henhao
Sign in
NeurosciencebioRxiv
Preprint

A widespread internal brain state for fentanyl withdrawal

Karim AbdelaaliD, Kathryn K. Walder-ChristenseniD, Cameron Blount, Kellie Williford, marisella Adams-Grimaldi, Stephen D. Mague, David CarlsoniD, Kafui DzirasaiD

Abstract

Opioid addiction is characterized by escalating drug use, driven in part by negative reinforcement from withdrawal, but the neural processes linking withdrawal to increased drug-taking remain poorly understood. Here, we use multisite local field potential recordings and interpretable machine learning to identify large-scale brain networks engaged by repeated opioid exposure and withdrawal. After discovering that repeated fentanyl exposure induces a network of widespread high beta and low gamma oscillations, we then identified a distinct brain network that selectively encodes the emergence and severity of opioid withdrawal. This latter network, termed EN-Withdrawal, is characterized by regional gamma oscillations and widely synchronized delta/theta oscillations. Its activity patterns predict the emergence of spontaneous and naloxone-precipitated withdrawal across multiple independent cohorts, generalizing across mice, sex, opioids, and dosing regimens, while persisting over multiple days of withdrawal. Using a novel, data-driven severity index, we find that network activity scales with individual behavioral severity without simply reflecting ongoing somatic behaviors or general aversion, suggesting that EN-Withdrawal underlies a withdrawal-induced internal state. Strikingly, network activity predicts the escalation of fentanyl self-administration on a mouse-by-mouse basis in experienced, but not drug-naive, animals. These findings reveal a neurophysiological substrate of the negative reinforcement cycle of addiction that shapes individual vulnerability.

PDFDOICode
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
0
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Sonnet 4.6 · xhigh effort · 7 Nov 2026 · from the full text of version 2 (6 Oct 2026) · CC BY 4.0

A multisite LFP network (EN-Withdrawal), found with supervised dCSFA-NMF in fentanyl-withdrawing mice, classifies morphine withdrawal, tracks behavioral severity, and its rise predicts fentanyl self-administration escalation in experienced mice.

Prior work on opioid withdrawal mostly examined single regions or circuits whose activity is transient and tied to specific motor output, so no sustained, multidimensional neural signature of withdrawal severity had been captured. This paper derives a distributed multiregion oscillatory network and links it to individual escalation of drug intake.

Key results

  1. ShownUnsupervised tensor component analysis on repeated fentanyl dosing found one factor, dominated by high beta and low gamma power and coherence, that ramps across days and was expressed only in fentanyl-treated mice; it replicated in a new cohort.

    Fentanyl vs saline expression t=8.210, df=15.77, p<0.0001 (Welch). Holdout cohort: fentanyl slope across days β=0.112±0.009, p<0.001 vs saline β=0.003±0.007, p=0.70; Day x Treatment β=-0.109±0.014. · Repeated fentanyl exposure alters widespread…

  2. ShownA supervised dCSFA-NMF model separated naloxone-precipitated withdrawal from non-withdrawal and yielded a supervised network, EN-Withdrawal, led by orbitofrontal gamma power with widespread delta/theta coherence; many recorded units were coupled to it.

    Trained on 14 FN, 7 S and 4 AFN mice, with a five-network model chosen by cross-validation. Strongest feature: OFC gamma power (~40-50 Hz). 21.1% of units significantly coupled, a nearly fourfold enrichment over chance (p<0.0001), highest during precipitated withdrawal. · A distributed brain network encoding fentanyl…; Widespread cellular activity couples to…

  3. ShownEN-Withdrawal, trained on fentanyl data, classified withdrawal in morphine-treated mice, was higher in withdrawing mice than in saline controls, followed pharmacological reversal, and stayed elevated at 24 hours but not at 5 days.

    Morphine cohort (N=10): per-mouse AUC=0.622±0.046, p=0.019, supervised network only. Group F1,14=7.948, p=0.0137 vs saline. After morphine rescue p=0.3585 vs saline; 24 h p=0.028; 5 days p=0.473. · EN-Withdrawal predicts morphine withdrawal

  4. SuggestedNetwork activity scaled with a PCA-derived behavioral withdrawal severity index. It did not follow moment-to-moment behaviors and did not rise after a single fentanyl-naloxone pairing in drug-naïve mice that developed place aversion.

    PC1 explained 54.2% of behavioral variance. Correlation with severity r=0.758 (p=0.011) spontaneous and r=0.711 (p=0.021) precipitated (small cohort); escalating-morphine cohort r=0.43, p=0.032 (one-tailed). Drug-naïve network change p=0.210 (N=6). · EN-Withdrawal predicts morphine withdrawal; EN-Withdrawal reflects withdrawal severity

  5. SuggestedIn fentanyl self-administration, baseline EN-Withdrawal activity rose across days, and its per-mouse escalation rate correlated with fentanyl-intake escalation rate in the maintenance phase but not in training.

    N=22 mice. Maintenance phase Spearman ρ=0.46, p=0.03; training phase ρ=0.22, p=0.33. Baseline network activity increased across days (β_Day=0.003±0.001); post-access activity was lower than baseline on the last day (p=0.0004). · Daily EN-Withdrawal activity predicts urgency…; EN-Withdrawal magnitude predicts escalation of…

Methods, data and reproducibility

Multisite LFP power and coherence features (13 regions) were recorded in mice during repeated opioid dosing, withdrawal and self-administration. They were analyzed with unsupervised tensor component analysis and with supervised dCSFA-NMF (a variational-autoencoder-based nonnegative factorization), and tested on held-out mice, other opioids and behavioral severity.

  • Fentanyl dosing and precipitated-withdrawal cohort used for TCA and dCSFA-NMF training (14 fentanyl, 7 saline; 4 saline mice reused as acute fentanyl-naloxone controls) · mouse (C57BL/6J) · n = N=21
  • Holdout fentanyl-versus-saline cohort for the TCA network · mouse (C57BL/6J) · n = fentanyl N=17; saline N=12
  • Constant-dose morphine withdrawal holdout cohort · mouse (C57BL/6J) · n = N=16 (10 morphine, 6 saline)
  • Escalating-dose morphine withdrawal cohort, used for the severity index and as a second holdout · mouse (C57BL/6J) · n = N=23 (15 morphine, 8 saline); severity index cohort described as N=19
  • Fentanyl intravenous self-administration with chronic LFP recording (8 days continuous access, 14 days intermittent access) · mouse (C57BL/6J) · n = N=22
  • Conditioned place aversion to acute fentanyl-naloxone, plus an independent drug-naïve holdout for network activity · mouse (C57BL/6J) · n = N=8 (CPA); N=6 (holdout)
Methods & tools
tensor component analysis (CP decomposition, PyTorch), dCSFA-NMF, MATLAB, PyTorch, Plexon Offline Sorter, Ethovision X12, Cerebus acquisition system, GraphPad Prism 10.5.0, SciPy, statsmodels
Code
https://github.com/carlson-lab/lpne-data-analysishttps://github.com/carlson-lab/lpne/https://github.com/carlson-lab/FentanylWithdrawal
Software & environment
MATLAB R2022a for feature extraction (R2021a for artifact removal); GraphPad Prism 10.5.0; Python with SciPy and statsmodels; no containers or requirement files stated.
Reproducibility gaps
The analysis repository (FentanylWithdrawal) is stated to be made public upon publication. No data deposit or accession is given.
Limitations (6)
  • The study does not test whether stronger expression of the repeated-fentanyl network predicts later EN-Withdrawal activity or withdrawal severity in the same animals. · stated by the authors
  • Unit counts are modest, and no tool exists to directly modulate network-level dynamics, so the network's causal role is not tested by perturbation. · stated by the authors
  • Males and females were included but sex differences were not explicitly tested. Only acute somatic withdrawal was studied, not protracted withdrawal. · stated by the authors
  • Volume conduction of LFP means activity attributed to the basolateral amygdala may include neighboring amygdalar subregions. · stated by the authors
  • Per-mouse AUC on the held-out morphine cohort was modest (0.622±0.046, N=10), and severity correlations rest on small cohorts (r=0.758 and 0.711 with about 10 mice). · noted in this summary
  • The escalation-rate link comes from one self-administration cohort (N=22) at ρ=0.46, p=0.03, without external replication. · noted in this summary

Who should read:Read this if you work on multisite LFP or electome-style network models and want a worked example linking an interpretable supervised factor model to behavioral severity and drug-intake escalation.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISFox '22Fragale '20Dejeux '26Wooldridge '25Downs '24Xie '26
76% semantic matchNeuroscience
Megan E. Fox, Andreas B. Wulff, Daniela Franco+9 more·bioRxiv·May 15, 2022· 2 citations