Temporal microstructure of dyadic social behavior during relationship formation in mice

Socially competent animals must learn to modify their behavior in response to their social partner in a contextually appropriate manner. Dominant-subordinate relationships are a particularly salient social context for mice. Here we observe and analyze the microstructure of social and non-social behaviors as 21 pairs of outbred CD-1 male mice (Mus Musculus) establish dominant-subordinate relationships during daily 20-minute interactions for five consecutive days in a neutral environment. Firstly, using a Kleinberg burst detection algorithm, we demonstrate aggressive and subordinate interactions occur in bursting patterns followed by quiescent periods rather than being uniformly distributed across social interactions. Secondly, we identify three phases of dominant-subordinate relationship development (pre-, middle-, and post-resolution) by utilizing two statistical methods to identify stability in aggressive and subordinate behavior across these bursts. Thirdly, using First Order Markov Chains we find that dominant and subordinate mice show distinct behavioral transitions, especially between tail rattling and other aggressive/subordinate behaviors. Further, dominant animals engaged in more digging and allogrooming behavior and were more likely to transition from sniffing their partner’s body to head, whereas subordinates were more likely to transition from head sniffing to side-by-side contact. Lastly, we utilized a novel method (Forward Spike Time Tiling Coefficient) to assess how individuals respond to the behaviors of their partner. We found that subordinates decrease their tail rattling and aggressive behavior in response to aggressive but not subordinate behavior exhibited by dominants and that tail rattling in particular may function to deescalate aggressive behavior in pairs. Our findings demonstrate that CD-1 male mice rapidly establish dominance relationships and modify their social and non-social behaviors according to their current social status. The methods that we detail also provide useful tools for other researchers wishing to evaluate the temporal dynamics of rodent social behavior.


54
A number of methods have been proposed for studying the temporal dynamics of social interaction in 55 laboratory animals but these are still underutilized. One approach has been to analyze the First Order 56 Markov transitions between successive behaviors produced by one individual to identify those behaviors 57 that transition between each other more frequently than expected by chance. This method has been  Forward STTC (FSTTC) = 1 Where is the proportion of behavioral onsets from that lie within time after any behavioral onset ∆ 227 from . is the proportion of total observation time which lies within time after any spike from . ∆ 228 is the proportion of total observation time which lies within time after any spike from . We used ∆ 229 throughout. The algorithm was written in C++ with a ready application in R. We calculated ∆ = 2

230
FSTTC values among behavior states (see Table 1   behaviors are not uniformly distributed as they occur in bursts of activity followed by periods of 271 quiescence. We used Kleinberg's burst detection algorithm to formally identify time periods containing bursts of aggressive and subordinate behavior (yellow bars in Fig 2). The median total number of bursts 273 per dyad over all five days was 20 with an IQR of 11-27 (

307
Note that we excluded any bursts with less than 6 behavior bouts to determine the relationship 308 resolution but those bursts are still shown in the graph.

309
As shown in Fig 3, Table S2).   419 We used timed-window cross-correlation to determine the likelihood of behavioral states being 420 exhibited by an animal within 2 seconds of when their partner exhibited a specific behavioral state 421 during the pre-and post-resolution phases. In Fig 7,   contingencies showing subordinate behavior almost always in response to each of these behaviors (Fig   440  7A).

441
We also find evidence that as relationships resolve, dominant animals continue to respond to aggressive 442 behaviors exhibited by their partner with aggression, but subordinate animals decrease this 443 contingency. We found no significant differences in contingency values pre-resolution between 444 dominant and subordinates in biting/lunging or tail rattling leading to biting/lunging. Post-resolution, 445 subordinate animals show significant reductions in these contingencies whereas dominant animals 446 maintain high values (Fig 7B). Interestingly, both pre-and post-resolution dominant and subordinate 447 animals were highly likely to respond to subordinate behavior by their partner with biting/lunging or tail 448 rattling, indicating that even subordinate animals may seek to be aggressive if their partner shows any 449 sign of yielding. Further, during the pre-and post-resolution phases both dominant and subordinate 450 animals respond to lunging and biting by their partner with tail rattling suggesting that tail rattling is an 451 indicator of arousal for all individuals. Interestingly, this behavior may actually be used as a signal to 452 deescalate aggressive behavior in pairs. Post-resolution, subordinates significantly decrease their tail rattling in response to the tail rattling by dominant animals whereas dominants continue to show high 454 contingencies of tail rattling in response to their partner's tail rattling (Fig 7C).