Perceptual Learning Improves Contrast Sensitivity of V1 Neurons in Cats

Summary

Background

Perceptual learning has been documented in adult humans over a wide range of tasks. Although the often observed specificity of learning is generally interpreted as evidence for training-induced plasticity in early cortical areas, physiological evidence for training-induced changes in early visual cortical areas is modest, despite reports of learning-induced changes of cortical activities in fMRI studies. To reveal the physiological bases of perceptual learning, we combined psychophysical measurements with extracellular single-unit recording under anesthetized preparations, and examined the effects of training in grating orientation identification on both perceptual and neuronal contrast sensitivity functions of cats.

Results

We have found that training significantly improved perceptual contrast sensitivity of the cats to gratings with the spatial frequencies near the ‘trained’ spatial frequency, with stronger effects in the trained eye. Consistent with behavioral assessments, the mean contrast sensitivity of neurons recorded from V1 of the trained cats was significantly higher than that of neurons recorded from the untrained cats. Furthermore, in the trained cats, the contrast sensitivity of V1 neurons responding preferentially to stimuli presented via the trained eyes was significantly greater than that of neurons responding preferentially to stimuli presented via the ‘untrained’ eyes. The effect was confined to the trained spatial frequencies. In both trained and untrained cats, the neuronal contrast sensitivity functions derived from the contrast sensitivity of the individual neurons were highly correlated with behaviorally determined perceptual contrast sensitivity functions.

Conclusions

We suggest that training-induced neuronal contrast-gain in area V1 underlies behaviorally determined perceptual contrast sensitivity improvements.

Introduction

Perceptual learning has been documented over a wide range of perceptual tasks [1-4]. The observed specificity to the trained task or stimulus in perceptual learning has been generally interpreted as evidence for representation enhancement in early cortical areas [2, 5, 6]. On the other hand, while cortical plasticity following extended practice has been documented in both auditory and somato-sensory cortices [7, 8], evidence for such plasticity in early visual cortices is modest in neurophysiology, although some evidence has been reported in several functional magnetic resonance imaging (fMRI) studies [9-11].

Single-unit recording from monkey early visual areas [12-15] has demonstrated that perceptual learning is not clearly associated with increased neuronal recruitment, or major changes of receptive field parameters in V1 and V2 [13, 14]. Schoups et al. [14] reported changes of the slopes of the neuronal tuning curves, but the magnitude of the changes did not provide a compelling account of the large behavioral improvements. Ghose et al. [13] found no such changes in tuning curves in early visual areas, while only modest sharpening of tuning curves has been reported in V4 [15]. Other task-specific tuning changes observed in V1 [16], which may reflect selection of task-relevant stimulus features by attention, are incompatible with the representation enhancement hypothesis that predicts persistent and task-independent tuning changes. Law and Gold [17] found that perceptual learning in motion direction discrimination does not involve neuronal response changes in the middle temporal area (MT), but rather in the lateral intraparietal area (LIP), a brain area related to selective readout of MT neurons. Finally, the related literature on cortical plasticity following lesions suggests that sensory-cortical recruitment or remapping is nearly absent in V1 [18, 19]. In sum, these reports found that early visual representations showed either no change or modest changes in the slopes of tuning functions following perceptual learning.

In this study, we investigated the physiological bases of perceptual learning in adult cats. Cats have a highly developed visual system and have been widely used as an animal model in visual neuroscience [20, 21]. We have recently developed an effective training paradigm to compare perception and neural activity [22-24]. Existing neurophysiological investigations of cortical plasticity related to perceptual learning have primarily used non-primate subjects in auditory and somato-sensory studies and primate subjects in visual studies. It's possible that primates exhibit less training-induced plasticity than non-primates in early visual areas [25, 26].

Previous neurophysiological studies on visual perceptual learning have predominantly used orientation discrimination tasks, which fix grating contrast and vary grating orientation to measure orientation thresholds [but see [17]]. We trained cats to identify gratings at two fixed, widely separated grating orientations (±45 deg), and varied grating contrast to measure contrast sensitivities [27-29]. Human subjects have demonstrated significant learning-induced improvements in contrast sensitivity functions that are specific to the trained spatial frequency and partially specific to the trained eye, suggesting primary visual cortex as the possible locus of learning [4, 30].

Conditioning was used to train two cats to identify the orientation of a high contrast ±45° sinusoidal grating (Fig. 1). Subsequently, the same procedure was used to measure monocular contrast sensitivity functions (CSF) in both eyes. The cats were then trained monocularly to perform a near-threshold orientation identification task (Supplementary Fig. S1). After approximately forty days of training, monocular CSFs were measured again, followed by extracellular recordings of single-unit activities from the primary visual cortex (V1) of anesthetized cats. Contrast response functions to the preferred stimuli were measured for isolated neurons. The combined contrast sensitivities of individual neurons were then used to construct the neuronal CSFs for neuronal populations that responded preferentially to the stimuli presented via trained or untrained eyes.

Fig. 1.

Visual stimuli used in conditioning training of cat1 (A; 0.2 c/deg) and cat2 (B; 0.4 c/deg). All stimuli are 80% contrast sine wave gratings oriented at ±45 degrees.

We found that (1) training improved perceptual contrast sensitivity, with some degree of specificity for the training spatial frequency and training eye, (2) training also improved the contrast sensitivity of V1 neurons responding preferentially to the trained spatial frequency, (3) perceptual and neuronal CSFs were highly correlated both before and after training, and (4) training increased neuronal contrast-gain.

Results

Perceptual learning of contrast detection

Both cats completed conditioning training in 3-4 months. Prior to near- threshold training, the cats performed at 94±4% correct on average (range: 90-100%) in the high contrast test across the full range of spatial frequencies. Their CSFs exhibited a significant main effect of spatial frequency (Cat1: F(6,84)=105.11; Cat2: F(6,84)=142.35; both p<0.0001), but no significant effect of eye (Cat1: F(1,84)=0.163; Cat2: F(1,84)=0.057, both p>0.5), nor frequency/eye interaction (cat 1: F(6,84)=0.164; cat 2: F(6,84)=0.215, both p>0.5) (Fig. 2).

Fig. 2.

Contrast sensitivity functions in the trained and untrained eyes before and after training for cat1 (A) and cat2 (B). Smooth curves represent the best fitting Gauss functions. The green arrows indicate the trained spatial frequency, and the error bars represent 1 SD.

Training significantly increased contrast sensitivity at the trained spatial frequency in the trained eye for both cats (Fig. 3): sensitivity at the trained spatial frequency increased from 17.15±2.83 (mean±SD) to 55.29±7.38 for cat1 (F(1,12)=162.55, p<0.0001), and from 12.65±3.43 to 39.76±5.33 for cat2 (F(1,12)=128.01, p<0.0001). After training, the cats performed at 96±4% correct on average (range: 89-100%) in the high contrast tests, comparable to the pre-training performance levels (p>0.25).

Fig. 3.

Contrast sensitivity at the trained spatial frequency (expressed as mean ± SD) versus training days for cat1 (filled circles) and cat2 (open circles).

Training also significantly improved the CSFs in the trained eye of both cats (cat1: F(1,84)=285.14, cat2: F(1,84)=317.7; both p<0.0001). The magnitude of learning depended significantly on spatial frequency (cat1: F(6,84)=73.55; cat2: F(6,84)=56.53; both p<0.0001) (Figs. 2A & B): the ratio of post- versus pre-training sensitivity ranged between 1.15 and 3.31 with maximal improvement at 0.4 c/deg for cat1, and between 1.06 and 3.37 with maximal improvement at 0.6 c/deg for cat2. That the maximal sensitivity improvement was observed at the trained spatial frequency is indicative of spatial frequency specificity of perceptual learning [4, 30].

Perceptual learning in the trained eye also partially transferred to the untrained eye for both cats: Contrast sensitivity of the untrained eye at the trained spatial frequency improved from 16.53±3.02 to 31.94±3.81 for cat1 (F(1,12)=70.35, p<0.0001), and from 13.56±2.94 to 27.81±4.09 for cat2 (F(1,12)=55.9, p<0.0001), although the magnitude of improvement was significantly smaller than that in the trained eye for both cats (Cat1: F(1,12)=55.18, p<0.0001; Cat2: F(1,12)=22.13; p<0.01) (Figs. 2A&B). In fact, partial transfer of perceptual learning happened in all spatial frequencies, i.e., the CSFs in the untrained eye improved following training (Cat1: F(1,84)=80.98; Cat2: F(1,84)=112.16; both p<0.0001) with significant training/spatial frequency interactions (Cat1: F(6,84)=23.77; Cat2: F(6,84)=21.19; both P<0.0001) (Figs. 2A & B). The improvements occurred largely around the trained spatial frequency of each cat, around 0.2-0.6 c/deg for cat1 and 0.4-0.8 c/deg for cat2. The ratio of post-versus pre-training sensitivity ranged between 1.02 and 1.99 with maximal improvement at 0.4 c/deg in cat1, and between 1.02 and 2.09 with the maximal improvement at 0.6 c/deg in cat2.

In summary, training greatly increased contrast sensitivity at the trained spatial frequency in the trained eye, with a certain degree of specificity for spatial frequency and eye, and significant partial transfer to untrained spatial frequencies and the untrained eye. Because the cats performed the task with high contrast stimuli presented to either eye and across a wide range of spatial frequencies at a comparable level prior to and after training, the specificity results were not due to their inability to perform the task in different spatial frequencies or with untrained eyes before training.

Training-induced plasticity of V1 neurons in trained cats

We systematically compared contrast sensitivities of V1 neurons of the trained and untrained cats, as well as neurons responding preferentially to stimuli presented via the trained and untrained eyes of the trained cats. A total of 142 and 117 cells in the trained and untrained cats were studied (Table 1). Cells recorded from each group of cats were at the same range of depth from the pial surface of the brain, representing random samples of neurons in all cortical layers. All cells had receptive fields within 8 degrees from the area centralis. The eccentricity distribution of the receptive fields of cells recorded from V1 of trained cats was not significantly different from that of untrained cats (χ²(15) = 10.725, p>0.5). Similarly, the eccentricity distributions of the receptive fields of cells responding preferentially to stimuli presented via the trained and untrained eyes of the trained cats were not significantly different (Cat1: χ²(15) = 21.614, p>0.10; Cat2: χ²(15) = 22.857, p=0.09). In addition, the distributions of preferred spatial frequency and orientation were not significantly different between cells recorded from the trained and untrained cats (χ²(34) = 45.41, p=0.09), cells responding preferentially to stimuli presented via the trained and untrained eyes of the trained cats (χ²(30) = 23.560, p>0.50), and cells responding preferentially to stimuli presented via the trained eyes of the trained cats and cells from the untrained cats (χ²(33)=40.17, p >0.15).

Table 1.

Contrast sensitivity of V1 neurons.

Subjects	Eye	CS	Spatial frequency (cycles/deg)
			0.1	0.2	0.4	0.6	0.8	1.2	1.6
Control cats	Both	N	15	28	32	18	8	9	7
		TCS	8.6±1.3	11.2±1.6	16.4±2.3	8.8±1.8	4.5±0.7	4.1±0.7	4.9±1.3
		C50CS	2.9±0.4	4.1±0.6	5.1±0.6	3.0±0.3	2.6±0.3	1.9±0.2	1.8±0.3
Trained cat1	Trained	N	9	13	15	6	3	3	2
		TCS	8.8±2.1	26.4±6.7	50.3±6.7	13.6±2.2	6.4±0.9	4.7±1.0	4.2±1.1
		C50CS	3.8±0.3	6.0±0.4	13.3±1.2	5.4±0.6	2.7±0.3	2.9±0.6	2.6±0.8
	Naive	N	4	7	13	7	6	3	2
		TCS	7.3±2.6	20.3±3.8	27.5±7.2	11.1±2.5	5.3±0.7	4.2±2.0	4.3±2.1
		C50CS	3.3±0.9	4.8±0.6	8.6±1.4	4.6±0.8	3.1±0.5	2.5±1.1	2.6±1.1
Trained cat2	Trained	N	2	3	9	7	3	2	2
		TCS	8.7±1.6	10.0±1.5	47.7±3.9	34.1±1.8	11.9±2.0	2.9±0.5	2.7±0.2
		C50CS	2.8±0.3	4.4±1.3	11.4±1.2	11.3±1.0	4.6±0.5	1.6±0.1	1.5±0.2
	Naive	N	2	3	5	5	2	2	2
		TCS	9.1±0.8	10.4±2.2	31.8±3.2	18.6±2.7	8.1±0.2	3.8±1.2	3.8±1.5
		C50CS	3.0±0.6	4.1±0.6	7.8±1.0	6.5±0.6	4.7±1.3	1.9±0.3	1.7±0.2

Note: CS represents contrast sensitivity. TCS and C₅₀CS are TC-contrast sensitivity (1/TC) and C₅₀-contrast sensitivity (1/C₅₀), respectively. Their values are expressed as Mean ± SEM (standard error of mean). N: Number of cells.

Comparison between trained and untrained cats

We first compared the contrast sensitivities of neurons from the trained and untrained cats. Neurons were grouped by their preferred spatial frequencies. We ignored their preferred orientations, temporal frequencies, and motion directions because perceptual learning in contrast detection is largely non-specific in those dimensions [4, 13, 31]. Additional analysis on the response characteristics of the neurons with preferred orientation near and away from the trained orientations was also performed. For each neuron, two measures of contrast sensitivity were computed from its contrast response function (Fig. 4): TC-contrast sensitivity is the inverse of each neuron's threshold contrast, which evokes 1.414× its spontaneous activity. C50-contrast sensitivity is defined as the inverse of C₅₀, which evokes half of a cell's maximal response (Supplementary Figs. S2 & S3). Comparisons were made between cells from each trained cat and all the untrained cats.

Fig. 4.

A typical neuron's response to its optimal visual stimulus. A: The voltage trace of the neuron's response to the optimal stimulus at 64% contrast. A spike with amplitude surpassing the horizontal broken line is counted as an action potential. The neuron's response is evoked by 5 cycles of grating stimulation, equivalent to a stimulus duration of about 1.7 seconds, and the spontaneous activity (M) is acquired 1 second prior to visual stimulus presentation. The arrowhead indicates the stimulus onset time. B: Contrast response function of the neuron (mean±SD). The smooth curve represents the best fitting Naka-Rushton equation (r²=99.5%). M and R_max represent the neuron's spontaneous activity and maximal visually-evoked response to visual stimuli. TC (threshold stimulus contrast) represents the stimulus contrast that evokes a neuron's response that is 1.414 times of its spontaneous activity. C₅₀ corresponds to the stimulus contrast that evokes half of the neuron's maximal response. N represents the slope of the neuron's response-contrast tuning curve.

The mean TC-contrast sensitivity of neurons recorded from the trained cats was significantly higher than that of cells from the untrained cats (Cat1: F(1,196)=6.181, p<0.05; Cat2: F(1,152)=15.374, p<0.0001), with strong dependence on spatial frequency (Cat1: F(6,196)=4.056, p<0.001; Cat2: F(6,152)=9.591, p<0.0001) (Fig. 5A-B, Table 1). Similarly, the mean C₅₀-contrast sensitivity of neurons recorded from the trained cats was also significantly increased compared with that of cells from the untrained cats (Cat1: F(1,196)= 13.131, p<0.0001; Cat2: F(1,152)=14.873, p<0.0001), also with strong dependence on spatial frequency (Cat1: F(6,196)=5.171, p<0.0001; Cat2: F(6,152)=5.901, p<0.0001) (Fig. 5C-D, Table 1, Supplementary Fig. S3).

Fig. 5.

(AB) TC-contrast sensitivity functions of V1 neurons recorded from cat1 (A) and cat2 (B). (CD) C₅₀-contrast sensitivity functions of V1 neurons recorded from cat1 (C) and cat2 (D). Green arrows indicate the trained spatial frequency. All values are displayed as mean ± SEM.

Eye and spatial frequency specificity of training-induced plasticity of V₁ neurons

For both trained cats, significant differences were found between the mean contrast sensitivity of neurons responding preferentially to stimuli presented via the trained and untrained eyes of the trained cats at their respective training spatial frequency, 0.4 c/deg for cat1 (TC: F(1,26)=5.32. p<0.03; C₅₀: F(1,26)=6.744, p<0.02) and 0.6 c/deg for cat2 (TC: F(1,10)=24.61, p<0.001; C₅₀: F(1,10)=15.03, p<0.003), although non-significant or marginal difference was found between the mean neuronal contrast sensitivities if neurons responding to the full range of spatial frequencies were included (Cat1: TC: F(1,79)=1.118, p > 0.25; C₅₀: F(1,79)=1.800, p>0.15. Cat2: TC: F(1,35)=4.042, p = 0.05; C₅₀: F(1,35)=2.94, p = 0.09) (Fig. 5A-D, Table 1, Supplementary Fig. S3). We conclude that training-induced plasticity of V1 neurons exhibited a degree of specificity to the trained eye and trained spatial frequency, consistent with our psychophysical results.

Comparing perceptual and neuronal CSFs

We constructed neuronal CSFs by averaging contrast sensitivities of all the neurons with the same preferred spatial frequencies (Fig. 6). For the trained eyes, the pre-training perceptual CSFs of cat1 and cat2 were significantly correlated with the average neuronal CSFs of the three untrained cats using either TC (Cat1: R=0.960; Cat2: R=0.951; both p<0.001) or C₅₀ contrast sensitivity (Cat1: R=0.970; Cat2: R=0.981; both p<0.0001) as the index of neuronal contrast sensitivities. After training, the post-training perceptual CSFs of the trained cats were also significantly correlated with the neuronal CSFs of neurons responding preferentially to stimuli presented via the trained eyes, based on both TC (Cat1: R=0.994; Cat2: R=0.963, both p<0.001) and C₅₀ contrast sensitivity measures (Cat1: R=0.962, p<0.001; Cat2: R=0.977, p<0.0001). Moreover, the observed magnitude of improvements of the perceptual CSF of the trained eye was also highly correlated with changes of the neuronal CSFs of neurons responding preferentially to stimuli presented via the trained eye of each trained cat relative to the three untrained cats, based on TC (Cat1: R=0.906, p<0.01; Cat2: R=0.971, p<0.0001) and C₅₀ contrast sensitivity (Cat1: R=0.700, p<0.05; Cat2: R=0.967, p<0.0001).

Fig. 6.

Scatter plots of mean neuronal contrast sensitivity versus mean psychophysical contrast sensitivity across different spatial frequencies (0.1, 0.2, 0.4, 0.6, 0.8, 1.2 and 1.6 c/deg) for the trained and untrained eyes of the trained cats (red circle: trained eye of trained cat1; purple circle: untrained eye of trained cat1; blue circle: trained eye of trained cat2; green circle: untrained eye of trained cat2). Neuronal contrast sensitivity is based on TC in Panel I, and based on C₅₀ in panel II. (A) and (B) contrast sensitivities before training. (C) and (D) contrast sensitivities after training. Colored lines in each subplot represent the best linear fits (Red: trained eye of trained cat1; Purple: untrained eye of trained cat1; Blue: trained eye of trained cat2; Green: untrained eye of trained cat2.).

In the untrained eyes, the pre-training perceptual CSFs of cat1 and cat2 were significantly correlated with the average neuronal CSFs of the three untrained cats using either TC (Cat1: R=0.952, p<0.0001; Cat2, R=0.945; both p<0.01) or C₅₀ contrast sensitivity (Cat1: R=0.964; Cat2: R=0.965, both p<0.0001) as the index of neuronal contrast sensitivity. After training, the post-training perceptual CSFs in cat1 and cat2 were also significantly correlated with the neuronal CSFs of neurons responding preferentially to stimuli presented via the corresponding untrained eyes based on both TC (Cat1: R=0.992; Cat2: R=0.951, both p<0.0001) and C₅₀ contrast sensitivity measures (Cat1: R=0.95, p<0.001; Cat2: R=0.983, p<0.0001). Further, the observed magnitude of CSF improvements of the untrained eye were also highly correlated with changes of the neuronal CSFs of neurons responding preferentially to stimuli presented via the untrained eye of each trained cat relative to three untrained cats based on TC contrast sensitivity (Cat1: R=0.827, p<0.05; Cat2: R=0.947, p<0.01).

Mechanisms of contrast sensitivity enhancement of V1 neurons

Four potential mechanisms may underlie the training-induced contrast sensitivity improvements [32-36]: (1) Decreased spontaneous activities (M), (2) Increased responsiveness (R_max), (3) Increased slopes of contrast-response functions (N), and (4) Increased contrast-gain (C₅₀). We systematically compared the best fitting parameters of the Naka-Rushton equation in different neuronal populations (Fig. 4B, Supplementary Fig. S2).

No significant difference was found between the trained and untrained cats in terms of spontaneous activities (F(1,254)= 1.946, p>0.1), maximum responses (F(1,254)=0.05, p>0.5), and slopes of contrast response functions (F(1,254)=3.319, p=0.07), nor was there significant difference between neurons responding preferentially to stimuli presented via the trained and untrained eyes of the trained cats in spontaneous activities (Cat1: F(1,91)=0.093, p>0.50; Cat2: F(1,47)=0.141, p>0.50), maximum responses (Cat1: F(1,91)=3.196, p=0.08; Cat2: F(1,47)=1.333, p > 0.25), and the slopes of contrast response functions (Cat1: F(1,91)=2.691, p > 0.1; Cat2: F(1,47)=3.232, p=0.08) (Fig. 7A-C). In contrast, there was significant difference between the trained and untrained cats in terms of C₅₀ (F(1,254)=37.487, p<0.0001) (Fig. 7D), and between neurons responding preferentially to stimuli presented via the trained and untrained eyes of the trained cats around their respective training spatial frequency (Fig. 7E-F), 0.2-0.6 c/deg for Cat1 (F(1,59)=8.855, p<0.01) and 0.4-0.8 c/deg for Cat2 (F(1,29)=9.774, p<0.01).

Fig. 7.

Parameters of the best-fitting Naka-Rushton equation to the neuronal contrast response functions. (A) Spontaneous activities. (B) Maximum responses. (C) Slopes of the contrast response functions (SL). (D) Contrast-gain (C₅₀) of cells in each trained and control cat. (E) Contrast-gain of cells with optimal SF 0.2-0.6 c/deg in trained cat1. (F) Contrast-gain of cells with optimal SF 0.4-0.8 c/deg in trained cat2. TE and NE denote cells responding preferentially to the stimuli presented via the trained eye and the naïve eye. All values are expressed as mean ± SEM.

We also compared post-training maximal response, TC contrast sensitivity and C50 contrast sensitivity between neurons with preferred orientation near (within ±15° from the trained orientations) and away from the trained orientations in two trained cats. We found no significant difference (Maximal response: F(1,140)=0.211, p=0.647; TC contrast sensitivity: F(1,140)=0.134, p=0.715; C50 contrast sensitivity: F(1,140)=0.0001, p=0.995). We also compared post-training maximal response, TC contrast sensitivity and C50 contrast sensitivity of neurons with preferred orientation near the trained orientations between the trained and untrained cats. The mean maximal response of neurons with preferred orientation near the trained orientations in trained cats was not significantly different from that in untrained cats (F(1,39)=0.247, p=0.622). However, the mean TC contrast sensitivity (F(1,39)=4.589, p=0.038) and C50 contrast sensitivity (F(1,39)=7.189, p=0.011) showed significant difference between the trained cats and untrained cats.

We conclude that the improved contrast sensitivity in the trained eyes of the trained cats can be attributed to increased contrast-gain of neurons responding preferentially to stimuli presented via the trained eyes, corresponding to a leftward shift of the neuronal contrast response functions [32, 34, 35] and consistent with the stimulus enhancement mechanism in human behavioral studies [36].

Conclusions and Discussion

In this study, we examined the physiological bases of perceptual learning by combining psychophysical assessment of contrast sensitivity and extracellular single-unit recording on cats. We found that training significantly improved the contrast sensitivity of the cats to gratings with spatial frequencies near the trained spatial frequency. The learning effect also exhibited specificity to the trained eye, although there was partial transfer to the untrained eye. Consistent with the psychophysical observations, the mean contrast sensitivity of V1 neurons with preferred spatial frequency near the trained spatial frequency was significantly increased in the trained cats relative to the untrained cats, specifically, the mean contrast sensitivity of neurons responding preferentially to stimuli presented via the trained eye of the trained cats was significantly higher than the untrained eye of the trained cats. Moreover, the perceptual and neuronal CSFs in the trained and untrained cats, and the trained and untrained eyes of the trained cats, showed a remarkable degree of correlation prior to and after training. The magnitude of neuronal contrast sensitivity improvements is also highly correlated with that of performance improvements at the whole animal level. These results suggest that training in grating orientation identification lead to improvements of contrast sensitivity of V1 neurons and therefore improved perceptual contrast sensitivity. We further determined that increased contrast-gain underlay improved neuronal contrast sensitivities.

The results of our study are different from other visual physiological studies on visual perceptual learning. There are three critical differences between our study and those in the literature:

1. Different species: whereas neurophysiological studies of visual perceptual learning in the literature all used primate subjects, the current study used non-primates (cats). This difference may be very important because significant training-induced plasticity has been widely reported in auditory and somato-sensory cortices of non-primates [7, 8], and a short-term (a few hours) intracortical microstimulation could lead to drastic changes of orientation preference maps in the visual cortex of adult cats [37].

2. Different tasks: As shown by some fMRI studies, training-induced neural plasticity may depend considerably on visual tasks [9-11]. Previous electrophysiological studies exploring training-induced visual cortical plasticity generally used orientation threshold as the dependent measure. Yet, the current study used contrast thresholds as the dependent measure and yielded a result quite different from others, suggesting different neural networks might be involved in orientation discrimination and contrast detection.

3. Different animal states: We recorded the response of V1 neurons in anesthetized and paralysed cats, whereas previous studies made recordings in awake-behaving monkeys. Compared to studies on anesthetized cats, recordings from early visual cortical areas of awake monkeys may include substantial top-down influences from higher visual cortical areas[16, 38, 39].

We are conducting new studies to further investigate all these factors.

That contrast sensitivity of visual perception in the trained cats improved significantly near the trained spatial frequency is consistent with previous observations in human studies [1,4, 40]. Similar results were also obtained in our physiology recordings, suggesting that perceptual learning in contrast detection is likely mediated through spatial frequency channels in the primary visual cortex in cats. Mixed results on eye specificity of visual perceptual learning have been reported in the literature [2, 4, 5, 23, 41]. In this study, we found that perceptual learning of contrast detection transferred partially to the untrained eye, with a certain degree of eye specificity. The results suggest that the training-induced plasticity may occur both before and after binocular combination. Subsequent experiments are needed to examine the learning effects that may have likely occurred in the dorsal lateral geniculate nucleus (dLGN). For example, one can examine the differences in training-induced plasticity between X- and Y-channels by presenting low spatial frequency stimuli to the animal and recording single neuron activity from area 18 (V2), which receives substantial direct Y-channel (but not X-channel) input from the dLGN. Another way is to examine whether training can modify the suppressive surround receptive fields of V1 and V2 neurons with preferred spatial frequency near the trained spatial frequency.

Our neurophysiological data showed that perceptual learning in contrast detection led to enhanced contrast sensitivities of V1 neurons with preferred spatial frequency near the trained spatial frequency. A systematic analysis of the parameters of the neuronal contrast response functions indicated that this training-induced plasticity was caused by increased contrast-gain of the neurons associated with training, not by lowered spontaneous activity, nor increased responsiveness of V1 neurons, nor increased slopes of neuronal contrast response functions. The increased contrast-gain resulted in a parallel leftward shift of the neuronal contrast response functions, consistent with decreased post-synaptic polarization [42, 43], but not changes in presynaptic processes [44, 45], which would have resulted in changes of the maximum response level and slope of contrast response functions. Future experiments with in vivo patch-clamp recordings are necessary to further elucidate the underlying cellular mechanisms of perceptual learning.

We conclude that training in grating orientation identification increased contrast-gain of neurons with preferred spatial frequency near the trained spatial frequency and responding preferentially to stimuli presented via the trained eye, and therefore improved contrast sensitivity of cats with certain degree of specificity to the trained spatial frequency and trained eye.

Experimental Procedures

Subjects

Five adult male cats (age: 1-3 years old; body weight: 2.2-3 kg) with no apparent optical or retinal problems served as subjects. Cat1 and Cat2 received training; the other three cats were control subjects. Animal treatments were strictly in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Psychophysical procedures

The training apparatus was similar to that used in the literature [24, 46]. At first, Cat1 and Cat2 received monocular conditioning training in a two-alternative forced-choice grating detection task with fixed, high contrast (80%) grating stimuli at a single spatial frequency, which were oriented either + or -45 degrees. The spatial frequency was set at 0.2 c/deg for Cat1 and 0.4 c/deg for Cat2 (Figs. 1A&B). The mean luminance of grating stimuli was kept at 19 cd/m². The untrained eye was covered with a special mask that blocked light.

The experimenter triggered the first trial in the beginning of each training block when everything was ready. Each trial started with a bright fixation dot (0.1° visual angle) appeared in the center of CRT for 1s. This was followed by a 4-second stimulus presentation with a 1-second response denied period (RDP) during which pushing the nose keys triggered no food reward. Because large-size sine-wave gratings were used in the study, eye fixation was not important and not monitored. A four-second inter-stimulus interval (ISI) was provided between trials.

Cat1 and Cat2 concluded their conditioning training after > 90% mean correct performance was attained in 6 consecutive days. This was then followed by measurement of pre-training contrast sensitivity functions (CSFs) in the trained and untrained eyes, monocular training at a single spatial frequency for 40 days, measurement of post-training CSFs, and tests of their performance using high contrast sine wave gratings. The same grating orientation identification task was used, except contrast thresholds were measured. Contrast thresholds at 0.1, 0.2, 0.4, 0.6, 0.8, 1.2 and 1.6 c/deg (560 trials of test for each spatial frequency; all intermixed) were measured to construct seven repeated measures of CSFs, one from each 80 trials per spatial frequency. Gratings at 0.4 c/deg and 0.6 c/deg were used to train Cat1 and Cat2, respectively.

A 2 down/1 up staircase procedure was used to measure contrast thresholds (Supplementary Fig. S1). Contrast sensitivity at each spatial frequency was defined as the log of the reciprocal of mean threshold contrast value. The contrast sensitivity functions were fit with the Gaussian equation:

$$\mathit{\text{CSF}}(f)={\mathit{\text{CSF}}}_0+A\text{exp}[-\frac{{(f-f_0)}^2}{{\mathit{\text{width}}}^2}],$$ (1)

where f is the spatial frequency of the grating, f₀ is the peak spatial frequency, A is the maximum sensitivity. The CSFs prior to and after learning were compared using ANOVA based on repeated measurements.

In each daily training session, the subject was administered 1000 to 1500 trials in 10-15 100-trial blocks. Subjects took 5-10 minute breaks between blocks.

Electrophysiological recording

Following the psychophysical experiment, all five cat subjects were prepared for extracellular single-unit recording using procedures described in previous publications [22](Supplementary Fig. S2). The response of a cell to a drifting sinusoidal grating was defined as the mean response value (after subtracting the baseline) corresponding to the time of stimulus modulation, which was used to draw the spatial frequency, orientation and contrast tuning curve (Figs. 4A&B).

The contrast response function of each neuron was fit with the Naka-Rushton equation [47]:

$$\text{R}(\text{C})={\text{R}}_{\text{max}}\frac{{\mathrm{C}}^{\text{N}}}{({\mathrm{C}}^{\text{N}}+{\text{C}}_{50}^{\text{N}})}+\text{M},$$ (2)

where R_max is the maximal response, M is the spontaneous activity, C₅₀ is contrast that evokes half of the maximal response, and N represents the slope of the contrast response function. Cells with less than 95% goodness of fit were not included in our data analysis. 7.2% of cells were excluded.