Research ArticleNEUROSCIENCE

Unsupervised experience with temporal continuity of the visual environment is causally involved in the development of V1 complex cells

See allHide authors and affiliations

Science Advances  29 May 2020:
Vol. 6, no. 22, eaba3742
DOI: 10.1126/sciadv.aba3742
  • Fig. 1 Experimental design.

    (A) Two groups of rats, control (top) and experimental (bottom), were born in dark and housed in lightproof cabinets until EO (black bars). Afterward, the control rats were subjected to daily 4-hour-long exposures to natural videos inside the virtual cages (blue bar), while the experimental rats were subjected to the frame-scrambled versions of the same movies (orange bar). Starting from P60, neuronal recordings from V1 of both the control and experimental rats were performed under anesthesia (gray bars), while the animals were exposed to drifting gratings and movies of spatially and temporally correlated noise. (B) Left: The mean correlation between the image frames of one of the natural movies [same as in (A)] is plotted as a function of their temporal lag (blue curve). The dashed line shows the best exponential fit to the resulting autocorrelation function (τ is the time constant of the fit). Right: The autocorrelation function obtained for the frame-scrambled version of the movie (orange curve) is shown along with its best exponential fit (dashed line). (C) Left: The autocorrelation functions of all the natural movies used during postnatal rearing of the control rats (blue curves) are shown along with the autocorrelation functions of their frame-scrambled versions (orange curves) used during postnatal rearing of the experimental rats. Right: The time constants of the best exponential fits to the autocorrelation functions of the natural movies (blue dots) and their frame-scrambled versions (orange dots) were significantly different (P < 0.001, one-tailed, unpaired t test). Photo credit: Giulio Matteucci and Mattia D’Andola, SISSA (Trieste, Italy).

  • Fig. 2 Postnatal rearing in temporally discontinuous visual environments results in an impoverished population of V1 complex cells but spares the development of orientation tuning.

    (A) A representative V1 complex cell of the control group (left, blue lines) is compared to a representative simple cell of the experimental group (right, orange lines). For each neuron, the graph shows, from top/left to bottom, (i) the linear RF structure inferred through STA, (ii) the direction tuning curve, (iii) the raster plot with the number of spikes (dots) fired across repeated presentations of the most effective grating stimulus, (iv) the corresponding peristimulus time histogram (PSTH) computed in 10-ms-wide time bins, and (v) its power spectrum with its mean (dotted line), its mean + SD (dashed line), and the 4-Hz frequency of the grating stimulus (vertical line) indicated. (B) Left: Distributions of the MI used to distinguish the poorly phase-modulated complex cells (MI < 3; gray-shaded area) from the strongly modulated simple cells (MI > 3), as obtained for the control (blue; n = 50) and experimental (orange; n = 75) V1 populations (only units with an OSI of >0.4 included). Both the distributions and their medians (dashed lines) were significantly different (P < 0.02, Kolmogorov-Smirnov test; ***P < 0.001, Wilcoxon test). Right: The fraction of units that were classified as complex cells (i.e., with an MI of <3) was significantly larger for the control than for the experimental group (***P < 0.001, Fisher’s exact test). (C) Left: Distributions of the orientation selectivity index (OSI), as obtained for the control (blue; n = 105) and experimental (orange; n = 158) V1 populations. No significant difference was found between the two distributions and their medians (P > 0.05, Kolmogorov-Smirnov test; P > 0.05, Wilcoxon test). Right: The fraction of sharply orientation-tuned units (i.e., units with an OSI of >0.6) did not differ between the two groups (P > 0.05, Fisher’s exact test). n.s., not significant.

  • Fig. 3 Unaltered spatial tuning of V1 neurons following the controlled rearing in temporally discontinuous visual environments.

    (A) Examples of linear RFs inferred through STA for the control (blue frame) and experimental group (orange frame). In every STA image, each pixel intensity value was independently z-scored on the basis of the null distribution of STA values obtained through a permutation test (see Materials and Methods). (B) Left: Distributions of the CI used to measure the sharpness of the STA images, as obtained for the control (blue; n = 105) and experimental (orange; n = 158) V1 populations. No significant difference was found between the two distributions and their medians (P > 0.05, Kolmogorov-Smirnov test; P > 0.05, Wilcoxon test). Right: Mean values (± SEM) of the CIs computed separately for the simple (dark bars; n = 20, control; n = 49, experimental) and complex (light bars; n = 30, control; n = 26, experimental) cells of the two groups (only units with an OSI of >0.4 included). Within each group, the mean CI was significantly larger for the simple than for the complex cells (**P < 0.01, two-tailed unpaired t test). (C) The color maps showing the distributions of lobe counts (abscissa) for the units with well-defined linear RFs [i.e., within the top quartiles of the CI distributions shown in (B)] in the control (blue; n = 27) and experimental (orange; n = 37) populations are plotted as a function of the binarization threshold (ordinate) used by the lobe-counting algorithm (see Materials and Methods). For every choice of the threshold, the control and experimental distributions were not significantly different (P > 0.05, Fisher’s exact test). (D) Same analysis as in (C) but applied to the distributions of RF sizes obtained for the control (blue; n = 27) and experimental (orange; n = 37) populations, as a function of the binarization threshold. Again, no significant difference was found between the two distributions at any threshold level (P > 0.05; Fisher’s exact test).

  • Fig. 4 Postnatal rearing in temporally discontinuous visual environments leads to the development of complex cells with abnormally fast response dynamics.

    (A) Top: PSTHs showing the average responses of the two example neurons of Fig. 2A to the contrast-modulated noise movies (see Materials and Methods). For both neurons, the firing rate was strongly modulated at the frequency of variation of the contrast of the movies (i.e., 0.1 Hz). Bottom: Distributions of interspike intervals (ISIs) of the spike trains evoked by the noise movies for the two example neurons. The resulting autocorrelograms were fitted with exponential decaying functions (dashed lines) to measure the slowness (i.e., the time constant τ of the exponential fit) of the responses. In this example, the complex cell (blue curve) displays slower dynamics (i.e., larger τ) than the simple cell (orange curve). The two units also differ in the number of counts at low ISIs, which is much larger for the simple cell, as expected for a unit firing tightly packed trains of spikes (see Fig. 2A). (B) Mean values (± SEM) of the time constants τ computed for the control (blue; n = 92) and experimental (orange; n = 143) populations (***P < 0.001, two-tailed unpaired t test). (C) Mean values (± SEM) of the time constants τ computed separately for the simple (dark bars; n = 43, control; n = 89, experimental) and complex (light bars; n = 49, control; n = 54, experimental) cells of the two groups. While the simple cells had equally fast dynamics, the complex cells were significantly slower in the control than in the experimental group (**P < 0.01, two-tailed unpaired t test).

  • Fig. 5 Postnatal rearing in temporally discontinuous visual environments reduces the ability of complex cells to support phase-tolerant discrimination of grating orientation.

    (A) The cartoon illustrates the expected outcome of the decoding analysis to test the ability of simple and complex cells to support phase-tolerant discrimination of grating orientation. In the case of simple cells (top), a linear classifier built at time t0 (middle; light gray shading) to successfully discriminate a vertical from a horizontal drifting grating (left; the filled and empty dots are well separated, within the neuronal representational space, by the linear decision boundary) will generalize poorly when tested at a later time t1 (middle; dark gray shading), with the accuracy dropping even below chance (right; the filled and empty dots swap sides of the linear decision boundary) due to the strong phase dependency of the responses ri (middle; some neurons firing at t0 will stop firing at t1, while some other units that are silent at t0 will respond at t1). By contrast, for a population of complex cells (bottom), given the greater stability of the responses ri (middle), the decision boundary resulting from the training at t0 (left) will generalize better at t1 (right; the filled and empty dots are still mostly on the original side of the decision boundary). (B) Decoding accuracy yielded by the four populations of control simple (dark blue), control complex (light blue), experimental simple (brown), and experimental complex (orange) cells in the vertical (i.e., 90°) versus horizontal (i.e., 0°) grating discrimination task in 33-ms-wide test bins located at increasingly larger time lags from the training bin (i.e., bin with a lag of 0). The solid curves are the averages of many resampling loops of the neuronal population vectors and the training bins (see Materials and Methods). The shaded regions are the bootstrap-estimated 95% confidence intervals of the solid lines (see Materials and Methods).

Supplementary Materials

  • Supplementary Materials

    Unsupervised experience with temporal continuity of the visual environment is causally involved in the development of V1 complex cells

    Giulio Matteucci and Davide Zoccolan

    Download Supplement

    This PDF file includes:

    • Figs. S1 to S5

    Files in this Data Supplement:

Stay Connected to Science Advances

Navigate This Article