An artificial neural network for automated behavioral state classification in rats

Accurate behavioral state classification is critical for many research applications. Researchers typically rely upon manual identification of behavioral state through visual inspection of electrophysiological signals, but this approach is time intensive and subject to low inter-rater reliability. To overcome these limitations, a diverse set of algorithmic approaches have been put forth to automate the classification process. Recently, novel machine learning approaches have been detailed that produce rapid and highly accurate classifications. These approaches however, are often computationally expensive, require significant expertise to implement, and/or require proprietary software that limits broader adoption. Here we detail a novel artificial neural network that uses electrophysiological features to automatically classify behavioral state in rats with high accuracy, sensitivity, and specificity. Common parameters of interest to sleep scientists, including state-dependent power spectra and homeostatic non-REM slow wave activity, did not significantly differ when using this automated classifier as compared to manual scoring. Flexible options enable researchers to further increase classification accuracy through manual rescoring of a small subset of time intervals with low model prediction certainty or further decrease researcher time by generalizing trained networks across multiple recording days. The algorithm is fully open-source and coded within a popular, and freely available, software platform to increase access to this research tool and provide additional flexibility for future researchers. In sum, we have developed a readily implementable, efficient, and effective approach for automated behavioral state classification in rats.

Interindividual differences in memory system local field potential activity predict behavioral strategy on a dual-solution T-maze

Individuals can use diverse behavioral strategies to navigate their environment including hippocampal-dependent place strategies reliant upon cognitive maps and striatal-dependent response strategies reliant upon egocentric body turns. The existence of multiple memory systems appears to facilitate successful navigation across a wide range of environmental and physiological conditions. The mechanisms by which these systems interact to ultimately generate a unitary behavioral response, however, remain unclear. We trained 20 male, Sprague-Dawley rats on a dual-solution T-maze while simultaneously recording local field potentials that were targeted to the dorsolateral striatum and dorsal hippocampus. Eight rats spontaneously exhibited a place strategy while the remaining 12 rats exhibited a response strategy. Interindividual differences in behavioral strategy were associated with distinct patterns of LFP activity between the dorsolateral striatum and dorsal hippocampus. Specifically, striatal-hippocampal theta activity was in-phase in response rats and out-of-phase in place rats and response rats exhibited elevated striatal-hippocampal coherence across a wide range of frequency bands. These contrasting striatal-hippocampal activity regimes were (a) present during both maze-learning and a 30 min premaze habituation period and (b) could be used to train support vector machines to reliably predict behavioral strategy. Distinct patterns of neuronal activity across multiple memory systems, therefore, appear to bias behavioral strategy selection and thereby contribute to interindividual differences in behavior.

Cortical zeta-inhibitory peptide injection reduces local sleep need

Local sleep need within cortical circuits exhibits extensive interregional variability and appears to increase following learning during preceding waking. Although the biological mechanisms responsible for generating sleep need are unclear, this local variability could arise as a consequence of wake-dependent synaptic plasticity. To test whether cortical synaptic strength is a proximate driver of sleep homeostasis, we developed a novel experimental approach to alter local sleep need. One hour prior to light onset, we injected zeta-inhibitory peptide (ZIP), a pharmacological antagonist of protein kinase Mζ, which can produce pronounced synaptic depotentiation, into the right motor cortex of freely behaving rats. When compared with saline control, ZIP selectively reduced slow-wave activity (SWA; the best electrophysiological marker of sleep need) within the injected motor cortex without affecting SWA in a distal cortical site. This local reduction in SWA was associated with a significant reduction in the slope and amplitude of individual slow waves. Local ZIP injection did not significantly alter the amount of time spent in each behavioral state, locomotor activity, or EEG/LFP power during waking or REM sleep. Thus, local ZIP injection selectively produced a local reduction in sleep need; synaptic strength, therefore, may play a causal role in generating local homeostatic sleep need within the cortex.

Infraslow coordination of slow wave activity through altered neuronal synchrony

Slow wave activity (SWA; the EEG power between 0.5 and 4 Hz during non-rapid eye movement sleep [NREM]) is the best electrophysiological marker of sleep need; SWA dissipates across the night and increases following sleep deprivation. In addition to these well-documented homeostatic SWA trends, SWA exhibits extensive variability across shorter timescales (seconds to minutes) and between local cortical regions. The physiological underpinnings of SWA variability, however, remain poorly characterized. In male Sprague-Dawley rats, we observed that SWA exhibits pronounced infraslow fluctuations (~40- to 120-s periods) that are coordinated across disparate cortical locations. Peaks in SWA across infraslow cycles were associated with increased slope, amplitude, and duration of individual slow waves and a reduction in the total number of waves and proportion of multipeak waves. Using a freely available data set comprised of extracellular unit recordings during consolidated NREM episodes in male Long-Evans rats, we further show that infraslow SWA does not appear to arise as a consequence of firing rate modulation of putative excitatory or inhibitory neurons. Instead, infraslow SWA was associated with alterations in neuronal synchrony surrounding "On"/"Off" periods and changes in the number and duration of "Off" periods. Collectively, these data provide a mechanism by which SWA can be coordinated across disparate cortical locations and thereby connect local and global expression of this patterned neuronal activity. In doing so, infraslow SWA may contribute to the regulation of cortical circuits during sleep and thereby play a critical role in sleep function.

Sleep/wake dependent changes in cortical glucose concentrations

Most of the energy in the brain comes from glucose and supports glutamatergic activity. The firing rate of cortical glutamatergic neurons, as well as cortical extracellular glutamate levels, increase with time spent awake and decline throughout non rapid eye movement sleep, raising the question whether glucose levels reflect behavioral state and sleep/wake history. Here chronic (2-3 days) electroencephalographic recordings in the rat cerebral cortex were coupled with fixed-potential amperometry to monitor the extracellular concentration of glucose ([gluc]) on a second-by-second basis across the spontaneous sleep-wake cycle and in response to 3 h of sleep deprivation. [Gluc] progressively increased during non rapid eye movement sleep and declined during rapid eye movement sleep, while during wake an early decline in [gluc] was followed by an increase 8-15 min after awakening. There was a significant time of day effect during the dark phase, when rats are mostly awake, with [gluc] being significantly lower during the last 3-4 h of the night relative to the first 3-4 h. Moreover, the duration of the early phase of [gluc] decline during wake was longer after prolonged wake than after consolidated sleep. Thus, the sleep/wake history may affect the levels of glucose available to the brain upon awakening.

Extracellular levels of lactate, but not oxygen, reflect sleep homeostasis in the rat cerebral cortex

STUDY OBJECTIVE

It is well established that brain metabolism is higher during wake and rapid eye movement (REM) sleep than in nonrapid eye movement (NREM) sleep. Most of the brain's energy is used to maintain neuronal firing and glutamatergic transmission. Recent evidence shows that cortical firing rates, extracellular glutamate levels, and markers of excitatory synaptic strength increase with time spent awake and decline throughout NREM sleep. These data imply that the metabolic cost of each behavioral state is not fixed but may reflect sleep-wake history, a possibility that is investigated in the current report.

DESIGN

Chronic (4d) electroencephalographic (EEG) recordings in the rat cerebral cortex were coupled with fixed-potential amperometry to monitor the extracellular concentration of oxygen ([oxy]) and lactate ([lac]) on a second-by-second basis across the spontaneous sleep-wake cycle and in response to sleep deprivation.

SETTING

Basic sleep research laboratory.

PATIENTS OR PARTICIPANTS

Wistar Kyoto (WKY) adult male rats.

INTERVENTIONS

N/A.

MEASUREMENTS AND RESULTS

Within 30-60 sec [lac] and [oxy] progressively increased during wake and REM sleep and declined during NREM sleep (n = 10 rats/metabolite), but with several differences. [Oxy], but not [lac], increased more during wake with high motor activity and/or elevated EEG high-frequency power. Meanwhile, only the NREM decline of [lac] reflected sleep pressure as measured by slow-wave activity, mirroring previous results for cortical glutamate.

CONCLUSIONS

The observed state-dependent changes in cortical [lac] and [oxy] are consistent with higher brain metabolism during waking and REM sleep in comparison with NREM sleep. Moreover, these data suggest that glycolytic activity, most likely through its link with glutamatergic transmission, reflects sleep homeostasis.