Causal oscillations in the visual thalamo-cortical network in sustained attention in ferrets

Summary:

Sustained visual attention allows us to process and react to unpredictable, behaviorally relevant sensory input. Sustained attention engages communication between the higher-order visual thalamus and its connected cortical regions. However, it remains unclear if there is a causal relationship between oscillatory circuit dynamics and attentional behavior in these thalamo-cortical circuits. By using rhythmic optogenetic stimulation in the ferret, we provide causal evidence that higher-order visual thalamus coordinates thalamo-cortical and cortico-cortical functional connectivity during sustained attention via spike-field phase locking. Increasing theta but not alpha power in the thalamus improved accuracy and reduced omission rates in a sustained attention task. Further, the enhancement of effective connectivity by stimulation was correlated with improved behavioral performance. Our work demonstrates a potential circuit-level causal mechanism for how the higher-order visual thalamus modulates cortical communication through rhythmic synchronization during sustained attention.

eTOC:

Huang et al. demonstrate that visual thalamic theta oscillations causally coordinate the posterior visual network to enhance sustained attention. This coordination is frequency and state-dependent.

Introduction

Sustained attention is crucial for survival and learning, from capturing errors in routine work to responding to unexpected threats ¹. This process allows the brain to allocate cognitive resources to respond to infrequent, but task-relevant stimuli over an extended period of time ². Dysfunction of this process is linked to attention deficit and impulsivity in many psychiatric disorders ^3–8. Unlike bottom-up attention driven by the presentation of a stimulus, sustained attention is a form of attentional control that stresses the maintenance of focus in anticipation of a goal-related stimulus before its presentation ⁹. This process requires the coordination of sensory cortices and higher-order cortical areas for top-down control and sensory gating ¹⁰.

A candidate circuit for this top-down control of sustained attention is the higher-order thalamo-cortical visual circuit ^11–13. Cortical oscillations have been widely implicated in attention, especially as a top-down control signal generated in the fronto-parietal network ^14–16. Such cortico-cortical synchronization may be driven by subcortical structures such as the higher-order visual thalamus which mediates sensory gating and attentional modulation. Anatomically, the primate pulvinar has highly specialized subdivisions that are reciprocally connected to the visual cortex, superior colliculus, prefrontal cortex, parietal association cortex, and temporal lobe ¹⁷. Thus the pulvinar is ideally situated to act as a major hub and potential synchronizer of cortical activity. Functionally, the pulvinar serves an important role as a secondary visual system ¹⁸ and coordinates cortical activity during visual attention behavior ^{11–13,19,20}. In primates, pulvinar activity was found to help predict visual responsiveness and high-frequency synchronous activity in V4 (a mid-level processing area along the ventral visual pathway) ²¹ and was reported to synchronize the activity of interconnected V4 and temporo-occipital cortex during the delay period in a visuospatial attention task ²⁰. Moreover, structural abnormalities in the pulvinar were found to be related to atypical pulvinar-cortical functional connectivity and attention deficits in people with ADHD ²². Together, these studies suggest that thalamic oscillations coordinate and modulate cortical activity during attentional tasks. However, it remains unclear to what extent pulvinar activation and synchronization with cortical areas represents an epiphenomenon or a causal mechanism. Here, we hypothesized that higher-order thalamic oscillations modulate thalamo-cortical and cortico-cortical communication in a frequency-specific way to causally facilitate sustained visual attention.

To test this hypothesis, we combined frequency-specific optogenetic targeting of rhythmic activity in the higher-order visual thalamo-cortical network with multisite electrophysiology in ferrets performing a sustained attention task. We found thalamic theta oscillations and functional connectivity with the cortex to be associated with sustained attention. Optogenetic stimulation of the higher-order visual thalamus modulated not only rhythmic thalamic activity but also thalamo-cortical and cortico-cortical functional interactions when compared to sham stimulation. Importantly, optogenetic stimulation in theta but not alpha frequency improved task performance, and this improvement correlated with stimulation-induced enhancement of theta oscillation in the posterior visual thalamo-cortical network. In summary, rhythmic enhancement of the higher-order visual thalamic activity increased frequency-specific communication in the posterior network and improved sustained attention performance by reducing attentional lapses.

Results

Identification of the oscillatory features associated with sustained attention

To probe the network substrate of sustained attention, we employed the five-choice serial reaction time task (5-CSRTT) ²³ in ferrets ¹⁵. This task was modified from the human continuous performance task (CPT) which is used to clinically measure sustained attention performance ^24–26 in psychiatric disorders ^3–8. In the 5-CSRTT, animals initiate trials at an infrared sensor, then after a variable delay period (4, 5, or 6 sec), a light stimulus is presented randomly in one of the five touchscreen windows. If the animal touches the correct window during stimulus presentation or thereafter within a 2-second hold period, a tone is played and the animal can retrieve a water reward at the lick spout and subsequently start a new trial. Otherwise, a premature touch (before stimulus onset), an incorrect touch (the wrong window was touched), or an omission (no touch was made) all lead to a 5-second timeout. (Figure 1A) To contrast low versus high attentional demand, we introduced long (2 sec) versus short (1 sec) stimulus presentation durations. We recorded electrophysiology signals from three nodes of the posterior thalamo-cortical visual network: the lateral posterior nucleus/pulvinar complex (LP/Pul), posterior parietal cortex (PPC), and visual cortex (VC). The signals were processed to yield local field potentials (LFP) and single-unit activity (Figure 1C example spike waveforms and LFP traces).

Figure 1. Sustained attention task in freely-moving ferrets and histological verification of anatomical connections and recording sites.

See also Figure S1, Table S1. (A) Illustration of the 5-CSRTT timeline and the behavioral equipment setup. Animals sustained attention during a variable delay period in anticipation of a brief presentation of a visual stimulus in one of five spatial locations. Touching the correct stimulus window resulted in reward delivery. (B) Tracing studies revealed structural connectivity between LP/Pul, PPC, and VC. The middle panel depicts the location of PPC and area 17 (VC) in the atlas brain and summarizes the reciprocal connectivity between the three regions. Left: Injection of retrograde tracer (CTB 488) reveals PPC input from LP/Pul and VC. Right: Anterograde virus AAV-CaMKII-GFP into PPC reveals projections to LP/Pul and VC. (C) Example sections with Nissl staining of studied regions show electrode implantation location in LP/Pul, PPC, and VC. Example single-unit waveforms as well as raw traces of local field potentials for the delay period in the three studied regions from a recording session without optogenetic stimulation. For LP/Pul, insets show a zoomed-in view of Nissl staining (left) and rAAV5-CaMKII-ChR2-mCherry expression (right) of LP/Pul.

We focused on LP/Pul, PPC, and VC in the posterior network due to their anatomical connectivity which provides a foundation for direct functional interaction. Anatomical connections between these regions were verified in a separate set of animals which showed anatomically reciprocal connections between LP/Pul, PPC, and VC (Figure 1B). The reciprocal connectivity between LP/Pul and VC was shown in a previous study ²⁷. The electrode implantation locations, stimulation fiber sites, and viral expression were histologically verified after the experiment (Figure 1C). Black squares in Figure 1C denote the implant locations indicated either by tissue damage (in PPC and VC) or electrode tracks (in LP/Pul). Histological verification of the viral injection location and electrode implantation location in LP/Pul are listed in Table S1. Only electrodes that were verified to be in the targeted location were used for the analysis.

Task-modulated intra- and inter-regional neural activity

Oscillations represent a mechanism for neural populations to communicate ²⁸, especially during attentional tasks where rhythmic sampling and gating of the sensory input are essential ^10,11. First, we identified the oscillatory features associated with performance during the 5-CSRTT task (N=85 sessions from 4 animals). Trials were aligned to trial initiation (“0”) and visual stimulus onset (“Stim”), respectively (Figure 2). The spectrogram of the local field potential (Figure 2A) showed distinct theta, alpha, and gamma bands. Baseline ([−2, −1] s before trial initiation) normalized spectrogram showed time-frequency clusters that were significantly different from baseline (p < 0.05 via permutation test, see Method: Permutation testing) which expanded from trial initiation throughout the entire delay period (black contours in Figure 2B). In general, after trial initiation, there was an increase in theta band power in LP/Pul and PPC (for better visualization of the distribution of data: Figure 2C, specifically, LP/Pul: Baseline = −0.29 vs. Delay = 2.71, P < 0.001, PPC: Baseline = −0.13 vs. Delay = 1.14, P < 0.001, VC: Baseline = 0.16 vs. Delay = −0.39, P = 0.331, 1-way ANOVA with Holm–Bonferroni correction for multiple comparison of 3 regions; individual animal data in Figure S1A) as well as a shift in frequency (from ~4 Hz to ~5.5 Hz) in the posterior network (specifically, LP/Pul: from 3–4.5 Hz to 4.5–7.5 Hz, PPC: from 3.5–4.5 Hz to 4.5–6.6 Hz, and VC: from 3–5 Hz to 5–7 Hz). This suggested that theta oscillations were associated with initiating and sustaining attention. In addition, we observed a significant cluster for increased gamma-band power across all 3 regions (black contours in Figure 2B: 45–80 Hz in LP/Pul, 43–128 Hz in PPC, 45–128 Hz in VC) which confirmed that these regions were functionally engaged during the sustained attention period. There was also a significant power decrease in the alpha frequency band around 16 Hz in the posterior cortical regions (black contours in Figure 2B: 7–27 Hz in PPC, and 8–30 Hz in VC), which is consistent with the role of alpha oscillation in modulating neural activity during visual-attention tasks ^29,30.

Figure 2. Task-modulated power and functional connectivity in the network during 5-CSRTT.

See also Figure S2. (A) Power spectrogram during the task, aligned to trial initiation “0” and visual stimulus onset “Stim”, respectively, for studied regions: LP/Pul, PPC, and VC. (B) Same representation but for baseline-normalized power spectrogram. (C) Comparison of the session means of normalized power in the theta (4–7Hz), alpha (14–18 Hz), and gamma (40–75 Hz) frequency band during the baseline period and the delay duration (i.e. sustained attention period) for each studied region. (D) Baseline normalized phase-locking value (PLV) between studied region-pairs during the task. (E) Comparison of the session means of PLV in the theta, alpha, and gamma frequency band during the baseline period and the delay duration for each studied region-pair. N=85 sessions from 4 animals. For (B) and (D), black contours delineate statistically significant spectral features compared to baseline activity via permutation testing (number of iterations = 1000). For (C) and (E), P-values were computed from a 2-tailed paired t-test comparing 2 populations and then corrected for multiple comparisons of regions or region-pairs. The error bar represents the standard error of the mean. Each dot represents one session. * P < 0.05, ** P < 0.01.

Given the concurrent change in theta oscillations in LP/Pul and PPC, we investigated synchrony between these regions with the phase-locking value (PLV) which measures the phase consistency across trials between different brain regions. Similar to the spectrogram, the PLV was also normalized against the baseline period ([−2, −1] s before trial initiation). Consistent with the enhanced theta oscillation in each posterior region (Figure 2, A and B), we found significant enhancement in inter-regional functional connectivity after trial initiation, measured by PLV (Figure 2, D and E, LP/Pul-PPC: Baseline = 0.71 vs. Delay = 0.78, P < 0.001, LP/Pul-VC: Baseline = 0.64 vs. Delay = 0.73, P < 0.001, PPC-VC: Baseline = 0.64 vs. Delay = 0.74, P < 0.001, 1-way ANOVA with Holm–Bonferroni correction for multiple comparison of 3 regions; individual animal data in Figure S1B), accompanied by an increase in the theta frequency (from 3–4.5 Hz to 4.5–7.5 Hz). In addition, there was significant alpha suppression from trial initiation in PPC-VC around 16 Hz (Figure 2D), consistent with the alpha suppression observed in PPC and VC individually (Figure 2, A and B).

Attentional demand modulates inter-regional neural synchrony

To further isolate the effect of attention, we tested the effects of varying the attentional demand in easy versus hard sessions (N = 42 easy and N = 42 hard randomly interleaved sessions from 3 animals). This was operationalized by adjusting the stimulus duration, where shorter stimulus durations (i.e. hard sessions) required animals to sustain attention more consistently in order to successfully complete a trial, and longer stimulus durations (i.e. easy sessions) required less attentional effort (Figure 3A). Hard sessions were associated with significantly lower accuracy (mean: Easy = 72% vs. Hard = 59%, P < 0.001 after Holm–Bonferroni correction for 4 behavioral outcomes) which mainly resulted from an increased omission rate in hard sessions (mean: Easy = 17% vs. Hard = 25%, P = 0.002 after Holm–Bonferroni correction for 4 behavioral outcomes) (Figure 3B). In addition, hard sessions were associated with shorter reaction time for correct trials (median: Easy = 1.14 s vs. Hard= 0.95 s, P < 0.001), indicating the successful recruitment of additional attentional effort (Figure 3C). In addition to the previous analyses using PLV, we examined the directionality of the functional interaction in the circuit when more attentional resources were recruited (hard vs. easy level). Strikingly, only LP/Pul showed significantly increased drive for the hard level compared to the easy level measured by conditional Granger Causality (CGC) (Figure 3D: LP/Pul->VC: median: Easy = 0.064 vs. Hard = 0.169, P < 0.001), which computes directed information flow between two regions while controlling for other regions studied (Figure 3E dark inter-regional arrow) (LP/Pul->VC showed a significant cluster around 5 Hz, but not in the other direction). Thus the functional interaction in the thalamo-cortical network that was associated with higher attentional demand was mainly driven by LP/Pul and not the cortical regions (PPC and VC). This supports a specific role of thalamic theta oscillations in sustained attention and illustrates the central role of inter-regional neural synchrony orchestrated by the thalamus.

Figure 3. Task-difficulty modulated change in directional functional connectivity.

(A) Experimental timeline for easy versus hard sessions. (B) Comparison of trial outcomes (percentage of correct, premature, omission, and incorrect trials) between easy (light green) and hard (dark green) sessions. (C) Comparison of reaction time for correct trials between easy (light green) and hard (dark green) sessions. (D) Directed functional connectivity determined by conditional Granger causality (CGC) between studied region-pairs for hard versus easy conditions, revealing theta drive from LP/Pul to cortical regions was modulated by attentional demand. (E) Comparison of session means of CGC in the theta frequency band (4.5–7.5 Hz) during the delay duration between easy (light green) and hard (dark green) sessions. N=42 easy, 42 hard sessions from 3 animals. For (H), black contours delineate statistically significant spectral features of easy vs. hard conditions via permutation testing (number of iterations = 1000). For (E), P-values were computed from a 2-tailed paired t-test comparing 2 populations and then corrected for multiple comparisons of regions or region-pairs. For (B) and (C), the P-values were calculated from a linear mixed effect model: PercentTrialOutcome (B) or ReactionTime (C) = 1 + EasyHardContrast + (1+EasyHardContrast|AnimalID) controlling for the random effect of different animals, and then corrected for multiple-comparisons for four trial outcome types using Holm–Bonferroni method. The error bar represents the standard error of the mean. Each dot represents one session. * P < 0.05, ** P < 0.01.

Causal engagement of the attention-related oscillatory features by optogenetics

Optogenetic stimulation entrains single-unit firing in a frequency-specific way

Thus far, we demonstrated that higher attentional demand (Figure 3E) was associated with higher theta drive from LP/Pul to the receiving cortical region VC. To test whether the thalamic theta enhances attention instead of just being a byproduct of the process, we implemented optogenetic stimulation by injecting rAAV5-CaMKII-ChR2-mCherry in LP/Pul and stimulating LP/Pul using an optrode as well as recording from all three regions during the last 3 seconds of the sustained attention (i.e. delay) period before the presentation of the visual stimulus for N = 51 sessions from 4 animals (Figure 4A). In each trial, optogenetic stimulation was applied randomly in one of three ways: the individualized task-relevant theta or alpha frequency bands, or a “no light” sham control. The verification of viral expression and implantation location is shown in Figure 1C. First, we verified that the single unit (SU) activity was entrained in LP/Pul and PPC (with less frequency-specificity in VC) by optogenetic stimulation in LP/Pul (Figure 4, B–D, example SU, and G), which demonstrated that the optogenetic stimulation successfully activated both the target and its connected region. One example SU waveform per region is shown (Figure 4B). The raster plot (Figure 4C) and trial-averaged peristimulus time histogram (PSTH; Figure 4D) of these example units showed more rhythmic organization of the spiking activity during theta (blue) and alpha (red) optogenetic stimulation conditions than sham (black), especially in LP/Pul (see Method: optical tagging test ³¹, theta vs. sham: P < 0.001, alpha vs. sham: P < 0.001) and in PPC (theta vs. sham: P < 0.001, alpha vs. sham: P < 0.001). Further, to investigate the SU entrainment at a population level, we averaged the PSTHs across all single units recorded during all sessions from the four animals (Figure 4E). To quantify the entrainment, we computed the Fourier transform of the baseline normalized PSTH (zPSTH) of SU during the optogenetic time window and found entrainment of spiking activity at the simulation frequencies (see peaks in spectra at around 5.6Hz and 16Hz) both in LP/Pul and PPC (Figure 4F). After averaging the spectra power across the studied frequency bands (theta and alpha), we computed the percentage of SUs with increased theta and alpha entrainment compared to an unaffected frequency-band baseline (see Method: Data analysis). Consistent with our hypothesis, there were significantly more SU entrained in the theta frequency band for theta stimulation compared to alpha stimulation or sham conditions in LP/Pul, PPC, and VC (LP/Pul: percentage of entrained SU under optogenetic stimulation at theta = 30%, alpha = 0%, sham = 1.3%; theta vs. sham: P < 0.001; theta vs. alpha: P < 0.001; PPC: percentage of entrained SU under optogenetic stimulation at theta = 11.2%, alpha = 2.9%, and sham = 1%; theta vs. sham: P < 0.001; theta vs. alpha P < 0.001; VC: percentage of entrained SU under optogenetic stimulation at theta = 9%, alpha = 1.1%, and sham = 0.6%; theta vs. sham: P = 0.002; theta vs. alpha P = 0.006, X² test with Holm–Bonferroni correction for multiple comparison of 2 frequency bands, 3 optogenetic conditions, and 3 regions). Similarly, there were significantly more SU entrained in the alpha frequency band for the alpha optogenetic stimulation condition compared to sham conditions in LP/Pul and PPC (LP/Pul: percentage of entrained SU under optogenetic stimulation at alpha = 13.7% and sham = 1.3%; alpha vs. sham: P < 0.001; PPC: percentage of entrained SU under optogenetic stimulation at alpha = 13.5% and sham = 0.4%; alpha vs. sham: P < 0.001) (Figure 4G). Interestingly, during theta stimulation, there were also significantly more SU entrained in the alpha frequency band in LP/Pul and VC, suggesting potential theta-alpha coupling mechanism in the circuit (LP/Pul: percentage of entrained SU under optogenetic stimulation at theta = 9.3% and sham = 1.3%; theta vs. sham: P = 0.002; VC: percentage of entrained SU under optogenetic stimulation at Theta = 7.3%, sham = 0.6%; Theta vs. Sham: P = 0.009). In summary, we confirmed that SU activity was entrained by optogenetic stimulation in a frequency-specific way in both the stimulated region LP/Pul as well as non-stimulated cortical regions connected to LP/Pul.

Figure 4. Frequency-specific entrainment of local single-unit activity and oscillatory features by optogenetic stimulation.

See also Figure S3, Figure S5. (A) Experimental timeline with optogenetic stimulation in the blue bar, for easy level only. The (B) waveform, (C) raster plot, (D) baseline normalized firing rate from an example single unit (from 1 session, Animal A) in each region: LP/Pul, PPC, and VC. (E) Average baseline normalized firing rate from all single units in each region (mean±sem). (F) Spectra of baseline normalized firing rate during optogenetic stimulation period ([−3, 0] s to visual stimulus onset) for theta (blue), alpha (red), and sham (black) optogenetic stimulation conditions (mean±sem). Lightly colored shadings represent sem. (G) Percentage of significantly entrained SU in theta (first row) and alpha (second row) frequency bands in each region. (H) Power spectrogram of the LFP for each region from −4 to 2 seconds around visual stimulation, contrasting individualized theta-band stimulation (first row) or alpha-band stimulation (second row) against the sham condition. (I) Average of theta (first row) and alpha (second row) frequency band power across the optogenetic stimulation period in theta (blue) and alpha (red) optogenetic stimulation condition compared to sham. Each dot represents one session. N = 51 sessions from 4 animals. The normalization baseline in D-G was chosen to be −5 to −3.5 seconds before laser stimulation onset to visualize the optogenetic effect. * P < 0.05, ** P < 0.01.

Optogenetic stimulation entrains oscillations in the posterior network

We next showed that this single-unit entrainment related to the LFP network signals by investigating the effect of optogenetic stimulation on the mesoscopic oscillatory features. We contrasted the power spectrogram during theta or alpha (Figure 4H) optogenetic stimulation to the spectrogram during the sham condition (i.e. theta-sham and alpha-sham) and averaged the theta and alpha band power across the 3-second optogenetic stimulation period (Figure 4I). For theta stimulation, there was a significant enhancement of theta oscillation in LP/Pul (Figure 4H first row and Figure 4I first row, 5.3–7 Hz, Theta-Sham = 1.03, P < 0.001, 2-tailed t-test with Holm–Bonferroni correction for multiple comparisons of 2 frequency bands, 2 optogenetic contrasts, and 3 regions), PPC (5–7 Hz, Theta-Sham = 3.65, P < 0.001), and VC (Theta-Sham = 0.39, P = 0.009) that was time-locked to the optogenetic stimulation window. The non-targeted alpha frequency band (close to the 2^nd and 3^rd harmonic of theta stimulation frequency) was also significantly enhanced (Figure 4I, blue bars in the second row: alpha frequency band LP/Pul: Theta-Sham = 1.95, P < 0.001, PPC: Theta-Sham = 2.16, P < 0.001, and VC: Theta-Sham = 0.78, P = 0.002). During alpha stimulation, as expected, we saw a distinct enhancement of alpha oscillations in all 3 regions in the posterior network (Figure 4H, second row and Figure 4I, red bars in the second row: alpha frequency band LP/Pul: Alpha-Sham = 4.5, P < 0.001, PPC: Alpha-Sham = 5.28, P < 0.001, and VC: Alpha-Sham = 3.21, P < 0.001, all centered around 16 Hz) comparing to sham, confirming the frequency-dependent optogenetic enhancement. The power increase in VC under alpha stimulation demonstrated that it is possible to enhance power (on the LFP level) without changing the firing rate (Figure 4F right-most bar). Although the entrainment was mainly limited to the simulation frequency, there were also power increases in non-targeted bands. We next investigated which enhanced oscillatory features engaged circuit-level synchrony and if there were frequency-specific effects on functional connectivity.

Optogenetic stimulation induces directed and frequency-specific information flow from LP/Pul to the posterior network

To examine how optogenetically-induced changes in oscillations altered inter-regional communication, we computed single-unit phase-locking values (SUPLV) between studied region pairs. During theta stimulation (Figure 5A, Theta-Sham), there was significant phase-locking at the theta frequency band (centered around 5.5 Hz) driven by LP/Pul to PPC, and VC. During alpha stimulation (Figure 5A, Alpha-Sham), there was a strong phase-locking at the alpha frequency band driven by LP/Pul to all cortical regions (all centered around 16 Hz). In addition, there was also feedback communication from PPC to LP/Pul and between PPC and VC, indicating that alpha stimulation engaged the posterior network. This was consistent with the finding that alpha oscillations dominated in the posterior network before trial initiation (Figure 2A: alpha oscillations before trial initiation in PPC and VC, which was also shown as an alpha suppression after trial initiation in baseline normalized spectrogram and PLV in Figure 2 B and C). Further quantification of this enhancement during the 3-second optogenetic stimulation period showed frequency-specific enhancement of the thalamo-cortical SUPLV (Figure 5B, LP/Pul->PPC Theta-Sham = 0.086, P < 0.001; Alpha-Sham = 0.035, P = 0.002; LP/Pul->VC, Theta-Sham = 0.011, P = 0.045; Alpha-Sham = 0.019, P = 0.050 2-tailed t-test with Holm–Bonferroni correction for multiple comparisons of 2 frequency bands, 2 optogenetic contrasts, and 6 directed region pairs). Interestingly, the theta enhancement in SUPLV during theta optogenetic stimulation was not observed when we only looked at inter-regional functional connectivity based on the LFP (Figure S2). In summary, optogenetic stimulation in LP/Pul in theta or alpha frequency bands induced inter-regional thalamo-cortical spike phase locking in theta or alpha frequency bands, respectively.

Figure 5. Frequency-specific entrainment of cortico-thalamo-cortical functional connectivity by optogenetic stimulation.

See also Figure S6 for hard sessions. (A) The single-unit phase-locking value between studied region pairs from −4 to 2 seconds around visual stimulation, contrasting theta frequency laser stimulation or alpha frequency laser stimulation against the sham condition. (B) Average of single-unit phase-locking value in the theta and alpha frequency band across the optogenetic stimulation period in theta (blue) and alpha (red) optogenetic stimulation condition compared to sham. Each dot represents one session. N = 51 sessions from 4 animals. Black contours delineate statistically significant differences in spectral features between corresponding optogenetic conditions and sham conditions via permutation testing (number of iterations = 1000). Background arrows in black (in contrast to gray) indicate significant connections. Arrows point from the region of single-unit to the region of LFP (i.e. A->B indicates A’s spikes phase-locking to B’s LFP). An identical number of action potentials were used for all connections. Single units with low counts of action potentials were excluded. Different signal-to-noise level is the result of fewer single units in VC. * P < 0.05, ** P < 0.01.

Causal manipulation of sustained attention performance

Theta optogenetic stimulation increases correct response and decreases omission

To determine the effect of optogenetic stimulation on sustained attention and to demonstrate a causal brain-behavior relationship, we quantified how frequency-specific optogenetic stimulation changed behavioral performance during the 5-CSRTT. We determined the change in the percentage of different trial outcomes (e.g. correct, premature, omission, and incorrect) in each optogenetic stimulation condition compared to the sham condition (Figure 6A). Theta stimulation significantly increased the percentage of correct trials (Theta vs. Sham = 77.8 – 70.1 = 7.1 percentage points, P = 0.0038; Alpha vs. Sham = 73.5 – 70.1 = 2.8 percentage points, P = 0.47, after Holm–Bonferroni correction for 4 behavioral outcomes) and reduced the percentage of omission trials (Theta vs. Sham = 12.9 – 19.2 = −6.3 percentage points, P < 0.001; Alpha vs. Sham = 18.4 – 19.2 = −0.8 percentage points, P = 0.85). Both observations were consistent across all 4 animals (for individual results see shaded bars in Figure S4). There was no significant change in premature responses trials (Theta vs. Sham = 6.5 – 6.1 = 0.4 percentage points, P = 0.78; Alpha vs. Sham = 5.2 – 6.1 = −0.9 percentage points, P = 0.48) or incorrect responses trials (Theta vs. Sham = 2.8 – 3.9 = −1.1 percentage points, P = 0.23; Alpha vs. Sham = 2.8 – 3.9 = −1.1 percentage points, P = 0.22). However, there was no significant effect of alpha-frequency optogenetic stimulation on sustained attention. In addition, either theta or alpha optogenetic stimulation did not alter incorrect touches and premature touches, which are an indicator of impulsivity. To test if optogenetic stimulation affected motor function or motivation level during the optogenetic stimulation period, reaction period, and reward retrieval period, we compared the animal velocity and reaction time under each optogenetic condition (Figure 6D). It turned out that optogenetic stimulation did not affect average velocity during optogenetic stimulation or velocity during reward retrieval (V_Opto: Theta vs. Sham = 0.118, P = 0.798, Alpha vs. Sham = −0.58, P = 0.210; V_Retrieval: Theta vs. Sham = −0.277, P = 0.458, Alpha vs. Sham = 0.602, P = 0.847) (Figure 6B) or reaction time in correct trials (Theta vs. Sham = −0.028, P = 0.140, Alpha vs. Sham = −0.069, P = 0.214) (Figure 6C). This suggests that theta-frequency optogenetic stimulation reduced attentional lapses (i.e. omission) and shifted the system towards a more attentive state with higher accuracy without affecting motor function, arousal, or motivation level.

Figure 6. Behavioral results of optogenetic stimulation in LP/Pul.

See also Figure S4, Figure S5, and Figure S6 for hard sessions. (A) Optogenetic stimulation-induced change of the percentage of correct / premature / omission / incorrect trials compared to sham conditions. (B) No change of average velocity during optogenetic stimulation (V_Opto) and reward retrieval (V_Retrieval) from the sham condition. (C) No change of reaction time for correct trials from the sham condition. (D) Illustration of definition of velocity variables in B and C. For A-C, the p-values were calculated from linear mixed effect models: PercentChange (data from A) or VelocityChange (B) or ReactionTimeChange (C) = 1 + OptoContrast + (1+OptoContrast|AnimalID) controlling for the random effect of different animals, and then corrected for multiple comparisons using Holm–Bonferroni method. Each dot represents one session. N = 51 sessions from 4 animals.

Optogenetically-induced effective connectivity change correlates with behavioral performance change

Sustained attention can be viewed as a gating mechanism for visual information to be transmitted to the higher-order cortices³². This is supported by our findings of LP/Pul-VC connectivity being modulated by different levels of attentional demand (Figure 3, D and E). Thus we next causally tested how the optogenetically-induced change in LP/Pul-VC connectivity correlated with behavioral performance (Figure 7, A and B). Behavioral performance was measured by the difference in percentage points of accuracy in verum compared to the sham condition. We focused this analysis on theta and gamma oscillations due to the network-wide task-relevant enhancement of their power and functional connectivity (Figure 2) during the sustained attention period. The theta stimulation-induced theta spike-PLV change from VC to both LP/Pul (Spearman correlation coefficient r = 0.36, P = 0.02) and PPC (r = 0.27, P = 0.08) was significantly and positively correlated with stimulation-induced change in accuracy (Figure 7A). However, this correlation was absent during alpha-stimulation (VC->LP/Pul: r = 0.16, P = 0.31, VC->PPC: r = −0.02, P = 0.92) (Figure 7B). One hypothesis is that alpha-stimulation suppressed information transmission, which impaired the influence of theta spike-PLV on gating sensory input. As a result, the positive correlation between theta spike-PLV and the behavioral performance was no longer present for the alpha optogenetic stimulation condition. Interestingly, both theta and alpha stimulation-induced gamma spike-PLV change correlated with accuracy change except for VC to LP/Pul during theta stimulation (theta stimulation: VC->LP/Pul r = 0.35, P = 0.02, VC->PPC r = 0.18, P = 0.26; alpha stimulation: VC->LP/Pul r = 0.50, P < 0.001, VC->PPC r = 0.57, P < 0.001). This implies that successful enhancement of communication in specific frequency bands (theta and gamma) from the visual cortex to higher-order visual regions contributed to performance improvement. The choice of connection to include in this analysis was driven by our main hypothesis.

Figure 7. Brain-behavior correlation underlying the performance change.

See also Figure S6 for hard sessions, Figure S7, Table S2. (A-B) Correlation between optogenetics-induced accuracy change (y-axis) and optogenetics-induced average spike-field coupling (PLV) change during the optogenetic stimulation window ([−3,0] second before visual stimulation) (x-axis) for theta-sham (A) and alpha-sham (B), respectively. The first row represents spike-PLV averaged across the theta frequency band (5–7 Hz), and the second row represents spike-PLV averaged across the gamma frequency band (40–75 Hz). A->B indicates A’s spikes phase-locking to B’s LFP. Each dot represents one session. N = 51 sessions from 4 animals. Correlation coefficient r and significance value P were calculated using Spearman correlation methods and verified using percentage bend correlation (See Table S2).

Blocked optogenetic stimulation and delta stimulation do not induce performance change

To control for the effect of laser delivery and potential confounds caused by light leakage, we conducted control sessions with the tip of the optogenetic cable blocked but still plugged into the implant (Figure S5, A and B). We did not find any significant change in trial outcomes or reaction time for optogenetic stimulation condition compared to sham (Accuracy: Theta vs. Sham: P = 0.385, Alpha vs. Sham: P = 0.094; Reaction time: Theta vs. Sham: P = 0.167, Alpha vs. Sham: P = 0.934). This finding confirms that the optogenetic effects in this study were not due to potential light leakage at the interface of the optic cable and the implant.

Since both theta and alpha stimulation increased alpha power, we investigated the effect of targeting a different frequency band that is not in proximity to alpha and theta frequencies. Delta optogenetic stimulation (at 2.5 Hz) did not drive a distinct alpha-band power change (Figure S5E), indicating that increased alpha power observed in theta and alpha stimulation conditions was an outcome of frequency-specific optogenetic stimulation and possible cross-frequency coupling. Further, delta stimulation did not significantly change any behavioral outcome (Delta vs. Sham: Accuracy: P = 0.409, Reaction time: P = 0.595) (Figure S5, C and D).

Optogenetic effect on the circuit and behavior is state-dependent

We next investigated whether similar stimulation effects could be observed in the hard sessions when attentional demand was higher. Unlike in easy sessions, in hard sessions, alpha stimulation significantly reduced omission but did not significantly change accuracy (Figure S6A).

Interestingly, this behavioral change significantly correlated with optogenetics-induced changes in SUPLV in theta-band but with alpha stimulation from VC->PPC (r = 0.47, P = 0.02). This implies a state-dependent mechanism where when baseline theta SUPLV is low as in the easy session, external theta stimulation would improve this coupling. The theta coupling from the sensory area then correlates with higher accuracy (Figure 7A VC->LP/Pul). In a hard session where the theta oscillation is already strong, it is hard to further improve functional theta coupling (i.e. SUPLV is higher but not correlated with performance). However, alpha stimulation was able to increase both alpha-band and theta-band coupling (to a lesser degree) and the increased theta-band coupling from the sensory area correlated with higher accuracy (Figure S6E) and lower omission (Figure S7). This suggests that increased alpha-band coupling between sensory area and thalamus might have contributed to the suppression of distractions and thereby reduced omissions.

Discussion

The role of the posterior thalamo-cortical network in sustained attention

In primates, the pulvinar is well-connected with multiple cortical regions and forms cortical-pulvino-cortical input-output loops, providing the anatomical foundation for a central role in synchronizing thalamo-cortical networks. Pulvino-cortical (temporo-occipital area and V4, respectively) synchrony is modulated by a selective attention task and this synchrony is directed from the pulvinar to the cortical regions ²⁰. Both the temporo-occipital area and V4 are considered higher-order sensory processing areas and visual association areas. Further, deactivating the pulvinar with muscimol reduces visual responsiveness as well as high-frequency synchrony within V4 ²¹. These studies elucidated the important role of pulvinar in modulating cortico-pulvino-cortical interactions, especially during attention-related tasks. In addition, theta-rhythmic sampling was shown during attentional engagement ^33,34. The pulvinar has also been shown to rhythmically engage or disengage the fronto-parietal network based on its theta phase during a spatial attention task ³⁴. However, the causal relationship was either demonstrated using conditional Granger causality and spike-LFP phase coupling instead of a causal manipulation, or via a non-specific silencing of the pulvinar. Further, none of the studies have directly examined the pulvinar-induced cortico-cortical interactions and their relationship with behavioral performance, so until this study, it was inconclusive whether the modulation occurs via pulvinar interaction with each cortical region individually or via successful synchronization of two cortical regions.

In our study, we were able to explicitly and causally test the role of cortical-pulvino-cortical oscillations by using frequency-specific optogenetics. Ferrets are ideally suited for such an investigation since they exhibit a sufficiently sophisticated thalamo-cortical visual system that generates alpha oscillations ³⁵. In addition, the LP/Pulvinar complex (LP/Pul) in ferrets resembles the primate pulvinar nucleus as it reciprocally connects with VC and the posterior parietal network ³⁶. This provides an anatomical foundation for the higher-order visual thalamus to modulate cortical communication and visual information integration. We found that LP/Pul synchronized cortical regions PPC and VC during the sustained attention period at the theta frequency band. This is consistent with the role of theta in facilitating long-range communications between brain regions ^37,38. It is known that alpha oscillation in the posterior network is the most prominent feature in eyes-closed conditions and that it plays an important role in modulating neural activity during visual-attention tasks ^29,30. Thalamic alpha is also shown to have an antagonist effect with theta oscillations as a function of arousal level ³⁵. In our study, we have shown a significant power decrease in the alpha frequency band in the delay period compared to the baseline in the posterior cortical regions (PPC and VC) in Figure 2B as well as between PPC-VC measured by phase-locking values in Figure 2DE. This is consistent with the finding that during a high arousal state (i.e. the sustained attention period) there are stronger theta and weaker alpha oscillations in the posterior network.

In addition, this directional theta drive was stronger for the hard condition of the task with higher attentional demand compared to the easy condition, demonstrating the significance of this thalamo-cortical theta drive in maintaining and modulating the magnitude of sustained attention. Further, using frequency-specific optogenetic stimulation, we successfully entrained the posterior network (both pulvino-cortical interaction and cortico-cortical interaction). Importantly, only theta but not alpha frequency stimulation caused significant improvement in the accuracy of the sustained attention task for the easy condition, and alpha frequency stimulation reduced omissions in the hard condition.

Further, this causal relationship was mainly explained by a successful engagement of the VC-LP/Pul pathway in the theta stimulation condition which increased accuracy, and this brain-behavior correlation was absent in the alpha stimulation condition. It should also be noted that there was a trend-level correlation between VC-PPC spike phase locking at the theta frequency band and accuracy, indicating that cortico-cortical interactions might also play a role in modulating behavioral performance. In summary, our results suggested that LP/Pul theta oscillation is modulated by attentional demand and causally enhances theta synchrony in the pulvino-cortical pathway (i.e. LP/Pul-PPC and LP/Pul-VC) as well as cortico-cortical pathway (i.e. PPC-VC), which in turn improves sustained attention performance.

Limitations

Like all scientific research, out study has limitations. First, we did not parameterize the task difficulty based on individual task performance. To mitigate this issue, our results are mostly focused on contrasts that indicate attentional demand or treatment effect so that the baseline performance is of less importance. Second, our study did not differentiate layer-specific connections and oscillations. For cortical regions, we primarily targeted the layer IV, V neurons due to the presumable abundance of thalamo-cortical input and cortical-thalamic output from the neocortex. However, it is still under investigation whether this is true for ferrets in the cortices we are interested in and we might underrepresent some cortico-cortical connections. Third, we recognize that an ideal control for the side-effect of optogenetic stimulation is to use another set of control animals injected with a control viral vector without opsin. However, given the complexity of the experiment, we decided to use two types of within-subject control conditions: within each session, we use a no-light sham condition randomly interleaved with verum stimulation conditions; in addition, we also collected control sessions where the optic cable tip was blocked throughout the session to control for the effect of any potential light leakage. Fourth, there are other possible pathways where the thalamus could have an impact on attention, such as LP/Pul->PMC->PPC. In our study, we could not rule out effect from other circuits, but only can look at the cumulative effects of the LP/Pul ->PPC or VC. Lastly, our task design does not allow to directly examine the neural signatures of unsuccessful trials such as omission trials since the trial number for these conditions is too low for the type of analyses used in this study.

Implications

Through a progression of target identification, engagement, and validation ³⁹, we demonstrated that the higher-order visual thalamus causally modulates thalamo-cortical and cortico-cortical synchrony to facilitate sustained attention via theta oscillations. Thalamic theta oscillations act as a gating mechanism for visual sustained attention by top-down control of spike-phase alignment as well as phase coupling to the gamma amplitude in the posterior network to presumably prepare the communication channel and amplify the bottom-up signal from the sensory cortices. The magnitude of communication in the cortico-cortical and cortico-thalamic circuit (e.g. VC->PPC, VC->LP/Pul) in turn predicts the sustained attention performance. Strikingly, performance enhancement is driven by increased rhythmic organization mediated by spiking activity in the visual cortex during stimulation of higher order-visual thalamus. This demonstrates the key role the of higher-order thalamus in organizing cortical drivers of the neural synchronization that underlies sustained attention and cognitive control. Since the thalamus is composed of multiple specialized nuclei that are connected to lower and higher sensory cortices, this cortico-thalamo-cortical interaction could be a mechanism underlying thalamic control of cortical synchrony during sensory processing in general.

STAR Methods

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Flavio Frohlich (Flavio_frohlich@email.unc.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Electrophysiological and behavioral data have been deposited at https://osf.io/5ac7j/ and are publicly available as of the date of publication.

Custom MATLAB code for the presented analyses has been deposited at https://gitlab.com/FlavioFrohlich/5CSRTT_analysis is publicly available as of the date of publication.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals

Four adult female ferrets (Mussel putorius furo, 16–19 weeks old at the beginning of the experiment, weighing 0.7–1kg, group-housed in a 12hr light/12hr dark cycle, spayed, Marshall BioResources, North Rose, NY) were included in this study. All animal procedures were performed in compliance with the National Institute of Health guide for the care and use of laboratory animals (NIH Publications 8^th edition, 2011), and were approved by the Institutional Animal Care and Use Committee of the University of North Carolina at Chapel Hill, and the United States Department of Agriculture (USDA Animal Welfare). Only female animals were used in this experiment due to the fact that the skull of male ferrets continues to grow during the experimental period which impairs the stability of the chronic cranial implants whereas the skull of the female ferret ceases to grow earlier in development.

There were 85 sessions of easy level (Animal A = 16, Animal B = 21, Animal C = 21, Animal D = 27) and 42 sessions of hard level (Animal A = 16, Animal C = 12, Animal D = 14, animal B was excluded due to this animal only completed 3 hard level sessions) without optogenetic stimulation. In easy vs. hard level contrast, 42 easy sessions were randomly selected from animals A, C, D to match the number of hard sessions. There were another 51 sessions at easy level with optogenetic stimulation (Animal A = 9, Animal B = 11, Animal C = 13, Animal D = 18).

METHOD DETAILS

Behavioral Training

A detailed description is similar to papers ²⁷ and ¹⁵.

Briefly, each animal was trained individually in a touch-screen-based 5-Choice Serial Reaction Time Task (5-CSRTT) ²³. There is one difference from the above-mentioned papers. To increase the probability of animals facing the screen during the delay period, we installed an IR beam across the chamber in front of the water spout. When the animal is at the spout after initiation, its body would block the IR beam and a new trial would be pending. Only when the animal turned back and unblocked the IR beam, thus correctly facing the screen, the trial and the delay period would start.

There are five main stages of training: 1) Chamber habituation, 2) Touch-reward association, 3) Stimulus-reward association, 4) Task initiation, and 5) 5-CSRTT with delay period. Within each stage, there were substages where the parameters were tuned so that the task difficulty gradually increase. The training stage progressed gradually once the animal met the criteria to move on to the next stage. Animals were trained and tested 1–2 times daily on weekdays during which they were water restricted. Animals were not trained and had free access to water at weekends. During the training and testing, animals received a water reward for correct responses. At the end of each training day, animals also received supplemental water to ensure their total water intake was no less than 60mL/kg/day. Each animal’s weight was monitored daily and never dropped below 85%.

Accuracy per session was measured as the main outcome variable (defined as the fraction of trials that the animal touches the correct window over all of the trials completed), together with the reaction time for correct trials.

Virus injection and electrode implantation surgery

The initial induction of anesthesia of each animal was performed by injecting a ketamine/xylazine cocktail (30 mg/kg of ketamine, 1–2 mg/kg of xylazine, intramuscular). After confirming a stable plane of anesthesia by the absence of toe-pinch response, the animal was intubated for subsequent mechanical ventilation with vaporized anesthetic maintenance (0.5–2% isoflurane in 100% oxygen). The animal was administered pain relief medication (meloxicam, 0.2 mg/kg, subcutaneous). The animal was then positioned into a stereotaxic frame, with its head fixed using a mouthpiece and ear bars to ensure the stability of the head throughout the procedure and the precise targeting of studied regions for virus injection and electrode implantation. Coordinates for virus injections and electrode implantations were based on the ferret atlas ⁴⁰. The vital signs including electroencephalogram, pulse, respiratory rate, partial oxygen concentration, end-tidal CO2, and rectal temperature were continuously monitored and recorded throughout the surgical procedure to maintain the animal in a stable state. Aseptic conditions were maintained during the following surgical procedures. Craniotomies were performed over the left hemisphere of PPC, and VC. The craniotomy over PPC was enlarged for implant Microelectrode arrays in PPC and optrode in LP/Pul. The dura and pia were removed before virus injection and electrode implantation. Each of the PPC, VC regions was implanted with a microelectrode array (2 × 8 tungsten electrodes, 35 μm diameter, 5 mm length, 200 μm spacing; a local reference electrode that is 500 μm shorter than recording electrodes Innovative Neurophysiology, Durham, NC) at the depth of around 600 μm. For optogenetic stimulation in LP/Pul, 0.3 μL of viral vector rAAV5-CaMKII-ChR2-mCherry was injected into LP/Pul. Then, an optrode (i.e. a circular electrode array with an optical fiber in the middle) was implanted (electrode tips and optical fibers were implanted 100 and 400 μm above the virus injection site, respectively) (16 channel circular Platinum/Iridium electrode, 125 μm diameter, 10 mm length, 250 μm spacing; fiber optic 9.5 mm length, 200 μm core outer diameter, 0.48 numerical aperture; Microprobes for life science, Gaithersburg, MD). Custom-designed plastic cylinders were implanted around the light fibers to anchor laser cables during behavior and to prevent laser light leakage. Each microelectrode array contained a silver grounding wire that was connected to bone screws on the skull. Additional bone screws were implanted for extra security of the headcap. A headcap was built using dental acrylic (Lang Dental, Wheeling, IL) to secure electrodes and cover the skull and the craniotomy. After hardening of the dental cement, skin and connective tissue were sutured together. Triple antibiotic ointment was applied around the suture to prevent infection. During post-operative care, pain medication (meloxicam, 0.2 mg/kg, subcutaneous) and antibiotics (clavamox, 12.5–13 mg/kg, oral) were given to the animal in addition to regular headcap monitoring and cleaning. The animals recovered with free water access in their home cage for at least a week before retraining and recording.

Animal in vivo recording and optogenetics

After recovery from surgery, animals were retrained on the final level of 5-CSRTT task until they reached stable criteria level (80% accuracy on the easy level) performance again (which took about 1–2 weeks). This time period also allowed for viral expression. At the beginning of a recording session, animals were placed into a custom-designed body fixation tube for stabilization while connecting the implanted multichannel electrode arrays with a data acquisition system (INTAN technologies). There were two conditions for recording sessions: easy condition (with 2-sec visual stimulus presentation) and hard condition (with 1-sec visual stimulus presentation). The two conditions were randomly assigned to sessions to avoid confounding from the training effect. A same number of sessions of easy and hard conditions were included for a fair comparison.

Within each session, the animals are allowed to complete as many as 100 trials or until they are satiated or not engaging in the task for 2 minutes (on average 50–60 trials). All of the correct trials (about 70–90% of all trials) were further examined for their electrophysiology recording quality. The trials with significant noise in any of the recording regions (i.e. those cannot be cleaned by our preprocessing denoising pipeline) were excluded (about 10–20% trials). A session is considered a valid session if it meets the following criteria: 1. accuracy >= 70% (for both hard and easy), 2. Number of accurate trials with good quality electrophysiology recording >= 30.

For sessions with optogenetic stimulation, the laser power was tuned to be 30 mW out of the optic fiber cable tip. The optic fiber cable was also connected to the optrode implant interface. A custom-made black light-proof cylindrical sheath was applied surrounding the optic cable connection end to prevent light leakage. There were three conditions of trials randomly interleaved during each session: optogenetic stimulation at individual theta frequency band (animal A: 5.2 Hz, B: 5.2 Hz C: 5.6 Hz D: 5.6 Hz), alpha frequency band (animal A: 15 Hz, B: 15 Hz C: 16 Hz, D: 16 Hz), and sham condition where no optogenetic stimulation was delivered. The laser was delivered at 50% duty cycle. The data acquisition system and the optic fiber cable were connected to a commutator (Tucker-Davis Technologies) positioned on the top of the box to allow freely-moving behavior. From the commutator, another optic patch cable was used to connect to the laser source (Shanghai Laser & Optics Century Co.). The broadband electrophysiological data (1Hz to 5kHz, 20kHz sampling rate) was collected simultaneously with the time-locked event signals (behavioral events and laser pulses) for offline processing. During each session, a high-resolution infrared video was collected throughout and was later synchronized with the other data via frame time matching of the first stimulus onset. The task and behavioral events were executed and recorded by a custom-written Matlab script.

For the optogenetics control session (Figure S5), the optic cable tip was covered by light-proof tape and was attached on top of the optrode implant interface so that the setup was the same as in verum conditions but there was no laser going into the optrode implant.

Histology and verification of electrode locations

After reaching the scientific endpoint, animals were euthanized by injecting an overdose of ketamine/xylazine cocktail (30 mg/kg of ketamine, 1–2 mg/kg of xylazine, intramuscular) and perfused with 4% paraformaldehyde in phosphate-buffered saline. The brain was extracted from the skull and kept in the same solution for post-fixation overnight. Then, the brain was submerged in 30% phosphate-buffered sucrose solution for cryoprotection till it sank. After shock-froze in dry ice, the brain was sliced into 50 μm sections using a cryostat (CM3050S, Leica Microsystems).

Sections were collected into three series: one was stained by Nissl for visualizing cell nuclei, one was for fluorescent imaging, and the last one was stained by cytochrome oxidase for better visualization of the thalamic nuclei. Nissl and CO series were imaged with a bright-field slide scanner at 10x magnification (Aperio VERSA). The fluorescent series was imaged with a widefield fluorescence microscope (Nikon Ti2). Implantation and viral expression locations were verified by overlapping the images with the ferret atlas planes ⁴⁰. Only electrodes that were verified to be in the targeted location were used for the analysis.

QUANTIFICATION AND STATISTICAL ANALYSIS

Custom scripts were used to execute all of the offline data analyses in Matlab 2020a (Mathworks).

Data preprocessing:

Raw extracellular potential data were first filtered by a low-band pass filter at 300 Hz to obtain local field potentials (LFPs). LFPs were then downsampled from 30 kHz to 1 kHz sampling rate. Further, data with movement artifacts were interpolated and outlier channels were removed from the dataset using Artifact Subspace Reconstruction (ASR) method (EEGLAB). At last, LFPs were epoched into trials, and trials with high amplitude artifacts were manually removed. On average, 2–3 trials out of 40–60 trials were removed in each session.

Multi-unit spiking activity was exacted from raw extracellular potentials by first applying a band-pass filter between 300 and 5000 Hz and then applying a −3.5 standard deviation threshold below the potential mean. Single-units were extracted using a template-matching algorithm implemented in KiloSort and KiloSort 2 ⁴¹ followed by manual curation and template matching via the phy software ⁴².

Spectral analysis:

The spectral decomposition of the LFPs was done with wavelet transformation. We used Morlet wavelets $\left(w\left(t,f_0\right)\right)$ that have Gaussian shape envelop in both time and frequency domains:

$$w\left(t,f_0\right)=A\cdot \mathrm{e}\mathrm{x}\mathrm{p}\left(-t^2/2{\sigma }_t^2\right)\cdot \mathrm{e}\mathrm{x}\mathrm{p}\left(-2i\pi f_{0}t\right)$$

where $f_0$ represents the central frequency of the wavelet, $A={\left({\sigma }_t\sqrt{\pi }\right)}^{-1/2}$ is a constant to normalize wavelet energy to 1, and ${\sigma }_t=1/2\pi {\sigma }_f$ is the standard deviation in the time domain as a function of the standard deviation in the frequency domain $\left({\sigma }_f\right)$. After convolving with the wavelets, LFPs in the time domain were transformed into complex-valued analytical signals $X\left(t,f_0\right)$ at each carrier frequency of $f_0$. We used a family of 150 wavelets with logarithmically spaced central frequencies between 2 and 128 Hz. The power spectrogram was calculated for each valid channel (by preprocessing and anatomical validation), each studied region, and each animal and then averaged by taking the median across channels and then across sessions and animals.

LFP phase synchronization:

Our main measurement of functional connectivity between regions is phase-locking-value (PLV). We apply PLV to measure phase synchronization between LFPs in different regions ⁴³. We have also calculated coherence, which confounds the consistency of phase difference with amplitude modulation. In general, the two methods yield similar results. Due to the sample-size bias of PLV measurement (with more sample biases PLV to 0, and fewer sample biases PLV to 1), we randomly sampled the same number of trials from each session to be included in the PLV calculation and dropped the sessions that don’t have enough trials. The number of subsampled trials $N$ was chosen so that at least 85% of the sessions will be included in each condition (easy vs. hard). Briefly, the phase angles (e.g. ${\theta }^a$ and ${\theta }^b$) of the analytical signal $X\left(t,f_0\right)$ were calculated for studied channel pairs (e.g. a and b). The PLV at the carrier frequency of $f_0$ for $N$ samples was defined as:

$$\left.\mathrm{P}\mathrm{L}\mathrm{V}\left(f_0\right)=\frac{1}{N}\right|\sum _{n=1}^N \mathrm{e}\mathrm{x}\mathrm{p}i[{\theta }_n^{\mathrm{a}}\left(f_0\right)-{\theta }_n^{\mathrm{b}}\left(f_0\right)]\mid$$

PLV for each channel pair was then averaged (median) to produce PLV between a region pair in each session. PLV results were then averaged across sessions for each condition (e.g. easy vs. hard). Because the delay duration of each trial is randomly selected from one of three values, we aligned all of the trials at trial initiation and stimulus onset respectively to obtain time-resolved PLV.

Spike-LFP phase synchronization

To quantify the timing dependence between spikes (apply to both single-unit and multi-unit activity) and the LFP phase, we computed Spike-LFP phase synchronization ⁴⁴ both within and between regions using simultaneously recorded spike and LFP data from all channels. For each region pair, we use LFPs from a representative channel (correlation between channel LFPs within the same region is greater than 0.8), and spikes from all valid channels. LFP phases were computed using wavelet convolution at each frequency of interest. For each spike channel, to calculate the PLV of spike-locked LFP phases across trials, we randomly selected the same number of spikes $K$ per time bin to control for any sample-size bias of the PLV measurement. The spike PLV at a carrier frequency of $f_0$ for $K$ spikes was defined as:

$${\mathrm{P}\mathrm{L}\mathrm{V}}_{\text{spike}}\left(f_0\right)=\frac{1}{K}\mathrm{e}\mathrm{x}\mathrm{p}[i{\theta }_k^{\text{spike}}\left(f_0\right)]$$

where ${\theta }^{\text{spike}}$ represents the instantaneous LFP phase at the occurrence of each spike. This random sampling procedure was repeated 200 times to yield a robust average PLV value for each time bin. The channels that do not have enough spikes per time bin were removed from the Spike-LFP analysis. The time-frequency resolved Spike-LFP phase-locking value will be averaged across channels within the same region, then averaged across sessions to yield the final result for each animal.

Conditional Granger Causality

Conditional Granger Causality (CGC) is a variant of the Wiener-Granger causality algorithm ^45–47. It measures the influence of one time-series $y$ on another time-series $x$, conditioned on (i.e. taken into consideration) the information of other time-series $z_1,z_2$, etc. The algorithm builds two vector autoregressive (VAR) models to predict the current value of $x$: a full model that uses the previous values of $x,y$, and $z_i$, and a reduced model that uses only the previous values of $x$ and $z_i$. If the full model results in less residual error than the reduced model, we can infer that time-series y contains critical information that can be used to help predict $x$, i.e. there is directed information flow from $y$ to $x$. We applied the MVGC (Multivariate Granger Causality) Matlab toolbox ⁴⁸ to implement the CGC in the frequency domain on the signals from three brain regions we are interested in: LP/Pul, PPC, and VC. For every studied region-pairs, we calculated CGC for both directions, conditioned on the signals from the other two regions. The LFP time series were downsampled to 200 Hz first and the CGC was calculated for a moving window with 1-second width and 0.1-second step. The MVGC toolbox works by first determining the model order using Akaike or Bayesian information criteria or cross-validation. We used Akaike method for model order estimation with a maximum allowed model order of 20. After the suitable model order is determined, a VAR full model is fitted to the time-series data and the parameters are estimated for both the full model and the reduced model. Finally, the estimators of residuals of covariance matrices are used to calculate the CGC.

Optical-tagging test

A detailed description of this method is in ³¹. Briefly, the optical-tagging test, i.e. Associated spike Latency Test (SALT) is used to identify light-activated neurons. We compared the spike timing of a unit during the optogenetic stimulation period ([−3, 0] s around visual stimulus onset) to that during a sham baseline ([−8, 5] s around visual stimulus onset) and yielded a p-value indicating whether there were optogenetically-induced significant changes in spike timing.

Quantification of entrained SU

To quantify the degree of entrainment by optogenetic stimulation for each SU, we computed a fast Fourier transform on the PSTH (with time-bin = 0.02 s) of each SU during the 3-second optogenetic stimulation window. The spectra signified the frequency bands that were enhanced due to the rhythmic firing activity induced by optogenetic stimulation. Then, we averaged across all of the SUs to show the group-level entrainment pattern of the SU under optogenetic stimulation as well as sham condition in each region (Figure 4F). To classify an SU as “entrained” in a frequency band (e.g. theta or alpha), we averaged the PSTH spectra power within that frequency band and compared it against a clean frequency baseline. This baseline was chosen to be two standards from the mean of the spectra in 20–25Hz frequency band because this is a stable band that is free from optogenetic effect and is representative of the mean and variance level of the spectra of PSTH without stimulation. We also tested using a different frequency band as baseline, the results are very similar. If the PSTH spectra power at a certain frequency band of an SU is higher than the above-defined baseline, we label this SU as “entrained” at that frequency band. After calculating the percentage of entrained SU under each optogenetic condition, we compared the percentage between optogenetic conditions using X² test. The resulted the P-values were corrected for multiple comparisons using Holm–Bonferroni method.

Permutation testing

To test for the significant time-frequency differences between conditions of interest (eg. easy vs. hard, theta optogenetic stimulation vs. sham), we applied nonparametric statistical testing across sessions and animals. The permutation testing starts with first computing the cluster size for each condition for each studied region or region pair. Then the conditions were randomly swapped for half of the sessions to establish a baseline where the difference between the conditions should be eliminated. We calculated the difference in contrast between these new condition labels and record the largest negative and positive cluster sizes. We then repeated this randomization process 1000 times to form a null distribution of the largest negative and positive cluster sizes. We used the 25th and 95th percentile values from the null distribution as the critical size values and considered the more extreme values to be statistically significant with an α level of 0.05. This way, we are able to control for multiple comparisons across time-frequency clusters ^49,50.

Brain-behavior correlation

A correlation was done after excluding outliers defined to be 3 standard deviations from the median of all sessions across 4 animals (although the statistical results did not change much when including outliers). Correlation coefficient r and significance value p were calculated using Spearman correlation methods and the results agree with percentage bend correlation, which is a correlation method that is robust to outliers ⁵¹.

Statistical analysis

Comparison of spectral features between different conditions (eg. easy vs. hard) was done at the session level using a 2-tailed paired t-test. A comparison of the means of behavioral measures between different conditions was done at the session level across animals. To control for the random effect of animal behavioral differences, P-values were calculated from linear mixed effect models “Measure = 1 + Condition + (1+Condition|AnimalID)” with significance levels of 0.05. If multiple measures were calculated as a group (i.e. different trial outcomes), then the P-values were corrected for multiple comparisons using Holm–Bonferroni method.

Frequency bands of interests

For statistical testing, the frequency bands of interests are defined as: theta frequency band (5–7 Hz), alpha frequency band (14–18 Hz), and gamma frequency band (40–75 Hz), unless specified otherwise.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
AAV-CaMKII-GFP	UNC Vector Core	N/A
rAAV5-CaMKII-ChR2-mCherry	UNC Vector Core	N/A
Biological samples
Ferret	Marshall BioResources, North Rose, NY	https://www.marshallbio.com/animals
Deposited data
Raw and analyzed data	This paper	https://osf.io/5ac7j/
Software and algorithms
MATLAB 2020a	Mathworks	https://www.mathworks.com/products/matlab.html
Customized MATLAB code	This paper	https://gitlab.com/FlavioFrohlich/5CSRTT_analysis
ImageJ	Schneider et al.	https://imagej.nih.gov/ij/

Highlights:

• Thalamo-cortical theta oscillation was modulated by a sustained attention task.

• Frequency-specific optogenetic stimulation enhanced connectivity and performance.

• Stimulation-induced connectivity changes correlated with performance.

• The effect of optogenetic stimulation was state-dependent.