Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Jul 19;37(29):7036-7047.
doi: 10.1523/JNEUROSCI.0702-17.2017. Epub 2017 Jun 19.

Social Context-Dependent Activity in Marmoset Frontal Cortex Populations During Natural Conversations

Affiliations
Free PMC article

Social Context-Dependent Activity in Marmoset Frontal Cortex Populations During Natural Conversations

Samuel U Nummela et al. J Neurosci. .
Free PMC article

Abstract

Communication is an inherently interactive process that weaves together the fabric of both human and nonhuman primate societies. To investigate the properties of the primate brain during active social signaling, we recorded the responses of frontal cortex neurons as freely moving marmosets engaged in conversational exchanges with a visually occluded virtual marmoset. We found that small changes in firing rate (∼1 Hz) occurred across a broadly distributed population of frontal cortex neurons when marmosets heard a conspecific vocalization, and that these changes corresponded to subjects' likelihood of producing or withholding a vocal reply. Although the contributions of individual neurons were relatively small, large populations of neurons were able to clearly distinguish between these social contexts. Most significantly, this social context-dependent change in firing rate was evident even before subjects heard the vocalization, indicating that the probability of a conversational exchange was determined by the state of the frontal cortex at the time a vocalization was heard, and not by a decision driven by acoustic characteristics of the vocalization. We found that changes in neural activity scaled with the length of the conversation, with greater changes in firing rate evident for longer conversations. These data reveal specific and important facets of this neural activity that constrain its possible roles in active social signaling, and we hypothesize that the close coupling between frontal cortex activity and this natural, active primate social-signaling behavior facilitates social-monitoring mechanisms critical to conversational exchanges.SIGNIFICANCE STATEMENT We provide evidence for a novel pattern of neural activity in the frontal cortex of freely moving, naturally behaving, marmoset monkeys that may facilitate natural primate conversations. We discovered small (∼1 Hz), but reliable, changes in neural activity that occurred before marmosets even heard a conspecific vocalization that, as a population, almost perfectly predicted whether subjects would produce a vocalization in response. The change in the state of the frontal cortex persisted throughout the conversation and its magnitude scaled linearly with the length of the interaction. We hypothesize that this social context-dependent change in frontal cortex activity is supported by several mechanisms, such as social arousal and attention, and facilitates social monitoring critical for vocal coordination characteristic of human and nonhuman primate conversations.

Keywords: communication; marmoset; natural behavior; social; vocalization.

Figures

Figure 1.
Figure 1.
Antiphonal conversations in marmosets. Spectrograms of antiphonal and independent phee calls. Top, the virtual marmoset phee stimuli broadcast to the marmoset subject. Bottom, Phee calls from the live marmoset Subject M. The first virtual marmoset phee call is an independent stimulus, characterized by the absence of response from M within the antiphonal period of 10 s as denoted by a gray dashed line. The next two calls from the virtual marmoset are antiphonal stimuli, characterized by phee responses from M within the antiphonal period. The final call from the virtual marmoset is independent, with no vocal response from M within 10 s.
Figure 2.
Figure 2.
Frontal cortical activity separates vocalization social contexts. A, Sample frontal cortical population responses simulated from test datasets plotted in the first and second PCs from training simulations by phee social context (top), randomly shuffled contexts (middle), and phee-stimulus lengths (bottom). The population responses to phee social contexts form distinct clusters for antiphonal and independent contexts, but this is not the case for randomly shuffled contexts, or to stimuli separated by phee length. B, ROC analysis measures the separation of population responses by social context (top), randomly shuffled contexts (middle), and stimulus length (bottom), in the first three PCs from independent training datasets. An area under the ROC curve of 0.5 indicates stimulus categories are not separable, and 1 indicates they are completely separable. Population responses to independent and antiphonal phee calls are highly separable in PC1, and remain significantly separable in the PC2. The remaining PCs show no separation. Population responses to randomly shuffled contexts (middle) or to different phee stimulus lengths (bottom) are not separable. Error bars are 95% CIs. C, The median coefficients of PC1 from all 500 training simulations of population responses to antiphonal and independent stimuli. Coefficients are organized by unit (columns) and time period (rows). Units are sorted by the sum of the coefficient magnitudes over all time periods, such that recording 1 contributes the least to PC1 and recording 258 contributes the most. D, Mean and 95% CIs for median PC1 coefficient contributions from each time period, calculated by summing the coefficient magnitudes across all units. Error bars are 95% CIs.
Figure 3.
Figure 3.
Social context classification from PC1 emerges from the population activity. A, Histograms of individual unit classifier accuracies and the distribution of accuracies of the population classifier performed on all 500 population response simulations. B, The distribution of average pairwise correlation coefficients over all units estimated under two conditions: with responses to each phee stimulus maintained across all units within a session (normal) and with responses to phee stimuli shuffled within each context across units (shuffled). C, The change in classifier accuracy for each behavioral session with responses to each stimulus maintained across all units in that session (normal) and with responses to each stimulus shuffled, within social context, across units (shuffled). D, Accuracy of the population classifier is much greater than chance even when predicting stimulus context from population activity at only one time period relative to that stimulus. Error bars are 95% CIs.
Figure 4.
Figure 4.
Differences in unit activity between vocalization social contexts. A, Sample raster (1 ms resolution; top) and normalized spike rates (0.5 s time bins; bottom) for a unit with large positive PC1 coefficients. In the raster plot, red lines indicate independent phee stimuli, blue lines indicate antiphonal phee stimuli, and brown lines mark subject replies (when within the axis limits). Binned, normalized, firing rates are shown below, with blue points for antiphonal stimuli and red points for independent stimuli. Gray rectangles indicate the mean phee pulse times. Error bars are 95% CIs. B, Sample raster with same conventions as A, except for a recording with preference for independent stimuli, which had large negative PC1 coefficients. C, Mean firing rates of all 172 single units in response to antiphonal stimuli compared with independent stimuli. Firing rates were averaged over all four time periods and plotted on a logarithmic scale. D, Mean firing rates of all 86 multiunits using the same conventions as C.
Figure 5.
Figure 5.
Frontal cortical activity distinguishes between vocalization social contexts. A, The context preference index, given by the difference in mean normalized firing rate between preferred and nonpreferred stimuli contexts, is plotted for each unit. Blue points indicate preference for the antiphonal context and red points for the independent context. Error bars are one-tailed 95% CIs. The inset provides the percentage of units preferring each social context, indicated by color, for the units that reached significance (Sig) and for all units (All). The example units from Figure 4A,B have respectively colored error bars. B, Mean normalized firing rates over all units for the preferred phee context (black) and nonpreferred phee context (white), in 0.5 s time bins. Error bars are 95% CIs. Mean phee-stimulus pulse timings are indicated by gray rectangles. C, D, Same conventions as A and B except context preference index was calculated when phee contexts were randomly assigned by shuffling context identity. E, F, Same conventions as A and B except context preference index was calculated based on length of the phee stimulus instead of its social context. In C and E, unit positions were kept the same as in A, with colors corresponding to the phee context preferences of each unit in A.
Figure 6.
Figure 6.
Discrimination of social contexts by location of electrode arrays in the frontal cortex. The anatomical layout of the four electrode arrays are shown. B01 and B02 are arrays from Subject B, and F01 and R01 are from Subjects F and R. Each electrode is colored according to its context preference index. Channels with multiple units only show the highest value; channels with no units are light gray with no border.
Figure 7.
Figure 7.
Population responses during conversations A, A schematic of an antiphonal conversation between a virtual monkey (VM) and marmoset subject (M) showing a bout of three independent stimuli (red), followed by a conversational bout of three antiphonal stimuli (blue) with subject replies (black), and then another independent bout. Below, Neural population responses to stimuli within the conversation. Responses to phee stimuli at the end and start of independent bouts (red) are greater than responses to stimuli at the start, middle, and end of an antiphonal conversation (blue). *p < 0.05 differences for antiphonal bouts compared with independent bouts. B, Population responses to antiphonal (blue) and independent (red) stimuli are compared with responses during antiphonal and independent bouts of specified lengths. Below each comparison n is the number of units with data for the bout length. *p < 0.03 for differences between independent and antiphonal contexts. All error bars are 95% CIs.

Similar articles

See all similar articles

Cited by 10 articles

See all "Cited by" articles

Publication types

LinkOut - more resources

Feedback