SK2/3 CHANNELS COUPLE WITH T-TYPE CA 2+ CHANNELS TO GATE SPINAL LOCOMOTOR RHYTHM GENERATION

Locomotion, Spinal cord, Interneurons, Central pattern generator, Rhythm, Pacemaker, SK channels, T-type channels. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 2 SUMMARY Initiating locomotion requires central pa ttern generator (CPG) interneurons to transition from tonic firing to persistent sodium current (I NaP)-dependent bursting. While I NaP provides the rhythmogenic drive, the conductanc es gating this transition remain unclear. Here, we show that functional coupling between small-conductance calcium-activated potassium (SK2/3) channels and low-threshold T-type Ca 2+ channels, notably Cav3.2, gates locomotor rhythm generation. Pharmacological or genetic disruption of this SK-T-type axis triggers intrinsic bursting in Hb9 inter neurons, a genetically identified rhythmogenic population of the locomotor CPG, and initiates fictive locomotion, whereas SK activation silences ongoing rhythmic output. Immunohistochemical co-expression of SK2, SK3 and Cav3.2 in Hb9 interneurons provides an anatom ical basis for this functional coupling. Simulation-based inference further shows that, beyond this gating mechanism, burst diversity is primarily determined by the balance between I NaP and M-type potassium conductances. Together, these findings identify SK-T-type coupling as a tunable brake on CPG activation, defining a biophysical module that controls the initiation and termination of locomotor rhythmic activity, with potential relevance for rhythmogenic circuits beyond locomotion. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 3

Locomotor activity arises from spinal circuits organized into a central pattern generator (CPG) capable of producing rhythmic movement in the absence of sensory or supraspinal drive 1, 2. At the core of this rhythmogenic network lie glutamatergic interneurons 3, including a subset of pacemaker neurons that intrinsically generate membrane oscillations 4, 5 . These neurons rely on the persistent sodium current (I NaP) 6-9 whose critical role in locomotor rhythm generation is well established across vertebrate CPGs 6, 10-15 . However, while the "engine" of these rhythms, I NaP, is well-characterized, the mechanisms that gate the transition from silence to rhythmic locomotor output remain incompletely understood. During locomotor onset, extracellular calcium ([Ca 2+]o) and potassium ([K +]o) concentrations undergo fluctuations that promote the emergence of pacemaker bursting 16. These ionic changes enhance I NaP while reducing outward potassium currents that normally oppose it 16. While Nav1.1 and Nav1.6 driving I NaP are well-characterized 17, the potassium channels that restrain burst emergence and the subsequent initiation of the locomotor rhythm remain poorly defined. The Kv7.2-mediated M-current (I M) contributes to the regulation of locomotor speed 18, but it does not fully account for the potassium-dependent brake that limits CPG activation. Given the dynamic changes in [Ca 2+]o and [K +]o during locomotor onset 16, small- conductance calcium-activated (SK) channels are ideally positioned to convert these ionic fluctuations into changes in firing mode. SK channels are classically viewed as burst- terminating conductances that mediate spik e-triggered afterhyperpolarization (AHP) and regulate locomotor output 15, 19-21 . However, emerging evidence suggests a more complex role for these channels in initiating rhythmic activity 22, 23 , while the relevant isoforms, their subcellular organization, and their calcium triggers within the locomotor CPG remain unknown. Here, through a combination of patch-clamp recordings, isoform-specific viral knockdown, high-resolution immunohistochemistry, and state-of-the-art simulation-based inference, we challenge the classical view of SK channels as simple feedback regulators. We .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 4 show that functional coupling between SK2/3 channels and T-type Ca 2+ channels (notably Cav3.2) gates pacemaker bursting and locomotor rhythm initiation. These findings identify an atypical SK-T-type mechanism that may represent a broader biophysical module for gating oscillatory activity across rhythmogenic circuits. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 5

Potassium channels constrain INaP-driven burst expression. We first examined whether inhibiting potassium channels unmasks intrinsic bursting in interneurons from the rhythmogenic locomoto r CPG region (ventromedial lamina VII-VIII, L1-L2) by performing whole-cell recordings in spinal slices from neonatal mice (P5-12). In standard conditions (aCSF: [Ca2+]/i1 = 1.2 mM and [K+]/i1 = 3 mM), interneurons fired tonically and did not burst 6, 16 . However, when intracellular K + currents were blocked with a CsCl- based internal solution, bursting emerged in 70% of interneurons (21/30 cells; Fig. 1a). Bursts were characteri zed as long-lasting plateau-like depolariz ations (duration: 5.7 ± 0.9 s; amplitude: 37.9 ± 3.9 mV; area: 171.9 ± 50.2 mV·s), occurring at low frequency (0.2/i1 ±/i1 0.03/i1 Hz; Fig. 1b). Following individual spikes, the afterhyperpolarization (AHP) was reduced or absent, often replaced by a slow afterdepolarization (Fig. 1c, arrowhead) which could evolve into plateau potentials upon brief depolarizing steps (Fig. 1d). Tetrodotoxin (TTX; 0.5 µM) abolished bursting and eliminated plateau potentials prior to spike attenuation (Fig. 1d,e), consistent with the in volvement of a TTX-sensitive I NaP in this intrinsic bursting activity. SK channels restrain burst emergence in the rhythmogenic region of the locomotor CPG. Because intracellular Cs + attenuated the AHP and unmasked bursting, we tested whether Ca 2+-activated K ⁺ channels specifically constrain burst initiation. We applied selective blockers of their three main subtypes: BK (IbTx, 200 nM), IK (Tram-34, 5 µM), and SK (apamin, 100-200 nM) channels 24-26. While IbTx shortened AHP duration (P 0.05; Supplementary Fig. 1c) and broadened action potentials by slowing repol arization (P < 0.01; Supplementary Fig. 1d- e), effects consistent with BK channel function 27, it nonetheless failed to convert tonic spiking into bursting (0/11 cells; Supplementary Fig. 1g,h). Similarly, IK blockade with Tram- 34 had no effect on the AHP, spike waveform or firing mode (0/12 cells; Supplementary Fig. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 6 1a-h). These results indicate that while BK channels shape spike repolarization and AHP kinetics, neither BK nor IK channels gate burst initiation. In contrast, SK channel blockade with apamin induced rhythmic bursting in approximately half of the interneurons (36/73 cells; Fig. 2a-d), an effect accompanied by a reduction in AHP amplitude (P < 0.01; Fig. 2f,g) and duration (P < 0.001; Fig. 2f,h), alongside increased firing frequency (P < 0.001; Fig. 2f,i). This apamin-induced bursting was abolished by riluzole (5 µM), confirming its I NaP dependence (Fig. 2e). Compared to Cs+-evoked bursts, those induced by apamin were shorter (P < 0.05; Fig. 2j), displayed smaller amplitude and area (P < 0.001; Fig. 2k,l), yet occurred at a higher frequency (P < 0.001; Fig. 2m). A similar effect was observed in rats, where apamin triggered bursting in 50% of ventromedial interneurons (38/76 cells; Fig. 2b) with burst characteristics comparable to those recorded in mice (P > 0.05; Fig. 2n-q). To further validate SK involvement, we used UCL-1684, a broad-spectrum SK channel blocker 28, 29 structurally distinct from apamin 30. UCL-1684 induced dose-dependent bursting (Supplementary Fig. 2a-c), reaching ~50% of bursters at 1 µM (8/15 cells) with features similar to those evoked by apamin (P > 0.05; Supplementary Fig. 2d-g). Together, these findings indicate that SK channels are the main Ca 2+-activated K+ conductance limiting INaP-dependent bursting in interneurons of the spinal locomotor CPG region. SK2/3 isoforms gate bursting in CPG interneurons. Although apamin and UCL-1684 unmask rhythmic bursting, their limited selectivity across SK1-3 isoforms precludes precise subunit attribution. In rodents, SK1 is relatively insensitive to these blockers 31, suggesting the involvement of SK2 and/or SK3. We therefore applied tamapin, a peptide toxin with high affinity for SK2 and SK3 32. Tamapin induced bursting in a concentration-dependent manner , nearly doubling the fraction of bursting interneurons at 10 nM versus 5 nM and reaching ~50% of the recorded population (Fig. 3a,b,d). This bursting was accompanied by reduced AHP amplitude (P < 0.01; Fig. 3e,f) and duration (P < 0.05; Fig. 3e,g), alongside increased firing frequency (P < 0.01; Fig. 3e,h). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 7 Furthermore, burst parameters matched those evoked by apamin (P > 0.05; Fig. 3i- k). Notably, tamapin also promoted bursting in genetically identified Hb9 interneurons (3/7 cells; Fig. 3d), indicating that SK2/3-dependent bursting also occurs in a genetically identified rhythmogenic population of the locomotor CPG 6, 33. To further isolate the role of SK2, we used lei-dab7, a derivative selective for SK2 over SK3 34. Lei-dab7 induced bursting in 41% of interneurons (7/17 cells; Fig. 3c,d) and reproduced the classic SK-blockade signature: AHP amplitude and duration decreased (P < 0.01; Fig. 3 e-g) with increased firing frequency (P < 0.001; Fig. 3e,h). Lei-dab7-induced bursts were significantly longer than those evoked by apamin (P 0.05; Fig. 3j,k). To assess the specific contribution of SK3, for which no selective pharmacological blocker is available, we turned to RNA interference. An SK3-targeting shRNA, validated in HEK-293 cells (~45% protein reduction; Supplementary Fig. 3 a,b), was delivered intrathecally at birth (T 13-L1) via an AAV9 vector. While control (scrambled shRNA) eGFP + interneurons remained tonically active (n = 8 cells; Fig. 3l), SK3 knockdown promoted bursting in ~42% of cells (14/33 cells). SK3 knockdown triggered bursting without altering AHP properties (P > 0.05; Supplementary Fig. 3c-e). SK3 reduction revealed three bursting profiles: spontaneous bursting at rest (n = 6 cells; Fig. 3m), depolarization-induced bursting (n = 4 cells; Fig. 3n), and an elliptic bursting pattern (n = 4 cells; Fig. 3o). Overall, SK3- knockdown bursts were shorter (P < 0.001; Fig. 3p), smaller (P < 0.01; Fig. 3q), and occurred at a higher frequency (P < 0.001; Fig. 3r) than those triggered by apamin. Together, these findings identify SK2 and SK3 as the principal SK isoforms constraining INaP-driven bursting in locomotor CPG interneurons. Distinct somatodendritic organization of SK2 and SK3 subunits in Hb9 interneurons To determine whether SK2 and SK3 display distinct subcellular distributions in Hb9 interneurons, we mapped their localization by high-resolution confocal immunofluorescence in eGFP-labeled Hb9 interneurons (Fig. 4a). Both antibodies produced punctate labeling .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 8 organized into discrete clusters, distributed ac ross somatic and dendritic compartments (Fig. 4b,e,f). At the soma, however, SK2 and SK3 showed distinct distribution patterns. SK2 assembled into prominent membrane-associated clusters outlining the somatic contour (Fig. 4b), whereas SK3 labeling was more diffuse and less enriched at the somatic membrane. In line with this asymmetric distribution, somatic SK2 clusters were significantly larger (P < 0.001; Fig. 4 c) and displayed a higher density per unit membrane area (P < 0.001; Fig. 4d). Co-localized SK2/3 clusters were rare at the soma (3.3 %) and occurred at the lowest density compared with SK2- or SK3-only clusters (P < 0.001; Fig. 4d). Because standard immunolabeling does not allow unambiguous attribution of dendritic processes to individual Hb9 interneurons in spinal cord sections, we performed whole-cell recordings with biocytin filling followed by post hoc immunolabeling to expand the analysis to their dendrites (Fig. 4f). This approach revealed intermingled SK2 and SK3 clusters distributed along dendritic shafts (Fig. 4f and Supplementary Fig. 4a,b). In contrast to the soma, SK2 and SK3 cluster sizes were similar in dendrites (P > 0.05; Fig. 4g). However, their densities diverged, with SK3 clusters being significantly more abundant along dendrites than SK2 clusters (P < 0.001; Fig. 4h and Supplementary Fig. 4b). Co-localized SK2/3 clusters again displayed the lowest density (P < 0.05; Fig. 4h), representing 17% of total dendritic clusters. Together, these data reveal a clear spatia l segregation of SK subunits in Hb9 interneurons, with SK2 enriched at the soma and SK3 predominating along dendrites. SK channels bidirectionally gate rhythmic bursting dynamics. We next asked whether SK channels also shape the dynamics of ongoing oscillations once bursting is established. Under locomotor-like ionic conditions ([Ca 2+]o = 0.9 mM; [K +]o = 6 mM), which elicit bursting in ~50% of CPG interneurons 16, apamin potentiated burst output (Fig. 5a). Specifically, it increased intraburst spike frequency (P < 0.01; 21.2 ± 3.7 Hz vs 49.2 ± 8.9 Hz, n = 10 cells), as well as burst amplitude and duration (P < 0.01; Fig. 5b,c) while decreasing burst frequency (P < 0.05; Fig. 5d). These results indicate that SK channels limit .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 9 burst strength and duration while contributing to the pacing of rhythmic activity once bursting is established. To determine whether increasing SK activi ty produced opposite effects, we applied 1- EBIO (200 µM), a positive allosteric modulator that left-shifts the Ca 2+-activation curve of SK and IK channels 35. In standard aCSF, 1-EBIO increased AHP amplitude and duration while reducing firing frequency (P < 0.01; Fig. 5e-h). Under locomotor-like conditions, 1-EBIO strongly reduced bursting: some cells transitioned to tonic firing (Fig. 5i), whereas others exhibited reduced burst duration (P < 0.05; Fig. 5j) and amplitude (P 0.05 vs. control; Fig. 5i-k), demonstrating that the effects of 1-EBIO are largely mediated by SK rather than IK channels. SK modulation produced comparable bidirectional effects on I NaP-dependent bursting evoked by veratridine in standard aCSF (Supplementary Fig. 5). Specifically, apamin enhanced burst amplitude and duration (P < 0.05; Supplementary Fig. 5a,b), whereas 1-EBIO reduced both (P < 0.05; Supplementary Fig. 5c,d). Together, these results indicate a bidirectional control of I NaP-driven rhythmicity by SK channels at the single-cell level. To refine the relevant SK isoforms, we applied CyPPA (3 µM), a selective positive modulator of SK2/3 with negligible effects on SK1 or IK 35, 36 . CyPPA similarly suppressed bursting, promoting a burst-to-to nic conversion (P < 0.05; Fig. 5l-n). Subsequent application of apamin reinstated bursting (n = 2; Fig. 5 l), supporting the conclusion that SK2/3 channels bidirectionally regulate burst propensity in locomotor CPG interneurons, capable of both initiating and terminating rhythmic output depending on their activity level. SK2/3-mediated burst gating depends on T-type Ca2+ channel coupling. To identify the Ca 2+ source activating SK2/3 channels in CPG interneurons, we first blocked voltage-gated Ca 2+ entry with cadmium (100 µM). Cadmium induced rhythmic bursting in 52 % of interneurons (n = 11/21 ce lls), with burst duration, amplitude, and frequency similar to those observed with apamin (P > 0.05; Fig. 6a-e). Thus, blocking Ca 2+ .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 10 influx mimics SK blockade, suggesting that tonic Ca 2+ influx normally sustains SK activation to restrain bursting. We next examined which Ca 2+ channel subtypes contri bute to this effect. Pharmacological blockade of high-voltage- activated (HVA) channels, including L-type (nifedipine, 20 µM; n = 10 cells), R-type (SNX-482, 500 nM; n = 14 cells), or P/Q-type ( ω - agatoxin-IVA, 400 nM; n = 14 cells), failed to induce bursting (Fig. 6 b). Among HVA blockers, only ω -conotoxin MVIIC (1 µM; P/Q + N-type) triggered bursting in a small subset of cells (14%, n = 2/14), a significantly lower proporti on than observed with apamin or cadmium (P < 0.05; Fig. 6b), suggesting at most a minor contributi on from N-type channels. In contrast, inhibiting low-voltage-activated (LVA) T- type channels robustly unmasked bursting. Mibefradil (10 µM) or nickel (Ni 2+; 200 µM) elicited bursting in 45% (n = 5/11 cells) and 64% (n = 9/14 cells) of interneurons, respectively (Fig. 6b,f,g). These bursts were indistinguishable from those induced by apamin or cadmium (P > 0.05; Fig. 6c-e). A submaximal Ni 2+ concentration (100 µM) was sufficient to reduce AHP amplitude and duration (P < 0.01; Supplementary Fig. 6a,b), and induced bursting in a subset of interneurons (10 %; 2/21 cells; Supplementary Fig. 6c). Consistent with this pharmacological profile, Cav3.2, the most Ni 2+-sensitive T-type isoform 37, was detected as discrete somatic clusters on Hb9 interneurons (Supplementary Fig. 6d,e), providing an anatomical substrate for its functional coupling with SK channels. Notably, riluzole abolished bursting evoked by T- type blockers, confirming its INaP dependence (Fig. 6f,g). We then investigated whether SK activation depends on spatially restricted Ca 2+ signaling. Both the fast Ca 2+ chelator BAPTA (10 mM) and the slower chelator EGTA (10 mM) induced riluzole-sensitive bursting in 44% (7/21 cells) and 50% (3/6 cells) of interneurons, respectively (Fig. 6b,h,i). Chelation-induced bursts were shorter than apamin- evoked bursts (P 0.05; Fig. 6l,m), consistent with SK activation within spatially restricted Ca 2+ microdomains in close proximity to T-type channels. Finally, blocking ryanodine (dantrolene; 50 µM) or IP 3 receptors .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 11 (xestospongin C; 2.5 µM) failed to induce bursting (Fig. 6j), arguing against a major contribution from intracellular Ca2+ stores. Together, these results identify T-type channels, notably Cav3.2, as the principal Ca 2+ source mediating SK2/3 activation and thereb y gating intrinsic bursting in locomotor CPG interneurons. Apamin unmasks a spectrum of intrinsic oscillatory phenotypes. Apamin revealed heterogeneous intrinsic responses rather than a uniform bursting response. Regardless of the concentration us ed, approximately half of the interneurons switched to rhythmic bursting, while the others remained tonic with increased firing rates (Fig. 7a). Within the bursting population, burst waveforms were highly heterogeneous, prompting a systematic characterization of this variability. We applied principal component analysis (PCA) followed by hierarchical clustering to map this heterogeneity. The first two components explained 81% of the total variance (PC1: 48%; PC2: 33%), with PC1 primarily capturing waveform properties (amplitude and slopes) and PC2 tracking temporal dynamics (duration and frequency; Fig. 7b,c). Hierarchical clustering in PCA space (Supplementary Fig. 7a) identified three distinct oscillatory phenotypes (Silhouette score = 0.43; Clusters 1-3; Fig. 7d,f). Cluster 1 (orange) and Cluster 2 (green) shared similar temporal dynamics (duration, area, frequency; P > 0.05; Fig. 7g-i), but differed in waveform. Cluster 1 exhibited the smallest amplitudes (P < 0.01 to P < 0.001; Fig. 7d,j), whereas Cluster 2 displayed the steepest depolarizing and repolarizing slopes (P < 0.01; Fig. 7e,k,l). Cluster 3 (blue) represented a distinct phenotype characterized by slow, plateau-like bursts with the longest durations and lowest frequencies (P < 0.01; Fig. 7f,g-i). While Cluster 3 matched Cluster 2 in amplitude, its slope kinetics more closely resembled the slower dynamics of Cluster 1 (Fig. 7j-l). Together, these comparisons define three distinct oscillatory modes: small/rapid (Cluster 1), large/sharp (Cluster 2), and slow/plateau-like bursts (Cluster 3), each with a characteristic waveform and temporal signature. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 12 Notably, the apamin concentration did not alte r the cell distribution in PCA space (P > 0.05; Supplementary Fig. 7b), nor did it affect either prin cipal component axis (P > 0.05; Supplementary Fig. 7c,d). This indicates that the identified phenotypes reflect intrinsic cellular variability rather than dose-dependent effects of SK blockade, prompting us to investigate the conductance combinations associated with each oscillatory mode. Simulation-based inference reveals distinct ionic signatures of bursting phenotypes. To uncover the biophysical basis of burst diversity, we used simulation-based inference (SBI) to fit an extended Hodgkin-Huxley model 16, 18 to the three experimental phenotypes. The model included SK (g SK) and T-type calcium (gCaT) conductances (Table 1), with gSK fixed at zero to mimic apamin application. Inference was constrained by summary features extracted from the burst envelope (Supplementary Fig. 8a). While an initial six-conductance model captured Clusters 1 and 2 phenotypes, it failed to reproduce the slow/plateau dynamics of Cluster 3 (Supplementary Fig. 8b). Because slowing INaP deinactivation prolongs bursts 6, we allowed the maximal I NaP deinactivation time constant ( /g2028 /g3035/g3015/g3028/g3017/g3040/g3028/g3051 ) to vary as a free parameter. This resolved the mismatch and yielded accurate fits for all three phenotypes (Fig. 8a,b and Supplementary Fig. 8a). Posterior distributions (Fig. 8c) indicated that persistent sodium (g NaP) and M-type potassium (g M) conductances were the primary determinants of cluster separation. Pairwise posterior distributions further revealed a consistent positive correlation between these conductances across all clusters (Supplementary Fig. 9). The gNaP/gM ratio was significantly lower in Cluster 1 than in Clusters 2 and 3 (0.62 vs. 0.93; P 0.05; Fig. 8c), they differed in their absolute conductance regimes (P < 0.001; Fig. 8c), with Cluster 3 additionally requiring increased /g2028 /g3035/g3015/g3028/g3017/g3040/g3028/g3051 to reproduce its slow/plateau dynamics (Supplementary Fig. 8a). By contrast, g CaT showed broad posterior with no significant differences across the three clusters (P > 0.05; Fig. 8c), indicating a limited contribution to phenotype separation. The remaining conductances, including g K, gNa and gL, also varied across clusters, although to a lesser extent (Fig. 8c). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 13 In silico pharmacological perturbations confirm ed the predictive validity of the model. Using cluster-specific parameter values (Table 2), reintroducing g SK converted bursting to tonic spiking across all clusters (Fig. 8d), while subsequent g CaT suppression restored bursting (Fig. 8e), supporting T-type calcium channels as the principal Ca2+ source driving SK activation. Suppressing gNaP abolished bursting regardless of cluster identity (Supplementary Fig. 10a), and mimicking EGTA chelation under the SK-on condition reinstated it (Supplementary Fig. 10b), in agreement with the experimental results. Together, these simulations confirm the predi ctive validity of the inferred parameters and establish a functional hierarchy in which T-type-SK coupling gates burst expression, I NaP provides the core depolarizing drive, and phenot ypic diversity emerges from the balance between gNaP and gM, further shaped by INaP kinetics. SK2/3 and T-type channel coupling bidirectionally modulates the onset and termination of locomotor rhythms. We next examined whether the SK2/3-T-type coupling identified at the cellular level scales up to the network level to control locomotor rhythm generation. We recorded fictive locomotion from bilateral L 5 ventral roots in isolated neonatal mouse spinal cords while applying drugs focally to the rhythmogenic L1-L2 segments (Fig. 9a-d). During ongoing fictive locomotion induced by NMDA/5-HT (5/10 µM), focal SK activation with 1-EBIO (5 mM) rapidly and reversibly silenced locomotor output in all preparations (n = 9; Fig. 9a). Similarly, the SK2/3-selective activator CyPPA (0.3-0.4 mM), produced comparable suppression (n = 7; Fig. 9 b), indicating that SK2/3 activation is sufficient to terminate ongoing locomotor-like rhythms. Conversely, we tested whether SK2/3 inhibition could trigger rhythmicity. Under subthreshold NMDA/5-HT conditions (0-1/10 µM), insufficient on their own to generate locomotor output, focal application of tamapin (1 µM) reliably induced locomotor-like activity in all preparations tested (n = 7; Fig. 9c). Consistent with our cellular findings, focal .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 14 application of the T-type channel blocker Ni 2+ (10-20 µM) mimicked SK2/3 inhibition and likewise elicited locomotor-like activity (n = 7; Fig. 9d). Together, these results indicate that the SK2/3-T-type axis operated from single neurons to the CPG network. Inhibition of either component initiates locomotor-like activity, whereas SK2/3 activation terminates ongoing rhythmic output. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 15

Our study establishes that SK channels and T-type Ca 2+ channels are functionally coupled to gate locomotor rhythm generation in the hindlimb CPG. Converging pharmacological, genetic, immunohistochemical and computational evidence shows that inhibiting SK channels, or blocking T-type Ca 2+ influx that recruits them, converts tonically firing interneurons into bursters and facilitates rhythmic locomotor output. Conversely, enhancing SK activity suppresses bursting and terminates ongoing locomotor rhythms. The unmasking of bursting by SK block ade has been reported in a broad range of mammalian brain regions 38-49, as well as in invertebrate CPGs 50. In these systems, SK currents constrain excitability, such that their removal allows persistent inward currents to dominate and promote rhythmic bursting 38, 46, 49, 51, 52 . Our recordings support this general framework, showing that apamin-induced bursting depends on I NaP, in agreement with our previous work 4-6, 16-18 . In this respect, the mammalian spinal locomotor CPG appears to follow the same general principle of SK-dependent rhythm control. We identify SK2 and SK3 as the principal isoforms involved. Their expression in two major locomotor CPG populations 53, 54, namely Hb9 interneurons as shown here, and Shox2 interneurons as reported previously 55, positions them as key regulators of rhythm generation. Whereas all Hb9 interneurons co-express SK2 and SK3, only a subset of Shox2 interneurons expresses both isoforms 55, suggesting that distinct rhythmogenic populations may rely on partially different SK-dependent regul atory strategies. This view is further supported by evidence that SK channels contribute to multiple oscillatory regimes. In NMDA- driven bursting, they shape both burst initiation and termination 56-60, paralleling what we observe here for INaP-dependent bursting. Given that Hb9 interneurons can alternate between these two modes 4, SK channels are well positioned to constrain both regimes in rhythmogenic CPG neurons. Our data indicate that SK2 and SK3 are not simply redundant. Immunohistochemistry revealed a clear somatodendritic segregation, with SK2 enriched at the soma and SK3 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 16 predominating in dendrites. Although SK2/ SK3 heteromers have been described in heterologous systems 61, 62 , the limited SK2/SK3 co-localization observed in our double immunolabeling, together with the lack of co-immunoprecipitation in rat brain 63, argues for predominantly homomeric assemblies in situ. Functionally, reducing either isoform promoted bursting, but SK2 inhibition reduced mAHP amplitude whereas SK3 knockdown did not. This difference is consistent with the stronger somatic enrichment of SK2, which would be expected to contribute more directly to the mAHP recorded at the soma, in line with previous evidence that SK2 loss is sufficient to alter the mAHP 64 and with the broader view that somatic SK channels are major determinants of mAHP amplitude 65. By contrast, the predominantly dendritic localization of SK3 may make its contribution less apparent in this somatic readout. This functional dissociation is also consistent with earlier evidence for isoform-specific roles of SK channels in burst regulation in midbrain dopamine neurons 66-69, and with observations in spinal motoneurons, where somatic SK2 regulates baseline AHPs while SK3 predominance in slow-type motoneurons prolongs the AHP and limits firing rates 70, 71. Our data further identify transmembrane Ca 2+ influx, rather than intracellular Ca 2+ stores, as the main source of SK activation in locomotor CPG interneurons. The promotion of bursting by calcium chelators, together with the la ck of effect of intracellular store inhibition, argues that SK recruitment depends primarily on voltage-dependent Ca 2+ entry. Pharmacological dissection points to T-type Ca 2+ channels as the dominant source of this influx, since nickel and mibefradil both reduced AHPs and triggered bursting, whereas blockade of high-voltage-activated Ca 2+ channels was largely ineffective. This interpretation is supported by previous evidence for T-type currents in locomotor-related interneurons 55 and by reports that T-type antagonists slow locomotor rhythms 16, 72. Interestingly, the effects of T-type channel blockade closely parallel those of SK inhibition, both in locomotor network activity 19-21, 73 and in the bursting dynamics of CPG interneurons observed here. Together, these observations strongly support a functional coupling between T-type Ca 2+ channels and SK channels. Among T-type channel subtypes, Cav3.2 appears as a plausible partner, as we .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 17 confirm its expression in Hb9 interneurons, consistent with previous work 72, and because the high nickel sensitivity we observe is compatible with its known pharmacological profile 37, 74. Computational modeling further refined the hierarchy of mechanisms underlying burst generation and diversification. Although T-type Ca 2+ channels are essential for recruiting SK current and thereby gating burst expression, si mulation-based inference indicates that they do not primarily account for burst di versity. Instead, the balance between g NaP and g M emerged as the main axis separating burst clusters , consistent with earlier work showing that the I NaP/IM ratio is a critical determinant of bursting propensity 18. Reproducing the slow/plateau phenotype further required adjustment of I NaP deinactivation kinetics, in line with evidence that veratridine prolongs burst duration by slowing I NaP deinactivation 6. In this framework, T-type-SK coupling determines whet her bursting is permitted, whereas the burst phenotype is shaped primarily by the balance between I NaP and IM, further modulated by I NaP kinetics. SK channels are more commonly associated with high-threshold Ca 2+ sources across the CNS 39, 48, 75-77 , including spinal motor circuits 19, 78 . Our findings therefore reveal an atypical partnership between SK channels and T-type Ca 2+ channels, so far reported in midbrain dopaminergic and thalamic neurons 79, 80 , and extend this mode of coupling to the locomotor CPG. Mechanistically, mild depolarization may activate Cav3.2 channels at subthreshold voltages, producing a local Ca 2+ signal sufficient to recruit nearby SK conductances. In turn, SK activation would oppose burst emergence by reinforcing AHP- dependent restraint and limiting the depolarizing drive required for regenerative bursting. Consistent with this view, both apamin and T-type antagonists produce similar effects on locomotor output, including burst prolongation and rhythm slowing in lamprey and rodent spinal preparations 16, 19-23, 72 . Our data extend these observations by showing that local modulation of SK2/3 within the CPG can either initiate (SK inhibition) or terminate (SK potentiation) locomotor rhythms, while T-type Ca 2+ channel blockade similarly triggers locomotor activity. In vivo studies support this bidirectional control; increasing SK activity .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 18 reduces locomotion in mice 81, while SK inhibition in Xenopus embryos increases episode frequency and swimming duration 73. More broadly, SK-T-type coupling may provide a flexible mechanism through which locomotor output is adjusted to behavioral de mands. Monoamines illustrate this potential by converging on SK-mediated AHP regulation. Sero tonin reduces AHPs in CPG interneurons and motoneurons 82, 83, either through direct inhibition of SK channels 83, 84 or by limiting Ca2+ influx 85, 86 , thereby prolonging bursts and slowing the rhythm 87. Dopamine exerts similar effects 88, 89, and noradrenaline may do likewise by reducing Ca 2+ sensitivity of SK2 gating 90. Cholinergic C-bouton inputs onto motoneurons similarly engage muscarinic receptors near SK channels to reduce the AHP and enhance locomotor output 70, 91, 92 . Beyond neuromodulation, activity-dependent shifts in ex tracellular ion concentrations may provide an additional layer of control. Local increases in [K +]o and decreases in [Ca 2+]o at locomotor onset 16 would reduce the driving forces for both I KCa and I Ca, thereby attenuating SK- mediated inhibition and favoring burst emergenc e. Such ionic changes could therefore provide a rapid gating mechanism for locomotion initiation, complementing the slower neuromodulatory pathways. In summary, our data identify SK-T-type coup ling as a central mechanism controlling the initiation, modulation, and termination of rhythmic bursting in spinal locomotor CPGs. As a tunable brake on burst propensity, SK-T-type coupling defines the operational range of the locomotor network. More broadly, this coupling may represent a conserved biophysical motif for rhythmogenesis, coordinating oscillatory activity across motor systems beyond locomotion. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 19

Animal models. Experiments were performed on Wistar rats and CD-1 IGS mice of either sex. Hb9::eGFP transgenic mice were used for immunohistochemistry and identification of rhythmogenic interneurons. Animals were used at postnatal day (P) 0-2 for fictive locomotion experiments, P5-12 for standard patch-clamp recordings, and P12-20 for AAV-injected mice. Animals were housed under a 12 h light/dark cycle with ad libitum access to water and food at 21-24°C and 40-60% relative humidity. All experimental procedures were conducted in accordance with French regulations (Décret 2010-118) and approved by the local ethics committee (Comité d'Ethique en Experimentation Animale, CEEA-071, authorization No. B1301404 and protocol No. 17485-2018110819197361 and 50133-2024060612594852). shRNA construct. An shRNA sequence targeting mouse SK3 (Kcnn3; 5'- TGAGTGACTATGCTCTGATTT-3') was cloned into an AAV9 vector under control of the U6 promoter, with eGFP expression driven by a CMV promoter for visualization of transduced cells (VectorBuilder, Chicago, IL, USA). A scrambled non-targeting shRNA sequence (5'- CCTAAGGTTAAGTCGCCCTCG-3') with no homology to known mouse genes served as control. AAV9 particles were produced at titers ≥ 1 × 1013 genome copies (GC)/ml and diluted 1:10 in sterile phosphate-buffered saline prior to injection. Intrathecal vector delivery. Neonatal mice (P0) were cryoanesthetized and positioned dorsal side up. The intervertebral space was widened by gently flexing the spine. A glass microcapillary preloaded with diluted AAV9 particles was lowered into the center of the T 13-L1 intervertebral space under visual guidance. A total volume of 0.75 µl was slowly injected manually over ~30 s. Animals recovered on a heating pad and were returned to their home cage. Experiments were performed 12-20 days post-injection to allow adequate transgene expression and knockdown. In vitro preparations. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 20 Slice preparation. Lumbar spinal cords were isolated in ice-cold (4 °C) sucrose-based dissection solution containing (in mM): 252 sucrose, 3 KCl, 1.25 NaH 2PO4, 4 MgSO 4, 0.2 CaCl2, 25 NaHCO 3, 20 D-glucose (pH 7.4, bubbled with 95% O ₂ /5% CO ₂ ). The spinal cord was embedded in 1% low-melting-point agar, cooled, and mounted on a vibrating microtome (Leica VT1000S, Leica Biosystems, Wetzlar, Germany). Transverse slices (325 µm) were cut from L1-L2 segments and transferred to a holding chamber containing oxygenated (95% O /i1 /5% CO/i1 ) aCSF at 30-32 °C composed of (in mM): 120 NaCl, 3 KCl, 1.25 NaH 2PO4, 1.3 MgSO4, 1.2 CaCl 2, 25 NaHCO 3, 20 D-glucose (pH 7.4). This solution is referred to as standard aCSF ([Ca 2+]/i1 = 1.2 mM; [K +]/i1 = 3 mM). After 60 min recovery, slices were transferred to a recording chamber continuously perfused with standard aCSF at 32 °C (2-3 ml/min). Whole-spinal cord preparation. The spinal cord was transected at T4, isolated with intact dorsal and ventral roots, and transferred to a recording chamber. The preparation was continuously superfused with oxygenated (95% O ₂ /5% CO₂ ) aCSF at 25-26 °C containing (in mM): 120 NaCl, 4 KCl, 1.25 NaH2PO4, 1.3 MgSO4, 1.2 CaCl2, 25 NaHCO3, 20 D-glucose (pH 7.4). Cell culture. HEK293 cells were maintained in DMEM, high glucose (Gibco) supplemented with 10% fetal bovine serum and penicillin/streptomycin (Gibco), incubated at 37°C in a humidified atmosphere containing 5% CO2. HEK 293 cells were transiently transfected using Lipofectamine 3000 (Thermo Fisher) according to the manufacturer’s recommendation. Cells were cotransfected with 2 plasmids: one encoding mKcnn3 (VB241013-1085awr, VectorBuilder) and the other either encoding shRNA targeting mKcnn3 (VB211028-1097hcr) or a control plasmid encoding shRNA against luciferase. The cells were then collected 72 h post transfection. In vitro recordings. Patch-clamp recordings. Whole-cell patch-clamp recordings were performed on ventromedial interneurons in lamina VII-VIII of L 1-L2 segments, adjacent to the central canal, a region .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 21 enriched in rhythmogenic neurons. Neurons were visualized using infrared differential interference contrast (IR-DIC) videomicroscopy with a 40x water-immersion objective (Olympus) mounted on an upright microscope (Olympus BX51WI) equipped with an infrared- sensitive CCD camera. Hb9::eGFP-positive neur ons were identified by epifluorescence. Recordings were performed using a Multiclamp 700B amplifier and Digidata 1550B interface (Molecular Devices). Signals were digitized on-line and filtered at 10 kHz using pClamp 10.7 software (Molecular Devices). Patch pipettes (4-6 M Ω ) were pulled from borosilicate glass capillaries (1.5 mm outer diameter, 1.12 mm inner diameter; World Precision Instruments, TW150-4) using a horizontal puller (Sutter P-97). The standard intracellular solution contained (in mM): 140 K +-gluconate, 5 NaCl, 2 MgCl 2, 10 HEPES, 0.5 EGTA, 2 ATP, 0.4 GTP (pH 7.3; 280-290 mOsm). For Ca 2+ chelation experiments, BAPTA (10 mM) or EGTA (10 mM) replaced the standard 0.5 mM EGTA. In experiments testing the contribution of intracellular Ca2+ stores, xestospongin C (2.5 µM) was added to the intracellular solution to block IP3 receptors. For experiments requiring K + channel blockade, we use a CsCl-based solution contained (in mM): 120 CsCl, 40 KCl, 2 MgCl2, 10 HEPES, 0.5 EGTA, 2 ATP, 0.4 GTP, pH 7.3. For post hoc morphological identification, biocytin (0.2%) was included in the intracellular solution, and filled neurons were processed as described below (see Immunostaining). The pipette offset potential and capacitance were nulled before seal formation. Data acquisition began at least 5 min after break-in to ensure dialysis and stability. To induce rhythmic activity, the recording ch amber was perfused with a modified “locomotor- like” aCSF characterized by reduced calc ium and elevated potassium concentrations ([Ca 2+]/i1 = 0.9 mM and [K +]/i1 = 6 mM). This ionic environment mimics the extracellular fluctuations observed during the onset of locomotor activity and promotes pacemaker-like membrane oscillations. To block fast synaptic transmission, neurons were isolated from excitatory glutamatergic, glycinergic, and GABAergic inputs using kynurenic acid (1.5 mM) or a combination of CNQX (10 µM) and AP-5 (50 µM), strychnine (1 µM), and picrotoxin (100 µM), respectively. Bicuculline was omitted to avoid blocking the calcium-activated potassium .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 22 current underlying the afterhyperpolarization 93, thereby potentially facilitating burst firing in spinal interneurons 4. Extracellular recordings. Fictive locomotion was recorded from L 5 ventral roots using glass suction electrodes connected to AC-coupled amplifiers (A-M Systems, model 1700). Signals were amplified (1000×), band-pass filtered (70 Hz to 3 kHz), digitized at 5 kHz, and simultaneously rectified and integrated online (t = 100 ms) using custom-built integrators. Locomotor-like activity was induced by bath appl ication of NMDA (5 µM) and 5-HT (10 µM). Prior to pharmacological testing, all spinal cord preparations were first tested for their ability to generate locomotor-like activity in respons e to NMDA/5-HT. Only preparations showing robust rhythmic ventral root activity were used for subsequent experiments. For focal drug application to the rhythmogenic region, drugs were delivered via pressure ejection through a glass micropipette positioned over the L 1-L2 ventromedial region using a micromanipulator under visual control. The ejection solution contained neutral red to visualize diffusion, and perfusion flow was arranged to prevent drug spread to the ventral root recording sites. Immunostaining. Spinal cords from P14-P20 Hb9::eGFP mice were dissected and immersion-fixed for 1 h in 0.25-4% paraformaldehyde (PFA). Tissues were rinsed in phosphate-buffered saline (PBS), cryoprotected overnight in 20% sucrose at 4 °C, and embedded in OCT medium (Tissue-Tek). Transverse cryosections (30 µm) were collected from L 1-L2 segments. Sections were rehydrated in Tris-buffered Saline (TBS) for 15 min at room temperature, permeabilized in TBS containing 10% normal horse serum and 0.4% Triton X-100 for 1 h, and incubated overnight at 4°C in blocking solution (10% normal horse serum, 0.2% Triton X-100 in TBS) with primary antibodies. Primary antibodies included: polyclonal rabbit anti-SK2 (1:400; #APC-028, Alomone Labs), rabbit anti-SK3 (1:400; #APC-025, Alomone Labs), guinea-pig polyclonal anti-SK2 (1:400; #APC-028-GP, Alomone Labs), and polyclonal rabbit anti-Cav3.2 (1:400; #ACC-025, Alomone Labs). For colocalization assays, sections were co-incubated with guinea-pig .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 23 polyclonal anti-SK2 and rabbit anti-SK 3. Endogenous GFP was enhanced using goat polyclonal anti-GFP (1:400; #ab6673, Abcam). For biocytin-filled neurons, slices containing recorded cells were fixed immediately after recording in 4% PFA overnight at 4 °C, wa shed in PBS, and processed as free-floating sections with permeabilization extended to 6 h (0.5% Triton X-100). After incubation with primary antibodies, sections were washed in TBS and incubated for 1-2 h at room temperature with secondary antibodies: donkey anti-guinea-pig Cy3 (1:400; #706- 165-148, Jackson ImmunoResearch), donkey anti-rabbit Cy5 (1:400; #711-175-152, Jackson ImmunoResearch), donkey anti-rabbit Alexa Fluor 555 (1:400; #A31572, Invitrogen), or donkey anti-goat Alexa Fluor 488 (1:400-1:800; #A11055, Life Technologies; #705-545-147, Jackson ImmunoResearch). For biocytin vi sualization, Alexa Fluor 405-conjugated streptavidin (1:1000; #S32351, Invitrogen) was added. Sections were rinsed, mounted in aqueous mounting medium, and coverslipped. Confocal images were acquired on a Zeiss LSM700 using x20 and x40 oil-immersion objectives. Z-stacks were collected with 0.3-1 µm optical steps and processed in ZEN 12.0 (Zeiss). Laser power, gain, and offset were adjusted for each tissue section to maximize signal-to-noise ratio while avoiding saturation. SK3 channel protein quantification. Transfected HEK293 cells were lysed in ice-cold buffer containing 1% Igepal CA-630 and 0.1% SD S, supplemented with protease inhibitors (Complete Mini, Roche Diagnostics). Lysates were centrifuged at 10,000 rpm for 10 min at 4 °C, and the supernatant was used as the total pr otein fraction. Protein concentrations were determined using a detergent-compatible assay (Bio-Rad). Equal amounts of protein (40 µg per lane) were separated on 4-15% gradient SDS-PAGE stain-free gels (Bio-Rad), transferred to nitrocellulose membranes, and probed overnight at 4 °C with either a polyclonal rabbit anti-SK3 antibody (1:500, Alomone Labs, APC-025) or an anti-actin antibody (1:1,000, A2066, Sigma-Aldrich) in Tris -buffered saline containing 5% fat-free milk. Membranes were then incubated for 1 h at room temperature with a goat anti-rabbit horseradish peroxidase-conjugated secondar y antibody (1:40,000; Thermo Fisher). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 24 Immunoreactive bands were visualized using enhanced chemiluminescence detection (Merck-Millipore) and quantified with Image Lab software (Bio-Rad). Pharmacological agents. All pharmacological agents were prepared as concentrated stock solutions and diluted to their final concentra tions in oxygenated aCSF immediately before use. Compounds were purchased from the followi ng suppliers: Sigma-Aldrich: apamin (100- 200 nM), mibefradil (10 µM), NiCl 2 (200 µM), CdCl2 (100 µM), nifedipine (10 µM), veratridine (60 nM), BAPTA (10 mM), EGTA (10 mM), N-methyl-D-aspartate (NMDA, 5 µM), serotonin (5-HT, 10 µM), strychnine (1 µM), picrotoxin (100 µM), biocytin (0.1%), and all salts for aCSF as well as agar/agarose. Alomone: tamapin (5-10 nM), 1-EBIO (200 µM), CyPPA (1-10 µM), and Tram-34 (5 µM). Smartox Biotech: le i-dab7 (5-10 nM), iberiotoxin (200 nM), and ω - Agatoxin IVA (100 nM). Tocris: UCL-1684 (0.2-1 µM), dantrolene (10 µM), xestospongin C (2.5 µM), and riluzole (5 µM). Hello Bio: SNX-482 (100 nM) and kynurenic acid (1.5 mM). Computational modeling Hodgkin Huxley Cell Model. For this study we extended a previously described single- compartment Hodgkin-Huxley (HH) type neuronal model 16, 18. Membrane potential dynamics were governed by the following equation: /g1829 /g1856/g1848 /g1856/g1872 /g3404 /g3398 /g1835 /g3015/g3028 /g3398 /g1835 /g3015/g3028/g3017 /g3398 /g1835 /g3012 /g3398 /g1835 /g3014 /g3398 /g1835 /g3020/g3012 /g3398 /g1835 /g3004/g3028/g3021 /g3398 /g1835 /g3013 /g3397 /g1835 /g3036/g3041/g3037 where /g;848 is the membrane potential (mV), /g;835 denotes membrane currents (pA), and /g;859̄ /g3025 denotes maximal conductances (nS). Individual ionic currents were defined as: .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 25 /g1835 /g3015/g3028 /g3404/g1859 /g3015/g3028 /g1308./g1308 /g1865 /g3015/g3028 /g2871 /g1308./g1308/g1860 /g3015/g3028 /g1308. /g1308/g4666/g1848 /g3398 /g1831 /g3015/g3028 /g4667 /g1835 /g3015/g3028/g3017 /g3404/g1859 /g3015/g3028/g3017 /g1308./g1308 /g1865 /g3015/g3028/g3017 /g1308./g1308 /g1860 /g3015/g3028/g3017 /g1308./g1308 /g4666/g1848 /g3398 /g1831 /g3015/g3028 /g4667 /g1835 /g3012 /g3404/g1859 /g3012 /g1308./g1308/g1865 /g3012 /g2872 /g1308. /g1308/g4666/g1848 /g3398 /g1831 /g3012 /g4667 /g1835 /g3014 /g3404/g1859 /g3014 /g1308./g1308/g1865 /g3014 /g1308./g1308 /g4666/g1848 /g3398 /g1831 /g3012 /g4667 /g1835 /g3020/g3012 /g3404/g1859 /g3020/g3012 /g1308./g1308/g1865 /g3020/g3012 /g1308./g1308 /g4666/g1848 /g3398 /g1831 /g3012 /g4667 /g1835 /g3004/g3028/g3021 /g3404/g1859 /g3004/g3028/g3021 /g1308./g1308/g1865 /g3004/g3028/g3021 /g2870 /g1308./g1308 /g1860 /g3004/g3028/g3021 /g1308./g1308 /g4666/g1848 /g3398 /g1831 /g3004/g3028 /g4667 /g1835 /g3013 /g3404/g1859 /g3013 /g1308./g1308 /g4666/g1848 /g3398 /g1831 /g3013 /g4667 /g;83; /g3015/g3028 , /g;83; /g3012 and /g;83; /g3013 denote the reversal potentials for sodium, potassium, and leak currents, respectively. /g;83; /g3015/g3028 and /g;83; /g3012 were calculated using the Nernst equation, whereas /g;83; /g3013 was computed using the Goldman-Hodgkin-Katz equation based on extracellular (out) and intracellular (in) ionic concentrations. I onic concentrations were assumed constant throughout the simulation: /g4670/g;840/g;853 /g2878 /g467; /g3042/g3048/g3047 /g3404 145 mM; /g4670/g;840/g;853 /g2878 /g467; /g3036/g3041 /g34041 5 mM; /g4670/g;837 /g2878 /g467; /g3042/g3048/g3047 /g34043 mM; /g4670/g;837 /g2878 /g467; /g3036/g3041 /g3404 140 mM; /g4670/g;829/g;864 /g2879 /g467; /g3042/g3048/g3047 /g3404 130 mM; /g4670/g;829/g;864 /g2879 /g467; /g3036/g3041 /g34048 mM. Relative permeabilities of sodium and chloride ions (with respect to potassium ions) were set to /g2025 /g3015/g3028//g3012 /g34040 . 0 3 , /g2025 /g3004/g3039//g3012 /g34040 . 1 . Activation and inactivation variables followed standard first-order kinetics: /g1856/g1876 /g1856/g1872 /g3404 /g1876 /g2998 /g3398/g1876 /g2028 /g3051 Where /g;876 /g1488 /g4668/g;865, /g;860/g4669 , /g;876 /g2998 represents the voltage-dependent steady-state value, and /g2028 /g3051 is the voltage-dependent time constant. These quantities are defined as: /g1876 /g2998 /g3404 1 1/g3397e x p /g2879 /g4666/g3023/g2879/g3023 /g3299/g3117//g3118 /g4667 /g3038 /g3299 /g2028 /g3051 /g3404 /g2028 /g3051/g3040/g3028/g3051 cosh /g3436 /g3398/g4666/g1848 /g3398 /g1848 /g3051/g2869//g2870 /g4667 /g1863 /g3099 /g3299/g3288/g3276/g3299 /g3440 where, /g;848 /g3051/g2869//g2870 and /g;863 /g3051 denote the half-activation voltage and the slope factor, respectively, and /g2028 /g3051/g3040/g3028/g3051 and /g;863 /g3099 define the voltage dependence of the corresponding time constant. Activation .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 26 kinetics for /g;835 /g3015/g3028 and /g;835 /g3015/g3028/g3017 were assumed instantaneous ( /g2028 /g3040/g3015/g3028 /g;308/g3404/g;308/g2028 /g3040/g3015/g3028/g3017 /g;308/g3404 /g;308 0 ), such that /g;865 /g3015/g3028 and /g;865 /g3015/g3028/g3017 were set to steady-state values. Unlike the voltage-dependent channels above, SK-channel gating depended on intracellular calcium concentration rather than membrane volt age. Intracellular calcium dynamics and SK activation were described as: /g1856/g4670/g1829/g1853 /g2870/g2878 /g4671 /g3036/g3041 /g1856/g1872 /g3404/g3398 /g2009 /g3004/g3028 ./g1835 /g3004/g3028/g3021 /g3398 /g4670/g1829/g1853 /g2870/g2878 /g4671 /g3036/g3041 /g3398/g4670/g1829 /g1853 /g2870/g2878 /g4671 /g3036/g3041/g3029/g3028/g3046/g3032 /g2028 /g3004/g3028 /g3398/g1863 /g3030/g3035/g3032/g3039/g3028/g3047/g3036/g3042/g3041 ./g4670/g1829 /g1853 /g2870/g2878 /g4671 /g3036/g3041 /g1865 /g3020/g3012 /g3404 /g4670/g1829/g1853 /g2870/g2878 /g4671 /g3036/g3041 /g3041 /g3268/g3260 /g1837 /g3005 /g3041 /g3268/g3260 /g3397/g4670/g1829 /g1853 /g2870/g2878 /g4671 /g3036/g3041 /g3041 /g3268/g3260 Because SK-channel activation is faster than intracellular calcium dynamics, it was assumed to be instantaneous. Here /g2009 /g3004/g3028 converts calcium current /g;835 /g3004/g3028 (pA) into intracellular concentration change /g4672 /g3091/g3014 /g3043/g3002/g1668/g3040/g3046 /g4673 . Extracellular calcium concentration was set to /g4670/g;829/g;853 /g2870/g2878 /g467; /g3042/g3048/g3047 /g3404 1.2 mM, basal intracellular calcium concentration to /g4670/g;829/g;853 /g2870/g2878 /g467; /g3036/g3041/g3029/g3028/g3046/g3032 /g34045 0 nM, and the calcium reversal potential fixed to /g;83; /g3004/g3028 /g3404 120 mV. To capture trial-to-trial variability, the injected current included a temporally correlated noise term modeled as an Ornstein-Uhlenbeck process, adapted from 94. /g1835 /g3036/g3041/g3037 /g4666/g1872/g4667 /g3404 /g1835 /g3010/g3041/g3037/g2868 /g3397/g1835 /g3041/g3042/g3036/g3046/g3032 /g4666/g1872/g4667 /g1856/g1835 /g3041/g3042/g3036/g3046/g3032 /g1856/g1872 /g3404/g3398 /g1835 /g3041/g3042/g3036/g3046/g3032 /g2028 /g3041/g3042/g3036/g3046/g3032 /g3397 /g1835 /g3041 /g2028 /g3041/g3042/g3036/g3046/g3032 ./g1840 /g4666 /g1872 /g4667 where /g;840/g4666/g;872/g4667 is a normally distributed random variable (/g;840/g4666/g;872/g4667 /g;533/g2280 /g4666 0 , 1 /g4667 ), /g2028 /g3041/g3042/g3036/g3046/g3032 /g34042 ms, and /g;835 /g3041 /g34045 . All model parameters are summarized in Table 1. Table 1. Biophysical parameters of the extended Hodgkin–Huxley model. Current Parameters Fast Na/g2878 Activation: /g;848 /g3040/g3015/g3028/g2869//g2870 /g3404 /g339843.8 mV ; /g;863 /g3040/g3015/g3028 /g34046 mV Inactivation: /g;848 /g3035/g3015/g3028/g2869 / /g2870 /g3404 /g339867.5 mV; /g;863 /g3035/g3015/g3028 /g3404 /g339810.8 mV; /g2028 /g3035/g3015/g3028/g3040/g3028/g3051 /g3404 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 27 Current Parameters 35.2 ms; /g;863 /g3099/g3015/g3028 /g3404 12.8 mV Persistent Na/g2878 Activation: /g;848 /g3040/g3015/g3028/g3017/g2869//g2870 /g3404/g3398 5 5 mV ; /g;863 /g3040/g3015/g3028/g3017 /g34043 mV Inactivation: /g;848 /g3035/g3015/g3028/g3017/g2869//g2870 /g3404/g3398 5 9 mV; /g;863 /g3035/g3015/g3028/g3017 /g3404/g3398 5 . 2 mV; /g;863 /g3099/g3035/g3015/g3028/g3017 /g3404 10.4 mV; /g2028 /g3035/g3015/g3028/g3017/g3040/g3028/g3051 /g3404 5000 ms K/g2878 delayed rectifier Activation: /g;848 /g3040/g3012/g2869//g2870 /g3404 /g339834.5 mV ; /g;863 /g3040/g3012 /g34045 mV; /g2028 /g3040/g3012/g3040/g3028/g3051 /g34044 ; /g;863 /g3099/g3040/g3012 /g34041 0 M-current Activation: /g;848 /g3040/g3014/g2869//g2870 /g3404/g3398 4 4 mV ; /g;863 /g3040/g3014 /g34044 . 3 mV; /g2028 /g3040/g3014/g3040/g3028/g3051 /g34047 ms; /g;863 /g3099/g3040/g3014 /g34042 5 mV Calcium T-type current Activation: /g;848 /g3040/g3004/g3028/g3021/g2869//g2870 /g3404/g3398 5 0 mV ; /g;863 /g3040/g3004/g3028/g3021 /g34045 mV; /g2028 /g3040/g3004/g3028/g3021/g3040/g3028/g3051 /g34041 ms; /g;863 /g3099/g3040/g3004/g3028/g3021 /g34047 mV Inactivation: /g;848 /g3035/g3004/g3028/g3021/g2869//g2870 /g3404/g3398 7 5 mV; /g;863 /g3035/g3004/g3028/g3021 /g3404/g3398 5 mV; /g2028 /g3035/g3004/g3028/g3021/g3040/g3028/g3051 /g34042 0 ms; /g;863 /g3099/g3035/g3004/g3028/g3021 /g34047 mV Small K current (SK) Activation: /g;837 /g3005 /g34040 . 5 /g2020 M; Hill coefficient /g;866 /g3020/g3012 /g34044 Intracellular calcium /g4670/g;829/g;853 /g2870/g2878 /g467; /g3036/g3041 /g2028 /g3004/g3028 /g34041 0 0 ms; /g;863 /g3030/g3035/g3032/g3039/g3028/g3047/g3036/g3042/g3041 /g34040 . 5 for BAPTA and 0.05 for EGTA; /g2009 /g3004/g3028 /g34040 . 0 1 /g2020 M/(pA.ms) Other /g;829/g34045 2 pF; /g;832 /g3404 96485 C/mol; /g;846 /g3404 305.15 K; Simulation-Based Inference (SBI). To fit the biophysical model to experimental data while accounting for parameter uncertainty, we used simulation-based inference (SBI), a Bayesian framework for parameter estimation 95. SBI estimates posterior distributions over model parameters from observed summar y features extracted from neuronal activity. Here, the inferred parameters initially comprised the si x conductances of the Hodgkin-Huxley model. Ten summary features were computed from the voltage envelope of each burst phenotype: burst duration, burst duration variance, maximum burst amplitude, inter-burst frequency, area under the curve (AUC), interburst period, the maximum and minimum of the first derivative of the envelope, the maximum and minimum of the envelope acceleration, and the minimum and maximum membrane depolarization. Unlike classical Bayesian approaches, SBI does not require explicit evaluation of the likelihood function relating model parameters to observed features, which is generally intractable for nonlinear systems such as the present Hodgkin-Huxley model. SBI circumvents this by approximating the like lihood using a normalizing flow, a class of deep generative models trained on forward simulations to learn an invertible mapping between parameter and feature spaces 96. To this end, parameter sets were sampled from prior .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 28 distributions defined over physiologically plaus ible ranges for the six conductances. Priors were uniform and identical across the th ree clusters. In total, 20,000 parameter configurations were simulated, and the corresponding voltage traces were generated using a Runge-Kutta integration scheme. Numerical simulation was performed using Runge-Kutta integration method for stochastic processes, as the injected current to the neuron (I inj) involved randomness, chosen to be of constant amplitude. Inference quality was assessed using standard posterior diagnostics. Posterior shrinkage, defined as the reduction in paramet er uncertainty from prior to posterior, was used to evaluate how strongly individual parameters were constrained by the observed features. Posterior predictive checks further verified that simulations drawn from the inferred posteriors reproduced the target features with in acceptable margins. Pairwise posterior distributions were additionally examined to assess parameter dependencies and residual identifiability issues. For the burst phenotype characterized by sustained plateau depolarization (Cluster 3; see Results), the initial six-parameter inference did not reproduce the waveform satisfactorily. We therefore introduced the maxima l time constant governing deinactivation of the I NaP inactivation gate ( /g2028 /g3035/g3015/g3028/g3017/g3040/g3028/g3051 ; see Hodgkin-Huxley model section) as a seventh free parameter, motivated by previous exper imental evidence that changes in I NaP kinetics strongly affect burst duration 6. This parameter was sampled from a uniform prior over 4,000- 80,000 ms, whereas all other priors were kept unchanged. In silico perturbation analyses. Inferred parameter sets for each cluster were used to perform targeted in silico perturbations. In order to repr oduce experimental results four perturbation paradigms were exam ined. First, SK conductance ( /g;859 /g3020/g3012 ) was progressively reintroduced to test whether restoration of SK current suppresses bursting and promotes tonic spiking. Second, persistent sodium conductance ( /g;859 /g3015/g3028/g3017 ) was set to zero to assess whether INaP is required for burst generation. Third, starting from the SK-restored condition in which bursting had been suppressed, T-type calcium conductance ( /g;859 /g3004/g3028/g3021 ) was removed to .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 29 test whether T-type calcium channel blockade wa s sufficient to restore bursting. Fourth, again from the SK-restored tonic-spiking condition, intracellular calcium handling was perturbed by modifying the chelation parameter /g;863 /g3030/g3035/g3032/g3039/g3028/g3047/g3036/g3042/g3041 to mimic the effects of fast and slow calcium chelators. Because the fitted parameter regimes differed across clusters, in silico perturbations were implemented using cl uster-specific parameter values, summarized in Table 2. Table 2. Cluster-specific parameter values used for in silico perturbation analyses. Experiment Cluster 1 Cluster 2 Cluster 3 SK reactivation /g;859 /g3020/g3012 /g34043 . 5 /g;859 /g3020/g3012 /g34043 . 5 ; /g;835 /g3036/g3041/g3037 /g34042 8 /g;859 /g3020/g3012 /g34043 . 5 ; /g;835 /g3036/g3041/g3037 /g34042 2 INap suppression /g;859 /g3015/g3028/g3017 /g34040 , /g;835 /g3036/g3041/g3037 /g34043 1 . 5 /g;859 /g3015/g3028/g3017 /g34040 ;/g;835 /g3036/g3041/g3037 /g34043 3 /g;859 /g3015/g3028/g3017 /g34040 ; /g;835 /g3036/g3041/g3037 /g34042 6 CaT suppression /g;859 /g3020/g3012 /g34043 . 5 ;/g;859 /g3004/g3028/g3021 /g34040 /g;859 /g3020/g3012 /g34043 . 5 ;/g;859 /g3004/g3028/g3021 /g34040 /g;859 /g3020/g3012 /g34043 . 5 ; /g;859 /g3004/g3028/g3021 /g34040 Calcium chelation /g;859 /g3020/g3012 /g34043 . 5 ;/g;863 /g3030/g3035/g3032/g3039/g3028/g3047/g3036/g3042/g3041 /g34040 . 0 5 /g;859 /g3020/g3012 /g34043 . 5 ; /g;863 /g3030/g3035/g3032/g3039/g3028/g3047/g3036/g3042/g3041 /g34040 . 0 5 /g;859 /g3020/g3012 /g34043 . 5 ; /g;863 /g3030/g3035/g3032/g3039/g3028/g3047/g3036/g3042/g3041 /g34040 . 0 5 Quantification and statistical analysis Data analysis: Electrophysiological data were analyzed off-line. For whole-cell recordings, several basic criteria were set to ensure optimum quality of intracellular recordings. Only cells exhibiting a stable resting membrane potential and an action potential amplitude larger than 40 mV were considered. Passive membrane properties of cells were measured by determining from the holding potential the largest voltage deflections induced by small current pulses that avoided activation of voltage-sensitive currents. We determined input resistance by the slope of linear fits to voltage responses evoked by small pos itive and negative current injections. Firing properties were measured from depolarizing current pulses of varying amplitudes. Action potential (AP) features were extracted with a custom Python pipeline (PyABF/SciPy). For each cell, the first spike at rheobase was analy zed. AP threshold was defined as the first sample within the 4 ms preceding the AP peak where dV/dt ≥ 10 mV·ms-1. AP amplitude was measured from threshold to peak. AP half-widt h was computed at half-amplitude with linear .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 30 interpolation at the rising and falling crossings. Depolarizing and repolarizing slopes (mV·ms - 1) were computed from threshold to peak and from peak to half-amplitude return, respectively. Instantaneous firing frequency was the reciprocal of interspike interval; a mean instantaneous frequency was calculated across spikes in each sweep. The medium afterhyperpolarization (mAHP) amplitude was measured as the voltage drop from threshold to the most negative membrane potential occurring after the spike. mAHP duration was defined as the time the membrane potential remained below the half-recovery level (midpoint between the mAHP minimum and the threshold voltage), with crossings estimated by interpolation. All reported memb rane potentials were offline-corrected for the liquid junction potential, computed for each experiment from the ionic composition of the intra- and extracellular solutions. Burst features were quantified from bursts evoked at rheobase current. Bursts were identified on raw membrane potential recordings as discrete episodes of action potential firing (active) separated by spike-free (silent) intervals. In some cells, spikes occurred only at burst onset and were followed by a depolarization block during the plateau; these events were still classified as bursts. Bursting properties were then quantified from low-pass filtered (3 Hz) and median-smoothed recordings to remove spikes while preserving the depolarizing envelope. Burst amplitude and duration were me asured using a threshold-based detection algorithm, which identified the burst onset, peak, and offset, as well as the corresponding maximum and minimum values of the derivative (dV/dt). Onset and offset were refined by tracking the sign of dV/dt around the thres hold crossings. For each burst, amplitude was measured as the maximal depolarization above the onset baseline; duration was the interval from refined onset to offset; area was the Simpson-integrated baseline-subtracted envelope between onset and offset. Burst frequency was calculated as the inverse of the interval between consecutive burst onsets. Quantitativ e burst features (duration, amplitude, frequency, and maximal depolarizing and hyperpolar izing slopes) extracted from individual neurons were subjected to principal component analysis (PCA) to identify dominant axes of variability. The first two components captured most of the variance and were used for .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 31 visualization and interpretation. Hierarchical agglomerative clustering using Ward’s linkage was then applied to the standardized variables, revealing three distinct bursting phenotypes. These analyses were performed in Python using scikit-learn and SciPy, and cluster assignments were projected onto the PCA space. For extracellular recordings, alternating activity between right/left L5 recordings was taken to be indicative of fictive locomotion. Quantitative image analysis was performed in FIJI (ImageJ, NIH). For somatic analysis, ROIs encompassing the somata of Hb9::eGFP-positive neurons or unidentified ventromedial interneurons located in the L ₁ -L₂ region near the central canal were manually delineated. For dendritic analysis, ROIs were drawn along the full extent of dendrites from biocytin-filled Hb9::eGFP-positive neurons. Within each ROI, SK2-, SK3-, and Cav3.2-immunopositive puncta were segmented using automated local thresholding after background subtraction. Puncta were then detected using the Analyze Particles plugin with compartment-specific parameters (40× objective). For the quantification of somatic clusters, inclusion criteria were adapted to the target protein. For SK channels, particles were included if they fell within a size range of 0.2-5 µm 2 with a circularity between 0.75 and 1.0. For Cav3.2 channels, detection parameters were adjusted to a size range of 0.04-0.5 µm 2 with a circularity between 0.55 and 1.0. For dendritic SK clusters, filt ering criteria were set to a size range of 0.12–2 µm 2 and a circularity of 0.55-1.0. Puncta density and mean particle diameter were extracted from the Analyze Particles outputs. Colocalization was assessed by generating binary masks for each channel and computing their intersection (logical AND) in ImageJ, which isolates overlapping puncta. The inte rsection mask was then analyzed with Analyze Particles using the same compartment-specific criteria. Statistics: Statistical analyses were performed using GraphPad Prism 7 and Python (with the statsmodels and scikit-lear n libraries). For two-group comparisons, Wilcoxon, Mann- Whitney or Kolmogorov-Smirnov tests were used as appropriate. The analysis of the estimated posteriors conductances distributions was conducted using the Kolmogorov- .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 32 Smirnov test with 50 randomly drawn samples. Multiple groups were analyzed using Kruskal- Wallis or Friedman tests with Dunn’s post hoc correction, and proportions were compared using Fisher’s exact test. One-way ANOVA was applied when data met parametric assumptions. Multivariate analysis of variance (MANOVA) was used to assess the effect of experimental factors on multiple dependent variables (PC1 and PC2 scores) simultaneously. Hierarchical agglomerative clustering was applied to the PCA-transformed dataset, using Ward’s linkage to ensure internal cluster homogeneity. All tests were two-sided, with P < 0.05 considered significant. Specific statistical tests, sample sizes ( n), and P values for each dataset are provided in the corresponding figure legends. Data availability All relevant data supporting the findings of this study are included in the Source Data or available from the corresponding authors upon request. Further information and requests for resources and reagents should be directed to and will be fulfilled by the corresponding author, Frédéric Brocard ([email protected] ). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 33

1. Kiehn O. Decoding the organization of spinal circuits that control locomotion. Nat Rev Neurosci 17, 224-238 (2016). 2. Grillner S, El Manira A. Current Principles of Motor Control, with Special Reference to Vertebrate Locomotion. Physiol Rev 100, 271-320 (2020). 3. Rancic V, Gosgnach S. Recent Insights into the Rhythmogenic Core of the Locomotor CPG. Int J Mol Sci 22, (2021). 4. Brocard F, Tazerart S, Vinay L. Do pacemakers drive the central pattern generator for locomotion in mammals? Neuroscientist 16, 139-155 (2010). 5. Brocard F. New channel lineup in spinal circuits governing locomotion. Current opinion in Physiology 8, 14-22 (2019). 6. Tazerart S, Vinay L, Brocard F. The persistent sodium current generates pacemaker activities in the central pattern generator for locomotion and regulates the locomotor rhythm. J Neurosci 28, 8577-8589 (2008). 7. Song J , et al. Multiple Rhythm-Generating Circuits Act in Tandem with Pacemaker Properties to Control the Start and Speed of Locomotion. Neuron, (2020). 8. Ziskind-Conhaim L, Wu L, Wiesner EP. Pe rsistent sodium current contributes to induced voltage oscillations in locomotor-related hb9 interneurons in the mouse spinal cord. J Neurophysiol 100, 2254-2264 (2008). 9. Crill WE. Persistent sodium current in mammalian central neurons. Annu Rev Physiol 58, 349-362 (1996). 10. Tazerart S, Viemari JC, Darbon P, Vinay L, Brocard F. Contribution of Persistent Sodium Current to Locomotor Pattern Generation in Neonatal Rats. Journal of Neurophysiology 98, 613-628 (2007). 11. Zhong G, masino MA, Harris-Warrick RM. Persistent sodium currents participate in fictive locomotion generation in neonatal mouse spinal cord. J Neurosci 27, 4507- 4518 (2007). 12. Ryczko D, Charrier V, Ijspeert A, Cabelguen JM. Segmental oscillators in axial motor circuits of the salamander: distribution and bursting mechanisms. J Neurophysiol 104, 2677-2692 (2010). 13. Tong H, McDearmid JR. Pacemaker and plateau potentials shape output of a developing locomotor network. Curr Biol 22, 2285-2293 (2012). 14. Svensson E, Jeffreys H, Li WC. The modulation of two motor behaviors by persistent sodium currents in Xenopus laevis tadpoles. J Neurophysiol 118, 121-130 (2017). 15. Zhang Q, Cheng Y, Zhou M, Dai Y. Locomotor Pattern and Force Generation Modulated by Ionic Channels: A Computational Study of Spinal Networks Underlying Locomotion. Front Comput Neurosci 16, 809599 (2022). 16. Brocard F , et al. Activity-dependent changes in extracellular ca(2+) and k(+) reveal pacemakers in the spinal locomotor-related network. Neuron 77, 1047-1054 ( 2013). 17. Drouillas B, Brocard C, Zanella S, Bos R, Brocard F. Persistent Nav1.1 and Nav1.6 currents drive spinal locomotor functions through nonlinear dynamics. Cell Rep 42, 113085 (2023). 18. Verneuil J, Brocard C, Trouplin V, Villard L, Peyronnet-Roux J, Brocard F. The M- current works in tandem with the persistent sodium current to set the speed of locomotion. PLoS Biol 18, e3000738 (2020). 19. el Manira A, Tegner J, Grillner S. Calcium-dependent potassium channels play a critical role for burst termination in the locomotor network in lamprey. J Neurophysiol 72, 1852-1861 (1994). 20. Cazalets JR, Sqalli-Houssaini Y, Magoul R. Differential effects of potassium channel blockers on the activity of the locomotor network in neonatal rat. Brain Res 827, 185- 197 (1999). 21. Hill R, Matsushima T, Schotland J, Grillner S. Apamin blocks the slow AHP in lamprey and delays termination of locomotor bursts. Neuroreport 3, 943-945 (1992). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 34 22. Mahrous AA, Elbasiouny SM. Modulati on of SK channels regulates locomotor alternating bursting activity in t he functionally-mature spinal cord. Channels (Austin) 12, 9-14 (2018). 23. Mahrous AA, Elbasiouny SM. SK channel inhibition mediates the initiation and amplitude modulation of synchronized burst firing in the spinal cord. J Neurophysiol 118, 161-175 (2017). 24. Candia S, Garcia ML, Latorre R. Mode of action of iberiotoxin, a potent blocker of the large conductance Ca(2+)-activated K+ channel. Biophys J 63, 583-590 (1992). 25. Wulff H, Miller MJ, Hansel W, Grissmer S, Cahalan MD, Chandy KG. Design of a potent and selective inhibitor of the intermediate-conductance Ca2+-activated K+ channel, IKCa1: a potential immunosuppressant. Proc Natl Acad Sci U S A 97, 8151- 8156 (2000). 26. Grunnet M, Jensen BS, Olesen SP, Klaerke DA. Apamin interacts with all subtypes of cloned small-conductance Ca2+-activated K+ channels. Pflugers Arch 441, 544-550 (2001). 27. Müller A, Kukley M, Uebachs M, Beck H, Dietrich D. Nanodomains of single Ca2+ channels contribute to action potential repolarization in cortical neurons. J Neurosci 27, 483-495 (2007). 28. Strobaek D, Jorgensen TD, Christophersen P, Ahring PK, Olesen SP. Pharmacological characterization of small-conductance Ca(2+)-activated K(+) channels stably expressed in HEK 293 cells. Br J Pharmacol 129, 991-999 (2000). 29. Fanger CM , et al. Calcium-activated potassium channels sustain calcium signaling in T lymphocytes. Selective blockers and manipulated channel expression levels. J Biol Chem 276, 12249-12256 (2001). 30. Ma B , et al. Structural mechanisms for inhibition and activation of human small- conductance Ca(2+)-activated potassium channel SK2. Nat Commun, (2026). 31. Weatherall KL, Seutin V, Liégeois JF, Marrion NV. Crucial role of a shared extracellular loop in apamin sensitivit y and maintenance of pore shape of small- conductance calcium-activated potassium (SK) channels. Proc Natl Acad Sci U S A 108, 18494-18499 (2011). 32. Pedarzani P , et al. Tamapin, a venom peptide from the Indian red scorpion (Mesobuthus tamulus) that targets smal l conductance Ca2+-activated K+ channels and afterhyperpolarization currents in central neurons. J Biol Chem 277, 46101- 46109 (2002). 33. Wilson JM , et al. Conditional Rhythmicity of Ventral Spinal Interneurons Defined by Expression of the Hb9 Homeodomain Protein. Journal of Neuroscience 25, 5710- 5719 (2005). 34. Shakkottai VG , et al. Design and characterization of a highly selective peptide inhibitor of the small conductance calcium-activated K+ channel, SkCa2. J Biol Chem 276, 43145-43151 (2001). 35. Pedarzani P, Stocker M. Molecular and cellular basis of small--and intermediate- conductance, calcium-activated potassium channel function in the brain. Cell Mol Life Sci 65, 3196-3217 ( 2008). 36. Hougaard C , et al. Selective positive modulation of the SK3 and SK2 subtypes of small conductance Ca2+-activated K+ channels. Br J Pharmacol 151, 655-665 (2007). 37. Lee JH, Gomora JC, Cribbs LL, Perez-Reyes E. Nickel block of three cloned T-type calcium channels: low concentrations selectively block alpha1H. Biophys J 77, 3034- 3042 (1999). 38. Nishimura Y , et al. Ionic Mechanisms Underlying Burst Firing of Layer III Sensorimotor Cortical Neurons of the Cat: An In Vitro Slice Study. Journal of Neurophysiology 86, 771-781 (2001). 39. Hallworth NE, Wilson CJ, Bevan MD. A pamin-sensitive small conductance calcium- activated potassium channels, through their selective coupling to voltage-gated calcium channels, are critical determinants of the precision, pace, and pattern of .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 35 action potential generation in rat subthalamic nucleus neurons in vitro. J Neurosci 23, 7525-7542 (2003). 40. Bennett BD, Callaway JC, Wilson CJ. Intrinsic Membrane Properties Underlying Spontaneous Tonic Firing in Neostriatal Cholinergic Interneurons. Journal of Neuroscience 20, 8493-8503 (2000). 41. Shepard PD, Bunney BS. Repetitive firing properties of putative dopamine-containing neurons in vitro: regulation by an apamin-s ensitive Ca(2+)-activated K+ conductance. Exp Brain Res 86, 141-150 (1991). 42. Shepard PD, Bunney BS. Effects of apamin on the discharge properties of putative dopamine-containing neurons in vitro. Brain Res 463, 380-384 (1988). 43. Ping HX, Shepard PD. Apamin-sensitive Ca(2+)-activated K+ channels regulate pacemaker activity in nigral dopamine neurons. Neuroreport 7, 809-814 (1996). 44. Rouchet N , et al. SK channel blockade promotes burst firing in dorsal raphe serotonergic neurons. Eur J Neurosci 28, 1108-1115 (2008). 45. Gu X, Blatz AL, German DC. Subtypes of substantia nigra dopaminergic neurons revealed by apamin: autoradiographic and electrophysiological studies. Brain Res Bull 28, 435-440 (1992). 46. Johnson SW, Wu YN. Multiple mechanisms underlie burst firing in rat midbrain dopamine neurons in vitro. Brain Research 1019, 293-296 (2004). 47. Yanovsky Y, Zhang W, Misgeld U. Two pathways for the activation of small- conductance potassium channels in neurons of substantia nigra pars reticulata. Neuroscience 136, 1027-1036 (2005). 48. Goldberg JA, Wilson CJ. Control of spontaneous firing patterns by the selective coupling of calcium currents to calcium-ac tivated potassium currents in striatal cholinergic interneurons. J Neurosci 25, 10230-10238 (2005). 49. De Waele C, Serafin M, Khateb A, Yabe T, Vidal PP, Muhlethaler M. Medial vestibular nucleus in the guinea-pig: apamin -induced rhythmic burst firing--an in vitro and in vivo study. Exp Brain Res 95, 213-222 (1993). 50. Harris-Warrick RM, Johnson BR. Potassium channel blockade induces rhythmic activity in a conditional burster neuron. Brain Res 416, 381-386 (1987). 51. Wilson CJ. The mechanism of intrinsic amplification of hyperpolarizations and spontaneous bursting in striatal cholinergic interneurons. Neuron 45, 575-585 (2005). 52. Shepard PD, Stump D. Nifedipine blocks apa min-induced bursting activity in nigral dopamine-containing neurons. Brain Res 817, 104-109 (1999). 53. Dougherty KJ , et al. Locomotor rhythm generation linked to the output of spinal shox2 excitatory interneurons. Neuron 80, 920-933 (2013). 54. Caldeira V, Dougherty KJ, Borgius L, Kiehn O. Spinal Hb9::Cre-derived excitatory interneurons contribute to rhythm generation in the mouse. Sci Rep 7, 41369 (2017). 55. Singh S, Shevtsova NA, Yao L, Rybak IA, Dougherty KJ. Properties of rhythmogenic currents in spinal Shox2 interneurons across postnatal development. J Physiol 603, 3201-3221 (2025). 56. Seutin V, Johnson SW, North RA. Apamin increases NMDA-induced burst-firing of rat mesencephalic dopamine neurons. Brain Res 630, 341-344 (1993). 57. Johnson SW, Seutin V. Bicuculline methiodide potentiates NMDA-dependent burst firing in rat dopamine neurons by blocking apamin-sensitive Ca2+-activated K+ currents. Neurosci Lett 231, 13-16 (1997). 58. Hopf FW, Martin M, Chen BT, Bowers MS, Mohamedi MM, Bonci A. Withdrawal from intermittent ethanol exposure increases probability of burst firing in VTA neurons in vitro. J Neurophysiol 98, 2297-2310 (2007). 59. Wallén P, Ekeberg O, Lansner A, Brodin L, Tråvén H, Grillner S. A computer-based model for realistic simulations of neural networks. II. The segmental network generating locomotor rhythmicity in the lamprey. J Neurophysiol 68, 1939-1950 (1992). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 36 60. Tell F, Jean A. Ionic basis for endogenous rhythmic patterns induced by activation of N-methyl-D-aspartate receptors in neurons of the rat nucleus tractus solitarii. J Neurophysiol 70, 2379-2390 (1993). 61. Stocker M. Ca(2+)-activated K+ channels: molecular determinants and function of the SK family. Nat Rev Neurosci 5, 758-770 (2004). 62. Monaghan AS , et al. The SK3 subunit of small conductance Ca2+-activated K+ channels interacts with both SK1 and SK2 subunits in a heterologous expression system. J Biol Chem 279, 1003-1009 (2004). 63. Sailer CA , et al. Regional differences in distribution and functional expression of small-conductance Ca2+-activated K+ channels in rat brain. J Neurosci 22, 9698- 9707 (2002). 64. Bond CT , et al. Small conductance Ca2+-activated K+ channel knock-out mice reveal the identity of calcium-dependent afterhyperpolarization currents. J Neurosci 24, 5301-5306 (2004). 65. Bock T, Honnuraiah S, Stuart GJ. Paradoxical Excitatory Impact of SK Channels on Dendritic Excitability. J Neurosci 39, 7826-7839 (2019). 66. Ohtsuki G, Hansel C. Synaptic Potent ial and Plasticity of an SK2 Channel Gate Regulate Spike Burst Activity in Cerebellar Purkinje Cells. iScience 1, 49-54 (2018). 67. Lee Y , et al. Impairment of a parabolic bursting rhyt hm by the ectopic expression of a small conductance Ca2+-activated K+ channel in Aplysia neuron R15. Neurosci Lett 349, 53-57 (2003). 68. Kasumu AW , et al. Selective positive modulator of calcium-activated potassium channels exerts beneficial effects in a mouse model of spinocerebellar ataxia type 2. Chem Biol 19, 1340-1353 (2012). 69. Soden ME , et al. Disruption of dopamine neuron activity pattern regulation through selective expression of a human KCNN3 mutation. Neuron 80, 997-1009 (2013). 70. Deardorff AS , et al. Expression of postsynaptic Ca2+-activated K+ (SK) channels at C-bouton synapses in mammalian lumbar -motoneurons. J Physiol 591, 875-897 (2013). 71. Deardorff AS, Romer SH, Fyffe REW. Location, location, location: the organization and roles of potassium channels in mammalian motoneurons. J Physiol 599, 1391- 1420 (2021). 72. Anderson TM, Abbinanti MD, Peck JH, Gilmour M, Brownstone RM, masino MA. Low-threshold calcium currents contribute to lo comotor-like activity in neonatal mice. J Neurophysiol 107, 103-113 (2012). 73. Wall MJ, Dale N. A slowly activating Ca(2+)-dependent K+ current that plays a role in termination of swimming in Xenopus embryos. J Physiol 487 ( Pt 3), 557-572 (1995). 74. Perez-Reyes E. Molecular physiology of low-voltage-activated t-type calcium channels. Physiol Rev 83, 117-161 (2003). 75. Marrion NV, Tavalin SJ. Selective activation of Ca2+-activated K+ channels by co- localized Ca2+ channels in hippocampal neurons. Nature 395, 900-905 (1998). 76. Sah P, Davies P. Calcium-activated potassium currents in mammalian neurons. Clin Exp Pharmacol Physiol 27, 657-663 (2000). 77. Womack MD, Khodakhah K. Somatic and dendritic small-conductance calcium- activated potassium channels regulate the output of cerebellar Purkinje neurons. J Neurosci 23, 2600-2607 (2003). 78. Li X, Bennett DJ. Apamin-sensitive calcium-activated potassium currents (SK) are activated by persistent calcium currents in rat motoneurons. J Neurophysiol 97, 3314- 3330 (2007). 79. Wolfart J, Roeper J. Selective coupling of T-type calcium channels to SK potassium channels prevents intrinsic bursting in dopaminergic midbrain neurons. J Neurosci 22, 3404-3413 (2002). 80. Cueni L, et al. T-type Ca2+ channels, SK2 channels and SERCAs gate sleep-related oscillations in thalamic dendrites. Nat Neurosci 11, 683-692 (2008). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 37 81. Vick KAt, Guidi M, Stackman RW, Jr. In vivo pharmacological manipulation of small conductance Ca(2+)-activated K(+) channels influences motor behavior, object memory and fear conditioning. Neuropharmacology 58, 650-659 (2010). 82. Van Dongen PA, Grillner S, Hökfelt T. 5-Hydroxytryptamine (serotonin) causes a reduction in the afterhyperpolarization following the action potential in lamprey motoneurons and premotor interneurons. Brain Res 366, 320-325 (1986). 83. Wallén P, Buchanan JT, Grillner S, Hill RH, Christenson J, Hökfelt T. Effects of 5- hydroxytryptamine on the afterhyperpolar ization, spike frequency regulation, and oscillatory membrane properties in lamprey spinal cord neurons. J Neurophysiol 61, 759-768 (1989). 84. Grunnet M, Jespersen T, Perrier JF. 5-HT1A receptors modulate small-conductance Ca2+-activated K+ channels. J Neurosci Res 78, 845-854 (2004). 85. Bayliss DA, Umemiya M, Berger AJ. Inhibition of N- and P-type calcium currents and the after-hyperpolarization in rat motoneurones by serotonin. J Physiol 485 ( Pt 3) , 635-647 (1995). 86. Díaz-Ríos M, Dombeck DA, Webb WW, Harris-Warrick RM. Serotonin modulates dendritic calcium influx in commissural in terneurons in the mouse spinal locomotor network. J Neurophysiol 98, 2157-2167 (2007). 87. Perrier JF, Cotel F. Serotonergic modulation of spinal motor control. Curr Opin Neurobiol 33, 1-7 (2015). 88. Barriere G, Mellen N, Cazalets JR. Neuromodulation of the locomotor network by dopamine in the isolated spinal cord of newborn rat. European Journal of Neuroscience 19, 1325-1335 (2004). 89. Schotland J , et al. Control of lamprey locomotor neurons by colocalized monoamine transmitters. Nature 374, 266-268 (1995). 90. Maingret F , et al. Neurotransmitter modulation of small-conductance Ca2+-activated K+ channels by regulation of Ca2+ gating. Neuron 59, 439-449 (2008). 91. Miles GB, Hartley R, Todd AJ, Brownstone RM. Spinal cholinergic interneurons regulate the excitability of motoneurons during locomotion. Proc Natl Acad Sci U S A 104, 2448-2453 (2007). 92. Nascimento F, Spindler LRB, Miles GB. Balanced cholinergic modulation of spinal locomotor circuits via M2 and M3 muscarinic receptors. Sci Rep 9, 14051 (2019). 93. Pflieger JF, Clarac F, Vinay L. Picrotox in and bicuculline have different effects on lumbar spinal networks and motoneurons in the neonatal rat. Brain Res 935, 81-86 (2002). 94. Mainen ZF, Sejnowski TJ. Reliability of spike timing in neocortical neurons. Science 268, 1503-1506 (1995). 95. Goncalves PJ, et al. Training deep neural density estimators to identify mechanistic models of neural dynamics. El ife 9, (2020). 96. Hashemi M , et al. Amortized Bayesian inference on generative dynamical network models of epilepsy using deep neural density estimators. Neural Netw 163, 178-194 (2023). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 38

We are grateful to Geneviève Rougon for her valuable input to the manuscript. We thank the GeneXprINT platform at the Institut de Neurosciences de la Timone for genotyping services, Anne Duhoux and Mélanie Huc for excellent animal care. This research was supported by Agence National de la Recherche Scientifique (SpasT- SCI-T ANR-21-CE17-0060; MotoBIS ANR-24-CE16-1548 and RhythMIC ANR-24- CE14-4160 to F.B.). AUTHOR CONTRIBUTIONS F.K. designed, performed and analyzed the vast majority of in vitro electrophysiological experiments, contributed to computational modeling and wrote the first draft of the manuscript. C.D. designed, performed and analyzed computational modeling experiments and simulation-based inference analyses under the supervision of M.G. C.B. designed, performed and analyzed immunohistochemistry experiments related to SK2 and SK3 channels. V.T. designed, performed and analyzed immunohistochemistry experiments related to Cav3.2 and validated the shRNA knockdown in cell culture. B.D. performed some intracellular recordings. J.-D.L. and M.H. contributed to the development of the model fitting pipeline. M.G. designed, performed, analyzed and supervised computational modeling experiments. F.B. conceptualized, administrated, designed, supervised and funded the whole project, updated the Hodgkin-Huxley model, performed and analyzed some in vitro experiments and wrote the manuscript. DECLARATION OF INTERESTS The authors declare no competing financial interests. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 39 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 40 FIGURE LEGENDS Figure 1. Potassium channels restrain INaP-driven burst expression. a Voltage trace from ventromedial interneuron of the rhythmogenic CPG region (L 1-L2) recorded with a CsCl- based intracellular solution, showing rhythmic bursting. b Raincloud plots with box-and- whisker overlays (median, interquartile range) quantifying burst duration, amplitude, area, and frequency. Each dot represents a single cell; sa mple size is indicated in parentheses. c Representative action potential evoked by a near-threshold depolarizing pulse. Inset highlights the slow afterdepolarization (ADP, arrow) replacing AHP. d Voltage responses to a 2-ms suprathreshold pulse before (black) and after (pink) tetrodotoxin (TTX, 0.5 µM) application. e Transition from bursting to tonic firing during TTX wash-in. Figure 2. SK channels gate I NaP-dependent bursting activity. a Example trace showing the transition from tonic firing to rhyt hmic bursting following apamin application. b Bar graph showing the proportion of bursting neurons with CsCl-based intracellular solution or under bath-applied apamin. c-d Expanded segments from the same in a showing activity in standard aCSF ( c) and after apamin ( d). e Bursting abolished by riluzole (5 µM). f Voltage traces from ventromedial interneurons of the rhythmogenic CPG region (L 1-L2) evoked by near-threshold depolarizing pulses before (black) and after (pink) apamin application (200 nM). g-i Raincloud plots with boxplots (median and interquartile range) quantifying afterhyperpolarization (AHP) amplitude ( g), AHP duration ( h), and spiking frequency ( i) before and after apamin. j-m Raincloud plots with boxplots (median and interquartile range) quantifying burst duration ( j), amplitude ( k), area (l), and frequency ( m) in neurons recorded with CsCl-based intracellular solution (grey) or under bath-applied apamin (pink). n-q Raincloud plots with boxplots (median and interquartile range) quantifying burst duration ( n), amplitude ( o), area ( p), and frequency ( q) after apamin in mouse (pink) and rat (blue) interneurons. Data points plotted beyond the dashed vertical line indicate values outside the axis range. Numbers in parentheses indicate recorded cells; each dot represents a single cell. n.s., not significant; *P < 0.05; **P < 0.01; ***P < 0.001 (two-sided Fisher’s exact test for .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 41 b; two-sided Wilcoxon paired test for g-i; two-sided unpaired t-test for j-q). For detailed P values, see Source data. Figure 3. Complementary roles of SK2 and SK3 in burst initiation. a-c Voltage traces from ventromedial interneurons of the rhythmogenic CPG region (L1-L2) recorded under control conditions ( a), after tamapin application (10 nM, b), or after lei-dab7 application (10 nM, c). d Bar graph showing the proportion of interneurons displaying bursting in response to increasing concentrations of lei-dab7 (teal) or tamapin (purple). e Representative action potentials evoked by near-threshold current in jections under control conditions (black), tamapin (purple), or lei-dab7 (teal). f-h Raincloud plots with box-and-whisker overlays (median, interquartile range) quantifying AHP amplitude ( f), AHP duration ( g), and firing frequency (h) under the three conditions. i-k Raincloud plots with box-and-whisker overlays (median, interquartile range) quantifying burst duration ( i), amplitude ( j), and frequency ( k) across interneurons treated with apamin (p ink), tamapin (purple), or lei-dab7 (teal). l-o Representative voltage traces from interneurons expressing control shRNA ( l) or SK3- targeting shRNA ( m-o), showing heterogeneous bursting phenotypes: bursts at rest/rheobase (m), bursts requiring stronger depolarization (n), and elliptic bursting dynamics (o). p-r Raincloud plots comparing burst duration ( p), amplitude ( q), and frequency ( r) in interneurons after SK3 knockdown (orange) and after apamin application (pink). Numbers in parentheses denote recorded cells; each dot r epresents a single cell. Data points plotted beyond the dashed vertical line indicate values outside the axis range. n.s., not significant; *P < 0.05; ** P < 0.01; *** P < 0.001 (two-sided Fisher’s exact test for d; Kruskal-Wallis with Dunn’s post hoc test versus control for f-h and i-k; two-sided Mann-Whitney test for p-r). For detailed P values, see Source Data. Figure 4. Distribution of SK2 and SK3 channels in Hb9-positive interneurons. a R epresentative low-magnification confocal image of GFP-expressing Hb9 interneurons in the ventromedial spinal cord. The asterisk marks the soma of an identified interneuron. The central canal (cc) is indicated by the dashed line. Scale bar, 25 µm. b High-magnification .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 42 images of the soma indicated in a, showing GFP fluorescence (top), SK2 immunolabeling (middle), and a merged GFP/SK2 overlay (bottom). Dashed lines delineate the somatic membrane. Scale bar, 10 µm. c-d Raincloud plots with box-and-whisker overlays (median, interquartile range) quantifying the size (c ) and density ( d, clusters per µm 2) of SK2, SK3, and co-localized SK2/3 clusters at the somatic membrane. e Confocal images of an Hb9 interneuron soma showing GFP (left), SK3 immunolabeling (middle), and a merged GFP/SK3 overlay (right). Dashed lines delineate the somatic membrane. Scale bar, 20 µm. f Confocal images of an intracellularly recorded HB9 interneuron filled with biocytin (magenta), showing GFP immunofluorescence (green), SK2 (red), and SK3 immunolabeling (cyan), alongside a merged GFP/SK3 signal. Scale bar, 20 µm. g-h Raincloud plots with box-and-whisker overlays (median, interquartile range) quantifying the size ( g) and density ( h) of SK2, SK3, and co-localized clusters along the dendrites. Numbers in parentheses denote the number of cells (c, d) or dendrites ( g, h) analyzed; each dot represents a single measurement. *** P < 0.001 (two-sided unpaired t-test for c and g; One-way ANOVA for d and h ). For detailed P values, see Source data. Figure 5. SK channels bidirectionally regulate intrinsic bursting dynamics. a Voltage traces from a bursting ventromedial interneuron (L 1-L2) recorded under locomotor-like ionic conditions ([Ca 2+]/i1 = 0.9 mM; [K + ]/i1 = 6 mM) before (black) and after apamin application (pink). b-d Raincloud plots with box-and-whisker overlays (median, interquartile range) quantifying burst amplitude ( b), duration ( c), and frequency ( d) before and after apamin. e Voltage responses to a suprathreshold depolarizing current step recorded in standard aCSF ([Ca2+]/i1 = 1.2 mM and [K +]/i1 = 3 mM) under control conditions (black) and after 1-EBIO application (blue). f-h Raincloud plots quantifying AHP amplitude ( f), AHP duration ( g), and firing frequency (h) before and after 1-EBIO. i Transition from rhythmic bursting to tonic firing during 1-EBIO application (blue gradient), followed by the restoration of bursting upon apamin co-application (pink) in the same cell. j, k Raincloud plots comparing burst duration (j) and amplitude (k) across the three conditions (Control, 1-EBIO, Apamin). l Voltage traces .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 43 under locomotor-like conditions showing baseline bursting (black), transition to tonic firing after CyPPA application (blue), and recovery of bursting with apamin (pink). m, n Raincloud plots quantifying burst amplitude ( m) and duration ( n) under control and CyPPA conditions. Numbers in parentheses denote recorded cells; each dot represents a single cell. n.s., not significant; * P < 0.05; ** P < 0.01 (two-sided Wilcoxon paired test for b-d, f-h, m and n; Friedman test with Dunn’s post hoc for j and k). For detailed P values, see Source data. Figure 6. T-type Ca 2+ channels couple to SK channels to gate intrinsic bursting. a Voltage traces from a ventromedial interneuron (L 1-L2) recorded in normal aCSF (black) and after cadmium (Cd 2+, 100 µM, light blue) application. b Bar graph showing the proportion of bursting interneurons under the indicated pharmacological conditions. c-e Raincloud plots with box-and-whisker overlays (median, inte rquartile range) quantifying burst duration ( c), amplitude (d), and frequency ( e) induced by apamin, Cd 2+, Ni2+, and mibefradil. f, g Voltage traces showing rhythmic bursting induced by mibefradil (10 µM, orange, f) or nickel (Ni²⁺ , 200 µM, purple, g) followed by the transition to tonic firing after riluzole application (5 µM, black). h, i Voltage traces from neurons recorded with intracellular BAPTA (10 mM; teal, h) or EGTA (10 mM; teal, i). Note the transition to tonic firing after riluzole application in h. j Representative voltage trace showing the maintenance of tonic firing (absence of bursting) in the presence of dantrolene (50 µM). k-m Raincloud plots with box-and-whisker overlays comparing burst duration ( k), amplitude ( l), and frequency ( m) between chelation-induced bursts (BAPTA or EGTA) and apamin-induced bursts. Numbers in parentheses denote recorded cells; each dot represents a single cell. Data points plotted beyond the dashed vertical line indicate values outside th e axis range. n.s., not significant; * P < 0.05; **P < 0.01; ***P < 0.001 (Kruskal-Wallis with Dunn’s post hoc for c-e; two-sided Mann-Whitney test for k- m). For detailed P values, see Source data. Figure 7. Diversity and biophysical classification of apamin-induced burst patterns. a Bar graph showing the proportion of interneurons displaying bursting in response to increasing concentrations of apamin. b Principal component analysis (PCA) of burst .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 44 parameters (duration, amplitude, area, fr equency, depolarizing and repolarizing slopes). Hierarchical clustering identified three di screte phenotypes: Cluster 1 (orange), Cluster 2 (green), and Cluster 3 (blue). Red crosses indicate cluster centroids. c Correlation circle showing the contribution of each burst parameter to PC1 and PC2. d-f Representative recordings for Cluster 1 ( d), Cluster 2 (e), and Cluster 3 ( f). Top: voltage traces showing raw signals with action potentials (shaded) and low-pass filtered signals highlighting burst envelopes (solid lines). Note the different time scales. Bottom: corresponding phase-plane trajectories (dV/dt vs voltage). g-l Raincloud plots with box-and-whisker overlays (median, interquartile range) showing burst duration ( g), area (h), frequency (i), amplitude (j), maximal depolarizing slope (k), and maximal repolarizing slope ( l) across the three clusters. Numbers in parentheses denote recorded cells; each dot r epresents a single cell. Data points plotted beyond the dashed vertical line indicate values outside the axis range. * P < 0.05; *** P < 0.001 (two-sided Fisher’s exact test for a; Kruskal-Wallis with Dunn's post hoc test for g-l). For detailed P values, see Source data. Figure 8. Simulation-based inference identifies the ionic basis of bursting phenotypes. a Representative experimental burst for Cluster 1 (red), Cluster 2 (green), and Cluster 3 (blue). Light curves: raw voltage traces; dar k lines: burst envelopes used for summary feature extraction. b Representative model fits for each phenotype generated from the inferred parameter sets. c Posterior distributions of six inferred conductances for each cluster (colored curves) overlaid on the prior distribution (grey shading). d-e In silico pharmacological manipulations applied to fitted models, consisting of reintroducing SK conductance in d (values in Table 2) followed by T-type calcium channel suppression (g CaT = 0) in e. For detailed P values, see Source data. Figure 9. SK2/3 and T-type channel modulation gates locomotor rhythmogenesis. a-b Representative ventral root recordings showing the effect of the broad-spectrum SK activator 1-EBIO ( a; 5 mM) and the SK2/3-selective activator CyPPA ( b; 0.3-0.4 mM) on ongoing fictive locomotion induced by NMDA/5-HT (5/10 µM). Top left in a and b: Schematic of the .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint 45 isolated spinal cord preparation illustrating focal drug puff application onto rhythmogenic segments (L 1-L2) and bilateral L5 ventral root recordings (L5R, L5L). Bottom: Raw ventral root activity (black) superimposed with low- pass filtered envelopes (blue for 1-EBIO, orange for CyPPA); vertical arrows indicate puff onset. Top right: Expanded time windows ( a’, b’) of the recordings shown below, highlighting rhythm suppression. c-d Representative recordings showing the induction of locomotor-like acti vity in quiescent preparations (subliminal NMDA/5-HT: 0-1/10 µM) by the SK2/3 inhibitor tamapin ( c; 1 µM) or the T-type calcium channel blocker nickel ( d; 10-20 µM). The experimental layout and display are identical to a, b. Low-pass filtered envelopes are shown in purple (tamapin) and teal (nickel). Arrows indicate puff onset. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 20, 2026. ; https://doi.org/10.64898/2026.03.19.712770doi: bioRxiv preprint