An Electroencephalogram Signature of Melanin-Concentrating Hormone Neuron Activities Predicts Cocaine Seeking

Abstract

Background: Identifying biomarkers that predict substance use disorder propensity may better strategize antiaddiction treatment. Melanin-concentrating hormone (MCH) neurons in the lateral hypothalamus critically mediate interactions between sleep and substance use; however, their activities are largely obscured in surface electroencephalogram (EEG) measures, hindering the development of biomarkers.

Methods: Surface EEG signals and real-time calcium (Ca²⁺) activities of lateral hypothalamus MCH neurons (Ca²⁺_MCH) were simultaneously recorded in male and female adult rats. Mathematical modeling and machine learning were then applied to predict Ca²⁺_MCH using EEG derivatives. The robustness of the predictions was tested across sex and treatment conditions. Finally, features extracted from the EEG-predicted Ca²⁺_MCH either before or after cocaine experience were used to predict future drug-seeking behaviors.

Results: An EEG waveform derivative-a modified theta-delta-theta peak ratio (EEG_TDT ratio)-accurately tracked real-time Ca²⁺_MCH in rats. The prediction was robust during rapid eye movement sleep (REMS), persisted through vigilance states, sleep manipulations, and circadian phases, and was consistent across sex. Moreover, cocaine self-administration and long-term withdrawal altered EEG_TDT ratio, suggesting shortening and circadian redistribution of synchronous MCH neuron activities. In addition, features of EEG_TDT ratio indicative of prolonged synchronous MCH neuron activities predicted lower subsequent cocaine seeking. EEG_TDT ratio also exhibited advantages over conventional REMS measures for the predictions.

Conclusions: The identified EEG_TDT ratio may serve as a noninvasive measure for assessing MCH neuron activities in vivo and evaluating REMS; it may also serve as a potential biomarker for predicting drug use propensity.

Keywords: Biomarker; Cocaine; EEG; MCH; REM sleep.

Figure 1.

Ca²⁺_MCH transients correlate with coordinated changes in EEG θ and δ activities in sleep. (A) EEG and fiber photometry dual recording diagram. (B) Ca²⁺_MCH transients occurred predominantly during REMS. (C) Ca²⁺_MCH peaks were accompanied by an increase in θ (red) and a decrease in δ (blue) power, with a shift in the peak of θ frequency (yellow circle) towards higher θ frequency. (D) EEG spectrogram aligned with Ca²⁺_MCH over NREM and REM episodes, showing increase in θ and decrease in δ power during Ca²⁺_MCH activities. Red dots in upper panel represent peak frequencies of EEG power (0–20 Hz) at 1-sec resolution. (E) Peak frequencies of θ activities during a REMS episode from (D). (F) 5–9 Hz θ peak frequencies showed a positive correlation with Ca²⁺_MCH levels over a 2-h recording period. (G) θ/δ ratio showed a positive correlation with Ca²⁺_MCH levels over the same 2-h recording period. (H) θ peak × (θ/δ ratio) showed a positive correlation with Ca²⁺_MCH levels over the same 2-h recording period, with an improved r-value.

Figure 2.

Ca²⁺_MCH is correlated with EEG_TDT Ratio: empirical formula and cross-validation. (A) Example Ca²⁺_MCH aligned to EEG and individual EEG features calculated over 10-sec sliding windows at 1-sec steps. Ca²⁺_MCH was smoothed using the same sliding window. EEG_TDT Ratio = (Power θ/ Power δ) × (F _peak/7_Hz); threshold (pink dotted line) = amplitude at 60% dwelling time over 2-h recordings. (B) r-values for EEG_TDT Ratio – Ca²⁺_MCH correlations from an example 2-h recording, which were calculated using 36 × 91 combinations of extended θ and δ ranges at 0.5 Hz increments for optimization. (C) Cross-validation for training sample size k = 1,⋯, 7. When k ≥ 6, the performance became stable. (D) Example EEG_TDT Ratio calculated using the optimized formula: EEG_TDT Ratio = (Power θ_{6.5–9.5 Hz} / Power δ_{1–5.5 Hz}) × (F_peak / 7 Hz), and aligned to Ca²⁺_MCH. (E) EEG_TDT ratio – Ca²⁺_MCH correlation over a 2-h recording in an example rat. (F) Supra-threshold EEG_TDT Ratio – Ca²⁺_MCH correlation over the same 2-h recordings in the example rat. (G) r-values at different EEG_TDT Ratio – Ca²⁺_MCH time lags in the example rat.

Figure 3.

Ca²⁺_MCH can be predicted from EEG signals using machine-learning (ML). (A) Example Ca²⁺_MCH over time aligned with ML-predicted, amplitude-normalized (i.e. Z-scored), Ca²⁺_MCH, and the corresponding empirically calculated EEG_TDT Ratio. (B) ML using r-values to optimize # of layers and # of units per layer for the multilayer perceptron network, with 10-fold cross-validation (shown in circles) for each condition. These networks yielded similar performances on the pilot data (layer × unit interaction: F_{4, 90}=0.020, p=0.999; two-way ANOVA), therefore we chose the one with the best average performance: networks with 3 hidden layers and 20 units/layer. (C) EEG-ML– Ca²⁺_MCH correlation over a 2-h recording in the example rat using an optimized multilayer perceptron network with 3 hidden layers and 20 units/layer. (D) r-values at different EEG-ML – Ca²⁺_MCH time lags in the example rat. (E) EEG_TDT ratio – Ca²⁺_MCH correlation over a 2-h recording in the example rat. (F) r-values at different EEG_TDT Ratio – Ca²⁺_MCH time lags in the example rat. (G) Among the testing set (n=36 new 2-h recordings), EEG-ML versus EEG_TDT Ratio performance was not significantly different (t₃₅=0.838, p=0.408, n=36 rats each, male and female, paired t-test). ns = not significant. (H) The r-values for ML-Ca²⁺_MCH predictions versus EEG_TDT Ratio Ca²⁺_MCH predictions were correlated. Each circle represents one rat and/or condition. n=47 rats and/or conditions, male and female. (I) Normalized Ca²⁺_MCH (x-axis), EEG_TDT Ratio (y-axis), and ML-predicted Ca²⁺ (z-axis) triplet-pairs were plotted within a 1×1×1 cube, using data from the example rat. Each dot represents a 10-sec sample from the 2-h recording. Color gradient represents the distance to the diagonal line (x=y=z). (J) Same plot as i viewed along the diagonal line.

Figure 4.

EEG_TDT Ratio – Ca²⁺_MCH correlations across sex, treatment conditions, and light/dark phases. (A) EEG and fiber photometry dual recording diagram and timeline. (B) EEG_TDT Ratio – Ca²⁺_MCH correlation r-values were similar in males (n=9) and females (n=5) (t₁₂=0.112, p=0.913, t-test). (C) EEG_TDT Ratio – Ca²⁺_MCH correlations persisted under REMS manipulations by environmental warming in naïve/saline-treated rats or following cocaine experience (interaction: F_{1, 33}=1.913, p=0.176, two-way ANOVA). (D) The slope of the EEG_TDT Ratio – Ca²⁺_MCH plot was not altered following cocaine exposure or REMS manipulations by environmental warming (interaction: F_{1, 10}=0.612, p=0.452, two-way RM ANOVA). (E) %Changes in EEG_TDT Ratio versus % changes in Ca²⁺_MCH responses under sleep disturbance/rebound, and environmental warming to increase REMS; within-subject comparisons. (F) In the dark phase during explorative behaviors in wakefulness, there was asynchronous Ca²⁺_MCH activity accompanied by high-frequency, low-amplitude EEG_TDT events. (G-H) EEG_TDT Ratio-Ca²⁺_MCH correlation from the example 2-h recordings. Each dot in (G) represents a 10-sec sample from the recording. (I) Overall reduced correlation r-values during the randomly sampled dark phase hours compared to light phase (t₂₅=3.650, p<0.01, t-test). arrows in (F) indicate occasions of EEG_TDT events when Ca²⁺_MCH was absent. Each circle represents one rat. n=5–14 male and female rats.

Figure 5.

Cocaine-induced long-term changes in EEG_TDT Ratio. (A) Cocaine self-administration training and EEG recording timeline. (B) Example 24-h EEG_TDT Ratio extracted from baseline sleep before cocaine exposure (left) or after 21 days of withdrawal (right) from repeated cocaine self-administration (1 overnight + 2-h daily × 5 d, 0.75 mg/kg/infusion, FR1). (C, D) The 24-h EEG_TDT Ratio exhibited reduced activities in the dark phase and increased activities in the light phase at withdrawal d21, measured by the total event time (C) or total AUC (D). (C) interaction: F_{1, 54}=8.804, p<0.01; (D) interaction: F_{1, 54}=9.815, p<0.01; two-way RM ANOVA with Sidak post-hoc test. (E, F) Similar changes were observed in REMS, measured by the total REMS time (E) or θ energy (F) during REMS (θ energy = θ power × duration). (E) interaction: F_{1, 54}=12.97, p<0.001; (F) interaction: F_{1, 54}=13.79, p<0.001; two-way RM ANOVA with Sidak post-hoc test. (G) Cluster analysis of EEG_TDT Ratio showed a decrease in the average cluster durations in the dark phase and an increase in the light phase. (H) Long (>50 sec) EEG_TDT event clusters showed a decrease in durations in both dark and light phases. (G) interaction: F_{1, 54}=5.053, p<0.05; (H) interaction: F_{1, 54}=3.458, p=0.068; two-way RM ANOVA with Sidak post-hoc test. (I) In contrast to (G), average REMS bout duration did not change in either dark or light phase. (J) In contrast to (H), average long-bout REMS did not change in light phase. (I) interaction: F_{1, 54}=11.67, p<0.01; (J) interaction: F_{1, 54}=14.99, p<0.001; two-way RM ANOVA with Sidak post-hoc test. # p<0.05, ## p<0.01, ### p<0.001 post-hoc test. Each circle represents a rat. n=25–31 male rats.

Figure 6.

Baseline EEG_TDT Ratio features predict future cocaine intake. (A, B) Dark-light distribution ratio of EEG_TDT Ratio events at baseline before cocaine exposure showed correlations with cocaine intake during subsequent self-administration training. (C, D) Same EEG_TDT Ratio features at withdrawal d21 showed correlations with cocaine intake during self-administration training. (E-H) Corresponding REMS features from before cocaine exposure (E, F) or withdrawal d21 (G, H) showed correlations with cocaine intake during self-administration training. (I, J) Cluster analysis of EEG_TDT events showed that the average durations (I) or inter-cluster intervals (J) from before cocaine exposure were correlated with cocaine intake during self-administration training. (K, L) Corresponding REMS features from before cocaine exposure showed varying extent of correlations with cocaine intake during self-administration training. (M, N) Long (>50 sec)-EEG_TDT event clusters showed inverse correlations between average cluster durations and the amount of cocaine intake during self-administration training, both at baseline (M) and on withdrawal d21 (N). (O, P) Long-REMS episodes showed inverse correlations between average bout durations and the amount of cocaine intake during self-administration training, both at baseline (O) and on withdrawal d21 (P). RNF = reinforcement (i.e. cocaine infusions). Each circle represents a rat. n=31 male rats.

Figure 7.

EEG_TDT Ratio features after long-term withdrawal are correlated with incubation of cocaine craving. (A) Cocaine self-administration training and EEG recording timeline. (B, C) 24-h EEG_TDT Ratio total event AUC was negatively correlated with the incubation index (B) or change-of-ANP (active nose-pokes on withdrawal d45-d1) incubation (C) on withdrawal d45. (D, E) Total # of EEG_TDT Ratio long clusters was negatively correlated with #ANP on withdrawal d1 (D) and on withdrawal d45 (E). (F, G) Corresponding REMS features such as 24-h total REMS θ energy did not show correlations with the incubation index (F) or change-of-ANP incubation (G) on withdrawal d45. (H, I) Total # of long REMS bouts did not show correlations with #ANP on withdrawal d1 (H) or on withdrawal d45 (I). Each circle represents a rat. n=16 male rats.