Color-detection thresholds in rhesus macaque monkeys and humans

Abstract

Macaque monkeys are a model of human color vision. To facilitate linking physiology in monkeys with psychophysics in humans, we directly compared color-detection thresholds in humans and rhesus monkeys. Colors were defined by an equiluminant plane of cone-opponent color space. All subjects were tested on an identical apparatus with a four-alternative forced-choice task. Targets were 2° square, centered 2° from fixation, embedded in luminance noise. Across all subjects, the change in detection thresholds from initial testing to plateau performance (“learning”) was similar for +L − M (red) colors and +M − L (bluish-green) colors. But the extent of learning was higher for +S (lavender) than for −S (yellow-lime); moreover, at plateau performance, the cone contrast at the detection threshold was higher for +S than for −S. These asymmetries may reflect differences in retinal circuitry for S-ON and S-OFF. At plateau performance, the two species also had similar detection thresholds for all colors, although monkeys had shorter reaction times than humans and slightly lower thresholds for colors that modulated L/M cones. We discuss whether these observations, together with previous work showing that monkeys have lower spatial acuity than humans, could be accounted for by selective pressures driving higher chromatic sensitivity at the cost of spatial acuity amongst monkeys, specifically for the more recently evolved L − M mechanism.

Keywords: color vision; human; macaque monkey.

Figure 1

Chromatic-detection task paradigm. (A) Left: Stimuli colors defined by the cardinal axes of cone-opponent color space (Derrington et al., ; MacLeod & Boynton, 1979). Numbers show the L − M and S − (L+M) Weber cone contrast elicited by the most saturated colors along the cardinal axes; for example, the cone contrast associated with Color 1 was (L₁ − L_g)/L_g − (M₁ − M_g)/M_g, where L₁ was the L-cone activation elicited by the stimulus, L_g was the L-cone activation elicited by the neutral gray adapting field, M₁ was the M-cone activation by the stimulus, and M_g was the M-cone activation by the adapting field. Right: Four-alternative forced-choice test paradigm. Subjects were rewarded for making an eye movement to a colored target in one of four locations (dashed lines) after fixation. (B) Left: Example eye-position trace for a human (gray) and a monkey (black) subject performing the task. The central cluster corresponds to fixation, the four clusters around the central cluster represent eye movements to the edge of the target, and the spurious traces show eye movements between trials. Right: Proportion of saccades made by monkeys (dark gray) and by humans (light gray) during an example session, separated by the location of the target (correct choice shown by black arrow). Error bars show ±1 standard deviation across subjects. (C) Mean lapse rate for monkey (left) and human (right) subjects shown over testing sessions. Shaded region shows standard deviation over colors.

Figure 2

Performance on color-detection task improves with training, monkey data (A) and human data (B). Left: Psychometric detection curves from a single session for color 1 (blackened square in the icon inset identifies the color shown in Figure 1A), for an early session (curve fit R²_human = 0.96, R²_monkey = 0.95) and a session following plateau performance (R²_human = 0.98, R²_monkey = 0.97). Vertical error bars show ±1 standard deviation and horizontal error bars show the 95% confidence interval of the threshold value. Contrast of the target is given in device-dependent units (D.D.U.) and in (L − M)-cone contrast units calculated as described in Figure 1A. Right: Detection threshold over time for color 1; shaded region shows 95% confidence interval. Dashed black line indicates reciprocal function fit (R²_human = 0.49, R²_monkey = 0.74). Arrows show which session was used to generate the psychometric curves (gray, early session; black, late session).

Figure 3

Change in chromatic-detection thresholds with task training. (A) The distance from the origin shows the log difference between the detection threshold obtained on the first two testing sessions and that of the last two testing sessions, for the colors plotted at different angles, for all naïve subjects. The change in detection thresholds with learning was similar for L − M and M − L colors (paired t test, p = 0.3) but differed for +S and −S colors (paired t test, p = 0.01). (B) The distance from the origin shows the number of testing sessions required to reach plateau performance for colors plotted at the different angles. Subjects required more sessions to reach plateau along the +S direction compared to the −S direction (paired t test, p = 0.003), but equal numbers of sessions along the +L − M and −L + M directions (paired t test, p = 0.54).

Figure 4

Reaction times for detecting chromatic targets. Average monkey (black) and human (gray) response-time histograms averaged over all colors for a given saturation level. Reaction times were calculated for correct trials except in the gray condition, when all trials were included in the analysis. Dashed lines show mean values of histograms. Stimulus saturation decreases from left to right.

Figure 5

Monkeys detect colored targets of all saturations faster than humans, and faster still for colors along the LM axis. The ratio of the reaction time to S stimuli obtained in monkeys to the reaction time obtained in humans (S_{reaction time monkeys}/S_{reaction time humans}) is shown as a function of the ratio of the reaction times to L/M stimuli (LM_{reaction time monkeys}/LM_{reaction time humans}). Darker colors correspond to data originating from trials with higher saturation targets. Error bars show standard error over seven subjects.

Figure 6

Psychometric curves for the detection of eight colors evenly sampling the equiluminant color plane, for monkeys and humans, using stimuli embedded in 0.2° luminance noise. Average monkey (black) and average human (gray) performance is shown for all colors (icons refer to Figure 1A). Contrast of the target is given in device-dependent units and in cone-contrast units (curve fits: color 1, R²_human = 0.99, R²_monkey = 1.0; color 2, R²_human = 0.99, R²_monkey = 0.99; color 3, R²_human = 0.99, R²_monkey = 0.99; color 4, R²_human = 1.0, R²_monkey = 1.0; color 5, R²_human = 1.0, R²_monkey = 0.99; color 6, R²_human = 0.98, R²_monkey = 0.98; color 7, R²_human = 0.99, R²_monkey = 0.98; color 8, R²_human = 1.0, R²_monkey = 1.0). Curves obtained for monkeys were considered significantly different from those obtained in humans if the 95% confidence intervals of the threshold values did not cross (asterisks).

Figure 7

Detection thresholds for chromatic targets embedded in 0.2° luminance noise. (A) Average threshold value in device-dependent units and cone-contrast units (see Figure 1A for formula), shown as the distance from the origin for each color direction tested. Error bars on curves for individual subjects show 95% confidence intervals. Asterisks mark colors that showed a significant difference between average monkey performance and average human performance (see Figures 6 and 7C). Monkeys showed lower detection thresholds in general (ANOVA, α = 0.05, p = 7 × 10⁻⁴). Inset shows an icon of the stimulus (see Figure 1A, right panel). (B) Average threshold for monkeys (dark gray) and humans (light gray) for the cardinal colors in cone-contrast units during initial testing, prior to exhausting perceptual learning, and (C) at plateau performance after exhausting perceptual learning. Error bars show 95% confidence intervals, asterisks indicate significant differences (see Figure 6). Values were considered significantly different if their 95% confidence intervals did not cross.

Figure 8

Psychometric curves for the detection of eight colors evenly sampling the equiluminant color plane, for monkeys and humans, using stimuli embedded in 2° luminance noise. Conventions as for Figure 6 (curve fits: color 1, R²_human = 0.97, R²_monkey = 1.0; color 2, R²_human = 0.99, R²_monkey = 0.98; color 3, R²_human = 1.0, R²_monkey = 0.97; color 4, R²_human = 1.0, R²_monkey = 1.0; color 5, R²_human = 1.0, R²_monkey = 1.0; color 6, R²_human = 0.97, R²_monkey = 0.95; color 7, R²_human = 0.99, R²_monkey = 0.99; color 8, R²_human = 1.0, R²_monkey = 0.99).

Figure 9

Detection thresholds for chromatic targets embedded in 2° luminance noise. Conventions as for Figure 7A. Monkeys showed lower detection thresholds (n-way ANOVA, α = 0.05, p = 6 × 10⁻³). Inset shows an icon of the stimulus.

Figure 10

Detection thresholds increase when targets are embedded in 2° versus 0.2° luminance noise for humans (A) and for monkeys (B). Error bars show standard deviation of the calculated threshold value. Color of marker corresponds to the hue direction. Multiple markers of the same color correspond to performance of different subjects.

Figure 11

Impact of luminance noise on chromatic-target detection threshold. Average human threshold values are shown as the distance from the origin for each color direction obtained in Experiment 1 (thick solid line; 0.2° luminance noise), Experiment 2 (thin solid line; 2° luminance noise), and the pilot experiment (dashed line; no luminance noise) for humans (A) and monkeys (B). Normalized difference in threshold value between the pilot study and Experiment 1 (C) and between the pilot study and Experiment 2 (D) are shown as distance from the origin for each color direction for humans (gray) and monkeys (black) as given by the following equation: