Abstract
Optogenetic neuromodulation combined with functional MRI (opto-fMRI) enables noninvasive monitoring of brain-wide activity and probes causal connections. In this study, we focused on the secondary somatosensory (S2) cortex, a hub for integrating tactile and nociceptive information. By selectively stimulating excitatory neurons in the S2 cortex of monkeys using optogenetics, we observed widespread opto-fMRI activity in regions beyond the somatosensory system, as well as a strong spatial correspondence between opto-fMRI activity map and anatomical connections of the S2 cortex. Locally, optogenetically evoked fMRI BOLD signals from putative excitatory neurons exhibited standard hemodynamic response function. At low laser power, graded opto-fMRI signal changes are closely correlated with increases in local field potential (LFP) signals, but not with spiking activity. This indicates that LFP changes in excitatory neurons more accurately reflect the opto-fMRI signals than spikes. In summary, our optogenetic fMRI and anatomical findings provide causal functional and anatomical evidence supporting the role of the S2 cortex as a critical hub connecting sensory regions to higher-order cortical and subcortical regions involved in cognition and emotion. The electrophysiological basis of the opto-fMRI signals uncovered in this study offers novel insights into interpreting opto-fMRI results. Nonhuman primates are an invaluable intermediate model for translating optogenetic preclinical findings to humans.
Significance Statement
The neocortex consists of excitatory and inhibitory neurons, each playing distinct roles in cognition and contributing to brain disorders. This study marks the first to use optogenetics to selectively stimulate excitatory neurons of the primate’s secondary somatosensory (S2) cortex, establishing a spatial correspondence between brain-wide optogenetic neuromodulation combined with functional MRI (opto-fMRI) activity and anatomical tracer-based connections in the S2 hand region. The differential relationships between laser power and changes in BOLD, local field potentials (LFPs), and spiking activity suggest that LFP signals better reflect opto-fMRI signals at the excitatory neuron population level. This work has translational potential for developing targeted neuromodulation therapies for neurological and psychiatric disorders. Success in nonhuman primates, along with advances in noninvasive adeno-associated viral vector delivery, paves the way for future clinical applications.
Introduction
Functional MRI (fMRI) has become the primary tool for mapping distinct functional circuits and understanding human brain functions. The proper interpretation of human fMRI findings relies on the assumption that fMRI signals reflect the underlying activity of neurons in each brain region. Given that the neocortex is primarily composed of excitatory and inhibitory neurons, each playing distinct roles in cortical information integration and cognition, it is crucial to determine to what extent activity from specific neuron subtypes influences neuron–hemodynamic relationship. Understanding this relationship is particularly significant, as imbalances between excitatory and inhibitory activities have been implicated in numerous neurological and psychiatric disorders, including autism, schizophrenia, depression, and anxiety (Yizhar et al., 2011; Culotta and Penzes, 2020; Kaul et al., 2022), as well as chronic pain (Schweinhardt et al., 2009).
Optogenetic photostimulation of neural populations in the brain in vivo is a cutting-edge technique that uses light-sensitive proteins to achieve precise spatial and temporal control of specific cell types, enabling the investigation of cellular functions and neural circuits, and their causal relationships with behaviors (Boyden et al., 2005; Cavanaugh et al., 2012; Gerits et al., 2012; Kim et al., 2017). While fMRI provides whole-brain coverage, researchers can noninvasively map activity related to particular brain functions at local and global networks. By combining fMRI with optogenetics [termed optogenetic neuromodulation combined with functional MRI (opto-fMRI)], researchers can causally manipulate specific brain circuits and observe the resulting changes in brain activity across regions (Chuang et al., 2023). This causal approach helps clarify the roles of different brain circuits in behavior and disease, by linking activities of specific neuron types (such as excitatory or inhibitory neurons) to the functions of large-scale brain circuits.
To date, most of the opto-fMRI studies have been conducted in rodent models (Lee et al., 2010; Desai et al., 2011; Chuang et al., 2023; Kim et al., 2023). In recent years, optogenetic studies in nonhuman primates (NHPs) have grown rapidly (Klink et al., 2021); for a review, see Galvan et al. (2017). Given the greater complexity of brain circuits and neuronal types in NHP, which more closely resemble those in the human brain, establishing opto-fMRI in NHP is a crucial step toward potential human applications, especially as technology continues to advance.
We focused our study on the secondary somatosensory (S2) cortex of NHPs because it is a critical region involved in somatic sensory processing and pain perception. Both animal and human fMRI studies suggest that the S2 cortex and its corresponding parietal operculum region in the human brain serve as key hubs for gating sensory information to high cognitive and emotional circuits (Taub et al., 2024), making it a potential target for neuromodulation. For instance, our previous research in the same NHP model used in this study identified brain regions and networks involved in processing touch and painful information, including the S2 cortex and posterior insula (Wu et al., 2017, 2022a; Ye et al., 2021), through stimulus-driven and resting-state fMRI. Therefore, gaining a better understanding of the causal functional and anatomical connections of the S2 cortex is essential for providing novel insights into the circuit-level pathology in chronic pain disorders and related neurological conditions (Maihofner et al., 2006; Condylis et al., 2020; Ishida et al., 2024; Taub et al., 2024), as well as for developing new therapies.
In this study, we focus on the S2 cortex with the following aims: (1) to exploit the ability of optogenetics to selectively activate excitatory neurons and map its functional connections, (2) to understand how excitatory neuron activity influences the BOLD signal and hemodynamic function, (3) to examine the extent to which opto-fMRI BOLD signal changes reflect underlying electrophysiological local field potential (LFP) and spiking activities of cortical excitatory neurons, and (4) to investigate whether the functional connectivity patterns monitored by fMRI correspond to underlying anatomical connections in the NHP brain. Since the compositions and arrangements of excitatory and inhibitory neurons vary across brain regions—contributing to the specialized function of each area, such as the S2 cortex—our current study allowed us to elucidate the specific circuits through which the innocuous tactile and nociceptive sensory information are processed and integrated. These studies provide novel insights into the interpretation of opto-fMRI data and help delineate the causal functional and anatomical circuits of the primate S2 cortex.
Materials and Methods
Four adult squirrel monkeys (Saimiri sciureus, one female, three males) were studied, involving a total of five hemispheres. Three monkeys received AAV5/AAV9 or AAV9 (adeno-associated viral vector: AAV) injections along with optical probe implants in the S2 hand regions on the left hemispheres. Additionally, two probes were implanted in two virus-free S2 regions (without AAV injection) in the opposite (right) hemisphere of one monkey that had received AAV injection in the left hemisphere and one naive monkey. All procedures adhered to National Institutes of Health and ARRIVE guidelines and were approved by the Institutional Animal Care and Use Committee of Vanderbilt University.
Animal preparation
Animals were initially sedated with ketamine hydrochloride (10 mg/kg, i.m.) and given atropine sulfate (0.05 mg/kg, i.m.), followed by anesthesia with isoflurane (in 1.5–2% range for surgery and 0.8–1.2% range for fMRI) delivered in a 70:30 N2O/O2 mixture. After intubation, the animal was placed in either a stereotactic frame or a custom-designed MR cradle, with the head secured by ear bars and head bars. Ringers’ solution was infused intravenously (3 ml/kg/h) to prevent dehydration. The animals were artificially ventilated. Rectal temperature was maintained between 37.5 and 38.5°C using a circulating water blanket. Heart rate, peripheral capillary oxygen saturation (SpO2), respiration pattern, and end-tidal CO2 (22–26 mmHg) were continuously monitored and maintained within appropriate ranges throughout the procedure.
Viral injection after electrophysiology mapping of the S2 cortex and based on fMRI activation maps
Prior to virus injection, the hand-digit region in the S2 cortex of each hemisphere was identified electrophysiologically by mapping the somatotopy organization and characterizing the neuronal properties of the receptive field and preferred peripheral stimuli using an intracranial microelectrode. LFP and spiking activity were recorded with a linear array, followed by virus injection.
For S2 mapping, a craniotomy was made at the juncture of the lateral and central sulci, where somatosensory area 3b and S2 cortices are located. The dura matter was removed to visualize the blood vessel patterns on the brain surface, which were used as landmarks for subsequent repeated microelectrode recordings. Single epoxylite-coated tungsten microelectrodes (FHC Inst.) with a sharp tip (<3 μm) and ∼1 MΩ impedance were inserted, first through the cortical layers of the area 3b facial regions, and then further advanced into the upper bank of the lateral sulcus, where the S2 digit region is located. At each electrode penetration site, the microelectrode was advanced in 300 µm increments, with its tip depth logged.
The preferred stimulus type (i.e., stroke, squeeze, and light touch) and response strength (scored on a six-level scale) were characterized along the cortical depth, as well as the receptive field properties. Receptive fields of neurons were identified by palpating and squeezing the contralateral hand while listening to an audio amplifier for spike activity and viewing action potential traces on a display. Based on these characteristics—receptive field properties, preferred stimuli, and the somatotopic organization of the digits, as established in this species (Kaas et al., 1984; Jain et al., 2008; Wang et al., 2013a)—we identified the S2 digit region.
We specifically mapped and localized clusters of both low-threshold mechanoreceptive neurons and heat-sensitive neurons. A detailed characterization of these two regions was previously described in Ye et al. (2021).
Injections of AAV, validation, and optical fiber implantation
We chose to inject AAVs, AAV5 and AAV9 (pAAV-CaMKIIa-hChR2(H134R)-mCherry; www.addgene.org), which have demonstrated transfection efficacy in NHPs. Guided by the electrophysiologically defined S2 hand representation map, AAV was injected into two locations: AAV5 into the hand tactile neuron cluster and AAV9 into a posterior heat-sensitive neuron cluster (∼1 mm posterior), both within the S2 digit region, at two cortical depths: the middle granular layer (3,500–4,500 µm) from the brain surface and the infragranular layer (500 µm above the middle layer). A total of 2–3 injection tracks were made in the hemisphere of the dominant hand, with 2 ml of virus solution injected per track. Figure 1 A and B illustrate the digit representation map and the injection sites in one representative monkey. We allowed 4–6 weeks for virus transfection before conducting optoelectrophysiology testing. Before implanting the optical fiber, we used optoelectrodes to deliver laser light stimulation and record neural activity to confirm the successful expression of humanized channelrhodopsin-2 in S2 neurons. The tip of the optical fiber was placed at a specific cortical depth (SM5411, 4,200 µm; SM6599, 4,000 µm). The same S2 identification procedure was performed on the virus-naive cortex, and an optical fiber was implanted at this virus-naive control site.
Optogenetically elicited neuronal spiking activity in the S2 cortex. A, Postmortem brain image showing the location of the electrophysiology mapping of S2 digit representation, B, which was used to guide the injection of viral vectors into the S2 cortex. C, A single-run coronal fMRI activation map shows the tissue volume corresponding to the electrophysiology mapping. D–I, Examples of simultaneous optogenetic stimulation and recorded spiking and LFP activity during three stimulation conditions: blue light (D, E), green light (F, G), and 8 Hz vibrotactile stimulation of the distal pad of digit 2 (H, I). The spiking panels show the raster plot across trails (top) and PSTH (bottom). The x-axis represents time, and the y-axis represents spike rate or LFP amplitude. Schematic inserts to the right depict the recording channel (red dot and number) on the electrode and tactile finger stimulation. J, Illustration of the quantification scheme of PSTH constructed from the response to blue laser illumination. The red line is the upper confidence interval of 99% for determining the response onset time. K, M, O, Box plots of peak amplitude, response latency, and time to peak of spiking activity as a function of blue light stimulation at seven power intensities. L, N, P, Group-level comparisons of peak amplitude (L), response latency (N), and time to peak (P) between optogenetic and tactile stimulation conditions. Q, R, S, Comparison of spike widths (through to peak) between neurons responsive to blue laser stimulation and those responsive to tactile stimulation, across three groups: all neurons (Q), excitatory neurons (R), and inhibitory neurons (S). Unpaired t test: *p < 0.05; ****p < 0.0001.
Electrophysiological data acquisition, preprocessing, and analysis
Four to six weeks after AAV injection, a 32-channel NeuroNexus (A32-Poly2, optical fiber core, 125 mm with 0.22 NA) or Cambridge (H7b, optical fiber core, 100 mm with 0.37 NA) optoelectrode was inserted into the AAV injection site to record laser-evoked spiking and LFP activity (CerePlex Direct, BlackRock Neurotech). During recording, the exposed cortex was covered and stabilized with 4% agar.
The Fieldtrip data analysis toolkit was used to process LFP signals. Raw LFP data were filtered with 60 and 120 Hz band-stop Chebyshev filters to remove power line frequency interference and then bandpass filtered between 1 and 150 Hz for quantification. Stimulus-evoked LFP signals were averaged across 30–50 trials within each run. Spike activity was initially sorted offline using Plexon's Offline Sorter software (Plexon), and the results were presented as scatterplots and peristimulus time histograms (PSTHs; 2 ms bin width, Neuroexplorer). PSTHs from 30 to 50 trials were averaged within each run and across runs within each recording session (day). The 99% confidence intervals for the spike count were computed to determine whether increases in spiking rate during stimulation were statistically significant, as well as to detect the response latency at which the spike count met or exceeded the upper confidence interval.
To classify extracellular recordings from putative excitatory or inhibitory neurons, the trough-to-peak duration was measured from the mean of sorted spike waveforms and used as a classification criterion. Excitatory neurons typically exhibit longer trough-to-peak durations compared with inhibitory neurons, with commonly used cutoffs of 0.2–0.22 ms (Hussar and Pasternak, 2009; Jacob et al., 2013; Thiele et al., 2016). In our study, we employed a cutoff of 0.22 ms: waveforms wider than 0.22 ms were categorized as putative excitatory neurons, while those narrower were considered putative inhibitory neurons.
MRI data acquisition and analysis
Structural and fMRI images were acquired using a 9.4 T Varian scanner (Varian Medical System) with a custom two-channel head coil. T2*-weighted structural images were obtained using a gradient echo sequence (FOV = 55 × 55 mm2, matrix size = 512 × 512, in-plan resolution 0.107 × 0.107 mm2, 24 slices, thickness = 1 mm). BOLD fMRI images were acquired using a 4-shots T2*- weighted 2D GE-EPI sequence (TR = 750 ms, TE = 16 ms, FOV = 55 × 55 mm2, matrix size = 128 × 128, in-plan resolution = 0.43 × 0.43 mm2, 24 slices, thickness = 1 mm). The stimulus protocol included a 30 s baseline, followed by interleaved 6 s ON and 24 s OFF periods of blue and green light stimuli, delivered to the left AAV-treated and right AAV-free hemispheres, respectively. Each fMRI run consisted of 12 epochs (six for each condition) and yielded a total of 250 image volumes.
fMRI images underwent standard preprocessing steps, including regression of respiration noise using the RETROICOR method (Glover et al., 2000), slice time correction, and 3D motion correction with AFNI. Functional EPI images were spatially smoothed using an isotropic Gaussian filter kernel with a full width at half maximum of 0.8 mm. These images were then upsampled from 0.43 × 0.43 × 1 to 0.215 × 0.215 × 1 mm3 and coregistered with corresponding T2*-weighted anatomical images for display. The fMRI BOLD EPI data were temporally smoothed with a low-pass filter ranging from 0.01 to 0.25 Hz (fslmaths, FSL). fMRI activation maps were generated using a cross-correlation function between the signal time courses of each voxel and the boxcar predictor of the hemodynamic response function (HRF)-convolved stimulus presentation paradigm (3dDeconvolve, AFNI). For each run, activated voxels were defined as those exhibiting stimulus-related signal changes at a statistically significant level of p ≤ 0.01, corrected for false discovery rate with a q value of 0.05 and a minimum cluster size of 10 consecutive significant voxels. To evaluate the reliability of fMRI activation maps, probabilistic activation maps were generated to quantify the spatial reproducibility of activity patterns across scan sessions. Voxels showing activation in >40% of EPI runs (out of a total of five runs; Fig. 6A) were overlaid onto T2*-weighted anatomical images for visualization. For methodology details, see Yang et al. (2018). This 40% probability criterion was chosen based on signal-to-noise and contrast-to-noise ratios, as well as the number of data acquisition runs achievable within each imaging session, in our previous fMRI studies at 9.4 T MRI using comparable stimulation parameters (Chen et al., 2012).
BOLD signal time courses were extracted from the five peak voxels (identified by maximal t-value) within each selected brain region, including the S2 cortex, insula, anterior cingulate cortex (ACC), midcingulate cingulate cortex (MCC), amygdala (Amy), putamen (Put), area 3b, primary motor cortex (M1), and thalamic ventroposterior lateral nucleus (VPL). BOLD response amplitudes to stimuli were quantified, with measurements from each experimental run averaged across runs and animals. Statistical significance was evaluated using a Brown–Forsythe and Welch ANOVA test, with a p-value of <0.05 considered statistically significant. ROI-based BOLD time course results are presented as mean ± standard error of the mean (SEM) unless otherwise specified.
Laser and tactile stimulation
Our laser illumination system can deliver two wavelength lights. Blue light, provided by a commercial blue light-emitting diode solid-state laser (473 nm, 80 mW, LuxX laser, Omicron-Laserage Laserprodukte), was used to activate channelrhodopsin 2 (ChR2), while a green laser (561 nm, 100 mW, OBIS LS/LX laser, Coherent) was used as a control. To assess the possible influence of thermal effects on neuronal processing, hybrid multibaseline thermometry was employed to measure the temperature around the optical probe. There was only a ∼0.02°C increase around the probe under 20 mW laser power during the functional scan (250 image volumes and 12.5 min in each functional scan; Luo et al., 2023). The temperature increase was negligible under 1, 2, and 4 mW (8 Hz) conditions.
For optogenetic stimulation, blue or green light was delivered to the S2 cortex, measured at the fiber tip (200 µm, NA = 0.39), corresponding to the following laser power settings: of 1 mW (tip, 1 mW, 8 mW/mm2), 2 mW (tip, 2 mW, 16 mW/mm2), 4 mW (tip, 4 mW, 32 mW/mm2), 8 mW (tip, 5 mW, 40 mW/mm2), 16 mW (tip, 9.5 mW, 76 mW/mm2), 20 mW (tip, 12 mW, 95 mW/mm2), and 30 mW (tip, 17 mW, 135 mW/mm2). Tactile stimulation, provided at 8 Hz by a rounded probe delivering 0.43 mm vertical displacement, was presented on the monkey's distal finger pads in the same manner as laser stimulation.
Postmortem immunohistology
The AAV injection and recording sites were later confirmed by postmortem immunohistology and anatomical reconstructions of the electrode tracks and viral expression in two monkeys (SM5411, 464 d post-AAV injection, and SM6599, 856 d post-AAV injection). After data collection was completed, animals were killed with pentobarbital overdose and then transcardially perfused with phosphate-buffered saline, pH 7.4, followed by fixative solutions containing 4% paraformaldehyde (PFA) with 10% sucrose. The brains were subsequently extracted and immersed in 30% sucrose in PB at 4°C overnight for cryoprotection and sectioned coronally into 40–50 µm thick slices using a freezing microtome.
The slices underwent immunochemical staining as described by De et al. (2020), using Living Colors mCherry monoclonal antibody (Takara Bio #632543), amplified with donkey anti-mouse IgG (H + L) highly cross-adsorbed polyclonal secondary antibody conjugated with Alexa-Fluor fluorescent tag (Invitrogen), and counterstaining with 4′,6-diamidino-2-phenylindole dihydrochloride (Invitrogen). Sections were mounted and coverslipped with ProLong Diamond Antifade Mountant. Fluorescence images of opsin expression were scanned and digitized for visualization by the Digital Histology Shared Resource at Vanderbilt University Medical Center. Slides were imaged on an Aperio Versa 200 automated slide scanner (Leica Biosystems) with whole slide images at 20× magnification to a resolution of 0.323 µm/pixel.
Results
Spiking activity elicited by optogenetic stimulation
To ensure precise and effective stimulation of excitatory neurons in the S2 cortex, we first mapped the hand region in the S2 cortex using microelectrode electrophysiology and receptive field mapping, followed by AAV injection into the left hemisphere, leaving the opposite right hemisphere as a virus-free control. Before the injection, we first localized the S2 cortex using fMRI and then mapped the hand region by identifying cortical territories containing neurons with receptive fields on the hand (Fig. 1A–C), using intracranial microelectrode on both hemispheres (Fig. 1A,B; Ye et al., 2021). The injected AAV, similar to that used by Lee et al. (2010), utilized a calcium–calmodulin-dependent protein kinase II alpha (CaMKIIα) promoter to drive the expression of blue light-sensitive ChR2 along with a cherry fluorescent tag into excitatory neurons.
Six weeks after AAV injection, we placed an optrode at the injection site to concurrently record spiking and LFP signals under three stimulus conditions: (1) delivery of blue (473 nm) light, (2) green (561 nm) light to the S2 digit region, and (3) tactile stimulation of the distal fingerpad (Fig. 1D–I). Representative raster plots of spiking activity across trials and PSTHs showed robust spiking activity during blue light and tactile stimulation (Fig. 1D,H). The neuron exhibited low spontaneous spiking activity during rest but responded rapidly to 10 ms blue laser illumination (at 4 mW, with an intensity of 32 mW/mm2). The spiking rate gradually increased after the onset of light stimulation, peaking ∼20 ms after the illumination ceased (Fig. 1D).
Similarly, the LFP signal amplitude (in microvolt) increased sharply following brief blue laser stimulation, peaking ∼20 ms after stimulus onset (Fig. 1E). In comparison, the LFP signal evoked by tactile stimulus (delivered at 8 Hz frequency with 20 ms pulse duration) was weaker and shorter (Fig. 1I). Both the spiking and LFP responses to tactile stimulation differed from those evoked by direct laser stimulation of S2 neurons. Spiking activity ceased immediately when the tactile stimulus stopped, indicating a rapid adapting response with minimal poststimulus activity. The LFP amplitude changes were also much shorter in duration compared with those elicited by blue laser stimulation. As a control, no significant spiking (Fig. 1F) or LFP (Fig. 1G) activity was observed during or after green light stimulation. The same experiment was conducted on the AAV-free S2 region in the opposite hemisphere and in AAV-free animals. In these cases, spiking and LFP activities were only elicited by natural tactile stimulation of digits (data not shown). Collectively, these studies confirmed that the blue light-sensitive opsin was successfully expressed in the neurons in the hand representation area in the S2 cortex as intended.
At the group level, we quantified and compared three dynamic measures of spiking activity based on the PSTH plots: peak amplitude, response latency, and time to peak, under both blue light and tactile stimulation (Fig. 1J). Two key features emerged. First, there was no apparent trend for peak amplitude changes (1 mW, 109.70 ± 55.5 spikes/sec; 2 mW, 93.42 ± 51.13 spikes/sec; 4 mW, 105.30 ± 43.28 spikes/sec; mead ± SD; linear fit slope = −0.4; R2 = 0.01; Fig. 1K), but there were decreasing trends for response latency (1 mW, 6.83 ± 3.92 ms; 2 mW, 8.10 ± 3.82 ms; 4 mW, 4.72 ± 2.64 ms; mead ± SD; linear fit slope = −0.8; R2 = 0.57; Fig. 1M) and time to peak (1 mW, 15.33 ± 7.30 ms; 2 mW, 12.43 ± 4.13 ms; 4 mW, 9.36 ± 5.19 ms; mead ± SD; linear fit slope = −1.9; R2 = 0.97; Fig. 1O) when the blue laser power was increased from 1, 2, and 4 mW. No apparent trend was observed when the laser power exceeded 8 mW. Second, compared with the tactile stimulation, blue laser-elicited spiking activity was significantly stronger in amplitude (tactile, 59.96 ± 47.12 spikes/sec; blue laser, 107 ± 58.89 spikes/sec; mean ± SD; Fig. 1L), exhibited much shorter response latency (mean ± SD, 6.1 ± 3.9 ms for laser vs 9.7 ± 2.6 ms for tactile; Fig. 1N), and peaked faster (mean ± SD, 12.0 ± 6.6 ms for laser vs 13.5 ± 4.5 ms for tactile; Fig. 1P). These results indicate that blue light directly activates S2 excitatory neurons, eliciting robust and faster responses compared with natural tactile stimulation, where signals are transmitted from the peripheral skin and afferents to the S2 cortex via ascending pathways.
Confirm blue light-responsive neurons are excitatory neurons
We confirmed that blue light-responsive neurons are excitatory neurons based on two characterizations: layer distribution and spike waveform analysis. It is known that excitatory neurons are distributed differently across cortical layers (Lubke and Feldmeyer, 2007; Radnikow and Feldmeyer, 2018). To examine the distribution of light-responsive neurons, we plotted the spiking activity of neurons at different cortical depths using a 32-channel optrode microarray under both blue and green laser illumination (Fig. 2). Blue light-responsive neurons were scattered across various cortical depths, as indicated by significant spiking activity in a subset of channels. For instance, channels 21a, 22a, and 12a (Fig. 2A, red *) exhibited increased firing rates during and after blue laser illumination, while other channels showed lower firing rates, supporting the notion that excitatory neurons are not uniformly distributed across cortical depth. Importantly, no significant spiking activity was observed during and after green laser illumination (Fig. 2B).
Distribution of blue light-responsive neurons across the cortical depths in the S2 cortex. A, PSTH plots illustrate the firing patterns of neurons detected at different channels in response to blue laser illumination. The PSTH display follows the microarray contact layout in a top-to-bottom arrangement, corresponding to the inferior-granular to granular layers. Red asterisks (*) on the microarray probe indicate the channels with the most robust spiking activity to blue light stimulation. The corresponding PSTH plots for these channels (Chan021a, Chan022a, and Chan012a) are also marked with red asterisks. B, PSTH plots show spiking activity during green laser illumination. The mean waveform is inset in the PSTH plots in both panels A and B.
Next, we classified neurons as putative excitatory or inhibitory based on spike widths (trough-to-peak duration), using 0.22 ms as the cutoff. Neurons with spike width between 0.22 and 0.98 ms were classified as putative excitatory, while those between 0.1 and 0.21 ms were considered putative inhibitory. The classification method and distribution of the spike widths are shown in Figure 3A, based on methods from the literature (Mitchell et al., 2007; Hussar and Pasternak, 2009; Jacob et al., 2013; Thiele et al., 2016).
Classification of neurons responding to optogenetic and tactile stimulation. A, The spike waveform width was calculated for each neuron by measuring the duration between the trough and peak of the average waveform. Exemplary waveforms for excitatory (blue) and inhibitory (red) neurons are shown below, along with the corresponding trough-to-peak time measurement (bottom). B, C, Distribution spike counts as a function of spike waveform widths, induced by blue laser illumination (B) or tactile stimulation (C). A trough-to-peak duration of 0.22 ms was used to classify putative excitatory and inhibitory neurons.
A total of 221 blue light-responsive neurons were isolated in the S2 cortex of two monkeys, with 205 neurons (93%) classified as excitatory neurons and 16 neurons (7%) as inhibitory neurons (Fig. 3B). The distribution of spike widths in tactile neurons differed: 55 out of 64 neurons (86%) were excitatory neurons, while 9 (14%) were inhibitory neurons (Fig. 3C). Regardless of neuron type, the spike widths of blue light-responsive neurons (mean ± SD, 0.47 ± 0.14 ms) were significantly wider than those of tactile neurons (0.33 ± 0.12 ms; unpaired t test: two-tailed, p < 0.0001, t = 6.94, df = 283; Fig. 1Q). When comparing excitatory neurons specifically, the spike widths of the blue light-responsive excitatory neuron were significantly wider than those of tactile neurons (unpaired t test: two-tailed, p < 0.0001, t = 7.03, df = 249; mean ± SD: blue laser = 0.5 ± 0.11 ms, tactile = 0.37 ± 0.12 ms; Fig. 1R). However, the spike widths of blue light-responsive inhibitory neurons did not differ significantly from those of tactile-responsive inhibitory neurons (unpaired t test: two-tailed, p = 0.0507, t = 2.062, df = 23; mean ± SD: blue laser = 0.16 ± 0.04 ms, tactile = 0.19 ± 0.02 ms; Fig. 1S). Additionally, plots of excitatory neuron distribution by spike widths (Fig. 3B) revealed two distinct peaks at 0.39 and 0.54 ms, suggesting the presence of two different excitatory neuronal populations within the S2 cortex. Together, the layer distribution pattern and spike waveform analysis support the successful expression of the ChR2 opsin in S2 cortex excitatory neurons.
Characterization of optogenetic stimulation-evoked LFP activity
Optogenetic stimulation-evoked LFP signal changes reflect the electrophysiological activity of a population of excitatory neurons. As the blue laser power was increased from 1 to 16 mW, the LFP signals exhibited progressively larger negative deflections (Fig. 4A,B). However, further increases in laser power beyond 16 mW (up to 30 mW) did not result in a significant increase in the evoked response amplitude. In comparison, the LFP amplitude evoked by natural tactile stimulation of the monkey's finger was significantly weaker than that evoked by the laser (Fig. 4A, black line; Fig. 4B, black bar).
The relationship between varying laser powers and evoked LFP amplitudes, compared with LFP signals in virus-free S2 and during tactile stimulation. A, Average LFP signal time course in response to blue light stimulation at varying powers and 8 Hz tactile stimulation. Inset: bar plots of the time-to-peak measures of the LFP response. B, Bar plots of LFP peak amplitudes (mean ± standard error) at different blue laser powers. C, Average LFPs to green laser stimulation at varying powers. Tactile-evoked LFP signals are shown as a control. D, Bar plots of the LFP response peak amplitudes (mean ± standard error) at different green laser powers. E, G, Average LFPs induced by blue (E) and green (G) laser stimulation in the virus-free S2 cortex (without AAV injections). F, H, Bar plots of the peak LFP signal amplitude during blue (F) and green (H) laser stimulation at varying powers. One-way ANOVA with Tukey post hoc, ****p < 0.0001.
To better understand the speed of information processing under blue laser versus tactile stimulation, we measured the time to peak of LFP responses. We found that the latencies of LFP peaks, regardless of laser intensity, were consistently shorter than those evoked by tactile stimulation (Fig. 4A inset). When green laser light stimuli were delivered at the same power levels, no LFP signal changes were observed (Fig. 4C). Additionally, we replicated the experiment in the AAV-naive S2 cortex, which lacks ChR2 opsin expression. No apparent LFP signal changes were detected during blue laser (Fig. 4E,F) or green laser (Fig. 4G,H) illumination. As a control, tactile stimulation evoked distinctive LFP responses from the S2 cortex, confirming that the recording site was functioning properly. The difference in the temporal features of light-evoked versus tactile stimulation-evoked LFP signals may suggest the involvement of different populations of excitatory neurons.
Graded fMRI BOLD responses to varying laser intensity stimulation in the S2 cortex
For the opto-fMRI experiments, blue light was delivered via an implanted optical fiber (Fig. 5A), which elicited robust, laser power-dependent BOLD signal increases. The location of the optical fiber is visible on MRI images as a signal void, indicated by the magenta arrows in Figure 5 B and C. BOLD activation was detected in the cortex beneath the optical fiber in the upper bank of the lateral sulcus, where the S2 cortex resides. A zoomed-in view illustrates the spatial relationship between the tip of the optical fiber and the BOLD signal increase, indicated by an orange–yellow patch outlined with a white dotted line (Fig. 5C).
Optogenetic BOLD fMRI activation setup and signal changes at the S2 cortex. A, Setup for optogenetic fMRI data acquisition. A single-run opto-fMRI activation maps are shown on two coronal brain slices B, with C, providing a zoomed-in view of the S2 region with the implanted optical fiber (indicated by arrows). PCC, posterior cingulate cortex; 3b, primary somatosensory area 3b. D, Left: averaged BOLD time course extracted from the S2 cortex with AAV injection during blue light stimulation at three power levels (1 mW, green; 2 mW, orange; 4 mW, red) compared with those from the AAV-free S2 cortex (control: black line) during the same stimulation. E, Plots of normalized signal changes (mean ± SEM) of BOLD, LFP, and spikes (dotted color lines) at three different blue laser powers, with linear fitting (solid color lines). r: correlation coefficient between the three curves of signal changes: BOLD and LFP (0.98), BOLD and spikes (−0.05), and LFP and spikes (−0.25). F, Comparison of fMRI BOLD signal changes in off-target brain regions during optogenetic stimulation of S2 excitatory neurons. Each bar represents the mean peak % BOLD signal changes for each region of interest, with error bars indicating the standard error of the mean (SEM). *p < 0.05, Brown–Forsythe and Welch ANOVA test.
The time courses of BOLD signal changes extracted from the voxels surrounding the optical probe exhibited a standard HRF during 6 s of blue light stimulation, with peaks occurring between 9 and 12 s, depending on laser power (Fig. 5D). Plots of normalized % BOLD signal changes as a function of laser power intensities (1 mW, 0.30%, 2 mW 0.33%, and 4 mW, 0.40%) showed a monotonically increasing trend (Fig. 5E, dotted blue line, linear slope = 0.06, R2 = 0.8). In addition to the BOLD activation detected at the stimulated S2 location, a brain-wide global fMRI activation pattern was apparent (Figs. 5B, 6A). Blue light stimulation of virus-naive S2 cortex did not elicit detectable fMRI signal changes in the opposite side of the hemisphere or in virus-free animals (Fig. 5D, black line).
Relationship between opto-BOLD fMRI and optogenetically evoked spiking and LFP signals
Understanding how the activity of different types of neurons, such as the excitatory neurons at the individual level (via single unit spike activity) or as a population (via LFP signal), contributes to opto-fMRI is essential for properly interpreting opto-fMRI observations. To address this, we normalized the response amplitude of each measure to that observed during 1 mW light stimulation and plotted the relationship between signal changes as a function of blue light intensity on the same graph to compare their dose–response curves (Fig. 5E; dotted blue line, BOLD; orange line, LFP; green line, spikes). Both BOLD and LFP signals exhibited linear response curves (solid line, BOLD: R2 = 0.99, slope = 0.11; LFP: R2 = 0.97, slope = 0.23), while spikes followed a nonlinear curve (R2 = 0.005; slope = −0.004). Importantly, the changes in BOLD and LFP signals were highly correlated (correlation coefficient r = 0.98), whereas spiking rates showed no significant correlations with either BOLD (r = −0.24) or LFP (r = −0.05) signals (Fig. 5E). This finding indicates that when the excitatory neuron activity dominates the source of BOLD and LFP signals, their changes in response to increased laser power are closely associated.
Global fMRI and anatomical network of the S2 cortex and correspondence between fMRI and anatomical tracer maps to the S2 cortex
In addition to robust local BOLD signal changes at the S2 region, we observed global BOLD fMRI activation in numerous cortical and subcortical regions during blue laser illumination of S2 excitatory neurons (Fig. 6A). These activated regions include somatosensory area 3b, M1, Put, VPL, insula cortex, Amy, ACC, and MCC. The BOLD time course sampled at the S2 and other selected regions exhibited a standard HRF (Fig. 6B,C). Significant differences were observed in the BOLD response amplitudes between the stimulated S2 cortex and off-target sensorimotor cortical regions, such as area 3b and M1 (*p < 0.05, Brown–Forsythe and Welch ANOVA test), as well as cortical and subcortical regions involved in the processing of touch, affective, and emotional aspects of pain, such as the ACC, MCC, Amy, and Put (Fig. 5F). Notably, the BOLD signal was stronger in off-target regions such as the insula and ACC compared with sensorimotor areas such as the VPL and M1. These fMRI results suggest that excitatory neuron activity elicited by optogenetic stimulation of S2 propagates robustly to remote cortical and subcortical regions, inducing stronger BOLD responses in perception- and emotion-related regions than in sensorimotor regions.
Spatial correspondence between optogenetically evoked fMRI BOLD activation maps and global anatomical connections of AAV tracer injected into the S2 cortex. A, Multirun BOLD activation probability maps from monkey A during 30 mW blue laser stimulation of the S2 cortex, which received the AAV injection. B, C, Averaged BOLD time courses extracted from voxels surrounding the implanted optical fiber in the S2 (B, shadow: SEM (standard error)) and eight brain regions: Insula, insular cortex; ACC, anterior cingulate cortex; MCC, midcingulate cortex; Amy, amygdala; Put, putamen; 3b, somatosensory area 3b; M1, primary motor cortex; VPL, thalamic ventroposterior lateral nucleus. D–I, Histological evaluation of AAV expression in one subject. Cortical and subcortical regions showing mCherry expression from the AAV injection (E, G, and I) are annotated on corresponding brain slices from a squirrel monkey atlas (D, F, and H). The small letters in the bottom right corners indicate the stereotaxic AP coordinates of each sectional image. E1–E3, High magnification views of mCherry immunoreactive fibers in three deeper brain regions: the Put, THA, and Amy. The AAV injection site in the S2 digit region is depicted in A. J, K, Overlays of blue laser-evoked fMRI activations (blue patches) and mCherry expression (pink patches) obtained from monkeys A and B. Brain structure outlines were modified from images in A Stereotaxic Atlas of the Squirrel Monkey's brain, Public Health Service Publication NO 933, 1962.
At the conclusion of fMRI and electrophysiology data acquisition, we performed postmortem immunohistological staining of the brain tissue to validate the successful expression of AAV vectors and delineated anatomical tracer connections in the two animals studied. The uptake of the AAV and expression of mCherry fluorescent tag were very strong at the injection site in the S2 hand region. Additionally, the coronal histological sections of the brain revealed widespread fluorophore-positive neurons and axons in many brain regions, as shown in Figure 6, D and I. Importantly, the cortical and subcortical regions exhibiting robust opto-fMRI signals also showed fluorophore uptake, indicating strong anatomical tracer connections with the S2 cortex where the AAV tracer was injected. Zoomed-in images highlight axons with fluorescence expression in key regions, including the thalamus (THA), Put, and Amy (Fig. 6E1–E3). There was strong spatial correspondence between the S2 opto-fMRI activation map and the S2 anatomical tracer connectivity map (Fig. 6J,K).
Discussion
Neurovascular relationship during direct manipulation of excitatory neuron activity
Our findings not only demonstrated the robustness of opto-fMRI of excitatory neurons in the S2 of squirrel monkeys but also revealed a disassociated dose–response relationship between opto-fMRI, LFP, and spiking activity. This relationship differs from those observed in the normal cortex (Wang et al., 2013b) but is similar to those seen in the deafferented cortex, such as after sensory input deprivation due to traumatic spinal cord injury (Wu et al., 2022b). Taken together, our findings using opto-fMRI of excitatory neurons, both under normal conditions and with sensory input deprivation, support the notion that the relationship between fMRI signals and underlying electrophysiology activity—whether at the individual neuron level or as a population—is significantly altered when specific neuron types are selectively manipulated by optogenetics or affected by lesions. These results highlight the need for caution when using fMRI signals to infer underlying pathological processing at the neuronal level in both animal and human fMRI studies.
Unlike natural peripheral stimuli, such as tactile stimulation of body parts or electrical stimulation of the cortex, which activate both excitatory and inhibitory neurons and axons, optogenetic fMRI allows for the selective activation of primarily one type of neuron—excitatory neurons, which are responsible for interareal communication in this study. To help interpret the opto-fMRI results, we characterized the dose (laser power) versus response curves of opto-fMRI, LFP, and spiking signals in the S2 cortex and examined the correlation strengths between changes in these three signals. We found that changes in excitatory neuron opto-fMRI signals strongly correlated with LFP signals, whereas corresponding changes in spiking activity were not detected at lower laser power levels (1, 2, 4, and 8 mW; Fig. 5).
When delivered into the brain by AAV vectors with the CaMKIIα promoter, ∼93% of blue light-responsive neurons were determined to be putative excitatory neurons, based on spike width analysis, consistent with previous reports (Scheyltjens et al., 2015). We conclude that opto-fMRI signals reflect underlying neuronal activity at the population level, as measured by LFP. The relationship between opto-fMRI signal amplitude and spiking rates suggests that single-neuron activity strength is not the primary contributor to increases in opto-fMRI signals. The success of this opto-fMRI study opens the door for future research aimed at identifying the fundamental networks involved in integrating tactile and nociceptive inputs and exploring how disruptions in the balance between excitatory and inhibitory activity contribute to abnormal somatosensory and nociceptive behavior.
Differences in LFP and spiking activity of S2 cortical neurons associated with natural tactile process and optogenetic stimulation
In the somatosensory S2 cortex, excitatory neurons primarily function as conduits for relaying, processing, and integrating sensory information, while inhibitory neurons play a critical role in modulating and shaping sensory signals. The proportion of excitatory (86%) and inhibitory (14%) neurons sampled in the S2 cortex during tactile information processing closely mirrors that observed in the rat's medial prefrontal cortex (Gabbott et al., 1997) and generally aligns with typical cortical neuron populations (Markram et al., 2004; Rudy et al., 2011; Hayashi et al., 2018). The response onset to blue light stimulation was significantly faster than that of tactile stimulation, suggesting more immediate activation of excitatory neurons. Examining response latencies across multiple brain regions could provide valuable insights into the causal connections of the S2 cortex in future studies.
Similarly, blue light-evoked LFP signals were robust, with amplitudes increasing gradually as laser power increased, presumably leading to the activation of either a greater number of excitatory neurons or stronger spiking activity. These LFP signal changes associated with excitatory neuron activation differed from those elicited by natural tactile stimulation. These electrophysiological features confirm that the opto-fMRI signals detected in the S2 and across the brain are indeed associated with the selective activation of excitatory neurons in this region (Fenno et al., 2011; Deisseroth et al., 2015).
Neuron-type-specific structural and functional relationships of the S2 cortex and beyond
We and others have previously demonstrated that the S2 cortex functions as a hub, connecting somatosensory cortices to other brain networks during the processing of both tactile and painful information (Taub et al., 2024). In this study, we showed that selectively activating S2 excitatory neurons leads to brain-wide fMRI responses, indicating that the S2 cortex is casually and functionally connected to numerous brain regions with distinct roles. For example, S2 cortex activation was directly associated with fMRI responses in the MCC, ACC, and subcortical regions such as the Put. Previous studies in macaque monkeys have provided evidence of S2 connections with the cingulate cortex (Morecraft et al., 2004, 2012) and the Put (Yeterian and Pandya, 1993).
In our experimental setting, we were able to directly link the brain regions exhibiting opto-fMRI signals with the functional roles of the S2 cortex, as AAV vectors were injected into a functionally defined hand region. This hand region contains clusters of low-threshold tactile and high-threshold thermal nociceptive neurons, so the brain regions showing opto-fMRI signals are involved in the processing and integration of touch and pain (Ye et al., 2021). Moreover, AAV tracer uptake was detected in axons within these brain regions showing strong opto-fMRI signals, confirming their extensive anatomical connections to the S2 cortex (Fig. 6). Many of these regions are not classical somatosensory regions but are involved in higher-order brain functions. Our data provide the first evidence supporting the direct connection of the S2 cortex to networks engaged in distinct functions such as perception and cognition.
It is worth noting that AAV uptake may have included nearby regions, such as the insular cortex, which has known connections with S2 in macaque monkeys, and projections to cognitive and emotional processing regions, including the Amy, as part of a “corticolimbic pathway for touch” (Friedman et al., 1986). These regions are affected in patients with impaired touch perception (Preusser et al., 2015) and support a model of ventral stream of somatosensory processing (Dijkerman and de Haan, 2007). This strong correspondence between opto-fMRI and tracer-based anatomical networks of the S2 cortex supports the anatomical basis of opto-fMRI networks of the S2 cortex (Wu et al., 2017).
At the circuit level, opsin expression pattern illustrated a brain-wide connection to the S2 cortex, including in S1, M1, the insula, and VPL, which is largely consistent with known anatomical connections between S2 and area 3b (Liao et al., 2013), area 1 (Cerkevich and Kaas, 2019), area 3a (Guldin et al., 1992), M1 (Gao et al., 2013), the insula, and the THA (Stevens et al., 1993). For reviews, see Delhaye et al. (2018). A widely accepted notion is that the therapeutic effects of neuromodulation arise from modifying dysfunctional neural circuits, either by compensating for or restoring lost brain functions, repairing disrupted functional connections, or leveraging homeostatic plasticity within affected brain networks (Krack et al., 2010; Edwardson et al., 2013; Lozano et al., 2019). Given the critical role of the S2 cortex within the primate nociceptive network and its involvement in chronic pain (Wager et al., 2013; Wu et al., 2017, 2022a; Weber Ii et al., 2019; Zheng et al., 2020; Ye et al., 2021), the S2 cortex may serve as a potential therapeutic target for pain relief through neuromodulation interventions (Kuroda et al., 2001; Kong et al., 2024).
Optogenetic fMRI in nonhuman primates
The growing use of optogenetics in NHPs has significantly advanced our understanding of brain function and holds promise for developing therapeutic interventions for human disorders (O'Shea et al., 2018; Tremblay et al., 2020). This approach has been especially valuable for dissecting the neural circuits underlying complex behavior and providing new insights into the brain's functional architecture—from specific neurons in particular brain regions to large-scale functional and anatomical connections. Combined opto-fMRI and intracranial electrophysiology enabled causally manipulating defined excitatory neurons in the S2 cortex to observe the resulting changes in brain-wide BOLD signals and improved our understanding of the relationship between opto-fMRI, LFP, and the spiking activity of excitatory neurons.
Our findings suggest that the observed opto-fMRI signals are driven by excitatory neurons, as both spike waveform analysis and postmortem immunohistology confirmed the successful expression of opsin in the S2 cortex neurons. The robust opsin expression, observed both at the injection site and in downstream regions (Fig. 6), closely resembles the expression patterns reported in the S1 and M1 cortex of the same species (O'Shea et al., 2018). This demonstrates the reliability of our approach and its potential for future preclinical and clinical studies of primate sensorimotor behavior and brain circuits.
One limitation of the study is its sample size—four NHP animals and five hemispheres—which is relatively small compared with typical human and small-animal studies due to ethical and economic constraints. To address this challenge, we maximized statistical power by increasing the fMRI and electrophysiology data sampling from each individual subject and employing a mixed-effects model that accounts for both within- and between-subject variables.
Conclusion
This study represents the first successful application of a combined fMRI, electrophysiology, and optogenetic approach to selectively stimulate excitatory neurons within the S2 cortex of monkeys. Our results reveal clear differential relationships between increased optical power and changes in BOLD, LFP, and spiking activity, underscoring the unique underlying properties of excitatory neuron-driven opto-fMRI signals. Additionally, we established a strong spatial correspondence between brain-wide opto-fMRI activity and anatomical tracer-based connections in the S2 hand region. These findings offer novel insights into the electrophysiological underpinnings of excitatory neuron-dominated opto-fMRI signals and the structure–function relationship of S2 circuits.
The study makes three key contributions: First, it establishes a strong correlative relationship between opto-fMRI and LFP signals in populations of a specific type of neuron (excitatory neurons) in the NHP cortex. Second, we achieved precise, neuron-type-specific activation of excitatory neurons through optogenetic stimulation, enabling the dissection of causal connections in the primate S2 circuits. This advancement opens new avenues for future research into neuron-type-specific microcircuitry and large-scale cortical networks using opto-fMRI. Third, our findings provide both electrophysiological and fMRI evidence supporting the pivotal role of the S2 cortex in connecting functionally distinct brain networks for the integration and processing of touch and pain-related information. The translational relevance of this work lies in its potential to inform the development of targeted neuromodulation therapies for neurological and psychiatric disorders in humans. The success of this technique in NHPs, combined with advancements in noninvasive AAV delivery tools, highlights promising future clinical applications for treating related brain disorders.
Data Availability
The lead contact will share all data, documentation, and code reported in this paper upon request.
Footnotes
This study was funded by NIH NINDS Grant 2R01NS078680. We thank Chaohui Tang for her assistance with animal care and data acquisition, and we acknowledge the valuable contributions of Huixin Qi and Laura Trace to the immunohistology work.
↵*J.C.G and L.M.C. equally directed the study.
The authors declare no competing financial interests.
- Correspondence should be addressed to Li Min Chen at limin.chen{at}vanderbilt.edu.
