Synchronized LFP rhythmicity in the social brain reflects the context of social encounters

The study addresses how mammalian brains modulate social behavior according to context, a key but poorly understood question in social neuroscience. Social interactions are complex and depend on partner identity and contextual internal states (arousal, motivation, emotion, homeostasis). Prior work has outlined a distributed "social brain" network comprising cortical, striatal, hippocampal, septal, amygdaloid, and hypothalamic regions that can support diverse, even opposing, social behaviors. Because many of these regions are bidirectionally interconnected and implicated across different social functions, system-level mechanisms are needed to explain flexible, context-dependent behavior. Oscillatory neural activity, especially theta (4–12 Hz) and gamma (30–80 Hz), increases with cognitive demands, internal states (e.g., fear, anxiety, attention), and social communication, and abnormalities are reported in neurodevelopmental disorders. A prominent hypothesis posits that coherence of theta/gamma rhythms dynamically binds distributed ensembles into functional subnetworks for specific tasks. The authors therefore hypothesized that coherent theta/gamma rhythms couple social brain regions into context-dependent functional subnetworks, such that distinct social contexts elicit distinct patterns of coordinated rhythmic activity supporting context-specific processing and behavior. They tested this by recording LFPs simultaneously from multiple social brain regions in mice during three social discrimination tasks that dissociate stimulus identity, valence (preference), and context.

The paper situates its work within several lines of research: (1) System-level encoding of social behavior: multi-site recordings in limbic networks predict social preferences, intentions, and decisions better than single-region activity. (2) Oscillations and internal states: theta/gamma power increases with learning, social communication, and internal states including fear, anxiety, and attention; aberrant rhythms occur in disorders such as ASD. (3) Communication-through-coherence framework: coherent oscillations coordinate inter-regional communication and bind distributed neuronal ensembles during cognitive operations. (4) Social brain connectivity: extensive bidirectional connections among mPFC, hippocampus (including ventral DG/CA1), amygdala, septum, accumbens, and hypothalamus enable flexible behavior. (5) Ventral dentate gyrus (vDG) and context: prior work links DG to contextual novelty and pattern separation, suggesting a potential role in context-dependent social processing. This literature collectively motivates examining theta/gamma power and coherence across a social brain network as mechanisms for context-sensitive social behavior.

Subjects and tasks: Adult male CD1 mice performed three 10-min sessions per task (baseline 5 min; encounter 5 min) in a common arena with two triangular chambers: (1) Social Preference (SP): social (group-housed male) vs object (Lego). (2) Emotional-state Preference (EsP): isolated male (7–14 days) vs group-housed male. (3) Sex Preference (SxP): female vs group-housed male. Behavioral tracking quantified investigation bouts, transitions, distance, speed, and a relative discrimination index (RDI). Only investigation bouts ≥2 s were analyzed. Electrophysiology: Custom 16-wire tungsten modular electrode arrays (EAr) targeted up to 18 regions across neocortex (PrL, IL, Pir), striatum (AcbC, AcbSh, VP), ventral hippocampal formation (vDG, vCA1), lateral septum (LS), amygdala (BLA, MeAD, CeA, EA/AhiAL), and hypothalamus (DMD, PVN, PLH). Electrode sites were verified post-mortem. Signals sampled at 20 kHz (downsampled to 5 kHz), low-pass 300 Hz; spectrograms computed with DPSS 2 s windows, 50% overlap, 0.5 s time bins. Power analysis: Theta (4–12 Hz) and gamma (30–80 Hz) band powers were averaged for baseline and encounter; ΔθP and ΔγP defined as encounter minus baseline. For bout-aligned power (Z-score), the 5 s pre-bout period served as baseline; only bouts ≥2 s considered. Coherence analysis: Magnitude-squared coherence (mscohere, Welch’s method) for all region pairs recorded in ≥5 sessions and ≥2 subjects per task (99 pairs). Baseline coherence and encounter changes were quantified; encounter change normalized as (Co_enc − Co_base)/(Co_enc + Co_base). Bout-specific ΔθCo and ΔγCo computed by subtracting baseline chamber investigation coherence from encounter bout coherence and averaging across bouts/sessions. Granger causality (GC): Pairwise conditional multivariate GC (MVGC toolbox) computed separately for baseline and encounter. LFP downsampled to 500 Hz; median VAR model order 38 (BIC); autocovariance acmaxlags 1500. GC integrated within theta or gamma bands; encounter vs baseline GC changes examined among regions over-represented among strong coherence-bias pairs (|mean±1.5 SD|). Behavioral-event alignment: Theta/gamma power in candidate hub regions was aligned to specific events (start/end of bouts to each stimulus; transitions between stimuli vs repeated investigations) with Z-score normalization relative to the 5 s pre-event baseline. Decoding analysis: Multi-class Random Forest (TreeBagger, 80 trees) with leave-one-mouse-out cross-validation. Features: ΔθCo or ΔγCo across pairs, or combined with ΔθP. Data normalization per mouse; bouts averaged per stimulus per session; missing pair entries imputed via a MICE-like linear regression approach. Confusion matrices aggregated across iterations; significance via Mann–Whitney tests with FDR correction. Statistics: Normality tests (Kolmogorov–Smirnov, Shapiro–Wilk). Paired/unpaired t-tests or Wilcoxon/Mann–Whitney; ANOVA/Welch’s ANOVA/Kruskal–Wallis with appropriate post-hoc (Šídák, Dunnett’s T3, Dunn’s); two-way or mixed-models ANOVA; Pearson/Spearman correlations; binomial tests; FDR correction throughout.

- Behavior: Across tasks, total investigation time, transitions, and distance traveled did not differ (Kruskal–Wallis: time H=1.702, P=0.427; transitions H=2.133, P=0.3442; distance H=0.6782, P=0.776). Preference (RDI) was lower in EsP vs SP (H=8.509, P=0.0142; Dunn’s post-hoc *P<0.03). Each task showed significant preference for one stimulus (e.g., SP social vs object: Wilcoxon W=495, P<0.0001; EsP isolated vs grouped: paired t(27)=2.374, P=0.025; SxP female vs male: paired t(25)=5.75, P<0.0001). - Session-wide LFP power: Baseline theta/gamma powers did not differ across tasks. Encounter-induced ΔθP was largest in SP (Welch’s ANOVA W(2,31.80)=14.67, P<0.0001; SP>EsP P=0.0018; SP>SxP P=0.0002). ΔγP also higher in SP (W(2,33.65)=5.134, P=0.0113; SP>SxP P=0.0127). No SP session-order effect for ΔθP (paired t(13)=1.722, P=0.1087) or ΔγP (t(13)=1.577, P=0.1388). Theta and gamma ΔP across regions were correlated in SP (r=0.72, P=0.0007) and SxP (r=0.50, P=0.0335); trend in EsP (r=0.45, P=0.059). - Bout-level power: During investigation bouts (≥2 s), theta and gamma modulations dissociated. Example EA showed gamma Z-score increases specific to certain stimuli per task with significant differences (SP: Wilcoxon, ***P=0.002; EsP: paired t, ****P<0.0001; SxP: paired t, P=0.001). Across regions, ΔθP bias correlated with behavioral preference (RDI) in a task-specific manner: negative correlations in EA and LS during SP (e.g., EA r=−0.66, P<0.05; LS r=−0.59, P<0.05) and positive correlations in AcbSh, AhiAL, VP, DMD during EsP/SxP (e.g., VP EsP r=0.84, P<0.05; VP SxP r=0.93, P<0.001). ΔγP showed broad bias toward grouped/male (less-preferred in EsP/SxP), while ΔθP biases differed by task; vDG uniquely biased toward object/isolated/female across tasks for both ΔθP and ΔγP. - Session-wide coherence: Baseline theta/gamma coherence did not differ across tasks. Encounter-induced ΔθCo differed (Kruskal–Wallis H=47.5, P<0.0001; SP>EsP Z=4.723, P<0.0001; SP>SxP Z=6.709, P<0.0001). ΔγCo also differed (H=23.58, P<0.0001; SP>EsP P<0.0001; SP>SxP P<0.0001). Patterns of ΔθCo across pairs correlated strongly between EsP and SxP, but not between SP and SxP; all ΔγCo correlations across tasks were significant. - Bout-level coherence reflects context: Pairs showing strong ΔθCo bias during bouts did not overlap by stimulus type or valence, but many pairs overlapped within the same task, indicating context dependence. Across all pairs, ΔθCo (and ΔγCo) correlations were strongest for stimulus pairs sharing the same context; correlations were weaker or absent for shared type or valence. A Random Forest classifier predicted social context above chance using ΔθCo (only SxP significant; other contexts near chance), while identity classification across six stimuli performed poorly. Adding ΔθP improved context decoding (notably SP and SxP). - Granger causality and hubs: Regions over-represented among strong coherence-bias pairs showed distinct, context-specific encounter vs baseline GC changes. vDG and AcbC exhibited significant GC changes in both theta and gamma across all tasks; PrL and AhiAL in theta across all tasks. Directional analysis (FDR-corrected) identified a significant directional asymmetry only for vDG→LS in gamma during EsP, supporting vDG as a coordinating hub. - Event-related vDG modulation: vDG power differed by event and context. In SP, theta and gamma power were significantly reduced at the end of social vs object bouts (**P<0.01). In EsP, transitions from isolated→grouped showed decreased vDG power vs repeated grouped investigations (*P<0.05). In SP, transitions social→object showed increased vDG theta power vs repeated object (*P<0.05).

The findings support the central hypothesis that coherent theta/gamma rhythms organize social brain regions into functional subnetworks in a context-dependent manner. Although social encounters evoke a global internal state that elevates theta/gamma power across regions with similar temporal dynamics, the pattern and magnitude of inter-regional coherence changes, particularly in theta, depend strongly on social context rather than stimulus identity or valence. During active investigation, theta and gamma power decouple, suggesting an active sensing state with distinct top-down (theta) and bottom-up (gamma) processes that differentially reflect stimulus characteristics. Coherence patterns during investigation were most similar within tasks, and a classifier leveraged ΔθCo to identify context above chance, indicating that coordinated rhythmic interactions encode contextual information that can influence behavioral outcomes to the same stimulus presented in different contexts. Granger causality revealed putative hubs, notably vDG and AcbC, with vDG showing directional influence (vDG→LS) and unique event-locked modulations, aligning with its known role in contextual processing and social memory. Together, the results indicate that social context is embedded in the network-level synchronization structure, offering a mechanism by which identical social cues evoke different neural dynamics and behaviors across contexts.

This work demonstrates that LFP theta/gamma power and, especially, inter-regional coherence within the social brain vary with social context. A global internal state elevates rhythmicity during encounters, but context-specific synchronization patterns, rather than stimulus identity or valence, dominate the coordination among regions. The ventral dentate gyrus emerges as a key hub, exhibiting consistent GC involvement across tasks, directional influence, distinct bout-related biases, and precise event-related modulations. These findings provide a systems-level account of how context can reconfigure social brain subnetworks to shape behavior. Future studies could expand recordings to additional regions and cell types, integrate spiking and cross-frequency coupling, causally manipulate candidate hubs (e.g., vDG) to test network control, examine female subjects and other strains/contexts, and develop improved decoders of social context from broader network features.

- Subjects were adult male CD1 mice only; results may not generalize to females, other strains, or species. - Electrode targeting accuracy was limited; not all regions were recorded in every subject, necessitating data imputation for some pairwise measures and potentially reducing decoding performance. - LFP measures provide mesoscopic signals without cellular specificity; mechanisms underlying theta/gamma sources and cross-frequency interactions were not directly probed. - Investigation-bout analyses were restricted to bouts ≥2 s, possibly omitting shorter but behaviorally relevant interactions. - Random forest context decoding accuracy was modest and significant mainly for SxP using ΔθCo alone, suggesting limited feature coverage due to the subset of recorded regions. - The nature of the global internal state (e.g., arousal, motivation) was not directly identified or manipulated.