Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2022 Mar 29;38(13):110606.
doi: 10.1016/j.celrep.2022.110606.

Coarse-to-fine processing drives the efficient coding of natural scenes in mouse visual cortex

Rolf Skyberg  1 , Seiji Tanabe  1 , Hui Chen  1 , Jianhua Cang  2
Affiliations

Coarse-to-fine processing drives the efficient coding of natural scenes in mouse visual cortex

Rolf Skyberg et al. Cell Rep. .

Abstract

The visual system processes sensory inputs sequentially, perceiving coarse information before fine details. Here we study the neural basis of coarse-to-fine processing and its computational benefits in natural vision. We find that primary visual cortical neurons in awake mice respond to natural scenes in a coarse-to-fine manner, primarily driven by individual neurons rapidly shifting their spatial frequency preference from low to high over a brief response period. This shift transforms the population response in a way that counteracts the statistical regularities of natural scenes, thereby reducing redundancy and generating a more efficient neural representation. The increase in representational efficiency does not occur in either dark-reared or anesthetized mice, which show significantly attenuated coarse-to-fine spatial processing. Collectively, these results illustrate that coarse-to-fine processing is state dependent, develops postnatally via visual experience, and provides a computational advantage by generating more efficient representations of the complex spatial statistics of ethologically relevant natural scenes.

Keywords: CP: Neuroscience; experience-dependent development; natural scene statistics; population coding; single-unit recording; spatial frequency tuning; temporal dynamics; visual neuroscience.

PubMed Disclaimer

Conflict of interest statement

Declaration of interests The authors declare no competing interests.

Figures

Figure 1.
Figure 1.. Mapping spatiotemporal receptive fields of mouse V1 cells
(A) Diagram of the subspace forward correlation method. Top, the stimulus is a rapid sequence of sinusoidal gratings of varying spatial frequencies (SF), orientations, and phases. Each neuron’s spike responses following the onset of all repeats of a given stimuli (black) or blanks (gray) are binned and summed into individual peri-stimulus time histograms (PSTHs). Bottom, example PSTHs to a particular stimulus (black) and blank (gray). (B) Representative one-peak cell from V1 of an awake mouse. Left, heatmap of responses to each SF. Black and red brackets indicate corresponding responses plotted to the right. Scale bar denotes firing rate in spikes/s. Right, low-SF (black) and high-SF (red) responses of the same cell. c/d, cycles/degree. (C) Same as (B) but for a representative two-peak cell. (D) Distribution of latencies to peak response for cells from awake mice. Orange, one-peak cells; dark blue, two-peak cells first peak (τpk1); light blue, two-peak cells second peak (τpk2). See also Figure S7.
Figure 2.
Figure 2.. State-dependent coarse-to-fine SF processing in mouse V1
(A) Proportion of one- and two-peak cells from V1 of awake (blue) and anesthetized (gray) adult mice. Each point represents one animal (18 awake mice, nine anesthetized mice). (B) Scatterplot comparing the preferred SF at τpk1 and τpk2 of two-peak cells from awake (blue circles) and anesthetized (gray diamonds) mice. Black dotted line marks where the preferred SFs are equal at τpk1 and τpk2. A small amount of scatter is added to the data to differentiate points at the same coordinates. (C) Bottom, cumulative distribution of Δfpk for all cells from awake (blue) and anesthetized (gray) mice. Arrows mark proportion of cells from awake and anesthetized mice with Δfpk ≤ 0. Red dots denote the mean Δfpk values. Top, mean Δfpk values for individual mice (18 awake, nine anesthetized). Statistical comparisons were done using individual mice mean values shown at top of (C). (D) Average normalized SF response at six distinct latencies for all cells from awake mice. 30 ms (black dotted line) is used as baseline. (E) Same as (D) but for all cells from anesthetized mice. Note that slightly different latencies were used here compared with (D) to account for the effects of anesthesia on response timing. (F) Average normalized time-integrated SF tuning of all V1 cells from awake (blue) and anesthetized (gray) mice. Error bars denote SEM throughout. ***p < 0.001. c/d, cycles/degree. See also Figures S2 and S7.
Figure 3.
Figure 3.. Awake mice process natural scenes in a coarse-to-fine manner
(A) Three variants of 60 natural scenes were shown to six awake mice; unfiltered natural scenes with 1/f spatial statistics (black), low-pass filtered natural scenes with attenuated high SFs (yellow), and high-pass filtered natural scenes with attenuated low SFs (purple). (B) Average SF power spectrum of unfiltered (black), low-pass filtered (yellow), and high-pass filtered (purple) natural scenes. SEM were not included in this plot as they were smaller than the thickness of the lines. Note the log10 y axis. (C) Heatmap of responses of one neuron to unfiltered (top) and low-pass (middle) or high-pass filtered (bottom) natural scenes. Scale bar denotes firing rate in spikes/s. (D) The average firing rate of the same cell in (C) to all unfiltered (black), low-pass filtered (yellow), and high-pass filtered (purple) natural scenes. (E–F)Another representative cell plotted in the same way as (C)–(D). (G) Average normalized response of all 171 cells to unfiltered (black), low-pass filtered (yellow), and high-pass filtered (purple) natural scenes. Error bars represent SEM. (H) Average normalized response (y axis) plotted against the average time at the peak response (x axis) of all cell’s responses to unfiltered (black), low-pass filtered (yellow), and high-pass filtered (purple) natural scenes. Error bars represent SEM. (I) Histogram of difference in peak response time between average high-pass and average low-pass responses from each cell. Positive values indicate that high-pass filtered image responses peak later than that of low-pass filtered images. Dotted line demarks 0. See also Figure S3.
Figure 4.
Figure 4.. Coarse-to-fine SF processing reduces redundancy in neural representation of natural scenes
(A) Hypothetical two-neuron example of redundant (black) and efficient (red) neural codes. Top, schematic of responses of a redundant (black) and efficient (red) neural code to five natural scenes. Dotted lines mark the neural response manifold generated by these hypothetical neural codes. Bottom, eigenspectrum generated from these hypothetical redundant (black) and efficient (red) neural codes. (B) Average firing rate of all cells (n = 629) to all natural scenes from V1 of eight awake mice. Solid and dotted lines mark the time points τ1 (121ms) and τ2 (165ms) used for subsequent analyses in (C)–(D). (C) Variance explained by the first six PCs over the response period when measuring the structure of the neural response manifold generated by all 629 cells. Solid and dotted lines mark the time points τ1 (121ms) and τ2 (165ms) used for subsequent analyses in (D). (D) Eigenspectra at 121 ms (solid line) and 165 ms (dotted line). Red line is the eigenspectrum of shuffled dataset averaged overtime. ΔSlope indicates the change in the eigenspectrum slope from τ1 (−0.49) to τ2 (−0.33). Note the log10 x axis. Dashed box indicates portion of eigenspectra plotted in inset. Inset, same eigenspectra but zoomed in to highlight the point at which the eigenspectra of the original data at τ1 and τ2 cross the eigenspectrum of the shuffled data. Dotted vertical lines mark the number of large PCs for τ1 and τ2. ΔLarge PCs is the difference in the number of large PCs from τ1 to τ2. (E) Time at τ1 and τ2 for individual recordings from V1 of eight different awake mice. (F) Eigenspectrum slope at τ1 and τ2 for individual mice (gray points connected by line). Blue bar plots indicate mean values. Error bars denote SEM. (G) Same as (F) but for the number of large PCs at τ1 and τ2 for individual mice. (H) Same as (F)–(G) but for the percentage of total variance explained by PC1 at τ1 and τ2 for individual mice. **p < 0.01. PC, principal component. See also Figure S4.
Figure 5.
Figure 5.. Experience-dependent development of coarse-to-fine SF processing
(A) Timeline of visual development in normally reared (top) and dark-reared mice (bottom). (B) Proportion of one- and two-peak cells in V1 of normally reared adult (blue), P17 (light blue), and dark-reared adult (black) mice. Each dot represents one mouse (18 adult mice, 12 DR mice, 10 P17 mice). (C) Bottom, cumulative distribution of Δfpk for all cells from normally reared adult (blue), P17 (light blue), and dark-reared adult (black) mice. Arrows mark proportion of cells with Δfpk ≤ 0. Red dots denote the mean Δfpk values. Top, mean Δfpk values for individual mice (18 adult mice, 12 DR mice, 10 P17 mice). Statistical comparisons illustrated in legend were done using individual mice mean values shown in top of (C). (D) Average normalized time-integrated SF tuning of the three groups. Error bars denote SEM throughout. ***p < 0.001. c/d, cycles/degree. See also Figures S5 and S7.
Figure 6.
Figure 6.. Disrupted redundancy reduction in natural scene responses from mice with attenuated coarse-to-fine SF processing
(A) Average eigenspectrum slope at τ1 and τ2 for individual recordings from V1 of six dark-reared mice (white bars) and V1 of six anesthetized mice (gray bars). Error bars denote SEM. Values from individual mice are marked with gray points. (B) Same as (A) but for number of large PCs at τ1 and τ2. (C) Same as (A)–(B) but for the percentage of total variance explained by PC1 at τ1 and τ2. (D) Average ΔSlope for normally reared (blue), dark-reared (white), and anesthetized (gray) mice. Error bars denote SEM. Values from individual mice are marked with gray points. (E) Same as (D) but for ΔLarge PCs. (F) Same as (D)–(E) but for ΔPercentage of total variance explained. *p < 0.05, **p < 0.01. c/d, cycles/degree; PC, principal component; ns, not significant. See also Figure S6.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Allen EA, and Freeman RD (2006). Dynamic spatial processing originates in early visual pathways. J. Neurosci 26, 11763–11774. - PMC - PubMed
    1. Andermann ML, Kerlin AM, Roumis DK, Glickfeld LL, and Reid RC (2011). Functional specialization of mouse higher visual cortical areas. Neuron 72, 1025–1039. - PMC - PubMed
    1. Atick JJ, and Redlich AN (1990). Towards a theory of early visual processing. Neural Comput 2, 308–320.
    1. Attneave F (1954). Some informational aspects of visual perception. Psychol. Rev 61, 183–193. - PubMed
    1. Ayaz A, Saleem AB, Schölvinck ML, and Carandini M (2013). Locomotion controls spatial integration in mouse visual cortex. Curr. Biol 23,890–894. - PMC - PubMed

Publication types

LinkOut - more resources