Compensatory coupling between leak and HCN conductances defines a low dimensional solution manifold in GPe neuron subtypes Authors: Matheus Phellipe Brasil de Sousa, Gabriel Moreno Cunha, Gabrielle Emily Boaventura Tavares, Gilberto Corso, Karina Possa Abrahao, Gustavo Zampier dos Santos Lima Published: 2026 ABSTRACT AbstractThe external segment of the globus pallidus contains distinct neuronal subtypes, primarily classified as prototypical and arkypallidal neurons, which exhibit different anatomical, electrophysiological, and functional properties. This cellular heterogeneity plays a central role in shaping basal ganglia activity under both physiological and pathological conditions, including neurodegenerative disorders such as Parkinson’s disease. Alterations in intrinsic membrane conductances, particularly leak and hyperpolarization activated cyclic nucleotide-gated (HCN) currents, have been implicated in abnormal neuronal excitability and dysfunctional circuit dynamics. In this study, we used a Hodgkin–Huxley-like computational model to investigate how variations in leak () and HCN () conductances shape the intrinsic dynamics of these pacemaker neuronal subtypes. By systematically exploring the parameter space, we characterized key electrophysiological features, including firing rate, sag ratio, and trough potential, and constrained these responses using experimental benchmarks. We then introduced an intersection-based methodology to identify subsets of conductance values that simultaneously satisfy multiple physiological constraints. Within this constrained space, we uncovered a robust inverse relationship between and , defining a low-dimensional solution manifold that captures coordinated interactions between these conductances. This manifold preserves physiological excitability through balanced adjustments in ion channel properties, consistent with experimental observations and the principle of ion channel degeneracy. Importantly, deviations from this regime provide a mechanistic framework to understand how imbalances in and may drive to abnormal excitability and pathological dynamics in basal ganglia circuits. Overall, this work establishes a quantitative link between intrinsic conductance regulation and neuronal function, offering a framework to investigate how disruptions in these mechanisms may contribute to neurodegenerative conditions.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-56455-z.Subject terms: Biological techniques, Computational biology and bioinformatics, Engineering, Neuroscience Introduction The external segment of the globus pallidus (GPe) plays a central role in basal ganglia circuits, contributing to motor control and the modulation of cortical activity through subcortical loops1–4. Importantly, dysfunctions within basal ganglia circuits, particularly those involving the GPe, have been strongly associated with neurological disorders such as Parkinson’s disease, where abnormal neuronal excitability and pathological oscillations emerge5,6. Once considered a homogeneous structure, the GPe is now recognized as a heterogeneous nucleus composed of distinct neuronal subtypes with divergent molecular profiles, projection targets, and intrinsic electrophysiological properties7–9. Two main subtypes have been characterized: prototypic neurons, often identified by parvalbumin (PV) expression, and arkypallidal neurons, marked by genes such as Npas1, FoxP2, or pre-proenkephalin7,8. These populations exhibit contrasting physiological signatures: prototypic neurons typically present high spontaneous firing rates, low sag ratios, and shallow hyperpolarization-induced troughs, whereas arkypallidal neurons display lower firing rates, more pronounced sag responses, and deeper trough potentials9,10. Although these subtype-specific differences are well documented, the underlying biophysical mechanisms responsible for shaping these electrophysiological signatures remain incompletely understood. In particular, the role of two key ionic conductances, leak () and hyperpolarization-activated cyclic nucleotide-gated (HCN) channel conductance (), has not been fully characterized in the context of subtype identity. Notably, both conductances are directly involved in regulating subthreshold membrane dynamics, and their imbalance has been associated with altered neuronal excitability and the emergence of abnormal rhythmic activity in basal ganglia circuits. HCN channels are known to modulate sag responses, subthreshold oscillations, and pacemaking properties11–13, while leak conductance contributes to membrane stability and determines the depth of hyperpolarization-induced troughs14. Given their complementary roles, the balance between and may define a critical axis of intrinsic regulation, whose disruption could contribute to pathological states such as Parkinsonian network dysfunction. To understand how these conductances interact to shape the functional identity of GPe neuronal subtypes, it is crucial to utilize computational modeling approaches that complement experimental findings. Several computational models of the GPe have been developed13,15,16. However, most of these models do not explicitly incorporate the neuronal heterogeneity evidenced in experimental studies8,10. In general, these approaches rely on networks composed of identical neurons and focus primarily on reproducing average firing rates. As a result, key subthreshold electrophysiological properties, such as sag dynamics, sag ratio, and trough potential, are often underexplored. Furthermore, the intrinsic variability observed within neuronal subtypes is rarely incorporated, limiting the ability of such models to capture biologically realistic dynamics. Therefore, there is a need for computational frameworks that account for both subtype-specific properties and intrinsic variability at the single-neuron level. Such an approach enables the identification of electrophysiological markers that distinguish neuronal subtypes. In addition, it allows for a systematic exploration of how variability in intrinsic conductances shapes neuronal function within physiologically plausible regimes. A central concept explored in this study is the notion of a solution manifold. From the perspective of dynamical systems theory, a manifold can be interpreted as a low-dimensional geometric structure embedded in a higher-dimensional space, which is locally Euclidean (i.e., locally flat), but may exhibit a complex global organization17,18. Intuitively, rather than assuming that a single set of ion channel conductances uniquely determines neuronal behavior, we consider that multiple combinations of conductances can produce equivalent electrophysiological outputs. This idea is closely related to the well-established principle of ion channel degeneracy, whereby similar activity patterns emerge from distinct underlying biophysical configurations19–21. In this context, the solution manifold corresponds to the subset of conductance combinations that simultaneously satisfy experimentally constrained electrophysiological properties, such as firing rate, sag ratio, and trough potential. Importantly, this manifold is not imposed a priori, but emerges from the intersection of multiple physiological constraints within the parameter space. As a result, it defines a structured, low-dimensional region embedded in a higher-dimensional conductance space, capturing the compensatory relationships that preserve neuronal function despite underlying variability. This framework provides a bridge between experimental observations and computational modeling, offering a principled way to interpret variability as an organized and biologically meaningful feature of neuronal systems. By varying and across a defined two-dimensional parameter space, we computed the firing rate, sag ratio, and trough potential for each conductance pair. These results were systematically compared to ex vivo electrophysiological recordings obtained from mouse GPe neurons10. This comparative analysis enabled the identification of specific regions in parameter space that reproduce experimentally observed neuronal behaviors. In particular, electrophysiological features such as firing rate, sag ratio, and trough potential were used as constraints to define biologically plausible regimes. Consistent with the concept introduced above, the intersection of these constraints defines a restricted region of conductance combinations, which we interpret as a low-dimensional manifold of physiological solutions embedded within a broader parameter space. Our simulations further revealed a robust inverse correlation between sag ratio and trough potential, consistent with experimental observations. This emergent relationship supports the biological plausibility of the model and suggests an underlying compensatory mechanism between intrinsic conductances. Altogether, these findings indicate that coordinated variations in a reduced set of conductances can account for subtype-specific electrophysiological behavior. Finally, this framework provides a mechanistic basis to investigate how disruptions in the balance between and may contribute to dysfunctional basal ganglia dynamics in pathological conditions such as Parkinson’s disease and drug addiction22–24. Methods Experimental ex-vivo data GPe neurons data were obtained from recordings performed for a prior publication10. All animal procedures were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the National Institute on Alcohol Abuse and Alcoholism Animal Care and Use Committee. For electrophysiological slice preparation experiments, mice were anesthetized with isoflurane and decapitated prior to rapid brain extraction. For immunohistochemistry procedures, animals were anesthetized with pentobarbital (50 mg/kg, intraperitoneally) before transcardial perfusion. Male C57BL/6J mice were used to identify PV-expressing neurons via post hoc immunostaining following whole-cell patch clamp electrophysiological recordings and 1% Neurobiotin filling procedures. Additionally, male and female FoxP2-Cre mice were used to target FoxP2-expressing neurons, after an AAV9.CAG.Flex.tdTomato.WPRE virus injection into the GPe 3–4 weeks prior to slice preparation and recordings. Briefly, brain slices (250–300 μm) containing the GPe were prepared in ice-cold, oxygenated cutting artificial cerebrospinal fluid (aCSF) containing (in mM): 194 sucrose, 30 NaCl, 4.5 KCl, 26 NaHCO, 1.2 NaHPO, 10 D-glucose, and 1 MgCl, using a vibratome and incubated for 30–40 minutes at 32°C in recovery aCSF composed of (in mM): 124 NaCl, 4.5 KCl, 26 NaHCO, 1.2 NaHPO, 10 D-glucose, 1 MgCl, and 2 CaCl. After recovery, slices were maintained at room temperature until recording. Electrophysiological recordings were performed in whole-cell current clamp mode at 30–32°C. Micropipettes were pulled from borosilicate glass capillaries (2–4 MΩ) and filled with the following pipette solution (in mM): 140 potassium gluconate, 10 Hepes, 0.1 CaCl2, 2 MgCl2, 1 EGTA, 2 ATP-Mg, 0.2 GTP-Na, pH 7.25 (290–295 mOsm). Neurons were visualized using an upright microscope with a LUMPlanFL N 40×/0.80 W objective. Data was acquired using a Multiclamp 700A amplifier and a Digidata 1322A digitizer (Molecular Devices, San Jose, CA, USA). Whole-cell current-clamp recordings (I = 0) were performed for five minutes, and the spontaneous firing rate (Hz) was quantified from the final minute of each trace. To assess intrinsic membrane properties, hyperpolarizing current steps of −200, −150, and −100 pA (1 s duration each) were applied, but only the −200 pA step was used to calculate the sag ratio and trough potential (mV). These electrophysiological parameters, measured in PV- and FoxP2-expressing neurons, served as empirical constraints for the computational model. Data were analyzed using pClamp 10.3 software (Molecular Devices, San Jose, CA, USA). Hodgkin–Huxley model for ark and proto: GPe neurons We developed a computational model to represent the PV and FoxP2 neuronal subtypes of the external globus pallidus (GPe) based on the Hodgkin-Huxley formalism25–28,by incorporating elements from the models proposed by Gunay et al.13 and Fujita et al.15. The model includes nine ionic currents: fast sodium (NaF), persistent sodium (NaP)29, delayed rectifier potassium (Kv2), fast-activating potassium (Kv3), high-voltage-activated calcium (CaH), calcium-activated potassium (SK), leak current, and HCN currents subdivided into fast (HCNf) and slow (HCNs) components12. This division of the HCN current is one of the main innovations of this model, allowing a more accurate representation of the distinct dynamic profiles of the neuronal subtypes. The general membrane equation is expressed as: where: The expressions for each current are as follows: Additionally, the term represents an external current (external stimulus) applied to the neuron. Gating variable dynamics The temporal evolution of the gating variables dynamics (m, h e s) follows the differential equation: The steady-state activation function () and time constant () are defined as: For the slow inactivation variable s of the NaP channel, the time constant is calculated by: Intracellular calcium dynamics The intracellular calcium concentration varies according to: where γ is the ratio between the membrane area and the cell volume, Z is the valence of the calcium ion, and F is the Faraday constant. Calcium-dependent SK channel The SK channel depends on the calcium concentration: the time constant is: Time constant for HCN channel For the HCN channels, the time constant follows: Numerical integration Simulations were performed using the fourth-order Runge-Kutta method, with time step of , over of 5 s, with the first 2 s were discarded to remove transient effects and ensure model stabilization. The simulations were implemented in the MATLAB environment using the DynaSim toolbox30. Simulated biophysical parameters for arkipallidal and prototypical neurons The electrophysiological properties chosen for the two neuronal subtypes of the GPe—arkypallidal (Ark) and prototypic (Proto)—were characterized based on available experimental data (ex vivo) from the literature8,10,12,24. According to the experimental data: (i) the firing rate of Ark neurons shows a distribution with a first quartile (Q1) of 8 Hz, median (Q2) of 10 Hz, and third quartile (Q3) of 18 Hz. Proto neurons, on the other hand, fire at higher frequencies, ranging from 35 Hz (Q1) to 80 Hz (Q3), with a median of 53 Hz. The sag ratio index,defined as the ratio of the steady-state voltage deflection to the peak (minimum) voltage deflection, also differentiates the subtypes: Ark neurons present higher values, between 1.28 (Q1) and 1.46 (Q3), with a median of 1.35, while Proto neurons show sag ratios between 1.05 and 1.13, with a median of 1.10. Additionally, the minimum potential (trough potential) of Ark neurons exhibits more hyperpolarized values, ranging from − 140 mV to − 175 mV, with a median of − 150 mV. Proto neurons, however, concentrate between − 85 mV and − 102 mV, with a median of − 95 mV. In the simulations (in silico), conductance densities of selected ion channels were systematically varied to assess the impact of heterogeneity on the membrane properties of GPe neurons. These parameters were defined based on previous computational models and electrophysiological data13,15,16. NaF, Kv3, CaH, SK channels were simulated with fixed conductances for both neuronal GPe neuron subtypes. According to24, prototypic PV-positive neurons (corresponding to the Proto subtype in our model) display prominent intrinsic conductances, including high densities of persistent sodium (NaP), HCN , Kv4 , and delayed rectifier potassium conductances. In contrast, FoxP2-positive neurons (arkypallidal subtype, corresponding to Ark subtype in our model) exhibit considerably lower levels of these same conductances. Based on this distinction, we selected higher conductance values for NaP, Kv, HCN, and delayed rectifier channels when simulating Proto neurons, and lower values when simulating Ark neurons. This parametrization was crucial to reproduce the electrophysiological phenotypes observed experimentally for each neuronal subtype. For the persistent sodium (NaP) channel, the range of conductance densities observed in the literature is between: 13,15. So, we used for simulating the Ark neuron dynamics and for simulating Proto neuron dynamics. The Kv2-type potassium current, which contributes to repolarization and the control of action potential duration, has been reported in the literature between 13,15. The minimum value was used to simulate Ark neuron dynamics and the maximum value to Proto neuron. The HCN channels, which participate in modulating the resting potential and membrane resonance, show conductances from 13,15. We chose the range ****to simulate the Proto neuronal dynamics and to simulate the Ark neuronal dynamics to capture the different functional profiles associated with the variability in HCN expression observed among GPe neuronal subtypes. These intervals were chosen to better visualize the electrophysiological responses consistent with experimental values. Finally, the leak conductance, responsible for maintaining the passive resting membrane potential, was simulated within the range of . These intervals were systematically adopted to compose the parametric value space of the simulations, enabling characterization of the individual and combined influence of each channel on the electrophysiological dynamics of the neuronal models. Table 1 presents the basic set of biophysical parameters used in the conductance-based model of GPe neurons, including the Proto and Ark subtype. These parameters define the maximal conductance densities (in ) for the main ionic channels included in the model, such as the NaP, the Kv2, the HCN, and the passive leak current. For each ionic channel, the table lists the range of conductance values used in our simulations to explore how different configurations reproduce experimentally observed firing behavior and membrane potential dynamics. Multiple combinations were tested to identify conductance sets that best match ex vivo electrophysiological recordings. Kinetic parameters are provided in the supplementary material Table 4. Table 1Parameters used. Bold values indicate parameters that differ between Ark subtype and Proto subtype neurons. The dash “–” indicates that the parameter was varied over a specific range.ParameterArk subtypeProto subtype110.22.01001000.12.032320.40.40.20.2–00–––50 mV50 mV− 90 mV− 90 mV− 60 mV− 60 mV130 mV130 mV− 30 mV− 30 mVγ3000030000F96485 sA/mol96485 sA/molZ220.4 ms0.4ms0.010.014.64.60.350.3555 (SK channel)4 ms4 ms (SK channel)76 ms76 ms Parameters used. Bold values indicate parameters that differ between Ark subtype and Proto subtype neurons. The dash “–” indicates that the parameter was varied over a specific range. Figure 1 provides a schematic overview of the methodology adopted in this study, which integrating computational simulations with ex vivo electrophysiological data to investigate the intrinsic dynamics of distinct GPe neuron subtypes. The top part of the figure illustrates the modeling approach, in which a conductance-based neuron model is used to simulate the membrane potential dynamics of prototypical and arkypallidal neurons in response to current injection protocols. Panels (a) and (b) display representative Hodgkin-Huxley model schematics and corresponding voltage traces obtained from simulations, with parameters tuned to reproduce characteristic features such as sag ratio, trough potential (arrows), and firing rate for each neuronal subtype. Validation of GPe neuronal subtypes through computational modeling and experimental electrophysiological recordings. (a) Schematic overview of the computational model based on the Hodgkin-Huxley (HH) formalism, used to simulate the electrophysiological properties of Protopypical and Arkypallidal neurons in the GPe. The model incorporates multiple ion channels with distinct conductance profiles to reflect the heterogeneity of neuronal subtypes. (b) Simulated membrane potential responses to hyperpolarizing current injection for prototypical (left) and Arkypallidal (right) neuron models. The Arkypallidal subtype displays a more pronounced sag (arrow) compared to the prototipycal subtype. (c) schematic of the ex vivo experimental protocol study10. Whole-cell patch-clamp recordings were performed on acute coronal brain slices from C57Bl/6J mice and FoxP2-Cre transgenic mice. (d) Experimental membrane responses to hyperpolarizing current injection recorded from PV (left) and FoxP2 (right) neurons. Consistent with the computational predictions, FoxP2 neurons exhibited a significantly larger sag response, validating the model’s ability to distinguish the functional characteristics of GPe neuronal subtypes. The bottom part of the figure illustrates the experimental approach, highlighting whole-cell patch-clamp recordings performed in PV-positive neurons from brain slices of wild-type (C57BL/6J mice) and FoxP2-positive neurons from brain slices of transgenic FoxP2-Cre mice. Panels (c) and (d) show how specific GPe neuron subtypes were targeted and recorded, yielding ex vivo voltage responses that serve as empirical counterparts to the simulated data. These recordings exhibit distinct electrophysiological profiles for PV and FoxP2 neurons, which are used to validate the computational model. This combined approach enables a systematic exploration of how different combinations of intrinsic conductances account for the observed variability in neuronal behavior, thereby linking specific ion channel properties to the functional diversity of GPe neuron subtypes. Additionally, to enable a tractable exploration of the parameter space, we adopted a reductionist approach in which a subset of conductances (NaF, kv3, CaH, and SK) was held fixed while gleak and gHCN were varied. This approach reduces the dimensionality of the parameter space and enables a systematic analysis of compensatory interactions between conductances that primarily govern subthreshold dynamics. Parametric space and subset permitted electrophysiological response Once the conductance ranges and have been defined, phenotypic responses of interest (k) are calculated through numerical simulation, such as firing rate, trough potential, and sag ratio. The simulation, for the chosen values of vs. , generates parametric space . These results are visualized in the heatmap graphs below for the Proto neuron (Fig. 3)e Ark (Fig. 4). To ensure biological plausibility, we define a subset () values from the parameter space of in which each electrophysiological response remains within the experimentally observed limits (for example, for the Prototype neuron, the firing rate values found in the literature vary between a value of Q1=35Hz and Q3=80Hz approximately). We then adjust the and intervals, for k = firing rate, and find a subset (), within the set, defined by the new sub-interval and (double arrows in the figures). For each phenotypic response k, we define the set of allowed conductances as: Intersection set of physiological response Once we define , we observe that each response (k) produces a different subinterval and . Thus, we need to find a region in the parameter space that simultaneously satisfies all (k). So, we define an intersection set for the conductances (red double-arrows in Figs. 3 and 4d), for a specific neuron: and, is the intersection subspace for each electrophysiological response (), yielding a constrained biophysical domain consistent with all experimental observations for a given neuronal subtype = {Proto, Ark}. Heatmap exploration of the electrophysiological characteristics of the Proto neuron. (a) Firing rate as a function of leak conductance () and slow-activating HCN channel conductance (). Warmer colors (red) indicate higher firing frequencies, which increase with both and . The black double-headed arrows along each axis indicate the conductance ranges that define the subset , corresponding to regions where the firing rates match experimental observations. This valid region spans the entire range of , but requires lying between 0.14 and 0.35 mS/cm2. (b) Heatmap of the trough potential as a function of and . The double black arrows indicate the acceptable conductance ranges where the trough potential remains within physiologically plausible limits. This region spans the full range of , with restricted between 0.17 and 0.35 mS/cm2. (c) Sag ratio map as a function of and . In this case, the constrained region is defined by between 0.07 and 0.2 mS/cm2, and between 0.17 and 0.35 mS/cm2. (d) Intersection region of the physiologically valid regions from panels (a), (b), and (c). Red double-headed arrows indicate the final conductance limits defining the intersection region (). Heatmap exploration of the electrophysiological characteristics of the Ark neuron. (a) Firing rate as a function of leak conductance () and fast-activating HCNf conductance (). The black double-headed arrows along the horizontal and vertical axes indicate the conductance ranges that yield viable firing frequencies for this neuronal subtype. This valid region is restricted to specific combinations of and and defined according to electrophysiological experimental recording10. (b) Trough potential map as a function of and . The valid region for trough potential includes all values of while constraining to the range between 0.08 and 0.2 mS/cm2. (c) Sag ratio map as a function of and . The permissible region for sag ratio is delimited by values between 0.02 and 0.06 mS/cm2, and values between 0.07 and 0.20 mS/cm2. (d) Intersection region of the valid conductance ranges obtained from panels (a), (b), and (c). The red double-headed arrows indicate the intersection limits, defining the intersection region (). Manifold of physiological solutions: functional coupling between conductances In this work, we use the term solution manifold to describe the subset of parameter space in which the electrophysiological properties of GPe neurons emerge as consistent and physiologically plausible solutions. In the mathematical literature, a solution manifold denotes a continuous set of solutions that satisfy a system of constraints, typically forming a lower-dimensional structure embedded in a higher-dimensional parameter space. In conductance-based neuronal models, this concept arises from the well-established principle of ion channel degeneracy, whereby similar electrophysiological outputs can be produced by multiple distinct combinations of underlying conductances19,20. In this context, the solution manifold represents a structured subset of conductance pairs that jointly reproduce experimentally constrained electrophysiological features, such as trough potential and sag ratio. This framework moves beyond isolated parameter tuning and instead captures the intrinsic interdependence between ionic conductances, whereby variations in one channel can be compensated by adjustments in another while preserving physiological function, an expected consequence of degeneracy in neuronal systems8. Within the restricted two-dimensional projection explored in this study, this compensatory relationship can be approximated by a first-order functional dependence: where a < 0 reflects the compensatory nature of the interaction between these ion channels. We interpret this relationship as defining a one-dimensional projection of a higher-dimensional solution manifold embedded within the reduced parameter space . Importantly, this reduced manifold should not be interpreted as the full space of solutions, but rather as a low-dimensional slice of a broader degenerate solution space. Therefore, the observed linear structure does not arise as an artifact of parameter restriction, but instead reflects an organizing principle of the system within the explored subspace, consistent with compensatory mechanisms expected in conductance-based neuronal models. Formally, the solution manifold is given by: This formulation provides a tractable representation of the solution space, highlighting the compensatory coupling between conductances while preserving consistency with experimentally observed neuronal dynamics. Similar low-dimensional compensatory structures have been reported in conductance-based models, where correlated variations in ionic conductances define structured subspaces that preserve neuronal activity19,20. Sampling of the solution manifold for comparison with experimental data Once the solution manifold is defined as a constrained subspace of physiologically plausible conductance combinations, we describe how this manifold was sampled to generate simulated datasets for comparison with experimental observations. To enable quantitative comparison with experimental data, we generated simulated datasets by sampling discrete combinations of conductance pairs along the solution manifold. For each neuronal subtype, we sampled multiple points along the manifold to capture the intrinsic variability permitted within the constrained parameter space. This ensemble of simulations was then compared with the empirically observed variability. Individual membrane potential traces correspond to representative realizations of the model (i.e., single simulations) selected along the solution manifold, whereas the boxplots (see Fig. 6) are constructed from multiple independent simulations, following the same statistical framework applied to the biological data. Importantly, simulated data points do not correspond to individual neurons, but rather to distinct parameter configurations that reproduce the same set of electrophysiological constraints. This sampling strategy is consistent with the principle of ion channel degeneracy. Accordingly, the variability observed in the simulated datasets reflects structured variability within a constrained solution space, rather than random dispersion. Comparison between electrophysiological properties in computational models and ex vivo recordings. Panels (a–c) display results from computational simulations comparing Ark subtype (red) and Proto subtype (green) neurons, while panels (d–f) present ex vivo experimental data comparing FoxP2-positive (red) and parvalbumin-positive (PV, green) neurons, from Abrahao and Lovinger10. (a) Spontaneous firing rates: Proto subtype neurons exhibit significantly higher firing rates and greater variability compared to Ark subtype neurons (p<0.001). (b) Trough potential: Proto neurons display less hyperpolarized minimum membrane potentials relative to Ark neurons (p<0.01). (c) Sag ratio: Ark V neurons show significantly higher sag ratios, suggesting a more prominent HCN-mediated depolarizing response (p<0.05). (d) ex vivo firing rates: PV neurons show substantially higher firing frequencies than FoxP2 neurons (p<0.001), with broader variability. (e) Trough potential: PV neurons exhibit more depolarized trough values compared to FoxP2 neurons (p<0.001). (f) Sag ratio: FoxP2 neurons have significantly higher sag ratios than PV neurons, consistent with differential HCN channel expression or activation (p<0.001). Asterisks indicate statistical significance: *p<0.05, **p<0.01, ***p<0.001. To further ensure that the observed relationships are not artifacts of parameter restriction, we verified that nearby points in parameter space (local perturbations around the manifold) preserve the qualitative compensatory relationship between and , indicating robustness of the identified structure within the explored subspace. Statistical analysis To quantify differences in electrophysiological properties between neuronal subtypes in, both computational simulations and ex-vivo experiments, we used the nonparametric Wilcoxon rank-sum test (Mann–Whitney U test)31, which does not assume data normality or a specific data distribution. This test evaluates whether two groups differ significantly by comparing the ranks of the observed values. Statistical significance was denoted by asterisks based on p values p: p < 0.001 (***), p < 0.01 (**), p < 0.05 (*); and p \ge 0.05 was considered not significant (ns). Analyses were performed using the ranksum function in MATLAB. To assess the relationship between sag ratio and trough potential across neuronal subtypes, we performed both linear regression and Pearson correlation analyses on simulated and experimental datasets. For the simulated neurons, data points were extracted along the diagonal of the intersection region in the parameter space—defined as the set of conductance pairs that simultaneously satisfy empirical constraints for firing rate, sag ratio, and trough potential. These points correspond to physiologically plausible configurations derived from the Proto and Ark models. For the experimental data, we used ex vivo whole-cell recordings from genetically identified PV and FoxP2 neurons, from which sag ratio and trough potential were extracted using standardized protocols under a − 200pA hyperpolarizing current steps. In both datasets, each neuronal subtype was independently fitted using a simple linear model: . Regression was performed using least-squares minimization, and the residual norm was computed to assess goodness of fit. In addition, the Pearson correlation coefficient (r) was calculated to quantify the strength and direction of the linear association between the two variables. These analyses enabled the comparison of slope, intercept, residuals, and correlation strength across neuronal subtypes, providing a quantitative basis for interpreting the observed co-variation as a structured, subtype-specific phenomenon. Results The initial analyses in this study, using the Hodgkin-Huxley like model, focused on investigating the effects of HCN and leak ion channels on neuronal dynamics for Ark and Proto neurons8,24 [see Fig.(1)]. Thus, the results demonstrate not only a distinction in the responses of the simulated electrophysiological properties, but also a correlation between HCN and leak channels for each neuronal subtype. Impact of HCN and leak channels on electrophysiological response Figure 2 presents a simulation of a generic neuron, demonstrating how the application, for varying intensities of the HCN and leak ion channel conductances, affects electrophysiological responses, including firing rate, sag, and plateau voltage, during a current clamp protocol (500 ms of hyperpolarizing current step). In Fig. 2a, a high leak conductance condition is applied, and the membrane potential is compared in the absence (black solid line) and presence (blue dashed line) of fast HCN channels (). Without HCNf, the response remains mostly static during the step. With HCNf, a pronounced sag emerges, characterized by a rapid partial recovery of the membrane potential during the stimulus, followed by a rebound depolarization after the step, both hallmark features of HCN activation.Furthermore, plateau voltage is higher compared to the condition without HCN, indicating the rapid modulatory role of the channel. The dashed horizontal line represents the firing threshold for reference. Figure 2b maintains the same HCNf configuration as in (a) but with a markedly reduced leak conductance. Under this condition, the membrane potential becomes more hyperpolarized during the stimulus (current step), and the sag promoted by is more prominent. The reduced increases the relative contribution of HCNf to the recovery dynamics, reflecting a more sensitive balance between inward and outward currents. In Fig. 2c, high leak conductance is maintained, as in (a), but is replaced with slow HCN channels (). In this case, the slower kinetics of lead to a delayed recovery, with reduced sag amplitude. Finally, Fig. 2d explores the combination of low leak conductance and . As in panel (b), the membrane potential reaches deeper hyperpolarization levels during the step current. However, because of the slow activation kinetics of , the recovery is delayed, as same observed in (c). Furthermore, the decrease in the leakage channel intensity results in a lower plateau value compared to the plateau in (b). Effects of HCN channel subtype and leak conductance on neuronal membrane dynamics via Hodgkin Huxley simulation. This figure compares the influence of fast-activating HCN () and slow-activating HCNs () channels, in conjunction with high or low leak conductance (), on membrane potential responses to a hyperpolarizing current step (clamp), during 500ms for all four panels. (a) Under high leak conductance, two conditions are shown: absence of HCNf (solid black line) and high-conductance HCNf expression (dashed blue line). The presence of HCNf enhances membrane potential recovery, leading to a faster return to a steady-state voltage level (Plateau voltage). The dashed black horizontal line denotes the threshold potential. (b) With low leak conductance, the same HCNf conditions (null and with HCNf) produce more hyperpolarized responses compared to panel (a). The sag amplitude is marked, defined as the difference between the lower (minimum) hyperpolarized membrane potential and the steady-state level reached during the current step. (c) Under high leak conductance and with only HCNs (slow-activating), the recovery of membrane potential is slower and less pronounced. The arrow indicates the trough potential, i.e., the minimum voltage reached during the hyperpolarization. (d) With low leak conductance and HCNs the membrane potential becomes even more hyperpolarized, but the slow recovery dynamics characteristic of HCNs remain. The arrow indicates the plateau voltage, reflecting the relatively sustained hyperpolarization before recovery. After simulating the individual effects of leak conductance and HCN on neuronal responses [see Fig.(2)], we constructed a set of electrophysiological responses ((referred to as parameter space - ), shown in heatmaps, to explore how these two ion channels influence the electrophysiological behavior of Proto and Ark neurons, based on systematic variations of the leak conductance () versus the slow HCN conductance (). The conductance intervals of and were kept the same for the three simulated parameter spaces. In Fig. 3a, the spontaneous firing rate of the Proto neuron is mapped as a function of and . Lower firing rates ≈ 15Hz (blue regions) are observed when both conductances are small, while higher rates ≈ 85Hz (red regions) emerge when and are increased. The double-headed horizontal and vertical arrows define the boundaries of , shown in Eq. (19), according to experimental observations10. Figure 3b shows the heatmap for the distribution of the trough potential. The most hyperpolarized values () occur at low intensities of and , while more depolarized troughs () appear at higher conductance levels. Additionally, it is observed that leak conductance predominates as the primary factor modulating the trough potential; that is, changes in the trough potential values occur (from left to right). Again, the arrows represent the parameter ranges that match experimental observations. Panel (c) presents the sag ratio. An increase in leads to a higher sag ratio, particularly for moderate values of . In this figure, we observe that HCN conductance predominates as the primary factor in modulating the sag ratio; that is, changes occur mainly as the gHCNs increases (from bottom to top). The double-headed arrows delineate the acceptable parameter range based on the Proto neuron’s experimental sag ratio. Finally, Fig. 3d shows the results by plotting the intersection subset regions () of valid parameter ranges identified in panels (a), (b), and (c). The shaded region highlights the conductance combinations that simultaneously satisfy all three electrophysiological constraints. The red double-headed arrows mark the boundaries of this intersection space, according to the Eq. 21. Analogous to Figs. 3, 4 illustrates the electrophysiological properties of the Ark neuron. Figure 4a presents the firing rate response across this conductance space. Lower firing rates are observed at low values of and , while higher frequencies occur with increasing conductance values. The horizontal and vertical double-headed arrows indicate the acceptable ranges of and that reproduce the experimentally observed firing frequency of Ark neurons. Panel (b) shows how the trough potential varies across the same parameter space. The most hyperpolarized values appear in regions that depend primarily on low , while depolarized valleys emerge at higher conductance values (left to right shift). The double-headed arrows delineate the conductance ranges consistent with the experimental trough potential, specifically between = 0.08–0.2 mS/cm2 and = 0–0.1 mS/cm2. Figure 4c maps the sag ratio as a function of and . As expected, increases in lead to higher sag ratios, especially when leak conductance is low. The double-headed arrows again highlight the conductance values that yield sag ratios consistent with electrophysiological measurements. Finally, panel (d) displays the intersection of the valid conductance ranges defined by the previous three panels. The shaded region indicates the parameter space where all electrophysiological properties—firing rate, trough potential, and sag ratio—are simultaneously satisfied. The red double-headed arrows mark the boundaries of this feasible region. Intersection of the valid conductance ranges To better visualize the solution manifold underlying the intersection region, we extracted all conductance pairs that simultaneously satisfied the experimental constraints defining and represented them in the two-dimensional parameter space. Figure 5a shows the resulting scatter plot for Proto neurons, where each point corresponds to a valid conductance combination within the intersection region. By overlaying these data, a linear regression reveals a clear negative correlation between and , defining a one-dimensional manifold of solutions embedded in the two-dimensional parameter space. Visualization of the solution manifold in the leak–HCN conductance space. The figure shows scatter plots of all valid conductance pairs that simultaneously satisfy the experimentally observed electrophysiological properties for Proto and Ark neurons. In Fig. 5a, the distribution of valid solutions for prototypical neurons reveals a significant inverse linear relationship between and (slope = -1.025 ± 0.455, , ), indicating a structured compensatory interaction between these conductances. Bootstrap analysis (1000 resamples) confirmed the robustness of this relationship, with a 95% confidence interval of the slope given by [− 1.4663, − 0.6175]. In Fig. 5b, arkypallidal neurons also exhibit an inverse relationship between and (slope = -0.075 ± 0.060, , p = 0.0386), although with greater dispersion, suggesting a weaker compensatory trend within a broader conductance subspace. Bootstrap analysis yielded a wider 95% confidence interval ([− 0.1418, 0.0216]), reflecting increased variability in this subtype. Solid black lines indicate linear regression fits. A similar analysis for Ark neurons is shown in Fig. (5)b. Although the admissible parameter region is shifted toward lower and values compared to Proto neurons, the valid solutions again align along an inverse linear relationship. This compensation curve indicates that increases in leak conductance must be counterbalanced by corresponding decreases in HCN conductance to preserve the electrophysiological properties of Ark neurons. Manifold electrophysiological statistics: simulated vs. ex vivo data Following the heatmap analyses and the development of the intersection methodology [see Figs. 3, 4], we extracted the solution manifold values () representing the correlated variations between and that simultaneously preserved key electrophysiological features. From this constrained parameter space, we collected the simulated measurements for Ark and Proto neuron subtypes, which were then statistically compared. To assess the validity of our model, these results were directly compared with ex vivo electrophysiological data from PV (n=9) and FoxP2 (n = 14) neurons reported by Abrahao and Lovinger10. Figure 6 presents a comparative analysis of electrophysiological properties derived from our computational simulations and experimental ex vivo recordings. Panels (a–c) display results obtained from the computational model, based on the manifold approach (Eq. 23) of the intersection region identified in previous figures. Panels (d–f) show corresponding empirical data from the study by Abrahao and Lovinger10. Panel (a) shows the spontaneous firing rates of Ark (red) and Proto (green) subtypes neurons, revealing a highly significant difference (p < 0.001), with Proto neurons firing at considerably higher rates. Panel (b) displays the trough membrane potential for both subtypes, with Ark neurons exhibiting more hyperpolarized values compared to Proto neurons (p < 0.01). In (c), the sag ratio is significantly greater in Ark neurons (p < 0.05), indicating enhanced subthreshold depolarizing responses. In the biological dataset (panels d–f), similar trends are observed. Panel (d) shows the spontaneous firing activity of FoxP2-positive (red) and PV-positive (green) neurons, with PV neurons firing at significantly higher rates (p < 0.001). Panel (e) reveals that PV neurons have less hyperpolarized trough potentials than FoxP2 neurons (p < 0.001), while panel (f) shows that FoxP2 neurons display higher sag ratios (p < 0.001). Overall, these results demonstrate consistent patterns across simulated and experimental data, supporting the functional analogy between Ark/FoxP2 and Proto/PV neuronal subtypes. The statistical correspondence between the two datasets supports the validity of our modeling approach in capturing key electrophysiological distinctions observed in biological neurons. Correlation between trough potential and sag ratio Figure 7 illustrates the relationship between sag ratio and trough potential, using data from both computational simulations (panel a) and ex vivo experiments (panel b). In panel (a), results are shown for Ark subtype (red inverted triangles) and Proto subtype (green squares) neurons. These values were extracted from the manifold according to Eq. (23) of the intersection region highlighted in Figs. 3d and 4d, where the leak conductance () and fast HCN conductance () co-vary within a physiologically plausible range. Panel (b) displays the same relationship derived from ex vivo data for FoxP2-positive (red triangles) and PV-positive (green circles) neurons. In both panels, a clear negative correlation is observed between the trough potential and the sag ratio: neurons with more hyperpolarized trough potentials tend to exhibit higher sag ratios. This inverse relationship suggests that electrophysiological properties, particularly those influenced by and , are tightly coupled and vary systematically across neuron subtypes. Importantly, Ark subtype neurons (simulation) and FoxP2 neurons (ex vivo) exhibit more hyperpolarized trough potentials and higher sag ratios than their respective counterparts (Proto subtype and PV neurons), reinforcing the functional distinction between these neuronal populations. These findings align with the electrophysiological signatures used to differentiate fast-spiking from slow-spiking subtypes. Comparison of the relationship between trough potential and sag ratio in simulated and ex vivo data. (a) Scatter plot showing the relationship between trough potential and sag ratio obtained from computational simulations for Proto subtype neurons (green squares) and Ark subtype neurons (red inverted triangles). (b) Corresponding relationship based on ex vivo experimental recordings, comparing PV neurons (green circles) and FoxP2-positive neurons (red upright triangles). In both panels, a negative correlation is observed between trough potential and sag ratio, suggesting that more hyperpolarized membrane potentials are associated with higher sag responses. Colored dashed lines represent linear regression fits for each neuronal subtype. This convergence between simulation and experiment supports the consistency of the modeled biophysical mechanisms with empirical observations. To further support the hypothesis that the physiological parameter space is structured along a solution manifold, we performed linear regression analyses on the relationship between two key electrophysiological features: sag ratio and trough potential. These analyses were conducted using both simulated and ex vivo data, as illustrated in Fig. 7. Our goal was to assess whether these variables exhibit coordinated subtype-specific variation, consistent with a biologically constrained subspace. In all cases, we observed a consistent negative linear correlation between sag ratio and trough potential: neurons with more hyperpolarized troughs tend to exhibit greater sag amplitudes, consistent with the stronger activation of HCN-mediated inward currents. Table 2 summarizes the results of the linear fits and includes the corresponding Pearson correlation coefficients (r), which quantify the strength and direction of this relationship: Table 2Linear regression and Pearson correlation results for sag ratio as a function of trough potential.ConditionSlope (a)Intercept (b)Residual normPearson rSimulation — Proto-0.0040.7030.107-0.628Simulation — Ark-0.0060.4770.070-0.630ex vivo — PV-0.0010.9920.049-0.628ex vivo — FoxP2-0.0030.8650.198-0.823 Linear regression and Pearson correlation results for sag ratio as a function of trough potential. The values of r confirm a robust negative correlation in all conditions, particularly for the FoxP2 subtype in ex vivo recordings (r = -0.823), indicating a tight functional link between the two metrics. The steeper slopes observed for Ark/FoxP2 neurons (both simulated and experimental) suggest a subtype-specific enhancement in HCN sensitivity, resulting in more pronounced sag effects as the membrane potential becomes more hyperpolarizedhyperpolarizes. Together, these findings quantitatively reinforce the interpretation that the solution manifold observed in the intersection region represents a biologically meaningful axis of co-regulation, rather than a modeling artifact. This linear correlation defines a constrained yet flexible subspace in which electrophysiological variability is achieved through coordinated adjustments in passive and active membrane properties. Discussion The GPe is a key structure within the basal ganglia, involved in diverse motor and cognitive functions. The discovery of cellular heterogeneity in the GPe, particularly the identification of two major neuronal subtypes, arkypallidal (Ark) and prototypical (Proto) neurons, has prompted more detailed investigations into the biophysical mechanisms underlying their distinct electrophysiological profiles2,3,7. We constructed a detailed Hodgkin-Huxley-like computational model to simulate the intrinsic dynamics of Ark and Proto neurons. Model parameters were constrained using experimentally observed electrophysiological features, rather than selected as unique optimal values, reflecting the non-uniqueness of conductance-based solutions. The modeling focused on three key electrophysiological properties: spontaneous firing rate, trough potential, and sag ratio, all measured under current-clamp protocols and compared to ex vivo empirical data reported by Abrahao and Lovinger10. By constraining the model through empirical benchmarks, we restricted the parameter space to physiologically plausible regions, without assuming parameter uniqueness, thereby explicitly accounting for the inherent indeterminacy of conductance-based models. Although traditional models often isolate the influence of individual ionic currents, emerging perspectives emphasize the importance of integrative conductance configurations21,32. Despite recent advances, there is still no consensus in the literature regarding which ion channels are primarily responsible for modulating specific electrophysiological features such as the sag ratio, trough potential, and firing rate. This lack of agreement stems from several factors, including variations in experimental protocols, species and strain differences, heterogeneous expression of channel isoforms (e.g., different HCN subtypes), interactions between ionic currents, and differing assumptions in computational modeling frameworks13,33. Correlation between leak and HCN ion channels Experimental and computational modeling studies indicate that the sag is an emergent electrophysiological property resulting from the activation of HCN channels in the neuronal membrane, which generate a depolarizing current during hyperpolarizing stimuli12,33,34. In contrast, the leak conductance () plays a fundamental role in determining the resting membrane potential and the depth of hyperpolarization in response to negative current injections, thereby modulating the basal excitability of the neuron13,14. Consistent with previous biophysical findings10, our simulations confirmed the complementary roles of leak and HCN channels in shaping sag dynamics and rebound depolarization. More specifically, the interplay between HCN kinetics and background conductance contributed to reproducing the distinct response latencies and amplitudes observed in Ark and Proto neurons, although alternative conductance configurations may also produce similar outputs. The agreement between simulations and experimental recordings supports the physiological plausibility of the model, although it does not imply parameter uniqueness. Not only were central tendencies (e.g., median sag ratio and trough potential) replicated, but the direction and strength of inter-variable correlations were consistent across conditions. This convergence suggests that the model captures key organizational constraints of GPe excitability within the explored parameter subspace. Our results revealed a restricted admissible region in the space for Proto neurons, where experimental firing rates, sag ratios, and trough potentials intersect. This constrained subset space defines a physiologically meaningful regime. Within this restricted region, linear regression revealed an inverse relationship between and , which should be interpreted as a local approximation rather than a universal law. We interpret this observation within the framework of solution manifolds, where compensatory relationships between conductances emerge as a consequence of degeneracy: increases in one conductance can be offset by decreases in another while preserving electrophysiological function. In Ark neurons, simulations also revealed a correlated compensation between leak and fast HCN conductances, but within a distinct subdomain compared to Proto neurons. The Ark solution manifold is shifted toward lower and higher values, consistent with their increased sag responses. These findings are consistent with prior transcriptomic studies suggesting differential expression of HCN2/HCN4 across subtypes12. To quantitatively assess this relationship, we performed linear regression analyses on the valid conductance pairs (Fig.5). For Proto neurons, we found a significant inverse relationship between and (slope = -1.025 ± 0.455, , ), indicating a relatively well-constrained compensatory interaction between these conductances. In contrast, arkypallidal neurons also exhibited a statistically significant inverse relationship (slope = -0.075 ± 0.060, , p = 0.0386), but with substantially greater dispersion around the fitted line. To further evaluate the robustness of these relationships, we performed a bootstrap analysis (1000 resamples). For prototypical neurons, the 95% confidence interval of the slope remained strictly negative ([− 1.4663, − 0.6175]), confirming a stable and robust inverse relationship between and . In contrast, for arkypallidal neurons, the bootstrap interval was broader and included values close to zero ([− 0.1418, 0.0216]), although the mean slope remained negative. This result indicates a weaker and more variable compensatory trend, consistent with a less constrained conductance relationship in this neuronal subtype. This suggests that, while a compensatory trend is present in both subtypes, Ark neurons occupy a broader region of the solution space, consistent with weaker constraints along the explored projection. Furthermore, beyond its geometric interpretation, this inverse relationship between leak and HCN conductances likely reflects a homeostatic rule that helps GPe neurons preserve their intrinsic firing properties and maintain desynchronized network activity, even in the presence of variations in ionic channel gene expression, levels of neuromodulation, or metabolic state19,35. From a functional perspective, such compensatory interactions may contribute to the stability of input–output properties, although multiple conductances are likely involved in this regulation. This type of compensation naturally fits within the broader framework of ion channel degeneracy, according to which distinct combinations of biophysical parameters can support equivalent electrophysiological behaviors36. In this context, our results suggest that GPe neurons operate along a restricted yet robust region of the conductance parameter space, which preserves consistent functional behavior within the basal ganglia circuitry despite underlying biological heterogeneity. Importantly, focusing on the leak–HCN relationship does not imply that other conductances play a negligible role in shaping neuronal stability. Although the present work highlights the leak–HCN relationship as a key axis of intrinsic homeostasis, it is plausible that other channels, including different Kv subtypes, Ca currents, and other potassium currents as the BK channels conductances, which strongly shape spike repolarization and fast afterhyperpolarization, contribute additional dimensions to this compensatory manifold13. These channels likely contribute additional dimensions to a higher-dimensional compensatory structure, reinforcing the idea that neuronal stability emerges from coordinated adjustments across multiple conductances. Intersection subspace and solution manifolds: biological and computational implications The identification of a solution manifold within the electrophysiological parameter space represents an important conceptual advance in the modeling of GPe subtypes. Rather than considering ionic conductances in isolation, the manifold emphasizes the interdependent nature of biophysical parameters: variations in can be systematically compensated by reciprocal changes in , yielding stable and physiologically consistent neuronal responses. This observation is consistent with the idea that neuronal identity is not defined by single conductance values, but by a continuum of combinations that collectively preserve function37,38. This perspective aligns with the concept of degeneracy, where different sets of biophysical parameters can produce equivalent dynamics, conferring robustness and flexibility to the nervous system39,40. From a mathematical standpoint, the solution manifold can be interpreted as a low-dimensional projection of a higher-dimensional conductance space41,42. This approach provides a natural explanation for the electrophysiological variability observed in experimental recordings, since neurons of the same subtype may occupy different points along the manifold while maintaining comparable output features20,43. Furthermore, this framework suggests that pathological states, such as those observed in Parkinson’s disease, may be associated with shifts or distortions of this manifold, although this hypothesis requires direct experimental validation. In this sense, the concept of a solution manifold strengthens the methodological link between simulation and experiment, while also offering a testable hypothesis of how intrinsic conductance variability contributes both to resilience and to vulnerability in neuronal dynamics, across scales from membranes to behavior44. Across both simulated and empirical datasets, we identified a subtype-specific correlation between sag ratio and trough potential, consistent with coordinated regulation of depolarizing (HCN-mediated) and stabilizing (leak) currents. From a functional standpoint, this trade-off supports a form of dynamic compensation, whereby changes in excitability are constrained along an axis of preserved responsiveness. The identification of a solution manifold in the (, ) parameter space suggests a biologically meaningful axis of compensatory regulation. Such manifolds align with the concept of conductance degeneracy20,38, where different ionic configurations produce similar electrophysiological outputs. Our findings reinforce that GPe neurons likely exploit this degeneracy to preserve functionality in the face of biological variability or pathological insult. More broadly, the modeling framework presented here illustrates how physiologically grounded parameter spaces can be used to reveal compensatory structures that would be difficult to infer from data alone. As such, the approach provides not only mechanistic insight but also a flexible platform for hypothesis testing, pharmacological prediction, and further extensions to network-level models. Conclusion We presented a biophysically grounded Hodgkin-Huxley model to investigate how the interaction between leak conductance () and hyperpolarization-activated cyclic nucleotide-gated (HCN) channels () modulates the intrinsic properties of GPe neuronal subtypes. Our simulations reproduced key physiological features observed in ex vivo recordings, including firing rate, sag ratio, and trough potential for both prototypic (PV-expressing) and arkypallidal (FoxP2-expressing) neurons, within a constrained region of the parameter space. A central methodological contribution was the definition of an intersection region in the conductance space, which revealed a low-dimensional projection of a broader manifold of physiological solutions. This structure captures a compensatory relationship between leak and HCN currents within the explored subspace, consistent with the principles of degeneracy and robustness in neural systems. Furthermore, we identified a subtype-specific correlation between the sag ratio and trough potential, observed both in silico and ex vivo, highlighting a coordinated balance between depolarizing and stabilizing membrane mechanisms. These results provide a theoretical framework linking ion channel co-regulation to subtype-specific electrophysiological properties within the GPe, without assuming unique parameter configurations. Although our model focused on PV and FoxP2 subtypes, the framework can be extended to other GPe neuronal populations, such as those expressing NPas1 and Lhx6. Together, these subtypes, characterized in detail by Abrahao and Lovinger10, suggest a broader landscape of GPe diversity that may be governed by similar conductance-based organizing principles, although this generalization requires further investigation. From a translational perspective, this framework provides a principled approach to investigate how imbalances in the – axis may contribute to pathological dynamics, particularly in neurodegenerative diseases such as Parkinson’s disease45,46. Disruptions to this compensatory regime may destabilize neuronal excitability and facilitate aberrant oscillatory activity in basal ganglia circuits, although direct causal links remain to be experimentally validated. Future extensions of this model may incorporate synaptic input variability, network coupling, and plasticity to assess how the manifold behaves under realistic activity regimes and disease conditions, including additional mechanisms such as synaptic interactions and extracellular effects (e.g., ephaptic coupling), which may modulate pathological beta-band oscillations47–50. Taken together, our results support the role of structured compensatory mechanisms in shaping the electrophysiological diversity of GPe neurons and suggest that conductance manifolds may provide a useful framework for interpreting both physiological variability and its disruption in disease. Supplementary Information Supplementary Information 1. Acknowledgements MPBS is the recipient of the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES - Brazil) fellowship (#88887.900715/2023-00). GMC is the recipient of the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq - Brazil) fellowship (#140895/2021-3). GZDL is the recipient of the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq - Brazil) fellowship (#309440/2022-0). We acknowledge Dr. David Lovinger for his permission to include data from our previous publication (Abrahao and Lovinger, 2018) in this article. The biological data used were supported by the Sao Paulo Research Foundation (FAPESP) [grant numbers: 2019/01686-0; 2022/15972-7]. Author contributions M.S. conceived the code model and results, G.C, G.T, G.C, K.A and G.L analyzed the results. All authors wrote and reviewed the manuscript. Data availability The datasets generated and analysed during the current study are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Information 1. Data Availability Statement The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.