Microelectrode locations, unit counts, and experimental design.

(A) Behnke Fried-style macro/micro depth electrode (left) and microelectrode bundle locations projected in MNI space (right). (B) The proportion of units recorded from each brain area (left) and the proportion of units that met the criteria for inclusion in analyses (average pre-trial baseline firing rate ≥ 0.1 Hz) (right). (C) Counts of total (grey) and included (colored) units within each region. (D) Intracranial recording and stimulation took place in the context of a two-phase (encoding, retrieval) visual recognition memory task. A series of neutral valence images were shown (3 s), half of which were followed by direct electrical stimulation (1 s). Retrieval memory was tested during a self-paced task ∼24 hours later. (E) Simulated theta-burst stimulation trace (left) and individual stimulation pulse (right); charge-balanced, bipolar, biphasic rectangular pulses were delivered over a 1 s period. HIP = hippocampus (coral), OFC = orbitofrontal cortex (yellow), AMY = amygdala (blue), ACC = anterior cingulate cortex (purple).

Example raster plots depicting heterogeneous responses to stimulation.

(A) Representative example of modulation during stimulation. The highpass-filtered, trial-averaged LFP from the corresponding microwire is shown (top) above the spike raster for an example unit located in the hippocampus (middle); the grey shaded region depicts the duration of stimulation with onset at t = 0. The average firing rate across trials was estimated by convolving the binned spike counts (100 ms bins) with a Gaussian kernel (bottom). (B) The difference in the number of spikes in the 1 s peri-stimulation epochs for each trial is shown (top). We subsequently performed a Wilcoxon signed-rank test on the during- and post-stimulation spike counts for each trial vs. the pre-trial baseline and compared the empirical test statistic against a null distribution generated by shuffling the epoch labels 1,000 times (bottom); the grey-shaded region represents the distribution containing 95% of observed values. (C) Some units (left, left-middle) exhibited increased firing rates, whereas others (right-middle, right) had their firing suppressed. The temporal dynamics of the firing rate modulation (e.g., onset, duration) were highly variable across units. The averaged waveform for each of the visualized units is shown below its corresponding peri-stimulation raster plot (WFs = waveforms); the shaded region represents standard deviation across waveforms.

Characterization of modulation in neuronal firing rate.

(A) Percent of modulated units observed across trials separated by stim (purple) vs. no-stim conditions (orange). (B) Percent of modulated units as a function of recording region. (C) Comparison of baseline firing rate in units separated by condition (stim vs. no-stim) and outcome (NS = not significant, Mod = modulated). (D) Venn diagram depicting the shared and independent proportions of units modulated by image onset (Image) and the two experimental conditions (stim vs. no-stim). (E) Scatterplot of pre-stimulation firing rate relative to the firing rate during the two contrast windows (during, post) for the stim (left) and no-stim (right) conditions. Modulated units are highlighted in purple (stim) or orange (no-stim), whereas units without a significant change are shown in grey. (F) Temporal dynamics of pseudo-population coactivity within each condition, represented by the first three principal components of the trial-averaged firing rates. The grey-shaded region depicts the duration of stimulation with onset at t = 0. Images were presented on screen for 3 s, with onset at t = -3. * p < 0.05, *** p < 0.001, NS = not significant.

Separation of presumed excitatory and inhibitory neurons by waveform morphology

(A) Two metrics were calculated using the averaged waveforms for each detected unit: the valley-to-peak width (VP) and peak half-width (PHW). (B) Scatterplot of the relationship between VP and PHW; note that units with identical metrics are overlaid. Using k-means clustering, we identified two distinct response clusters, representing presumed excitatory (E, blue) and inhibitory (I, red) neurons. The units from which the example waveforms were taken are outlined in black. Probability distributions for each metric are shown along the axes. (C) Total number of units within each cluster, separated by region. (D) Comparison of baseline firing rates, separated by cluster. (E) Percent of modulated units in each cluster. * p < 0.05, NS = not significant.

Summary of behavioral performance during memory task.

(A) Memory performance for each session is quantified using d’ (left); grey lines connect d’ scores across conditions for an individual session. Boxplot of the observed difference in d’ scores across conditions (right). (B) Hit rate (percent of old images correctly recognized) and false alarm rate (percent of new images incorrectly labeled as old) across conditions. (C) Change in recognition memory performance was split into two categories using a d’ difference threshold of ± 0.5: responder (positive or negative; Δd’, pink) and non-responder (∼d’, grey). Individual d’ scores are shown (left) with points colored by outcome category; dotted lines demarcate category boundaries, and the grey-shaded region represents negligible change. The number of sessions within each outcome category (middle) and the proportion of modulated units as a function of outcome category, separated by region (right). NS = not significant.