All-optical voltage interrogation for probing synaptic plasticity in vivo

The study addresses how to directly measure synaptic efficacy and plasticity rules in identified connections in vivo, a key requirement for understanding learning and memory. Traditional intracellular recordings are brief, laborious and low-yield, leading many to rely on two-photon calcium imaging as a proxy for voltage dynamics. Recent genetically encoded voltage indicators (GEVIs) suitable for two-photon microscopy enable direct voltage measurements in genetically defined neurons. Here, the authors present an all-optical strategy combining an optimized GEVI (JEDI-2Psub), two-photon imaging, optogenetic activation of granule cell inputs, and sensory stimulation of climbing fiber inputs in awake mice. This approach aims to resolve both subthreshold and suprathreshold signals across Purkinje cell dendrites, probe correlations within and between neurons, and define timing-dependent rules for plasticity induction in vivo.

The paper situates its contribution within the need for direct voltage measurements to understand synaptic information processing and storage. Prior methods include intracellular recordings (sharp microelectrodes, patch clamp) which are technically challenging and short-lived in vivo, and two-photon calcium imaging used as a proxy for spikes and synaptic activation. Advances in GEVIs and two-photon-compatible sensors have enabled population voltage imaging. In the cerebellum, microzone architecture and synchronized complex spikes have been documented using electrophysiology and calcium imaging. Feed-forward inhibition from granule cell activation via interneurons shapes Purkinje cell output, and classical plasticity paradigms pair parallel fiber and climbing fiber activation to induce synaptic changes. The study builds on these foundations by introducing an optimized GEVI (JEDI-2Psub) with enhanced subthreshold sensitivity and a protocol to measure inhibitory synaptic plasticity all-optically in awake mice.

- GEVI optimization: JEDI-2P was modified by inserting a tryptophan immediately after cpGFP to yield JEDI-2Psub, increasing single-spike fluorescence responses and shifting sensitivity toward negative potentials for enhanced subthreshold reporting. In vitro validation in HEK293A cells included voltage-clamp protocols (AP waveforms, spike trains, subthreshold depolarizations, voltage steps from -120 to +50 mV) under two-photon (940 nm; 440 Hz imaging; ~61 mW) and one-photon excitation (475 nm). Brightness and photostability were assessed via a high-throughput two-photon screening platform, normalizing fluorescence to cyOFP1 and calculating photostability from the area under the fluorescence curve. - Animals and surgical procedures: Male and female Math1-Cre mice (3–5 months) were implanted with a headplate and cranial window over cerebellar vermis (lobules V–VI). AAVs were injected to co-express JEDI-2Psub under CaMKII promoter (selective PC expression; AAV2/1; 5 × 10^11 VG/mL) and Cre-dependent ChRmine-mScarlet in granule cells (AAV 9/2; 3 × 10^12 VG/mL). Four 100 nL injections were made at depths of 200 µm and 400 µm. A coverslip sealed the craniotomy; for pharmacology, a custom coverslip with a small access port was used. Mice recovered and expression was confirmed. - In vivo imaging and stimulation: Two-photon resonant-scanning microscope (940 nm excitation for JEDI-2Psub) with a widefield 590 nm LED for optogenetic stimulation was used. Fields of view were 208 × 24 µm at 0.8 µm/pixel (440 Hz frame rate). Detector gating around LED pulses (±1 ms) produced a 5–8 ms deadtime between LED onset and recording. Typical imaging power was ~70 mW; optogenetic LED intensity ~8 mW/mm^2. Sensory stimulation was via whisker-pad airpuffs (25 ms) to drive CF inputs. - Data processing: Images were motion-registered (Suite2p). Putative PC dendrites were masked, intensity-normalized and segmented into ~4.8–5 µm segments via polynomial fitting. Cross-correlation matrices clustered segments belonging to the same dendrite. Fluorescence was converted to −ΔF/F0; baseline was extracted via low-pass filtering (Butterworth 3–5 Hz) after removing LED-gating artifacts. Spikes were detected when signal exceeded 3 SD over baseline; spike times were determined by maxima within 25 ms windows. Triggered averages were upsampled using DAQ timing and Gaussian fits to spike windows. - Gabazine control: Topical gabazine (200 µM) was applied via the access port; optogenetic LED stimuli (2 ms, 0.5 Hz × 45) were repeated across matched fields of view. IPSP amplitude was measured trial-by-trial as the minimum baseline signal 20–200 ms post-stimulation. - Plasticity protocol: Pre-pairing characterization used optogenetic GrC activation (2 ms LED, 0.5 Hz × 60). Plasticity induction paired CF activation via a 25 ms airpuff followed after 150 ms by 2 ms optogenetic GrC activation, repeated 300 times at 1 Hz. Post-pairing optogenetic characterization was performed 40 min later. Cells were matched pre/post via morphology, segment-wise correlations, complex spike shape and firing rate; controls omitted pairing or reversed stimulus order. Features extracted included IPSP amplitude (minimum baseline 20–200 ms post LED) and spontaneous complex spike amplitude. Statistics included Wilcoxon signed-rank tests, Mann–Whitney U-tests, linear regression/Wald tests, and linear mixed-effects models (animal and FOV as random effects).

- GEVI enhancement: JEDI-2Psub produced larger single-spike responses than JEDI-2P (−ΔF/F0 = −34.1 ± 6.8% vs −23.4 ± 3.5%, mean ± 95% CI; n = 6 vs 7 HEK293A cells), improved photostability, and shifted voltage sensitivity toward more negative potentials, yielding ~3.5× larger responses around resting membrane potential. Linear regression of subthreshold responses: JEDI-2P slope ~0.43; JEDI-2Psub slope ~1.5; R² ~0.91–0.92. - In vivo PC dendritic signals: Spontaneous complex spikes were prominent (−ΔF/F0 = −31.2 ± 4.4%; d′ = 5.9 ± 1.3; FWHM = 10.5 ± 1.8 ms; rate 1.3 ± 0.4 Hz; n = 43 cells, 4 mice) and scaled with baseline depolarization. Sensory airpuffs evoked diverse responses across PCs, including inhibitory and excitatory signals; excitatory responses had time-to-peak 65.1 ± 13.1 ms (n = 43 cells, 4 mice) and matched spontaneous complex spike waveforms. - Optogenetic GrC activation: Typically evoked graded hyperpolarizing IPSPs in PCs, with onset latency 9.6 ± 4.8 ms to 10% of peak and time-to-peak 85.4 ± 24.4 ms, scaling with LED intensity. Gabazine significantly reduced IPSP amplitude (control −ΔF/F0 = 4.1 ± 2.8% vs post-gabazine 1.6 ± 1.6%; Wilcoxon p = 1.26 × 10⁻⁴; n = 19 cells, 2 mice), confirming GABAergic feed-forward inhibition. Occasionally, parallel fiber EPSPs/depolarizations were observed (−ΔF/F0 = −10.5 ± 2.9%; time-to-peak 10.3 ± 0.6 ms; n = 7 cells), but were rare due to rapid feed-forward inhibition and 5–8 ms detector deadtime. - Microzone synchrony: Neighboring PCs showed correlated CF activity; sensory-evoked complex spike peak latency and probability correlated between neighbors (R = 0.490, p = 1.15 × 10⁻³; R = 0.590, p = 4.85 × 10⁻⁵; n = 49 cells in 20 FOVs, 4 mice). Complex spike amplitudes were not significantly correlated between neighbors (spontaneous: R = 0.286, p = 0.0695; sensory-evoked: R = 0.219, p = 0.170). - IPSP correlations: Mean IPSP amplitudes in neighboring PCs were highly correlated for both sensory-evoked and optogenetically evoked responses (R = 0.843, p = 4.55 × 10⁻¹²; R = 0.807, p = 1.90 × 10⁻¹⁰; N = 23 FOVs, 4 mice). IPSP peak latencies were also correlated (R = 0.500, p = 8.71 × 10⁻⁴; R = 0.474, p = 1.74 × 10⁻³). Cross-correlogram peaks for spontaneous complex spikes within microzones were narrow (mean width ~6.4 ms). - Spatial distribution: Across dendritic segments, complex spikes were uniformly distributed (CV sensory-evoked 0.079 ± 0.032; CV spontaneous 0.095 ± 0.043; n = 40 cells), whereas IPSPs were more heterogeneous (CV 0.130 ± 0.061), consistent with discrete inhibitory synapse activation. - Plasticity: A timing-dependent pairing protocol (CF activation via airpuff followed by GrC activation 150 ms later; 300 repetitions at 1 Hz) induced long-term potentiation (LTP) of IPSPs lasting ≥40 min (mean IPSP amplitude pre 7.9 ± 4.1% vs post 9.9 ± 4.6%; Wilcoxon p = 8.80 × 10⁻³; n = 32 cells, 4 mice). Controls (pairing omitted) showed no potentiation (LME p = 4.92 × 10⁻⁵), and reversing order of stimuli produced no significant effect. LTP magnitude was inversely correlated with baseline IPSP amplitude (R = −0.543, p = 1.32 × 10⁻³; control r = 0.349, p = 0.050). Neighboring PCs exhibited correlated LTP magnitudes (R = 0.611, p = 1.95 × 10⁻³; N = 9 FOVs, n = 24 cells, 3 mice), increasing IPSP correlations post-pairing (Mann–Whitney p = 4.40 × 10⁻⁴). LTP reduced dendritic IPSP CV (Mann–Whitney p = 3.94 × 10⁻²), while complex spike spatial distribution remained unchanged.

Combining optogenetic control with two-photon voltage imaging using JEDI-2Psub enabled direct interrogation of synaptic signals and plasticity across dendritic arbors and neighboring neurons in awake mice. Voltage imaging provided a cell-type-specific, subthreshold-resolving readout of synaptic efficacy without relying on calcium proxies, and allowed repeated measurements over extended times. The discovery of timing-dependent LTP at inhibitory synapses following conjunctive CF and GrC activation underscores the importance of inhibitory plasticity as a substrate for cerebellar learning. Such inhibitory LTP may counterbalance excitatory LTP and regulate CF-driven calcium signals, potentially modulating future plasticity. The approach is compatible with high-speed imaging developments and, with further refinement to resolve full dendritic trees and single spines during behavior, offers a path toward linking synapse-specific plasticity to learning in vivo.

This work introduces an all-optical framework for measuring synaptic strength and plasticity in identified cerebellar circuits by optimizing a GEVI (JEDI-2Psub) for subthreshold sensitivity and integrating two-photon imaging, optogenetics, and sensory stimulation. The method reveals microzone synchrony, spatially heterogeneous inhibition, and robust LTP of inhibitory IPSPs induced by timing-specific CF–GrC pairing, with coordinated plasticity across neighboring PCs and normalization of inhibitory responses across dendrites. Future directions include elucidating cellular mechanisms and timing rules underlying inhibitory plasticity, expanding monitoring to entire dendritic trees and single spines, integrating with high-speed microscopy, and applying the strategy during learning tasks to causally link synapse-level plasticity to behavior.

- Optical-electronic constraints: Detector gating around LED stimulation introduced a 5–8 ms deadtime, limiting detection of rapid optically evoked EPSPs. Imaging frame rates constrained somatic signal capture mainly to complex spikes. - Experimental design: Experiments were not randomized; investigators were not blinded during data collection (blinded to timing condition during analysis). PCs not responding to optogenetic activation were excluded. Sample sizes were modest but consistent with prior in vivo two-photon voltage imaging studies. - Generalizability and mechanism: Findings are in cerebellar Purkinje cells of awake mice; underlying cellular mechanisms of inhibitory LTP were not dissected. Potential photobleaching and opsin activation by imaging were assessed and found minimal, but remain considerations. - Spatial coverage: Segment-based analysis provided partial dendritic coverage; full-tree and spine-level resolution during behavior were not achieved in this study.