Main

Direction selectivity is the capacity of neurons to respond more significantly to one principal direction of stimulus motion than any other5. In the retino-geniculo-cortical pathway, it is first expressed at the level of columnar circuits in the primary visual cortex (V1), where it is organized into a map of direction preference6,7. In ferrets, direction columns emerge shortly after the onset of visual experience in a process that dark-rearing experiments indicate requires visual experience3. To gain insight into the mechanisms by which stimulus-driven neural activity shapes the emerging properties of cortical neurons, we asked whether exposure to a moving visual stimulus was sufficient to induce the development of cortical direction columns and, if so, how this experience altered the response properties of individual cortical neurons.

We began by studying the emergence of direction columns using intrinsic signal imaging techniques in juvenile ferrets (n = 9; postnatal day 30–35) that had less than one day of visual experience. Consistent with previous reports, the visual cortex of these visually naive ferrets exhibited a well-defined system of orientation columns, but lacked the columnar pattern of direction-selective responses that normally develops to mature levels 7–10 days following eye opening3 (Fig. 1). To determine whether exposure to a moving visual stimulus was sufficient to induce the emergence of direction-selective responses, animals were exposed to a ‘training’ stimulus, namely a single sine- or square-wave grating (spatial frequency, 0.06–0.08 cycles per degree; temporal frequency, 4 Hz) drifting back and forth along an axis of motion orthogonal to the grating orientation. As illustrated in two example cases (Fig. 1b, d), no changes were apparent for the first 8–10 h of stimulation; at later times, however, small domains of directional preference became evident in difference images. This direction difference signal continued to intensify and the number of emergent domains increased across the area of the responsive cortex as the experiments continued. The strength of the direction difference signal was quantified by calculating a direction selectivity index (DSI; see Supplementary Methods), and a notable and progressive increase in DSI accompanied exposure to the training stimulus (Fig. 1e).

Figure 1: Rapid emergence of direction columns with motion training.
figure 1

a–d, Visually naive animals (postnatal days 34, 35) with orientation columns (a, c) were trained using moving vertical (b) or horizontal (d) gratings. Direction domains emerged after 8 h. e, Time course of training-induced increases in direction selectivity (top, b; bottom, d). f, After training, direction domains were present only for the trained directions of motion (red arrows) (left, b; right, d). g, Direction selectivity before and after training; colours indicate different animals and arrows indicate median DSI, which increased significantly for trained direction (t-test, P < 0.001; top), but not orthogonal motion directions (t-test, P = 0.12; bottom).

PowerPoint slide

Full size image

Evidence that exposure to the training stimulus was responsible for the rapid emergence of direction selectivity came from experiments, performed following the training period, in which cortical responses to eight directions of motion were examined. Direction-selective responses were evident for stimulus directions that matched that of the training stimulus in seven of nine animals, but were weak or absent for other directions of motion in all animals that were trained (Fig. 1f, g). In comparison with the initial conditions, the average DSI values after training were significantly increased for responses to stimuli whose properties matched the training stimulus (t-test, P < 0.001), but DSI values associated with responses to stimuli orthogonal to the training stimulus were not significantly different from those at the start of training (t-test, P = 0.12). Moreover, at the end of the training period, there was no significant difference in the average response of the trained and untrained orientation columns to a drifting stimulus of the preferred orientation (Supplementary Fig. 1b), indicating that the emergence of direction columns cannot be attributed to a general increase in the activity of the trained columns. Because orientation selectivity is not a prominent feature of pre-cortical sites in the visual pathway (retina, lateral geniculate nucleus)8,9,10,11,12, the orientation selectivity of these training effects indicates that the mechanisms responsible for the rapid emergence of direction selectivity must include events that reside at the level of cortical circuits.

To better understand changes at the cellular level that underlie the emergence of direction columns, we used in vivo two-photon imaging of calcium signals13 to explore the direction selective properties of individual layer-2/3 neurons (Fig. 2a). As a first step, we examined the magnitude of single-neuron direction tuning in visually naive animals and compared this to direction tuning in animals that had visual experience sufficient to achieve mature levels of direction selectivity as measured by intrinsic signal imaging. Neurons in visually naive animals were highly responsive and tuned for stimulus orientation, but were at best only weakly selective for direction of motion (Fig. 2b, c). The median direction index value for visually naive animals was 0.15, and only 6% of neurons in visually naive animals exhibited a direction index value of 0.5, which corresponds to a preferred/opposite response ratio of 2:1 (Fig. 2c). By contrast, many neurons in animals with visual experience were well tuned for direction of motion, such that the median value of the direction index was 0.56. Thus, the normal emergence of direction columns that ensues following eye opening in ferrets reflects a significant increase in the percentage of neurons that exhibit strong tuning for direction of motion.

Figure 2: Direction selectivity of cells in visually naive and experienced ferrets, demonstrated by two-photon calcium imaging.
figure 2

a, Left, average change in fluorescence (ΔF/F) in response to different directions of motion at 160-µm depth in visual cortex of an animal with three weeks of experience; right, cells labelled with Oregon Green 488 BAPTA-1 acetoxymethyl ester, at same depth. Bottom, tuning curves, direction indices (DI) and orientation indices (OI) for cells circled in the right-hand subpanel (dashed lines indicate mean response to grey screen). b, Plots of orientation/direction selective cells in two-photon multi-depth imaging fields from naive and experienced animals. Cells with direction indices <0.5 are depicted with green bars indicating preferred orientation; cells with direction indices ≥0.5 are depicted with red arrows indicating preferred direction. c, Cumulative histograms of direction index and orientation index for naive and experienced animals. Differences between groups are significant (Kruskal–Wallis test, *P < 0.001).

PowerPoint slide

Full size image

We then examined whether the rapid emergence of direction columns that occurs under the influence of a training stimulus is accompanied by an increase in the direction selectivity of individual cortical neurons. In these experiments, we limited our analysis to 262 cells that had significant orientation tuning (vector test; see Supplementary Methods) and could be unambiguously identified both before and after the training period (Fig. 3a). Because the calcium-sensitive dye generally fades over time, the training period was shorter (3–6 h) than that used for intrinsic signal imaging so that reliable tuning curves could be measured at the conclusion of training (Supplementary Methods, Supplementary Figures 2–5).

Figure 3: Motion training increases direction selectivity in individual cells.
figure 3

a, Representative two-photon images (135-µm depth) showing labelled cells evident before (top left) and after (top right) 6-h motion training. Cell history over the course of training (persisted/disappeared/appeared) is depicted in the bottom-middle panel. Tuning curves and direction index (DI) values for circled cells are shown at bottom left (before training) and bottom right (after training). b, Plots of cells from four animals before and after 3–6-h motion or flash training; icons indicate trained directions or orientation of flashing stimulus. c, Cumulative histograms of direction index and orientation index for, variously, naive (N = 8; cell number, 951), motion-trained (N = 5, 262), flash-trained (N = 3, 135) and experienced (N = 5, 551) conditions. Direction index increased significantly following motion training (Kruskal–Wallis test, *P < 0.01), but not after flash training (Kruskal–Wallis test, P = 0.27). Orientation index exhibited a small but significant increase following both motion and flash training (*P < 0.01).

PowerPoint slide

Full size image

The distribution of direction index values measured after the training period was significantly higher than the values measured in the same neurons before training (Fig. 3b, c; Kruskal–Wallis test, P < 0.001): overall, median direction index increased to 0.39, and 36% of cells exhibited direction index values greater than 0.5. On an individual basis, median direction index values increased in all five cases. This training effect was strong enough that statistically significant increases in direction selectivity were observed when the responses of individual cells were compared before and after training (Fig. 3a, Supplementary Figures 2, 3): 59 cells (23%) distributed among four of five animals exhibited a statistically significant increase in direction index over that found before training, and none exhibited a significant decrease in direction selectivity (bootstrap test; see Supplementary Methods).

To test whether motion of the training stimulus was necessary to induce these rapid changes in direction-selective responses, we examined the effects of training with an identical grating stimulus that was flashed (sinusoidal modulation between grating and grey screen, 4 Hz). This stimulus was effective in driving cortical activity (Supplementary Fig. 5a), but no significant increase in direction selectivity was found after flash training (Fig. 3b, c; Kruskal–Wallis test, P = 0.27). Both motion and flash training resulted in small increases in the degree of orientation tuning (Kruskal–Wallis test, P < 0.001) that were equal in magnitude (Fig. 3c; Kruskal–Wallis test, P = 0.29).

These results provide strong evidence that visual experience increases the magnitude of direction selectivity; by itself, this could explain the rapid emergence of columnar structure visualized with intrinsic signal imaging. However, a comparison of the spatial organization of direction preferences in visually naive and experienced animals suggests that visual experience also has a role in determining which direction of motion is preferred by individual cortical neurons (Fig. 4a). Animals with visual experience exhibited a robust clustering of neurons with similar direction preference, consistent with descriptions of the fine-scale mapping of direction preference in cats14. By contrast, in visually naive animals, the spatial organization of direction preferences was weak, and neighbouring neurons often exhibited opposite direction preferences.

Figure 4: Impact of normal experience and motion training on direction preference.
figure 4

a, Plots of direction preference in naive and experienced animals. All cells are represented by arrows, which indicate preferred direction; arrow length indicates magnitude (DI, direction index). Colour differentiates cells with opposite preferences (±90°). b, c, Spatial coherence of direction preference (local coherence index (LCI)) increased—relative to naive (N = 8; cell number, 951)—with experience (N = 5, 551) or motion training (N = 5, 262), but not with flash training (N = 3, 135) (s.e.m. calculated across animals). Significant relations among curves and with distance (analysis of variance), and differences from zero in naive and flash traces (sign test) indicated by asterisks. Error bars are s.e.m. calculated across animals. d, Direction preference of individual cells before (left) and after (right) motion training (numbers refer to cells in e). Cells in circles and squares appeared to reverse their preference, coming to prefer rightward and leftward motion, respectively. e, Distributions of preferred directions in simulations for three cells, after (bottom row) and before (top row) motion training (unc., preference uncertainty; reverse, likelihood of a preference reversal). Some cells (1) initially were uncertain but developed a consistent preference after motion training; other cells exhibited biases that strengthened (2) or reversed (3). f, g, Influence of initial local coherence index on motion training (f) and flash training (g) effects. Cells whose preferred direction differed from their neighbours were most likely to reverse (f). No systematic relationship was observed with flash training (g).

PowerPoint slide

Full size image

To quantify this difference, we developed a measure of local coherence: for each neuron, we calculated the percentage of neighbouring cells with similar direction preferences (within 45°) minus the percentage with opposite preferences (within 45°). The direction preference of weakly selective cells is acutely sensitive to trial-to-trial variability; we accounted for this variability in local coherence values by employing the bootstrap technique to create 100 simulations of each cell’s responses drawn randomly from the original data with replacement. The median local coherence value from these 100 simulations is reported here. In animals with mature direction selectivity and robust columnar structure, average local coherence values were positive and highest for samples within a radius of 50 µm, falling gradually to near zero at distances of 200–250 µm (Fig. 4b, c). The local coherence values for visually naive animals were significantly lower at all distances; nevertheless, they were significantly different from zero, indicating a weak tendency for nearby neurons in visually naive animals to exhibit similar direction preferences. To determine whether experience with moving stimuli was sufficient to explain increases in the spatial coherence of direction preference, we examined the local coherence indices in animals that had undergone training with a motion stimulus and those that received flash training. Motion training produced a significant increase in local coherence values over that found for visually naive animals (Fig. 4c; Kruskal–Wallis test, P < 0.001), whereas flash training produced a significant decrease (Kruskal–Wallis test, P < 0.001).

These results imply that as a result of motion training, the direction of motion preferred by individual cortical neurons changes to become more like that of their neighbours. Examination of the preference of individual neurons before and after training revealed a number of examples in which this appeared to be the case (Fig. 4d). However, this conclusion rests on the confidence with which a direction preference can be assigned to the weakly selective neurons in visually naive animals. From our bootstrap simulations, we derived a measure of uncertainty in direction preference, defined to be the percentage of simulations that differed from the mean direction by more than 90°. We observed a wide range of uncertainty in initial direction preferences. The direction preference of some cells was highly uncertain before motion training but became more certain after training (Fig. 4e, left-hand subpanel). For example, 68 cells with uncertainties greater than 25% before motion training exhibited uncertainties that were less than 25% after training. Other cells exhibited moderate direction preference biases at the onset of training that could be either strengthened (Fig. 4e, middle subpanel) or reversed (Fig. 4e, right-hand subpanel). Of the 148 cells that had initial uncertainties less than 25%, 74 were likely to have maintained their initial direction preference (likelihood of reversing, <25%), whereas 20 were more than 75% likely to have reversed their preference.

We then asked whether the diverse effects of motion training produced a more coherent map of direction preference by building on the weak spatial organization that existed at the onset of training. Consistent with this idea, we found a significant correlation between the likelihood of direction preference reversal and local coherence before training (Fig. 4f; regression F-test, P < 0.001, R2 = 0.22), but not with flash training (Fig. 4g; regression F-test, P = 0.34, R2 = 0.006). Thus, in regions with strong local coherence, there was a predictable impact of motion training: cells that were surrounded by neighbouring neurons expressing the opposite direction preference before training (high negative local coherence) were likely to reverse their direction preference, whereas cells whose neighbours expressed the same preference (high positive local coherence) were unlikely to reverse. Furthermore, the impact of motion training on the likelihood of reversal was unpredictable in regions characterized by weak local coherence (local coherence index values near zero). These findings implicate local cortical interactions in the mediation of the effects of motion training on direction preference. This systematic relationship between the likelihood of preference reversal and initial coherence values also rules out the possibility that changes in uncertainty alone are sufficient to explain the increase in local coherence. Indeed, on average, we estimate that training-induced reversals in direction preference accounted for 53% of the total training-induced increase in spatial coherence, whereas changes in uncertainty accounted for 34%; the remaining 13% was contributed by slight but significant changes in orientation preference that accompanied motion training (Supplementary Methods, Supplementary Figure 4). Although the number of cells that appeared to reverse direction was less than the number of cells that had reduced uncertainty, the impact of a preference reversal on the local coherence index is greater than the impact of a reduction of preference uncertainty.

We conclude that early experience with moving visual stimuli exerts a strong, rapid and selective impact on response properties of developing cortical neurons, transforming a weakly biased array of poorly selective neurons into a more mature state that exhibits stronger direction selectivity and enhanced spatial coherence of direction preference. The rapid time course of these effects and the fact that they are sensitive to the spatio-temporal structure of the stimulus are consistent with activity-dependent mechanisms of synaptic plasticity15,16. However, what differentiates these observations from most previous demonstrations of activity-dependent alterations in stimulus preference17,18,19,20,21,22,23,24,25 is that the information present in the training stimulus is ambiguous: opposite directions of motion are presented and yet neurons rapidly acquire and/or strengthen their preference for a single direction. Thus, the spatio-temporal cues present in a bidirectional motion training stimulus, which is more consistent with the balanced stimulation an animal might receive in nature, are sufficient to drive the development of cortical circuits that represent each direction of motion. Evidently, sufficient asymmetry in functional architecture exists to facilitate symmetry-breaking and seed the formation of direction columns from stimulus patterns that are equally balanced. Our spatial analysis has identified a constructive mechanism that operates locally and is reflected in the preferences of neighbouring neurons in layer 2/3. This mechanism—possibly mediated by local recurrent or feedforward circuits—effectively disambiguates the bidirectional motion energy of the training stimulus by influencing the probability that a neuron’s initial preference will be reinforced or reversed.

Whether the weak direction bias present at the onset of training emerges through visual experience through closed lids26, endogenous activity27,28 or activity-independent mechanisms29,30 remains unclear. Nevertheless, the evidence presented here indicates that the events preceding eye opening are insufficient to account for either the magnitude of direction selectivity or the preference exhibited by mature cortical neurons, and that early experience with moving stimuli has a significant impact on both features.

Methods Summary

Ferrets were anaesthetized with ketamine (50 mg kg-1) and isoflurane (2% for surgery, 0.8–1% during imaging). The training protocol consisted of a 5-s stimulation followed by a 10-s interstimulus interval. The protocol continued for 20 min and was followed by 10 min of no stimulation. This entire procedure was repeated for several hours. For intrinsic signal imaging experiments, cortex was illuminated with 610-nm light and data was acquired using the Imager 2001/3001 (Optical Imaging)3. For two-photon experiments, Oregon Green 488 BAPTA-1 acetoxymethyl ester (Invitrogen) was pressure injected into cortex and changes in calcium fluorescence were monitored with an Ultima IV two-photon microscope (Prairie Technologies) driven by a mode-locked Chameleon laser (810 nm, Coherent). Stimulation and analysis were performed using custom software for Matlab (Mathworks). See Supplementary Methods for details.