Importance of a species’ socioecology: Wolves outperform dogs in a conspecific cooperation task

Significance

A popular hypothesis is that during the course of domestication, dogs acquired a tamer temperament, showing increased tolerance and cooperative inclinations compared with their wolf relatives. This “domestication effect” is suggested to mirror how humans evolved a more tolerant and cooperative nature compared with chimpanzees. However, whereas wolves rely heavily on cooperation for hunting, pup-rearing, and territorial defense, dogs’ reliance on cooperation is much reduced. Here we compared similarly raised and kept wolves and dogs on a cooperative string-pulling task and found that, in line with the different socioecology of the two species, wolves better cooperate with their conspecifics than dogs. Furthermore, cooperation in wolves was more successful among partners of similar rank and with a close social bond.

Abstract

A number of domestication hypotheses suggest that dogs have acquired a more tolerant temperament than wolves, promoting cooperative interactions with humans and conspecifics. This selection process has been proposed to resemble the one responsible for our own greater cooperative inclinations in comparison with our closest living relatives. However, the socioecology of wolves and dogs, with the former relying more heavily on cooperative activities, predicts that at least with conspecifics, wolves should cooperate better than dogs. Here we tested similarly raised wolves and dogs in a cooperative string-pulling task with conspecifics and found that wolves outperformed dogs, despite comparable levels of interest in the task. Whereas wolves coordinated their actions so as to simultaneously pull the rope ends, leading to success, dogs pulled the ropes in alternate moments, thereby never succeeding. Indeed in dog dyads it was also less likely that both members simultaneously engaged in other manipulative behaviors on the apparatus. Different conflict-management strategies are likely responsible for these results, with dogs’ avoidance of potential competition over the apparatus constraining their capacity to coordinate actions. Wolves, in contrast, did not hesitate to manipulate the ropes simultaneously, and once cooperation was initiated, rapidly learned to coordinate in more complex conditions as well. Social dynamics (rank and affiliation) played a key role in success rates. Results call those domestication hypotheses that suggest dogs evolved greater cooperative inclinations into question, and rather support the idea that dogs’ and wolves’ different social ecologies played a role in affecting their capacity for conspecific cooperation and communication.

In recent years, building on the hypothesis that dogs and humans may have gone through a process of convergent evolution, researchers have suggested that dogs—in addition to nonhuman primates—might be a good model for investigating the evolution of human social cognition (1–4). Due to intense selection during domestication, dogs are thought to have evolved genetic predispositions for cooperative interactions (5–7), a process suggested to mirror the “self-domestication” of humans, whereby having become more tolerant of one another, during our evolutionary history we evolved stronger cooperative tendencies compared with other members of our great ape family (8–10; but see ref. 11). Based on such hypotheses, compared with wolves, dogs are expected to show a higher propensity for cooperation, not just with humans but also with conspecifics (12).

Interestingly however, the socioecology of wolves and dogs would suggest the reverse. Wolves live in tight-knit family groups that strongly rely on cooperation for hunting, pup-rearing, and territorial defense (13–15). In contrast, studies of free-ranging dogs [which form 80% of the world-dog population (16, 17)] show that, although group hunting can occur, foraging is mostly carried out solitarily on human refuse (18, 19) and that there is little allomaternal care of pups (20–22). Indeed cooperation in free-living dogs appears to be largely limited to territorial defense (23, 24). Hence, based on the socioecology of the two species, we would expect wolves to perform at least as well as dogs, if not better, in cooperative tasks with their conspecifics (25, 26).

To assess these contrasting hypotheses, we tested similarly raised pack-living wolves and dogs housed at the Wolf Science Center in Vienna (www.wolfscience.at/en/) in the cooperative loose-string paradigm (Fig. 1). In this task, food is placed on an out-of-reach tray. A loose string is looped through rings on the tray, with the two ends of the string placed within the animal’s enclosure at such a distance that a single individual cannot reach both ends and pull them simultaneously. In test trials, two individuals are given access to the enclosure at the same time, and cooperation is observed if they coordinate their actions so as to simultaneously pull on the two ends of the rope, thereby moving the platform forward, allowing them to access the out-of-reach food. However, if only one end of the string is pulled, the other end becomes inaccessible and the tray cannot be moved forward anymore, rendering the trial unsuccessful. This task has been used with a wide range of species, from ravens to elephants [chimpanzees (27–29), macaques (30), elephants (31), gray parrots (32), rooks (33), ravens (34), kea (35, 36), and dogs (37)], with many succeeding in solving the task after being initially trained individually to pull the tray out by pulling both ends of the rope together.

Fig. 1.

Wolves working in the loose-string paradigm. Image courtesy of Rooobert Bayer (Wolf Science Center, Ernstbrunn, Austria).

Results

At the Wolf Science Center in Vienna, wolves and dogs live in conspecific packs, composed of between two and six individuals. Overall, we tested 12 wolves (8 male, 4 female) and 14 dogs (7 male, 7 female): that is, a total of 16 wolf dyads and 10 dog dyads (i.e., all of the available within-pack dyads) in different conditions. Initial dyads (tested in the Spontaneous condition, see below) were chosen based on the closeness of their affiliative bonds, giving precedence—where possible—to mixed-sex dyads over single-sex dyads.

Animals were tested in a number of different conditions (Fig. 2 for a decisional flowchart of conditions). First, each dyad was presented with the string-pulling apparatus with no prior training (Spontaneous condition). Dyads were simultaneously allowed into the enclosure and presented with six to eight sessions of six trials each, depending on performance/motivation [average of 36 (range 9–48) trials] (SI Materials and Methods for details). If animals were successful on at least four trials in each of the last two sessions, the dyads were presented with the Two-tray condition, where two identical apparatuses were presented 10-m apart in the same enclosure and animals were released at the same time (six sessions, six trials per session). This condition tested whether partners could coordinate their actions in both space and time. Finally, following the Two-tray condition and regardless of the dyads’ performance, subjects were presented with a Delay condition (six sessions, six trials per session). In this condition, one animal was released 10 s after the other, allowing us to test whether the subject released earlier would wait for their partner’s arrival before pulling the rope.

Fig. 2.

Schematic depiction of the experimental procedure.

If a dyad failed to solve the Spontaneous condition, the two individuals were tested in a different dyad. If they succeeded in the new dyad, they were retested with their former partner (for retesting information, see Table S1). In packs, where no dyad was successful, similarly to previous studies using the loose-string paradigm (37), each individual went through a training procedure, whereby the individual learned that when ropes were placed close enough to each other, holding both in the mouth and pulling would allow it to solve the task (SI Materials and Methods). Following this training procedure, dyads were tested again with a single apparatus (Fig. 2) (SI Materials and Methods).

Table S1.

All wolf dyads (sex combination and pack membership) tested in all conditions

Pack	Dyad	Sex	Spontaneous total_trials (success)	Posttraining (P)/retesting (R) total_trials (success)	Two-tray total_trials (success)	Delay total_trials (success)
1	Kaspar_Shima	M–F	45 (25)	NA	42 (11)	Kaspar 36 (23) –Shima 36 (3)
1	Kaspar_Tala	M–F	36 (34)	NA	48 (24)	Tala 36 (22)
1	Aragorn_Chitto	M–M	35 (27)	NA	36 (29)	Chittto 36 (20)
1	Aragorn_Shima	M–F	36 (23)	NA	36 (3)	Aragorn* 36 (8)
1	Kaspar_Aragorn	M–M	36 (31)	NA	36 (4)	Aragorn* 36 (26)
1	Tala_Shima	F–F	36 (10)	NA	36 (14)	NA
1	Kaspar_Chitto	M–M	36 (7)	R-36 (36)	36 (36)	NA
1	Chitto_Shima	M–F	37 (3)	R-36 (32)	36 (36)	NA
1	Chitto_Tala	M–F	34 (3)	R-36 (36)	36 (36)	NA
1	Aragorn_Tala	M–F	33 (2)	R- 36 (36)	36 (31)	NA
2	Amarok_Kenai	M–M	32 (1)	P-36(0)	NA	NA
2	Geronimo_Amarok	M–M	37 (0)	NA	NA	NA
2	Geronimo_Kenai	M–M	36 (0)	NA	NA	NA
3	Geronimo_Vukon	M–F	NA	P-36 (28)	36 (31)	Geronimo 36 (33) –Yukon 36 (27)
3	Wamblee_Yukon	M–F	40 (2)	P-36 (5)	NA	NA
4	Nanuk_Una	M–F	48 (3)	P-36 (33)	36 (28)	Nanuk 36 (34) –Una 36 (32)

F, female; M, male. Retesting refers to dyads that failed the first time they were tested together; each individual was tested with another partner, and then retested with the initial partner. Individuals were tested in the Delay condition only once, with the first partner with whom they were successful. The exception to this was Aragorn*, who was first tested with Shima and performed rather poorly. Upon inspection of the movies, it became obvious that his poor performance was attributed to Shima; he was tested again in the Delay condition with Kaspar.

Dog–Wolf Comparison.

Seven wolf and eight dog dyads were compared in the Spontaneous condition. Wolf and dog dyads had comparable experiences with the task, in that either (i) both partners had never been exposed to the apparatus (i.e., they were completely task-naïve) or (ii) one individual had been tested previously with another pack member but had failed to solve the task (Table S2 for details). All animals had previous experience of pulling a string to obtain an attached piece of food. Five of the seven wolf dyads succeeded in at least one trial (and across dyads success rates were in between 3% and 56% of trials), while only one of the eight dog dyads succeeded and only in one trial.

Table S2.

Wolf and dog dyads (sex combination and pack membership) compared in their success rates when tested in the Spontaneous and Posttraining conditions, including number of trials completed by each dyad, number of successful trials, and number of failures due to one individual pulling out the rope (the default failure being a time-out)

Pack	Dyad	Sex	Species	Condition	Total trials	No. success trials	No. rope-out fails	Condition	Total trials	No. success trials	No. rope-out fails
1	Nuru-Layla*	M–F	Dog	Spontaneous	40	0	2	NA
1	Zuri-Layla*	F–F	Dog	Spontaneous	36	0	0	NA
1	Nuru_Zuri	M–F	Dog	Spontaneous	32	0	1	Posttraining	36	0	0
2	Asali_Bora	M–F	Dog	Spontaneous	38	0	3	Posttraining	36	0	22
3	Maisha_Binti	M–F	Dog	Spontaneous	32	0	1	Posttraining	36	0	0
4	Meru_Nia	M–F	Dog	Spontaneous	33	0	1	Posttraining	36	1	6
2	Asali_Banzai	M–F	Dog	NA				Posttraining	36	0	20
2	Bora_Banzai	F–F	Dog	NA				Posttraining	36	0	11
5	Imara_Hiara	F–M	Dog	Spontaneous	36	1	26	NA
6	Sahibu_Gombo†	M–M	Dog	Spontaneous	9	0	6	NA
1	Kaspar_Shima	M–F	Wolf	Spontaneous	45	25	14	NA
1	Chitto_Tala	M–F	Wolf	Spontaneous	34	3	23	NA
2	Geronimo‡_Amarok	M–M	Wolf	Spontaneous	37	0	26	NA
2	Geronimo‡_Kenai	M–M	Wolf	Spontaneous	36	0	20	NA
2	Amarok_Kenai	M–M	Wolf	Spontaneous	32	1	7	Posttraining	36	0	22
3	Wamblee_Yukon	M–F	Wolf	Spontaneous	40	2	26	Posttraining	36	5	9
3	Geronimo‡_Yukon	M–F	Wolf	NA				Posttraining	36	28	7
4	Nanuk_Una	M–F	Wolf	Spontaneous	48	3	24	Posttraining	36	33	2

In bold are dyads tested in the Spontaneous condition, where one or both partners had been tested with a previous partner but unsuccessfully.

^*Layla, when being trained in the individual training condition, stopped showing any interest in the task and was dropped from the study.

^†Testing for this dyad was interrupted after only nine trials because during the last three trials they showed an escalation in the intensity of threatening behaviors toward each other.

^‡After being tested in the Spontaneous condition, Geronimo was removed from the pack due to an illness. After recovery it was no longer possible to integrate him with his former pack mates; hence, it was not possible to retest him in the Posttraining condition with Amarok and Kenai. Geronimo was, however, tested with his new pack-mate Yukon.

Next, four wolf dyads (three of which had been tested in the Spontaneous condition but with less than 10% success rate) (Table S2) and six dog dyads (four of which had been tested in the Spontaneous condition) (Table S2) were submitted to a training procedure similarly to previous studies, where they learned to solve the string-pulling task alone by taking both ends of the rope in their mouth and pulling (SI Materials and Methods for details on training regimen). All dyads were then retested in six sessions of six trials each (Fig. 2 and Table S2). Three of the four wolf dyads succeeded in 14–92% of trials, while only two of the six dog dyads succeeded in a single trial (3%).

Overall, the wolves outperformed the dogs regardless of condition [generalized linear mixed-model (GLMM) χ² = 10.418, P < 0.0001] and dyads were significantly more successful after individual training (GLMM: χ² = 38.64; P < 0.0001; no species × condition interaction: χ² = 1.94; P = 0.16) (Fig. 3).

Fig. 3.

Boxplots (median, interquartile range, outliers) showing the success rates (in percentage of trials in which dyads succeeded) for wolves and dogs in the Spontaneous and Posttraining conditions.

To investigate why wolves, but not dogs, successfully cooperated, we analyzed the behaviors exhibited by both individuals in a dyad during the first test session of both the Spontaneous and Posttraining conditions (i.e., when the wolves’ and dogs’ experiences with the task was still comparable).

Two elements are important for animals to be able to pull both ends of the rope at the same time and thus to succeed: (i) they need to explore and manipulate the apparatus sufficiently to discover the “correct” behavior and (ii) they need to tolerate each other’s presence and activity at the apparatus.

To assess the first aspect, we compared wolves and dogs on the frequency with which individuals showed (i) rope-pulling behaviors (exhibited not at the same time as the partner, so it did not lead to success) and (ii) other nonfunctional behaviors (i.e., biting, scratching, pawing, and so forth), which allowed us to assess the rate of manipulation in wolves and dogs.

As regards individual rope pulling, a species-by-condition interaction emerged (GLMM: χ² = 4.01, P = 0.045); hence, we ran separate models on the Spontaneous and Posttraining conditions. In the Spontaneous condition, wolves did significantly more individual rope-pulling than dogs (GLMM: χ² = 4.57, P = 0.03; Spontaneous: dog mean 0.5, range 0–6; wolf mean 0.7, range 0–10). In Posttraining trials, no effect of species emerged (GLMM: χ² = 0.84, P = 0.35; Posttraining: dog mean 1.6 range 0–11; wolf mean 1.7.5, range 0–13).

The relative frequency of the other nonfunctional manipulations of the apparatus did not differ between wolves and dogs (no effect of species GLMM: χ² = 1.47, P = 0.22) and was not affected by condition (effect of condition GLMM: χ² = 1.09, P = 0.029; no species × condition: χ² = 0.96, P = 0.32).

To assess whether wolf and dog dyads differed in their level of tolerance, we analyzed: (i) the latency it took both animals to be within one body length of the apparatus and the relative duration of trial time both individuals were simultaneously present at the apparatus, (ii) the likelihood that in a trial both animals would simultaneously manipulate the apparatus (i.e., biting, scratching, and pawing it), and (iii) the likelihood of dominant, aggressive, and submissive behaviors occurring during testing.

Wolf and dog dyads did not differ in the latency until both partners were close to the apparatus nor in the relative duration both individuals spent in proximity to the apparatus (latency: species: GLMM: χ² = 1.52, P = 0.22; no interaction species × condition GLMM: χ² = 2.09, P = 0.15; duration: GLMM: species: χ² = 0.008, P = 0.93; no species × condition interaction: χ² = 1.42, P = 0.23). Wolf and dog dyads approached the apparatus faster in Posttraining than Spontaneous trials (GLMM: χ² = 4.9, P = 0.03) and both species spent more time in proximity of the apparatus in Posttraining than Spontaneous trials (GLMM: χ² = 5.51, P = 0.02).

However, wolf dyads were significantly more likely to simultaneously manipulate the apparatus, even if these were nonfunctional behaviors (e.g., biting, pawing, scratching, and so forth) than dog dyads (GLM: χ² = 4.85, P = 0.027). Of course, wolves also significantly more frequently pulled the rope ends at the same time, leading to their higher success rate (see results relating to successful performance outlined above).

Dominant and aggressive behaviors (SI Materials and Methods) were rare in both wolves and dogs, occurring in a total of 12% of trials (wolves: 8 trials, 4 dyads; dogs: 10 trials, 5 dyads); the likelihood of these behaviors occurring was not affected by species (GLMM: χ² = 0.0001, P = 0.99), but they were more likely to occur after the training (GLMM: χ² = 4.5, P = 0.044; no species × condition interaction: χ² = 0.047, P = 0.83). Submissive behaviors (e.g., including withdrawing from the apparatus when another approached) also occurred rarely [i.e., in 11% of trials (wolves: 10 trials, 6 dyads; dogs: 6 trials, 3 dyads)] and the likelihood of their occurrence was affected by neither species nor condition (GLMM: species χ² = 1.59, P = 0.2; condition χ² = 0.25, P = 0.61; species × condition interaction: χ² = 0.08, P = 0.78).

Finally, as a potential measure of their coordination abilities, we evaluated gaze alternation behaviors (i.e., the rate of gazing from partner to apparatus and vice versa). No effect of species emerged (GLMM: χ² = 0.2, P = 0.65 and no species × condition χ² = 2.17, P = 0.14), but gaze alternation was more frequent in both species after training (GLMM: χ² = 38.81, P < 0.0001), potentially suggesting they had a better understanding of the need for the partner after individual string-pulling training.

Factors Affecting Wolves’ Success.

Due to the wolves’ success, we continued to investigate their cooperative performance. A total of 16 wolf dyads were tested in the Spontaneous and Posttraining conditions and of these, 12 dyads succeeded in passing the set criterion (i.e., four of six trials in the last two sessions) to be tested in the Two-tray condition (Table S1); the latter condition requiring them to coordinate their actions in both space and time. To evaluate their understanding of the need for a partner, the wolves were subsequently presented with a Delay condition in which the subject was released into the testing enclosure 10 s before the partner (Fig. 2).

We first compared success rate across conditions for those wolf dyads that did indeed pass criteria in the Spontaneous/Posttraining (One-tray), and therefore were presented with the subsequent Two-tray and Delay conditions and found no effect of condition (n = 12, GLMM: χ² = 0.08, P = 0.96) (Fig. 4).

Fig. 4.

Boxplots (median, interquartile range, outliers) showing the success rates (in percentage of trials in which dyads succeeded) for wolves across the three main conditions (Single-tray: Spontaneous, Retesting, and Posttraining; Two-tray and Delay conditions).

However, a learning effect was evident within each condition, since success rates increased across sessions in the Spontaneous and Posttraining (GLMM: χ² = 7.17, P < 0.0001), Two-tray (GLMM: χ² = 19.19, P < 0.001), and Delay (GLMM: χ² = 7.87, P = 0.005) conditions.

In the Two-tray condition, dyads succeeded on both trays on average in 74% of the trials (range 20–100%) (Table S1). In accordance with other studies (e.g., refs. 27–29, 34, and 37), we found that the stronger the affiliative bond measured during daily observations (SI Materials and Methods) (GLM: χ² = 8.6019, P = 0.003) (Fig. 5) and the smaller the rank distance (GLM: χ² = 25.82, P < 0.0001) (Fig. 6), the better dyads were able to coordinate their actions and obtain rewards from both trays.

Fig. 5.

The higher the wolf dyad’s affiliation score, the more successful they were in coordinating their actions to solve both apparatuses in the Two-tray condition.

Fig. 6.

The smaller rank distance between partners in a wolf dyad, the more successful they were in coordinating their actions to solve both apparatuses in the Two-tray condition.

In the Delay condition, one individual (Shima) performed rather poorly (8% success). A second individual (Aragorn), who was tested with her, also performed poorly (22% of trials); however, on inspecting videos, it was clear that the responsibility for failures was mostly due to Shima. Hence, Aragorn was retested with another partner. The remaining subjects’ performance (n = 8; including Aragorn’s retest with another partner) showed a success rate of between 55% and 94% of trials in the Delay condition. A subject’s performance (all tests included) was affected by the prior success rate they had with their partner in the Two-tray condition (GLM: χ² = 14.53, P < 0.0001) and in the Spontaneous/Posttraining conditions (GLM: χ² = 21.15, P < 0.0001).

SI Materials and Methods

Subjects.

All available dog and wolf subjects were tested in all possible pairs. For the wolf–dog comparison, a subsample of wolf dyads was used to ensure that the wolf and dog dyads had comparable experiences: all completely naïve pairs of both species were included, and dyads in which one partner had attempted the task before with another partner but without success (Table S2). Because in some cases wolf (but not dog) dyads were composed of an individual who had been previously successful in another dyad, and this could have affected their success rates, these wolf dyads were not included in the dog–wolf comparison (Table S1 for all wolf dyads tested in all conditions).

Apparatus and Test Location.

The apparatus was a 1.5-m × 75-cm food delivery tray with a rope passing through loops in the tray. The apparatus was placed on one side of the fence, and the ropes were placed in such a way that the ends lay on the ground in the testing enclosure. The rope was approximately 520-cm long, with 120 cm dangling from each end of the apparatus. The apparatus was such that if only one rope was pulled, the tray could not move, but rather the rope slid out of the loop system and the nonpulled end of the rope became unavailable. Adjacent to each rope, 20-cm apart from each other, two food-delivery areas were set, each containing one dead chick and one chunk of raw meat. To move the tray toward them and hence successfully obtain the food, both subjects needed to pull the two ends of the rope at the same time.

Testing took place in the two main testing enclosures at the Wolf Science Center, Vienna. The starting location of the animals was on the opposite side of the testing enclosure, both animals placed within the same cubicle (Fig. S1, showing the start location of each wolf, the location of the apparatus, and the location of the experimenter behind the barrier). Before running the test, the animals were allowed to explore the empty test enclosure for 5 min. During testing the experimenter was behind a screen, out of sight of the animals. Before the trial started the experimenter called the animals’ attention and showed them the food being placed on the apparatus.

Fig. S1.

Test set-up in the Spontaneous/Posttraining conditions (A) and the two-tray (B) condition. Figure courtesy of Akshay Rao and Corinna Kratz (Wolf Science Center, Ernstbrunn, Austria).

Procedure of Experimental Conditions in Detail.

Spontaneous condition.

The food tray was placed outside the animals’ reach, with the ropes going through the bars of the enclosure into the testing area. The members of the dyad were then simultaneously released into the enclosure. The animals’ spontaneous behavior toward the apparatus and toward each other was recorded. Trials lasted 2 min, or ended either once the task was completed successfully and the animals had finished the food, or when the rope was pulled solely on one side, making the other end unavailable and hence making the tray impossible to move. At the end of each trial the animals were called back to the start position, while the experimenter set up the task again for the following trial. Following test trials, if an individual never pulled the rope during the trial, motivation string-pulling trials were presented (see details below). All dyads were given an average of two sessions per week for a total of six to eight sessions.

Sessions 1 and 2 consisted of six test trials in which the dyad was allowed access to the apparatus. In sessions 3–6, the number of test trials was no longer fixed but varied from between two and six trials, depending on the animal’s behavior. If neither of the animals pulled the rope for two consecutive trials, the session was stopped to avoid demotivating the animals. All animals received motivation string-pulling trials following these sessions (see details below). If in session 6 animals were presented with fewer than four trials, two additional sessions were presented. In sessions 7 and 8 the animals again received a fixed number of six trials, independent of their performance, followed by motivation string-pulling trials if animals did not pull the ropes for two consecutive trials. Overall, the mean number of trials across animals was 36, with a range between 32 and 48 trials (Table S2).

Motivation string-pulling trials varied depending on session. In all cases, each animal was allowed in the enclosure without its partner. After sessions 1 and 2, a piece of meat was placed on the ground on the opposite side of the fence, out of direct reach of the animal. The meat was attached to a rope, which could be pulled by the animal to retrieve it. Following test trials in sessions 3 and 4 the food was placed on a wooden box outside the test enclosure, with the rope dangling within reach of the animal. These changes to the presentation were done to encourage the use of the mouth rather than the paw during string pulling. From sessions 5–8 in motivation string-pulling trials, the meat was placed on the test apparatus, alternating the position of the food between the two food slots, with the meat directly attached to the rope. These changes to the presentation were carried out to further encourage animals to view the apparatus as a potentially reinforcing object, even if they had had no success during testing. The number of motivation string-pulling trials depended on the animals performance in that as many trials were given as necessary for the animal to successfully retrieve the meat in three consecutive trials, with no prompting or encouragement.

Following the Spontaneous condition, dyads that had successfully obtained the food on at least four of six trials in the last two sessions conducted were presented with the Two-tray condition.

In packs where no dyad was successful, a training regimen (SI Materials and Methods) involving positive reinforcement was set up to allow animals to acquire the behavior of pulling both ends of the rope simultaneously when alone before testing them in dyads again (36).

Individual string-pulling training.

The overall aim of the training was to facilitate animals’ understanding of the mechanics of the task: that is, that pulling on a single rope would not allow them to obtain food, but pulling on both simultaneously would. Hence, to achieve this aim we adapted the procedures previously adopted by Seed et al. (33) with rooks and Ostojic and Clayton (37) with dogs.

In the first training phase: A 90-cm-long rope, folded in two so that the two ends were dangling from the end of the apparatus (the length of the dangling rope was 30 cm), was used. The ends of the rope were kept close together by cable ties placed at 10-cm intervals along the whole length of the rope. The rope was fixed in the middle of the apparatus so that the food tray moved forward (allowing subjects to obtain two pieces of meat) whenever the animals simultaneously pulled both ends of the rope. If animals reached criteria (see below) on this first step, the first cable tie (closest to the subject) was removed, so that the two ends dangled a few centimeters apart. In successive progressions we removed all of the cable ties, so that gradually the ends of the rope were further apart (the removal of the cables consisted of four training steps).

In the second training phase a small mesh (1-cm × 1-cm holes) was attached to the fence. This allowed us to have precise control of the distance between rope ends presented to the animals. In the first step of this second training phase, rope ends were first put through two holes next to each other (distance 0 cm). When the animals reached criteria (see below), they were presented with the rope ends 1-cm apart. We progressively increased the distance between rope ends until dogs successfully pulled the tray forward when presented with rope ends 6-cm apart.

In all cases (both for training phase 1 and training phase 2), the criterion to move on to the next step of training was that animals successfully pulled the platform in a minimum of five consecutive trials. However, if animals performed three consecutive unsuccessful trials, rather than continue with the current step, animals went back to the previous training step. Furthermore, after reaching criterion on the final step of each training phase, subjects had to successfully solve the step again on the next training session without going back to the previous phase before moving onto the next phase of the test (37).

Training was considered complete when animals successfully pulled the tray alone, in four of six trials with the rope ends 6-cm apart in two consecutive sessions. Having completed the training, individuals were paired with a partner who had also completed the training, and the dyads were tested in the Posttraining condition.

Two more steps were presented before retesting animals in pairs. For the first step, with the ropes 6-cm apart, the animals were released from the shifting system from the same location used during testing (i.e., on the opposite side of the enclosure in relation to where the apparatus was placed). Animals had to reach criteria (four successful trials of six) in this step to move onto the final step. The second step was identical to the previous step. However, in this case, as in test trials, the experimenter was behind a screen and not visible to the animals. This final step was carried out to ensure animals would still perform the string-pulling task independently (with no human social contact), after a long training period where the trainers were sitting next to the apparatus. Animals had to be successful again in four of six trials.

Posttraining condition.

The Posttraining condition was identical to the Spontaneous condition, in which each dyad was presented with a single apparatus and both animals were simultaneously released into the test enclosure. A total of six sessions consisting of six trials per session were conducted. The same pass criterion of four of six successful trials in three consecutive sessions was set. If this criterion was reached the dyad was presented with the Two-tray condition.

Two-tray condition.

Dyads that were successful in the Spontaneous or Posttraining condition were presented with two identical apparatuses placed 10-m apart from each other, in the same enclosure. The animals were released simultaneously from the start position as described above. A total of six sessions consisting of six trials each were carried out per dyad and again, trials lasted 2 min, or ended either once the task was completed successfully and the animals had finished the food or when the rope was pulled solely on one side, making the other end unavailable and hence making the tray impossible to move. A dyad was considered successful in a trial if they were able to obtain food from both trays. Regardless of performance in this condition, individuals were then presented with the Delay condition.

Delay condition.

In this condition, a single tray with two food sources was presented; however, instead of simultaneously opening the enclosure door for the two animals, partners were placed in adjacent enclosures and the subject was released 10 s before the partner. A total of six sessions each consisting of six trials was presented.

Behavioral Analyses.

Assessment of dyad’s relationship.

Daily observations of all pack members were carried out at the Wolf Science Center by long-term students, following a training period and interrater reliability assessment with senior staff. Ten-minute focal animal sampling, focused on the social interactions with other pack members, was carried out for each member of a pack during different times of day (Table S3). The ethogram adopted comprises behaviors which have been found to be good indicators of dominance in a number of studies in both wolves and dogs (53–55).

Table S3.

Relevant subset of the ethogram used at the Wolf Science Center to allow evaluation of partner relationship in terms of affiliation and rank distance

Behavior	Definition
Dominance behaviors
Stand tall	Subject straightens up to full height, with a rigid posture and tail, may include raised hackles, ears erect and tail perpendicular or above the back.
Paw on	To place one or both forepaws on the other’s back.
Ride up	To mount another one from behind or from the side, exhibiting a thrusting motion with the hips.
Head on	The subject approaches another’s shoulder/back and puts its head on it.
Muzzle Bite	To grasp the muzzle of another subject either softly or with enough pressure to make the other whimper.
Approach dominant	To approach another subject within one body length for at least 5 s, with the tail perpendicular or above the plane of the back and the ears erect and pointed forward and with a rigid posture.
Submissive behaviors
Crouch	Lowering the head, sometimes bending the legs, arching the back, lowering the tail between the hind legs, and avoiding eye contact.
Passive submission	To lie on the back showing the stomach and holding the tail between the legs. The ears are held back and close to the head and the subject raises a hind leg for inguinal presentation.
Active submission	The subject has its tail tucked between the hind legs sometimes wagging it while he is in a crouched position (with hindquarters lowered) and may attempt to paw and lick the side of actors’/aggressor’s muzzle. The behavior may also include urination.
Withdrawing	The subject withdraws from another moving away slowly in the opposite direction, displaying a submissive posture. It occurs when a subject has been threatened or attacked by another, or a fight has taken place.
Flee	To run away from another with tail tucked between the legs and body ducked. It occurs when a subject has been threatened or attacked by another, or after a fight.
Avoidance	In response to another reducing the distance toward it, the subject moves away displaying a submissive posture. The subject may also look at the individual he is trying to avoid.
Approach submissive	To slowly approach another within one body length remaining within that distance for at least 5 s. The approach is characterized by a ducked posture and tail between the legs. Subject can also be moving in a wavy line and in a hesitant (stop–start) manner.
Affiliative behaviors
Grooming	To nip, lick or scratch the fur or skin of another.
Lie friendly	To lie on the back, tail-wagging, maybe kicking with the foreleg against another subject sometimes with open mouth.
Stand friendly	The subject stands with tail perpendicular to or below the plane of the back, wagging it, ears pointed forward, while another is approaching it or orienting/looking toward it.
Body contact	Two subjects stay (for at least 10 s) with at least a part of their bodies in contact and in a relaxed position.
Social sniff	To sniff another’s body part except its anogenital area.
Body rubbing	To rub one’s body against any part of the receiver’s.

The Pocket Observer program (3.2 Software) was used for data collection, then imported into the Observer XT 10.5 program (both from Noldus Information Technology) for further analyses.

Based on such observations, we calculated: (i) an “affiliation score” for each dyad: that is, the bidirectional frequency of affiliative behaviors exchanged by individuals A and B, divided by observation time (hours) for subjects A + B; and (ii) the rank distance between members of the dyad: that is, subtracting the David’s score of individual A from that of individual B. The David’s score for each individual was calculated based on the frequency of dominant and submissive behaviors displayed toward other pack members. The David’s score is considered the most accurate measure of an individuals’ dominance status within the pack, since it takes into account the relative strength of all pack members, thereby also allowing an assessment of relative strength across groups with different numbers of members within (56).

No significant difference emerged between the wolves’ and dogs’ affiliation score (χ² = 0.3, P = 0.58) and rank distance (χ² = 1.04, P = 0.31) and no correlation emerged between rank distance and affiliation score for wolves (R = 0.27, P = 0.395).

String-pulling success.

In all conditions, a trial was considered to end in a successful cooperative interaction if the dyad succeeded in obtaining the food by both individuals simultaneously pulling on the rope ends so that the tray moved forward sufficiently for them to reach the food. On a number of occasions (fewer than five trials overall), an individual obtained a piece of food by reaching their paws or snout through the fence and moving the tray forward enough to obtain the food; this was not considered as a success and the trial in such cases was repeated.

Behavior coding of session 1 of the Spontaneous and Posttraining conditions.

The first session of both the Spontaneous and Posttraining conditions were observed on video and a detailed coding of the animals’ behaviors was carried out (Table S4 for ethogram). For analyses purposes, dominant and aggressive behaviors were summed in a single category, as were submissive and withdrawing behaviors. Interobserver reliability was carried out on 20% of the data, and was found to be high for all behaviors (all Spearman’s ρ > 0.93). Due to video malfunction, the Spontaneous and Posttraining sessions for one dyad (Wamblee–Yukon) could not be coded in detail; therefore, this dyad is not included in these analyses.

Table S4.

Ethogram used for the behavior coding of session 1 of Spontaneous and Posttraining conditions

Behaviors	Measure	Description
Dominant	Frequency	Stand tall, paw on, ride up, head on, muzzle bite
Aggressive	Frequency	Growl, snarl, snap, lunge
Submissive	Frequency	Crouch, active and passive submission
Withdraw	Frequency	Subject is within two body lengths of the apparatus and withdraws to outside of two body lengths from the apparatus within 2 s of partner’s approach or following an dominant/aggressive interaction
Close apparatus	Duration	Both partners simultaneously within one body length of the apparatus
Pull rope	Frequency	Subject pulls the rope (taut, not with slack), so that a movement in the rope is detected (not just holding in the mouth)
Nonfunctional behaviors on tray	Frequency	Subject bites, paws or scratches at the tray or the fence immediately in front of it
Gaze apparatus-partner	Frequency	Changing head orientation from the apparatus to the partner or vice versa

Statistical Analyses.

Wolf–dog comparison.

To compare the success rate of wolves and dogs in the task, we ran a GLMM with proportion of successful trials over all trials (since this could vary between dyads), including trials from both the Spontaneous and the Posttraining conditions, since we had a comparable number of wolf and dog dyads in both conditions (Table S2). Species and condition (and their interaction) were included as explanatory factors and dyad as the random factor (since some dyads were tested both in the Spontaneous and Posttraining conditions).

We further evaluated whether wolves and dogs showed a different pattern of behaviors during session 1 (i.e., when all animals were comparable in their experience with the apparatus). Hence, considering only session 1 for both Spontaneous and Posttraining conditions, we ran a number of GLMM models with the frequency (pulling the rope, nonfunctional behaviors, gaze alternation from partner to apparatus) and durations (time spent in proximity of the apparatus) of specific behaviors (normalized by trial duration) as the dependent factor, the species and condition (and their interaction) as the explanatory factor and the dyad as random factor. We additionally ran a GLMM model looking at whether species and condition (and interaction) affected the (i) latency it took both animals to be in proximity of the apparatus and (ii) the likelihood they would display dominant/aggressive interactions in a trial (binomial distribution). Finally, we assessed whether species had an effect over all trials in the likelihood that both individuals would simultaneously manipulate the apparatus (display nonfunctional behaviors) in a trial (binomial distribution). For this, we ran a GLM with the number of trials a dyad was shown to be simultaneously active on the apparatus corrected for the total number of trials that dyad was tested in (since this could vary) as the dependent variable and species as the explanatory factor.

Factors affecting wolf dyad cooperative success.

To evaluate performance across conditions for all wolf dyads (Table S1), we ran a GLMM with rate of success as the dependent variable, condition as the explanatory factor, and dyad identity as the random factor.

To assess potential learning effects within each condition (Spontaneous/Posttraining conditions, Two-tray and Delay), we ran a GLMM with rate of success as the dependent variable, session (including only the first six sessions of the Spontaneous condition, since these were carried out by all dyads) as the explanatory variable, and dyad as the random factor.

To evaluate the potential effect of the animal’s social relationship on the dyad’s performance, we considered data from the Two-tray condition for two reasons: (i) this condition better controls for the different levels of experience of the animals since they must all have had a measure of success in the Spontaneous condition before being tested in the Two-tray condition; and (ii) this condition requires a greater level of coordination between partners, since they need to synchronize their action both in space and time to successfully obtain food from both apparatuses. The latter level of coordination is more likely to be sensitive to the type of relationship between individuals. Accordingly, we ran a GLMM with the rate of success (i.e., number of trials in which animals solved both apparatuses) as the dependent measure, the dyad’s affiliation score and rank distance as the explanatory variables, and the dyad identity as a random factor.

Finally, we analyzed the dyad’s performance in the Delay condition, assessing whether the success rate of a dyad in the Spontaneous and Two-tray conditions affected the subject’s success in the Delay condition. We ran a GLM with the individuals’ rate of success in the Delay condition as the dependent variable, and the dyads’ percentage of successful trials in the Spontaneous and Two-tray conditions as explanatory variables.

All analyses were carried out in R v3.2.2 (57) package “lme4.” Model assumptions were checked and corrected for if not met. A model reduction procedure based on P values was adopted, starting from dropping interactions between factors if these were found not to be significant.

Discussion

Overall, in line with the different socioecologies of dogs and wolves, and the latter’s more conspicuous dependence on cooperative activities, results show that wolves consistently outperformed dogs in the cooperative string-pulling task.

Interest in the apparatus was comparable, since no differences emerged in the latency and duration of both animals being in proximity of the apparatus, nor the individual frequency of biting and pawing at it. However, in the Spontaneous condition, wolves manipulated the ropes more frequently than dogs, which likely increased the probability of dyads achieving success and learning from this experience. Differences in string-pulling frequencies between wolves and dogs are in line with studies showing that wolves tend to be more persistent in object manipulation than dogs, both in problem-solving tasks involving food (38, 39) and when exposed to novel objects (40, 41), and such basic differences may play an important role when assessing wolves and dogs in cognitive tasks (42).

Another possibility is that wolves had a better understanding of the task requirements, thereby more frequently performing the “correct” string-pulling behavior. However, wolves and dogs have been shown to be on a par in tests of means-ends understanding (43), which is the task most related to the string-pulling apparatus used in the present study. Therefore, it seems unlikely that dogs and wolves differed in their understanding of the task.

Indeed, in the present study, differences in the rate of string-pulling per se were not sufficient to explain differences in cooperative success, since after individual training on the apparatus, the rate of string-pulling between wolves and dogs was comparable yet wolves continued to outperform dogs. Hence, in this condition dogs and wolves showed similar levels of interest and individuals equally frequently displayed the appropriate behaviors on the apparatus, but crucially wolves were better at coordinating these behaviors so that they both pulled at the same time, thereby succeeding.

Tolerance has been shown to be a key factor in cooperative success in a number of species (28, 29, 34). In the present test, in the first session, we found no obvious difference in the time wolves and dogs spent simultaneously in proximity of the apparatus, in the frequency of agonistic behaviors, and whether the dominant or subordinate animal manipulated the ropes, which indicates a comparable degree of tolerance around the apparatus. Furthermore, wolf and dog dyads tested did not differ in their social relationships, in that no significant effect of species emerged for either their affiliation score or their rank distance (SI Materials and Methods). However, crucially, whereas wolves pulled each end of the rope at the same time (leading to success), dogs did not, and hence never succeeded. Indeed, dogs tolerated each other’s presence at the apparatus but they were significantly less likely than wolves to both engage in the task at the same time, most likely as a conflict-avoidance strategy over a coveted resource. This avoidance of a coveted resource in a potentially competitive context has been observed more often in dogs than wolves (44, 45), suggesting that there may be different social strategies adopted by the two species to avoid/resolve conflicts. Difference in such strategies may in turn affect cooperative success, where a shared resource is the ultimate aim.

Dogs in the present study performed very poorly, which contrasts with a study carried out with pet dogs, where the five dyads tested all succeeded in the cooperation task (37). It is likely that differences in the prior experience of dogs (pet dogs were all highly trained), and potentially methodological differences in the procedure, contributed to these discrepancies. But most importantly, pet dyads were composed of dogs living in the same household, where typically owners train their dogs not to engage in conflicts over resources, promoting a level of tolerance, which may facilitate cooperation. In contrast, pack dogs at the Wolf Science Center develop their social relationships with minimal human interference, in that animals are removed only if levels of aggression lead to potentially serious injuries. In this setting, competition over resources is likely to be higher, and conflict-avoidance strategies (constraining cooperation) may be more prevalent than in a pet dog environment. The large variability in results across the two very different dog populations studied so far highlights the high behavioral plasticity of dogs, which is likely one of the major ingredients of their success as human’s companions. In line with this, it would be of great value to test dog populations from the most diverse backgrounds on such a task (e.g., dogs selected for pack hunting, free-ranging dogs, highly trained working dogs, and so forth) to start teasing apart the factors promoting/inhibiting dogs’ conspecific cooperation.

Nevertheless, current results of dogs and wolves living under comparable conditions show wolves being strikingly more prone to coordinate their actions in a cooperative task. This is in line with the notion that wolves’ reliance on cooperative activities, such as group hunting, continues to place a high-selection pressure on their capacity for tolerance and coordinated actions. During the course of domestication, the alteration in dogs’ socioecology, leading to a reduced reliance on conspecific cooperative activities, likely relaxed the selective pressures for such skills (26). Future comparisons of wolves’ and dogs’ cooperative abilities in more naturalistic tasks, such as territorial defense or third-party conflict support, may help to disentangle whether such differences relate specifically to coordinated actions on an apparatus (i.e., where a resource is at stake) or are more generally extended to other aspects of the animal’s social environment.

In contrast to dogs, wolves overall showed a striking capacity to coordinate their actions in this task. Interestingly, when comparing the success rate of the 12 dyads across all three conditions, we found no significant difference (Fig. 4). These results suggest that during the single-apparatus condition, dyads learned the basic requirements of the task and were then able to flexibly apply them in the new contexts (Two-tray and Delay conditions). However, a number of elements appeared to be important in contributing to the wolves’ success. Dyads significantly improved their performance across sessions within each condition, suggesting that after an initial success, the positive feedback allowed them to rapidly acquire the necessary associative rules to continue to succeed. Indeed, in the Delay condition, where the individual’s understanding of the need for a partner was tested, the higher success rate of the subject was directly related to how often they had succeeded in the prior Two-tray and single apparatus (Spontaneous or Posttraining) conditions, suggesting that the animals learned the contingencies of the task during testing in these conditions.

Interestingly, besides the learning process, the social relationship played an important role in how well dyads succeeded in the Two-tray condition. Indeed in line with previous results in other species (28, 29, 34), we found that the strength of the social bonds was associated with the success with which wolves were able to coordinate their actions to obtain a reward from both apparatuses. Furthermore, the closeness of the rank between individuals was also related to higher cooperation success, similarly to results from chimpanzees (46) and hyenas (47) [although studies in other species found the opposite effect, with higher cooperative success being related to increased rank distance between partners (30, 34)]. In the Two-apparatus condition, the coordination between partners is particularly important; therefore, it is possible that animals closer in rank paid closer attention to one another compared with partners in dyads with larger rank distances. Indeed studies in dogs suggest that subordinates will learn a task from a dominant more readily than vice versa (48), and in rhesus macaques, gaze-following has been observed to be more likely to occur the closer the rank distance between partners (49). Future studies will be needed to test this hypothesis further.

Overall, the present study questions the hypotheses that dogs, during the process of domestication, have become better cooperators than wolves, and cautions against using the wolf–dog comparison as a model for hypotheses regarding human “self-domestication.” Indeed, studies on captive pack-living wolves and dogs suggest that considering dogs a “tamer/friendlier” version of wolves is an oversimplification. Rather, dogs appear to exhibit different behavioral strategies than wolves when interacting with conspecifics: showing fewer formal signals of dominance but higher intensity of aggression (50–52), a more persistent use of avoidance and distance maintenance in managing conflicts in the feeding context (44, 45), and a reduced inclination to coordinate actions in a cooperative task (present results). Taken together, such results suggest that changes in dogs’ socioecology, in particular their reduced dependence on conspecific cooperation in hunting and pup-rearing, may have significantly affected their intraspecific social behavior in a number of ways, highlighting the importance of taking socioecology into account in theories about domestication (26).

Materials and Methods

Details of the subjects, testing, training, coding of test and observations, as well as statistical analyses carried out are included in the SI Materials and Methods, Fig. S1, Tables S1–S4, Movies S1–S3, and Datasets S1 and S2. This study was discussed and approved by the institutional Ethics and Animal Welfare Committee at the University of Veterinary Medicine Vienna, in accordance with Good Scientific Practice guidelines and national legislation (Protocol number ETK-01/04/97/2014).