Fig 1.
Foraging task in environmental contexts.
(A) Example trial from monkey PA with a safe environment (#121 in Fig 1C). Top: sequence of events starting with blank screen during the ITI. Bottom: eye position (magenta: horizontal; green: vertical) and spike activity of one amygdala neuron (spike timing, blue). In this trial, the bad object appeared twice, followed by the good object (top). The monkey made saccades from the fixation point to both objects but quickly returned to the center for the bad object and kept fixating the good object to get a reward. The neuron was nearly silent (only two spikes). (B) Trial from monkey PA with a dangerous environment (#172 in Fig 1C). After the bad object, the robber object appeared and remained until the good object appeared, to which the monkey made a saccade quickly enough to trigger reward delivery. On other trials when the saccade was delayed, the robber object jumped to the good object and no reward was available. See S1 Fig and Materials and methods for detailed procedures. See also S1 and S2 Movies. The same neuron continued to be active until the reward delivery (unlike in the safe environment in A). Performing these trials correctly required the monkey to retrieve the memories of the objects contained in the scene. We refer to each scene as an “environment.” (C) Multiple sets of example scenes (n = 56) and objects (n = 140) for monkey PA. Each scene contained one good object (associated with a big or small reward) and one bad object (associated with no reward). Some scenes contained a third object that acted as a robber object (D/R) or a distractor object (some in S/R and S/P). The robber object tried to attack the good object and forestall the reward (as in S1 Fig). The distractor remained while another object (good or bad) appeared, but never attacked the other object. Each good object was consistently associated with either a big or a small reward. We thus classified these scenes into three groups. D/R: dangerous (with robber) and rich (big reward). S/R: safe (no robber) and rich (big reward). S/P: safe (no robber) and poor (small reward). The trial continued with objects appearing in random sequence and position until the monkey ended the trial by holding fixation on either the good object (followed by reward) or the bad object (followed by no reward). The scene remained until the end of the trial. These example scene images were derived from OpenAerialMap (https://openaerialmap.org). D/R, dangerous and rich; ITI, intertrial interval; S/P, safe and poor; S/R, safe and rich.
Fig 2.
SacRTs in safe and dangerous contexts.
(A) Safe scene (top, #121 in Fig 1C) and dangerous scene (bottom, #172 in Fig 1C), which appeared occasionally and randomly among many others (Fig 1C). The first object can be a good or bad object, or a robber object (in the dangerous scene). (B) Distribution of the SacRT in monkey PA to the first object (good or bad) in the dangerous (D/R) scenes (black) and the safe (S/R) scenes (red). The data are based on all scenes in each group (n = 24 for D/R, n = 20 for S/R, Fig 1C). SacRT distribution is shown using reciprobit plot, in which a straight line indicates a normal (Gaussian) distribution of the speed of the saccade preparation process [16]. (C) SacRT distribution in the dangerous scenes, shown separately for good objects and bad objects. Example scene images were derived from OpenAerialMap (https://openaerialmap.org). D/R, dangerous and rich; S/R, safe and rich; SacRT, saccade reaction time.
Fig 3.
Distribution of SacRTs in different contexts and for different targets.
(A) Distributions of the SacRT for three groups of scenes are superimposed: D/R, S/R, and S/P (see Fig 1C). They are shown separately for the good and bad objects. Data are shown for monkeys PA (left), PI (center), and SO (right) after learning (>350, >200, and >200 trials for each environment, respectively). (B) The same SacRT data are shown separately for the three groups of scenes (D/R, S/R, S/P), but data for good (G) and bad (B) objects are superimposed. The numbers of saccades examined are shown in each graph. The effects of scenes (A) and objects (B) on SacRT were independent and orthogonal in all subjects (two-way ANOVA, monkey PA [scene, F(2,18693) = 68.004, P < 0.001; object, F(1,18693) = 381.645, P < 0.001; scene*object, F(2,18693) = 5.966, P = 0.003], monkey PI [scene, F(2,12428) = 18.313, P < 0.001; object, F(1,12428) = 1,216.167, P < 0.001; scene*object, F(2,12428) = 21.892, P < 0.001], monkey SO [scene, F(2,27266) = 84.198, P < 0.001; object, F(1,27266) = 584.585, P < 0.001; scene*object, F(2, 27266) = 37.405, P < 0.001]). Data used to generate these plots can be found at https://osf.io/2yq8p/?view_only=97c4b290514348bb91cdbb9ec1c85e09. B, bad; D/R, dangerous and rich; G, good; S/P, safe and poor; S/R, safe and rich; SacRT, saccade reaction time.
Fig 4.
Responses of one example neuron in the amygdala to scene environments.
Neuron (#73) in monkey PA that was selectively active in the dangerous context (partially shown in Fig 1). (A–B) The neuron’s activities in the three groups of scenes: dangerous (D/R), rich (S/R), poor (S/P). They are shown separately as spike rasters (A) and are superimposed as SDF. (B). For SDF, each spike was replaced by a Gaussian curve (σ = 10 ms) in this and the following figures. Triangle indicates the onset of the scene response (white) and the onset of the context-dependent differentiation (black). (C) Time course of the neuron’s activity bias in two dimensions of context: (1) dangerous–safe (D versus S): difference in activity between D/R and S/R, (2) rich–poor (R versus P): difference in activity between S/R and S/P. In each dimension, the difference of activity (ΔFR score) was calculated in a sliding 300-ms window (10 ms steps) if it was statistically significant (tested by one-way ANOVA and Tukey–Kramer post hoc test); otherwise the score was put as 0. Then, the scores were divided by the maximum of the difference during free-viewing and fixation periods. ΔFR > 0: D > S (top) and R > P (bottom). (D) The neuron’s response (z-score) to individual scenes in the three groups examined (see Fig 1C). In this and the following figures, the neuronal data are focused on the activity before the first object appeared (see Fig 1A and 1B). This period was divided into the two parts: free-viewing period (from scene onset to fixation start) (left) and fixation period (from fixation start to object onset) (right). Data used to generate these plots can be found at https://osf.io/2yq8p/?view_only=97c4b290514348bb91cdbb9ec1c85e09. D/R, dangerous and rich; FR, firing rate; S/P, safe and poor; S/R, safe and rich; SDF, spike density function.
Fig 5.
Responses of amygdala neurons in three monkeys to scene environments.
(A) The numbers of neurons (excited type, inhibited type, others) in monkeys PA, PI, and SO. (B) Average activities in the three groups of environments: dangerous (D/R), rich (S/R), and poor (S/P). They are shown separately for the excited-type neurons (top) and the inhibited-type neurons (bottom) for each monkey. The averaging was based on the normalized z-scores of individual neurons (see Materials and methods). Shaded gray area indicates the free-viewing (left) and fixation (right) periods. (B, bottom) Time course of the activity biases of individual neurons in two dimensions of environmental context: dangerous–safe dimension (D versus S) and rich–poor dimension (R versus P) (same format as Fig 4C). Individual neuron data are sorted by their mean ΔFR scores in the fixation period, high to low scores. In each dimension, the averaged ΔFR scores are shown by a cyan line with black dots. Data used to generate these plots can be found at https://osf.io/2yq8p/?view_only=97c4b290514348bb91cdbb9ec1c85e09. D/R, dangerous and rich; FR, firing rate; S/P, safe and poor; S/R, safe and rich.
Fig 6.
Amygdala neuronal activity and SacRT modulated by two dimensions of scene context.
(A–B) Neuronal activity during the fixation period (A) and SacRT (B) in three monkeys (PA, PI, SO), shown separately for three groups of scenes: S/P, D/R, and S/R. The data (ordinate) are plotted against the predicted reward value (abscissa), which was defined as: reward amount × success rate (see Materials and methods). Neuronal activity was based on the normalized z-scores; SE is shown by a vertical bar. Asterisk (*) indicates statistically significant contrasts at P < 0.05 (scene context: one-way ANOVA, post hoc: Tukey–Kramer [see S1 Table]; predicted versus actual in D/R: one-sample t test). (C) Relation between the neuronal activity (abscissa) and SacRT (ordinate) for individual scenes. Statistics (Pearson’s correlation) are shown in each graph. The color of each data point indicates the scene context, as used in A and B. Data are based on excited-type neurons (Fig 5). Data used to generate these plots can be found at https://osf.io/2yq8p/?view_only=97c4b290514348bb91cdbb9ec1c85e09. D/R, dangerous and rich; S/P, safe and poor; S/R, safe and rich; SacRT, saccade reaction time.
Fig 7.
Pupil size and heart rate modulated by two dimensions of environmental context.
Pupil size (A) and heart rate (B) during the fixation period in all three monkeys, shown separately for three groups of environments: S/P, D/R, and S/R. The physiological measures (ordinate) are plotted against the predicted reward value (abscissa), defined as reward volume × success rate (see S2D Fig). Same format as Fig 6A and 6B. Asterisk (*) indicates statistically significant contrasts at P < 0.05 (environment context: one-way ANOVA, post hoc: Tukey–Kramer [see S2 Table]; predicted versus actual in D/R: one-sample t test, pupil size [PA]: P < 0.001, t = 14.962, pupil size [PI]: P < 0.001, t = 10.639, pupil size [SO]: P < 0.001, t = 39.775; heart rate [PA]: P = 0.001, t = 2.571; heart rate [PI]: P = 0.986, t = 0.017; heart rate [SO]: P = 0.018, t = 2.377). Data used to generate these plots can be found at https://osf.io/2yq8p/?view_only=97c4b290514348bb91cdbb9ec1c85e09. D/R, dangerous and rich; n.s., not significant; S/P, safe and poor; S/R, safe and rich.
Fig 8.
Estimated positions of recorded neurons.
Recording sites are shown in five coronal MR images spanning 0–4 mm posterior to the AC. (A) Amygdala and surrounding brain areas. (B–D) The neurons’ positions are shown, based on different features. (B) Neurons excited and inhibited by visual scenes during the fixation period, and neurons showing no response to the scenes (Visual −). (C) Excited-type neurons showing significantly different activity between dangerous (D) and safe (S) contexts (D > S; S > D), and others (D = S). (D) Excited-type neurons showing significantly different activity between rich (R) and poor (P) contexts (R > P; P > R), and others (R = P). AC, anterior commissure; BL, basolateral complex of the amygdala; CE, central nucleus of the amygdala; D, dangerous; HP, hippocampus; L, lateral nucleus of the amygdala; LV, lateral ventricle; MR, magnetic resonance; OC, optic chiasm; OT, optic tract; P, poor; PU, putamen; R, rich; S, safe.
Fig 9.
How emotional–environmental contexts affect saccadic eye movement—hypothetical neural circuits.
cdlSNr, caudal-dorsal-lateral part of the substantia nigra pars reticulata; cdlSNc, caudal-dorsal-lateral part of the substantia nigra pars compacta; CDt, caudate nucleus; cvGPe, caudal-ventral part of the globus pallidus externus; SC, superior colliculus.