The probands are sisters, age 27 and 25, who both presented with epilepsy around 6 weeks of life. Seizures were described as eye fluttering during the transition between wakefulness and sleep. EEG showed generalized and multifocal epileptiform abnormalities. Seizure types consist mostly of atypical absence seizures with activity arrest and stereotypical ictal behavior including eye flutter, drooling, and head drop, and infrequent myoclonic seizures. Both were diagnosed with DEE, hypotonic cerebral palsy, autism, and intellectual disability, and ultimately failed multiple treatments for epilepsy, including phenobarbital, ACTH, prednisolone, carbamazepine, rufinamide, clobazam, felbamate, and Epidiolex. Their seizures partially responded to the ketogenic diet and valproic acid. Brain MRI of sibling #1 was normal except a cavum septum pellucidum, and sibling #2 had bilateral ventriculomegaly, prominent in the posterior horns and trigones of the lateral ventricles with slight thinning of the posterior body of the corpus callosum. There was no family history of epilepsy. Diagnostic genetic testing, which was not available until they were nearly adults, demonstrated de novo KCNC2 missense variant resulting in a V473A substitution in both siblings as previously identified (Mehinovic et al., 2022). Developmentally, sibling #1 walked independently at age 6, uses gestures, vocalizations, and augmentative devices for communication, and follows 2-3 step commands. Sibling #2 walked independently at age 2 and communicates with short sentences and follows multi-step commands.

Treatment with fluoxetine was initiated in sibling #1 at age 26, Her atypical absence seizures occurred 3–5 times/day with incontinence and post-ictal period in the month prior to starting fluoxetine 20 mg/day and were reduced to 0–1 times/day with no incontinence or post-ictal period during the following month. The seizure frequency was anecdotal and not formally quantified by either caregiver or provider. By the family’s report, she also demonstrated marked improvement in stamina that eliminated the need for an adaptive stroller, verbalizations, mood, and new skill development, including opening doors and initiating independent ambulation. Overnight EEG prior to fluoxetine, at age 25, showed nearly continuous 3–4 Hz frontally predominant generalized discharges, with recording of 2 atypical absence seizures associated with electrodecrement (Figure 1A). A routine EEG (age 27) was improved and showed no seizures (Figure 1B). After fluoxetine was initiated, she was able to discontinue Epidiolex and reduce Depakote and Diazepam (Figure 1C).

Treatment with fluoxetine 20 mg was initiated in sibling #2 at age 21 for impulsivity and irritability. Seizure frequency was reduced from 1-2 seizures/day in the month prior to 1-2 seizures/week during the following year. The seizure frequency was anecdotal and not formally quantified. By the family’s report, she had marked improvement in expressive communication, alertness, and reduction in irritability, outbursts, and aggression. Prior to fluoxetine, at age 20, overnight EEG showed nearly continuous 3–4 Hz frontally predominant generalized discharges and 6 brief electroclinical atypical absence seizures (Figure 1D). Overnight EEG at age 25 showed slow background activity with no epileptiform abnormalities or seizures (Figure 1E). Ethosuximide, Epidiolex, and risperdone were discontinued, and diazepam and valproic acid dosages were reduced (Figure 1F). Higher doses of fluoxetine were tried but were not tolerated due to irritability.

The KCNC2 V473A de novo variant is located within the sixth membrane spanning domain (S6) that supports the contour of the internal vestibule and is allosterically linked to the voltage sensor, regulating the voltage-dependent opening and closing of the channel (Figure 2A). The channel’s main activation gate, which closes off the potassium conductance pathway, is located at the helical bundle crossing region of the S6 segments on the cytoplasmic face of the channel (Chi et al., 2022; Kim and Nimigean, 2016; Mukherjee et al., 2022). Notably, the proline-valine-proline (PVP) motif, also called the S6 gating hinge, is located between 468 and 470. The V473A mutation is located very close to the bundle crossing (Figure 2B) which is involved in the mechanical flexibility of the pore and regulation of the activation energy necessary to open the channel. Kv3.2, encoded by KCNC2, is a highly similar paralog of Kv3.1 (KCNC1) and has an identical S6 amino acid sequence. In fact, there are only 2/174 amino acid differences in sequence inclusive of S4 through S6 so the structure is assumed to be similar. The residue position V473 in the Kv3.2 sequence is equivalent to residue V436 in Kv3.1 because of a length difference in an amino acid segment near the amino terminal that shifts the residues by 37. While not all GOF variants in Kv3.1 and Kv3.2 are located in this region, the clustering of GOF mutations in S6 (Figure 2A) may indicate that GOF mutations in this region shift the equilibrium among the open and closed states to favor the open states.

To investigate the functional effect of the siblings KCNC2 V473A variant, we used heterologous channel expression in mammalian HEK293 cells together with whole cell patch clamp recording and found that currents from human WT Kv3.2 channels activate with an inflection delay (Figure 3A) and remain activated for the duration of the examined depolarizations, as described previously (Kaczmarek and Zhang, 2017; Lien and Jonas, 2003). The half-activation voltage (V_1/2) of the WT Kv3.2 currents was −1.6 ± 1.4 mV, which is more positive than that of most other voltage gated channels (Lien and Jonas, 2003; Pak et al., 1991a; Pak et al., 1991b). A V_1/2 near 0 mV for the WT Kv3.2 channels would permit the rapid unimpeded rise of the action potential (Lien and Jonas, 2003). In comparison, currents associated with the pathogenic Kv3.2-V473A variant differed markedly from the WT Kv3.2 currents. Currents for the Kv3.2-V473A variant rose more rapidly and the voltage for half-activation of current was left-shifted by more than 50 mV along the voltage axis to more negative potentials compared to WT Kv3.2 currents.

Because all reported KCNC2 pathogenic variants, including the Kv3.2-V473A variant described here, occur in the heterozygous state, random assembly of WT and variant subunits would be expected to result in five types of channels with different stoichiometries in an individual patient. Three of the five types would be hybrid channels comprised of both WT and variant subunits in various combinations, with lessor numbers of WT and homomeric mutant channels (Supplementary Figure S1; MacKinnon, 1991; Blaine and Ribera, 1998). To mimic a heterozygous variant, HEK cells were transfected 1:1 with equal amounts of WT Kv3.2 and variant Kv3.2-V473A expression vectors. As might be expected for these 1:1 “heterozygous” transfections, the V1/2 for the G/V curve (Figure 3A, blue line in plotted data) falls between that of the WT Kv3.2 current (black line) and the Kv3.2-V473A currents (red line), likely mimicking the currents present in cells of the heterozygous sisters. These currents were well described by a single Boltzmann function (blue line). We showed previously that currents generated by the random assembly of WT and variant subunits of an analogous GOF BK channel mutation produced a mix of five channel stoichiometries where the G/V curves were well approximated by a single Boltzmann function (Geng et al., 2023). In contrast, the G/V curve of the hybrid currents was less well described by a single Boltzmann function if we assumed that the 1:1 currents did not form hybrid currents and arose from two separate populations of channels, half homomeric WT channels and half homomeric variant channels (Figure 3A plotted data, green curve).

Channels formed exclusively from four mutant subunits express a current with an initial component of inactivavation. Significantly, such inactivation is not seen in the 1 to 1 expression of the mutant and WT subunits. Thus, such inactivation is not likely to be a significant component of the potassium current in the heterozygous patients which may have a mix of channels with stoichiometries resembling those represented in Supplementary Figure S1. It is possible that only a small percent of channels in the patients (∼6%) consisting of four mutant subunits have such inactivation.

In addition to their different voltage range of activation, the Kv3.2-V473A variant currents activated 5-fold more rapidly than WT Kv3.2 currents (Figure 3B) following depolarizing voltage steps and then deactivated 5.6-fold more slowly than WT Kv3.2 currents following negative voltage steps (Figure 3C). The negative shift in the voltage range of activation, 5-fold faster activation and 5.6-fold slower deactivation, would all increase the time that Kv3.2-V473A variant currents were active, indicating that Kv3.2-V473A is a gain of function variant. These changes are consistent with a shift in the gating properties of the variant channels favoring the open state.

Previous work demonstrated that fluoxetine resulted in clinical improvement in a patient with a de novo KCNC1 GOF missense variant (V425M) (Ambrosino et al., 2023), and inhibited both WT and mutant Kv3.1 channels (as well as hybrid channels), with a slight bias toward inhibition of the variant current (Ambrosino et al., 2023). Because KCNC1 and KCNC2, the paralogous genes encoding the Kv3.1 and Kv3.2 Shaw family potassium channels, share large regions of identical structure and closely overlapping anatomical distributions in the brain, we decided to investigate whether there might be a similar channel blocking effect of fluoxetine on Kv3.2 currents, which might be clinically useful for individuals with KCNC2 GOF variants. As shown in Figures 4A, C, fluoxetine similarly blocked the K⁺ currents carried by homomeric WT Kv3.2 channels, homomeric V473A channels, and channels assembled from WT and V473A subunits to mimic a heterozygous mutation, where most of the channels could be heteromeric (Geng et al., 2023). Thus, the effect of fluoxetine on Kv3.2 channels is similar to that on their close paralogue Kv3.1.

Norfluoxetine, a metabolite of fluoxetine (Eli-Lilly and Company, 2006), has been shown to be considerably more effective in blocking Kv3.1 channels than fluoxetine itself (Choi et al., 2001). Fluoxetine is converted into norfluoxetine in the liver by removal of a methyl group and only a small percentage of both agents (<5%) is estimated to be eliminated by the kidney (Eli-Lilly and Company, 2006). In patch clamp experiments using the heterologous HEK293 cell expression system, we found that norfluoxetine was indeed a more potent blocker of Kv3.2 currents than fluoxetine, by a factor of about 3.6 for WT currents (Figure 4). Furthermore, norfluoxetine discriminates between WT and variant Kv3.2 currents; the inhibition by norfluoxetine on variant currents is approximately seven times greater than on WT currents (Figure 4).

Kv3.2 and also Kv3.1 channels activate (open) at voltages that are far more positive than the action potential threshold (4). Indeed, these channels have a specialized mechanism of conductance called a “resurgent” current in which the channels, upon depolarization, rapidly transit into a specialized closed state and then rapidly open on the downward (repolarizing) phase of the action potential (Labro et al., 2015). The channels then rapidly deactivate to minimize the refractory period of the action potential. This resurgent mechanism thus facilitates rapid repolarization while allowing the rapid regeneration of the next action potential. Rather than producing a delayed resurgent Kv3.2 current to repolarize the action potential, the mutant V473A channel produces a voltage dependent potassium current which activates at a more negative membrane voltage (Figure 3A) and activates more rapidly than WT Kv3.2 channels (Figure 3B). Thus, its more rapid activation at negative voltages supersedes the action of the resurgent current (Labro et al., 2015), which normally does not greatly contribute to action potential repolarization until the downward, repolarizing phase of the action potential. The left shift in activation, more rapid activation and decreased rate of deactivation of the pathological GOF variant Kv3.2 (V473A) (Figure 3) would all negatively alter the properties of fast spiking neurons by removing specializations in Kv3.2 that support fast spiking. To visualize the impact of Kv3.2V473A on the neuronal excitability of fast spiking neurons, we simulated the action potential response of a PV interneuron model embedded with WT or mutated Kv3.2 channels to current injection. The result shows that the pathogenic phenotype could both reduce the frequency of firing in a train of action potentials and potentially eliminate repetitive firing (Supplementary Figure S2; Murakoshi and Trimmer, 1999; Santi et al., 2006; Liu and Bean, 2014; Olah et al., 2022). Notably, the complete elimination of the Kv3.2 current in the simulation is less damaging to repetitive firing than including the pathological currents contributed by the GOF Kv3.2 (V473A) variant, which is consistent with the effects of a genetic deletion of the Kv3.2 current in mouse models (Lau et al., 2000).