Gamma rhythms and visual information in mouse V1 specifically modulated by somatostatin-positive neurons in reticular thalamus

Visual perception in natural environments depends on the ability to focus on salient stimuli while ignoring distractions. This kind of selective visual attention is associated with gamma activity in the visual cortex. While the nucleus reticularis thalami (nRT) has been implicated in selective attention, its role in modulating visual perception remains unknown. Here we show that somatostatin-(SOM) but not parvalbumin-expressing (PV) neurons in the nRT preferentially project to visual thalamic nuclei. In freely behaving mice, single-unit and field recordings reveal powerful modulation of both visual information transmission and gamma activity in primary visual cortex (V1), as well as in the dorsal lateral geniculate nucleus (dLGN). These findings pinpoint the SOM neurons in nRT as powerful modulators of the visual information encoding accuracy in V1, and represent a novel circuit through which the nRT can influence representation of visual information.


INTRODUCTION
ed that they may play a causal role in perception and the focusing of attention (Gray et al., 1992;Singer and Gray, 1995;Kreiter and Singer, 1996) or in enabling a time-division multiplexing of cortical responses to multiple simultaneous stimuli (Stryker, 1989). Electro-corticographic (ECoG) studies have shown a clear correlation between focal increases in high gamma and cortical responses to stimulation measured by other means (Ray et al., 2008a;Yazdan-Shahmorad et al., 2013). Whether gamma oscillations play a causal role in brain function or merely represent the "ringing" of an insufficiently damped response of the cortical circuit to a strong input is not clear. However, studies have shown that the effective output of primary sensory cortical areas to peripheral stimulation is much greater during a particular phase of gamma, and that increasing gamma and synchronized cortical activity by optogenetic stimulation at the appropriate phase with respect to a sensory stimulus can change the number of In order to study the effect of nRT activity on the thalamocortical visual system, we transfected SOM and PV cells in the portion of the nRT that projects the visual thalamic nuclei with channelrhodopsin-(ChR2) or halorhodopsin-(eNpHR) containing viruses as described in (Clemente-Perez et al., 2017). This verified that the expression of opsins was located in the nRT and cell-type specific and the opsins were well expressed (Figure 1-figure supplement 1).

Optogenetic activation of SOM but not PV nRT neurons reduces single-cell responses and gamma power in V1 both with and without visual stimulation
To determine whether disrupting activity of SOM nRT neurons affects visual responses in V1, we injected an AAV viral construct containing ChR2 in the nRT of SOM-Cre mice. Thereafter, extracellular recordings of single unit activity and local field potentials (LFPs) were made using a double-shank 128-channel microelectrode array placed in the V1 of mice that were free 1 2 3 4 to stand or run on a polystyrene ball floating on an air stream (Figure 2A) (Du et al., 2011;Hoseini et al., 2019). Mice viewed a gray blank screen while a blue light (473 nm, ~63 mW/mm 2 ) was delivered using an 1 2 3 4 Figure 1. Visual relay thalamic nuclei are preferentially targeted by SOM and not PV GABAergic neurons from nRT. A, Representative example sections of SOM-Cre and PV-Cre mice after injection of floxed AAV in nRT, which results in eYFP expression in cell bodies and projections of SOM or PV neurons, respectively. Yellow boxes indicate locations chosen for 63x confocal imaging and putative bouton quantification. Inset: nRT injection site as seen in an adjacent section. B, 63x confocal images showing the entire field of view (FOV) and a zoomed cropped region ('High mag') to show details of axonal boutons and nRT somata. LP: lateral posterior nucleus; dLGN: dorsal lateral geniculate nucleus; PO: posterior medial nucleus; vLGN: ventral lateral geniculate nucleus; VPM: ventroposteromedial nucleus; nRT: thalamic reticular nucleus. The expression of the viral constructs in different brain regions was confirmed using the mouse brain atlas (Paxinos and Franklin, 2001). C, Number of eYFP-labeled boutons present in thalamic nuclei of representative mice shown in panel A. Data taken from three consecutive sections from each mouse. D, Number of eYFP-labeled boutons present in thalamic nuclei of all mice imaged (n = 2 SOM-Cre, 3 PV-Cre, 3-4 sections per mouse). Differences are significant between genotypes for all regions except for nRT after correction for multiple comparisons. *p<0.05, **p<0.01.  Table 1). Consistent with previous findings, 1 2 3 4

Figure 2. Optogenetic activation of SOM nRT neurons reduces gamma activity in the primary visual cortex (V1) both with and without visual stimulation. A,
Neural activity was recorded from V1 in freely-moving mice. Mice were presented with a gray blank screen while a blue light (473 nm, ~63 mW/mm2) was delivered to ChR2-expressing SOM cells in nRT using an optical fiber implanted above the nRT. Mouse movement was tracked over the course of the experiment. B, Representative extracellular raw voltage trace is shown along with its power spectrum. Blue shading areas indicate optogenetic activation. Mouse movement speed is shown at the bottom. C, Across-trial average power of all channels in the absence (black) and presence (blue) of optogenetic activation in one representative mouse shows a moderate decrease across theta (4-8 Hz) and beta (15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) bands and a strong decrease in gamma band (30-80 Hz) (See Table 1 for statistics). D, Across-trial average power of all channels in one representative mouse indicates that locomotion slightly modulates theta and beta powers, while causing a strong enhancement in gamma power (Table 1). E, Average power across all channels shows that optogenetic activation of SOM nRT cells greatly reduces power across the three bands in the still condition (black vs. blue marks). Optogenetic activation of SOM nRT cells strongly modulates gamma power when the mice are running (red vs. blue marks) ( Table 1). F, Visual responses were recorded while mice were presented with moving gratings (8 directions, each moving in one of two possible directions; 2 s duration; randomly interleaved with optogenetic stimulation) in the visual field contralateral to the recording site. G, Firing rate (averaged over all 8 drifting directions) of an example cell during the course of experiment. Black marks: visual responses when the laser is off; Blue marks: visual responses when visual stimuli and optogenetic activation of SOM nRT cells are coupled. Red shadings: locomotion state. Error bars: SEM. H, Stimulus evoked (average over all 20 trials) minus ongoing power of all channels when the laser is off (black circles) versus the laser-on condition (blue circles) indicates a significant shift across all frequencies (Table 1). I, Same as in Fig. 2E in the presence of visual stimulus (Table 1). J, Using the three parameters calculated from average waveforms, cells were classified into narrow-(NS, cyan) or broad-(BS, magenta) spiking (height of the positive peak relative to the negative trough: -0.20 ± 0.01, -0.34 ± 0.02 (p=1.02e-9, Wilcoxon rank-sum test); the time from the negative trough to the peak: 0.73 ± 0.02, 0.32 ± 0.02 ms (p=3.9e-33), slope of the waveform 0.5 ms after the negative trough: 0.01 ± 0.00, -0.01 ± 0.00 (p=5.94e-35), BS (n=169) and NS (n=73) cells respectively). Subplot: average spike waveforms for all units, aligned to minimum, demonstrating BS (magenta) and NS (cyan) cells. K, Firing rate of BS (magenta) and NS (cyan) cells across different conditions (Table 1). L, Percentage change in firing rate of both cell types versus percentage change of power in channels that each cell is recorded from for still and running states (  Locomotion caused a moderate power increase at low frequencies (<10 Hz), a moderate decrease in the beta band (15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30), and a dramatic power increase at higher frequencies in the gamma band (30-80 Hz) ( Figure 2D and Table 1). Averaging power across all channels showed that optogenetic activation of SOM neurons in nRT nearly abolished power across the three bands in the still condition ( Figure 2E, Table 1).
In contrast, during locomotion optogenetic activation of SOM neurons in nRT most strongly reduced gamma power in V1 ( Figure 2E, Table 1).
To investigate how evoked visual responses are affected by optogenetic activation of SOM nRT cells, visual responses were recorded to drifting sinusoidal gratings presented in the visual field contralateral to the recording site, and SOM nRT neurons were ac-   1  2  3  4  5  6  7  8  9  10  11  12  13   14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30 as NS) consist of fast-spiking interneurons, whereas broad-spiking (abbreviated as BS) cells are 90% excitatory and 10% inhibitory cells (Barthó et al., 2004;Atencio and Schreiner, 2008). Optogenetic activation of SOM nRT neurons significantly suppressed activity of both BS and NS cell types in V1 during both stationary and locomotion states ( Figure 2K, Table 1). However, locomotion continued to increase the firing rates even in the presence of optogenetic activation compared with firing rates during stationary states ( Figure   2K, Table 1).
Strikingly, the effect of activating them produced com-    (Table 1). C, Across-trial average power of all channels in one mouse indicates that locomotion slightly modulates theta and beta powers, while causing a strong enhancement in gamma power (Table 1). D, Visual responses were recorded while mice were presented with moving gratings. E, Average power across all channels shows that light delivery in eYFP-expressing SOM nRT cells does not affect power in V1 (Table 1). F, Firing rate of BS (magenta) and NS (cyan) cells across different conditions are not affected by the light in mice in which nRT does not express the opsin ( Table 1)     Changes in the V1 ongoing power due in response to optogenetic inhibition of SOM (maroon) and PV (dark yellow) neurons in nRT across the three frequency bands (SOM vs. PV; theta, still: -3.8e3 vs. 0.2e3, p=1, run: -15.9e3 vs. 3.5e3, p=0.13; beta, still: 6.4e3 vs. 4.1e3, p=0.80, run: 3.9e3 vs. -0.7e3, p=0.53; gamma, still: 11.9e3 vs. 5.4e3, p=0.4, run: 9.8e3 vs. 3.8e3, p=0.13; all powers in uV2/Hz). B. Effect of the optogenetic inhibition of SOM and PV neurons in nRT on the ratio of laser on to laser off firing rates in NS and BS cells in V1, in still and running conditions (SOM vs. PV; BS, still: 1.33 vs. 1.27, p=3.7e-2, run: 1.29 vs. 1.10, p=3.1e-4; NS, still: 1.47 vs. 1.26, p=5.6e-2, run: 1.35 vs. 1.17, p=5.7e-4). *p<0.05, **p<0.01, ***p<0.001. Spike-triggered averages of a multi-(A) and a single-(B) unit in dLGN were calculated using responses to low-frequency filtered noise stimulus. C, Spike raster for a representative single-unit in dLGN across several trials (rows) with visual stimulus only (top) or visual stimulus coupled with optogenetic activation of SOM nRT neurons (bottom). Grey shaded area shows the duration of the visual stimulus. Blue shading shows the duration of optogenetic activation. Firing rates of dLGN single-units in response to visual stimulation only (D) and in response to visual stimulation coupled with optogenetic activation of SOM nRT neurons (E). The firing rate was normalized to its average ongoing firing rates averaged across all stimulus conditions. Red trace in D and E is the example cell shown in C.      14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41 contrast, despite the fact that PV cells are present in the visual portion of the nRT (see Figure 1- 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44 Comparison with carnivore and primate. The perigeniculate nucleus in the carnivore and primate is thought to be the portion of the nRT related to the thalamocortical visual system, although this is contested (Ahlsén et al., 1982). The carnivore perigeniculate consists of inhibitory neurons that receive excitatory input from ascending thalamocortical axons of the dLGN as well as from layer 6 cells of V1, and they project back to inhibit the principal cells of the dLGN in a highly focused topographic fashion. This focal projection of the nRT is consistent with the Crick (1984) "Searchlight" hypothesis for nRT function. It is not known whether the nRT projection to the mouse dLGN has sufficiently precise topography to play such a focal role in directing attention to particular areas of the field. It is possible that the portion of the carnivore nRT referred to by Ahlsén et al., (1982) as "reticular neurons" may be analogous or even homologous to the nRT of the mouse. Such an arrangement could be consistent with a role for the mouse nRT in switching attention between modalities rather than among different loci in the visual field.
Implications of our findings in disease. Sensory stimulation in the gamma range has been shown to enhance cognition in a mouse model of Alzheimer's disease (Adaikkan et al., 2019). Given that nRT is involved in sensory processing and attention, and that its dysfunction has been implicated in attention disorders (Zikopoulos and Barbas, 2012;Wells et al., 2016), we propose that SOM nRT cells could be a target for modulating gamma power in V1 and visual attention.  (Figs. 2K, 2L, 4F).

Animals
We

Viral delivery in nRT for optogenetic experiments
We performed stereotaxic injections of viruses into was validated by histology after euthanasia in mice whose brains we were able to recover and process.

Headplate surgery and implanting fiber optic
Three to six weeks after viral injections in the nRT, we  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44 mm posterior to bregma, 2.5 mm lateral to the midline) or dLGN (-2.0 mm posterior to bregma, and 2.0 mm lateral to the midline). The base of the fiber optic and the entire skull, except for the region above V1 or dLGN, was covered with Metabond (Parkell Co.). One week after the recovery from this surgery, the animal was allowed to habituate to the recording setup by spending 15-30 minutes on the floating ball over the course of one to three days, during which time the animal was allowed to run freely. About two weeks following this surgery (i.e. ~4-6 weeks after viral injection in nRT), the animal's head was fixed to a rigid crossbar above a floating ball. The polystyrene ball was constructed using two hollow 200-mm-diameter halves (Graham Sweet Studios) placed on a shallow polystyrene bowl (250 mm in diameter, 25 mm thick) with a single air inlet at the bottom. Two optical USB mice, placed 1 mm away from the edge of the ball, were used to sense rotation of the floating ball and transmit signals to our data analysis system using custom driver software.
These measurements are used to divide data into still and running trials and analyze them separately.

Microelectrode recordings in alert mice
To control for circadian rhythms, we housed our animals using a fixed 12 hr reversed light/dark cycle and performed recordings between roughly 11:00 AM and 6:00 PM. All the recordings were made during wakefulness in awake, head-fixed mice that were free to run on the floating ball (Figure 2A 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44 face and inserted to a depth of 2.5-3.0 mm (Piscopo et al., 2013). An optical fiber (200 μm diameter) coupled to a light source (green laser for eNpHR, peak intensity ~104 mW/mm 2 at 532 nm; blue laser for ChR2, peak intensity ~63 mW/mm 2 at 473 nm) was connected to the implanted fiber optic in order to deliver light into nRT. Laser power (3-20mW) was measured at the end of the optical fiber before connecting to the animals.
Recordings were started an hour after electrode insertion.

Visual stimuli
Stimuli were displayed on an LCD monitor (Dell, 30x40 cm, 60 Hz refresh rate, 32 cd/m 2 mean luminance) placed 25 cm from the mouse and encompassing azimuths from -10° to 70° in the contralateral visual field and elevations from -20° to +40°. In the first set of recordings, no stimulus was presented (uniform 50% gray) while nRT was exposed to the optogenetic light for 4 s every 20 s. For the second set of recordings, drifting sinusoidal gratings at 8 evenly spaced directions (20 repetitions, 2 s duration, 0.04 cycles per degree, and 1 Hz temporal frequency) were generated and presented in random sequence using the MATLAB Psychophysics Toolbox (Brainard, 1997;Kleiner et al., 2007) followed by 2-second blank period of uniform 50% gray. This stimulus set was randomly interleaved with a similar set in the presence of optogenetic light.
Optogenetic stimulation was delivered for 2 s periods beginning simultaneously with the onset of the visual stimulus, overlapping the entire stimulus period and turns off by the end of the stimulus.

Data acquisition
Movement signals from the optical mice were acquired in an event-driven mode at up to 300 Hz, and integrated at 100-ms-long intervals and then converted to the net physical displacement of the top surface of the ball. A threshold was calculated individually for each experiment (1-3 cm/s), depending on the noise levels of the mouse tracker, and if the average speed of each trial fell above the threshold, the mouse was said to be running in that trial. Running speed of the animal was used to divide trials into running and still states that were analyzed separately. Data acquisition was performed using an Intan Technologies RHD2000-Series Amplifier Evaluation System, sampled at 20 kHz; recording was triggered by a TTL pulse at the moment visual stimulation began. Spike responses during a 1000 ms period beginning 500 ms after stimulus onset were used for analysis.

Single-neuron analysis
The data acquired using 128-site microelectrodes the height of the positive peak relative to the negative trough, the slope of the waveform 0.5 ms after the negative trough, and the time from the negative trough to the peak (see Figure 2J). For dLGN recordings, spike triggered averages (STAs) were used to classify units into single-and multi-units (Figure 7A, B).

Mutual information
Neuronal responses are considered informative if they are unexpected. For example, in the context of visually evoked neural activity, if a neuron responds strongly to only a very specific stimulus, e.g. photographs of Jennifer Aniston (Quiroga et al., 2005)  where r and s are particular instances from the set of neural responses (measured as spike counts) and stimuli (grating movement directions), respectively. We used Information Theory Toolbox in MATLAB to compute mutual information (https://www.mathworks.com/ matlabcentral/fileexchange/35625-information-theory-toolbox)

Population-based analysis: decoding the visual stimulus from population responses
Data trials were separated into equal numbers of laser-off and -on trials. We then randomly subsampled from each 50 times to get a distribution of decoding errors based on the data included. We trained a linear discriminant analysis (LDA) classifier to classify single-trial neural responses, assuming independence between neurons (a diagonal covariance matrix). We used a leave-one-out approach to train and test classification separately for each condition (LDA-LOOXV).
The classifier was trained and tested using MATLAB's fitcdiscr and predict functions. To decode only grating orientation and not movement direction, we grouped stimuli moving 180° apart into the same class.

Population-based analysis: decoding with equal population spike counts
To determine whether optogenetic manipulation of fir-  1  2  3  4  5  6  7  8  9  10  11  12  13   14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42 but have many samples of laser-off and laser-on trials with the same population spike counts. We used an LDA-LOOXV to train and test classification separately for each subset. For each number of neurons, we subsampled with replacement 100 times from the population, yielding 100 combinations of neurons. Classifiers were trained separately on each subsample and for each condition.

Decoding stimulus orientation in dLGN recordings
Since most dLGN units are not orientation or direction selective, predicting stimulus orientation using the LDA-LOOXV approach is not appropriate. Instead, we decided to use patterns of correlations in single-unit responses to decode stimulus identity. First, we selected all the on-center single-units that were not orientation selective from all the recordings in each condition.
Using these non-simultaneously recorded neurons, we computed correlations among all possible pairs (28 pairs with optogenetic inhibition and 21 with optogenetic activation of SOM nRT neurons) across roughly randomly selected trials for each stimulus orientation. Then the data was divided into 3 sets of training (80%), cross-validation (10%) and testing (10%). A deep neural network was trained on the training set and its hyperparameters were tuned using its performance on the cross-validation dataset. Finally, the network predictions were tested using the test set. This process was repeated for each condition. A complete description of the model and codes are available at: https://github.com/Mahmood-Hoseini/Decoding-stimulus-orientation-in-dLGN-recordings
Sections were mounted in an antifade medium (Vectashield; from Vector Laboratories, H-1000) and imaged using either a Biorevo BZ-9000 Keyence microscope or a confocal microscope. The expression of the viral constructs in different brain regions was confirmed with reference to two standard mouse brain atlases (Paxinos and Franklin, 2001) and the Allen Brain Atlas (Lein et al., 2007).