL) was used for fitting.
To obtain activation time constant ( on ), activation curves (voltage was increased from Ϫ80 to ϩ30 mV) were fitted with a single exponential function raised to a power of 4, I(t) ϭ A(1 Ϫ exp(Ϫt/ on )) 4 . In studies of deactivation kinetics of Kv channels, the cells were held at Ϫ80 mV, given a 2-ms pre-pulse to ϩ60 mV and 20-ms voltage pulses from Ϫ100 to Ϫ10 mV. To obtain deactivation time constants ( off ), tail currents (voltage decreasing from ϩ60 to Ϫ60 mV) were fitted with the equation, I(t) ϭ A exp (Ϫt/ off ). Clampfit 10.0 was used for fitting to get on and off .
Current Clamp Recording of Action Potentials-The same internal solution and Hanks' buffer were used for recording of primary cultured neurons. Both mature neurons (older than 16 DIV) and young neurons (7 to 10 DIV) were recorded from either the E18 or the P8 cultures. The membrane resistance, capacitance, and resting membrane potentials of the neurons were measured. These values are consistent within each age group. APs induced by either puffing (0.5-1 s in duration) 2 M glutamate (Glu) onto neuronal soma, or current injection from the recording pipette, were recorded under the current clamp mode. For long pulse stimulations, 1000-ms duration currents of increasing amplitude (from 5 to 145 pA with increments of 10 pA) were injected. Due to the variation of expression levels of transfected channel constructs, only the neurons carrying clear after-hyperpolarization were used for quantification of spiking frequency. For short pulse stimulations, 2-ms duration currents of 800 pA with increasing frequency (from 50 to 300 Hz with increment of 50 Hz) were injected. Because the current injection is very short for short pulse stimulations, we found that only the large current (800 pA) can reliably induce the first AP in control neurons, consistent with previous studies (33)(34)(35).
Post Hoc Immunostaining of Hippocampal Neurons from P8 Culture-To examine the AP firing frequency of the endogenous Kv3.1 channels role, we performed current clamp recording on the P8 neurons from 14 to 16 DIV. For each neuron, induced AP traces by long-pulse stimulation were recorded and its morphology was imaged with transmitted light. The neuron was fixed immediately after recording, permeabilized with 0.2% Triton X-100, and stained for endogenous Kv3.1b and NaV channels with a mouse monoclonal anti-Kv3.1b antibody (clone number B16B/8; University of California Davis/NIH NeuroMab Facility, Davis, CA) and a rabbit polyclonal anti-pan NaV channel antibody (Millipore), respectively. The maximal AP frequency and immunostaining intensity of endogenous Kv3.1b in proximal axons are correlated.
Local Drug Application and Effects on Action Potential Firing-Glass pipettes with a tip diameter around 1 m for patch clamp recording were pulled with a model P-1000 Flaming/Brown micropipette puller (Sutter Instrument, Novato, CA). Glass pipettes with diameters around 50 m, made by adjusting the pulling parameters and filled with either 2 M Glu or 1 mM tetraethylammonium (TEA), were used for perfusion of whole neurons. Glass pipettes with diameters around 3 m, pulled with a designated program and filled with 1 mM TEA, were used for local perfusion onto axons or dendrites.
Immunostaining, Imaging, and Quantification-The immunocytochemical procedures were previously described (16,21). Briefly, neurons were stained under nonpermeabilized conditions (without Triton X-100) to label the surface pool and under permeabilized conditions (with 0.2% Triton X-100) to label total proteins. Fluorescence images were captured with a Spot CCD camera RT slider (Diagnostic Instrument Inc., Sterling Heights, MI) in a Zeiss upright microscope, Axiophot, using Plan Apo objectives ϫ20/0.75 and ϫ100/1.4 oil, saved as 16-bit TIFF files, and analyzed with NIH ImageJ and Sigmaplot 10.0 for fluorescence intensity quantification. Exposure times were controlled so that the pixel intensities in dendrites and axons were below saturation, but the same exposure time was used within each group of an experiment. The quantification procedure was described previously (16,23). Only transfected neurons with clearly separated dendrites and axons, and isolated from other transfected cells, were chosen for analysis. To obtain the axonal polarity index (F axon /F dend ) reflecting polarized targeting on the neuronal surface, we performed anti-HA staining under the nonpermeabilized condition. Using NIH ImageJ, we laid a line along the major axon to acquire its average fluorescence intensity (in arbitrary unit) (F axon ), and laid lines along proximal dendrites 10 m away from the soma to obtain the average fluorescence intensity on dendritic membranes (F dend ). To obtain the relative axonal level of total proteins (F axon /F sd ), we performed anti-HA staining under permeabilized conditions. Using NIH ImageJ, we laid a line along the major axon to acquire F axon , and laid lines along proximal dendrites starting from the soma to obtain F sd . The background fluorescence intensity was measured for each image and subtracted.
Simulation of Action Potential Firing Using NEURON Software-The computation model was implemented using NEURON 7.1 (36). We assumed a simplified morphology, largely consistent with a cultured hippocampal neuron at 10 DIV. The cell was constructed using three cylinders, each representing a distinct region of the neuron: a dendrite (100 m in length and 4 m in diameter), a soma (20 m in length and 20 m in diameter), and an axon (400 m in length and 2 m in diameter). The AIS is located in the first 10 -45 m of the axon.
Each simulation was compared three model neurons (control, dendritic Kv3.1, and axonal Kv3.1) with different Kv3 channel distributions but otherwise identical, that is, they had the same morphology, NaV channel distribution, and leak channel distribution.
In dendritic Kv3.1-expressing neurons, Kv3 current density is relatively higher on the somatodendritic surface, whereas in axonal Kv3.1-expressing neurons, Kv3 current density is relatively higher on the axonal surface including the AIS. This is largely consistent with the immunostaining results under the nonpermeabilized condition.
In Fig. 8, A and B, we provide one example of the simulation results. The parameters used in this figure are as follows: NaV channels carrying transient Na ϩ currents were present throughout the cell, with density in the AIS (37,300 pS/m 2 ) five times higher than along the rest of the axon (7,460 pS/m 2 ) and 20 times higher than the rest of the cell (1,865 pS/m 2 ). In the control neuron, Kv3.1 channels were absent. In the dendritic Kv3.1-expressing neuron, Kv3.1 channels were present on somatic and dendritic membranes (20,000 pS/m 2 ) and axonal membranes including the AIS (1,000 pS/m 2 ). In the axonal Kv3.1-expressing neuron, Kv3.1 channels were present on somatic and dendritic membranes (5,000 pS/m 2 ) and axonal membranes including the AIS (15,000 pS/m 2 ). The ratios of Kv3.1 channels used here were adopted from the most highly polarized examples observed in our immunostaining results (F axon /F dend ) under the nonpermeabilized condition. It was assumed here that the anti-HA immunofluorescence signal is in a linear relationship with the actual amount of Kv3.1 channels. Leaky channel conductance was present throughout the cell at 7.1 pS/m 2 .
The biophysical properties of NaV channels, Kv3.1 channels, and leaky channels used in our model cells were adopted from Golomb et al. (37) with modifications based on experimentally observed time constants. The current balance equation is, where V is the membrane potential of the neuron, C ϭ 1 F/cm 2 is the membrane capacitance, r is the section radius, R a ϭ 35.4 ⍀-cm is the cytoplasmic resistivity, I app denotes external current injected into the neurons, and I Na , I Kv3 , and I l represent the sodium (NaV), potassium (Kv3.1), and leak currents, respectively. These are modeled as: I Na (V,h,x) ϭ g Na (x) m ∞ 3 (V)h(V Ϫ V Na ); I Kv3 (V,n,x) ϭ g k (x)n 2 (V Ϫ V K ); and I l (V) ϭ g l (V Ϫ V l ), where g Na (x), g K (x), and g l represent the maximum NaV, Kv3.1, and leak conductances, the first two vary with space depending on the region of the cell. The sodium, potassium, and leak reversal potentials are V Na ϭ 50 mV, V K ϭ Ϫ77 mV, and V l ϭ Ϫ70 mV, respectively.
Long pulse experiments were performed as follows. First we administered a current at the midpoint of the soma for 5 s; then

Kv3.1 Polarized Targeting and Spiking Frequency
we counted the number of spikes; and finally divided the total count by 5 to get the firing rate. We established the robustness of the qualitative result that axonal Kv3.1 leads to a higher firing rate than dendritic Kv3.1 by keeping the expression ratios constant and randomly choosing base conductances in the following ranges: g Na (peak expression) between 0 and 50,000 pS/m 2 , g K (lowest expression) between 0 and 10,000 pS/m 2 , and g l between 0 and 20 pS/m 2 .

Spiking Frequency of Hippocampal Neurons and Endogenous
Kv3 Channels-To determine AP firing patterns of various mature hippocampal neurons (from 16 to 28 DIV), we performed whole cell current-clamp recordings on neurons from the E18 and P8 cultures (Fig. 1A). To mimic excitatory inputs, 2 M Glu was puffed (0.5-1 s) onto neuronal dendrites and soma. Glu puffing depolarized the membrane potential and induced AP firing (Fig. 1, B and C). In the same neuron, we also injected square-wave current pulses into the soma to induce APs. Similar spiking frequency resulted from the two different induction methods. In a slow-spiking neuron, Glu puffing and current injection led to a few APs per second (Fig. 1B). In contrast, in a fast-spiking like neuron, both Glu puffing and current injection induced much higher firing rates (Fig. 1C). Our finding that these cultured neurons display a variety of spiking frequencies is consistent with the fact that the hippocampus contains different types of neurons with distinct firing properties, including fast-spiking, parvalbumin-positive GABAergic interneurons (38).
To determine the role of Kv3 channels in regulating the spiking frequency, we examined the effect of a low concentration (1 mM) TEA, which preferentially blocks Kv3 channels (28). Indeed, the fast spiking of recorded neurons were markedly and reversibly suppressed by 1 mM TEA (normalized frequency control, 100%; 1 mM TEA, 44.4 Ϯ 13.4%; wash, 71.1 Ϯ 11.3%; n ϭ 5) (Fig. 1D). Before TEA application, these neurons exhibited spiking patterns that are characteristic of fast-spiking neurons, which are nearly constant AP amplitudes, uniform interspike intervals, and high firing frequencies (Fig. 1D). The TEA application broadened AP duration; this was especially prominent in subsequent APs (Fig. 1D). Due to broadening of the duration of previous APs, initiation of the 5th AP was much delayed and its amplitude also decreased (Fig. 1D). However, it is important to note that TEA does block other Kv channels at 1 mM concentration, although with less efficacy compared with Kv3.
To further examine the potential role of Kv3.1 channels, we combined the current-clamp recording with post hoc immunostaining. Neurons were imaged with transmitted light during recording, immediately fixed afterward, and stained for endogenous Kv3.1b and NaV channels. A slow-spiking neuron often did not express any Kv3.1b at all (Fig. 1E), whereas a fast-spiking neuron expressed endogenous Kv3.1b in neuronal soma and proximal axons (Fig. 1F). Expression levels of Kv3.1b along proximal axons correlate with the maximal spiking frequency (Fig. 1G). Overall, the percentage of fast-spiking-like neurons was low (fast, ϳ11%, Ͼ31 Hz; medium ϳ28%, 21-30 Hz; slow, ϳ61%, Ͻ20 Hz; n ϭ 218), consistent with the staining result. Taken together, our study is consistent with the notion that Kv3 channels play a critical role in generating fast spiking within neurons (28).
Expression of Kv3.1b Effectively Converts Slow-spiking Young Neurons into Fast-spiking Neurons-Next, we examined whether Kv3 expression is sufficient to induce fast spiking by using hippocampal neurons from the E18 culture. At different developmental stages (from 4 to 22 DIV), there was maturation of neuron morphology, with increased axonal and dendritic length and branching patterns (supplemental Fig. S1). This was accompanied by APs displaying increased amplitude and shortened duration (supplemental Fig. S1), which is consistent with the AP maturation revealed from Xenopus spinal neurons and rodent auditory neurons, where increased expression of NaV and Kv channels plays a major role (39 -42). We transfected various Kv channel constructs into neurons at 5 DIV and then recorded them from 7 to 10 DIV. This is because there were sufficient NaV channels (supplemental Fig. S2). On the other hand, the expression levels of endogenous Kv channels is low and the outward currents increased by over 5-fold when Kv3.1 was transfected (control, 1.56 Ϯ 0.22 nA, n ϭ 10; Kv3.1aHA, 11.43 Ϯ 1.09 nA, n ϭ 10; Kv3.1bHA, 9.79 Ϯ 0.60 nA, n ϭ 6), rendering a relatively clean background (supplemental Fig. S2). Importantly, we have never observed fast spiking from these cultured neurons at this stage.
The Mechanism Underlying Kv3.1 Axonal Targeting and Its Impact on Spiking Frequency-We wondered whether the polarized targeting of Kv3.1 channels plays an important role in regulating the spiking frequency. Two splice variants of Kv3.1, Kv3.1a and Kv3.1b, differ in their polarized axon-dendrite targeting (16). Whereas Kv3.1a is restricted on somatodendritic membranes, Kv3.1b is more enriched on axonal membranes (16). To determine the impact of channel targeting on spiking frequency, we first made Kv3.1 channel mutants with unchanged channel activity but altered targeting. Our protein biochemistry and cell biology studies suggest that the splice One-way analysis of variance followed by a Dunn test, **, p Ͻ 0.01. E, relationship of injected current and AP firing frequency in neurons expressing different Kv channels, between Kv3.1aHA and the Kv3.1bHA t test: **, p Ͻ 0.01; *, p Ͻ 0.05. The p values were adjusted by the Holm-Bonferroni method from repeat measures of Kv3.1bHA. F, APs induced by short-pulse current injections from a control neuron and a Kv3.1bHA-expressing neuron. Short current pulses (2 ms; 800 pA) with 100 Hz frequency were injected into the neuronal soma. G, relationship of the frequency of short-pulse current injections and AP success rate. The "n" numbers are provided in parentheses.
Next, we determined the biophysical properties of channel deletion constructs using voltage-clamp recordings on transfected HEK293 cells. All deletion constructs except Kv3.1bHA 1-513 had unchanged channel properties compared with wild type Kv3.1bHA, including current amplitude, activation threshold, conductance-voltage relationship, and activation and deactivation time constants (supplemental Figs. S6 and S7). To determine their effects on spiking frequency, we transfected the three channel deletion constructs with

Kv3.1 Polarized Targeting and Spiking Frequency
unchanged activity into neurons. Kv3.1bHA 1-513 and previously identified mutants at the ATM (16) had altered channel activity and therefore were not further examined for their impact on the spiking frequency (supplemental Fig. S7). Inter-estingly, Kv3.1bHA 1-554 and Kv3.1bHA 1-530 effectively enhanced the spiking frequency induced by prolonged current injection just like Kv3.1bHA, whereas Kv3.1bHA 1-502 functioned as Kv3.1aHA (Fig. 4F and supplemental Fig. S8A). No dramatic difference was observed among these channel constructs using the short-pulse current injection protocol (Fig. 4G  and supplemental Fig. S8B). Therefore, these results support that axonal targeting of Kv3.1 channels significantly increases the spiking frequency. However, the truncation strategy cannot rule out the possibility that an unidentified binding protein to the deleted region of the splice domain may regulate channel activity.
After comparing the C termini of Kv3.1a and Kv3.1b, we found a striking difference, the net charge of the residues. The net charge between residues 501 and 530 in Kv3.1b is Ϫ5, whereas the net charge of the last 10 residues in Kv3.1a (residues 502-511) is ϩ3 (Fig. 4A). Given the net charge of the C-terminal ATM is ϩ8, we wondered if the electrostatic interaction plays an important role in the regulation of the N/Cterminal interaction by the splice domain, which may be essen-  Fig. 2E. G, summary of firing frequency following the short-pulse and high frequency inputs.
Next, we examined the effects of these point mutations on spiking frequency. The expression of Kv3.1bHA DE-KK failed to enhance the maximal firing frequency (Fig. 5, C and D). In sharp contrast, both Kv3.1bHA ED-KK and Kv3.1bHA ED-AA were similar to Kv3.1bHA in enhancing the maximal spiking frequency (Fig. 5, C and D). Interestingly, Kv3.1bHA DE-AA , similar to Kv3.1aHA in targeting, did not markedly enhance the maximal spiking frequency as effectively as Kv3.1bHA (Fig. 5, C and D).
We further analyzed the channel biophysical properties of these mutants expressed in HEK293 cells. Kv3.1bHA DE-KK indeed had markedly smaller current and slower activation time constants, whereas the other three mutants had identical channel biophysical properties as the wild type Kv3.1bHA (Fig. 6). Taken together, our mutagenesis studies (Figs. 4 -6) suggest that electrostatic attraction and repulsion of the splice domain to the N-terminal T1 domains regulate the masking and unmasking of the C-terminal ATM. The axonal targeting of Kv3.1 channels is critical for the maximal AP firing frequency.
Locally Blocking Kv3.1b in Proximal Axons but Not Dendrites Decreases the Spiking Frequency-To determine whether suppressing Kv3.1 channel activity on either axonal or dendritic membranes affects spiking frequency, we applied 1 mM TEA using a fine-opening glass pipette to the neurons expressing Kv3.1b and displaying fast spiking (Fig. 7A). When 1 mM TEA was applied to a large area including soma, as well as proximal dendrites and axon, the spiking frequency was markedly reduced and could partially recover after washing (control as 100%; TEA, 15.90 Ϯ 3.89%; wash, 42.63 Ϯ 19.78%) (Fig. 7, B and  G). When TEA was applied to the proximal axon, the spiking frequency was also markedly reduced and the inhibitory effect was reversible (control as 100%; TEA, 46.50 Ϯ 7.37%; wash, 89.87 Ϯ 3.16%) (Fig. 7, C, D, and G). In contrast, the application of TEA at the proximal dendrite had no effect on the spiking frequency (control as 100%; TEA, 93.14 Ϯ 3.13%; wash, 99.55 Ϯ 4.74%) (Fig. 7, E-G). This result further supports our conclusion that Kv3.1 channels along the proximal axon, including the AIS, are the most critical for the fast spiking of neurons.
Simulations Reveal the Effect of Kv3 Polarized Targeting on Spiking Frequency-To further elucidate whether the axonal targeting of Kv3.1 channels is critical for fast spiking, we per-

Kv3.1 Polarized Targeting and Spiking Frequency
formed simulation analysis. By using the NEURON software, we built three model neurons with identical geometry and NaV and leaky channel distribution, but with different amounts and localization of Kv3.1 channels. In the control neuron that has NaV channels at the AIS but no Kv3.1 channels, a simulated prolonged depolarization induced only a small number of APs (30 Hz with 150 pA current injected) (Fig. 8, A and B). In the dendritic Kv3.1 neuron that has Kv3.1 channels mainly enriched on somatodendritic membranes, a simulated depolarization induced more APs (80 Hz with 150 pA current injected) (Fig. 8, A and B). Importantly, in the axonal Kv3.1 neuron that has Kv3.1 channels enriched on axonal membranes, a simulated depolarization induced markedly more APs (110 Hz with 150 pA current injected) (Fig. 8, A and B). Qualitatively similar results were obtained with varied absolute values of current densities of NaV, Kv3.1, and leaky channels (see details under "Experimental Procedures"). Therefore, the simulation results suggest that Kv3.1 axonal targeting is a powerful means to increase the maximal spiking frequency, consistent with our experimental results.

DISCUSSION
In this study, we show that both biophysical properties and axonal targeting of Kv3.1 channels are critical to enable neurons to fire APs at the maximal frequency (Fig. 8C). We further reveal novel mechanistic insights into Kv3.1 axonal targeting, which includes electrostatic interactions between the N/C termini of the channel (Fig. 5A).
The maximal spiking frequency requires both rapid biophysical kinetics and axonal targeting. Blocking Kv3 effectively decreased the spiking frequency in both mature neurons and young neurons expressing Kv3.1bHA (Figs. 1 and 7), consistent with the critical role of Kv3 channels in fast-spiking neurons. The uniquely high activation threshold (about Ϫ20 mV, although other Kv channels about Ϫ40 mV) and fast deactivation kinetics (at least 10 times faster than other Kv channels) of Kv3 channels are thought to be essential for fast spiking (28). Our results are consistent with this notion and further show that expression of axonal Kv3.1 channels is sufficient to induce fast spiking in young hippocampal neurons (Fig. 2). In sharp contrast, none of the neurons expressing YFP-Kv1.2 or YFP-Kv4.2 displayed fast spiking (Fig. 2), although both channels were expressed on plasma membranes (20,22) and provided large outward currents (Fig. 3). Interestingly, YFP-Kv1.2 had a similar activation threshold with Kv3.1bHA, but much slower in activation and deactivation (Fig. 3). YFP-Kv1.2 expression did somewhat increase the spiking frequency, but never induced fast spiking (Fig. 2). Furthermore, a Kv3.1b mutant, Kv3.1bHA DE-KK , displaying markedly increased activation time constant and threshold, but unaltered deactivation (Fig. 6), also failed to induce fast spiking (Fig. 5). Therefore, both activation and deactivation kinetics are critical for fast spiking. As expected, YFP-Kv4.2, displaying low activation threshold, and slow activation and deactivation kinetics (Fig. 3), did not increase the spiking frequency at all (Fig. 2).
Having the right channel biophysical properties alone is not sufficient for effective fast spiking. Kv3.1a has the same channel biophysical properties as Kv3.1b, but fails to induce fast spiking as effectively as Kv3.1bHA. The maximal range of spiking frequency requires axonal targeting of Kv3.1 channels, especially localization at the AIS. The channel density at the AIS was much less for Kv3.1aHA (16). Furthermore, the Kv3.1bHA mutations, Kv3.1bHA 1-530 , Kv3.1bHA 1-554 , Kv3.1bHA ED-KK , and Kv3.1bHA ED-AA , with unchanged levels at the AIS can induce fast spiking, whereas the mutants, Kv3.1bHA 1-502 and Kv3.1bHA DE-AA , with unchanged channel activity but decreased AIS levels, failed to effectively increase the maximal spiking frequency (Figs. 4 and 5). This result is consistent with our hypothesis. Furthermore, blocking Kv3.1 activity locally at proximal axons but not dendrites reduced fast spiking (Fig. 7), also supporting our hypothesis. Finally, simulations show that localization of Kv3 channels at the proximal axon including the AIS is critical for inducing the maximal spiking frequency (Fig.  8). Because APs are initiated at the AIS in most neurons, and this is where NaV channels are concentrated, the presence of Kv3.1 channels at the AIS allows them to more effectively regulate spiking frequency. Somatodendritic Kv3.1aHA still increases the spiking frequency; this may be due to either: 1) the electrical field passively spreading from the soma, or 2) low levels of Kv3.1aHA existing at the AIS. Nonetheless, because many K ϩ channels have cytosolic ancillary subunits (47), we cannot rule out the possibility of an unidentified binding protein to differentially regulate channel activities of Kv3.1a and Kv3.1b in neurons.
Why was no significant difference observed between Kv3.1aHA-or Kv3.1bHA-expressing neurons when APs were induced by transient current injections? Both enabled neurons to follow high frequency stimulations (Fig. 2). Less Kv3.1 channel activity or an attenuated electrical field, provided by somatodendritic Kv3.1 channel activity, might be sufficient for transient stimulations. Alternatively, the protocol we used here may not be sensitive enough to detect the difference.
Both experimental data and computer simulations are consistent with the major conclusion that the axonal targeting of Kv3.1 channels regulates the maximal spiking frequency; however, there are minor differences between these two approaches. It is important to note that a native central neuron often expresses multiple types of voltage-gated ion channels, which may regulate the effectiveness of the function of the Kv3 channels in generating fast spiking. Our simulations were performed only with three types of channels: NaV, Kv3.1, and a leak (Fig. 8, A and B). The major conclusion of the simulations is that axonal Kv3.1 channels allow the neuron to fire at higher frequencies than dendritic ones (Fig. 8, A and B); this is consistent with the experimental data (Figs. 2, 4, 5, and 7). However, there are two differences between the experimental data and the simulations. First, the maximal spiking frequency for the axonal-Kv3.1-expressing neuron in the simulations is about 120 Hz (Fig. 8B), whereas it is less than 60 Hz on average, for Kv3.1bHA-expressing neurons (Fig. 2E). Second, at the low spiking frequency, the effect of expressed Kv3.1bHA is not as strong as predicted by simulations. It is likely that smaller effects are difficult to show statistically in experimental results. More importantly, because, 1) Kv3.1 channels are activated at a higher membrane potential compared with Kv1, Kv2, and Kv4 channels and 2) the need for Kv channel activity to repolarize the membrane potential is less at the low current injection condition, the endogenous Kv currents may play a bigger role (supplemental Fig. S2). Recent studies show that AIS localization of NaV and calcium channels regulate AP initiation, waveform, and unidirectional propagation (48 -52). Kv1 channels at the AIS of pyramidal neurons and interneurons regulate AP waveform and initiation (53,54). How different channels at the AIS affecting different aspects of APs interact with the Kv3 channels functionally is an interesting question for future investigation.
Electrostatic interaction of the Kv3.1 N/C termini regulates polarized axon-dendrite targeting via exposing the ATM. Our previous studies suggested that the Kv3.1b splice domain weakens the masking effect of the N-terminal T1 domains onto the ATM. Subsequently, the exposed ATM is recognized by ankyrin G at the AIS and directs Kv3.1b targeting into axons (16). How the splice domain interferes with T1-ATM binding was not known. In the present study, using mutagenesis we have identified a novel electrostatic interaction of the Kv3.1 N/C termini interfered by the splice domain. First, using deletion analysis, we mapped the critical region in the splice domain from residues 513 to 530 (Fig. 4). This region (Ϫ5) and the last 10 residues of Kv3.1a (ϩ3) are completely opposite in their net charges (Fig. 4A). Interestingly, the ATM in the C-terminal region had 8 positive net charges (16), which could mediate the masking effect by interacting with negative charges on the surface of the T1 domains, which contains a number of acidic residues. In Kv3.1a, the 3 positive charges of the C terminus may enhance T1-ATM binding to facilitate the masking effect of the ATM, whereas in Kv3.1b the negative charges within the splice domain repels the negatively charged T1 and weakens the masking of the ATM (Fig. 5A). The conclusion was supported by Kv3.1bHA point mutants (Figs. 5 and 6). This model, involving intramolecular interactions, is novel and deserves rigorous testing by other approaches in future studies, such as x-ray crystallography. Because we previously showed that the Kv3.1 N-/C-terminal interaction requires Zn 2ϩ , the potential interaction of the splice domain and the Zn 2ϩ -binding site is also an interesting question. Furthermore, the complex dynamic movement of membrane domains of Kv channels has been revealed (55). Our results that mutating D529E/D530E but not E517D/E518D affected Kv3.1bHA axonal targeting suggests an important role of conformational change(s), in addition to the electrostatic interaction, during the N-/C-terminal interaction and exposure of the ATM. This result is also consistent with a recent study showing the complex binding surfaces of the N-/C-terminal interaction in Kir2.1 channels (56).
Altering the spiking frequency via regulating Kv3 polarized targeting may underlie important physiological functions. Kv3mediated fast spiking is critical for sensory and reflex functions in the central nervous system, for instance, in the sound location by the auditory circuit and in stabilizing images on the retina by the vestibule-ocular reflex circuit (57)(58)(59). Fast-spiking GABAergic interneurons in the hippocampus and cortex play important roles in learning and memory, and are implicated in epilepsy, schizophrenia, and drug addiction (38,60). Prolonged and transient alteration of the spiking frequency can lead to profound physiological consequences. Regulation of the levels of Kv3 channel proteins and phosphorylation has been proposed as an underlying mechanism in adjusting the spiking frequency of neurons in the suprachiasmatic nucleus and auditory circuits following the circadian rhythm and different acoustic environments, respectively (34,61). In this study, we show that Kv3.1 channel protein targeting is a novel strategy for regulating spiking frequency, which may be the basis for development-regulated firing rates. The expression of two splice variants of Kv3.1, Kv3.1a and Kv3.1b, is developmentally regulated (62). As a result, with neuronal maturation, dendritic Kv3.1 channels shifts to axonal ones (23) to enhance the spiking frequency. In future studies, it is of interest to identify the potential activity-dependent regulation of neuronal spiking via altering Kv3 polarized targeting.