Tuning in to real-time social interactions in macaques

Significance

Social motor entrainment refers to the alignment of behaviors in time between biological agents. This phenomenon is seen widely in people with typical neurodevelopment and rarely so in autism spectrum disorder. Although social motor entrainment plays a pivotal role in promoting closeness and connectedness, thereby contributing to group cohesion, its neural mechanisms remain elusive. This is mainly because suitable animal models have yet to be established in which direct neural recordings are available. Here, using task paradigms, we show that macaque monkeys exhibit entrainment that is best tuned to the real-time movement of social partners. Our paradigm provides a promising platform to study the neural underpinnings of evolutionarily conserved mechanisms that support social cohesion.

Abstract

Movement synchronization between individuals has been implicated in reinforcing their cohesion. How might such interindividual motor entrainment be controlled by the social brain? The answer remains elusive owing largely to the lack of suitable animal models in which direct neural recordings are available. Here, we show that macaque monkeys exhibit social motor entrainment without human prompting. We found that repetitive arm movements for horizontal bar sliding were phase coherent between two monkeys. The nature of motor entrainment was specific to animal pairs, consistent across days, dependent on visual inputs, and affected by social hierarchy. Notably, the entrainment was diminished when paired with prerecorded movies of a monkey making the same movements or bar motion alone. These findings demonstrate that motor entrainment is facilitated by real-time social exchanges, providing a behavioral platform to study the neural basis of potentially evolutionarily conserved mechanisms that support group cohesion.

Humans have a remarkable tendency to synchronize their movements with those of others. This interindividual motor synchronization is ubiquitous and robust, as it can occur without explicit instructions (1–3) or even under instructions not to do so (4). It has been argued that synchronized movements encourage group cohesion by exploiting neurobiological and psychological mechanisms that underlie the maintenance of social bonds (5). A number of studies have shown that movement synchrony with, or being mimicked by, other individuals promotes social bonding and prosocial behavior across various contexts in humans (6–12). Moreover, movement synchrony is disrupted in people with autism spectrum disorder (13, 14), which is characterized by deficient social communication and interaction. These findings raise the possibility that interindividual synchrony (hereafter referred to as “social motor entrainment”) has been selected during evolution to support the formation of much larger social networks than would be expected given the human brain size (5).

To understand the evolutionary origin of social motor entrainment, researchers have explored evidence for synchronous motor behaviors in nonhuman primates. They found that great apes—both chimpanzees and bonobos—demonstrate coordinated rhythmic entrainment with other conspecifics (15–19). Macaque monkeys also display a spontaneous tendency of synchronous arm motion between conspecifics (20). These findings suggest that the neural substrates for social motor entrainment are conserved between different primate species. The positive evidence in macaque monkeys, albeit only in a single report so far, is particularly significant for the probe of in-depth mechanisms at the neural level. In great apes, investigations of brain activity, either directly (e.g., electrophysiological recordings) or indirectly (e.g., functional neuroimaging), are not feasible due to ethical and practical reasons.

However, fundamental issues remain to be addressed in the studies of nonhuman primates described above. Specifically, it is unknown to what extent the movement of other biological agents is important for generating motor entrainment. There remains the possibility that the movement of nonbiological objects is equally effective. Furthermore, even if the motion of biological agents is proved to be critical, another question will arise regarding the extent to which real-time interactions are a necessary requisite. Do the movements of filmed (offline) agents induce motor entrainment in the same way as those of real-time agents? Addressing these issues can only lead us to determine whether the observed motor synchrony is a social phenomenon. Here, using a set of behavioral paradigms devised for macaque monkeys, we show that the entrainment is best tuned to the movements of real-time, social agents.

Results

Movement Entrainment between Two Monkeys.

Three monkeys (Macaca fuscata; G, M, and N) were trained individually (solo condition) to grasp a vertical bar with their right hand and move it from side to side along a straight track. The track length was 250 mm, and correct bar motion required whole arm movement. Each trial started when two light-emitting diodes (LEDs) at the track ends were illuminated simultaneously along with a high-pitched tone. After this trial-outset cue, one of the LEDs remained illuminated to indicate the first movement direction. When the bar entered a predetermined target zone, the lighted LED turned off, and the LED on the opposite side was illuminated. In this way, the monkeys continuously performed repetitive arm movements. The duration of one-way movement (shot) was required to be longer than 400 ms (i.e., <1.25 Hz when converted to a reciprocating motion; “speed requirement”). A reward was given to the monkeys for every 1 to 7 shots (reward cycle; typically 3 shots). The quantity of reward after each reward cycle was increased in proportion to the total number of shots within a trial (Materials and Methods). A trial was completed when 25 s had elapsed, which was signaled by a low-pitched tone (trial-end cue). Note that the direction of the forthcoming movement was externally instructed, albeit fully predictable, but the speed of the movement and, therefore, the timings of LED on-off responses were under spontaneous control of the monkeys.

After the training had progressed on an individual basis, two monkeys sat face to face at a distance of 1 m from each other and performed the same repetitive arm movements under the same speed requirement (paired condition; Fig. 1A; see SI Appendix, Fig. S1 for the distribution of shot durations). The first movement after the trial-outset cue was either in the same or opposite direction between the two monkeys. The monkeys were free to look wherever they wanted unless their lines of sight were out of a rectangular electronic window (15 × 30°) centered at the partner’s bar zone for longer than 2 s (“gaze requirement”). The correctness of task performance was judged independently from each other. Thus, when one of the two monkeys made an error (break in speed or gaze requirement), the LEDs for that monkey turned off immediately, indicating no chance of its reward. However, the other monkey was allowed to continue the task performance. The trial-end cue occurred when at least one of the two monkeys correctly completed a trial (i.e., 25 s) or when both monkeys made an error. No extra reward was given to the monkeys for the occurrence of movement entrainment between them.

Fig. 1.

Behavioral paradigm and analysis in the paired condition. (A) Schematic illustration of the paired, mutually visible condition. (B, Top) plot of bar positions for a representative single trial. Blue line, monkey (MK) 1. Red line, MK2. Small values in the ordinate indicate flexion positions for both monkeys. Dotted vertical lines indicate the start and end of data analysis. (Bottom) fast Fourier transform of continuous bar positions to compute a “dominant frequency” (inverted triangles) for each monkey (Left). If the ratio of the two dominant frequencies (Mk1:Mk2) was within the range of 3:2 to 2:3 (gray area; Right), the data were further analyzed. The green dot denotes data from the example trial. (C) Phase information extracted using Hilbert transform. (D) Phase difference (Mk1 minus Mk2). Inset, rose diagram showing the distribution of phase differences. The blue line indicates the mean resultant vector. Note that the scales for the rose diagram (orange) and the mean resultant length (blue) are different. Values in parenthesis denote the direction and length of the mean resultant vector.

The coordinated nature of the two monkeys’ movements was assessed using phase difference. In each trial, we monitored the bar positions for both monkeys (Fig. 1 B, Top), extracted phase information using Hilbert transform (Fig. 1C), and obtained a phase difference at each time point (Fig. 1D). Then, on the basis of multiple points of phase differences, we calculated the direction and length of the mean resultant vector (Fig. 1D, blue line in Inset). Here, the mean direction provided an index of how different the phase was between the two monkeys in a given trial, and the mean resultant length provided an index of how consistent the phase difference was in that trial. These analyses were conducted for trials in which both monkeys continued correct performance for at least 7 s, and the dominant movement frequencies were comparable between the two monkeys (Fig. 1 B, Bottom; see Materials and Methods). In the example trial shown in Fig. 1, a Rayleigh test rejected the null hypothesis that the phase difference was uniformly distributed (R = 0.57, n = 19,039, P = 4.9 × 10⁻³²⁴; Fig. 1 D, Inset). This finding provided evidence for the occurrence of social motor entrainment that was characterized by phase anisotropy at the level of single trials.

The existence of phase anisotropy extended beyond single trials. When the mean resultant vectors were pooled across all trials for each monkey pair, the grand mean direction—i.e., the direction of the grand mean resultant vector—exhibited significant anisotropy (pair M–N, R = 0.54, n = 205, P = 1.0 × 10⁻²⁸, Fig. 2 A, Top; pair G–N, R = 0.66, n = 113, P = 4.0 × 10⁻²⁵, Fig. 2 B, Top; pair G–M, R = 0.23, n = 554, P = 2.5 × 10⁻¹³, Fig. 2 C, Top; Rayleigh test). Thus, a significant phase relationship in motor entrainment occurred nonrandomly and was generally consistent across trials. Furthermore, the grand mean direction differed markedly depending on monkey pairs. Specifically, the direction was –0.38 rad for the pair M–N (Fig. 2 A, Top, red line), –0.12 rad for the pair G–N (Fig. 2 B, Top, red line), and –2.01 rad for the pair G–M (Fig. 2 C, Top, red line). These observations suggest that the nature of movement entrainment is inherent to monkey pairs.

Fig. 2.

Pair and vision-dependent motor entrainment. (A–C) Motor entrainment in each monkey pair. Blue dots indicate end points of mean resultant vectors in individual trials. Red lines indicate the grand mean resultant vectors. Distribution of the mean directions (Right). (Top) mutually visible trials. (Bottom) mutually invisible trials. Values on the circumference of rose diagrams denote the scale for both the mean resultant vectors (blue dots) and the grand mean resultant vectors (red lines). Values in parenthesis denote the direction and length of the grand mean resultant vector. P values from the Rayleigh test. n.s. not significant. Pairs M–N (A), G–N (B), and G–M (C). (D) Comparison of the mean resultant length between mutually visible and mutually invisible trials. Mean (bar) ± SD (black horizontal lines). P values from Welch’s t test.

Effect of Visual Information on Movement Entrainment.

Next, we examined how visual accessibility to the partner's movement affected social motor entrainment. For this purpose, we placed an optical sheet that switched between transparent and opaque states under experimental control in front of the face of each monkey (Materials and Methods). In randomly selected trials (approximately one-sixth of trials), the optical sheet was switched from transparent to opaque for both monkeys simultaneously. This operation blocked visual accessibility to each other’s movement, while auditory feedback remained unchanged. We found that when the repetitive movements were mutually invisible, the consistent phase difference was no longer observed in any pair (M–N, R = 0.29, n = 29, P = 0.09, Fig. 2 A, Bottom; G–N, R = 0.11, n = 7, P = 0.92, Fig. 2 B, Bottom; G–M, R = 0.27, n = 33, P = 0.10, Fig. 2 C, Bottom; Rayleigh test). Moreover, the mean resultant length (radius of blue dots) decreased significantly with visual occlusion in the pairs G–N (t = 2.93, df = 118, P = 0.004, Welch’s t test; Fig. 2D) and G–M (t = 5.09, df = 585, P = 4.8 × 10⁻⁷, Welch’s t test; Fig. 2D). These findings demonstrate that visual information plays a critical role in the generation or maintenance of social motor entrainment.

We asked whether the above findings could be reproduced using alternative procedures. First, we tested the significance of phase anisotropy using a bootstrap method (repeat n = 1,000; see Materials and Methods). This analysis revealed in all monkey pairs that in the mutually visible condition, the length of the grand mean resultant vector was significantly larger compared to the length that was computed from the same number of data points (from −π to π) sampled randomly from a uniform distribution (SI Appendix, Fig. S2A; all monkey pairs P < 0.05, two-sided, bootstrap). In the mutually invisible condition, however, statistical significance was not detected in any pair (SI Appendix, Fig. S2B; all pairs P > 0.05, two-sided, bootstrap). A direct comparison performed with the same number of trials revealed that the length of the grand mean resultant vector was significantly larger in the mutually visible condition than in the mutually invisible condition in the pairs M–N and G–N (SI Appendix, Fig. S2C; both P < 0.05, two-sided, bootstrap). Second, we calculated the total duration of time for which phase differences fell within a certain narrow range (the mean phase difference ± 1/6 π) and divided it by the total trial length during which two monkeys were constantly moving (SI Appendix, Fig. S3A). This index—the proportion of entrainment time—was significantly larger in the mutually visible condition than in the mutually invisible condition in the pairs G–N and G–M (SI Appendix, Fig. S3B; M–N, t = 0.73, df = 34, P = 0.47; G–N, t = 4.2, df = 10, P = 1.7 × 10⁻³; G–M, t = 7.6, df = 52, P = 5.6 × 10⁻¹⁰; Welch’s t test). Thus, analyses of the mean resultant length and the proportion of entrainment time yielded consistent results.

We further asked how long it took from the start of each trial to the onset of motor entrainment and whether the initiation of entrainment might differ between the mutually visible and invisible conditions. For this purpose, we defined the time when the phase difference first entered the aforementioned range (the mean phase difference ± 1/6 π) as the “entrainment start time.” The analysis revealed that the entrainment start times were mostly distributed within 5 s after trial start in both conditions (SI Appendix, Fig. S3C). The mean start time was not significantly different between the two conditions in any pair (M–N, t = 1.3, df = 58, P = 0.19; G–N, t = −1.0, df = 6.2, P = 0.34; G–M, t = −1.4, df = 36, P = 0.16; Welch’s t test). However, a close inspection of the distribution reveals that rapid-onset motor entrainment (<0.5 s) was rarely observed in the mutually invisible condition (SI Appendix, Fig. S3C).

To further clarify the extent to which the attention to the partner’s movement affected motor entrainment, we calculated the proportion of the time spent looking at the partner’s bar zone per total movement time in mutually visible trials. On the basis of the median values (pair M–N, M = 0.523, N = 0.525; pair G–N, G = 0.520, N = 0.597; pair G–M, G = 0.494, M = 0.510), trials were grouped into two conditions: high mutual attention and low mutual attention. We confirmed that significant phase anisotropy was present in all the pairs regardless of high or low mutual attention to each other’s performance (Fig. 3 A–C, Top and Middle). The bootstrap method reproduced these findings (SI Appendix, Fig. S4 A and B), and additionally showed that the length of the grand mean resultant vector in the pair M–N was significantly larger in high mutual attention trials than in low mutual attention trials (SI Appendix, Fig. S4 C, Left; P < 0.05, two-sided, bootstrap). Gross inspection also revealed that the grand mean direction (red line) was largely similar between the high (Fig. 3 A–C, Top) and low (Fig. 3 A–C, Middle) mutual attention conditions. In support of this, the directional difference was not significant in the pairs G–N (F = 0.04, df = 1, P = 0.84, Watson-Williams test) and G–M (F = 0.00, df = 1, P = 0.97, Watson-Williams test), suggesting that the pair-inherent phase difference was robust regardless of whether visual inputs were derived from a more central or peripheral field. However, the mean resultant length (radius of blue dots) was generally shorter in low mutual attention trials compared to high mutual attention trials; the difference reached significance in the pairs M–N (t = 3.28, df = 97, P = 1.4 × 10⁻³, Welch’s t test; Fig. 3 A, Bottom) and G–M (t = 4.75, df = 260, P = 3.4 × 10⁻⁶, Welch’s t test; Fig. 3 C, Bottom).

Fig. 3.

Effect of attention to partner’s performance on motor entrainment. (A–C) Motor entrainment in each monkey pair. Same conventions as in Fig. 2A. (Top) high mutual attention trials. (Middle) low mutual attention trials. P values from the Rayleigh test. n.s. not significant. (Bottom) comparison of the mean resultant length between high and low mutual attention trials. Mean (bar) ± SD (black horizontal lines). P values from Welch’s t test. Pairs M–N (A), G–N (B), and G–M (C).

In addition to the effect on the consistency of entrainment, visual accessibility affected the speed of the repetitive movement. We found that the executed movement was significantly faster in the mutually visible condition than the mutually invisible condition, regardless of whether the actual (SI Appendix, Fig. S5A; t = 5.25, df = 332, P = 2.7 × 10⁻⁷, Welch’s t test) or normalized (SI Appendix, Fig. S5B; t = 7.72, df = 339, P = 1.3 × 10⁻¹³, Welch’s t test) speed was analyzed. This finding suggests that the observed movements accelerate one’s own movements. The vision-dependent movement facilitation was also evident in the dominant frequency (SI Appendix, Table S1) in monkey N (paired with monkeys G and M) and monkey M (paired with monkey G). In monkey G, the dominant frequency was consistently higher when paired with the higher-ranked monkey N than when paired with the lower-ranked monkey M (SI Appendix, Table S2).

Effect of Social Rank on Movement Entrainment.

Social dominancy can affect behavioral expressions in various task contexts in the macaque (21, 22). In the present study, social ranks, as assessed using a competitive food-grab test (Materials and Methods), were in the following order: N (high ranked) > G (middle ranked) > M (low ranked). We examined the possibility that aspects of social motor entrainment, such as a tendency of tuning in to others, might be associated with social ranks. Such tendencies, if any, would be more readily appreciable under asymmetrical gaze conditions (Fig. 4A).

Fig. 4.

Possible dependency of motor entrainment on social ranks. (A) Schematic illustration of the paired, asymmetrically visible condition. In this example (N→M), N can observe M’s performance but not vice versa. The social rank was in the order of N > G > M. (B–D) Motor entrainment in each monkey pair. Same conventions as in Fig. 2A. P values from the Rayleigh test. n.s. not significant. Pairs M–N (B), G–N (C), and G–M (D).

In a subset of trials (approximately one-sixth of trials), the optical sheet turned opaque for only one of the two monkeys. Thus, two directions of visual accessibility could be generated in each pair, i.e., M→N (M can observe N) and N→M (N can observe M) directions in the pair M–N. Under this asymmetrically visible condition, the consistent phase difference was not observed, except for the following two cases: N→M (R = 0.73, n = 20, P = 5.1 × 10⁻⁶, Rayleigh test; Fig. 4 B, Bottom) and N→G (R = 0.80, n = 12, P = 1.3 × 10⁻⁴, Rayleigh test; Fig. 4 C, Bottom). These two cases were similar in that the most dominant monkey N was capable of viewing the subordinate partner’s movement. The bootstrap method applied to viewer-based trial groups (i.e., N→G/M, G→N/M, and M→N/G) confirmed that phase anisotropy was significant only in the trial group in which monkey N could observe its partner (N→G/M; SI Appendix, Fig. S6, Left; P < 0.05, two-sided, bootstrap). The grand mean directions in the N→G and N→M trials were not significantly different from those under the mutually visible condition in the same pairs (N→G, F = 0.10, df = 1, P = 0.75; N→M, F = 0.10, df = 1, P = 0.75; Watson–Williams test). These findings suggest that high-ranked monkeys might have a stronger tendency to coordinate their movement pace with the tempo of others.

To test this possibility more directly, we measured transfer entropy (TE) using bar positions between two monkeys in each pair. TE is a nonparametric statistic that can detect directed exchange of information between two systems (23, 24). Here, we subtracted from a lower-ranked monkey to a higher-ranked one from TE from a higher-ranked monkey to a lower-ranked one. This difference in TE takes a negative value if lower-ranked monkeys predominantly affect higher-ranked monkeys.

In the mutually visible condition, the observed frequencies of positive and negative values were significantly different from equal frequencies and were biased in the direction of negative values (SI Appendix, Fig. S7, Top Row) in the pairs N–M (P = 1.5 × 10⁻⁴, binomial test) and N–G (P = 2.3 × 10⁻⁸, binomial test). The same pattern was observed in these pairs in the high mutual attention condition, in which two monkeys paid close attention to each other’s bar zone (SI Appendix, Fig. S7, second row from top; N–M, P = 3.8 × 10⁻⁴; N–G, P = 3.5 × 10⁻⁴; binomial test). In the asymmetrically visible condition, a similar finding was obtained in the N→G condition (SI Appendix, Fig. S7, second row from bottom; P = 6.3 × 10⁻³, binomial test). These findings support the hypothesis that the movement of the most dominant monkey (N) was influenced more readily by the movement of subordinate monkeys.

Effect of Reward Size on Movement Entrainment.

In the present task, the monkeys received a reward for every several shots, and the reward size was increased in proportion to the total number of shots within a single trial (Materials and Methods). This reward setting raises the possibility that social motor entrainment might actually be reinforced by reward. If this is the case, motor entrainment would be more obvious with a larger reward size. To test this possibility, we calculated phase differences every time a reward was given and obtained the absolute value for their deviations from the mean phase difference in each trial. We then compared the obtained deviations between when the reward size was smaller than the median and when the reward size was larger than the median.

We found that the deviations from the mean phase difference were affected by the reward size in several conditions. However, the deviations were significantly smaller—i.e., better alignment to the pair-dependent phase difference—when the reward size was smaller (SI Appendix, Fig. S8). Moreover, this reward dependency was only observed in trials where the most dominant monkey (N) was involved (SI Appendix, Fig. S8). Thus, it seems likely that the reward dependency was not a common characteristic in motor entrainment but might be a feature associated with this dominant monkey. Although precise mechanisms underlying reward dependency remain to be determined, these findings demonstrate that motor entrainment in the present study was not reinforced by reward.

Reduced Movement Entrainment under Offline Conditions.

Thus far, we have demonstrated several properties of motor entrainment between two monkeys facing each other. We next asked whether motor entrainment was similarly elicitable when facing prerecorded videos of a moving monkey or object. In this offline condition, movies of a life-sized monkey performing the same repetitive movements (offline monkey condition) or movies of a moving bar alone (offline bar condition) were replayed on a large monitor in front of the subject monkeys (Fig. 5A). For analysis of movement entrainment, the phase difference was computed in the same manner as in the paired condition.

Fig. 5.

Diminished motor entrainment in offline video conditions. (A) Schematic illustration of two offline conditions. (B) Motor entrainment in the offline bar condition. n.s. not significant. Same conventions as in Fig. 2A. (C) Comparison of the mean resultant length between the offline bar condition, offline monkey condition, and paired condition. Mean (bar) ± SD (horizontal black lines). P values from Welch’s t test. (D) Motor entrainment in the offline monkey condition. P values from the Rayleigh test. Same conventions as in Fig. 2A.

In the offline condition, the mean resultant vector varied between trials despite facing the same monkey or bar across trials. Notably, in the offline bar condition, none of the subject monkeys exhibited significant anisotropy at the population level (Fig. 5B; monkey N, R = 0.08, n = 64, P = 0.67; monkey G, R = 0.10, n = 68, P = 0.48; monkey M, R = 0.07, n = 113, P = 0.57; Rayleigh test). In the offline monkey condition, however, monkeys N and G showed significant population anisotropy (Fig. 5D; monkey N, R = 0.26, n = 74, P = 6.8 × 10⁻³; monkey G, R = 0.16, n = 133, P = 0.028; Rayleigh test). In these two monkeys, there was a significant main effect of task conditions in the mean resultant length (Fig. 5C; monkey N, df = 1, F = 9.7, P = 2.0 × 10⁻³; monkey G, df = 1, F = 67.0, P = 9.9 × 10⁻¹⁶; one-way ANOVA). Specifically, compared to the paired condition, the mean resultant length for monkeys N and G was significantly shorter in the offline bar condition (monkey N, t = 4.39, df = 108, P = 2.6 × 10⁻⁵; monkey G, t = 5.73, df = 83, P = 1.6 × 10⁻⁷; post hoc Welch’s t test) and in the offline monkey condition (monkey N, t = 2.22, df = 111, P = 0.028; monkey G, t = 8.18, df = 215, P = 2.4 × 10⁻¹⁴; post hoc Welch’s t test). Thus, in the offline conditions, the movement entrainment was more readily elicitable when facing conspecifics compared to objects. However, phase difference was most consistently observed in the real-time paired condition. Note that the most subordinate monkey (M) demonstrated no sign of anisotropy in either the offline bar (R = 0.07, n = 113, P = 0.57, Rayleigh test) or offline monkey (R = 0.05, n = 80, P = 0.85, Rayleigh test) condition. Monkey M also showed the least difference in the mean resultant length between the offline and real-time paired conditions (df = 1, F = 0.094, P = 0.76; one-way ANOVA). These findings were reproduced in the analysis of the proportion of entrainment time (SI Appendix, Fig. S9A). The entrainment start time was consistently earlier in the real-time paired condition compared to the two offline conditions (SI Appendix, Fig. S9B).

Although movement entrainment was less remarkable in the offline conditions, the speed of the executed movement was affected by the speed of the observed movement. We used movies with three levels of speed, i.e., faster than, comparable to, and slower than the executed movement (Materials and Methods). We found that the executed movement speed increased significantly as a function of the observed movement speed (Fig. 6A; ρ = 0.38, P = 4.9 × 10⁻⁵, Spearman’s rank correlation). Notably, the executed movement speed was significantly faster when the monkey was present than when the monkey was absent (Fig. 6B; t = 2.69, df = 108, P = 8.7 × 10⁻³, Welch’s t test). Moreover, the movement of the bar alone had a facilitatory effect on motor performance; the executed movement was significantly faster in the offline bar condition than in a control condition in which a stationary image was presented (Fig. 6C; t = 7.29, df = 47, P = 2.9 × 10⁻⁹, Welch’s t test). Thus, the observation of movement per se, be it animate or inanimate, accelerates motor performance. However, the facilitatory effect is greater when an animate movement is observed.

Fig. 6.

Effect of observed movement on executed movement speed. (A) Increase in the speed of executed movement as a function of the speed of observed movement. P values from the Spearman correlation test. In box-and-whisker plots, the central rectangle spans the first to the third quartiles, and the colored segment inside the rectangle denotes the median; the ends of the whiskers represent the minimum and maximum of the data excluding outliers. The density plots show estimated density curves based on the concentration of data points. (B) Comparison of executed movement speed between the offline bar and offline monkey conditions. Same conventions as in A. P values from Welch’s t test. (C) Comparison of executed movement speed between trials with stationary images and those with offline bar motion. Same conventions as in A. P values from Welch’s t test.

Discussion

In the present study, we provided evidence for spontaneous motor entrainment between two monkeys under laboratory conditions. Our task involved repetitive arm movements for bar sliding, and the motor entrainment occurred without human prompting, i.e., without additional rewards for the occurrence of intersubject phase synchrony. These findings are generally consistent with previous observations that nonhuman primates—both great apes and macaques—exhibit motor synchrony between conspecifics (15–19). Our findings are also in line with prior work that the degree and timing of interindividual motor entrainment can depend on visual inputs and animal pairs in macaques (20).

Despite apparently ample evidence supporting interindividual entrainment in nonhuman primates, critical information is lacking. Specifically, the extent to which motor entrainment is a social phenomenon was unclear in previous studies. This question remains unsolved because the nature of entrainment, such as its frequency and magnitude, has not been compared between conditions in which observed movements are made by a conspecific and a physical object. This issue is not trivial considering that macaques can synchronize their movements with rhythmic presentation of visual objects in nonsocial contexts (25). Another fundamental issue is whether the movements of real conspecifics and videotaped conspecifics are equally effective at eliciting motor entrainment in the observer. Addressing this issue also helps to understand the social nature, as well as the conditions of occurrence, of movement entrainment. By comparing behavioral manifestations between several task conditions, we showed that motor entrainment is best tuned to real-time social interactions.

In the offline-monkey condition, the lack of significant anisotropy in the lower-ranked animal (monkey M) suggests that the higher-ranked animals (N and G) show better motor entrainment. However, there was a confounding factor that the animal in the video was the most dominant monkey (Materials and Methods). If the social rank of the monkey in the video had been lower than monkey M, monkey M might have demonstrated better motor entrainment. What seems more critical in the present study is the observation suggesting that the movement of the higher-ranked monkey is affected more readily by the movement of the lower-ranked monkey. Specifically, in the asymmetrically visible condition, significant anisotropy was observed when the higher-ranked monkey was capable of viewing the lower-ranked monkey but not vice versa. Moreover, the amount of information transfer between two monkeys, as indexed by TE, was greater in a direction from the lower-ranked monkey to the higher-ranked monkey. Thus, although there was a limitation that the number of trials was small in the G→N condition, our findings suggest that the higher-ranked monkey collects more abundant information about others’ movement, be it automatically or deliberately. We hypothesize that in this way, the higher-ranked monkey may become a driver of motor entrainment in social contexts.

Our findings suggest that spontaneous phase synchrony between biological agents emerges when they exchange visual information about body movements. Such visual information is likely to be processed through a “third” visual pathway on the lateral brain surface (26). This pathway projects from the primary visual cortex, via motion-selective areas, into the superior temporal sulcus (STS) (26) and exhibits a greater neural response to moving than stationary faces and bodies (27–29). The third visual pathway is hypothesized to play a special role in computing visual actions of other biological organisms to support constantly changing social interactions (26). It has been shown in macaques that single neurons in the dorsal bank and fundus of the middle STS respond to the movements of real more than videotaped conspecifics (30). The role played by the third visual pathway in social motor entrainment is certainly an interesting topic for future research.

It has been advocated that social motor entrainment promotes cooperation and group cohesion (5, 31). This notion is supported by a number of psychological studies in humans (6–12). However, the neural basis of the social bonding effects is yet to be clarified. Movement entrainment is characterized by perception–movement coupling between two agents. The primate brain is equipped with the so-called mirror system, in which movement observation and movement execution activate the same set of neurons (32). This neural machinery may allow others’ movements to automatically and instantaneously affect the motor system in the observer, thereby causing synchrony between two agents at both the neural (33) and behavioral (1, 2, 4) levels. Behavioral entrainment may lower the perceptual boundaries between the self and others. In support of this view, the synchronized others are perceived to be more similar to oneself (11). Such self-other equivalence or blurring is hypothesized to underlie the social bonding effects of synchronization (5). We conjecture that the mirror system is a critical node for social motor entrainment and social bonding. An important question for future work is how activity in the mirror system differs between conditions in which motor entrainment appears and disappears. A similarly important question is whether and how the mirror system might be functionally linked to subcortical regions controlling the release of neurohormones that affect social bonding, such as oxytocin, vasopressin, and endorphin (5, 34). Indeed, neural coordination between the social brain and hypothalamus—the central hub of the neuroendocrine system—was recently identified in the macaque (35). These unsolved questions are testable using our experimental paradigms. Whether and, if so, how the motor entrainment in the present task contributes to group cohesion will be an important question for future work.

Materials and Methods

Animals.

Three male monkeys were used for the experiments [M. fuscata; N (aged 9 y, 7.0 kg), M (aged 11 y, 7.6 kg), and G (aged 8 y, 6.4 kg)]. These monkeys were housed in individual cages but were capable of communicating with other monkeys both visually and verbally. The animal care and experimentation protocols were approved by the Institutional Animal Care and Use Committee of the National Institutes of Natural Sciences and were carried out in accordance with the guidelines described in the US NIH Guide for the Care and Use of Laboratory Animals. This study is reported in accordance with ARRIVE guidelines.

Task Training (Solo Condition).

A square panel was placed horizontally on the front of a primate chair (Fig. 1A). A straight stainless track was mounted on the panel. The monkeys were required to move a vertical bar on the track from side to side by extension and flexion of the elbow. The track was inclined upward to the right side (30°) on the horizontal plane for ease of movement. The position of the bar was continuously monitored using a digital laser sensor (Keyence LV-H32 and IL-600) with a sampling rate of 1 kHz. On the side of each track end, an LED was placed to indicate the initiation of a trial as well as the direction of an upcoming movement. For the trial initiation, both sides of LEDs turned red for 0.7 s along with a high-pitched tone. The completion of extension and flexion was detected every time the bar passed a predetermined boundary for each movement. Because of this procedure, “completion” did not necessarily indicate the extreme bar position for each movement. The duration of one-way movement (shot) was defined as a temporal difference between two consecutive completion times. For example, the duration of the current extension shot was the difference between the time of the current extension completion and the time of the preceding flexion completion. The monkeys were required to make the duration of each shot to be longer than 0.4 s (speed requirement). When the monkeys successfully performed the repetitive movements for 25 s, in accordance with previous human studies (36, 37), or failed to meet the speed or gaze requirement, LEDs went out along with a low-pitched tone (trial-end cue).

Paired Condition.

Two monkeys in primate chairs were seated face to face. The distance between their face was 1 m; direct body contact was not possible between them. The task requirement in the paired condition was exactly the same as in the solo condition. The outcome of one monkey did not affect the opportunity for the other monkey to perform the task. This means that even if one monkey failed, the other monkey was allowed to continue the task. For this purpose, the trial-end cue was delivered at the time when at least one of the two monkeys correctly completed a trial (i.e., 25 s) or when both monkeys made an error. An optical sheet (TANYO FOGLEAR, Cloudpoint Inc., Tokyo, Japan) was placed in front of the face of each monkey. We randomly switched the sheet between a transparent mode and an opaque mode. The opaque mode was used in less than one-third of trials and was used equally for one or both monkeys.

Offline Condition.

A large LCD monitor (W57 × H97 cm) was placed 1 m in front of the monkeys. The monitor replayed movies of a life-sized monkey performing the same repetitive movements (offline monkey condition) or movies of a moving bar alone (offline bar condition). At the end of each trial, the movies were replaced by a blank screen.

For the creation of movies in the offline monkey condition, we recorded monkey N performing the task in the solo condition using a video camera (HC-V480MS, Panasonic Co., Tokyo, Japan) at 30 frames per second with a resolution of 1,920 × 1,080 pixels. The original movie was 28 s long, consisting of 2 s of initial waiting, 25 s of motor execution, and 1 s of postexecution waiting. The dominant frequency (see below) of the repetitive movements was 0.47 Hz. Using commercially available software (Premiere Pro, Adobe Inc., San Jose, USA), the original movie was edited to create three movies with different dominant frequencies of 0.37 Hz, 0.74 Hz, and 1.27 Hz for the slower, comparable, and faster conditions, respectively. The final resolution of these movies was 1,536 × 864 pixels to make them life-sized. To record the original movie in the offline bar condition, the experimenter moved the bar using an invisible stick from behind black curtains. A total of four movies were edited with the same movement frequencies as in the offline monkey condition.

White marks were added to movie frames (out of the subject’s field of view) in which the bar was at the two extreme positions. The time of the bar reaching each extreme position was detected using photodiodes. Based on such temporal information, the bar positions were linearly reconstructed.

Competitive Food-Grab Test.

We performed a food-grab test to determine the social hierarchy of the monkeys (38). Two monkeys in each primate chair were seated face to face with an interface distance of 65 cm. Both animals were hungry at the start of the test. The experimenter put a small piece of sweet potato at the center of a table (27 × 17 cm) that was placed in between the two primate chairs. The monkeys were free to grab and eat the food piece on a first-come-first-served basis. The test was continued until the difference in the number of foods acquired by the two monkeys was >10, and a winning percentage of a dominant monkey was >70%. The hierarchical rank determined in this way was in the following order: N (winning percentage against G and M was 72.7% and 90.5%, respectively) > G (winning percentage against M was 92.9%) > M.

Reward.

During a single trial, the monkeys received a reward for 1 to 7 shots (reward cycle; typically 3 shots) through a spout under the control of a solenoid valve. The quantity of reward was gradually increased within a trial as follows:

$$\begin{array}{c}\mathrm{Reward} \mathrm{quantity} = \mathrm{baseline} + 0.03 \hfill \\ \times \mathrm{baseline} \times (\mathrm{cumulative} \mathrm{shot} \mathrm{number})\hfill \end{array}$$

No extra reward was given for the occurrence of movement entrainment.

Monitoring of the Eye Position.

The eye position of the monkeys was monitored using an infrared video tracking system (iRecHS2, Human Informatics Research Institute, National Institute of Advanced Industrial Science and Technology), with a sampling rate of 187 Hz. In trials in which the optical sheet was not opaque, the monkeys were required to gaze within the partner’s bar-motion zone (15 × 30°, paired condition; 9 × 15°; offline conditions) at least for 0.4 s. Once the eye position was out of the zone, the monkeys were required to redirect their gaze to the bar-motion zone no later than 2 s, and their eye position had to stay there for at least 0.4 s (gaze requirement).

Behavioral Recording Procedures.

The time of extension and flexion completion was monitored using Arduino uno Rev3 (Arduino SRL. Torino, Italy). Stimulus (LED) presentation, monitoring of speed requirement, switching of the optical sheet, start and end of video playback, reward delivery, and monitoring of gaze requirement were controlled using a personal computer running the MonkeyLogic MATLAB toolbox (39).

Statistics.

No statistical methods were used to predetermine sample sizes. Dataset S1 were assumed to be normally distributed, but this was not formally tested. Data collection and analysis were not performed blinded to the conditions of the experiments. No data were excluded, except for trials in which the duration of correct performance was shorter than 7 s. All statistical analyses were performed using two-tailed tests, unless otherwise stated, and performed using commercial software (MATLAB 2018a and 2020b; MathWorks Inc., MA, USA).

Data Analysis.

Movement entrainment was assessed in the following steps. First, we performed fast Fourier transform of bar positions and determined a frequency component with the maximum power (dominant frequency; Fig. 1 B, Bottom). This analysis was performed for the motion of two monkeys (paired condition) or for the motion of a monkey and a video (offline condition). Second, if the ratio of the two dominant frequencies was within the range of 3:2 to 2:3, we extracted phase information from the bar positions using Hilbert transform (Fig. 1C). We chose the above range of the frequency ratio so as not to include integer-fold rhythmic synchronization other than 1:1. Even when the movement frequencies of two monkeys are in integer multiples, e.g., 0.5 Hz versus 1.0 Hz (=1:2), and steady rhythmic synchronization occurs, the phase difference between them cannot be used as an indicator of entrainment because the value increases with time and is never constant. Finally, a phase difference was obtained at each time point (Fig. 1D). These analyses were conducted for trials in which both monkeys continued correct performance for at least 7 s.

The mean direction and resultant length of phase difference were calculated in each trial on the basis of relative frequency in rose diagrams (bin width =1/12π) as follows:

$$\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n} \mathrm{d}\mathrm{i}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n} \stackrel{-}{\mathrm{\Theta }}=\mathrm{a}\mathrm{r}\mathrm{g}\left(\sum _{j=1}^n,e^{i{\mathrm{\Theta }}_j}\right),$$

$$\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n} \mathrm{r}\mathrm{e}\mathrm{s}\mathrm{u}\mathrm{l}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t} \mathrm{l}\mathrm{e}\mathrm{n}\mathrm{g}\mathrm{t}\mathrm{h} \stackrel{-}{R}=\frac{1}{n}\left|\sum _{j=1}^n{e}^{i{\mathrm{\Theta }}_j}\right|,$$

where $\mathrm{\Theta }\mathrm{j}$ is a relative frequency in jth phase difference bin, n = 24, and i is an imaginary number. A Rayleigh test was used to determine nonuniformity of phase differences in each trial and the mean directions across trials. Welch’s t test was used to compare the mean resultant length between experimental conditions. The Watson–Williams test was used to determine whether the grand mean direction was different between experimental conditions.

The significance of phase anisotropy was also tested using a bootstrap method. Suppose there were N_A trials in a trial condition A in a certain monkey pair. For Dataset S1, we randomly sampled N_A data points with replacement from 10,000 points separated by equal intervals on the circumference of the unit circle (from −π to π). Using these N_A data points, we computed the length of the mean resultant vector. We repeated this procedure 1,000 times and obtained the 2.5 and 97.5 percentiles. If the actual length of the grand mean vector in condition A was below the 2.5 percentile or above the 97.5 percentile, we considered it to be statistically significant. Likewise, we compared the length of the grand mean resultant vector between different trial conditions using a bootstrap method. Suppose there were N_A and N_B (N_A > N_B) trials in trial conditions A and B, respectively, in a certain monkey pair. To correct for differences in sample size, we randomly sampled N_B trials with replacement from trial condition A and computed the grand mean resultant vector. We repeated this procedure 1,000 times and obtained the 2.5 and 97.5 percentiles. If the actual length of the grand mean vector in condition B was below the 2.5 percentile or above the 97.5 percentile, we considered it to be statistically significant.

To assess the time of occurrence of motor entrainment in each trial, we detected time points at which phase differences were within a certain narrow range (the mean phase difference ± 1/6 π). Then, we defined the first time point in each trial as the entrainment start time for that trial. We also defined the total duration of entrainment time divided by each trial length as the “proportion of entrainment time.”

To examine the reader–follower relationship, we calculated TE. TE is a measure of information theory and evaluates the direction of information transfer between two systems (23), i.e., the causality between movements of two paired monkeys for each trial. Specifically, TE is a gap between the entropy of a monkey’s future movement determined by the current movement of the monkey and the entropy of the monkey’s future movement determined by the current movements of the monkey and another monkey as follows:

$$\begin{array}{c}Entropy{1}_{\text{X}\leftarrow XY}=-{\displaystyle \sum }_{i,j,k}p\left(x_{t+\tau }^i,x_t^j,y_t^k\right) {\log }_2 p\left(x_{t+\tau }^i|x_t^j,y_t^k\right),\end{array}$$

$$\begin{array}{c}Entropy{2}_{\text{X}\leftarrow \text{X}}=-{\displaystyle \sum _{i,j,k}p\left(x_{t+\tau }^i,x_t^j,y_t^k\right) {\log }_2 p\left(x_{t+\tau }^i|x_t^j\right),}\end{array}$$

$${\mathrm{TE}}_{\mathrm{X}\leftarrow \mathrm{Y}}=\mathrm{Entropy}{2}_{\mathrm{X}\leftarrow \mathrm{X}}-\mathrm{Entropy}{1}_{\mathrm{X}\leftarrow \mathrm{XY}},$$

where i, j, and k indicate the state number of phase (−π to π, step size = 1/3π, 6 states), and $x_t$ indicates the current phase of higher-ranked monkey X. So $x_t^j$ indicates the phase of monkey X at a time t in the state j. $y_t$ indicates the current phase of lower-ranked monkey Y, and $x_{t+\tau }$ indicates the time-shifted phase of monkey X. $p(A|B)$ denotes the conditional probability, and $p(A,B)$ denotes the joint probability. Tau was set at from 1 to 500 ms. The same calculation was performed to assess the effect of the current phase of monkey X on the time-shifted phase of monkey Y as follows:

$$\begin{array}{c}Entropy{1}_{\text{Y}\leftarrow XY} =-{\displaystyle \sum _{i,j,k}p\left(y_{t+\tau }^i,x_t^j,y_t^k\right) {\log }_2 p\left(y_{t+\tau }^i|x_t^j,y_t^k\right),}\end{array}$$

$$\begin{array}{c}\text{Entropy}{2}_{\text{Y}\leftarrow \text{Y}} =-{\displaystyle \sum _{i,j,k}p\left(y_{t+\tau }^i,x_t^j,y_t^k\right) {\log }_2 p\left(y_{t+\tau }^i|y_t^k\right),}\end{array}$$

$${\mathrm{TE}}_{\mathrm{Y}\leftarrow \mathrm{X}} =\mathrm{Entropy}{2}_{\mathrm{Y}\leftarrow \mathrm{Y}} - \mathrm{Entropy}{1}_{\mathrm{Y}\leftarrow \mathrm{XY}}.$$

Then, we averaged the subtraction of TE_X←Y from TE_Y←X. When the causality from monkey X (higher ranked) to monkey Y (lower ranked) was larger, the value was positive. Therefore, the sign of subtraction is important to judge the relationship between two monkeys. We performed a binomial test to determine the deviation of information transfer from a distribution with no directional bias.

To determine a potential effect of reward on motor entrainment, we calculated phase differences at the time of reward delivery and obtained the absolute value for their deviations from the trial-averaged phase difference. The obtained deviations were compared between when the reward size was smaller than the median and when the reward size was larger than the median (Watson–Williams test).

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Data, Materials, and Software Availability

All data discussed in the paper are available in the main text and SI Appendix.