Abstract
Layer 6 corticothalamic (L6CT) neurons project to both cortex and thalamus, inducing multiple effects including the modulation of cortical and thalamic firing, and the emergence of high gamma oscillations in the cortical local field potential (LFP). We hypothesize that the high gamma oscillations driven by L6CT neuron activation reflect the dynamic engagement of intracortical and cortico-thalamo-cortical circuits. To test this, we optogenetically activated L6CT neurons in NTSR1-cre mice (both male and female) expressing channelrhodopsin-2 in L6CT neurons. Leveraging the vibrissal pathway in awake, head-fixed mice, we presented ramp-and-hold light at different intensities while recording neural activity in the primary somatosensory barrel cortex (S1) and the ventral posteromedial nucleus (VPm) of the thalamus using silicon probes. We found that the activation of S1 L6CT neurons induces high-frequency LFP oscillations in S1 that are modulated in frequency, but not in amplitude, across light intensities and over time. To identify which neuronal classes contribute to these oscillations, we examined the temporal evolution of firing rate in S1 and VPm. While most S1 neurons were steadily suppressed, VPm and S1 layer 4 fast spiking (L4 FS) neurons evolved from being suppressed to facilitated within 500 ms, suggesting differential recruitment of the intracortical versus cortico-thalamo-cortical pathways. Finally, we found that LFP frequency selectively correlates with VPm firing rate. Taken together, our data suggests that L6CT neurons sculpt the frequency of S1 LFP oscillations through cortico-thalamo-cortical circuits, linking the recurrent interactions mediated by L6CT neurons to the high gamma oscillations observed across physiological and pathological conditions.
Significance Statement
Layer 6 corticothalamic (L6CT) neurons are strategically positioned to modulate the cortex and the thalamus allowing them to engage distinct, yet interlocked, circuits. Here we show that the activation of L6CT neurons in the mouse primary somatosensory cortex induces fast cortical oscillations through the coordinated engagement of cortico-thalamo-cortical and intracortical pathways. Our work reveals that these two L6CT-mediated pathways exert competing effects: while intracortical connections suppress cortical spiking, the activity of the cortico-thalamo-cortical loop rapidly evolves, facilitating cortical spiking. We demonstrate that the cortico-thalamo-cortical pathway operates on a faster timescale than the intracortical pathway and critically shapes cortical oscillation frequency. These findings reveal how the unique position of corticothalamic neurons allows them to flexibly and dynamically modulate the thalamocortical network.
Introduction
Layer 6 corticothalamic (L6CT) neurons are an excitatory neuronal class with projections ramified throughout the cortex and spreading to higher and lower order thalamic nuclei. Thanks to these projections (Kirchgessner et al., 2020), L6CT neurons are strategically positioned for modulating the thalamocortical network (Briggs and Usrey, 2008; Bortone et al., 2014; Kim et al., 2014; Kirchgessner et al., 2021) and conveying top-down modulatory feedback. This feedback has been suggested to be critical for shaping signaling in the pathway, as it can enhance or suppress incoming sensory inputs and, thus, perception (Briggs and Usrey, 2009; Olsen et al., 2012). It has been proposed that this modulatory effect can be mediated by changes in the activity and synchronization at the population level (Dimwamwa et al., 2024), manifested in local oscillations (Ray et al., 2008a) implicated in important aspects of behavior and function. Given their pivotal role in corticothalamic communication, it is particularly important to illuminate the impact of L6CT neurons on other neurons and population-level oscillations and how they contribute to electrophysiological and behavioral functions.
While multiple previous studies have investigated how L6CT neurons affect neuronal firing rates, relatively few have examined their effects at the local field potential (LFP) level. Optogenetic activation of L6CT neurons has been reported to induce cortical LFP oscillations whose frequency peaks between 60 and 250 Hz, falling in the range of activity referred to as high-gamma oscillations or the “ripples” band (Ray et al., 2008a; Nevalainen et al., 2020). Given the central role of L6CT neurons in communication between cortex and thalamus, their ability to provoke fast oscillations aligns with studies implicating high gamma oscillations in thalamocortical communication (Hu et al., 2023). Furthermore, the relationship between high gamma oscillations and thalamocortical connections provides an anatomical substrate for their critical role in attention, memory processing, coordination across areas, and cognitive tasks (Herrmann et al., 2004; Canolty et al., 2006; Ray et al., 2008c; Gaona et al., 2011; Kucewicz et al., 2017). Studies in the somatosensory pathway in vitro and in the auditory pathway in vivo (Crandall et al., 2015; Guo et al., 2017) found that the phase of these oscillations is locked to the spikes of cortical (including L6CT) and VPm neurons, thus suggesting a relationship between these oscillations and neuronal firing activity. This observation suggests that L6CT-induced oscillations may be mediated by both intracortical and cortico-thalamo-cortical circuits. Yet, the extent to which these two circuits contribute to the emergence of high-gamma oscillations is still elusive and broadens the long-standing question of whether L6CT effects are mainly mediated by their intracortical or cortico-thalamo-cortical projections (Sherman and Guillery, 1996).
Given the above, the ability of L6CT neurons to generate high gamma oscillations raises several questions: do L6CT neurons originate high gamma oscillations also in the somatosensory pathway in awake mice? How are these oscillations shaped and modulated? How do they relate to neuronal spiking? Here we hypothesize that L6CT neurons drive high gamma LFP oscillations in the primary somatosensory cortex (S1) of awake mice through the coordinated interaction of intracortical and thalamocortical pathways. We tested this hypothesis by manipulating the activity of L6CT neurons through targeted optogenetic stimulation and examining the effects on the cortical LFP and the spiking activity of cortical and thalamic neurons. Our results show that the optogenetic activation of L6CT neurons induces high gamma oscillations in the somatosensory pathway of awake mice; while the amplitude of these oscillations is stable, their frequency is determined by the degree of L6CT activation and varies over time. To identify the neuronal circuits regulating this activity, we examined the firing rates of cortical and thalamic units, finding a similar evolution over time in the activity of VPm and layer 4 fast spiking (L4 FS) neurons that was distinctly different from other neurons in the circuit. Finally, we found that the evolution of VPm firing rate over time covaries with the frequency of S1 high gamma oscillations, thus suggesting that VPm may play a critical role in sculpting the frequency of S1 LFP oscillations.
Materials and Methods
Data acquisition
In this paper we leverage the experiments used to study L6CT-induced single-unit activity in Dimwamwa et al. (2024) to investigate how L6CT activation shapes cortical LFP oscillations and to explore how these oscillations relate to cortical and thalamic spiking activity. This dataset consists of the optogenetic stimulation of S1 in 11 NTSR1-cre (B6.FVB(Cg)-Tg(Ntsr1-cre)GN220Gsat/Mmcd, MMRRC) adult mice aged between 6 weeks and 6 months (7 males, 5 females). In brief, mice were kept in a reversed light/dark cycle at 65–75°F and 40–60% humility. Mice underwent a first surgical procedure, thinning the skull above S1 (as identified by stereotactic coordinates), intrinsic imaging during whisker stimulation to map the location of the barrels associated with each whisker, were injected in the primary somatosensory cortex (S1) with 800 nl of cre-dependent AAV5-EF1a-DIO-hChR2(H134R)-eYFP.WPRE.hGH (Pauzin and Krieger, 2018; Kirchgessner et al., 2020; UPenn Vector Core), and were habituated to the experimental rig. After at least 3 weeks from the injection, mice were anesthetized (induction, isoflurane 5%; maintenance, isoflurane 0.5–2%) to open two burr holes above VPm (as identified by stereotactic coordinates) and S1 (as identified by intrinsic imaging).
The day of the experiment, mice were head-fixed in the experimental rig, and the electrodes were inserted in S1 (A1x32-5mm-25-177-A32, A1x32-Poly3-5mm-25s-177-A32, A1x64-Poly2-6mm-23s-160, or A1x64-Edge-6mm-20-177-A64, NeuroNexus), VPm (A1x32-Poly3-5mm-25s-177-A32, NeuroNexus), and TRN (Buzsaki32 or A1x32-Poly3-5mm-25s-177-A32, NeuroNexus). Electrophysiological data were acquired through a Cerebus acquisition system (Blackrock Neurotech) or TDT RZ2 Bioamp Processor (Tucker-Davis Technologies) with a sampling frequency of either 30,000 or 24,414.0625 Hz, respectively.
An optical fiber (400 nm diameter) was connected to a blue LED source (470 nm wavelength) and positioned to shine light above the craniotomy in S1. L6CT neurons were activated using ramps-and-hold light waveforms (ramp, 250 ms; hold, 500 ms; light intensities, 8, 16, 24, and 30 mW/mm2). Additionally, to determine sensory responsiveness, mice underwent whisker stimulation through a galvanometer deflected with a sawtooth waveform at a velocity of 300°/s in the caudo-rostral direction. Whisker stimuli were administered with 10 Hz frequency in blocks of 1 s (i.e., a sequence of 10 sawtooth waveforms). Videography was acquired under infrared illumination using CCD cameras triggered at 30 or 200 Hz. The LED source, the galvanometer, and the camera were controlled through custom scripts in MATLAB and Simulink Real-Time (MathWorks) at 1 kHz. For a detailed description of the experimental procedures, see Dimwamwa et al. (2024).
Histology
After the final recording, mice were transcardially perfused with PBS (137 mM NaCl, 7.2 mM KCl, and 10 mM PB, VWR) and 4% PFA. The skull was removed, and the brain was extracted, post-fixed overnight in a PFA solution, and sliced to 100 μm sections using a vibratome (VT100S, Leica Biosystems). Slices were incubated with DAPI (2 mM in PBS, AppliChem) for 15 min, mounted on slides with a 1,4-diazabicyclo[2.2.2]octane solution (Sigma), and imaged with a confocal microscope (Laser Scanning Microscope 900, Zeiss).
Videography processing
Videography data were analyzed using FaceMap (Syeda et al., 2022). The analysis was restricted to the region of the video with the mouse whiskers, and that region was used to compute the motion energy (Wright et al., 2021; Pala and Stanley, 2022). The time series of the motion energy was then smoothed over time, and all trials with motion energy above a fixed threshold, indicating whisker movements, were excluded (Urbain et al., 2019; Wright et al., 2021), and all subsequent analyses were performed on the remaining trials (Dimwamwa et al., 2024). Thus, the data analyzed in this study were during periods of time with no appreciable self-induced whisker motion.
Single-unit data preprocessing
All analyses were conducted in Matlab (MathWorks). Recorded data were high-pass filtered (500 Hz cutoff frequency; third-order Butterworth filter), median filtered across channels, sorted using Kilosort (Pachitariu et al., 2016), and manually curated using Phy2 (Rossant and Harris, 2013). We considered only units with signal-to-noise ratio (SNR) of the mean spike waveform >3 and <2% of the spikes violating a 2 ms refractory period, where SNR here is defined as the peak-to-peak amplitude of the spike divided by the amplitude of the background or baseline “noise.” S1, VPm, and TRN units used for analyses were required to be sensory responsive (Pala and Stanley, 2022; Dimwamwa et al., 2024). S1 neurons were classified as regular spiking or fast spiking based on the mean spike waveform (trough-to-peak time >0.5 ms or <0.4 ms, respectively; Ganmor et al., 2010), and each neuron was assigned to a layer relative to the center of layer 4 (as identified by current source density analysis; Sederberg et al., 2019; Sofroniew et al., n.d.). Based on the location of the center of layer 4 along the silicon probe, each channel was assigned to a layer based on the following thickness: layer 2/3, 300 μm; layer 4, 200 μm; layer 5, 300 μm; layer 6, 300 μm (Hooks et al., 2013; Sederberg et al., 2019; Pala and Stanley, 2022).
L6CT neurons were identified using the stimulus-associated spike-latency test (SALT; Kvitsiani et al., 2013) to determine whether a particular neuron's spiking was significantly increased by 5 ms light pulses compared with an equivalent duration pre-stimulus window. Using the 10 Hz train of sawtooth whisker stimuli of 1 s, both VPm and TRN neurons were required to be significantly responsive to both the first and the last whisker stimulus of the train, as defined by a two-sided Wilcoxon signed rank test. Additionally, to ensure that the recorded neurons were not from the posteromedial nucleus, both VPm and TRN neurons were further identified by a response latency <15 ms for the first whisker stimulus and <20 ms for the last whisker stimulus (Ahissar et al., 2001; Dimwamwa et al., 2024; Liew et al., 2024). Furthermore, VPm neurons were required to have a spike waveform width >0.3 ms (Guo et al., 2017).
Single-unit data processing
We computed peristimulus time histograms (PSTHs) from −750 to 750 ms around the onset of the ramp-and-hold light with bins of 1.5 ms. PSTHs were smoothed by applying a moving average filter (15 ms smoothing window, movmean function from Matlab).
The mean firing rate of each neuron during the early (250–500) and late (500–750) window was computed by averaging the mean firing rate across trials in the respective window.
To perform the dimensionality reduction analysis, the average PSTH for each cell type was cut from −100 to +750 ms around the light onset, and the average PSTH of the baseline was subtracted from the entire PSTH. Dimensionality reduction was performed using principal component analysis (PCA; pca function from Matlab, algorithm: “svd”) on the average PSTH traces. To explain 99% variance, we selected the first two components (see scree plot in Fig. 3C) and reported the weights and time course of each component.
LFP data processing
LFP data were downsampled to 1,000 Hz (low-pass filter 200 Hz), high-pass filtered (65 Hz cutoff frequency; third-order Butterworth filter), and segmented from −750 to 750 ms around the onset of ramp-and-hold light manipulations.
The power spectral density (PSD) was computed using the pwelch method (pwelch function from Matlab) on the time window from 0 to 750 ms and averaged across trials. The PSD of the unfiltered signals was computed with steps of 1 Hz from 1 to 250 Hz. The PSD of the filtered signal was computed with steps of 1 Hz from 65 to 200 Hz.
To obtain the time course of the absolute voltage for each channel, we rectified the voltage and averaged it across trials. To obtain the oscillation peak frequency of each channel, we decomposed the voltage of each channel in the time–frequency domains [newtimef function from eeglab (Delorme and Makeig, 2004); number of wavelet cycles, 5; Padratio, 8; max frequency, 250 Hz] and identified, for each time bin and channel, the frequency showing the maximum power.
We quantified the absolute amplitude and the average peak frequency of each channel in the early and late window by averaging the absolute amplitude and the oscillation frequency, respectively, for each window. For these quantifications, we computed the Hanning windows of the wavelet filters used for the time–frequency decomposition and the moving average, and we accounted for it by reducing the early and late windows to 245–500 ms and 500–745 ms for the absolute voltage and 292–500 ms and 500–708 ms for the oscillation frequency.
To ensure that the observed high gamma activity reflected oscillations rather than contamination of LFP with spiking activity, we analyzed the LFP activity time locked to spikes from each class of cortical neurons. For each neuron, we extracted the LFP from −100 to +100 ms around the time of each spike and averaged it across layers and spikes, thus obtaining a spike-triggered LFP average (STA). Additionally, we trimmed the split signal from −64 to +64 ms and performed the matching pursuit analysis as described in Ray et al. (Ray et al., 2008b). To remove noise, Gabor atoms with frequency peak at 60 Hz were excluded. Afterward, to ensure that no atom exhibited the typical features of a spike contamination (i.e., a sharp peak at the time of the spike), we selected and inspected the reconstructed signal and the energy profile of the first five Gabor atoms peaking closest to the time of the spike.
Statistical comparisons
The statistical analyses comparing the LFP amplitude and frequency across different light intensities were performed using a repeated-measures ANOVA test (Muhammad, 2023).
Statistical analyses comparing early and late window in units firing rate, LFP amplitude, and LFP frequency between were performed using Wilcoxon signed-rank tests (Woolson, 2005). To account for the high number of comparisons for each cell type and light intensity, the relative p values were corrected using the Benjamini–Yekutieli false discovery rate algorithm.
The correlations between single-unit PSTH and LFP amplitude and frequency were computed as follows: PSTHs were resampled to the same frequencies of LFP amplitude and frequency; signals were cut in the stable light window (250–750 ms) and correlated over time with the simultaneously recorded LFP amplitude and frequency (Spearman's correlation; de Winter et al., 2016).
Results
To investigate how L6CT neurons affect the LFP activity in S1, we activated L6CT neurons through targeted optogenetic manipulation while simultaneously recording extracellular electrophysiological activity across layers in S1 with silicon probes in awake mice. Specifically, we injected NTSR1-Cre mice with a recombinant adeno-associated virus (rAAV) encoding eYFP fused with Cre-dependent Channelrhodopsin2 (ChR2) in S1 (Fig. 1A). To drive spiking activity in L6CT neurons, we administered blue light through a 400 μm fiber conducting light from a 470 nm LED driven with a ramp-and-hold waveform (ramp, 250 ms; hold, 500 ms; hold at 16 mW/mm2) positioned on the skull above S1 (Fig. 1B). This optogenetic activation of L6CT neurons drove high-frequency activity in the LFP of S1, shown across cortical depths before and during LED stimulation (Fig. 1C). Qualitatively, this high-frequency activity started briefly after the light onset and persisted as long as the light was on, although the oscillation pattern varied over time. Specifically, the light onset drove fast activity that progressively slowed over time. We computed the power spectral density (PSD) of the cortical LFP signal induced in the 250–750 ms window during L6CT activation and in periods of time with no L6CT activation of equivalent duration, finding a striking difference in the power of high-frequency activity, consisting of a large narrow-band peak at ∼150 Hz induced by the L6CT activation (Fig. 1D, top). This effect was highly reproducible across recordings, but the corresponding peak frequency changed from 80 to 180 Hz depending on the subject and light intensity, as illustrated in the color map of spectral power over this frequency band in Figure 1D (bottom).
Optogenetic activation of L6CT neurons induces high gamma oscillations in the local field potential of S1. A, The left panel shows the schematic setup for the insertion of silicon probes in S1 and VPm during optogenetic activation of L6CT neurons. The right panel shows in blue the S1 histological section with DAPI staining and in green L6CT neurons marked with the eYFP fluorophore. B, The light was administered through a ramp-and-hold waveform to minimize transient effects of light. C, Representative traces recorded from all the silicon probe channels in S1 in one trial showing that the optogenetic activation of L6CT neurons induces high-gamma activity in the broadband local field potential (LFP). At the time of the onset of the stimulus, the LFP shows a slow deflection superimposed to high frequency oscillations. In black a subset of traces was highlighted for ease of visualization. The insets show how the oscillation changes from baseline to the beginning and end of the L6CT activation. D, Top, Representative broadband power spectral density (PSD) of the activity induced by the optogenetic activation of L6CT neurons during the hold-phase (light blue trace) and without optogenetic stimulation (black trace) averaged throughout the probe depth. The PSD shows that L6CT optogenetic activation generates narrowband LFP oscillations in the high gamma range. Bottom, PSD of the activity induced by L6CT activation in each recording in the 65–200 Hz band.
To characterize the properties of the L6CT-induced high gamma oscillation, we high-pass filtered the signal above 65 Hz to remove line-noise and optoelectric artifacts, and separated for each time bin its amplitude, quantified as the absolute value of the voltage, from its frequency, quantified as the frequency with the maximum power in the time–frequency decomposition (Fig. 2A). Inspection of the amplitude and time–frequency dynamics across all light intensities revealed that the oscillation amplitude rapidly increased, followed by a decrease over the initial light ramp (0–250 ms) and then remained stable during the hold period (250–750 ms) of the light. In contrast, the oscillation frequency progressively increased throughout the light ramp and then progressively declined during the hold period. The time course of the oscillation amplitude and frequency averaged across subjects confirmed this pattern: the oscillation amplitude peaked early during the ramp phase and, after a brief reduction, remained stable over time; conversely, the oscillation frequency peaked at ∼200 ms and then steadily decreased over time. Importantly, these observations confirm that the ramp phase exhibits transient effects of the stimulation that disappear during the hold phase. Interestingly, these traces suggested that distinct light intensities drove distinct LFP oscillations (Fig. 2B). To investigate the difference in the oscillations induced by distinct light intensities, we averaged over time (250–750 ms) for each light intensity the oscillation amplitude and frequency and compared them across light intensities (Fig. 2C). We found that while the oscillation amplitude was invariant across light intensities (ANOVA repeated measures; p = 0.07; amplitude at 8 mW/mm2: 2.6 μV; amplitude at 30 mW/mm2: 2.8 μV), the oscillation frequency significantly increased with higher light intensities (ANOVA repeated measures; p < 0.001; frequency at 8 mW/mm2: 104 Hz; frequency at 30 mW/mm2: 130 Hz). Next, we examined the evolution of the oscillation over time in the hold period. To this aim, for each light intensity level and time window (early, 250–500 ms; and late, 500–750 ms, accounting for the relative Hamming windows; see Materials and Methods), we averaged over time the oscillation amplitude and frequency (Fig. 2D). We found that the oscillation amplitude was marginally modulated over time (Wilcoxon signed-rank test; p < 0.01 in 8 mW/mm2 and 16 mW/mm2; average amplitude at 30 mW/mm2: early, 2.85 μV; late, 2.82 μV), while the oscillation frequency consistently decreased from the early to the late time window (Wilcoxon signed-rank test; all p < 0.001; average frequency at 30 mW/mm2: early, 122 Hz; late, 136 Hz).
The high gamma LFP oscillations induced by L6CT neurons are intensity and time dependent. A, The top panels show that the optogenetic activation of L6CT neurons was administered through ramp-and-hold waveforms of four different light intensities (color coded by light intensity). The area outlined by the dashed line highlights the time window during the hold-phase of the light waveform. The shaded areas in light and dark gray highlight the early and late windows of the hold-phase of the light waveform, respectively. The bottom panels show one representative S1 LFP (left) and time–frequency decomposition (right) of the oscillation induced by the optogenetic activation of L6CT neurons showing the emergence of high gamma oscillations. The shaded regions in the bottom panel highlight smaller windows with respect to the top panels to account for the Hamming window of the filtering. B, The traces show the absolute voltage (left) and frequency (right; baseline window shaded) of the LFP over time across different light intensities and averaged across subjects (color code as in panel A). C, The boxplots show the distribution of the absolute voltage (left) and frequency (right) of the LFP during the hold-phase across different light intensities (line, median; box, 25 and 75 percentiles; whiskers, minimum and maximum value; color code as in panel A; ANOVA repeated measures). D, The boxplots show the distribution of the absolute amplitude (left) and frequency (right) across L6CT activation intensities between early (lighter) and late window (darker; color code as in panel A; Wilcoxon signed-rank tests).
Changes in the LFP oscillation frequency are associated with changes in the overall excitability of the circuit (Gao et al., 2017) that is known to be associated with different firing rates across distinct cell classes in the cortex (Tatti et al., 2017). Accordingly, we hypothesized that this evolution of excitability as inferred by changes in the LFP oscillation frequency would reflect in distinct modulation in the firing rates of distinct cell classes. L6CT neurons have two main projection targets: cortical neurons in deep layers (Bortone et al., 2014; Kim et al., 2014) and thalamic neurons [including the reticular (TRN) and ventroposteromedial nucleus of the thalamus (VPm); Sherman and Guillery, 1996]. To investigate the impact of the changes in excitability reflected in the LFP on different cell classes in cortex and in subcortical recipients of L6CT projections, we examined the spiking activity induced by the optogenetic ramp-and-hold activation of L6CT neurons across cortical layers and electrophysiological cell classes in the cortex [regular spiking (RS) and fast spiking (FS)], VPm, and TRN. We found that cortical and VPm units were initially suppressed by L6CT activation, while TRN units were enhanced across all light intensities, as shown in our previous study in this same dataset (Dimwamwa et al., 2024; Fig. 3A). The suppression of most cortical neurons persisted throughout the L6CT activation across all light intensities. One exception to this were layer 4 FS neurons in which, at higher light intensities, the initial suppression was followed by a remarkable and progressive increase in firing rate, to the point of becoming facilitated at the end of the light modulation with respect to their baseline. Another exception were L5 FS neurons that were suppressed by the hold phase of the light stimulus at lower light intensities and almost not modulated at intermediate light intensities. Interestingly, VPm neurons exhibited a similar pattern to L4 FS neurons, and TRN neurons were facilitated during the light activation. To characterize this evolution of the firing rate over time, we compared the average firing rate of the early (250–500 ms) and late (500–750 ms; Fig. 3B) windows of the constant light stimulation. We found a small but significant increase in the firing rates of most neuronal classes except for L6CT and L6 FS units (Wilcoxon signed-rank test, Benjamini–Yekutieli false discovery rate correction for multiple comparisons). This increase was particularly remarkable in VPm and L4 FS neurons at higher light intensities (increase at 8 mW/mm2: VPm, 0.4 Hz; L4 FS, 0.6 Hz; increase at 30 mW/mm2: VPm, 2.1 Hz; L4 FS, 3.4 Hz). Contrary to most other neurons, TRN neurons were initially facilitated by L6CT activation, and this facilitation reduced over time across all light intensities (reduction at 30 mW/mm2: 1.9 Hz). Interestingly, this effect was larger at higher light intensities.
L6CT activation induces two distinct firing profiles across regions, layers, and neuronal classes. A, The top left panel shows the ramp-and-hold light manipulation protocol showing the time windows classified as early (light gray) and late (dark gray). The other panels show the average firing rate over time induced by ramp-and-hold optogenetic stimulation of L6CT neurons across multiple layers and classes of neurons. B, Scree plot for the PCA analysis illustrating the percent variance explained by each PCA component sorted by their explained variance. C, Comparison between the early and late firing rates across layers and neuronal classes (Wilcoxon signed-rank test, Benjamini–Yekutieli false discovery rate correction for multiple comparisons; “x” indicates nonsignificant p > 0.05). D, The left panels show that PCA summarizes the effect of L6CT neurons on other neurons in two components, one stable (green, top panels) and one shifting over time (brown, bottom panels), explaining ∼99% of the variance. The central panels illustrate the back projection of the components on two representative firing rate traces (L23 FS and L4 FS, respectively), showing how firing rates in different neuronal classes can either reflect only the stable component or a sum of both components. The right panels show that the stable component is represented across all cortical neuronal classes, VPm, and TRN, while the shifting component is mainly represented in VPm and L4 FS neurons.
Since L4 neurons are the main target of VPm neurons, it is possible that similarities in their firing rate patterns may reflect shared underlying dynamics. To capture these underlying dynamics, we applied principal component analysis to the average time course of the firing rates of each class of neurons (PCA), finding that two components were sufficient to capture ∼99% of the variance across firing rates (Fig. 3C,D). Both components were characterized by a rapid onset in the first tens of milliseconds from the beginning of the ramp, but the first component was stable over time, as indicated by the relatively constant weight throughout the early and late window, while the second component increased rapidly over time, as indicated by the inversion of the weights' polarity from the early to the late window. These two patterns are reminiscent of the stable time course of most cortical neurons and of the rapidly shifting time course of VPm and L4 FS units, respectively, presented in Figure 3A. This was confirmed by the inspection of the weights of each component across the different cell-type classes, showing that the stable component was represented across all cortical layers, VPm and TRN, while the shifting component was selectively represented with positive weights in VPm and L4 FS units at higher light intensities. Interestingly, the stable component exhibited positive weights in L6CT and TRN neurons and negative weights in all other neurons.
The LFP analysis indicated that L6CT activation induced an oscillation in the high gamma range whose amplitude was stable over time, while its frequency shifted during the constant light window. The analysis of single-unit firing rates indicated that most cortical units exhibited a stable modulation of their firing rates over time, while the firing rates of VPm and L4 FS units shifted from suppression to facilitation. The observation of similar LFP and single-unit dynamics, both consisting of an element stable over time and an element shifting in the timescale of 500 ms, suggested a potential relationship between them. To investigate the relation between LFP and spiking dynamics, we correlated the firing rate of each neuron with the simultaneously recorded LFP oscillation amplitude and frequency (Spearman's correlation). For each class of neurons, we assessed the strength of the correlation by computing the percentage of neurons showing a significant correlation (p < 0.05) with the LFP amplitude and the average Spearman’s rho across neurons showing a significant correlation (Fig. 4A). First, this analysis revealed that L6CT activation increases the percentage of significant correlations between the firing rate of neurons from any class and the LFP amplitude. Additionally, we found that while most neurons in each neuronal class were significantly correlated with LFP amplitude during L6CT activation, the correlation strengths were close to 0 across all light intensities, suggesting a negligible relationship between the LFP amplitude and all cell class firing rates (max, 0.12; min, −0.1; Fig. 4A).
Relation between high-gamma LFP oscillation amplitude and frequency with spiking activity. A, The plot illustrates the relation between the firing rate of each class of neurons and the amplitude of the simultaneously recorded LFP oscillations (Extended Data Fig. 4-1). The circle size shows the percentage of units showing a significant correlation (Spearman correlation; p < 0.05). The color code displays the average strength of the correlation, as quantified by Spearman's rho. The plot shows that L6CT activation induces a nonspecific and weak correlation between all classes of neurons and LFP amplitude. B, Same as panel A for the correlation between firing rate and the frequency of the simultaneously recorded LFP. The plot shows a selective strong correlation between VPm firing rate and LFP frequency across all light intensities and a high correlation with L23 Fs neurons at low light intensities.
Figure 4-1
L6CT activation induces real cortical high gamma oscillations. A. The plots illustrate the spike triggered LFP average for each cell type. The spike triggered LFPs average computed on the spikes occurring during the baseline and late windows are reported in blue and yellow, respectively. All spike triggered averages exhibit either a lack of modulation or a modulation that persists for more than 10 ms. B. The plots illustrate, for the spikes of one representative L4 Fs, L6CT, and VPm neuron extracted from the late window, the five Gabor atoms with peak time closest to the spike peak. From top to bottom, the plots illustrate: the spike-triggered LFP average, the time course of the signal reconstructed from each atom with normalized amplitude, the energy of the 5 selected atoms, and the time course of the signal reconstructed from the 5 selected atoms. Download Figure 4-1, TIF file.
We next correlated the LFP frequency with the single-unit firing rate of all cell classes in a similar manner. In contrast to the relationship with LFP amplitude, the firing rates of most cortical neurons across all classes including L4 Fs neurons were not significantly correlated with LFP frequency, except for L2/3 FS neurons at low light intensity. Another exception was the VPm neurons, as most of them were significantly and negatively correlated with LFP frequency, and with remarkable correlation strengths across all light intensities (Fig. 4B). Although a thalamic spike-triggered analysis of the cortical LFP did not reveal a causal structure directly (Extended Data Fig. 4-1), this does not preclude a causal relationship between the thalamic spiking activity and oscillations in cortex, as suggested by the correlation between VPm firing rate and LFP frequency. Interestingly, while a relatively small number of L6CT neurons were significantly correlated with LFP frequency, they were the only class of neurons showing a positive correlation with the LFP frequency. The spike-triggered LFP average analysis performed for each neuron revealed that spikes were associated with either no cortical LFP activity or with activity that lasted >10 ms, indicating that the high gamma activity observed in the cortical LFP was not produced by the leakage of cortical spikes within the LFP signal (Extended Data Fig. 4-1A). Additionally, while the STA analysis exhibited spike-like waveforms, the matching pursuit analysis resulted in Gabor atoms whose waveform and spectrogram were temporally spread beyond spike artifacts, thus suggesting that the observed activities reflect primarily an oscillatory pattern in the cortical LFP that is not originating from the leakage of spiking activity itself into the LFP (Ray et al., 2008b; Extended Data Fig. 4-1B). Taken together, these analyses confirm that the investigated high gamma LFP activity reflects full-fledge oscillations.
Discussion
In this study, we show that L6CT neurons induce high gamma oscillations in S1 (Crandall et al., 2015; Guo et al., 2017) whose frequency, but not amplitude, depends on the activation level of L6CT neurons and varies over time. L6CT activation induced in most cortical neurons an initial suppression of firing rate that gradually decreases, while VPm and L4 FS neurons exhibited a remarkable switch from being suppressed to being facilitated. These spiking activity profiles were captured by two underlying components: one stable over time and represented in all neurons, and the other shifting over time and represented primarily in VPm and L4 FS neurons. We found that all spiking activity profiles were only weakly correlated with LFP amplitude, while LFP oscillation frequency was selectively correlated with VPm neuron activity.
Interestingly, we found that LFP oscillation frequency, but not amplitude, depends on L6CT activation intensity. This selective modulation of the LFP frequency complements the modulation of LFP oscillation amplitude exerted by pyramidal (Adesnik and Scanziani, 2010; Lu et al., 2015), parvalbumin (Cardin et al., 2009; Sohal et al., 2009; McNally et al., 2021), hippocampal (Butler et al., 2016), and thalamocortical neurons (Hu et al., 2023). The finding that L6CT neurons can sculpt high gamma oscillations frequency suggests that they are critical for cognitive functions (Herrmann et al., 2004; Canolty et al., 2006; Ray et al., 2008c; Gaona et al., 2011; Kucewicz et al., 2017) and diseases (Shin et al., 2011; McNally and McCarley, 2016; Mably and Colgin, 2018) associated with high gamma modulations, especially for conditions that modulate LFP oscillations in a narrow frequency band, such as attention (Pulvermüller et al., 1997), seizures (Fisher et al., 1992; Alvarado-Rojas et al., 2014; Shahabi et al., 2023), dyskinesia (Wiest et al., 2022), and perception (Dubey and Ray, 2020). The spike-triggered LFP average (Extended Data Fig. 4-1) suggests that this high gamma LFP activity may reflect both population-level oscillations and leakage of neuronal spikes (Ray et al., 2008b). The matching pursuit analysis, however, suggests that our observations result primarily from oscillations as opposed to leakage from neural spikes.
The temporal profile of the LFP oscillation frequency correlated with the evolution of VPm firing rate. This suggests that primary thalamic nuclei may mediate the modulation of cortical LFP frequency over time. This finding aligns with previous studies showing that thalamocortical communication can shape narrow-band high gamma oscillation frequency in the cortex (McAfee et al., 2018; Meneghetti et al., 2021). Future causal experiments are required to establish whether VPm activity causes, is caused by, or enables L6CT-induced LFP oscillations. The recurrent loop between S1 and VPm makes them reciprocally interdependent, and disentangling their roles through thalamic suppression may also disrupt intracortical activity (Lanier et al., 1975; Parnaudeau et al., 2013). Indeed, the thalamus serves both as a modulator by enabling intracortical communication and as a driver of cortical activity. Thus, these ambivalent effects would prevent disentangling whether thalamocortical neurons are driving or enabling cortical oscillations.
Our observation that at low light intensities L6CT neurons suppress VPm neurons agrees with previous reports of L6CT-induced suppression of thalamic neurons in vivo in the visual thalamus and in vitro in VPm (Olsen et al., 2012; Crandall et al., 2015). This may be explained by L6CT spikes recruiting inhibitory TRN inputs to VPm that dominates L6CT’s direct excitatory inputs (Cruikshank et al., 2010; Dimwamwa et al., 2024). Interestingly, L6CT spikes induce a prolonged suppression of VPm at the low light intensities tested here, clearly different from the pause-and-burst pattern observed after optogenetic VPm inhibition (Borden et al., 2022) and TRN excitation (Halassa et al., 2011; Russo et al., 2025). This difference may be explained by L6CT neurons simultaneously depolarizing and hyperpolarizing VPm, through their excitatory terminals or through TRN-mediated inhibition, respectively. Previous simulations suggest that the coexistence of TRN hyperpolarizing and cortical depolarizing inputs may keep thalamocortical neurons' potential below spiking threshold, but above the threshold de-inactivating T-currents and evoking bursts (Russo et al., 2025). Indeed, L6CT spikes may keep thalamocortical neurons in a “sweet spot” of suppression, maintaining reduced spiking activity over prolonged time windows, instead of switching to a bursting regime.
The finding that L6CT activation suppresses most cortical layers naturally raises a question about its underlying mechanisms. There are two mechanisms that may contribute to L6CT-induced cortical suppression that may coexist. The first mechanism is polysynaptic inhibition through cortical inhibitory interneurons, as suggested by L6CT neurons strongly recruiting local inhibitory neurons in the primary visual cortex (Olsen et al., 2012; Bortone et al., 2014) and in the deep layers of S1 (Dimwamwa et al., 2024). An alternative mechanism of cortical suppression is the withdrawal of excitatory inputs due to VPm suppression, since thalamic inputs are key to sustain cortical firing (Shu et al., 2003; Russo et al., 2025). Indeed, our results show that one factor captures firing activity in both VPm and L4 FS neurons, a primary VPm target. Hence, we suggest that the withdrawal of excitatory inputs from VPm contributes to the L6CT-induced suppression in L4 FS neurons. In contrast, other cortical neurons mainly present stable suppression, advocating for intracortical inhibition that slowly reduces over time, potentially reflecting the short-term plasticity pattern typical of cortical FS neurons (Ma et al., 2012). Overall, our analyses suggest that intracortical inhibition suppresses cortical activity across layers over time, while the withdrawal of VPm inputs suppresses neurons only for a relatively short time after L6CT activation.
The firing rate dynamics induced by L6CT activation is recapitulated by two principal components, with one stable component represented across all neurons and one component that shifts over time represented only in VPm and L4 FS neurons. These two patterns may reflect the short-term plasticity typical of intracortical and thalamic L6CT synapses, respectively. Indeed, within 1 s, the intracortical excitatory synaptic currents induced by the activation of L6CT neurons varied from a ∼60% increase to a ∼20% decrease, depending on the targeted layer (Kim et al., 2014). A similar activation induced strikingly larger effects on L6CT terminals in VPm, with a ∼400% increase of fast glutamatergic currents and a drop of ∼70% in disynaptic GABA hyperpolarizing currents (Jurgens et al., 2012; Crandall et al., 2015). While the stable component resonates with the relatively weak depression of intracortical circuits (Kim et al., 2014), the shifting component observed in VPm and L4 FS neurons resonates with the strong adaptation in thalamic synapses (Crandall et al., 2015). Indeed, our findings align with previous in vitro reports that L6CT activation switches, within one second, from suppressing to facilitating VPm neurons (Crandall et al., 2015). Overall, the two factors that we isolated may recapitulate the consequences of synaptic adaptation of L6CT intracortical and thalamic projections, respectively.
Interestingly, the rapidly shifting component described here is selectively expressed at the cortical level in L4 FS neurons. While the specific regain of activity in L4 neurons following the initial suppression may be explained by the strong thalamocortical projections from VPm, the selective increase of firing in FS, but not RS, neurons cannot be explained by structural connectivity (Kim et al., 2014; Sermet et al., 2019; Naskar et al., 2021). Instead, it may reflect neurons' synaptic properties: PV+ neurons, associated with fast spiking, can spike in response to one single excitatory synaptic input from VPm, while excitatory neurons require multiple synchronous excitatory inputs from VPm in order to spike (Swadlow, 1995; Jouhanneau et al., 2018; Sermet et al., 2019). Accordingly, the gradual increase in VPm spikes may drive FS neurons more easily than RS neurons.
L6CT neurons have been implicated in many functions and behaviors (Augustinaite and Kuhn, 2020; Kirchgessner et al., 2020; Voigts et al., 2020), as they strategically modulate both intracortical and thalamocortical loops. Furthermore, L6CT neurons can induce time- and intensity-dependent oscillations in the cortex, associated with both facilitatory and suppressive, synchrony-dependent effects on downstream neurons (Dimwamwa et al., 2024), corroborating the idea that L6CT neurons can flexibly adapt to exert diverse functions. Multiple studies implicate L6CT neurons in higher cognitive functions, in line with our finding that they can sculpt the frequency of high gamma oscillations, typical of these functions (Kaiser and Lutzenberger, 2005; Ray et al., 2008c; Gaona et al., 2011). Importantly, our work targets a L6CT population encompassing anatomically and physiologically diverse neurons (Tsumoto and Suda, 1980; Landry and Dykes, 1985; Sirota et al., 2005; Briggs and Usrey, 2009; Kwegyir-Afful and Simons, 2009) involved in heterogeneous behavioral functions (Augustinaite and Kuhn, 2020; Clayton et al., 2021; Dash et al., 2022). This heterogeneity may affect our findings, as distinct L6CT subclasses may engage intracortical and thalamic neurons differently. This limitation can be addressed by future works targeting specific corticothalamic subpopulations to outline their distinct roles in generating oscillations.
While our findings demonstrate that L6CT neurons can modulate the firing rate of downstream units and LFP frequency, they leave an open question as to whether they naturally serve this role, as their physiological relevance remains elusive. Additionally, we explored how L6CT activation affects how S1 LFP activity evolves over time, which raises the question of whether responses to sensory stimuli are differentially modulated over different L6CT activation time windows. Finally, while our study establishes a correlative link between VPm firing rate and LFP oscillation frequency, future experiments are needed to verify a causal link between them, to disentangle the contribution of intracortical and cortico-thalamo-cortical pathways in mediating physiological L6TC neurons effects.
Footnotes
This work was supported by National Institutes of Health National Institute of Neurological Disorders and Stroke BRAIN Grant (NINDS) R01NS104928, NINDS BRAIN Grant RF1NS128896, NINDS Grant R21NS112783, and National Institute of Biomedical Imaging and Bioengineering (NIBIB) BRAIN Grant R01EB029857.
S.R. is the Chief Scientific Officer of Manava Plus. The other authors declare no competing interests.
- Correspondence should be addressed to Garrett B. Stanley at garrett.stanley{at}bme.gatech.edu.
