Insights on the cellular mechanisms of intracranial electrical stimulation using optical imaging in preclinical models

Abstract

Intracranial electrical stimulation is increasingly used to treat neurological and psychiatric conditions. However, the underlying therapeutic mechanisms at the cellular and circuit levels remain poorly understood. Exciting progress in the development of genetically encoded neural activity indicators enables artifact-free, high-spatiotemporal-resolution optical analysis of the impact of intracranial stimulation on membrane potentials, cytosolic calcium, and neurotransmitter and neuromodulator dynamics. Using these optical imaging tools in vivo, in the mammalian brain, preclinical studies probed the effect of intracranial stimulation on individual neurons and population network dynamics across timescales ranging from milliseconds to minutes or more. These studies have revealed complex, stimulation parameter-dependent effects across cell types and provided experimental support for various therapeutic mechanisms. We discuss these studies in the context of clinical observations and highlight the exciting potential of optical imaging in advancing the mechanistic understanding of clinical electrical stimulation in epilepsy and other neurological and psychiatric orders.

Introduction

Intracranial electrical stimulation has been FDA approved for treating various neurological and psychiatric diseases, including epilepsy, Parkinson's disease, essential tremor, obsessive-compulsive disorder, and more [[1], [2], [3], [4], [5], [6]]. Two main types of clinical therapeutic stimulation are Deep Brain Stimulation (DBS) and Responsive Neurostimulation (RNS). DBS targets deep brain structures and delivers continuous or bursts of stimulation in an open-loop fashion, though progress in closed-loop DBS design has highlighted its exciting potential for improved therapeutic outcomes [7,8]. The more recently developed RNS therapy targets broad cortical and subcortical structures and delivers brief bouts of stimulation in response to detected pathological biomarkers in a closed-loop system [9].

Despite the exciting therapeutic benefits, the underlying cellular and circuit mechanisms of intracranial electrical stimulation remain unclear. Leading hypotheses include 1) axonal activation of local neurons and axons of passage [[10], [11], [12]]; 2) cellular activity suppression of firing or responsiveness to inputs [10,13,14]; 3) functional informational lesion via cellular activity disruption [10,15,16] and network desynchronization [9,10,17,18]; 4) synaptic and network plasticity [3], [19], [20], [21]; 5) glial recruitment [3], [22], [23], [24], [25], [26] (Fig. 1). These potential neurophysiological mechanisms have been thoroughly discussed in various review articles (e.g. [3,10,13,[27], [28], [29]]).

Fig. 1.

Hypothesized cellular and circuit mechanisms of electrical stimulation. Illustration of a locally implanted electrode, and its associated electric field in yellow, 1) directly activating a myelinated excitatory neuron and nearby afferent and efferent fibers via anti- and orthodromic axonal activation, which 2) suppresses neural activity via vesicle depletion or increased membrane conductance, both of which 3) cause a functional informational lesion in the stimulated cell that disrupts their coordination with other neurons and desynchronizes pathological network activity. Stimulation also 4) triggers cellular and network plasticity, and 5) activates glia which further modulate neural excitability and activity.

Because electrical stimulation generates electrical interference, real-time measurement of the stimulation effects is challenging using traditional electrophysiological approaches [27]. Additionally, electrophysiology techniques cannot report the genetic identity of the recorded cell, making it difficult to investigate cell-type-specific response to electrical stimulation. Progress in cellular optical neural activity imaging, particularly cytosolic calcium imaging and membrane voltage imaging, has enabled artifact-free optical recording of the real-time effects of electrical stimulation on specific cell types in preclinical animal models [28].

In this review, we highlight studies investigating the cellular effects of intracranial electrical stimulation with optical imaging techniques in vivo, in the mammalian brain (summarized in Table 1). Such studies have provided insights into the complex, stimulation parameter-dependent effects across cell-types (summarized in Table 2) and, furthermore, experimental support for various proposed therapeutic mechanisms (summarized in Table 3). We discuss these results within the context of clinical observations and emphasize the exciting potential for optical imaging to advance mechanistic understanding of clinical electrical stimulation for epilepsy and other neurological and psychiatric disorders.

Table 1.

Summary of in vivo cellular optical imaging studies of electrical stimulation effects.

Ref	Imaging modality	Cell type/Neurotransmitter	Brain region	Animal model	Imaging state	Parameters	Stimulation interface
Bekar et al, 2008 [30]	2P OGB-1AM calcium imaging	Astrocytes (OGB-1AM)	Cortex	FVB/NJ mice	Anesthetized	Freq: 25, 75, 125, 200 Hz	Concentric bipolar electrode (FHC)
						Amp: 15, 25, 50 μA
						PW: 60 μs
						Dur: 10 s
Histed et al, 2009 [31]	2P OGB-1AM Calcium imaging	Neurons and astrocytes (OGB-1AM and SR101)	Visual cortex	C57BL/6 mice; long Evans rats; Cat	Anesthetized	Freq: 250 Hz	Metal electrode (tungsten or Pt/Ir; D = 10–50 μm; SA = 79–1963 μm2; FHC) Or glass Pipettes (filled with ACSF; D = 3–5 μm; custom)
						Amp: <10 μA
						PW: 400 μs
						Dur: 100–815 ms
Michelson et al, 2019 [32]	2P GCaMP6s calcium imaging	Pyramidal neurons (Thy1-GCaMP6s)	Somatosensory cortex (layer II/III)	C57BL/6J-Tg(Thy1-GCaMP6s)	Anesthetized	Freq: 10–250 Hz	Single-shank 16-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: ∼50 μA
						PW: 100 μs
						Dur: 30 s
Eles et al, 2020 [33]	2P GCaMP6s calcium and GluSnfr glutamate imaging	Pyramidal neurons (Thy1-GCaMP6s) and glutamate release (hSyn-iGluSnFr)	Somatosensory cortex (layer II/III)	C57BL/6J-Tg(Thy1-GCaMP6s) and C57/BL6	Anesthetized	Freq: 100 Hz	Single-shank 16-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: 5, 10, 15, 20 μA
						PW: 100 μs cathodic, 50 μs interphase, 200 μs anodic
						Dur: 30 s
Stieger et al, 2020 [34]	2P GCaMP6s calcium imaging	Pyramidal neurons (Thy1-GCaMP6s)	Somatosensory cortex (layer II/III)	C57BL/6J-Tg(Thy1-GCaMP6s)	Anesthetized	Freq: 10 Hz	Single-shank 16-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: Co-varied
						PW: Co-varied
						Dur: 30 s
						2.48–2.56 nC/phase; varied phase asymmetry (PW/amplitude)
Trevathan et al, 2021 [35]	1P mini-scope GCaMP6m Calcium imaging	All cells (CAG-GCaMP6m)	Striatum (STN stimulation)	C57BL/6; 6-OHDA lesioned mice	Awake and anesthetized	Freq: 30, 80, 130 Hz	Bipolar electrode (Pt; D = 250 μm/wire, SA = 49,100 μm2/wire; PlasticsOne)
						Amp: Animal specific (130–220 μA anesthetized; 30–80 μA awake)
						PW: 140 μs or 200 μs
						Dur: 10 s
Wu et al, 2021 [36]	Fiber photometry GCaMP6f Calcium imaging	D1 and D2 MSNs (GCaMP6f)	Nucleus accumbens	C57BL/6; Drd1-cre; A2a-cre; Hedonic feeding	Awake	Freq: 30, 80, 130 Hz	Four PtIr wires around central optical fiber (D = 100 μm/wire, SA = 7900 μm2/wire; Doric lenses)
						Amp: 100–1000 μA
						PW: 90 μs
						Dur: 1–3 h continuous or responsive
Ma et al, 2021 [37]	2P OGB1-AM calcium imaging	Neurons and astrocytes (OGB-1 AM and SR101)	Somatosensory cortex (layer II/III)	C57BL/6	Anesthetized	Freq: 10–300 Hz	Non-penetrating cortical stimulation; glass micropipette (filled with ACSF; D = 20–35 μm; custom)
						Amp: 25–225 μA
						PW: 200–1800 μs
						Dur: 2, 20, 30, 40 s
Eles et al, 2021 [38]	2P calcium and 1P glutamate imaging	Pyramidal neurons (Thy1-GCaMP6s) and glutamate release (hSyn-iGluSnFr)	Cortex (layer II/III)	C57BL/6J-Tg(Thy1-GCaMP6s)and C57/BL6	Anesthetized	Freq: 10, 100 Hz (uniform), 100 Hz burst stimulation (averaged to 10 Hz)	Single-shank 16-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: 15 μA (cathodic), 7.5 μA (anodic)
						PW: 100 μs cathodic, 50 μs interphase, 200 μs anodic
						Dur: 30 s
Schor et al, 2022 [39]	Fiber photometry GCaMP6s calcium imaging	STN glutamatergic neurons, SNr GABAergic neurons, and STN projecting-M1 neurons (GCaMP6s)	STN, SNr, M1 hyperdirect pathway	VGlut2-Cre, VGAT-cre, C57/BL6; 6-OHDA lesioned PD mice	Awake	Freq: 5–40, 60–100, 120–180 Hz	Three bipolar metal electrode pairs (stainless steel; D = 76.2 μm/wire, SA = 4600 μm2/wire; custom)
						Amp: 200 μA
						PW: 60 μs
						Dur: 1 min
Wang et al, 2022 [40]	2P GCaMP6s calcium imaging	PV and all neurons (hSyn-GCaMP6s)	Motor cortex (layer II/III)	PV-Cre and C57BL/6	Anesthetized	Freq: 10, 100, 130 Hz	Bipolar electrode pair (Nichrome; D = 65 μm/wire 0.2–0.5 mm exposed length, SA = 16,300–35,800 μm2/wire; custom)
						Amp: 200 μA
						PW: 800 μs
						Dur: 5 s
Kim et al, 2022 [41]	1P miniscope GCaMP6s calcium imaging	Pyramidal neurons (Thy1-GCaMP6s)	CA1	C57BL/6J-Tg(Thy1-GCaMP6s)	Awake	Freq: Single pulse (<2 Hz)	Bipolar electrode pair (stainless steel D = 60 μm/wire, SA = 2800 μm2/wire; custom)
						Amp: 130–200 μA
						PW: 200 μs
						Dur: 200 μs
Stieger et al, 2022 [42]	2P GCaMP6s calcium imaging	Pyramidal neurons (Thy1-GCaMP6s)	Cortex (layer II/III)	C57BL/6J-Tg(Thy1-GCaMP6s)	Anesthetized	Freq: 10, 100 Hz	Single-shank 16-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: Co-varied
						PW: Co-varied
						Dur: 30 s
						2.5 nC/phase; phase asymmetry varied
Telega et al, 2022 [43]	Fiber photometry GRABDA2m	Extracellular dopamine	Nucleus accumbens (medial forebrain bundle stimulation)	Long-Evans rats	Awake	Freq: 130 Hz	Bipolar metal electrode (Pt/Ir; μm/wire, SA = 12,700 μm2/wire; custom)
						Amp: 100–300 μA
						PW: 100, 250, or 350 μs
						Dur: 5 or 20 s
Lowet et al, 2022 [44]	1P SomArchon voltage imaging	All neurons (syn-SomArchon with CoChR linked)	CA1	C57BL/6	Awake	Freq: 40, 140 Hz	Metal electrode and skull screw (stainless steel; D = 127 μm, SA = 12,700 μm2; custom)
						Amp: Animal specific (10–60 μA)
						PW: 400 μs
						Dur: 1 s
van den Boom et al, 2023 [45]	1P miniscope GCaMP6f calcium imaging	Putative cortical pyramidal neurons and striatal medium spiny neurons (Thy1-GCaMP6f)	Prefrontal cortex and striatum	Thy1-GCaMP6f; SAPAP3 −/− OCD model	Awake and anesthetized	Freq: 1–180 Hz	Bipolar electrode pair (Pt/Ir; D = 75 μm/wire, SA = 4400 μm2/wire; custom)
						Amp: 100, 200, 300 μA
						PW: 40, 80, 160 μs
						Dur: 60 s
						Also tested cyclic DBS (10 s ON and 10, 5, or 1 s OFF for 60 s)
Wu et al, 2023 [46]	2P GCaMP6s calcium imaging	Pyramidal neurons (Thy1-GCaMP6s)	Somatosensory cortex (layer II/III)	Thy1-GCaMP6s	Anesthetized	Freq: 2–200 Hz	Single-shank 16-channel silicon probes (SA = 177 or 703 μm2/sites; NeuroNexus)
						Amp: 1–100 μA
						PW: 200 μs (100 μs interphase)
						Dur: 10 s
Hughes and Kozai, 2023 [47]	2P GCaMP6s calcium imaging	Pyramidal neurons (Thy1-GCaMP6s)	Visual or somatosensory cortex (layer II/III)	C57BL/6J-Tg(Thy1-GCaMP6s)	Anesthetized	Freq: 0–200 Hz	Single-shank 16-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: 0–20 μA
						PW: 200 μs cathodic, 100 μs interphase, 400 μs anodic
						Dur: 1 s or 30 s
						Dynamic patterning of stimulation frequency and/or amplitude
Dadarlat et al, 2024 [48]	2P GCaMP6s calcium imaging	Excitatory and inhibitory neurons (syn-GCaMP6s with Tdtomato labeling)	Primary visual cortex (layer II/III)	GAD2-ires-cre × Tdtomato	Awake	Freq: 250 Hz	Microelectrode and chloridized silver wire in contralateral frontal lobe (Pt/Ir; SA = 3900 μm2; microprobes for life sciences)
						Amp: 3–50 μA
						PW: 200 μs cathodic, 200 μs inter-phase, 200 μs anodic
						Dur: 100 ms (25 pulses) every 10 s
Song et al, 2024 [49]	Fiber photometry GCaMP6s/m calcium imaging	GABAergic (mDlx-GCaMP6s) and dopaminergic (mTH-GCaMP6m) neurons	Ventral tegmental area (nucleus accumbens stimulation)	C57BL/6; chronic unpredictable mild stress model	Awake	Freq: 100 Hz	Bipolar metal electrodes (Tungsten; D = 220 μm/wire, SA = 38,000 μm2/wire; custom)
						Amp: 100 μA
						PW: 100 μs
						Dur: 1 h/day (30 min unilaterally per hemisphere) for 2 weeks
Ravasio et al, 2025 [50]	1P GCaMP7f Calcium imaging	All neurons (Syn-GCaMP7f)	CA1 and sensorimotor cortex	C57BL/6	Awake	Freq: 40, 140, 1000 Hz	Metal electrode and skull screw (stainless steel; D = 127 μm, SA = 12,700 μm2; custom)
						Amp: Animal specific (5–340 μA)
						PW: 90, 400 μs
						Dur: 5 s
Hughes et al, 2025 [51]	2P GCaMP7b Calcium imaging	All neurons (AAV-syn-GCaMP7b), inhibitory neurons marked (AAV-FLEX-tdTomato)	Visual cortex	VGAT-Ires-Cre	Awake	Freq: 10 or 100 Hz uniform, 1 Hz burst (10 pulses at 100 Hz), 5 Hz (theta) burst (2 pulses at 100 Hz)	Single-shank 4-channel silicon probes (SA = 703 μm2/sites; NeuroNexus)
						Amp: 10 μA cathodic, 5 μA anodic
						PW: 200 μs cathodic, 100 μs interphase, 400 μs anodic
						Dur: 30 s
Yuan et al, 2025 [52]	Fiber photometry GCaMP7s calcium imaging	CaMKII neurons (CaMKII-GCaMP7s) and GABAergic interneurons (DIO-GCaMP7s)	Anterior cingulate cortex	C57BL/6 and VGAT-cre (chronic restraint stress model)	Awake	Freq: 20 or 130 Hz	Bipolar metal electrodes (Ni/Cd; D = 75 μm/wire; SA = 4400 μm2/wire; custom)
						Amp: 100 μA or 180 μA (anodic) and −100 μA (cathodic)
						PW: 600 μs (symmetrical) or 100 μs (anodic) and 180 μs (cathodic)
						Dur: 2 h/day over 1 week or a single 2 h session
Shanazz et al, 2025 [53]	2P GRABAch3.0 acetylcholine imaging	Extracellular acetylcholine (hSyn-Ach4.3)	Somatosensory cortex (basal forebrain stimulation)	C57BL/6	Anesthetized	Freq: 20, 60, 100, or 130 Hz	Metal electrode and skull screw (Pt/Ir; D = 127 μm, SA = 12,700 μm2; A.M. Systems)
						Amp: Animal specific
						PW: 100 μs
						Dur: 5, 10, or 20 s
Zou et al, 2025 [54]	Fiber photometry GCaMP6m Calcium imaging	Glutamatergic neurons (GCaMP6m)	Ventromedial hypothalamus	vGlut2-Cre mice (various seizure and epilepsy models)	Awake	Freq: 1, 30 or 100 Hz	Bipolar metal electrodes (material not listed; D = 125 μm/wire, SA = 12,300 μm2/wire; A.M. Systems)
						Amp: Not listed
						PW: 100 μs
						Dur: 5, 10, 30, or 90 min depending on epilepsy model
Fabris et al, 2026 [29]	1P SomArchon voltage imaging	PV interneurons (syn-SomArchon)	Sensorimotor cortex	PV-Cre mice	Awake	Freq: 40, 140 Hz	Bipolar metal electrode pair (stainless steel; D = 127 μm/wire, SA = 12,700 μm2/wire; custom)
						Amp: Animal specific (50–300 μA)
						PW: 400 μs
						Dur: 1 s
Wang and Ravasio et al, 2025 [55]	1P ElectraOFF voltage imaging	CaMKII neurons (CaMKII-ElectraOFF)	CA1	C57BL/6	Awake	Freq: 40 Hz	Bipolar metal electrode pair (stainless steel; D = 127 μm/wire, SA = 12,700 μm2/wire; custom)
						Amp: 380 μA
						PW: 200 μs
						Dur: 1 min
Ye et al, 2026 [56]	2P GCaMP6s calcium imaging	CaMKII-GCaMP6s and GAD67-GCaMP6s (inhibitory)	Somatosensory cortex (layers II/III and V)	C57BL/6	Awake	Freq: 60, 160 Hz	Bipolar cortical stimulation; insulated metal electrode pair (details not listed in text; custom)
						Amp: 100 μA
						PW: 90 μs
						Dur: 2 min
Wang et al, 2026 [57]	1P GCaMP8m Calcium and SomArchon voltage imaging	All neurons (syn-SomArchon-P2A-GCaMP8m)	CA1	C57BL/6	Awake	Freq: 40, 140, 1000 Hz	Bipolar metal electrode pair (stainless steel; D = 127 μm/wire, SA = 12,700 μm2/wire; custom)
						Amp: 100–250 μA
						PW: 200 μs
						Dur: 700 ms

Abbreviations: 2P = two-photon; 1P = one-photon/widefield; Freq = frequency; Amp = Current Amplitude; PW = pulse-width; Dur = duration; PD = Parkinson's Disease, OCD = Obsessive Compulsive Disorder; STN = Subthalamic Nucleus; SNr = Substantia Nigra; M1 = Primary motor cortex; D = Diameter; SA = Surface Area.

Table 2.

Summary of optical cellular imaging investigation of stimulation parameters.

Tested effect	Description	Ref.
Parameter exploration
Frequency	Stimulation frequency is varied	[29,44,57,45,54,50,32,[46], [51], [56],35,[30], [37], [40], [52], [53],47,36,49,39]
Amplitude	Stimulation amplitude is varied	[45,48,46,33,37,30,47,36]
Pulse Width/Asymmetry	Pulse width and/or cathodic vs anodic phase asymmetry are varied	[45,42,37,52,34,43]
Duration	Stimulation duration is varied	[37,53,43,47]
Stimulation Patterning	Stimulation is delivered non-uniformly (burst, closed-loop, responsive, cyclic, dynamic etc.)	[48,38,51,47,36]
Spatial and cell-type characterization
Spatial Preference	The effect of relative location between neuron to stimulation site	[45,31,48,50,46,51,34]
Cell-Type Preference	The effect genetic identity (e.g. excitatory vs inhibitory) has on a neuron's response to electrical stimulation	[29,48,38,42,51,40,52,43,49]

Table 3.

Summary of optical cellular imaging investigation of specific electrical stimulation mechanism hypotheses.

Mechanistic hypothesis	Description	Ref.
Axonal activation	Axonal excitation releases neurotransmitters orthodromically and activates or blocks somatic firing antidromically	[45,31,32,51,35,34,39]
Suppression	Electrical stimulation decreases overall firing or responsiveness to inputs	[54,31,50,42,51,33,52,47]
Functional Informational Lesion at cellular and network levels	Electrical stimulation alters individual neuron's response to inputs and disrupts network synchrony	[29,[44], [55], [57],45,[36], [41], [49],39]
Plasticity/Network Reorganization	Electrical stimulation evokes persistent changes in connectivity, excitability, or structure	[55,47,49]
Glial Recruitment	Non-neuronal glia are affected by electrical stimulation and contribute to its overall therapeutic effect	[37,30]

Neuronal biophysical properties and electrical sensitivity

Neurons are electrically sensitive, and their intrinsic biophysical properties ultimately determine their response to electrical stimulation. The time-constant of the neuron governs how quickly its membrane potential responds to changes in current, which subsequently triggers neuronal firing. Neurons with smaller time-constants, such as smaller fast-spiking interneurons, can follow higher frequency inputs more readily than larger pyramidal cells with greater time constants. This difference in membrane properties suggests that different types of neurons can be differentially engaged by electrical stimulation [29,58]. Similarly, due to the low activation threshold at nodes of Ranvier with a high density of voltage-gated sodium channels, neurons with myelinated axons are more readily activated by rapid changes in the electric field when compared to neurons with unmyelinated axons [11]. However, with large enough changes in transmembrane potential, it is possible to affect the neural soma, dendrites, and other cellular compartments [29,[44], [55], [57]]. Thus, stimulation pulse parameters contribute to electric field dynamics and impact the evoked neuronal responses as discussed in detail below.

Stimulation parameter space

Intracranial electrical stimulation is comprised of charge-balanced biphasic square-wave electrical pulses. The stimulation parameter space for the delivery of these pulses is essentially infinite and can be specifically tuned for each patient to optimize individual therapeutic outcome. The main parameters are 1) frequency (the number of pulses per second, often referred to clinically as low ( < ∼100 Hz) and high ( > ∼100 Hz)), 2) amplitude (the current or voltage of each pulse), 3) pulse width (the duration per full biphasic pulse) and pulse phase asymmetry (relative difference in the duration of cathode vs anode phase, while maintaining charge-balance), 4) stimulation duration, and 5) stimulation patterning (continuous, burst, closed-loop, adaptive, responsive, etc.). As stimulation parameters are a critical consideration in programming DBS and RNS devices to achieve therapeutic benefit, many preclinical studies have investigated how these parameters modulate individual neurons and neural network dynamics (Table 2).

Preclinical optical imaging studies enable systematic investigation of the therapeutic mechanisms of intracranial stimulation with high spatiotemporal and cell-type resolution

Modern clinical DBS was pioneered in the early 1990s to treat tremor associated with movement disorders, which subsequently received FDA approval for continuous stimulation in the thalamic ventral intermediate nucleus and later in the subthalamic nucleus [[59], [60], [61], [62], [63]]. The overwhelming success of DBS in treating movement disorders led to its now FDA-approved use in several other neurological and psychiatric disorders [2,6,64,65], including obsessive compulsive disorder [4,66], depression [5,67], and epilepsy [1,[68], [69], [70]], and its exploratory use in Alzheimer's disease [[71], [72], [73],45]. In each case, different stimulation targets and intensities have been tested, but there has been limited exploration of the optimal stimulation parameters for these different neuropathologies and brain regions. The initial fully-implantable DBS studies delivered stimulation using modified Medtronic cardiac pacemakers with a maximum stimulation frequency of 130 Hz [59], leading to broad clinical adoption of ∼130 Hz despite the differences in targeted brain regions and underlying pathologies [2,5,6,65,74,75,66].

Preclinical models enable systematic analysis of stimulation parameters on cells and circuits, and such insights can help derive a principled understanding of stimulation effects and aid physician programming of stimulation parameters. When combined with disease-models, preclinical experiments may inform new clinical targets and strategies [20,58,45,54]. Furthermore, testing therapeutic efficacy in preclinical animal models allows for direct comparison of treatment versus sham while avoiding the concern associated with possible deterioration in human patients receiving sham stimulation [76,77].

Optical imaging using genetically encoded cellular activity indicators

Exciting progress in the development of chemical and genetically encoded neural activity indicators enables precise optical recording of membrane potentials, cytosolic calcium, and extracellular neurotransmitters and neuromodulators (e.g. GABA, glutamate, dopamine, and acetylcholine) [28,78,79]. Chemical activity indicators require intracranial or systemic injection to label the cells of interest, whereas genetically encoded cellular activity indicators are expressed in specific neuron types or glia using whole-body transgenic techniques, or viral-based gene transduction [28,78,80]. We largely focus on fluorescent indicators of cytosolic calcium and membrane voltage, as they are most often used to study the effects of intracranial stimulation, although important cell signaling information can also be gleaned from neurotransmitter and neuromodulator imaging.

Calcium indicators change their fluorescence in response to fluctuations in cytosolic calcium concentration [81,82]. As cytosolic calcium correlates with a wide variety of cellular signals, including action potentials, calcium concentration is used as an indicator of cellular activity. Further separation of calcium activity in neural soma versus neuropil allows more detailed analysis of stimulation effect in soma and neuronal processes [83,84]. It is important to note that increasing evidence indicates that the relationship between somatic cytosolic calcium and neuronal spiking is complex and biased toward spike bursts and slow membrane voltage dynamics [85,86]. In contrast, voltage indicators can directly measure neural spiking and subthreshold membrane potential fluctuations from individual neurons in the mammalian brain [[87], [88], [89]].

The slow kinetics of the calcium indicators and intrinsic cytosolic calcium dynamics limit the temporal resolution of calcium imaging to tens of milliseconds, but support the simultaneous imaging of hundreds to thousands of neurons, enabling sophisticated network analysis of neuronal ensembles (Fig. 2A‒D). In contrast, voltage indicators have sub-millisecond temporal resolution to capture detailed membrane potential information with single-spike precision and access to subthreshold membrane voltage dynamics (Fig. 2E). However, this technique requires high-speed imaging, typically beyond 500 frames per second which—due to hardware limitations—reduces the imaging field of view and limits the number of neurons that can be simultaneously imaged; although, advancements in microscopy techniques involving patterned light illumination and fast scanning enable the recording of tens of neurons simultaneously in awake, behaving mice [[90], [91], [92]].

Fig. 2.

Cellular calcium and voltage imaging in awake behaving mice to probe electrical stimulation effects. A) An example illustration of the animal preparation for testing the effect of electrical stimulation in the CA1 of behaving mice, showing optical imaging chambers with nearby stimulation and ground electrodes. B) Schematic of imaging experimental setup in freely locomoting, head-fixed mice. C). Schematic representation of a recorded neuron in the electric field generated by a nearby electrode. D) An example widefield GCaMP7f calcium imaging experiment investigating electrical stimulation effects. i) Max-min projected imaging field of view of the GCaMP7f fluorescence during a CA1 recording session testing 140 Hz stimulation, and ii) the same field of view overlaid with a heatmap depicting the theoretical electric field strength dissipation from the electrode tip. Solid white lines: electrode tip. Dotted white circles mark every 50 μm from the center of the electrode tip. iii) four example neurons' GCaMP7f fluorescence traces (outlined in yellow in i and ii), aligned to stimulation onset during 140 Hz (left) and four different example neurons' traces (n1-n4) during 40 Hz (right). Every trial is shown in gray (10 trials total); the average fluorescence trace across all trials is shown in red for activated neurons and blue for suppressed neurons. E) An example widefield SomArchon voltage imaging experiment investigating electrical stimulation effects. i) example SomArchon fluorescence before, during and after 140 Hz DBS, with a max-min projected image of the neuron shown on the left (scale bar, 15 μm). SomArchon trace is shown in black, detected spikes are marked by black ticks, and electrical stimulation pulse patterns are in gold. ii) similar to i, but for an example neuron during 40 Hz stimulation. A, B, D are adapted from Ravasio et al. [50], and C, E from Lowet et al. [44].

Summary of in vivo cellular optical imaging studies

Cellular optical imaging has been used to measure electrical stimulation effects in preclinical animal models for nearly two decades. Here, we review and summarize in vivo studies that evaluated the cellular and network responses to electrical stimulation across a range of stimulation parameters and tested specific mechanistic hypotheses. See Table 1 for an overview of the studies reviewed.

Consideration of brain regions, anesthesia, stimulation electrodes, and protocols

Stimulation-evoked effects become rapidly complex as the direct cellular effects induced by an electric field trigger subsequent indirect network responses which propagate through inter-connected circuits to further influence and modulate cellular responses [[31], [48], [93]]. For example, it has been shown that the relative distance from the electrode correlates to the strength of evoked calcium activation for frequencies less than 250 Hz [31,[32], [38], [42], [46], [50]], suggesting direct cellular effects. However, kilohertz electrical stimulation better activated neurons farther from the electrode than those nearer to the electrode, supporting stronger indirect network effects [50]. Likewise, in a study from Hughes et al., despite strong correlation between the neuron and neuropil fluorescence near the electrode (<225 μm) indicating direct cellular activation, the response profiles of neurons farther from the electrode become more diverse and display different temporal characteristics, indicating indirect network-driven cellular responses [51].

Electrical stimulation effects also critically depend on neural network dynamics and brain structures [29,44,50]. For example, 40, 140, and 1000 Hz electrical stimulation evoked different population calcium responses in the hippocampus versus the cortex [50]. 40 and 140 Hz stimulation induced a more balanced excitation and inhibition in the hippocampus, but predominantly induced inhibition in the cortex; meanwhile 1 kHz stimulation evoked population-level suppression in the hippocampus but not in the cortex. Similarly, voltage imaging of parvalbumin-positive interneurons showed that 140 Hz stimulation reliably paced subthreshold membrane potentials in the visual cortex, but not the motor cortex [29]. Likewise, calcium imaging throughout the primary somatosensory cortex demonstrated that within 50 μm of the electrodes, 60 and 160 Hz stimulation activated layer II/III neurons, with a longer delay during 60 Hz compared to 160 Hz [56]. However, in layer V neurons, 60 Hz stimulation suppressed activity and 160 Hz conversely evoked delayed activation after stimulation offset.

Similarly, brain-state (anesthetized vs awake) profoundly influences how a neural network processes information and how it responds to electrical stimulation [[94], [95], [96]]. Several of the reviewed preclinical studies demonstrate that anesthesia alters brain dynamics and attenuates electrical stimulation-evoked effects compared to awake states [57,45,35]. With continued technological advances, increasingly more studies have been conducted in awake animals (Table 1).

Finally, these studies also use different electrodes, surgical preparations, activity indicators (chemical or different versions of genetically encoded ones), mouse lines, and imaging modalities. The effect of electrode geometry and material in particular have been studied extensively (see review [97]). In addition to DBS and RNS, intracranial microsimulation also utilizes biphasic pulsatile electrical stimulation. It is often explored in brain computer interfaces and intraoperative testing in patients, but generally uses smaller electrodes and higher current density than DBS or RNS [[98], [99], [100], [101]]. Despite such variations, many of the reviewed studies reported similar findings—including stimulation-evoked waveform shape and response-strength correlation to stimulation parameter changes—pointing toward unified cellular mechanisms. For example, calcium imaging studies with differences in many experimental parameters all reported similar stimulation-evoked calcium activation shapes and patterns in single cells [45,50,46,33,37], highlighting the translational validity of these findings.

Investigation of the effects of electrical stimulation parameters

To explore the therapeutic parameter space, various studies have investigated the real-time effects of stimulation parameters paired with cellular imaging techniques to test the effects of pulse parameters and pulse train patterns in the brains of awake or anesthetized animal models (Table 2, Fig. 3).

Fig. 3.

Simplified overview of the effect of varying each stimulation parameter. A) A cartoon summary illustrating the effects varying frequency along a continuous spectrum on individual cells within a network. B-D) Same as A but for B) i) pulse phase asymmetry, ii) pulse width, C) current amplitude, and D) duration of electrical stimulation. The illustration includes an electrode placed among a network of inhibitory neurons (triangles), excitatory neurons (circles), and astrocytes (stars). Gray indicates unresponsiveness to stimulation, red indicates an increase in activity, and blue indicates a decrease in activity.

Effect of frequency

Single-cell calcium and voltage imaging revealed that stimulation frequency shapes both the magnitude and temporal characteristics of the evoked neuronal responses, often in a spatial-, cell-type-, brain region-, and circuit-dependent manner. Calcium imaging studies demonstrated that prolonged continuous electrical stimulation for 30 s broadly recruits both local and distant neurons at onset regardless of frequency [32,42]. However, at higher frequencies (>90 Hz), distant neurons fail to sustain activity over prolonged trains, while local soma and neuropil near the electrode remain densely activated; in contrast, lower frequency stimulation (<75 Hz) sustains more spatially distributed activation throughout the train [32,42]. Two-photon imaging of cortical extracellular acetylcholine revealed that 600 pulses of higher-frequency stimulation (60–130 Hz) released more acetylcholine than low-frequency stimulation (20 Hz) of the basal forebrain, but there was no difference in acetylcholine release above 60 Hz [53]. Interestingly, astrocytes also show greatest activation and farthest calcium wave propagation during high-frequency stimulation (100–200 Hz) and little response to low frequencies (10–25 Hz), though the responsivity to high-frequency stimulation appears to saturate despite any further increase in the total electric energy delivered [37,30].

Growing evidence suggests that increasing frequency may progressively activate more fast-spiking interneurons and glia and fewer excitatory neurons, thus inhibiting the targeted region and disrupting the resting-state excitation-to-inhibition balance. For example, calcium and voltage imaging studies showed that local cortical excitatory neurons are better activated by lower frequencies at 50–100 Hz [46,37,40], whereas parvalbumin-positive fast-spiking interneurons maintain strong responses at ∼130 Hz [29,40]. Using cellular voltage imaging, Lowet et al. showed that the membrane potentials of hippocampal pyramidal cells followed individual electrical stimulation pulses at 40 Hz, but not 140 Hz [44], whereas fast-spiking neurons in the visual cortex were reliably paced by 140 Hz electrical stimulation [29]. Additionally, a study of the effects of electrical stimulation in the anterior cingulate cortex using fiber photometry calcium imaging showed that at lower frequencies (5–40 Hz) local neurons were more likely to be activated and at higher frequencies (80–130 Hz) more likely to be suppressed [52].

Effect of current amplitude

Single-cell imaging studies have also explored how current amplitude and local electric field strength influence neural dynamics. Computational modeling and calcium imaging studies reveal that somatic and neuropil calcium signals increase with stimulation amplitude and electric field strength [31,46]. Electrical stimulation delivered at above-threshold currents activates axonal processes near the electrode which propagate anti- and ortho-dromically to synaptically recruit neurons up to several hundred micrometers from the electrode [31,33]. Increases in current amplitude and higher electric field strengths increase the density of neurons being modulated within this volume [31,50,33]. Furthermore, glutamate imaging reveals that sustained release of the neurotransmitter is confined to ∼20 μm of a microelectrode tip where the current is the highest [33]. As the electric field dissipates rapidly from the stimulation site, the spatial effects of stimulation on individual neurons are greatly influenced by network effects as distance from the electrode increases. For example, Hughes et al. found inhibitory neurons reached peak activation faster than excitatory neurons during stimulation when closer to the electrode (<225 μm), possibly due to a combination of faster dynamics and direct axonal and indirect synaptic recruitment [51]. However, when measured farther from the electrode, the evoked responses in inhibitory neurons reached peak slower than excitatory neurons, suggesting that distal inhibitory neurons may participate in more inhibitory feedback loops than excitatory neurons at this distance [51].

Effect of pulse width and phase asymmetry

Pulse width and phase asymmetry predominantly shape the temporal changes of the electric field gradient, especially around the electrode tip, and thus have been hypothesized to influence local dynamics around the stimulation site by altering sodium channel kinetics [102]. Indeed, two-photon calcium imaging in cortical layer II/III—a region with extensive horizontal axonal projections—showed that compared to symmetric pulses, cathode-leading, cathode-elongated asymmetric pulses produce dense calcium activation in nearby neuropil and soma, consistent with strong activation of local axons and dendrites near the electrode [34]. Meanwhile, anode-leading, anode-elongated asymmetric pulses better activate the passing horizontal axonal fibers which led to more robust distal activation [34]. In addition to evoking different soma-neuropil calcium activity correlation patterns, two-photon cortical calcium imaging shows that varying waveform asymmetry and pulse width—holding frequency and charge per phase constant—differentially affects neural responses by steering which cells are directly/indirectly activated [42]. Finally, fiber photometry recordings showed that increasing phase asymmetry in an anode-leading, cathode-elongated waveform (180 μA for 100 μs, and -100 μA for 180 μs) during 130 Hz stimulation in the anterior cingulate cortex evoked more sustained neural suppression for up to 25 min [52].

Meanwhile, increasing pulse width, but not phase asymmetry, was essential for evoking sustained activation in neurons—more consistently in excitatory than inhibitory neurons—in the anterior cingulate cortex with 20 Hz stimulation [52]. It has also been reported that increasing pulse width durations during 130 Hz stimulation of the internal capsule led to stronger suppression in the frontal and somatosensory cortices [45]. Furthermore, while short pulses (≤100 μs) at 100 Hz mainly recruited neurons, increasing pulse width additionally recruited astrocytes, likely through neuronal mediated glia engagement [37]. However, it is worth noting, that increasing the pulse width during 130 Hz stimulation of the medial forebrain bundle did not increase the strength of extracellular dopamine release in the nucleus accumbens [43].

Effect of pulse train duration

In addition to the effect of varying parameters of individual pulses, stimulation pulse train duration influences both the magnitude and temporal evolution of neural responses, with brief 1-s trains producing distinct transient versus sustained effects [29,44,55,47]. Voltage imaging studies in the hippocampus and cortex revealed that neurons exhibit a transient response to 40 and 140 Hz electrical stimulation in the first 100 ms followed by a sustained response over the rest of the 1 s stimulation train [29,44]. Furthermore, membrane voltage of CA1 pyramidal neurons demonstrated a gradual adaptation/desensitization to the 40 Hz stimulation over the course of prolonged 1 min electrical stimulation [55]. Meanwhile, cortical calcium imaging studies revealed that increasing 100 Hz stimulation duration from 2 s to 20–40 s did not change response amplitude, but did prolong neural activation, while astrocytic calcium increases remained transient but more astrocytes were recruited [37]. Similarly, short 1 s electrical stimulation trains at 150 Hz produced mainly onset-locked responses [47], whereas longer stimulation (>5 s) at the same frequency (140–150 Hz) led to pronounced suppression of calcium activity [50,47]. Thus, electrical stimulation seems to consistently drive a transient response within milliseconds to seconds of stimulation onset, which evolves into a somewhat “steady-state” during sustained stimulation lasting tens of seconds or longer, suggesting that prolonged stimulation—as often used in the clinic—likely engages diverse indirect network adaptation mechanisms over time.

In a similar manner, while both 5 s and 20 s bouts of electrical stimulation to the medial forebrain bundle induced transient increases in dopamine release in the nucleus accumbens peaking around 1 s, longer-duration stimulation resulted in lower post-stimulation dopamine levels than the shorter-duration stimulation, suggesting more suppression of dopaminergic activity [43]. Thus, brief bursts of stimulation, rather than continuous, uniform electrical stimulation may be better at increasing chronic levels of neuromodulators. Indeed, Shanazz et al. demonstrated that, in the basal forebrain, 60 Hz bouts of stimulation delivered for 20 s did not increase peak acetylcholine release beyond what was observed by 10–12 s [53].

Effect of pulse train patterns

In addition to uniform pulse frequencies, burst and non-uniform stimulation patterns have been explored to evoke more specific network effects. In the cortex, bursts of electrical stimulation evoke time-locked calcium and glutamate signals at the burst frequency, and differentially recruit excitation and inhibition depending on the state of the network [48,38,51]. Non-uniformly patterned stimulation paradigms have also been developed to better leverage the intrinsic biophysical properties of neurons for improved therapeutic benefit or more precise stimulation often with less charge delivered [42,34]. Finally, dynamically modulating electrical stimulation amplitude and/or frequency within a pulse train has been show to evoke faster onset and offset calcium responses in cortical neurons, and to reduce stimulation induced depression of neural excitability during repeated brief bouts of stimulation [47].

Experimental evidence for specific neurophysiological mechanistic hypotheses

Electrical stimulation operates through a combination of direct effects on soma, axons, dendrites, and synapses, as well as indirect circuit effects through synaptic integration [10,13,14]. Preclinical cellular imaging has been used to test various specific mechanistic hypotheses (see Table 3 and Fig. 1).

Axonal activation

A major mode of neural recruitment by electrical stimulation is believed to be direct axonal activation, since axons are the most electrically sensitive component of neurons. This theory is supported by results from Histed et al. which demonstrated robust, albeit sparse, activation of neurons hundreds of microns from the electrode and dense activation of neuropil around the electrode as current increased [31], consistent with the activation of passing axonal fibers near the electrode. Furthermore, neurons closer to the electrode—more likely to be driven by direct axonal recruitment—maintained stable sustained activity throughout stimulation, while neurons farther from the electrode—more likely to experience a combination of indirect recruitment via synaptic integration—displayed more diverse responses [32,51]. Some of these farther neurons only respond transiently at stimulation onset, particularly at higher frequencies, but theta-burst stimulation better maintains sustained activity in these distal neurons [51].

Similarly, subthalamic nucleus (STN) stimulation in anesthetized mice led to onset-locked calcium activity in the otherwise quiet striatum, consistent with the recruitment via axonal activation of projecting fibers passing through or near the STN [35]. However, internal capsule stimulation in anesthetized mice did not evoke any sustained activity in the downstream cortical networks, arguing that network activity is essential for stimulation-induced sustained effects [45]. By increasing the anodic phase of an anode-leading pulse waveform compared to the cathodic phase, axons can be even more selectively activated compared to soma and promote the spread of electrical stimulation effects farther from the electrode [34].

Suppression

Originally, electrical stimulation was believed to inhibit local neural activity, as its clinical benefits resemble a tissue lesion [103,104]. Indeed, electrical stimulation can induce axonal depolarization blockade, leading to spiking failure [103,104]. Furthermore, a study using fiber photometry in the anterior cingulate cortex of chronic stress model mice demonstrated that parameters could be specifically tuned to reliably evoke excitation or inhibition in the local circuit, but depressive behavior in mice was ameliorated only when electrical stimulation evoked local inhibition [52]. Similarly, calcium imaging in the ventromedial hypothalamus showed that more effective alleviation of seizures corresponded to greater stimulation-induced suppression of overactive glutamatergic neurons [54].

Additionally, calcium imaging studies have noted that high-frequency stimulation (>100 Hz) induced synaptic depression and/or depression of neural excitability and left the network less responsive to subsequent inputs [31,42,33,47]. Synaptic depression could be a result of synaptic vesicle or neurotransmitter depletion due to electrical stimulation orthodromically driving synaptic activity at high frequencies [14,105]. And, stimulation-induced depression of neural excitability is related to intrinsic changes in the excitability of the neuronal membrane after a period of stimulation driven hyperactivity [97]. Calcium imaging in various brain regions and at a wide range of stimulation parameters has further demonstrated a combination of activation and suppression at the single-cell level, but an overall inhibitory effect on the local population level [50,51]. However, recent cellular voltage imaging studies indicated neurons can often be directly excited and sometimes entrained/paced by electrical stimulation at frequencies under 100 Hz, and that some fast-spiking interneurons can be paced as fast as 130 Hz [29,44,55]. These results argue against a local inhibition mechanism alone and instead suggest a combination of excitation, inhibition, and disinhibition which increase in complexity with synaptic integration recruiting indirect network effects.

Cellular informational lesion via disrupting processing of inputs

A more nuanced theory than direct inhibition of neural activity is that electrical stimulation disrupts pathological activity through a functional informational lesion via both orthodromic (downstream) and antidromic (upstream) pathways [10,15,16]. For example, during electrical stimulation, a neuron's ability to respond to synaptic inputs or to produce spiking outputs is impaired, thus causing a functional informational lesion [15,44,106]. Using optical voltage imaging, Lowet et al. directly tested this theory by examining membrane voltage changes to optogenetically-driven inputs, and found that 40 Hz and 140 Hz electrical stimulation disrupted neuronal responses to optogenetic inputs in the mouse CA1 [44]. Fabris et al. further demonstrated that electrical stimulation disrupted visual-stimulus-driven activity of parvalbumin-positive fast-spiking interneurons in the visual cortex [29].

Similarly, Kim et al. disrupted CA1 sharp-wave ripple associated calcium transients by applying closed-loop electrical stimulation to the ventral hippocampal commissure within 30 ms of each sharp-wave ripple onset [41]. Furthermore, stimulating the nucleus accumbens reduced hedonic feeding in mice by disrupting the nucleus accumbens D1-expressing medium spiny neurons’ activity pattern associated with high-fat food approach [36]. Stimulating in the nucleus accumbens also ameliorated depressive behaviors in chronic stress model mice by disinhibiting ventral tegmental area dopaminergic neurons via disruption of ventral tegmental area GABAergic interneuron activity [49]. Finally, recent studies in the cortex and hippocampus using voltage-imaging and simultaneous voltage-calcium measurements indicated electrical stimulation produces subthreshold depolarization [29,[44], [55], [57]] and drives disproportionately large calcium influx during prolonged depolarizations compared with isolated neural spikes [57]. Together, these findings highlight that electrical stimulation influences spike timing and modulates the coupling between membrane voltage and intracellular calcium dynamics, resulting in altered cellular processing of neuronal inputs and a functional informational lesion.

Network informational lesion via desynchronization

Consistent with the functional informational lesion theory, electrical stimulation is thought to also disrupt pathological synchrony via network desynchronization. In Parkinson's disease, exaggerated increases in beta-band neural dynamics are closely linked to motor symptoms in the clinic, particularly akinesia [10,107]. It has been broadly documented that electrical stimulation in the STN or globus pallidus internus disrupts this abnormal beta-band synchronization and ameliorates motor deficits in patients [17,18]. Fiber photometry calcium imaging performed in hemi-Parkinsonian mice confirmed that rather than a simple suppression of the STN or basal ganglia, 100 Hz electrical stimulation disrupts pathological neural activity patterns in the STN [39].

Similarly, obsessive compulsive disorder (OCD) is thought to involve a pathological increase in neural synchrony and coherence in the cortico-striatal-thalamo-cortical circuit, which can be disrupted by electrical stimulation [108,109]. Calcium imaging in OCD model mice revealed electrical stimulation of the internal capsule reduced OCD-like grooming in OCD mouse models by modulating neural activity in neurons that were consistently active during grooming at baseline [45]. Moreover, in epilepsy, it is broadly recognized that seizures are driven by pathological synchrony of neural firing, which is once again disrupted by electrical stimulation to reduce severity and frequency of seizures [74,[66], [110], [111],70].

Plasticity and network reorganization

Chronic electrical stimulation is also hypothesized to engage plasticity and network reorganization due to its persistent effects beyond stimulation offset [3,[19], [20], [21]]. DBS-induced plasticity changes take time to manifest, which may underlie the differential time courses for the therapeutic effects to emerge on different disease symptoms [112]. Indeed, it has been shown that 100 Hz electrical stimulation of the nucleus accumbens in chronic unpredictable mild stress model mice reduced depressive behaviors in mice after 2 weeks, but not 1 week of repeated exposure [49]. Fiber photometry calcium imaging in the ventral tegmental area, downstream of the stimulated nucleus accumbens, revealed that this delayed onset of therapeutic effects is related to synaptic changes between local GABAergic and dopaminergic neurons. This synaptic effect was further confirmed by knocking out the GABA_A receptor, which blocked all anti-depressive effects of electrical stimulation [49]. In another study, bursts of electrical stimulation lasting no longer than 1 s in dopamine-depleted mice differentially modulated two subpopulations of globus pallidus external neurons to produce therapeutic benefits lasting several hours after electrical stimulation was turned off [58].

Electrical stimulation has also been shown to evoke non-synaptic plasticity in the form of stimulation-induced depression of neuronal excitability, which can last for days post-stimulation [97], but can be mitigated through careful design of dynamic, nonuniform stimulation waveforms [47]. Finally, a recent voltage imaging study by Wang and Ravasio et al. revealed that CA1 neurons experience prominent adaptation during electrical stimulation and become less sensitive to electrical stimulation over time [55]. Specifically, subthreshold membrane voltage, but not spiking, was robustly entrained by 40 Hz electrical stimulation, and this voltage membrane entrainment decreased progressively over 1 min of electrical stimulation, and continued to decrease across five repeated trials despite modest inter-trial recovery [55]. Such cellular adaptation to prolonged electrical stimulation may be relevant to changes in symptoms observed on the minutes scale, and perhaps even longer, in the clinic [[113], [114], [115], [116]].

Glial recruitment

Glia are responsive to electrode implantation, forming a glial scar around the electrode [[117], [118], [119]]. While glial scaring often leads to the clinical need to increase stimulation strength, there is evidence that glia play an active role in modulating neuronal excitability and facilitating neuroprotection, though they may also contribute to further neurodegeneration [3,[22], [23], [24], [25], [26],117,118]. Since glial dysfunction has been linked to neurodegenerative disease progression, it would be invaluable to examine whether glial dynamics can be intentionally altered using electrical stimulation to mitigate neurodegeneration and promote neuroprotective effects [117,[120], [121], [122]].

Bekar et al. tested cortical electrical stimulation at 25–200 Hz and 15–50 μA in a Parkinsonian mouse model, and found that higher frequencies and amplitudes (≥125 Hz and 50 μA) evoked calcium waves that propagated farther, indicating astrocytic regulation of extracellular adenosine is essential for reducing tremors [30]. However, further increasing the stimulation frequency to 200 Hz did not alter evoked astrocytic calcium waves, suggesting a frequency ceiling effect in astrocytes [30]. Ma et al. observed a similar saturation effect during cortical astrocytic calcium imaging, where further increases in the strength of stimulation did not significantly increase intracellular calcium response [37]. They also noted a similar preference for higher frequencies (100–150 Hz) and noted very little astrocytic response to low frequencies (∼10 Hz) [37]. Nonetheless, an optogenetic stimulation study in rats demonstrated astrocytic stimulation alone is not sufficient to ameliorate Parkinsonian motor deficits [123].

Preclinical disease models support disease-specific electrical stimulation mechanistic studies

As briefly mentioned before, various disease models have been used to probe the effect of electrical stimulation in specific pathological contexts; particularly in order to unravel the clinically therapeutic mechanisms of electrical stimulation. For example, Schor et al. investigated the mechanisms of 130 Hz STN stimulation in 6-hydroxydopamine neurotoxin lesioned hemi-Parkinsonian mice using fiber photometry calcium imaging and showed that stimulation disrupted STN neural activity patterns related to movement by activating the hyperdirect primary motor cortex to STN pathways rather than inhibiting the STN or basal ganglia output [39]. Likewise, using fiber photometry calcium imaging, Song et al. found that 100 Hz stimulation of the nucleus accumbens ameliorates depressive behaviors in chronic unpredictable mild stress mouse models, by inhibiting ventral tegmental area interneurons which reverse the stress induced suppression of ventral tegmental area dopaminergic neurons [49]. The observed neuronal disinhibition and anti-depressant behavioral effects were abolished upon GABA_A receptor knockdown, suggesting an important role of local interneurons in mediating stimulation efficacy [49]. Furthermore, van den Boom et al. performed calcium imaging of neocortical neurons during high-frequency electrical stimulation of the internal capsule while tracking grooming rates in an obsessive compulsive disorder transgenic mouse model [45]. They showed electrical stimulation mediated dose-dependent excitation and inhibition of corticostriatal neurons in the medial orbitofrontal cortex, which scaled with grooming suppression without disrupting basal neuronal activity rate or neural network dynamics [45].

In addition to investigating therapeutic mechanisms, preclinical optical imaging in disease models enables specific stimulation parameter exploration to facilitate clinical translation. For example, brief bursts of electrical stimulation in the nucleus accumbens—delivered in response to increased delta band activity—more effectively suppressed hedonic feeding than uniform, continuous electrical stimulation, possibly by producing repeated increases calcium activity in D1-expressing medium spiny neuron [36]. Similarly, Yuan et al. noted that when stimulating the anterior cingulate cortex in chronic restraint stressed mice, higher frequencies combined with more asymmetric pulses robustly suppressed both excitatory and inhibitory neurons in the anterior cingulate cortex and attenuated depressive behaviors [52]. In contrast, lower frequencies combined with wider, symmetric pulses consistently evoked more excitation in excitatory neurons, sometimes excited GABAergic interneurons, and exacerbated depressive behaviors [52]. Finally, to explore the efficacy of basal forebrain electrical stimulation in driving release of acetylcholine—a potential target for treating dementia—Shanazz et al. imaged extracellular acetylcholine release in the somatosensory cortex while changing stimulation frequency and duration [53]. They found that increasing frequency above 60 Hz did not significantly increase the amount of acetylcholine released, but that increasing the duration of stimulation from 5 to 15 s did [53].

Beyond stimulation parameter testing, efforts have also been made to probe disease-specific stimulation targets preclinically. Zou et al. tested the effectiveness of the ventromedial hypothalamus as a potential target in several seizure or epilepsy mouse models, including the pentylenetetrazol, maximal electric shock, kainic acid, and hippocampal-kindling models [54]. Interestingly, they found that 1 Hz stimulation more effectively suppressed hyperactivity of the local glutamatergic neurons and alleviated seizure severity across these models than higher frequency stimulation at 30 and 100 Hz [54], highlighting the potential involvement of stimulation-evoked acivation of local GABAergic neurons in mediating widespread suppression [124].

Conclusions and future directions

The exciting progress in cellular activity optical imaging has enabled precise analysis of the effects of electrical stimulation on individual neurons and population networks in the brains of awake, behaving rodents, free of electrical interference. These preclinical studies can rapidly and systematically explore the stimulation parameter space, assess therapeutic efficacy in disease models, and provide mechanistic insight into the engaged cellular and network effects. While extensive efforts have been directed to examining the effect of various aspects of electrical stimulation, a systematic exploration of stimulation frequencies, intensities, pulse patterns, and pulse train durations in healthy and disease models would provide a more comprehensive understanding of the impact of these parameters in the context of specific pathological conditions.

For example, when applied in epilepsy mouse models, cellular imaging techniques allow for systematic investigation of the stimulation parameter space, and its effects on animal behavior and neurophysiological mechanisms. Epileptic activity is thought to involve pathological dysregulation of the excitatory and inhibitory circuits [[125], [126], [127], [128], [129], [130], [131], [132]]. Clinically, a wide range of stimulation frequencies (1–333 Hz) have been shown to reduce seizure risk [[66], [110], [111],[70], [133], [134]]. It would be interesting to investigate preclinically whether different stimulation frequencies selectively restore excitation-to-inhibition equilibrium in epileptic mice. It is plausible that brief bursts (≤5 s) of electrical stimulation, as used in RNS, could disrupt network input processing, with higher frequency stimulation progressively suppressing more excitatory neurons and augmenting inhibitory neurons [29,44,46,37,40,52]. Finally, since electrical stimulation also induces seizures, as in the kindling model [135], it would be imperative to test the safe operating range of these parameters and determine the underlying mechanisms that support the transition from a therapeutic regime to a pathological one. Such understanding could provide principled understanding which would facilitate clinical parameter selection.

With continued improvement of the spatial and temporal resolution of optical imaging techniques, it is becoming feasible to probe the effect of electrical stimulation on different compartments of a neuron (e.g. axon, soma, dendrite and synapses etc.) and other non-neuronal cells, like astrocytes, microglia, and oligodendrocytes, to gain an even more detailed understanding of the cellular and network mechanisms. While optical imaging techniques provide an excellent opportunity for artifact-free analysis of the real-time effects of electrical stimulation, they require transgene expression, making them difficult to deploy in clinical studies. Thus, it is important to design animal experiments that integrate with clinical observations to guide the development of a more principled approach in advancing clinical neuromodulation. In addition to probing the cellular and network mechanisms of electrical stimulation, studies in pathological models would further clarify disease-specific effects to ultimately improve both patient safety and therapeutic stimulation parameter selection.

Author contribution

C.R. and X.H. wrote the manuscript. All authors edited the manuscript.

Declaration of competing interest

The authors declare no competing interests.