Layers of the monkey visual cortex are selectively modulated during electrical stimulation

Electrical stimulation is an emerging therapy for neurological and psychiatric conditions, yet how it modulates sensory-evoked local field potentials (LFPs) across cortical layers is not well understood. The neocortex consists of six layers with distinct connectivity: thalamic inputs predominantly target layer 4, propagate to layers 2/3, and then to layers 5/6, forming canonical feedforward and recurrent circuits. LFPs measured with laminar probes are widely used to study layer-specific microcircuits, particularly in primary visual cortex (V1), which has well-characterized laminar structure and connectivity. Prior work shows deeper layers often exhibit stronger evoked LFPs than superficial layers, potentially due to larger pyramidal neurons and higher firing rates in deep layers. This study asks whether transcranial alternating current stimulation (tACS) modulates visually evoked LFPs in a layer-specific and phase-dependent manner in nonhuman primates, and what biophysical mechanisms underlie any observed laminar selectivity. By combining laminar V1 recordings during visual flashes with concurrent low-frequency tACS and computational modeling of a cortical column, the authors aim to bridge single-neuron stimulation effects with population-level laminar dynamics relevant to human EEG.

• Layered neocortical organization supports distinct roles for layers in sensory processing, with feedforward thalamic input to layer 4, propagation to superficial layers 2/3, and outputs/recurrent dynamics involving layers 5/6.
• LFPs reflect transmembrane synaptic currents from populations; deeper layers often show stronger evoked LFPs, attributed to larger pyramidal cells and higher firing rates in layers 5/6.
• Nonhuman primates provide a close model to humans for studying biophysics and physiology of electrical stimulation across cortical layers.
• tACS can modulate neuronal activity in a phase-dependent manner at single-neuron and network levels, but laminar specificity of such modulation in evoked LFPs has been unclear.
• Modeling studies suggest layer-dependent firing patterns and LFP generation in V1, with deeper layers contributing strongly to LFPs. Prior experimental work indicates that electric field strength can vary across layers due to tissue properties, but physiological responsiveness may not follow field amplitude alone.

Subjects: Two female capuchin monkeys (Cebus apella), 15 years old, 1.5–3 kg. Approved by IACUC (AP2014-510). Under general anesthesia, each was implanted with a headpost and an occipital chamber for perpendicular penetrations into V1.

Visual stimulation: High-intensity strobe flashes at 2.3 Hz (every 435 ms) delivered while monkeys were lightly anesthetized with 1–1.5% isoflurane. Two sessions per animal: Flash only, then Flash + AC (with rest between).

Laminar recordings: 23-channel linear array (U-Probe, Plexon), 0.1 mm inter-contact spacing, impedance 300–750 kΩ. Referenced to dura. Signals amplified 10x, analog filtered 0.1–500 Hz for LFP, digitized and downsampled to 2 kHz (then to 1 kHz). Epochs from −50 to 250 ms relative to flash onset with baseline correction (−50 to 0 ms). P1 identified within 55–90 ms, N1 within 100–140 ms per trial. Trials exceeding mean ±5 SD rejected. Trials: Monkey 1 (Flash 320; Flash+AC 216); Monkey 2 (Flash 295; Flash+AC 389).

Electrical stimulation (tACS): Starstim (Neuroelectrics) Ag/AgCl electrodes (10 mm radius). One electrode over occipital scalp above probe, the other on right temporal area. Sine 1.5 Hz continuous, 30 s ramp-up/down. Current intensity peak-to-zero: 0.1 mA (Monkey 1), 0.2 mA (Monkey 2). Montage chosen to align fields parallel to probe (radial direction).

Artifact rejection: Applied ICA (SOBI) to identify 1.5 Hz sinusoidal components on low-pass and notch-filtered data pre-segmentation. Identified components via FFT; discrete Fourier transform filtering subtracted 1.5 Hz component; back-projected to contact level. Power spectra confirmed suppression of 1.5 Hz artifact in Flash+AC.

Layer assignment: Combined (i) CSD from Flash condition (second derivative of LFPs) to identify layer 4C sink/source patterns, (ii) electric field distribution to identify layer 1 onset (rise in field relative to extra-cortical space), and (iii) anatomical references. Contacts grouped: layer 1 (1–2 contacts), layers 2/3 (5–6), layer 4AB (2), layer 4C (4), layers 5/6 (4). White matter control obtained by lowering probe 1.3 mm to place distal 5 contacts in WM.

Electric field measurement: From raw data filtered 0.5–2 Hz, divided by gain (10) to obtain voltages; numerical gradient along contacts yielded radial electric field. Phase and amplitude at 1.5 Hz extracted with FFT. Table A (S1 Text) reported mean voltage/field per layer.

Phase dependency analysis: For Flash+AC, AC phase at flash onset labeled per trial at each contact; trials sorted into 20 phase bins (18°). Layer-averaged P1 and N1 amplitudes computed per bin. Mean direction and vector length quantified phase preference and strength. Monkey 1 showed bimodal circular distributions; axial mean methods (CircStat) were used. Control analysis: Virtual 1.5 Hz sinusoid used in Flash-only data; same analysis performed.

Statistics: Cluster-based permutation tests (2,000 permutations, alpha=0.01) compared Flash vs Flash+AC across contacts and time (0–250 ms), identifying significant clusters and their laminar extent. Phase dependency significance tested by permutation of phase labels (5,000 shuffles); vector length z-scored; one-tailed p-values determined; significant if original vector length >95% of surrogate values. MUA comparisons used paired t-tests for within-condition layer contrasts and unpaired two-sample t-tests for across-condition comparisons.

MUA analysis (Monkey 2): MUA band 300–5,000 Hz with zero-phase digital filter. Spike detection threshold = 3.5 × median(|x|/0.6745). Events above threshold classified as MUAs. Trials binned at 10 ms for PSTHs. Additional analysis contrasted trials aligned to AC peak vs trough (96 peak, 90 trough).

Computational modeling: Based on a biophysically detailed V1 cortical column model (Billeh et al., 2020) comprising 230,924 neurons (17 classes) with LGN inputs and background Poisson activity. Flash stimulus: 50 ms white between gray screens. LFP obtained from extracellular potentials low-pass filtered at 100 Hz, downsampled to 1 kHz, baseline corrected. AC integration: Quasi-static field with spatial component uniform at 16 mV/mm (consistent with experiments/simulations) and temporal 1.5 Hz sinusoid. tACS started at 250 ms; peak or trough aligned to flash onset at 500 ms to test phase effects. Recorded membrane currents (Imem) from passive basal dendrites of 500 randomly selected excitatory neurons per layer; subtracted currents from a tACS-only silent-network control to isolate LGN-driven effects. Firing rates computed at 2 ms bins.

• Layer-specific LFP modulation: LFP amplitudes were larger in deeper layers (4–6) than superficial (1–3). Under Flash+AC, LFPs after ~100 ms increased relative to Flash, especially in layers 4–6. Example normalized amplitudes (layers 5/6): Monkey 1 mean N1 0.59±0.1 (Flash+AC) vs 0.47±0.12 (Flash); P1 0.13±0.08 vs 0.12±0.09. Monkey 2 mean N1 0.38±0.12 vs 0.28±0.06; P1 0.10±0.09 vs 0.06±0.06.
• Statistical significance: Cluster-based permutation tests revealed significant differences between Flash and Flash+AC across contacts from layer 4AB through 5/6 in 100–250 ms window (Monkey 1 p=4.99×10⁻⁴; Monkey 2 p=9.99×10⁻⁴). No significant clusters in 0–100 ms. In Monkey 2, a significant difference also occurred in layers 2/3 (100–140 ms).
• Phase-dependent modulation: Sorting trials by AC phase (20 bins) showed phase preferences for P1 and N1 amplitudes only in deeper layers (4–6). Monkey 1 exhibited bimodal phase distributions; mean P1 directions across layers: −145° (L1), −135° (L2/3), −114° (L4AB), −118° (L4C), −113° (L5/6). N1 bimodal mean directions: −40° (L1), −61° (L2/3), −41° (L4AB), −37° (L4C), −40° (L5/6). Monkey 2 showed unimodal distributions with strong P1 directionality; P1 and N1 mean directions: L1 −159° and 100°; L2/3 −86° and 30°; L4AB −94° and 74°; L4C −59° and 107°; L5/6 −110° and 53°. Permutation tests confirmed significant phase locking in deeper layers only (p<0.05), with complementary phase preferences between P1 and N1.
• Control (virtual AC): No significant phase preferences for P1 or N1 in any layer in Flash-only data, ruling out spurious phase dependency.
• Electric field biophysics: Measured radial electric field peaked in layers 2/3 and decreased with depth. Monkey 1 mean fields (V/m): L1 0.62, L2/3 2.53, L4AB 1.18, L4C 0.62, L5/6 0.37. Monkey 2: L1 0.97, L2/3 2.76, L4AB 1.60, L4C 1.59, L5/6 1.42. Despite higher fields in layers 2/3, phase-dependent LFP modulation was absent there, indicating physiological layer properties outweigh field magnitude for LFP modulation.
• MUA (Monkey 2): Deeper layers had higher firing rates than superficial in both conditions: Flash 6.83±5.35 vs 5.32±2.22 spikes/s (p=2.03×10⁻⁶), Flash+AC 9.85±6.09 vs 5.03±2.03 spikes/s (p=6.09×10⁻⁴²). AC increased firing in layers 5/6 vs Flash (p=2.54×10⁻¹²); no significant AC effect in layers 2/3. Peak vs trough phase: no significant firing differences (layers 2/3 p=0.36; layers 5/6 p=0.51), though trough tended to be slightly higher.
• White matter control: No significant AC effects on LFPs or phase dependency in white matter.
• Modeling: The model reproduced higher deep-layer firing and LFPs. P1 (~50 ms) preceded spiking, suggesting subthreshold processes. Imem differences between AC peak and trough were largest during P1, especially in deeper layers, consistent with more positive LFP during peak and more negative during trough. This supports a mechanism where AC phase modulates membrane potential and electrochemical driving force at basal dendrites in deep-layer pyramidal neurons, leading to phase-dependent LFP components.

The study demonstrates that tACS selectively and phase-dependently modulates visually evoked LFPs in deeper cortical layers of primate V1, addressing the open question of laminar specificity of electrical stimulation effects on evoked population activity. Although the electric field amplitude was maximal in layers 2/3, modulation and phase dependency manifested predominantly in layers 4–6, implying that neurophysiological and anatomical properties—such as larger pyramidal neurons, higher firing rates, and stronger synchronization in deeper layers—govern sensitivity to stimulation more than field magnitude alone. The modeling results suggest a mechanistic account: the AC phase shifts membrane polarization in basal dendrites, altering the synaptic driving force and membrane currents that generate LFPs, thereby producing opposite phase preferences for P1 (positive deflection) and N1 (negative deflection) and stronger effects in deeper layers. The P1 preceding spiking links it to postsynaptic currents, whereas N1 correlates with spike-related activity, consistent with the observed complementarity between P1 and N1 phase preferences. Differences between animals (bimodal versus unimodal phase preference) likely reflect differences in deep-layer field strengths and the interplay with intrinsic low-frequency oscillations. The absence of effects in white matter emphasizes the specificity of gray-matter laminar microcircuits. Together, these findings connect single-neuron polarization by exogenous fields to layer-resolved population dynamics, informing strategies for targeted neuromodulation and interpretation of macroscopic signals like EEG that aggregate laminar LFPs.

This work shows that low-frequency transcranial AC selectively enhances and phase-modulates visually evoked LFPs in deeper layers of primate V1. Phase-dependent effects occur for both P1 and N1 components but are restricted to layers 4–6, despite higher field strengths in superficial layers, highlighting the dominant role of layer-specific physiology. A cortical column model explains these effects via phase-driven changes in membrane potential and synaptic driving force in deep-layer pyramidal neurons. These insights can guide more effective, layer-targeted neuromodulation protocols. Future directions: test a range of tACS frequencies and intensities to map dose–response and frequency specificity; extend experiments to awake conditions; incorporate spontaneous activity and variability into models to capture bimodality; develop and test strategies for selectively targeting deeper layers (e.g., ICMS, optimized multichannel TES, temporal interference); expand modeling to human cortical columns for translational applications; include additional recording modalities (single-unit) to further validate mechanisms.

• Animals were lightly anesthetized; effects in awake cortex may differ, though sensory-evoked LFPs are relatively preserved.
• Residual tACS artifacts are unlikely but cannot be entirely excluded; conclusions are supported by consistency across layers and null results in white matter.
• Computational model is based on a rodent V1 column and does not include spontaneous network dynamics, limiting its ability to capture bimodal phase preferences observed in one animal.
• Layer boundaries were defined via CSD, electric field changes, and anatomical references; alternative laminar markers might refine assignments.
• Phase-dependent modulation was clear in LFPs but not in MUA, likely reflecting differences in sensitivity of subthreshold versus suprathreshold activity.
• MUA analyses were conducted in one animal only.
• Only one stimulation frequency (1.5 Hz) and specific current intensities (0.1/0.2 mA) were used; generalizability across parameters requires further study.