Visual evoked feedforward-feedback traveling waves organize neural activity across the cortical hierarchy in mice

Abstract

Sensory processing is distributed among many brain regions that interact via feedforward and feedback signaling. Neuronal oscillations have been shown to mediate intercortical feedforward and feedback interactions. Yet, the macroscopic structure of the multitude of such oscillations remains unclear. Here, we show that simple visual stimuli reliably evoke two traveling waves with spatial wavelengths that cover much of the cerebral hemisphere in awake mice. 30-50 Hz feedforward waves arise in primary visual cortex (V1) and propagate rostrally, while 3-6 Hz feedback waves originate in the association cortex and flow caudally. The phase of the feedback wave modulates the amplitude of the feedforward wave and synchronizes firing between V1 and parietal cortex. Altogether, these results provide direct experimental evidence that visual evoked traveling waves percolate through the cerebral cortex and coordinate neuronal activity across broadly distributed networks mediating visual processing.

Fig. 1. Visual stimuli elicit strong intertrial phase coherence over large cortical areas.

a Schematic showing the 64 channel electrocorticography (ECoG) grid used to record local field potentials (LFPs) from the cortical surface of the left hemisphere of 13 awake mice. Stimuli consisted of 10 ms flashes of a green LED placed in front of the R eye (100 trials, intertrial interval 3–4 s). Created with Biorender.com. b Stereotaxic coordinates of ECoG electrodes from 13 animals (color coded by animal). V1 and PPA targets for laminar probes are shown by red and blue diamond respectively. The white circle marks bregma. The cortical surface is shaded by area according to the brain regions represented in the 3D Brain Explorer of the Allen Brain Atlas^,,: visual (orange), association (red), retrosplenial (yellow), somatosensory (green), motor/frontal (blue), and cerebellum (gray). c Single trials, average, and standard deviation of visual-evoked potentials (VEPs) over V1 are indicated by gray, solid red, and dashed red lines respectively. Stimulus onset is denoted by the green line. d Intertrial phase coherence (ITPC) computed at V1 and averaged over single trials and animals (0 ms marks stimulus onset). e Single trials, average, and standard deviation of visual-evoked potentials (VEPs) over PPA are indicated by gray, solid blue, and dashed blue lines respectively. Stimulus onset is denoted by the green line. f Intertrial phase coherence (ITPC) computed at PPA and averaged over single trials and animals (0 ms marks stimulus onset). g Average ITPC of 30–50 Hz oscillations within the first 100 ms of the VEP averaged over animals at each stereotaxic location. Locations in which ITPC does not meet Bonferroni corrected statistical significance compared to time shuffled surrogate data are shaded in gray. h Similar to E for average ITPC of 3–6 Hz activity within the first over 800 ms of the VEP. i Top: VEPs recorded over V1 and filtered at fast (30–50 Hz) oscillations (gray and red show single trials and trial average respectively). Bottom: Same data recorded from over PPA (gray and blue show single trials and trial average). Green line shows stimulus onset. Dashed lines highlight the phase offset between V1 and PPA. j Similar to G except the signals are filtered at 3–6 Hz. Slower (3–6 Hz) oscillations also show a phase shift between V1 and the PPA. k Histogram of phase differences between V1 and PPA computed for 30–50 Hz oscillations. l Same as in k but computed for 3–6 Hz oscillations. *Data in c, e, i, j, k, l are from a single representative mouse.

Fig. 2. Average filtered LFP illustrate traveling wave-like behavior.

a Average of the VEP filtered at 30–50 Hz from 10 electrodes along the anterior to posterior axis (−2.25 mm ML) in a representative mouse. The x axis denotes time relative to stimulus onset, the y axis indicates AP position of an electrode relative to bregma. Note evoked high frequency waves starting at −4.05 mm from bregma (V1) and traveling anteriorly over ~100 ms. b Average of the VEP filtered at 3–6 Hz from the same mouse arranged in the same format at b. Note the low frequency waves begin more anteriorly (~2 mm from bregma) relative to the fast oscillations and travel in the posterior direction. c Superimposition of the data in a, b (amplitude of the signals is normalized to highlight phase relationships between oscillations at different temporal frequencies). The fast wave begins posterior to the slow wave and travels rostrally towards the slow wave initiation zone.

Fig. 3. SVD identifies visual-evoked traveling waves that are consistently elicited both across trials and across animals.

a Histogram of the difference in the spatial phase of the most visually responsive 30–50 Hz mode between two electrodes in V1 (black and red diamonds in c) across trials and animals. b Histogram of phase angle difference of the most visually responsive 30–50 Hz mode between an electrode in V1 (black diamond in plot c) and PPA (blue diamond in plot c) across trials and animals. Note that the phase angle difference is increased as distance from V1 increases. c At each stereotaxic location, the average phase offset of the 30–50 Hz spatial mode relative to V1 (the black diamond) is plotted in color. Spatial phase gradient is depicted by black arrows. The direction of the arrows shows the direction of spatial phase gradient over trials and mice. The length of the arrows is 1- circular variance and therefore corresponds to the consistency of the angle of the spatial phase gradient over trials and animals (scale arrow for 100% phase coherence underneath color axis in d). Locations that are grayed out did not meet Bonferroni corrected statistical significance (p value < 0.0006, Rayleigh test). d Phase offset relative to V1 and the spatial phase gradient of the most visually responsive 3–6 Hz spatial mode at each stereotaxic location depicted as in d.

Fig. 4. Intra-laminar recordings reveal vertical propagation pattern of waves.

a Photograph of 64 channel electrocorticography grid with two 32 channel penetrating laminar probes placed in through holes in the grid into V1 and PPA. b Histological verification of laminar electrode localization in V1. Outline adapted from the Allen Mouse Brain Atlas and Allen Reference Atlas—Mouse Brain^,,. c Current source density (CSD) in V1 averaged over trials and mice. d V1 CSD filtered at 30–50 Hz and averaged across trials in a representative mouse. 30–50 Hz oscillations originate in layer IV and then propagate to superficial and deep cortical layers. In d, e purple lines show approximate location of layer IV defined by the earliest sink in the CSD. e Same as d but filtered at 3–6 Hz. The 3–6 Hz oscillations appear approximately simultaneously in the supra- and granular layers and propagate to deeper layers. f Histological verification of laminar electrode localization in PPA. Outline adapted from the Allen Mouse Brain Atlas and Allen Reference Atlas—Mouse Brain^,,. g CSD in PPA averaged over trials and mice. Purple lines show approximate location of superficial layers I–IV, and deep layers V/VI based on anatomy (Allen Brain Atlas). h Same as d for PPA. The 30–50 Hz oscillations are most prominent in the superficial layers. i Same as e for PPA. The 3–6 Hz oscillations begin in the superficial layers and propagate to deeper layers. *Data in d, e, h, i are from the same representative mouse.

Fig. 5. Fast and slow visual-evoked waves have similar spatial wavelengths but significantly different propagation velocities.

a Distribution of spatial wavelengths of 30–50 and 3–6 Hz most visually responsive modes. b Distribution of wave speeds of most visually responsive modes at 30–50 and 3–6 Hz.

Fig. 6. The phase of the slow wave modulates the amplitude of the fast wave both within V1 and throughout the cortex.

a Top: single trials (gray) and average (red) data filtered at 30–50 Hz over V1 of a representative mouse. Middle: same as above, but for 3–6 Hz. Note that fast oscillation bursts occur rhythmically in phase with slow oscillations. Bottom: Amplitude of high frequency oscillations at each phase of the low frequency oscillation, averaged over trials. The deviation of this distribution from a uniform distribution is summarized in the modulation index (MI) of 0.00991 (t = 36.59, p value ~4.9e−324, one-sided student’s t test, df = 99, compared to time shuffled surrogates). b Similar to (a) but for an electrode over PPA. The phase-amplitude MI = 0.00155 (t = 30.24, p value ~4.9e−324, one-sided student’s t test, df = 99, compared to time shuffled surrogates). Note that the phase of the 3–6 Hz oscillation at which the gamma amplitude is maximum is shifted compared to that in V1. c Modulation indices averaged over all 13 mice and plotted in color at each stereotaxic location. Locations that are grayed out did not meet Bonferroni corrected statistical significance (p value < 0.0006, Rayleigh test) compared to time shifted surrogate data. MI peaks near V1 but remains statistically significant over much of the cortical surface. d The phase of the slow 3–6 Hz oscillation at which the fast 30–50 Hz oscillation reaches maximum amplitude as shown for a representative mouse at each stereotaxic location. Grayed out locations did not meet statistical significance (p value < 0.0006, Rayleigh test) compared to time shifted surrogates.

Fig. 7. Probability of neural spiking in both V1 and PPA depend on the phase of the slow wave.

a Individual action potential waveforms of a representative PPA unit located in layer V (gray traces), with the average waveform superimposed in blue. b Same as a, but for a representative V1 neuron located in infragranular layers (gray traces), with the average waveform superimposed in red. c Raster plot (top) of 100 trials of PPA unit (green dashed line marks stimulus onset). The average CSD, filtered at 3–6 Hz, of the LFP at the same depth as the unit in a (middle). The peristimulus histogram of the same unit (bottom). d Same as c, but for the representative V1 unit in b. e Probability of firing as a function of phase of the slow oscillation for each unit in the PPA. Each row is an individual unit in the PPA that has statistically significant spike-field coherence (SFC) with the slow oscillation. Units above the black horizontal line are in the superficial layers of the PPA. Units below the black horizontal line are within the deep layers. f Same as e for V1 units with statistically significant spike-field coherence. The horizontal back lines highlight four sections in which V1 cells reside, in top-down order: layer II/II, layer IV, layer V, and layer VI. g Cross-correlograms between PPA and V1 neurons entrained by the slow wave during the 500 ms before visual stimulation. Each row is an individual PPA V1 pair, organized by laminar location of V1 cell and the depth of the PPA neuron from the surface (purple denotes most superficial to gray denotes deepest layers). Probability of firing is shown by color. h Same as in g but for 500 ms after the stimulus.