Monkeys
Two adult male rhesus macaques (Macaca mulatta) participated in the experiment. The number of monkeys (2) follows previous work using similar approaches21. Monkeys Si and Ch were between 8 and 11 years old and weighed approximately 12.7 and 10.7 kg, respectively. All of the experimental procedures were approved by the Princeton University Institutional Animal Care and Use Committee (protocol, 3055) and were in accordance with the policies and procedures of the National Institutes of Health.
Behavioural task
Each trial began with the monkeys fixating on a dot at the centre of the screen. During a fixation period lasting 500 ms–800 ms, the monkeys were required to keep their gaze within a circle with a radius of 3.25 degrees of visual angle around the fixation dot. After the fixation period, the stimulus and all four response locations were simultaneously displayed.
Stimuli were morphs consisting of both a colour and shape (Fig. 1a). The stimuli were rendered as three-dimensional models using POV-Ray and MATLAB (MathWorks) and displayed using Psychtoolbox on a Dell U2413 LCD monitor positioned 58 cm from the animal. Stimuli were morphed along circular continua in both colour and shape (that is, drawn from a four-dimensional ‘Clifford’ torus; Fig. 1b). Colours were drawn from a photometrically isoluminant circle in the CIELAB colour space, connecting the red and green prototype colours. Shapes were created by circularly interpolating the parameters defining the lobes of the ‘bunny’ prototype to the parameters defining the corresponding lobes of the ‘tee’ prototype. The mathematical representation of the morph levels adhered to the equation \({X}_{1}^{2}+{X}_{2}^{2}+{X}_{3}^{2}={P}^{2}\) where X represents the parameter value in a feature dimension (for example, L, a, b values in CIELAB colour space). We chose the radius (P) to ensure sufficient visual discriminability between morph levels. The deviation of each morph level from prototypes (0% and 100%) was quantified using percentage, corresponding to positions on the circular space from −π to 0 and 0 to π. Morph levels were generated at eight levels: 0%, 30%, 50%, 70%, 100%, 70%, 50%, 30%, corresponding to 0, π/6, π/2, 5π/6, π, −5π/6, −π/2 and −π/6, respectively. 50% morph levels for one feature were only generated for prototypes of the other feature (that is, 50% colours were only used with 0% or 100% shape stimuli and vice versa). Stimuli were presented at fixation and had a diameter of 2.5 degrees of visual angle.
The monkeys indicated the colour or shape category of the stimulus by saccading to one of the four response locations, positioned 6 degrees of visual angle from the fixation point at 45, 135, 225 and 315 degrees, relative to the vertical line. The reaction time was taken as the moment of leaving the fixation window relative to the time of stimulus onset. Trials with a reaction time below 150 ms were terminated, followed by a brief timeout (200 ms). Correct responses were rewarded with juice, while incorrect responses resulted in short timeouts lasting 1 s for monkey Ch and 5 s for monkey Si. After the trial finished, there was an intertrial interval lasting 2–2.5 s before the next trial began.
The animals performed three category-response tasks (Fig. 1c). The S1 task required the monkeys to categorize the stimulus by its shape. For a stimulus with a shape that was closer to the ‘bunny’ prototype, the animals had to make a saccade to the UL location to get rewarded. For a stimulus with a shape closer to the ‘tee’ prototype, the animals had to make a saccade to the LR location to be rewarded. For ease of notation, we refer to the combination of the UL and LR target locations as axis 1. The C1 task required the monkeys to categorize the stimulus by its colour. When a stimulus’ colour was closer to ‘red’, the animals made an eye movement to the LR location and when it was closer to ‘green’ the animals made an eye movement to the UL location. Finally, in the C2 task, the monkeys again categorized stimuli based on their colour but responded to the UR location for red stimuli and the LL location for green stimuli. Together, the UR and LL targets formed axis 2. This set of three tasks was designed to be related to one another: the C1 and C2 tasks both required categorizing the colour of the stimulus while the C1 and S1 tasks both required responding on axis 1.
The monkeys were not explicitly cued as to which task was in effect. However, they did perform the same task for a block of trials allowing the animal to infer the task based on the combination of stimulus, response and reward feedback. Tasks switched when the monkeys’ performance reached or exceeded 70% on the last 102 trials of task S1 and task C1 or the last 51 trials of task C2. For monkey Si, block switches occurred when their performance reached or exceeded the 70% threshold for all morphed and prototype stimuli in the relevant task dimension independently. For monkey Ch, block switches occurred when their average performance at each morph level in the relevant task dimension exceeded the 70% threshold (that is, average of all 30% morphs, average of all 70% morphs, and 0% and 100% prototypes in colour dimension were all equal or greater than 70% accuracy in the C1 and C2 tasks). Moreover, on a subset of recording days, to prevent monkey Ch from perseverating on one task for an extended period of time, the threshold was reduced to 65% over the last 75 trials for S1 and C1 tasks after the monkey had already done 200 or 300 trials on that block.
When the animal hit the performance threshold, the task switched. This was indicated by a flashing yellow screen, a few drops of reward and a delay of 50 s.
To ensure even sampling of tasks despite the limited number of trials each day, the axis of response always changed following a block switch. During axis 1 blocks, either S1 task or C1 task was pseudorandomly selected, interleaved with C2 task blocks. Pseudorandom selection within axis 1 blocks avoided three consecutive blocks of the same task, ensuring the animal performed at least one block of each task during each session. On average, animals performed of 560, 558 and 301 trials and 2.68, 2.77 and 5.4 blocks per day for the S1, C1 and C2 tasks, respectively. Task orders and trial conditions were randomized across trials within each session.
... continue reading