Abstract
BACKGROUND: Serotonergic psychedelics, which display a high affinity and specificity for 5-HT(2A) receptors like 2,5-dimethoxy-4-iodoamphetamine (DOI), reliably induce a head-twitch response in rodents characterized by paroxysmal, high-frequency head rotations. Traditionally, this behavior is manually counted by a trained observer. Although automation could simplify and facilitate data collection, current techniques require the surgical implantation of magnetic markers into the rodent's skull or ear. METHODS: This study aimed to assess the feasibility of a marker-less workflow for detecting head-twitch responses using deep learning algorithms. High-speed videos were analyzed using the DeepLabCut neural network to track head movements, and the Simple Behavioral Analysis (SimBA) toolkit was employed to build models identifying specific head-twitch responses. RESULTS: In studying DOI (0.3125-2.5 mg/kg) effects, the deep learning algorithm workflow demonstrated a significant correlation with human observations. As expected, the preferential 5-HT(2A) receptor antagonist ketanserin (0.625 mg/kg) attenuated DOI (1.25 mg/kg)-induced head-twitch responses. In contrast, the 5-HT(5A) receptor antagonists SB 699,551 (3 and 10 mg/kg), and ASP 5736 (0.01 and 0.03 mg/kg) failed to do so. CONCLUSIONS: Previous drug discrimination studies demonstrated that the 5-HT(5A) receptor antagonists attenuated the interoceptive cue of a potent hallucinogen LSD, suggesting their anti-hallucinatory effects. Nonetheless, the present results were not surprising and support the head-twitch response as selective for 5-HT(2A) and not 5-HT(5A) receptor activation. We conclude that the DeepLabCut and SimBA toolkits offer a high level of objectivity and can accurately and efficiently identify compounds that induce or inhibit head-twitch responses, making them valuable tools for high-throughput research.