Rats to the rescue!


Editor's Introduction

Empathy and pro-social behavior in rats.

annotated by

Previous studies have shown that non-primate mammals will help each other when they're in trouble. However, researchers have not yet been able to figure out why they do this. In this scientific report, rats learned how to open cages containing a trapped rat that they "knew." This action was selective in that they did not open empty cages or cages with a toy rat. The helping behavior occurred even when no social interaction after freeing was possible. Furthermore, the rats chose to free a trapped cagemate even when they had a choice to open a restrainer containing chocolate. Are nonprimate mammals really capable of feeling empathy and acting on those feelings?

Paper Details

Original title
Empathy and pro-social behavior in rats.
Original publication date
Vol. 334 no. 6061 pp. 1427-1430
Issue name


Whereas human pro-social behavior is often driven by empathic concern for another, it is unclear whether nonprimate mammals experience a similar motivational state. To test for empathically motivated pro-social behavior in rodents, we placed a free rat in an arena with a cagemate trapped in a restrainer. After several sessions, the free rat learned to intentionally and quickly open the restrainer and free the cagemate. Rats did not open empty or object-containing restrainers. They freed cagemates even when social contact was prevented. When liberating a cagemate was pitted against chocolate contained within a second restrainer, rats opened both restrainers and typically shared the chocolate. Thus, rats behave pro-socially in response to a conspecific's distress, providing strong evidence for biological roots of empathically motivated helping behavior.


Pro-social behavior refers to actions that are intended to benefit another. One common motivator of pro-social behavior in humans is empathic concern: an other-oriented emotional response elicited by and congruent with the perceived welfare of an individual in distress (12). Sharing another’s distress via emotional contagion can result in overwhelming fear and immobility unless one’s own distress is down-regulated, thus allowing empathically driven pro-social behavior (34). Building on observations of emotional contagion in rodents (510), we sought to determine whether rats are capable of empathically motivated helping behavior. We tested whether the presence of a trapped cagemate induces a pro-social motivational state in rats, leading them to open the restrainer door and liberate the cagemate.

Rats were housed in pairs for 2 weeks before the start of testing. In each session, a rat (the free rat) was placed in an arena with a centrally located restrainer in which a cagemate was trapped (trapped condition, n = 30 rats, 6 females). The free rat could liberate the trapped rat by applying enough force to tip over the restrainer door (Fig. 1A). If a free rat failed to open the door, the experimenter opened it halfway, allowing the trapped rat to escape and preventing learned helplessness. Rats remained in the arena together for the final third of the session. Door-opening only counted as such if the free rat opened the door before the experimenter opened it halfway. Sessions were repeated for 12 days. Control conditions included testing a free rat with an empty restrainer (empty condition,n = 20 rats, 6 females) or toy rat–containing restrainer (object condition, n = 8 males). As an additional control, for the number of rats present, we tested a free rat with an empty restrainer and an unrestrained cagemate located across a perforated divide (2+empty condition, n = 12 males). Free rats’ heads were marked and their movements were recorded with a top-mounted camera for offline analysis (11).

Fig. 1.  (A) Top views of the trapped and 2+empty conditions and side views of the restrainer and door. (B) The locations (0.5 frames per second) of representative free rats with respect to the restrainer (red box) are plotted for each condition on day 1 of testing. (C) Rats in the trapped condition spent more time (mean ± SEM) in the arena center (>5 cm away from the wall) than did rats in control conditions. (D) The velocity (mean ± SEM) of rats in the trapped condition was greater than that of control rats throughout the session. (E) The ratio of the average activity during the second half of sessions relative to the average activity during the first half (mean ± SEM) was greater for rats in the trapped condition on days 1 to 6 than for rats in control conditions.
Panel A

Schematic representations of the experimental set up. The leftmost image shows a large rectangular shaped arena into which a free rat and a rat trapped inside of a restrainer (small rectangle) were placed.

The image in the center depicts the design of the restrainer. Note the perforations on the side to allow olfactory, auditory, and tactile communication between rats. The restrainer was large enough for the trapped rat to move and turn around in.

The image on the right depicts the rectangular plexiglass arena divided in half by a perforated wall. One side contained a free rat, the other a free rat and an empty restrainer.

Panel B

In order to allow analysis of the rat's movements by video tracking, the rat's head was marked just below the ear line with either a black felt sticker or Sharpie mark of approximately the same size. Each blue dot represents the location of the free rat in the arena during each frame that the camera captured on the first day of testing.

See video of active rats here: [https://www.youtube.com/watch?v=9QdAH6qz240]

Panel C

Free rats in an arena containing a trapped cagemate spent more of their time close to the restrained rat in the center of the arena. This is measured by the time the free rat spent more than 5 centimeters away from the arena wall.

By testing day 12, rats in the trapped condition spent about 60%of their time at the arena center. The empty, object, and 2+empty conditions spent about 40%, 30%, and about 25% of their time in the arena center, respectively.

Panel D

For the entirety of the testing session, rats in the trapped condition moved faster than rats in any of the control conditions. Once the door is opened at the halfway point, the rats in the trapped condition move at a velocity at least two times that of any other control.


Activity during the last 30 minutes of the session versus activity during the first 30 minutes of the session is higher for the rats in the trapped condition for testing days 1–6. This ratio is a measure of persistence, as rats that display brief flurries of activity and then stop moving will have a lower value than rats who continue to move around.

Free rats circled the restrainer, digging at it and biting it, and contacted the trapped rat through holes in the restrainer (Fig. 1B and movie S1). They learned to open the door and liberate the trapped cagemate within a mean of 6.9 ± 2.9 days. Free rats spent more time near the restrainer in the arena center [P < 0.001, mixed model analysis (MMA), Fig. 1C] and showed greater movement speed (hereafter termed activity, P < 0.001, MMA, Fig. 1D) than did control rats. Before learning to open the restrainer door, free rats in the trapped condition stayed significantly more active in the second half of sessions relative to the first half than did rats in control conditions [P < 0.001, MMA, protected least significant difference (PLSD) test, Fig. 1E]. Thus, rats were motivated to move and act specifically in the presence of a trapped cagemate.

Video. Author Peggy Mason narrates some footage of the rat behaviors researched in this article.

In the trapped condition, the proportion of rats that opened the door increased (Fig. 2A), and the latency to door-opening decreased (Fig. 2B and movie S2) across sessions, which is evidence of learning. Significantly more rats in the trapped [23 out of 30 (23/30)] than control (5/40) conditions were classified as “openers” by the end of the experiment (P < 0.001, χ-square test), opening the door within minutes of placement in the arena (11). A sharp increase in the free rat’s activity was observed at the time of door-opening (Fig. 2C), suggesting that the liberation of a trapped cagemate is a salient event.

Fig. 2.  (A) The proportion of rats in the trapped condition that opened the door increased across the days of testing. (B) Only rats in the trapped condition opened the door at decreasing latencies across days of testing. (C) Rats in the trapped condition showed a sharp increase in activity when the restrainer door was opened (time 0). (D) Across days, free rats in the trapped condition developed a consistent opening style, lifting the door up with their heads. (E) As rats learned to open the door, they stopped freezing in response to door-opening. (F) More alarm calls were recorded in the trapped condition (n = 67 sample files) than in empty (n = 64) or object (n = 67) conditions.
Panel A

For each day of testing, the number of rats that opened the door was divided by the total number of rats tested.

For the first 4 days, all the different conditions had a similar proportion of door openers. This is not surprising because it took about 6 days for all rats to learn to open the door.

For the last 6 days, however, the proportion of door openers increased ONLY for the trapped condition as seen by the upward trend in the graph (solid rectangles).

QUESTION: How did the proportion of rats in the control conditions that opened on the last day compare to the proportion that opened on the first day?

Panel B

The y axis is a measure of the median time it takes the rat to open the door. Instead of representing this in regular units of time, the authors represent latency as a percentage of the total time the rat spent in the arena.

This was done because two different lengths of testing time—90 minutes and 60 minutes—were used (see Supplementary Materials). By reporting the latency as a percentage of total time, all the data can be represented on one graph instead of two.

Each data point (filled rectangles, open rectangles, etc.) represent the time elapsed before door opening for a particular testing day. There are 12 dots; one for each day of testing. Data for four groups are displayed, but only three are visible because the symbols for object and empty groups sit on top of each other and are overlapping.

From day 5, there is a decrease in the amount of time it takes a rat in the trapped condition to open the door to the restrainer. In addition, these rats opened the door themselves before the experimenter opened it halfway.

QUESTION: What is the difference between a mean value and median value? What are the benefits of presenting a median value rather than the mean and vice versa?

Panel C

This graph presents data from the trapped cagemate condition only. Data were collected every minute from 10 minutes before the rat opened the door to 10 minutes after the door was opened for a total of 21 data points.

Each dot represents the average activity at that minute for all the free rats tested.

QUESTION: The increase in activity precedes the door opening by a couple minutes. Why might that be?

Panel D

Again, these data refer to rats in the trapped cagemate condition only. For each day of testing the number of rats that opened the door via lifting it up with their heads, leaning on the heavy side of the door, or pushing down from the top of the restrainer is shown.

QUESTION: How many more rats lift the door with their heads rather than from the side on day 7?

Panel E

Mean freezing time on the y axis represents the average amount of time all rats tested remained immobile after the door of the restrainer fell over. This was calculated for every day of the 12 testing days.

QUESTIONS: What was the longest amount of time the rats spent frozen? The shortest? On which testing days did this occur?

Panel F

This graph shows the proportion of all sampled audiofile sessions in which rats emitted alarm calls (as determined by ultrasonic frequency of ~23kHz).

In the trapped condition the authors made sure to differentiate calls coming from the trapped rat versus the free rat.

QUESTION: Why would the free rats in the empty and object conditions make alarm calls at all?

Initially, rats in the trapped condition opened the door in any of three ways: tipping the door over from the side or top or pushing it up with their heads. However, on days 6 to 12, they consistently opened the door with their heads (Fig. 2D). Furthermore, whereas rats initially froze after the door fell over, later on they did not freeze (Fig. 2E), demonstrating that door-opening was the expected outcome of a deliberate, goal-directed action.

Ultrasonic (~23 kHz) vocalizations were collected from multiple testing arenas with a bat-detector and were analyzed to determine whether rats emitted alarm calls. Significantly more alarm calls were recorded during the trapped condition (13%) than during the empty and object conditions [3 to 5%, P < 0.05 analysis of variance (ANOVA), PLSD < 0.05, Fig. 2F] in randomly sampled files from all days of testing. Alarm calls occurred more frequently (20 to 27%) on days 1 to 3, when door-opening was rare. In 90% of files containing alarm calls on day 1, the trapped rat was identified as the source; in the remaining samples, we were not able to identify the caller. These data suggest that trapped rats were indeed stressed.

A greater proportion of female rats (6/6) than male rats (17/24) in the trapped condition became door-openers (P < 0.05, χ-square), which is consistent with suggestions that females are more empathic than males (71213). Further, female rats in the trapped condition opened the restrainer door at a shorter latency than males on days 7 to 12 (P < 0.01, MMA, Fig. 3A). Female rats were also more active than males in the trapped condition (P < 0.001, ANOVA) but not in the empty condition (Fig. 3B).

Fig. 3.  (A) Females in the trapped, but not empty, condition opened the door at consistently shorter latencies than did males on days 7 to 12. (B) Activity was greater for females than males in the trapped, but not empty, condition.
Panel A

The y axis represents the time it took for a rat to open the restrainer door relative to the length of the entire testing session.

This was then converted to a percentage. This was done so that data from the 90-minute-long and 60-minute-long sessions could be combined into one graph.

Female helper rats opened the door for their trapped cagemate much more quickly than male helper rats did.

However, the pool of female rats was much smaller (six) than the pool of male rats tested (24).

QUESTION: How much faster did the average female rat open the restrainer door than the the average male rat?

Panel B

Female helper rats moved around the arena faster than male helper rats.

QUESTION: Do you consider a difference of 2 cm/s by the average female helper rat to be noteworthy?

To examine whether individual differences in boldness influenced door-opening, we tested the latency for approach to the ledge of a half-opened cage before the experiment (11). Animals who became openers had lower approach latencies than nonopeners (P < 0.01, t test), suggesting that successful opening behavior correlates with boldness scores (fig. S1). This demonstrates that individual trait differences may factor into the expression of pro-social behavior.

To determine whether anticipation of social interaction is necessary to motivate door-opening, we tested rats in a modified setup in which the trapped animal could only exit into a separate arena (separated condition, Fig. 4, A and B). Rats (12 pairs) were first exposed to the trapped condition (12 days); three rats did not open the door on any of the last 3 days and were not tested further. Next, rats were placed in the separated setup with a restrainer that was either empty (separated empty) or contained a cagemate (separated cagemate) for 29 days of testing. Finally, conditions were reversed so that rats previously in the separated cagemate condition were tested in the separated empty condition and vice versa, for 27 days. Thus, all nine rats were tested in counterbalanced order with both an empty and a full restrainer. Rats placed in the separated cagemate condition either continued or returned to opening the door at short latency as they had in the trapped condition. In contrast, when rats were placed in the separated empty condition, they stopped opening the door of the empty restrainer (P < 0.001, MMA, PLSD, Fig. 4, A and B). Thus, rats opened the door of a cagemate-containing restrainer but not of an empty restrainer, indicating that the expectation of social contact is not necessary for eliciting pro-social behavior.

Fig. 4. (A and B) Rats opened the door for a trapped cagemate even when no social interaction was possible between the two animals after door-opening. Door-opening was extinguished when the restrainer was empty but either resumed (A) or persisted (B) when the restrainer contained a cagemate, regardless of the order of testing [n = 4 rats, (A); n = 5, (B)]. (C) On days 6 to 12, the latencies at which rats opened a restrainer containing a trapped cagemate and one containing chocolate chips were not different. (D) Rats in the chocolate empty condition opened the empty restrainer at significantly longer latencies than the chocolate restrainer.
Panels A and B

Y axis shows average time to open the door in minutes, rather than as a percentage of entire session time.

Helper rats were first placed in an arena with a trapped cagemate for 12 days. They opened the restrainer door in 5 minutes or less.

Next, the helper rats were placed into an arena with an empty restrainer for 32 days (day 13–44).They took more than 10 minutes to open the restrainer door, sometimes never opening the door at all.

Finally, for the remaining 24 days, the helper rats were placed into an arena with a trapped cagemate that is released into a separate container upon opening of the restrainer door.

The helper rats once again open the door in less than 10 minutes.

QUESTION: What makes Panel B different than Panel A? Why did the researchers examine different orders of the testing environments?

Panel C

Helper rats are as likely to open a cage with chocolate chips as the cage containing their trapped cagemate.

QUESTION: Look closely at the lines connecting the open circles (the chocolate chip condition). On what day did the helper rat in the chocolate chip condition open the door before the experimenter?

Panel D

Rats were in an arena that contained two restrainers. One was empty, the other contained chocolate chips. As in panel C, the y axis is measuring the time the rat took before opening the door.

The first 6 days of testing reveal that rats open the restrainer containing chocolate faster than they open the empty one.

Compare the times it took for door opening on days 7–12.

QUESTION: Why would the rat open the empty container at all?

In order to examine the relative value of liberating a trapped cagemate, we tested a cohort of rats in a cagemate versus chocolate paradigm. When given a choice, these non–food-deprived rats ate an average of >7 chocolate chips and no rat chow, indicating that they found chocolate highly palatable. The free rat was placed in an arena with two restrainers, one containing the trapped cagemate and the other containing five chocolate chips (chocolate cagemate condition, Fig. 4, C and D). As a control, one restrainer was empty while the other contained chocolate (chocolate empty condition). For rats in the chocolate cagemate condition, there was no difference in the door-opening latencies for the two restrainers during days 6 to 12 (Fig. 4C). In contrast, rats in the chocolate empty condition opened the chocolate-containing restrainer more quickly than the empty one (P < 0.01, t test, Fig. 4D). These results show that the value of freeing a trapped cagemate is on par with that of accessing chocolate chips. Like rats in the trapped condition, rats needed several days (5.8 ± 2.1) to learn to open the chocolate restrainer, which is evidence that door-opening was neither easy nor instinctual.

Although free rats in the chocolate cagemate condition could potentially eat all five chocolate chips, they shared them in half of all trials (52%) and in 61% of trials on days 6 to 12. Rats in the chocolate empty condition ate virtually all the chips (4.8 ± 0.7), whereas free rats in the chocolate cagemate condition ate fewer chips (3.5 ± 1.5,P < 0.01, t test), which allowed trapped rats to eat the remaining chips (1.5 ± 1.4).

Our study demonstrates that rats behave pro-socially when they perceive a conspecific experiencing nonpainful psychological restraint stress (1415), acting to end that distress through deliberate action. In contrast to previous work (591617), the present study shows pro-social behavior accomplished by the deliberate action of a rat. Moreover, this behavior occurred in the absence of training or social reward, and even when in competition with highly palatable food.

Our observations could have alternative explanations. Rats may have acted to stop the alarm calls of the trapped rats (18). Yet alarm calls occurred too infrequently to support this explanation. Alternatively, rats may have been attracted to the trapped cagemate by curiosity. However, door-opening in the separated cagemate condition persisted for over a month, a time period over which curiosity extinguishes (19). Finally, door-opening could be a coincidental effect of high activity levels. This is unlikely because once rats learned to open the door, they did so at short latency, using a consistent style, and were unsurprised by door-opening. Additionally, door-opening is not easy, rendering accidental openings unlikely. Thus, the most parsimonious interpretation of the observed helping behavior is that rats free their cagemate in order to end distress, either their own or that of the trapped rat, that is associated with the circumstances of the trapped cagemate. This emotional motivation, arguably the rodent homolog of empathy, appears to drive the pro-social behavior observed in the present study.

The presence of empathy in nonhuman animals is gaining support in the scientific community (2026), although skeptics remain (27). In the current study, the free rat was not simply empathically sensitive to another rat’s distress but acted intentionally to liberate a trapped conspecific. The ability to understand and actively respond to the affective state of a conspecific is crucial for an animal’s successful navigation in the social arena (4) and ultimately benefits group survival.

Supporting Online Material


Materials and Methods
Fig. S1
Table S1
Reference (28)

References and Notes

  1. D. C. Batson, in The Social Neuroscience of Empathy, J. Decety, W. J. Ickes, Eds. (MIT Press, Cambridge, MA, 2009), pp. 3–15.

  2. J. Decety, P. L. Jackson, The functional architecture of human empathy. Behav. Cogn. Neurosci. Rev. 3, 71 (2004).

  3. N. Eisenberg et al., Relation of sympathy and personal distress to prosocial behavior: A multimethod study. J. Pers. Soc. Psychol. 57, 55 (1989).

  4. J. Decety, M. Svetlova, Devel. Cogn. Neurosci., 10.1016/j.dcn.2011.05.003 (2011).

  5. R. M. Church , Emotional reactions of rats to the pain of others. J. Comp. Physiol. Psychol. 52, 132 (1959).

  6. D. Jeon et al ., Observational fear learning involves affective pain system and Cav1.2 Ca2+ channels in ACC. Nat. Neurosci. 13, 482 (2010).

  7. D. J. Langford et al., Social approach to pain in laboratory mice. Soc. Neurosci. 5, 163 (2010).

  8. C. Zalaquett, D. Thiessen, The effects of odors from stressed mice on conspecific behavior. Physiol. Behav. 50, 221 (1991).

  9. G. E. Rice, P. Gainer, “Altruism” in the albino rat. J. Comp. Physiol. Psychol. 55, 123 (1962).

  10. D. J. Langford et al., Social modulation of pain as evidence for empathy in mice. Science 312, 1967 (2006).

  11. Materials and methods are available as supporting material on Science Online.

  12. T. Romero, M. A. Castellanos, F. B. de Waal, Consolation as possible expression of sympathetic concern among chimpanzees. Proc. Natl. Acad. Sci. U.S.A. 107, 12110 (2010).

  13. C. Mohr, A. C. Rowe, O. Blanke, The influence of sex and empathy on putting oneself in the shoes of others. Br. J. Psychol. 101, 277 (2010).

  14. G. B. Glavin, W. P. Paré, T. Sandbak, H. K. Bakke, R. Murison, Restraint stress in biomedical research: An update. Neurosci. Biobehav. Rev. 18, 223 (1994).

  15. W. P. Paré, G. B. Glavin, Restraint stress in biomedical research: A review. Neurosci. Biobehav. Rev. 10, 339 (1986).

  16. J. F. Lucke, C. D. Baton, Response suppression to a distressed conspecific: Are laboratory rats altruistic? J. Exp. Soc. Psychol. 16, 214 (1980).

  17. G. E. J. Rice, Psychol. Rec. 14, 165 (1964).

  18. J. J. Lavery, P. J. Foley, Altruism or arousal in the rat? Science 140, 172 (1963).

  19. M. L. Reger, D. A. Hovda, C. C. Giza, Ontogeny of rat recognition memory measured by the novel object recognition task. Dev. Psychobiol. 51, 672 (2009).

  20. F. B. de Waal, Putting the altruism back into altruism: the evolution of empathy. Annu. Rev. Psychol. 59, 279 (2008).

  21. F. Warneken, M. Tomasello, The roots of human altruism. Br. J. Psychol. 100, 455 (2009).

  22. S. Preston, in Encyclopedia of Animal Behavior, Mark Bekoff, Ed. (Greenwood Press, Westport, CT, 2004), vol. 2, D-P.

  23. G. D. Wills, A. L. Wesley, F. R. Moore, D. A. Sisemore, Social interactions among rodent conspecifics: A review of experimental paradigms. Neurosci. Biobehav. Rev. 7, 315 (1983).

  24. J. Decety, The neuroevolution of empathy. Ann. N. Y. Acad. Sci. 1231, 35 (2011).

  25. E. Hatfield, R. L. Rapson, Y. C. Le, in The Social Neuroscience of Empathy, J. Decety, W. J. Ickes, Eds. (MIT Press, Cambridge, MA, 2009), pp. 19–30.

  26. S. D. Preston, F. B. M. de Waal, Empathy: Its ultimate and proximate bases. Behav. Brain Sci. 25, 1 , discussion 20 (2002).

  27. J. B. Silk, in The Oxford Handbook of Evolutionary Psychology, R. I. M. Dunbar, L. Barrett, Eds. (Oxford Univ. Press, Oxford, 2007), pp. 115–126.

  28. D. Drai, I. Golani, SEE: A tool for the visualization and analysis of rodent exploratory behavior. Neurosci. Biobehav. Rev. 25, 409 (2001).

  29. Acknowledgments: This research was supported by grants from the National Institute on Drug Abuse (DA022978 and DA022429) to P.M. and from NSF (BCS-0718480) to J.D. The assistance of I. Boni, A. Brimmer, U. Fonio, K. Hellman, A. Logli, J. Peralta, K. Ragsdale, D. Rodgers, M. Sales, J. Wang, A. Weiss, and K. Yuan is gratefully acknowledged.