Activity of striatal neurons reflects social action and own reward

Significance

A basic requirement for engaging in beneficial social interactions is to recognize the actor whose movement results in reward. We studied the neuronal correlates of social action and reward in the monkey striatum. Behaviorally, the animals distinguished between their own and their conspecific's reward and knew which individual acted. In a subset of striatal neurons, neuronal activity occurred preferentially when either the own or the conspecific's action was followed by own reward. Some of the social action activity disappeared when the conspecific's role was simulated by a computer, confirming a social rather than observational relationship. These findings demonstrate a role of striatal neurons in identifying the social actor and own reward in a social setting.

Abstract

Social interactions provide agents with the opportunity to earn higher benefits than when acting alone and contribute to evolutionary stable strategies. A basic requirement for engaging in beneficial social interactions is to recognize the actor whose movement results in reward. Despite the recent interest in the neural basis of social interactions, the neurophysiological mechanisms identifying the actor in social reward situations are unknown. A brain structure well suited for exploring this issue is the striatum, which plays a role in movement, reward, and goal-directed behavior. In humans, the striatum is involved in social processes related to reward inequity, donations to charity, and observational learning. We studied the neurophysiology of social action for reward in rhesus monkeys performing a reward-giving task. The behavioral data showed that the animals distinguished between their own and the conspecific’s reward and knew which individual acted. Striatal neurons coded primarily own reward but rarely other's reward. Importantly, the activations occurred preferentially, and in approximately similar fractions, when either the own or the conspecific's action was followed by own reward. Other striatal neurons showed social action coding without reward. Some of the social action coding disappeared when the conspecific's role was simulated by a computer, confirming a social rather than observational relationship. These findings demonstrate a role of striatal neurons in identifying the social actor and own reward in a social setting. These processes may provide basic building blocks underlying the brain's function in social interactions.

Social interactions enhance individual fitness by giving agents access to otherwise unobtainable resources (1). To be successful in such interactions, individuals need to recognize the social action resulting in reward. This process allows individuals to assign credit to their partners (2), which is crucial for establishing and maintaining mutually beneficial interactions.

Despite the importance of recognition of social action for obtaining reward, little is known about the underlying neuronal mechanisms. Recent evidence suggests separate coding of social reward and action in the neuronal activity of distinct frontal cortical areas of monkeys. Reward neurons distinguish between own and other's reward (3, 4) and show differential activity during social competition (5) and movement (6). Action neurons respond to the observation of movement of social partners (7), differentiate between own and other's movements (8), and detect other’s error commission (9). However, none of these studies investigated the neuronal coding of social action together with reward.

A candidate brain structure for coding social action and reward is the striatum. Its neurons code reward expectation and reception (10, 11), are activated during planning and execution of movements (12), and process influences of reward on movement coding (13, 14). In human imaging, the striatum is activated during social interactions, such as donation giving (15), observational learning (16), and detection of reward inequity (17). A social role of the striatum is further supported by actor specific activations without reward coding in cingulate cortex (2), which projects strongly to striatum (18). Thus, the striatum is engaged in separate reward, action, and social mechanisms, but it is unknown how the same striatal neurons may process social action and reward.

To investigate neuronal mechanisms underlying social action and reward, we recorded neuronal activity from the striatum while two rhesus monkeys interacted via a horizontal computer touchscreen mounted between them (Fig. 1 A and B). We used imperative trials to avoid choice confounds and dissociated two social dimensions: who was required to act, and who would receive reward. Only one animal acted at a time, and the actor's role switched after every correct trial. Hence, the recorded animal performed as actor or conspecific on alternate trials. There were no behavioral requirements for the animal taking the role as conspecific. The animals experienced four different payoffs: reward to neither, only to the recorded animal, only to the conspecific, or to both (Fig. 1C). We used fixed numbers of juice drops on each experimental day, thus testing reward presence vs. reward absence.

Fig. 1.

Reward-giving task and behavioral results. (A) Experimental setup. Two monkeys sat opposite each other at a horizontal computer touch screen, each holding a resting key. On each trial, light gray and black backgrounds indicated actor and conspecific roles to the respective animals. (B) Task sequence: shape of conditioned cue predicted absence (square) or presence (circle) of reward for each animal (yellow for left, purple for right animal). Appearance of a subsequent blue go signal was followed by key release, stimulus touch and reward for actor, and 1 s later for conspecific. (C) The four reward conditions used: reward for neither, own reward only, conspecific's reward only, and reward for both. (D) Mean reaction times for the four reward conditions (from go signal to stimulus touch) between the two animals. Error bars show SEM. (E) Eye fixation density between onset of conditioned stimuli and go signal. (F) Eye fixation density on conspecific's face and spout after reward delivery to conspecific.

Results

Behavior.

To investigate the neuronal coding of social variables, it is critical to first establish that the animals reacted to the social components of the task. Four behavioral measures fulfilled this requirement. Reaction times of the recorded animal varied between conspecific's rewarded and unrewarded trials (Fig. 1D, conspecific: purple vs. blue, own reward: red vs. blue; P < 0.05, post hoc Tukey tests following one-way ANOVA). There were also minor effects of conspecific's reward on error rates (Fig. S1A). Eye fixation patterns revealed that the recorded animal closely observed the conspecific's reward-predicting stimuli, face and spout (Fig. 1 E and F), confirming previous observations during social interaction of nonhuman primates (3, 8, 19). Without being required, the passive, nonacting animal maintained its hand on the resting key (Fig. 1A) during the conspecific’s turn on a fraction of trials (monkey A: 20%; monkey B: 60%) and released it in less than 3% of these trials (monkey A: 3%; monkey B, 2%). These results suggest that the animals distinguished between the active and the passive individual in the task, thus demonstrating crucial components of social behavior in our laboratory situation.

Neuronal Database.

We identified 457 statistically significant task-related activations in 192 of 273 tested neurons in the anterior, precommissural caudate, and putamen of two monkeys (A and B; Fig. S2; one-way ANOVA with post hoc Holm test against baseline activity). The activations occurred in one or more of six task epochs: before cues, during cues, during arm movement, after arm movement, before reward, and after reward. Table 1 presents a breakdown of the major types of activations, which failed to vary significantly between monkeys [χ²(20) = 24.97, P = 0.2].

Table 1.

Numbers of neuronal activations coding reward and actor

Actor	Animal	Reward
		Own	Conspecific	Both	None	Subtotal
Own	A	37 (27)	4 (4)	5 (4)	28 (23)	74 (45)
	B	30 (22)	2 (2)	2 (2)	29 (25)	63 (44)
Conspecific	A	20 (16)	1 (1)	9 (8)	16 (13)	46 (30)
	B	9 (7)	1 (1)	0 (0)	7 (6)	17 (13)
None	A	15 (13)	6 (6)	8 (8)	99 (55)	128 (63)
	B	25 (19)	2 (2)	1 (1)	101 (71)	129 (80)
Subtotal	A	72 (47)	11 (10)	22 (19)	143 (79)	248 (93)
	B	64 (41)	5 (5)	3 (3)	137 (83)	209 (99)

Activations were classified into the stated categories only when coding was statistically significant and, if applicable, consistent between main factors and interactions. Some neurons showed activations in multiple periods. Thus, in columns and rows labeled subtotal, the number of activations is the sum across the row or column, but not the sum of neurons as each neuron was counted only once. Three classes of activation were collapsed in the actor none, reward none cell: (i) activations that showed incongruence in the significance between the main and interaction factors; (ii) activations with significant interactions but insignificant main factors; and (iii) neuronal activations without any significant effects. Number of neurons in parentheses.

Reward Coding.

The next step of our statistical analysis revealed that 177 of the 457 significant activations (38%; 125 neurons) distinguished between rewarded and unrewarded trials for the recorded monkey, as described before (13, 14), or for the conspecific monkey (three-way ANOVA). Specifically, 136 of the 177 significant activations (87 neurons) reflected either the presence (n = 66 activations; Fig. 2 A–D, red-green vs. blue-purple) or the absence of own reward (n = 70; Table 1). By contrast, these activations failed to vary with the conspecific's reward (purple vs. blue). As a confirmation, the activations differed also when both animals received reward compared with the conspecific alone (green vs. purple) or when neither received reward (green vs. blue). Inversely, 16 activations were modulated by the conspecific’s, but not own, reward (9% of 177 activations; 15 neurons; Table 1; Fig. S3A; SI Text). Finally, 25 activations were modulated by both own and conspecific's reward (14% of 177 activations; 22 neurons). Hence, most task-related striatal neurons coded own reward predominantly and separately from conspecific's reward.

Fig. 2.

Coding of reward and actor in striatal neurons. (A and B) Activations coding own reward irrespective of actor. (A) Single neuron (activation after feedback onset). (B) Population; n = 20 activations showing increased activity with own reward presence (red-green) compared with no own reward (purple-blue). Activity was normalized to maximum firing rate of individual neurons irrespective of trial type and is shown as impulse density. (C and D) Coding of own action and own reward: higher activations with own compared with conspecific's action in single neuron (C; activation after cue onset) and population (D; n = 28). (E and F) Coding of conspecific's action for own reward in single neuron (E; activations after feedback onset) and population (F; n = 15). Interrupted axes underneath B, D, and F indicate noncontinuous analysis periods and are labeled in F.

Our regression analysis confirmed the predominant coding of own reward compared with conspecific's reward or reward to both (Fig. S4A; SI Text). The numbers of activations identified by the regressions (123, 32, and 2 activations for own, conspecific's, and both reward, respectively) corresponded to the numbers from the main analysis (three-way ANOVA; 136, 16, and 25 activations; Table 1).

The separate coding of own and conspecific's reward amounted to reward discrimination between the animals (Fig. 2, red vs. purple). To formally assess the discrimination, we subjected activity in these neurons to a receiver operating characteristic (ROC) distinguishing between reward only for the recorded monkey and reward only for the conspecific. A total of 107 of the 152 (70%) activations reflecting either own or conspecific's reward discriminated between reward recipients (P < 0.05, permutation test). These results suggest that a population of striatal neurons distinguished between own and conspecific's reward.

Coding of Social Action and Social Reward.

For adequate social interactions, it is important to distinguish whether reward coincides with one's own or a conspecific's action (2, 20). Indeed, a subset of striatal activations made this distinction. Ninety-six of the 136 own reward coding activations (70%, 70 neurons) differentiated between the actors (Table 1). Of the 96 activations, 67 (49 neurons) were significantly higher when the recorded animal received reward while acting compared with the conspecific's action (own action; Fig. 2 C and D, red-green, solid vs. dashed lines). Inversely, 29 of the 96 activations (23 neurons) were higher when the conspecific rather than the recorded animal made the movement and the recorded animal received reward (conspecific's action; Fig. 2 E and F, dashed vs. solid lines). The remaining 40 of the 136 activations (30%, 32 neurons) were indifferent to whose turn it was (Fig. 2 A and B). Correspondingly, a fraction of the activations reflecting conspecific's reward also distinguished between the individual performing the action (8 of 16 activations in 8 of 16 neurons; Fig. S3A; SI Text). Table S1 provides a complete breakdown of activations reflecting social action and reward for the six task epochs.

Neurons may simultaneously code two variables in several ways. In the statistical form of interaction, one variable affects the coding of the other variable. This form represents the most interesting part of our results, as the individual who acts affects the coding of the reward for each animal. In the other form, neurons may conjointly code both variables independently from each other. Fifty-four of the 96 activations (46 neurons) reflected the interaction between own reward and social action (28 for reward presence, 26 for reward absence). Of these, 35 activations (30 neurons) reflected the interaction of own action and own reward, and 19 (16 neurons) reflected the interaction of conspecific's action and own reward (Table S2). By contrast, 42 of the 96 activations (39 neurons) reflected own reward conjointly with a specific animal's action (20 for reward presence, 22 for reward absence). Of these, 32 activations (29 neurons) reflected conjointly own action and own reward, and 10 (10 neurons) reflected conjointly conspecific's action and own reward (Table S2).

The regression analysis confirmed the substantial sensitivity to social action and own reward in striatal neurons (Fig. S4A; SI Text). The numbers of activations identified by the regressions (95 and 7 activations for own and conspecific's reward, respectively) corresponded well to the numbers from the main, three-way ANOVA (96 and 8 activations; Table 1).

Taken together, these data demonstrate that a population of striatal reward neurons distinguished between social actors and between reward receivers.

Social Action Coding Irrespective of Reward.

The observed social neuronal coding might be specific for the combination of action and reward. However, earlier studies on human and monkey cingulate cortex reported differential coding of action observation without having tested reward variations (2, 8). Thus, would striatal neurons also code social action irrespective of reward? Indeed, 80 activations of the 457 task-related activations (17%, 65 neurons) discriminated between actors but were not modulated by reward. Of these, 57 activations (48 neurons) were higher when the recorded animal acted rather than the conspecific (Fig. 3 A, B, and E), whereas 23 activations (19 neurons) showed the opposite (Fig. 3 C–E; Table S1). This result was confirmed by our regression analysis. Of 75 actor sensitive activations, 58 reflected own action and 17 reflected conspecific’s action (Fig. S4A). In both the ANOVA and regression analysis, social action coding was not restricted to the movement period but could occur during any task period.

Fig. 3.

Sensitivity to actor in striatal neurons not coding reward and neuronal ROC. (A and B) Coding of own action (solid lines) compared with conspecific's action (dashed lines) in single neuron (A) and population (B; n = 57). Overlapping solid lines suggest lack of reward coding. (C and D) Coding of conspecific's action (dashed lines) rather than own action (solid lines) in single neuron (C) and population (D; n = 23). Overlapping dashed lines suggest lack of reward coding. (E) Neuronal ROC values for own reward vs. actor. Reward ROC varies between 0 and 1 for no reward vs. own reward and actor ROC varies between 0 and 1 for own action vs. conspecific’s action. (F) Same as E but for conspecific's reward vs. actor. Gray bars indicate 95% bootstrap CI. Number of members on each group for G and H (permutation test, P < 0.05): green rhomboids (n = 73), own reward and not actor; yellow squares (n = 11), conspecific's reward and not actor; turquoise triangles (n = 23), own and conspecific's reward and not actor; purple triangles (n = 61), own reward and actor; dark blue crosses (n = 3), conspecific's reward and actor; blue stars (n = 118), not reward but actor; pink triangles (n = 45), all categories; red dots (n = 122), all insignificant.

Neuronal ROC.

We used ROC analysis as a third independent measure to assess the degree of neuronal coding for reward recipient and actor. Activations reflecting the presence or absence of own or conspecific's reward (Fig. 3 E and F, horizontal axis), as well as activations not modulated by reward, were sensitive to the actor, being higher for either own (Fig. 3 E and F, below horizontal line) or conspecific's action (above horizontal lines; number of activations on each category is summarized in Table S3). Thus, the ROC analysis confirms the substantial sensitivity to social action for reward as identified by our ANOVA and regression analysis.

Social vs. Observational Nature of Actor Coding.

If these activations depended on the social nature of the other agent, they should change in its absence. We investigated this possibility by moving the conspecific out of the recorded monkey’s sight while a computer presented identical visual stimuli on the touch table in the same sequence of events used in conspecific's trials (except for reaction time being fixed to 1.2 s). The recorded animal received the same juice amount as the computer, but the computer's juice dropped visibly into an inaccessible bucket. We tested a further 22 actor-related activations from 12 striatal neurons. Of these, nine activations lost the discrimination between own and not own action when the conspecific was out of sight (Fig. 4 A and B, Center; Fig. 4C, empty bars). The activations recovered when the conspecific was returned to the task (Fig. 4 A and B, Right). Of the nine activations, seven occurred with own action (Fig. 4A) and two with conspecific's action (Fig. 4B). The remaining 13 activations differentiated between own and not own action irrespective of the conspecific being present or absent (Fig. 4C, filled bars).

Fig. 4.

Decrease of neuronal distinction between own and conspecific's action during computer control test. (A) Higher activation with own action compared with conspecific's action (Left) decreased when conspecific was replaced by computer (Center). The difference recovered with reinstatement of conspecific (Right). Data are from a single striatal neuron. (B) Same as A, but higher activation with conspecific's action. (C) Kolmogorov-Smirnov statistics (D) for influence of computer opponent on actor specific neuronal responses. Empty bars, decreased difference own vs. conspecific's action with computer (9 activations); filled bars, maintained difference own vs. conspecific's action with computer (13 activations). (D and E) Simple action relationships fail to explain neuronal sensitivity for social action (D, average of 58 activations coding own social action; E, 36 activations coding conspecific's social action). Solid bars represent population activity during action of recorded monkey (normalized to maximum firing rate of individual neurons irrespective of trial type). Dashed bars represent activity in recorded monkey during conspecific's action. Response inhibition refers to absence of movements of recorded monkey. In No response inhibition trials, the recorded animal performed movements without being required. In D, neuronal activity was high in own trials but low in conspecific's trials irrespective of own inhibition. In E, neuronal activity was low in own trials but high in conspecific trials irrespective of own movement. Error bars show SEM.

Arm movement reaction time was shorter in this control task (monkey A: from 588 ± 5 to 563 ± 6 ms, mean ± SEM, P > 0.05, t test; monkey B: from 542 ± 6 to 496 ± 13 ms, P < 0.05) and correlated only insignificantly with the reductions in actor coding of the nine social sensitive neuronal activations (r = 0.58, P = 0.09; Pearson’s correlation). Thus, the observed loss of actor coding in the computer control task was unlikely to be explained by faster arm movement reactions.

Of 14 agent differential activations recorded together with eye movements, only one showed relationships to ocular positions (P < 0.05, ANCOVA). Of the remaining 13 activations, six were social sensitive in being reduced during the computer test, whereas seven were maintained during the computer test. We conclude that a fraction of neurons distinguishing between own and conspecific's action stopped doing so when the social agent was removed, suggesting that these neurons were sensitive to the social context.

Response Inhibition.

Neuronal activations classified as being sensitive to social action might instead reflect differences in action vs. no action of the recorded animal. The task did not require specific behavioral action or inhibition by the nonacting animal. In particular, the passive player did not need to contact the resting key, nor release it after the cue, and thus was not required to inhibit a prepotent behavioral response. Nevertheless both animals contacted the key in a proportion of trials (monkey A: 20%; monkey B: 60%) and occasionally released it after the cue (monkey A: 3% of all trials; monkey B, 2%), and thus may not have inhibited a response in these trials. If activations classified as sensitive to own social action might be simply explained by performance of own movement, then neuronal activity during trials without response inhibition should be similar to own action trials as the recorded animal moved in both situations. Similarly, if responses sensitive to conspecific's social action were explained by inhibition of own movement, they should be stronger in trials with response inhibition compared with trials without response inhibition. This analysis included only neurons tested in greater than five uninhibited trials in which the recorded animal released the key during conspecific's trials. Of the 58 activations classified as reflecting own social action (with or without reward sensitivity), 53 remained significantly lower in conspecific's than own trials even when the recorded animal acted (Fig. 4D, Response inhibition and No response inhibition vs. Own action; one-way ANOVA followed by post hoc Tukey test; both P < 0.05). Similarly, of the 36 activations reflecting conspecific's social action, 29 remained significantly higher in conspecific's than own trials even when the recorded animal moved (Fig. 4E). Thus, the reported activations sensitive to social action were unlikely to be explained by own movement or response inhibition.

Eye Movement.

Given the known involvement of striatal neurons in eye movements (21), neuronal activations sensitive to social action might reflect oculomotor processes. Particularly important were eye position, which differed between own and conspecific's action (Fig. 1E), and eye velocity, which might have reflected differential motivation between own and conspecific's action (general neuronal relationships to saccadic eye movements are described in SI Text and Table S4). Of 81 actor-related activations recorded alongside eye movements, 61 (76%) were unrelated to eye position. Furthermore, 71 of the 81 activations (88%) were unrelated to eye movement velocity. The remaining activations could equally reflect actor or eye movements. Thus, differences in eye position or velocity were unlikely to explain the neuronal coding of social action and reward.

Reward Timing.

The acting monkey received the reward 1 s before the conspecific. Thus, would the activations coding social action and own reward rather reflect temporal discounting of reward value? To address this issue, we included a hyperbolic discounting model in the simultaneous testing procedure using a wide range of discounting parameters typical for rhesus monkeys (22–24) (SI Text). In 3 of 95 neuronal activations, we rejected the social action and own reward model in favor of the hyperbolic discounting model. Thus, temporal discounting did not seem to explain the large majority of activations reflecting social action and own reward.

The acting animal received reward 1 s before the passive animal to facilitate distinction between reward recipients. Despite this advantage, the temporal order of reward delivery might explain the actor sensitivity of reward responses. This possibility was investigated in conspecific's trials in which the recorded animal received reward first when the conspecific received no reward, but last when both animals received reward. Only 1 of the 29 neuronal activations reflecting conspecific's action and own reward varied significantly between these two trial types (P < 0.05; ROC with 2,000 permutations). A similar comparison was impossible for actions by the recorded animal, as it always received the reward first. Thus, at least for conspecific's actions, temporal order did not seem to explain actor-sensitive reward processing.

Reward Cost.

Movement effort constitutes an economic cost that reduces reward value (25). Thus, would activations coding social action and own reward be better explained by own reward minus cost? To test this possibility, we included an action-cost model in the simultaneous testing procedure with a wide range of cost parameters (SI Text). In only 14 of 95 neuronal activations originally reflecting social action and own reward (Fig. S4, Inset) was action cost a reasonable model of the data. Thus, movement effort did not seem to explain most activations reflecting social action and own reward.

Discussion

These experiments revealed that a population of striatal neurons coded social action and reward. Other striatal neurons coded social action without coding reward. Almost all reward related striatal neurons processed own rather than conspecific's reward. A fraction of social action-coding neurons required the presence of a social agent, suggesting a biological relationship. This spectrum of neuronal activity demonstrates an active involvement of the striatum in social processes.

During the nonsocial computer control task, the activity of some social action-coding neurons disappeared with the removal of the conspecific. This result implies that the activity was not related to the animal's action per se. Instead, it may have been related to the social context in which the action was embedded. In the subclass of neurons sensitive to conspecific's action, the response reduction could have been due to reduced visual stimulation generated by the conspecific's absence. However, the activity recorded during the computer control task was unrelated to eye position and thus insensitive to the visual stimulation elicited by different eye positions.

Our analysis of potential alternative interpretations included behavioral processes known to engage striatal neurons, including simple motor action, behavioral inhibition, eye position, eye velocity, reward delay, reward order, and effort (economic cost) (13, 21, 22, 26). Remarkably, the combined evidence from these limited control analyses suggests that only a few of the observed striatal activations reflecting social processes could be explained by these behavioral alternatives.

The sensitivity to own vs. conspecific's action could alternatively reflect differential coding of peri-personal space (within reach) during own trials vs. extrapersonal space (out of reach) during conspecific's trials. Movement-related neurons in the frontal cortex are known to code the distance at which task-relevant stimuli are presented (27). In the computer control task, stimulus positions in peri- and extrapersonal space were unchanged; hence, activations reflecting personal space should remain unaltered. However, the social responses were reduced during the control task in a good fraction of neurons. These data suggest sensitivity to social action rather than personal space in these neurons. By contrast, some of the activations not affected by the control task might reflect personal space, requiring additional experimentation for better characterization.

Our data on social action coding in the striatum might challenge the known role of this structure in straightforward movement processing (13, 14). However, previous studies already demonstrated the conditional nature of striatal movement-related activity. For example, groups of putamen neurons are only activated with the first of several consecutive movements (28), many movement-related striatal neurons are only activated with rewarded as opposed to unrewarded movements (14, 15), and some striatal neurons code action values rather than the actions themselves (29). Some of these neurons, and some of the many unmodulated neurons in the anterior striatum in which we recorded here, might have shown social action relationships if the testing involved a conspecific. Thus, rather than challenging the movement role of the striatum, the observed social action coding provides an extension of the conditional nature of movement-related activity of this structure.

Our data might challenge current views that position processing of social information and action primarily in the cerebral cortex and amygdala (30, 31). However, striatal neurons receive afferents from many cortical and subcortical structures involved in social processes (18, 32), which could give rise to the observed activities. As striatal neurons link own action with own reward (33), converging inputs and local interactions in the striatum (34) are also well suited to combine information about other’s actions and own reward. Thus, the current data extend social processing to a further subcortical structure.

The current observation of social action-related activity may help to assign credit to a social agent when receiving reward from other individuals (2). The recognition of social agency is an important factor for fostering reciprocal exchanges in primates and maintaining adequate social relationships (1, 2, 20). “Returning a favor” to someone relies on knowing whose action produced the original favor. This type of social interaction would break down without that knowledge. The present data suggest a possible role of the striatum in this form of social behavior.

Materials and Methods

Two adult male Macaca mulatta monkeys (monkeys A and B), weighing 9 kg, served in the main study. All experimental procedures were approved by the University of Cambridge Ethics Committee and the Home Office under the Animals (Scientific Procedures) Act 1986.

Monkeys A and B performed an imperative reward giving task under computer control (Fig. 1A). The task included two simultaneous cues predicting reward (circle) or no reward (square) separately for each animal (Fig. 1B, Left), followed by a blue go signal eliciting the actor's arm movement for touching the stimuli (Fig. 1B, second from left), and delivery of reward to the actor and 1 s later to the conspecific. We tested presence vs. absence of reward and kept the amount of reward fixed on each day (one to three drops indicated by one to three circles, respectively). We sampled extracellular activity from 273 slowly firing neurons (n = 115 and n = 158 in monkeys A and B, respectively) with known electrophysiological properties (21) from the anterior striatum of one monkey at a time with conventional electrophysiological techniques during at least 64 trials performed by each individual. We discriminated the activity of single neurons against background noise online with window discriminator hardware and offline with spike sorting software (Plexon).

Analysis of Neuronal Data.

We defined the control period as the last 0.5 s of the intertrial interval and defined six trial epochs of interest, namely (i) onset of gray background for acting animal (0.5 s before cue onset), (ii) cue onset (between 0.25 s after cue onset and go signal onset 0.75 s later), (iii) go signal onset (the first 0.5 s after go onset), (iv) movement feedback (0.5 s starting at go signal offset), (v) before reward (from end of movement feedback period for 1.5 s, which coincided with disappearance of the cue for the acting animal), and (vi) reward delivery (from first juice pulse to 0.5 s later).

In the first step, we defined task-related activations as a significant increase in neuronal activity during a task epoch compared with baseline using a one-way ANOVA followed by post hoc Holm test (P < 0.05). Subsequently, we explored the relationship of neuronal activity to reward receiver and actor of each task-related activation using a three-way ANOVA (P < 0.05). The factors and their levels were as follows: reward for recorded monkey (presence or absence), reward for conspecific (presence or absence), actor (action by recorded monkey or conspecific monkey), and interaction between reward receiver and actor. Because striatal neurons may show either increasing or decreasing activity with increasing reward value (35), this analysis distinguished between activations induced by reward presence (high reward value) and activations induced by reward absence (low reward value). We did not test the interaction between own and conspecific's reward or a triple interaction because they did not capture the variables of interest, namely actor and reward.

We classified activations as “own reward” or “conspecific's reward” when the respective factors were significant in the three-way ANOVA. Activations reflected own action when the factor actor was significant and mean firing rate during own trials was higher than during conspecific trials. Likewise, an activation reflected conspecific's action when activity was higher during conspecific's than own trials. Activations reflected actor for own reward when the statistical interaction of actor × own reward was significant. These activations revealed the most interesting aspect of the results, namely coding of reward for the recorded animal only when either that animal or the conspecific performed the action that led to reward for the recorded animal. By contrast, activations reflected conjoint coding of action and own reward when both factors were significant but the interaction of actor × own reward was insignificant. We classified, analogously, the few actor coding activations for conspecific's reward by the significant interaction of actor × conspecific's reward, whereas conjoint coding showed an insignificant interaction.

Eye Movement Analysis.

We used an ANCOVA to analyze potential oculomotor relationships in actor-related activations in monkey A. The ANCOVA included the same variables as the main ANOVA and additional horizontal and vertical eye position and, separately, eye velocity. Activations were considered as actor-related when the factors referring to an actor survived addition of the eye movement parameters (P < 0.05).

ROC Analysis.

To assess differences in neuronal activations in binary comparisons involving reward, actor, or spatial movement positions, we used ROC analysis. A significant ROC indicated discrimination (permutation test with 2,000 iterations; P < 0.05). CIs were established from a bootstrap distribution of ROC values (2,000 iterations).

Multiple Linear Regression.

We used multiple regressions structured as a simultaneous testing procedure (STP) for refinement and independent confirmation of the ANOVA results. This procedure adequately deals with multiple significant regressors without requiring corrections for multiple tests of significance. In addition, it allowed us to test temporal discounting and action cost. The principal hypothesis is stated in the unrestricted model

where Y is neuronal activity, A is actor (recorded animal acts = 1, conspecific acts = 0), W is reward for recorded animal (reward = 1, no reward = 0), and Z is reward for conspecific animal (reward = 1, no reward = 0). The two models of most interest were

which tests neuronal coding of actor for own reward, and

which tests neuronal coding of actor for conspecific's reward. We also included a model that tests coding of reward only

All models implied by the unrestricted model (Eq. 1), as well as the temporal discounting and action cost models, are detailed in Eqs. S5–S16 (for full details of the STP, see SI Text).

Computer Control Test.

To test the social nature of the observed actor-dependent coding, we moved the conspecific animal 50 cm laterally out of the recorded monkey’s sight and away from the spout and delivered its reward visibly into an inaccessible bucket. In these trials, the behavioral control computer timed all stimuli as in the regular trials and mimicked task performance in alternation with the recorded monkey; the go stimulus stayed on for a fixed 1.2 s. We conducted these tests in three consecutive trial blocks, namely with the conspecific present, the computer replacing the conspecific, and the conspecific back again. In each block, we assessed the difference in neuronal activation between the own and the conspecific's action using the Kolmogorov–Smirnov test and the direction of the difference using the ROC. Actor-related social activations were considered as biological if the differences between actors were significant during the initial and the retesting block but not during the computer block.