Observing hidden neuronal states in experiments Authors: Dmitry Amakhin, Anton Chizhov, Guillaume Girier, Mathieu Desroches, Jan Sieber, Serafim Rodrigues Published: 2025 ABSTRACT In this article we demonstrate a general protocol for constructing systematically experimental steady-state bifurcation diagrams for electrophysiologically active cells. We perform our experiments on entorhinal cortex neurons, both excitatory (pyramidal neurons) and inhibitiory (interneurons). A slowly ramped voltage-clamp electrophysiology protocol serves as closed-loop feedback controlled experiment for the subsequent current-clamp open-loop protocol on the same cell. In this way, the voltage-clamped experiment determines dynamically stable and unstable (hidden) steady states of the current-clamp experiment. The transitions between observable steady states and observable spiking states in the current-clamp experiment provide partial evidence for stability and bifurcations of the steady states. This technique for completing steady-state bifurcation diagrams in a model-independent way expands support for model validation to otherwise inaccessible regions of the phase space. Overlaying the voltage-clamp and current-clamp protocols leads to an experimental validation of the classical slow-fast dissection method introduced by J. Rinzel in the 1980s and routinely applied ever since in order to analyse slow-fast neuronal models. Our approach opens doors to observing further complex hidden states with more advanced control strategies, allowing to control real cells beyond pharmacological manipulations. Introduction When characterising the dynamics of nonlinear systems, one fundamental criterion for a model is if its stable states such as stationary solutions or periodic orbits match experimental observations. The ability to fit and validate models is, thus, greatly expanded by experimental tools with the capacity to unveil non-observable (sensitive or dynamically unstable) states that are otherwise inaccessible to standard measurements. The combination of observable and non-observable states gives access to an experimental equivalent of parameter-dependent families of stable and unstable states in a model, which are usually referred to as a bifurcation diagram. This article applies a new experimental technique of using feedback control to find unstable states to electrophysiology experiments on neuronal cells. Our aim is to support systematic validation of neuron models by comparing bifurcation diagrams and observing their between-cell variability. We focus on unstable parts of input-dependent families of steady-state solutions obtained by feedback-controlled experiments and compare them with indirect evidence from standard measurements from open-loop experiments. This extends recent work of Ori et al. [1,2] constructing phase diagrams from neuronal data, and complements other approaches such as using data to verify the bifurcation structure of neuronal models [3,4], model-based data analysis [5] or parameter estimation from data [6]. Our technique is a simplification of the so-called control-based continuation (CBC) method, an approach which has been recently demonstrated in mechanical experiments [7], vibrations and buckling experiments [8], pedestrian flow experiments [9], atomic-force microscopy [10], cylindrical pipe flow simulations [11], and feasibility studies for synthetic gene networks [12,13]. Indeed, CBC is a procedure that combines feedback control [14] and pseudo-arclength continuation [15] in a model-free environment, that is, only reliant upon noisy experimental data in a closed-loop control setup. The objective is to compute experimental bifurcation diagrams, that is, both families of stable and unstable states (either stationary or periodic) together with bifurcation points joining such families. Importantly, the control is noninvasive in the sense that it vanishes once an equilibrium or periodic of the uncontrolled system has been reached; the control is decreased iteratively using a Newton’s method. In our case, the approach is simpler and does not require the use of Newton iterations since we consider a slow ramp on the control target. Hence we exploit de presence of two timescales in the experimental procedure in order to obtain directly an approximation of the steady-state bifurcation curve, and we explain why such a bifurcation curve is obtained at low cost; see below. Our approach is hence useful for both modelers and experimentalists, as it relies upon standard protocols routinely used in patch-clamp electrophysiology. It can further decipher the excitability class of a given real neuron as well as help fitting a model to data by using both current-clamp and voltage-clamp protocols. In our electrophysiology experiments on entorhinal cortex neurons we treat the cell as an electric circuit, and apply a voltage clamp (VC) [16,17], followed by current clamp (CC). In the VC setup the electrode acts as a voltage source at the neural membrane, fixing the potential across the neural membrane, measuring the current, while the CC setup adds a fixed external current, measuring the resulting membrane potential; see the illustration for the experimental setup in Fig 1A and the S1 Text for further details. In-vivo neurons are subject to current signals that drive spiking (oscillatory) or rest (steady) states of the neural membrane potential [18]. This makes the CC setup the open-loop part of the protocol. In contrast, the VC part of our experiment is the closed-loop feedback-controlled part of the protocol because the voltage source regulates the external current to maintain the hold voltage. VC has been applied successfully to study neuronal nonlinear current-voltage relationships, so-called N-shaped I-V characteristics, which cause enhanced neuronal excitability and influence the regenerative activation of certain ionic currents (e.g., sodium) [19–23] across the neural membrane, into and out of the cell. We show that the VC protocol with a slowly varied reference voltage signal gives access to stable and unstable neuronal steady states of the neuron, which was hinted at in [24,25]. In contrast, the open loop CC protocol with a slowly varying applied current always follows stable (observable) states, driving the neuron to dynamically transition between its observable rest states and its observable spiking states. A: Sketch of experimental setup with brain slice (a), patch pipette (b), reference electrode (Ag-AgCl pellet) connected to ground (c), amplifier (e, Multiclamp 700B) with CV-7B headstage (d), AD-converter (f, National Instruments NI USB-6343) and standard PC computer (g, https://pixabay.com/vectors/computer-desktop-workstation-office-158675/). B: VC and CC protocol runs for cell 5: (Ivc(t),Vh(t))-curve for VC run (thin bright blue: unfiltered data with sampling time step 5×10−5 s, blue/red/brown: median of Ivc over moving windows of size Δw=4×103 steps equalling 0.2s) and (Ih(t),Vcc(t))-curve for CC run (orange, mean of Ih over moving windows. C: (Ivc(t),Vh(t))-curves for VC protocol of all 5 cells on waterfall Vh-axis (color coding indicates conjectured stability as indicated for (b)). We interpret these combined experimental protocols (VC and CC) using multiple-timescale dynamics, in particular, the dissection method [26], which reveals the dynamic bifurcations in the experiment, as demonstrated in the experimental bifurcation diagram in Fig 1B. In this slow-fast framework, the states traced by the VC protocol with slow variation correspond to the steady-state experimental bifurcation diagram of the so-called fast subsystem of the mathematical model describing the protocol [26], that is, the model system with constant input current. Hence, the N-shaped I-V relation of a neuron should be seen as a S-shaped V-I bifurcation diagram, such as in Fig 1B. Following this strategy, we demonstrate the feasibility of tracking a family of neuronal steady states (stable and unstable) via variations of reference signal and reparameterizing the obtained curve using the feedback current. Results The experimental neuronal bifurcation diagram in Fig 1B shows the time profiles (Ih(t),Vcc(t)) of the CC protocol run (orange, thin) and (Ivc(t),Vh(t)) of the VC protocol run (bright blue, thin) for cell 5 overlaid in the (I,V)-plane. For both runs Ih(t) and Vh(t) were slowly increased, respectively (see Materials and Methods). After smoothing, the VC time profile is the S-shaped curve (Ivc,sm(Vh),Vh) (blue/brown/red, thick). It equals the (I,V)-characteristic of the stationary neuronal states of the CC protocol, including dynamically unstable states (brown and parts of red). The transition to stable spiking states is compatible with (Hodgkin) class-I excitability, however we do not have sufficient data to distinguish the different classes of excitability. See Fig B in S1 Text for how the interspike intervals depend on Ih. The dynamical stability and instability of stationary states is inferred based on two pieces of evidence: (i) the negative slope of the (Ivc,sm,Vh)-curve (brown) after smoothing over a moving window with larger size (Δw=1.5×104 steps equaling = 0.75s in Fig 1C), or (ii) by the presence of slow-fast oscillations (neuronal firing) in the CC run at Ih equalling the Ivc,sm (red). Criterion (i) implies instability for topological reasons. Criterion (ii) provides only circumstantial evidence. In Fig 1B (cell 5) the darker shading of the (orange-colored) CC run indicates slow dynamics as identified by the norm |(Vcc′(t)Δt,Vcc′′(t))Δt| being less than a fixed threshold (after smoothing of Vcc(t) and Vcc′(t), see S1 Text for detailed definition). This shading shows that the firing oscillations in Fig 1B spend most time near their voltage minimum Vmin for the respective Ih(t). Fig 5C and 5E show an embedding of the firing oscillations of cell 5 into the (Vcc,Vcc′)-plane, and a zoom near Ih≈200 pA. The projection and zoom indicate that the firing oscillation passes slowly near the presumed fixed point, but leaves its neighborhood again. At the low-current end of the (Ivc,sm,Vh)-curve, the CC protocol for cell 5 shows no firing oscillations (recall that Ih is ramped up) in Fig 1B. Thus, the part of the (Ivc,sm,Vh) curve between fold and presence of slow-fast oscillations is labelled “undetermined” (colored yellow) as the combination of single-ramp VC and CC protocol do not provide evidence for or against stability of this part. We do not label the “transition” from stability label “undetermined” to “unstable” as a Hopf bifurcation as the slow parts of the firing oscillations are approximately 20 mV below the equilibrium indicated by the VC run. So, it is unclear if (and where precisely) a change of stability occurs between Idb≈365 pA and the fold of steady states at Ih≈3 pA. The stability boundaries of the stationary states are labelled as bifurcations in Fig 1B. The change of stability near the disappearance of stable spiking states at Idb is labelled as a Hopf bifurcation. We observe that for Ih>Idb small-amplitude oscillations are visible, which emerge from relaxation type oscillations with a slow phase near the stationary state for Ih