Machine Learning Based Classification of Deep Brain Stimulation Outcomes in a Rat Model of Binge Eating Using Ventral Striatal Oscillations

Neural oscillations in the ventral striatum reveal differences between the encoding of palatable food and ethanol consumption

BACKGROUND

Across multiple levels of investigation, there appear to be convergent neuronal processes underlying substance use and other motivated behaviors (i.e., the pursuit and consumption of rewarding substances). The consumption of alcohol and sweet, high-fat food engages many of the same brain regions, especially, the ventral striatum. In the current study, we hypothesized that ventral striatal local field potentials (LFPs) recorded during self-administration sessions could be used to detect when the consumption of 10% ethanol or sweet-fat food (SF) was occurring compared to all other behaviors, including naturalistic controls (i.e., water or house-chow).

METHODS

We used an intermittent limited access approach to condition Sprague-Dawley rats to consume either ethanol or SF while we recorded LFPs. We used machine learning and simple logistic regressions to determine whether LFP features could classify when consumption of each substance was occurring, and whether a general model could predict consumption of both substances. We report performance as the average area under the receiver operator characteristic curve (AUROC).

RESULTS

Consumption of a single substance was differentiable from all other behaviors, as evidenced by the AUROC (ethanol = 0.84 and SF = 0.83, p < 0.01). Models built from the combined dataset (general) did modestly overall (general → general = 0.68, p < 0.05), and did not detect the consumption of the two substances similarly (general → SF = 0.5 and general → ethanol = 0.63, p > 0.05).

CONCLUSIONS

Models successfully classified ethanol and SF consumption versus all other behavior/naturalistic controls. However, the findings highlight differences in how the ventral striatum represents the consumption of ethanol and SF and show that, although there is potential for finding biomarkers related to substance use, it may be difficult to build a model that performs well detecting multiple substances.

Maternal immune activation and adolescent alcohol exposure increase alcohol drinking and disrupt cortical-striatal-hippocampal oscillations in adult offspring

Maternal immune activation (MIA) is strongly associated with an increased risk of developing mental illness in adulthood, which often co-occurs with alcohol misuse. The current study aimed to begin to determine whether MIA, combined with adolescent alcohol exposure (AE), could be used as a model with which we could study the neurobiological mechanisms behind such co-occurring disorders. Pregnant Sprague-Dawley rats were treated with polyI:C or saline on gestational day 15. Half of the offspring were given continuous access to alcohol during adolescence, leading to four experimental groups: controls, MIA, AE, and Dual (MIA + AE). We then evaluated whether MIA and/or AE alter: (1) alcohol consumption; (2) locomotor behavior; and (3) cortical-striatal-hippocampal local field potentials (LFPs) in adult offspring. Dual rats, particularly females, drank significantly more alcohol in adulthood compared to all other groups. MIA led to reduced locomotor behavior in males only. Using machine learning to build predictive models from LFPs, we were able to differentiate Dual rats from control rats and AE rats in both sexes, and Dual rats from MIA rats in females. These data suggest that Dual "hits" (MIA + AE) increases substance use behavior and disrupts activity in reward-related circuits, and that this may be a valuable heuristic model we can use to study the neurobiological underpinnings of co-occurring disorders. Our future work aims to extend these findings to other addictive substances to enhance the translational relevance of this model, as well as determine whether amelioration of these circuit disruptions can reduce substance use behavior.

Cannabis Vapor Exposure Alters Neural Circuit Oscillatory Activity in a Neurodevelopmental Model of Schizophrenia: Exploring the Differential Impact of Cannabis Constituents

Cannabis use is highly prevalent in patients with schizophrenia and worsens the course of the disorder. To understand how exposure to cannabis changes schizophrenia-related oscillatory disruptions, we investigated the impact of administering cannabis vapor containing either Δ9-tetrahydrocannabinol (THC) or balanced THC/cannabidiol (CBD) on oscillatory activity in the neonatal ventral hippocampal lesion (NVHL) rat model of schizophrenia. Male Sprague Dawley rats underwent lesion or sham surgeries on postnatal day 7. In adulthood, electrodes were implanted targeting the cingulate cortex (Cg), the prelimbic cortex (PrLC), the hippocampus (HIP), and the nucleus accumbens (NAc). Local field potential recordings were obtained after rats were administered either the "THC-only" cannabis vapor (8-18% THC/0% CBD) or the "Balanced THC:CBD" cannabis vapor (4-11% THC/8.5-15.5% CBD) in a cross-over design with a 2-week wash-out period between exposures. Compared to controls, NVHL rats had reduced baseline gamma power in the Cg, HIP, and NAc, and reduced HIP-Cg high-gamma coherence. THC-only vapor exposure broadly suppressed oscillatory power and coherence, even beyond the baseline reductions observed in NHVL rats. Balanced THC:CBD vapor, however, did not suppress oscillatory power and coherence, and in some instances enhanced power. For NVHL rats, THC-only vapor normalized the baseline HIP-Cg high-gamma coherence deficits. NHVL rats demonstrated a 20 ms delay in HIP theta to high-gamma phase coupling, which was not apparent in the PrLC and NAc after both exposures. In conclusion, cannabis vapor exposure has varying impacts on oscillatory activity in NVHL rats, and the relative composition of naturally occurring cannabinoids may contribute to this variability.

Sex differences in the ability of corticostriatal oscillations to predict rodent alcohol consumption

Background

Although male and female rats differ in their patterns of alcohol use, little is known regarding the neural circuit activity that underlies these differences in behavior. The current study used a machine learning approach to characterize sex differences in local field potential (LFP) oscillations that may relate to sex differences in alcohol-drinking behavior.

Methods

LFP oscillations were recorded from the nucleus accumbens shell and the rodent medial prefrontal cortex of adult male and female Sprague-Dawley rats. Recordings occurred before rats were exposed to alcohol (n = 10/sex × 2 recordings/rat) and during sessions of limited access to alcohol (n = 5/sex × 5 recordings/rat). Oscillations were also recorded from each female rat in each phase of estrous prior to alcohol exposure. Using machine learning, we built predictive models with oscillation data to classify rats based on: (1) biological sex, (2) phase of estrous, and (3) alcohol intake levels. We evaluated model performance from real data by comparing it to the performance of models built and tested on permutations of the data.

Results

Our data demonstrate that corticostriatal oscillations were able to predict alcohol intake levels in males (p < 0.01), but not in females (p = 0.45). The accuracies of models predicting biological sex and phase of estrous were related to fluctuations observed in alcohol drinking levels; females in diestrus drank more alcohol than males (p = 0.052), and the male vs. diestrus female model had the highest accuracy (71.01%) compared to chance estimates. Conversely, females in estrus drank very similar amounts of alcohol to males (p = 0.702), and the male vs. estrus female model had the lowest accuracy (56.14%) compared to chance estimates.

Conclusions

The current data demonstrate that oscillations recorded from corticostriatal circuits contain significant information regarding alcohol drinking in males, but not alcohol drinking in females. Future work will focus on identifying where to record LFP oscillations in order to predict alcohol drinking in females, which may help elucidate sex-specific neural targets for future therapeutic development.

Corticostriatal Oscillations Predict High vs. Low Drinkers in a Rat Model of Limited Access Alcohol Consumption

Individuals differ in their vulnerability to develop alcohol dependence, which is determined by innate and environmental factors. The corticostriatal circuit is heavily involved in the development of alcohol dependence and may contain neural information regarding vulnerability to drink excessively. In the current experiment, we hypothesized that we could characterize high and low alcohol-drinking rats (HD and LD, respectively) based on corticostriatal oscillations and that these subgroups would differentially respond to corticostriatal brain stimulation. Male Sprague–Dawley rats (n = 13) were trained to drink 10% alcohol in a limited access paradigm. In separate sessions, local field potentials (LFPs) were recorded from the nucleus accumbens shell (NAcSh) and medial prefrontal cortex (mPFC). Based on training alcohol consumption levels, we classified rats using a median split as HD or LD. Then, using machine-learning, we built predictive models to classify rats as HD or LD by corticostriatal LFPs and compared the model performance from real data to the performance of models built on data permutations. Additionally, we explored the impact of NAcSh or mPFC stimulation on alcohol consumption in HD vs. LD. Corticostriatal LFPs were able to predict HD vs. LD group classification with greater accuracy than expected by chance (>80% accuracy). Moreover, NAcSh stimulation significantly reduced alcohol consumption in HD, but not LD (p < 0.05), while mPFC stimulation did not alter drinking behavior in either HD or LD (p > 0.05). These data collectively show that the corticostriatal circuit is differentially involved in regulating alcohol intake in HD vs. LD rats, and suggests that corticostriatal activity may have the potential to predict a vulnerability to develop alcohol dependence in a clinical population.

Corticostriatal Oscillations Predict High vs. Low Drinkers in a Rat Model of Limited Access Alcohol Consumption

Individuals differ in their vulnerability to develop alcohol dependence, which is determined by innate and environmental factors. The corticostriatal circuit is heavily involved in the development of alcohol dependence and may contain neural information regarding vulnerability to drink excessively. In the current experiment, we hypothesized that we could characterize high and low alcohol-drinking rats (HD and LD, respectively) based on corticostriatal oscillations and that these subgroups would differentially respond to corticostriatal brain stimulation. Male Sprague-Dawley rats (n = 13) were trained to drink 10% alcohol in a limited access paradigm. In separate sessions, local field potentials (LFPs) were recorded from the nucleus accumbens shell (NAcSh) and medial prefrontal cortex (mPFC). Based on training alcohol consumption levels, we classified rats using a median split as HD or LD. Then, using machine-learning, we built predictive models to classify rats as HD or LD by corticostriatal LFPs and compared the model performance from real data to the performance of models built on data permutations. Additionally, we explored the impact of NAcSh or mPFC stimulation on alcohol consumption in HD vs. LD. Corticostriatal LFPs were able to predict HD vs. LD group classification with greater accuracy than expected by chance (>80% accuracy). Moreover, NAcSh stimulation significantly reduced alcohol consumption in HD, but not LD (p < 0.05), while mPFC stimulation did not alter drinking behavior in either HD or LD (p > 0.05). These data collectively show that the corticostriatal circuit is differentially involved in regulating alcohol intake in HD vs. LD rats, and suggests that corticostriatal activity may have the potential to predict a vulnerability to develop alcohol dependence in a clinical population.

Finding the balance between model complexity and performance: Using ventral striatal oscillations to classify feeding behavior in rats

The ventral striatum (VS) is a central node within a distributed network that controls appetitive behavior, and neuromodulation of the VS has demonstrated therapeutic potential for appetitive disorders. Local field potential (LFP) oscillations recorded from deep brain stimulation (DBS) electrodes within the VS are a pragmatic source of neural systems-level information about appetitive behavior that could be used in responsive neuromodulation systems. Here, we recorded LFPs from the bilateral nucleus accumbens core and shell (subregions of the VS) during limited access to palatable food across varying conditions of hunger and food palatability in male rats. We used standard statistical methods (logistic regression) as well as the machine learning algorithm lasso to predict aspects of feeding behavior using VS LFPs. We were able to predict the amount of food eaten, the increase in consumption following food deprivation, and the type of food eaten. Further, we were able to predict whether the initiation of feeding was imminent up to 42.5 seconds before feeding began and classify current behavior as either feeding or not-feeding. In classifying feeding behavior, we found an optimal balance between model complexity and performance with models using 3 LFP features primarily from the alpha and high gamma frequencies. As shown here, unbiased methods can identify systems-level neural activity linked to domains of mental illness with potential application to the development and personalization of novel treatments.

As neuropsychiatry begins to leverage the power of computational methods to understand disease states and to develop better therapies, it is vital that we acknowledge the trade-offs between model complexity and performance. We show that computational methods can elucidate a neural signature of feeding behavior and we show how these methods could be used to discover neural patterns related to other behaviors and reveal new potential therapeutic targets. Further, our results help to contextualize both the limitations and potential of applying computational methods to neuropsychiatry by showing how changing the data being used to train predictive models (e.g., population vs. individual data) can have a large impact on how model performance generalizes across time, internal states, and individuals.