Shank3-deficient rats show a sign-inverted accumbal dopamine response during pouncing, and closed-loop optogenetic VTA-NAc stimulation persistently lengthens social play

In wild-type male rats, proactive play (pouncing, pinning) evokes nucleus accumbens dopamine surges, whereas forced submission suppresses dopamine; wild-type females instead show dopamine increases during evasion and rearing but not contact-heavy play. Prior dopamine measurements of juvenile social play largely stayed at the level of overall social contact; this work aligns dopamine dynamics to 800 ms behavioral syllables and to active versus passive roles, thereby separating opposite dopamine directions within the same social interaction by role and sex. Based on GRAB-DA3m fiber photometry of NAc dopamine synchronized with nine-camera volumetric tracking and Social-Seq behavioral syllable labels, with recording at PND 45-55 and behavioral testing at PND 35-45, 55-65, and 75-85.

Shank3+/- mutants show blunted dopamine responses during sniffing and chasing, a sign-inverted dopamine response during pouncing, and exaggerated dopamine during solitary rearing only in mutant males. Links Shank3 deficiency to dopamine encoding abnormalities at the level of specific behavioral syllables rather than only reporting reduced total social behavior, suggesting the mutation affects the sign and context specificity of the reward signal, not merely its magnitude. WT versus Shank3+/- comparison under the same synchronized recording pipeline, with behavioral syllables generated by a Seq2seq autoencoder and clustering and validated by manual annotation of reciprocal pairs; dopamine response defined as post-onset peak ΔF/F minus baseline mean.

A multi-agent reinforcement learning model whose rewards were parameterized with empirical dopamine amplitudes reproduced the mutant phenotype, including preserved sniffing, reduced play, and disrupted transitions. Translates behavioral-level dopamine abnormalities into a computable reward signal tested within a multi-agent game, indicating that changing reward calibration alone can generate the mutant behavioral profile without positing additional motor or sensory deficits. A 4×4 fully observable grid with PPO agents (RLlib, 2×256 tanh layers, 120k training steps, γ=0.95, lr=3e-4), evaluated over 30 rollouts, with rewards calibrated from empirical dopamine amplitudes.

In Shank3+/- males, closed-loop optogenetic stimulation of VTA-NAc projections at proactive play initiation persistently increased targeted play duration and reduced nonsocial leaving. Moves from correlational dopamine-behavior alignment to causal intervention, and because stimulation was triggered by real-time behavior detection rather than delivered at fixed times, it positions dopamine as an action-contingent reinforcement signal rather than an accompanying phenomenon. Closed-loop design with 3 baseline days, 8 days of behavior-triggered training, and a subsequent 10-day stimulation-free period; real-time detection used YOLOv8n-seg, DANNCE, and BiLSTM-FCN across 36 behavior classes, with 470 nm laser at 4 mW, 40 Hz, 0.5-s pulses.

This resource is intended for researchers studying nucleus accumbens dopamine encoding during juvenile social play and the effects of Shank3 deficiency, applicable to same-sex dyadic interaction, PND 35-85 rats, and analyses whose unit is the behavioral syllable. The data include raw 3D keypoint coordinates, behavioral syllable labels, synchronized Z-scored dopamine ΔF/F traces, optogenetic timestamps, and summary metrics (frequencies, durations, transition matrices), and can be used to replicate the ethogram, perform neural-behavior alignment, parameterize computational models, or evaluate intervention outcomes. The closed-loop optogenetic results demonstrate sufficiency, not exclusivity, of dopamine for targeted play.

The current read is an incomplete version, with figures and supplementary statistics not included, so effect sizes for the sign-inverted dopamine response, sample sizes per behavioral syllable, and statistical test details cannot be confirmed here. Sex must be treated as a covariate, behavioral labels rely on quantitative operational definitions, the Shank3 model is specific to Phelan-McDermid syndrome, and closed-loop optogenetics demonstrates sufficiency rather than exclusivity. The multi-agent model calibrates rewards from empirical dopamine amplitudes, and how its parameter choices affect the robustness of conclusions remains an open question.