Astronomy of the Brain: Neural Reorganization Across Persistent Fentanyl Use

Abstract

Persistent opioid use despite adverse consequences involves a series of events that unfold while the drug is already on board, yet preclinical models usually investigate choices made drug-free. To address this, we developed a delayed shock paradigm for oral fentanyl self-administration in rats. After a stable baseline of fentanyl intake was established, we escalated shock intensity across days (0.1, 0.2, 0.4, and 0.8 mA), spanning the mild intensities conventional in self-administration work and the higher intensities conventional in stress work, and then imposed a week of forced abstinence followed by a Relapse test at 0.8 mA. During the shock sessions the animals could reach their preferred fentanyl concentration during the Safe period (15 trials), which was followed by the Danger period (remaining 165 trials) in which any rewarded lever press delivered the reward and then a shock 4 s later. We found that the Danger-trial responding was indistinguishable from baseline through 0.4 mA and collapsed at 0.8 mA, so the same animals would have been called shock resistant by the self-administration convention and shock sensitive by the stress convention, and both readings would have been artifacts of an inherited intensity rather than facts about the animal. A within-session changepoint model showed that shock moved the timing of the strategy switch rather than whether it occurred, and ultrasonic vocalizations reorganized across the same ladder while dissociating from responding, so the switch was also captured by a measure that did not depend on lever pressing. This paradigm demanded multi-hour recordings and the ability to register the same neurons across weeks, and neither capability existed. We therefore built: 1) The Miniscope Processing Suite (MPS), which carried 7.26 TB and 77 h of raw video through preprocessing, source extraction, and deconvolution in 55.6 h on a single workstation and recovered every neuron of a matched synthetic ground truth, and 2) Stars2Cells (S2C), which registers neurons from the geometry of their four-neuron neighborhoods, reaching an F1 of 98.4% across a 1,262-run synthetic benchmark, reproducing the Allen Brain Observatory registration across 431 two-photon containers at a median F1 of 0.97, and retaining 41.6 +/- 3.4% of first-session neurons across a four-session chain against 13.0 +/- 3.9\% for footprint matching. We validated this tool stack in the dorsomedial striatum (DMS) during a fentanyl behavioral-economics task. Same-cell tracking revealed that the press-responsive fraction held near one half across the demand curve while roughly half of the tracked neurons switched class at every transition. Applying the same tools to the lateral habenula (LHb) in our aforementioned paradigm, we found that the shock was the most decodable event in the session (ROC-AUC 0.88 at 0.8 mA) and that a shock ensemble of comparable strength was recovered at Relapse a week later, yet the population reported a press only after it had occurred and forecast neither an upcoming press nor the switch to withholding. Thus, we establish that the LHb registers what a choice cost rather than computing what to do next under adverse consequences during active fentanyl self-administration. Furthermore, since a policy for stopping was acquired while the drug was on board and remained stable across weeks, we suggest that treatment needs to address the brain state under active drug use and not only during abstinence.