Motor cortical control of vocal interaction in neotropical singing mice

See allHide authors and affiliations

Science  01 Mar 2019:
Vol. 363, Issue 6430, pp. 983-988
DOI: 10.1126/science.aau9480

Turn-taking in singing mice

The ability to take turns is a hallmark of social interaction among animals. It occurs in many different species, from dueting birds to frogs, and is a notable part of human speech. Such rapid response requires a complex cascade of sensory and motor actions that has been difficult to characterize. Okobi et al. examined turn-taking in tropical singing mice, in which males interrupt, and alter, each other's songs (see the Perspective by Hage). They describe an orofacial motor cortex that mediates rapid transition from the motor cortex to the vocal motor apparatus and facilitates rapid vocal interactions.

Science, this issue p. 983 see also p. 926


Like many adaptive behaviors, acoustic communication often requires rapid modification of motor output in response to sensory cues. However, little is known about the sensorimotor transformations that underlie such complex natural behaviors. In this study, we examine vocal exchanges in Alston’s singing mouse (Scotinomys teguina). We find that males modify singing behavior during social interactions on a subsecond time course that resembles both traditional sensorimotor tasks and conversational speech. We identify an orofacial motor cortical region and, via a series of perturbation experiments, demonstrate a hierarchical control of vocal production, with the motor cortex influencing the pacing of singing behavior on a moment-by-moment basis, enabling precise vocal interactions. These results suggest a systems-level framework for understanding the sensorimotor transformations that underlie natural social interactions.

Adaptive behavior often requires adjusting action in response to a rapidly changing environment. Elucidating the mechanisms of these sensorimotor transformations has become a central focus of systems neuroscience, as researchers use simple and elegant behavioral tasks to explore the behavioral responses of traditional model species to sensory cues (13). Ultimately, however, we would like to understand such transformations in natural contexts; among such contexts, perhaps none is more challenging or interesting than social behavior. During social interactions, an animal must dynamically modulate complex actions in response to the changing behavior of a conspecific. For example, during conversation, we listen to the words of another person, interpret them, and respond appropriately (4). Indeed, acoustic exchanges are promising foci for the study of sensorimotor transformations that underpin social behavior. These exchanges are common across taxa, including insects (5, 6), amphibians (7, 8), birds (911), and mammals (1216); they serve a variety of essential social functions, including male-male competition and mate selection; and they require dynamic interaction as signalers avoid temporal overlap with one another (17).

Despite the ubiquity of acoustic interactions in the natural world, there are few existing models within neuroscience. Among mammals, for example, laboratory mice produce elaborate frequency-modulated vocalizations (18) but fail to exhibit robust turn-taking behavior (19). In contrast, marmoset pairs call antiphonally (14, 20), but the time scale of these interactions is relatively slow (3 to 5s) (4, 20, 21). In Alston’s singing mouse (Scotinomys teguina), we find a robust and rapid countersinging (~500 ms) that resembles the subsecond latencies of both conditioned sensorimotor transformations in laboratory settings (22) and the timing of vocal turn-taking evident in human conversation (4). We employ a range of techniques for manipulating neural dynamics to pinpoint a motor cortical locus that works hierarchically within the song production pathway to enable precise vocal interactions between conspecific pairs.

S. teguina is a small (~12 to 15 g), highly vocal neotropical rodent native to the cloud forests of Central America (2326) and is related to the genus Peromyscus and other New World rodents. Their family (Cricetidae) includes voles and hamsters and is in the same superfamily (Muroidae) as house mice and the Norwegian rat (27). Both male and female S. teguina produce vocal sequences consisting of a series of discrete frequency-modulated elements strung together, with characteristics that change predictably as the vocalization progresses (24) (Fig. 1, A to D, and movie S1). Following the convention of previous studies (23, 25), we refer to each vocal episode as a “song” and individual components as “notes.” We visualize this trend by plotting the duration of each note as a function of its onset time within the song: The song trajectory plot (Fig. 1D) provides a succinct representation of this motor sequence. We found that trajectory plots were highly stereotyped across renditions from individuals recorded in acoustic isolation (Fig. 1E). This degree of motor precision is reminiscent of vocalizations produced by a range of evolutionarily distant species (28, 29) but stands in stark contrast to the variable acoustic structure of ultrasonic vocal sequences produced by laboratory mice (18, 30, 31).

Fig. 1 Social context modulates vocalizations in S. teguina.

(A) An adult S. teguina in its natural habitat. (B) Spectrograms of three example notes from one individual. Frequency range: 0 to 125 kHz. (C) Spectrogram of a full S. teguina advertisement song. The colored arrows denote the onset time of the three corresponding notes from (B). (D) Trajectory plot in which individual note durations are displayed as a function of their onset times in the song, with colored circles indicating notes from (B). (E) Trajectories from one male S. teguina in different social contexts (n = 15 songs per condition). The vocal stereotypy exhibited during isolated singing (top and bottom) is significantly altered during social interaction (middle). Individual dots represent the duration of each displayed song, and the histogram quantifies the durations for all songs produced in a given context (day 1 alone: n = 57 songs; day 2 social: n = 388 songs; day 3 alone: n = 50 songs). The red line is the same trajectory plotted in (D). (F and G) The number of songs per hour (F) and the song duration variability (G), defined as the standard deviation of the song duration distributions, significantly increase during the social context (n = 8 animals). Red lines represent the example mouse from (E). Asterisks indicate a significant difference between conditions (*P < 0.01, Wilcoxon signed rank test; n.s., not significant).

We next examined whether the acoustic characteristics of S. teguina vocalizations are modulated by social context, as observed in other taxa (32). To investigate this, we staged a social encounter by relocating a male subject (a “recruit”) into a testing room occupied for at least 1 week by another male (a “resident”). The two mice were held in adjacent chambers with acoustic but not visual access to each other. In this configuration, recruit males altered their singing in two ways. First, recruits vocalized four times as often in the social context [social (day 2): 20.4 ± 4.8 songs/hour; mean ± SEM unless stated otherwise] as in isolation [alone (day 1): 4.7 ± 0.8 songs/hour, alone (day 3): 4.4 ± 0.7 songs/hour] (Fig. 1, E and F). Second, the variability of song trajectory plots increased significantly when recruits could hear the resident mouse (Fig. 1E). Consistent with this observation, we found that song duration variability was higher in the social context [social (day 2): 2.7 ± 0.3 s] than in isolation [alone (day 1): 1.5 ± 0.1 s, alone (day 3): 1.4 ± 0.2 s] (Fig. 1G).

To examine the fine structure of vocal interactions between male S. teguina, we simultaneously recorded the songs of both the resident and recruit mice in the social condition. We found extensive temporal coordination of singing behavior within vocal pairs (Fig. 2 and movie S2). Whereas exchanges could be initiated by either male, they typically ended with a recruit’s song (Fig. 2, A to E). Surprisingly, this asymmetry was observed across all recruit-resident pairs (n = 8) and was preserved for the entire ~24-hour social session (89 ± 10 interaction bouts per pair) (Fig. 2, B and E). For the remainder of this study, we restricted our analysis to the songs of the recruit mice to focus on this sensory-evoked vocal response. By aligning the interaction bouts to the songs of the resident mouse, we found that the recruit mouse precisely times his vocal onset to coincide with the end of the resident’s songs (Fig. 2B, left). This observation was robust across all pairs (average response latency = 0.81 ± 0.18 s) (Fig. 2C; n = 8). To estimate the amount of countersinging that one would expect by chance given the amount of singing observed in the social condition, we shuffled the song times and quantified the likelihood of such “spurious countersinging” to be nearly an order of magnitude less (Response probabilityData = 0.69 ± 0.09, Response probabilityShuffled = 0.07 ± 0.02, P < 0.01, Wilcoxon signed rank test). The recruit’s response probability distributions were significantly sharper when interaction bouts were aligned to the end of the resident’s songs rather than the start (jitterend-aligned = 2.94 ± 0.64 s, jitterstart-aligned = 5.19 ± 0.43 s, P < 0.05, Wilcoxon signed rank test; fig. S1), suggesting that the recruit mouse uses the end of the resident’s song as a sensory trigger. Additionally, the recruit mouse often stopped vocalizing immediately after the resident mouse started singing (Fig. 2B, right, and 2D). Thus, recruit males were capable of actively timing their vocalization onsets and offsets to avoid acoustic overlap with the resident (Fig. 2E), giving rise to turn-taking dynamics similar to those observed during human conversation (4). Furthermore, the recruit’s response precision correlated with the degree of social engagement, as quantified by countersinging probability (Fig. 2F) and the increase in song duration variability across social contexts (Fig. 2G), suggesting that active participation in an orderly vocal exchange contributed to these changes (Fig. 1). This result is consistent with recent findings demonstrating that context can influence the timing of vocal turn-taking in other species (33, 34).

Fig. 2 Temporal coordination of vocal interactions between conspecific males.

(A) One hour of continuous audio recordings from two interacting males. Two typical interactions are shown in detail: one initiated by resident mouse (black) and another by a recruit (red). (B) All vocal interactions (n = 101 interactions for this example pair) over a 24-hour period aligned to either the end (left) or the beginning (right) of the resident’s songs. The corresponding start and stop probability distributions for the recruit’s song are plotted below. (C) Summary of mean start latencies across all pairs (n = 8). For each, the circle represents the mean latency of the recruit mouse’s song with respect to the offset of the resident’s song, with horizontal line indicating song initiation jitter (full-width at half maximum of the probability distribution). (D) Mean stop latencies across all pairs with respect to onset of the resident’s song. (E) Probability of song occurrence at any given time point aligned to the end of the resident mouse’s song for the pair featured in (A) (top) as well as for all pairs (bottom), showing active avoidance of song overlap between conspecifics. In the bottom plot, dashed lines represent the SEM. (F and G) Song initiation jitter is negatively correlated with countersinging probability (F) (Pearson’s correlation, r = −0.78, P < 0.05) as well as the degree of song duration variability change from the alone condition (G) (Pearson’s correlation, r = −0.79, P < 0.05). Each dot represents the behavior of a single recruit mouse.

We next sought to explore the neural mechanisms contributing to countersinging. As a first step, we characterized the biomechanics of song production by examining the motor elements that make up a song. Singing resulted in a rapid cycle of inhalation and exhalation (fig. S2), a stark contrast from laboratory mice whose vocalizations are strongly coupled to ongoing sniffing activity (35). In singing mice, phonation is coupled to exhalation and jaw movements; electromyography (EMG) confirmed that individual vocalizations were produced during the exhalation phase and were preceded by robust flexion of the jaw muscle (digastricus) (Fig. 3A). The correlation between song production and jaw movement—similar to that previously observed in rats (36)—allowed us to use EMG activity to probe the relationship between specific brain centers and song-related musculature. In previous studies, stimulation of motor cortical centers in primates resulted in vocal fold adduction (37), suggesting a possible involvement of the motor cortex in vocalization. We used intracortical microstimulation (ICMS) over a large portion of the anterior cortex to identify areas leading to flexion of song-related musculature. The minimum current that reliably elicited a fixed EMG activity threshold (Fig. 3, B and C, and fig. S3) was used to define a functional hotspot that maps to the anterolateral aspect of the motor cortex (Fig. 3C, right), which corresponds to the orofacial motor cortex in Mus musculus (38). We therefore refer to this region as the orofacial motor cortex (OMC).

Fig. 3 Hierarchical control of song timing.

(A) Electromyograph from the digastricus muscle and simultaneous raw audio (log amplitude) of one advertisement song, showing increased muscle activity immediately before vocalization of individual notes. The inset shows three notes (marked by an asterisk) and accompanying EMG activity in greater detail. (B) ICMS of two different loci elicits short-latency EMG activity. The simulation artifact (four parallel lines) is truncated for clarity. (C) The minimum amount of current needed to elicit a significant (statistical significance, as defined in the methods) EMG response (threshold current) from each ICMS site is color coded for one example mouse (left) and across the population (right, n = 5 mice), revealing a “hotspot” on the anterolateral portion of the motor cortex, henceforth referred to as the orofacial motor cortex (OMC, right). The ICMS locations for examples in (B) are indicated by cyan and green squares. A, anterior; P, posterior; M, medial; L, lateral. (D) Example spectrograms from one individual in which song was truncated (top) or paused (bottom) in different trials by a 200-μA electrical stimulation of the OMC. Yellow lines indicate the onset and offset of electrical stimulation. (E) Two possible outcomes for song resumption after a brief electrical stimulation–induced pause. (F) Full trajectory of an example song before and after electrical stimulation–induced pause, with dots indicating the duration of each note. The inset is an expanded view of the peristimulation period. Dashed black lines are the estimated note duration slopes. Expected trajectories under outcomes 1 and 2 are depicted as red and green lines, respectively. The gray line indicates the actual change in note duration after song resumption. (G) Summary data for all paused songs in each animal. The majority of trajectories are consistent with outcome 2 (n = 12 of 14, 14 of 15, 18 of 18, and 14 of 14 trials). (H) Proposed effects of OMC cooling on song trajectory if OMC activity does not affect song timing (left), if OMC exclusively controls song timing (middle), or if OMC and subcortical structures share this control (right). (I) Spectrograms and trajectories of example songs during baseline and cooling sessions. Cooling of the OMC lengthens song durations by decreasing the rate of change of note duration (slope) during song. (J and K) Summary for all songs during the control (n = 27 songs) and cooling periods (−3°C: n = 10 songs, −6°C: n = 32 songs) for mouse C4 (J), as well as the mean ± SEM values of the entire population (K) (n = 10 animals). Arrows in (J) denote the mean values of each distribution. Cooling resulted in a decrease in the slope of the song trajectory (left) and an increase in the time needed to reach a threshold note length of 75 ms (middle) without changing the duration of individual notes (right). Asterisks indicate a significant difference between conditions (*P < 0.01, Wilcoxon signed rank test).

What is the functional role, if any, of the OMC on song production? Although OMC stimulation can elicit electrical activity in song-relevant musculature, this does not necessarily imply that the OMC can influence song production. To address this directly, we carried out a series of perturbations during singing in the alone condition, beginning with bilateral electrical stimulation of the OMC. Strong stimulation resulted in song truncation, whereas milder stimulation (200 to 500 μA) often produced brief pauses (range: 638 to 1448 ms), with songs resuming once stimulation ended (Fig. 3D). The precise stereotypy of alone S. teguina songs (Fig. 1E) provides an ideal opportunity to distinguish between two possible experimental outcomes. First, the song could resume at the expected point in the sequence, accounting for the time delay (outcome 1; Fig. 3E), consistent with the hypothesis that the vocal patterning is primarily driven by a pathway independent of the OMC. In nonhuman primates, for example, there is a vocal motor stream that begins in the cingulate cortex and acts via the periaqueductal gray (39). An alternative outcome of our experiment is that the song could resume at the same point in the motor sequence where it had paused (outcome 2; Fig. 3E), suggesting that the pathway leading from the OMC to vocal musculature is capable of sculpting the structure of song. For every trial, we used the 10 notes preceding the perturbation to estimate the note durations that would be expected in an uninterrupted song. We then compared the actual note duration with these predicted values and found that song typically resumes at the same point in the sequence where it had paused (Fig. 3F, outcome 2). Across the population, note durations after song resumption were significantly more similar to outcome 2 than outcome 1 in 58 out of 61 trials across four animals (Fig. 3G). These results refute the hypothesis that an OMC-independent pathway shapes song patterning in S. teguina.

Although our stimulation results functionally connect the OMC to the behavioral output, they do not elucidate the nature of this interaction. Previous reports suggest that most mammalian vocal communication does not involve the motor cortex and that subcortical structures are sufficient for this behavior (39, 40). To isolate the contributions of local neuronal dynamics in the OMC from those of downstream structures, we used mild focal cooling of the OMC during song production. Manipulating neural circuits with temperature has emerged as a useful experimental tool for maintaining behaviorally relevant activity while selectively slowing these dynamics (4145). We predict three possible outcomes of this manipulation. If song timing is exclusively governed by subcortical structures, as expected in standard rodent models (35), then the control and cooled song trajectories should completely overlap (Fig. 3H, model 1). Alternately, if OMC dynamics exclusively dictate the temporal structure of song, then cooling should lead to the dilation of vocal behavior on all time scales (i.e., note duration and song length) (Fig. 3H, model 2), as evident in both birdsong (41) and human speech (42). If motor control of the song is shared between the OMC and subcortical regions, then cooling may alter some temporal properties while preserving others (Fig. 3H, model 3). One possibility is that cooling may change the slope of the song trajectory, a parameter we observe to be socially modulated (Fig. 1). To test these models, we used a custom-built Peltier device capable of rapidly and reversibly cooling the OMC (fig. S4). Cooling strongly affected song timing by monotonically increasing the overall song duration (Fig. 3I). In contrast, cooling did not affect running speed (fig. S5), a behavior unlikely to require substantial cortical involvement (46, 47). OMC cooling resulted in a shallower song trajectory that took longer to unfold (Fig. 3, I to K; n = 10 animals). We found that cooling decreases the slope of the song trajectory (Slopecontrol = 0.013 ± 0.001; Slopecooling = 0.009 ± 0.001, P < 0.002, Wilcoxon signed rank test) as well as the time for the song trajectory to surpass an arbitrary threshold (Thresholdcontrol = 4.23 ± 0.16 s; Thresholdcooling = 4.88 ± 0.2 s, P < 0.002, Wilcoxon signed rank test). These changes demonstrate that the OMC contributes significantly to song patterning, thereby ruling out model 1 (Fig. 3H). In addition, a closer examination of song acoustic structure revealed that the distribution of individual note durations did not change with cooling (Lengthcontrol = 68.1 ± 1.5 ms; Lengthcooling = 67.8 ± 1.5 ms, P = 0.92, Wilcoxon signed rank test), which is inconsistent with the model that the OMC solely determines all aspects of song timing (Fig. 3H, model 2). Instead, we find that the OMC shapes song progression without influencing the structure of individual notes. Neither the starting nor the ending note durations change as the result of cooling, but it takes longer for this progression to occur, which is accomplished by increasing the total number of notes produced (Note numbercontrol = 44.9; Note numbercooling = 48.9, P < 0.01, Wilcoxon signed rank test). Therefore, these data suggest a hierarchy of motor timing control (Fig. 3H, model 3), with the OMC being capable of exerting moment-by-moment control over the pacing of a subcortically generated song sequence.

In our initial experiments, we observed that social interaction profoundly changed song progression (Fig. 1E) and that this song variability was driven by the degree of social engagement during vocal interactions (Fig. 2, F and G). Our stimulation and cooling studies suggest that the OMC is well positioned to contribute to such social coordination by altering song structure. Therefore, we proceeded to test this prediction by reversibly inhibiting the OMC with muscimol (a GABAA agonist). Preliminary injections of a high muscimol dose (100 mM, 100 nl) in the motor cortex led to gross movement abnormalities as well as the complete abolishment of singing behavior for an extended period of time (>4 hours). Such nonspecific motor deficits were not evident when we lowered this concentration to 10 mM (100 nl), a common dosage (46, 48, 49). Each animal was tested for both experimental conditions: muscimol (OMC inactivation) and saline (control; n = 6 mice) (Fig. 4, A and B). In both conditions, we found that five of six individuals produced spontaneous songs and that the rate of spontaneous singing was not significantly influenced by this manipulation [Fig. 4C (bar graphs); control, 4.1 ± 1.51 songs/hour; muscimol, 2.1 ± 1.2 songs/hour, P = 0.31, Wilcoxon signed rank test].

Fig. 4 The OMC is required for countersinging.

(A) Countersinging response to audio playback of a conspecific male song. (B) Design of experimental paradigm. (C) Song raster plot of all trials with either saline (top) or muscimol (bottom); each row denotes a different session. Ticks represent playback from a loudspeaker, and colored dots represent S. teguina songs. Bar plots indicate the total number of spontaneous songs per hour during the pretesting (nonsocial) period for each animal. (D and E) Compared with saline (control) injections, dosing with muscimol eliminates a precise countersinging response [top and middle: mouse M29; bottom: entire population (n = 5 mice, mean ± SEM)]. Mouse M28 was excluded because he did not countersing in either condition. (F) Countersinging response probability significantly decreases with muscimol treatment compared with saline dosing (*P < 0.05 for each animal, n = 5 mice, binomial test). Dashed lines represent cases where the muscimol session preceded the saline session. (G) In saline-injected animals, the total number of songs per hour significantly increases during the playback period compared with the pretesting alone period (*P < 0.05, n = 6 mice, Kruskal-Wallis test) and the playback period after muscimol inactivation (*P < 0.05, n = 6 mice, Kruskal-Wallis test). This increase of song rate during the playback condition was absent upon OMC inactivation with muscimol (P = 0.8068, n = 6 mice, Kruskal-Wallis test). Gray circles represent individual animals; black circles denote mean and SEM. (H) Song durations during the playback period are significantly higher for saline-dosed mice compared with muscimol-injected animals (saline: 8.55 ± 0.25 s; muscimol: 5.1 ± 0.41 s, *P < 0.00001, Wilcoxon rank sum test).

We next used playback to evaluate whether the OMC mediated social influences on singing behavior. In control (saline-injected) animals, song playback led to an increase in the amount of singing as well as song duration variability, as expected in a social countersinging context (Fig. 4, C and G; P < 0.05, Kruskal-Wallis test). In contrast, muscimol-injected animals did not sing more songs in response to playback (Fig. 4, C and G; P = 0.81, Kruskal-Wallis test), suggesting that the OMC affects context-dependent modulation of song rate, a phenomenon we have observed during natural social encounters (Fig. 1F). Similarly, the probability of eliciting a countersinging response was significantly greater in the control condition than in the OMC-inactivated condition for each mouse (Fig. 4, C to E; P < 0.05, binomial test) as well as across the population (Fig. 4F; n = 5, saline: 0.59 ± 0.13; muscimol: 0.09 ± 0.05; P < 0.05, one-sided Wilcoxon signed rank test). Using a permutation test, we found that this difference in response probability could not be explained by our observed changes in song rate across conditions (fig. S6). Moreover, in cases where residual singing behavior remained after muscimol injection to the OMC (Fig. 4C), we observed an increase in mean response latency of 2.2 ± 0.9 s relative to that of saline-injected controls (Fig. 4, D and E; P < 0.05, Wilcoxon rank sum test). These data demonstrate that the OMC is critical for rapid vocal responses to playback; such responses must be driven by sensorimotor coupling rather than by more general changes in motivation.

In this study, we examined vocal interactions between pairs of S. teguina to test a range of hypotheses concerning the neural mechanisms underlying complex sensorimotor interactions. Using four complementary lines of evidence (intracortical microstimulation, stimulation-induced vocal arrest, focal cooling, and pharmacological inactivation), we define a region of the motor cortex (the OMC) that influences vocalization and mediates rapid vocal interactions. Whereas previous studies have used immediate early genes or electrophysiological approaches to suggest cortical involvement in nonhuman primate communication (5054), our study represents the first direct demonstration of cortical dependence of precise vocal interactions in a mammal. Specifically, we have shown that the motor cortex is required for adaptive countersinging but not for song production itself. Additionally, our cooling results demonstrate that the motor cortex is capable of dynamically adjusting the pacing and duration of song sequences, consistent with the changes in these same parameters during social interactions. This finding provides evidence for recent proposals that the motor cortex informs subcortical structures to appropriately respond to unexpected sensory stimuli (47, 55) and is consistent with the idea that cortical control may be required for volitional vocal production in primates (56). In S. teguina, this executive role of the motor cortex may be bolstered by integrating information from other regions, potentially related to factors such as past history and social status. Future studies in which neural activity is monitored during countersinging will help to further refine our understanding of OMC’s contribution to this behavior.

The hierarchical control mechanism that appears to underlie countersinging in S. teguina features functionally distinct regions responsible for vocal production and coordination. By segregating the vocal motor pathway from cortical control, the structure of the individual notes remains tightly constrained, thus conveying context-invariant information, perhaps related to individual identity (57). A similar organizing principle appears in other taxa as well (6, 58, 59). For instance, cricket stridulation is controlled by a command neuron upstream from central pattern generators (60). In songbirds, specific pallial regions are necessary for precise vocal timing of innate calls that are likely to originate subcortically (61). These examples of hierarchical control across the animal kingdom suggest a common algorithm that may mediate a wide variety of social interactions.

There has been a recent emphasis on understanding brain function through the lens of complex, ethologically relevant behaviors (62, 63). Here we present S. teguina as a new rodent model for investigating neural mechanisms underlying vocal communication with a socially modulated, tractable, and cortically dependent behavior. Moreover, countersinging itself can be temporally segregated into distinct sensory and motor epochs (Fig. 2 and movie S2). Such segregation offers an enormous experimental advantage by recapitulating the organization of task structure typically engineered into standard laboratory sensorimotor paradigms (13) and will allow for the incorporation, testing, and extension of existing hypotheses for analogous brain regions.

Supplementary Materials

Materials and Methods

Figs. S1 to S6


Movies S1 and S2

References and Notes

Acknowledgments: We thank D. Aronov, J. Sakata, C. Ebbersen, S. Ghosh, C. Desplan, and members of the Long laboratory for comments on earlier versions of this manuscript. We thank N. Leal for help with recording the movies. We also thank the individuals involved in caring for and overseeing the NYULMC S. teguina colony, especially C. Jackson. Funding: This research was supported by the New York Stem Cell Foundation (M.A.L.), NIH grant T32GM007308 (D.E.O.), the Simons Foundation Society of Fellows (A.B.), and the Simons Collaboration on the Global Brain (M.A.L.). S.M.P.’s efforts were funded in part by the NSF (IOS 1457350). Author contributions: M.A.L., S.M.P., A.B., and D.E.O. conceived the study and designed the experiments; D.E.O. and A.B. conducted the research; A.B. and D.E.O. performed data analyses; A.B, D.E.O., and M.A.L. created the visualizations; A.M.M.M. collected preliminary data for the project; A.B. wrote the initial draft of the manuscript; M.A.L, A.B., S.M.P., D.E.O., and A.M.M.M. edited and reviewed the final manuscript; S.M.P. provided the starter S. teguina mice for establishing the NYU colony; M.A.L. acquired funding; and M.A.L. and S.M.P. supervised the project. Competing interests: None declared. Data and materials availability: For research and educational purposes, we will share all data collected as part of this project at

Stay Connected to Science

Navigate This Article