|Home | About | Journals | Submit | Contact Us | Français|
The role of specific cortical regions in sleep-regulating circuits is unclear. The anterior insula (AI) has strong reciprocal connectivity with wake and sleep-promoting hypothalamic and brainstem regions, and we hypothesized that the AI regulates patterns of sleep and wakefulness. To test this hypothesis, we lesioned the AI in rats (n = 8) and compared sleep, wake, and activity regulation in these animals with nonlesioned controls (n = 8) with 24-h sleep recordings and chronic infrared activity monitoring. Compared to controls, animals with AI lesions had decreased wakefulness and increased rapid eye movement (REM) sleep and non-REM (NREM) sleep. AI-lesioned animals had shorter wake bouts, especially during the active dark phase. AI-lesioned animals also had more transitions from NREM to REM sleep, especially during the inactive light phase. Chronic infrared monitoring revealed that AI-lesioned animals also had a disturbed temporal organization of locomotor activity at multiple time scales with more random activity fluctuations from 4 to 12 h despite intact circadian rhythms. These results suggest that the AI regulates sleep and activity and contributes to the regulation of sleep and motor behavior rhythmicity across multiple time scales. Dysfunction of the AI may underlie changes in sleep-wake patterns in neurological diseases.
Sleep is a necessary behavior shared by all mammals. Regions that regulate sleep are distributed throughout the neuraxis. For example, multiple brain regions promote the transition to or maintenance of sleep, including the ventrolateral preoptic hypothalamus (Sherin et al., 1996; Lu et al., 2000), the medullary parafacial zone (Anaclet et al., 2012, 2014), and nitric oxide synthase–containing neurons spread throughout the cerebral cortex (Gerashchenko et al., 2008; Morairty et al., 2013). Similarly, many brain regions promote the transition to or maintenance of wakefulness, including the parabrachial nucleus (Fuller et al., 2011), orexinergic neurons (Adamantidis et al., 2007), the locus coeruleus (Aston-Jones and Bloom, 1981; Foote et al., 1991; Carter et al., 2010), and others. These sleep-promoting and wake-promoting systems are thought to interact with circadian and other regulatory drivers to regulate the overall behavioral state of an animal (Saper et al., 2010), contributing to the temporal organization of activity at multiple time scales (Hu et al., 2007).
Sleep is often measured with the encephalogram (EEG), an ensemble readout of cortical activity. Surprisingly, the precise contribution of the cortex itself to overall sleep regulation is unclear. Removal of cortical influence in transection studies preserves the gross behavioral signs of sleep (Villablanca, 2004). Nonetheless, the cortex may play a key role in regulating sleep-wake patterns. For example, ablation of the frontal cortex reduces sleep, especially rapid eye movement (REM) sleep (Villablanca et al., 1976). Specific subregions within the frontal cortex may regulate sleep-wake states through projections to powerful sleep-regulatory regions in the brainstem (e.g., ventromedial prefrontal cortex projections to ventrolateral periaqueductal gray regions) to regulate REM sleep (Chang et al., 2014). Similarly, the anterior cingulate cortex, via projections to and from the locus coeruleus, is thought to promote wakefulness, specifically in response to novelty in the environment (Gompf et al., 2010).
In normal conditions, the cortex may help organize sleep-wake states and patterns of behavior across multiple time scales. Not surprisingly, cortical regions have also been implicated in sleep disorders. For example, the anterior insula (AI) and nearby orbitofrontal cortex gray matter have been previously implicated in insomnia (Stoffers et al., 2012). A recent study using functional magnetic resonance imaging found increased coactivation of the AI within a cortical network—including the wake-modulating anterior cingulate—that is thought to underlie attention and arousal (Chen et al., 2014). In addition to cortical projections, the AI projects heavily to the critical arousal-promoting parabrachial nucleus, which in turn projects back to the AI (Saper, 1982; Allen et al., 1991). Thus, the AI may be an intriguing locus of cortical sleep-wake control.
In the present study, we set out to determine the role of the AI in regulating sleep-wake behavior and patterns of activity across hours and days. We lesioned the AI in rats and measured sleep polysomnography and chronic locomotor behavior. We found that lesions of the AI in rats increase REM and non-REM (NREM) sleep and decrease wakefulness. AI lesions alter patterns of sleep-wake over a 24-h period and lead to a significant reduction in temporal correlations of behavior at multiple time scales (i.e., activity fluctuations at longer time scales become more random), especially at ultradian time scales. Collectively, these data suggest the AI plays a key role in regulating the timing and quantity of sleep-wake behavior.
For all experiments, we used pathogen-free, age-matched adult male Sprague-Dawley rats, 350 to 400 g (Harlan). Rats were individually housed in temperature- and humidity-controlled holding rooms with ad libitum access to food and water. Rats were housed on a 12:12 light-dark cycle (light: 0700-1900 h) and were cared for in accordance with National Institutes of Health standards. All procedures were preapproved by the Beth Israel Deaconess Medical Center Institutional Animal Care and Use Committee.
For lesion and EEG/EMG implantation, animals were anesthetized with ketamine-xylazine (intraperitoneally [i.p.], 100 mg/kg ketamine, 10 mg/kg xylazine; Med-Vet, Mettawa, IL) and then placed in a stereotaxic frame. Injections of 10% ibotenic acid (IBO; Tocris, Ellisville, MO) or 0.9% saline (Med-Vet) were administered directly into the brain using a fine glass pipette (1-mm glass stock, tapering slowly to a 10- to 20-μm tip) connected to an air compression system. A series of 20- to 40-psi puffs of air were used to deliver 30- to 45-nL injections into the AI (anterior-posterior (AP): +2.00 mm, dorsal-ventral (DV): −4.2 mm, right-left (RL): ±4.5 mm from the bregma) (Paxinos and Watson, 2006). Animals were then implanted with 4 EEG screw electrodes (Plastics One, Roanoke, VA) in the skull, 2 in the frontal bone and 2 in the parietal bone, and 2 EMG wire electrodes in the nuchal muscles. The EEG/EMG leads were placed into an electrode pedestal, and the entire implant was cemented to the skull using dental acrylic (Lang Dental, Wheeling, IL). Animals then received a subcutaneous injection of meloxicam (1.0 mg/kg; Med-Vet).
One week postsurgery, animals were placed in temperature- and light-controlled recording chambers for sleep recording. Flexible cables attached to rotating commutators were attached to electrode implants, and rats were given access to ad libitum food and water as before. After 24-h habituation, 24-h sleep-wake recordings with infrared video were made with a VitalRecorder (Kissei Comtec, Nagano, Japan). Following the completion of the sleep-wake recordings, animals were disconnected from commutators and placed into isolated holding containers with infrared sensors and recorded for at least 6 weeks.
EEG/EMG was recorded at 256 Hz and exported into SleepSign (Kissei Comtec) for analysis. EEG was band-pass filtered from 0.1 to 60 Hz, and a 10-Hz high-pass filter was applied to the EMG. EEG/EMG signals were divided into 10-sec epochs and scored using automatic scoring parameters with manual correction. Briefly, wake was identified with low-amplitude EEG and EMG activity with video-recorded movement. NREM sleep was identified with high-amplitude, low-frequency delta (0.5-4.5Hz) activity and low EMG, and REM sleep was identified with flat EMG and a high theta (5-10 Hz) to delta ratio in the EEG. EEG power spectra were calculated using a fast Fourier transform (FFT) for each 10-sec epoch from 0.5 to 30 Hz in 0.5-Hz bins (512-point, Hanning window), averaging across behavioral states (SleepSign). Average wake, NREM, and REM sleep, as well as the number, length, and transitions between sleep-wake episodes, were calculated over 24 h as well as in 12-h phases in light or dark. To quantify the dynamics of sleep-wake transitions, we examined the temporal correlations in the time series of sleep-wake stages, which were formed by assigning 0 (for sleep state) or 1 (for wake state) to each epoch. This was used to create a time series with a sampling rate of 10 sec, to which we applied detrended fluctuation analysis (see below).
To monitor locomotion, animals were placed into individual cages with under a 12:12 light-dark cycle for 6 weeks. Locomotor activity was monitored by an acquisition system that continuously received inputs from 2 infrared sensors (Panasonic EW, Panasonic, Newark, NJ) placed above each cage. The resulting data were sampled at 10 Hz and integrated over 60-sec epochs for further analysis. To assess activity patterns, we investigated 2 complementary dynamic properties in activity fluctuations: rhythms at discrete time scales and temporal correlations at different time scales from ~6 min to 10 h. To assess the rhythmicity in locomotor activity, we performed the power spectral analysis on each activity recording by using FFT. The powers from 1 to 24 h were extracted to quantify the rhythmicities at these discrete time scales.
To estimate temporal correlations in the fluctuations at different time scales, we used a widely accepted analytical tool—namely, detrended fluctuation analysis (DFA). This method quantifies the fluctuation amplitude, F(n), of activity fluctuations at different time scales n. As shown in many previous studies (Hu et al., 2004, 2007, 2009, 2012; Hsieh et al., 2014; Gu et al., 2015), locomotor activity fluctuations under normal healthy conditions display fractal structure with similar temporal correlations at different time scales that can be characterized by a power-law form of F(n), that is, F(n)~nα. The parameter α, called the scaling exponent, quantifies the correlation property in the signal: if α = 0.5, there is no correlation in the fluctuations (“white noise”); if α > 0.5, there are positive correlations, where large activity values are more likely to be followed by large activity values (and vice versa for small activity values). The exponent α = 1.0 indicates highest complexity in the control systems. Similar α values (~1.0) have been observed in many normal physiological outputs. Larger α (>1) indicates more correlations but more excessive regularity (less complexity) in the fluctuations. For motor activity, disrupted fractal activity control often leads to distinct patterns over 2 time scale regions. The boundary of the 2 regions is ~4 for rats (Hu et al., 2007; Hsieh et al., 2014). Thus, we calculated the scaling exponent over 2 regions for each data set—0.2 to 1 h (α1) and 4 to 12 h (α2)—while ignoring the transition region between 1 and 4 h. For sleep-wake stages, we calculated the scaling exponent only at time scales <1 h because the sleep recordings were too short to reliably estimate the fluctuation amplitude at time scales >4 h.
The sleep-wake parameters were analyzed using unpaired t tests (AI lesions vs. saline-injected controls) at a significance threshold of p < 0.05. To test the effects of lesion, light, and their interactions on the dynamics of sleep-wake transitions, we performed a mixed-model analysis of variance (ANOVA) with the scaling exponent α as the response and subject as a random effect for intercept. The statistical package JMP Pro version 11 (SAS Institute, Cary, NC) was used for all statistical tests.
After all recordings were completed, animals were anesthetized with 7% chloral hydrate (i.p. 500 mg/kg; Sigma, St. Louis, MO) and perfused with 10% buffered formalin (Fisher Scientific, Pittsburgh, PA) intracardially. Brains were removed and moved to 20% sucrose and phosphate-buffered saline (PBS) with 0.02% sodium azide overnight. Brains were sliced into 4 series of 40-μm sections using a freezing microtome, and these sections were stored in PBS–0.02% sodium azide at 20 °C. Tissue was washed with PBS, mounted onto slides, and stained with 0.1% thionin (Sigma). Slides were dehydrated, placed in xylene, cover-slipped, and imaged. All images were processed using the programs GIMP and Inkscape.
Injections of ibotenic acid into the AI (n = 8) resulted in confined lesions in all animals compared to saline-injected controls (n = 8) (Figure 1A). Lesions extended at the anterior level rostral to the anterior tip of the striatum and caudally several millimeters (Figure 1B). Lesions did not extend into the primary motor or somatosensory cortex. At more caudal levels, the lesion involved the piriform cortex and, to a smaller extent, some nearby structures such as the lateral orbitofrontal cortex or striatum. Animals in both lesion and control groups appeared healthy and showed no gross behavioral abnormalities. After recovery from surgery, sleep-wake polysomnography was recorded for 24 h. To assess the effects of AI lesion on sleep, we analyzed the scored polysomnography data and applied mixed models examining group and light-dark phase effects on wake, NREM sleep, and REM sleep.
For wake, animals in both groups had significantly more wake time during the active dark phase than the light phase, F(1, 14) = 311.06, t(14) = 17.64, p < 0.001. Over a 24-h phase, however, AI animals spent 13% less time awake, F(1, 14) = 12.72, t(14) = 3.53, p = 0.001 (Figure 2A). While the full model did not show a significant interaction between group and light-dark phase, F(1, 14) = 1.11, p > 0.05, examining the dark phase revealed the pronounced difference in wakefulness during the active dark phase, t(14) = 3.24, p = 0.03.
For NREM sleep, there was a significant effect of light-dark phase, F(1, 14) = 518.44, p < 0.001; animals in both groups had significantly more NREM sleep during the inactive light phase, t(14) = 22.77, p < 0.001. In addition, there was a significant effect of lesion, F(1, 14) = 12.19, p = 0.002 (Figure 2B); AI-lesioned animals had 10% more NREM sleep time, t(14) = 3.49, p = 0.002, consistent with the decrease in wakefulness. The full model also showed a significant interaction of group and light-dark phase, F(1, 14) = 6.43, p = 0.017. Specifically, during the dark phase, AI-lesioned animals had 27% more NREM sleep than control animals, t(14) = 4.26, p < 0.001. During the light phase, however, there was no significant difference between the groups, t(14) = 0.68, p > 0.05. REM sleep, like NREM sleep, was concentrated in the inactive light phase, F(1, 14) = 22.87, t(14) = 4.78, p < 0.001. In a 24-h period, AI-lesioned animals had 33% more REM sleep compared to controls, F(1, 14) = 7.36, t(14) = 2.71, p = 0.011 (Figure 2C). There was no significant interaction between the effects of group and light-dark phase on REM sleep time, although the differences between groups were pronounced during the REM sleep-enriched light phase, t(14) = 2.56, p = 0.016.
Changes in the total sleep or wake time may arise from changes in the number of sleep-wake episodes. A mixed model of the number of wake episodes showed a significant effect of light-dark phase on the number of wake episodes, specifically more wake episodes during the light phase, F(1, 14) = 141.60, t(14) = 12.23, p < 0.001 (Figure 2D). A model of NREM sleep episodes showed a corresponding increase in sleep episodes during the light phase, F(1, 14) = 135.10, p < 0.001. There were no significant effects of lesion group on the number of wake or NREM episodes (Figure 2E). Like wake and NREM sleep, the number of REM sleep bouts was light-dark phase dependent, F(1, 14) = 14.07, with more bouts during the light phase, t(14) = 3.75, p < 0.001. However, the model also revealed a significant effect of lesion, F(1, 14) = 8.46, p = 0.007. AI-lesioned animals had 35% more REM bouts. There was no significant interaction between group and light-dark phase, F(1, 14) = 1.23, p > 0.05, but examining the REM-enriched light phase revealed a 42% increase in REM bouts in AI-lesioned animals, t(14) = 2.84, p = 0.008 (Figure 2F). Plots of wake, NREM, and REM sleep (Figure 2G,I,K) show distributed changes across the states, along with the average FFT power spectra of the EEG during these states (Figure 2H,J,L). While the power spectra was similar between groups, AI animals had a distinctly elevated theta band (5-10 Hz) power in wakefulness, t(14) = 3.89, p = 0.002, and NREM sleep, t(14) = 3.19, p = 0.007, with a nonsignificant elevated theta in REM sleep. The power spectra also featured a reduced delta band (0.5-4.5 Hz) activity power in both wakefulness, t(14) = 3.44, p = 0.004, and NREM sleep, t(14) = 3.64, p = 0.003.
We next examined the average duration of each episode, or bout, of wake or sleep (Figure 3). A model of the average wake bout duration revealed significant effects of light-dark phase, F(1, 14) = 152.86, p < 0.001, lesion group, F(1, 14) = 7.14, p = 0.012, and a significant interaction between these 2 factors, F(1, 14) = 4.53, p = 0.042 (Figure 3A,B). Specifically, wake bout durations were much longer during the active dark phase, t(14) = 12.36, p < 0.001. In a 24-h period, AI-lesioned animals had 21% shorter average wake bouts, t(14) = 2.67, p = 0.012. This effect was pronounced during the dark phase, t(14) = 3.40, p = 0.002, but not during the inactive light phase, t(14) = 0.38, p > 0.05. NREM sleep bouts were longer during the inactive light phase across all animals, F(1, 14) = 12.65, t(14) = 3.56, p = 0.001. AI-lesioned animals had 10% longer NREM bouts than control animals, F(1, 14) = 4.94, t(14) = 2.22, p = 0.035 (Figure 3C,D). There was no significant interaction between these factors, F(1, 14) = 0.38, p > 0.05, although examination of the inactive dark phase revealed longer NREM bouts in AI-lesioned animals, t(14) = 2.08, p = 0.047. REM sleep bout duration was consistent across both groups, F(1, 14) = 0.01, and light-dark phase, F(1, 14) = 2.38, p > 0.05. Histograms of the frequency of bout duration lengths (Figure 3B,D,F) show mixed changes across wake bout durations, an increase in the number of longer NREM bouts, and no changes to the distribution of REM bouts (Figure 3E). For example, while AI-lesioned animals had increased frequency of REM bouts across many bout durations (Figure 3F), the cumulative percentage of REM bouts was unchanged compared to controls, whereas the cumulative frequency of shorter NREM bouts was higher in control animals than AI-lesioned animals.
A mixed model of wake to NREM sleep transitions showed a significant effect of light-dark phase, F(1, 14) = 134.55, p < 0.001, specifically more transitions during the light phase than the dark, t(14) = 11.60, p < 0.001, but no significant effect of AI lesions (Figure 4A). A model of NREM to wake transitions also showed significant effects of light-dark phase, F(1, 14) = 81.01, p < 0.001 (Figure 4B). However, for NREM to REM sleep transitions, we found significant effects of both light-dark phase, F(1, 14) = 13.25, p = 0.001, and lesion group, F(1, 14) = 8.89, p = 0.006. Specifically, there were more NREM to REM transitions during the inactive light phase, t(14) = 3.64, p = 0.001, but AI-lesioned animals also had 35% more NREM to REM transitions than control animals, t(14) = 2.98, p = 0.006. There was no significant interaction between these group and light-dark phase, F(1, 14) = 1.28, p > 0.05, but examining the REM-enriched light phase alone confirmed the significant increase in NREM to REM transitions, t(14) = 2.91, p = 0.007, mirroring the increase in the number of REM bouts as expected (Figure 4C, Figure 2C).
In summary, AI-lesioned animals have more NREM and REM sleep and less wakefulness than control animals. The decrease in wakefulness, evident during the active dark phase, was due to shorter bout durations; the increase in NREM sleep was due to longer bout durations. The increase in REM sleep was due to the increased number of REM episodes, especially during the light phase, but not a change in REM bout duration. AI-lesioned animals had more NREM to REM transitions, including during the inactive light phase, consistent with the observed increase in number of REM bouts. Example hypnograms during the dark (Figure 4D,E) and light (Figure 4F,G) phases from control animals (Figure 4D,F) and AI-lesioned animals (Figure 4E,G) illustrate these differences.
Furthermore, the fractal analysis also revealed significant effects of AI lesion and light-dark condition on the temporal correlations in the time series of sleep and wakefulness. Specifically, at time scales less than 1 h, the scaling exponent, α, was larger in the AI-lesioned animals, t(14) = 2.32, p = 0.036. Note that the α value in AI-lesioned animals was much greater than 1 (lesion: α = 1.11; control: α = 1.04), suggesting that AI lesion led to stronger correlations but more excessive regularity (less complexity) in the sleep-wake transitions. The mixed model also showed that α was much larger during the dark phase compared to the light phase, that is, t(14) = 11.83, p < 0.0001. No interaction was observed between the effects of group and light-dark condition (p > 0.9). All these sleep-wake results consistently suggest that AI lesions disrupt the organization of activity states, particularly in transitions between sleep states and the regulation of wakefulness.
The relatively short duration of the sleep recordings constrains the time scale windows that can be examined. To examine longer time windows, we recorded infrared-detected movement from individually housed animals with AI lesions (n = 7) and saline-injected controls (n = 7) for 6 weeks under 12:12 light-dark conditions as before. Representative patterns of activity show time scale–invariant properties or self-similar properties across multiple time scales (Figure 5A). To characterize the rhythms in activity, we conducted an FFT and extracted the amplitude from 1 to 24 h (Figure 5B). This analysis revealed a peak at 24 h in both AI-lesioned and control groups, confirming intact 24-h rhythms. Furthermore, ultradian rhythms remained intact, as illustrated by FFT peaks at 6, 12, etc., hours in both control and AI-lesioned groups. However, AI-lesioned animals had a reduced peaks at ultradian time scales, including a significantly reduced 12-h FFT peak compared to controls, t(8) = 2.34, p = 0.048, suggesting some degradation in the patterns of activity at that time scale.
The behavioral patterns in rats exhibit long-range temporal correlations (Hu et al., 2007; Hsieh et al., 2014). To better characterize the chronic disruption in activity caused by AI lesions, we performed a detrended fluctuation analysis (DFA) and examined the correlation property in activity recordings in 2 regions of time scales: region I (0.2-1 h) and region II (4-14 h). The degrees of the correlations are characterized by 2 scaling exponents: α1 (region I) and α2 (region II). Representative plots of the calculated DFA functions, plotted against time, show significant differences in fluctuation functions between control and lesion animals (Figure 5C). Examining the scaling exponents, we found that AI lesions had a significant reduction in α2 (t(8) = 3.47, p = 0.008); that is, α2 was closer to 0.5, indicating more random activity fluctuations at time scales between 4 and 14 h (Figure 5D). However, there was no group difference in α1 (t(8) = 0.99, p > 0.05).
In the present study, we examined the role of the AI in regulating sleep-wake states and chronic patterns of locomotor activity using polysomnography and infrared recordings, respectively. Both recordings show that AI-lesioned animals have disturbances in the organization of sleep, wake, and activity. AI-lesioned animals have more sleep, have less sustained wake, and are more likely to transition from NREM to REM sleep, especially during the inactive light phase. These animals also have increased theta power and decreased delta power in wakefulness and NREM sleep. Chronic infrared recordings show intact 24-h rhythms in AI-lesioned animals under 12:12 LD conditions but a reduction of 12-h FFT activity peaks and significantly reduced fractal patterns of activity organization from 4 to 12 h. Taken together, these results suggest that the cortex, and specifically the insula, plays a critical role in organizing the timing of sleep-wake states and locomotor behavior. Lesions of the insula have a greater effect on sleep-wake amounts than brainstem lesions to regions implicated in sleep-wake regulation, including the median and dorsal raphe nuclei, the laterodorsal tegmentum, and the locus coeruleus (Lu et al., 2006).
The AI is a key site in processing autonomic inputs, especially visceral (Cechetto and Saper, 1987; Bagaev and Aleksandrov, 2006) and gustatory afferents via the parabrachial nucleus (Yamamoto et al., 1980; Kosar et al., 1986). The role of the AI in both autonomic and sleep-wake behavior suggests possible mechanisms by which autonomic information and sleep-wake behavior are related. These mechanisms may depend on the inputs and outputs of the insula region. For example, AI neurons may receive and process visceral and gustatory inputs and in turn influence overall behavioral state via arousal-promoting circuits. The lesions here include some of the largely agranular and dysgranular subregions that receive visceral inputs from taste information, with sparser representation of gastric mechanoreception, inspiratory baroreception, and CO2 (Cechetto and Saper, 1987), while also receiving projections from thalamic nuclei that are relays from the parabrachial nucleus. However, the main lesion regions are the rostral agranular insular cortex that largely does not respond to visceral stimuli but rather receives direct (i.e., not relayed through thalamic nuclei) parabrachial projections. This region also receives input from nearby insular regions that process numerous visceral stimuli, thus acting as a site of converging multimodal information regarding bodily state (Shi and Cassell, 1998). The rostral agranular region targeted in this study may integrate visceral information from the animal and regulate sleep-wake states to tune the behavioral response to food, water, temperature, and other physiological needs (Saper, 2002).
In contrast to studies of autonomic or visceral processing, a direct role of the insula in sleep-wake states is largely unexplored. Strokes affecting the insula can result in tiredness and lack of energy or activity (Manes et al., 1999), and coactivation of the AI with salience networks has been implicated in insomnia (Chen et al., 2014). Imaging and EEG source modeling studies raise the intriguing possibility that the AI may be directly involved in the electrophysiological properties of sleep, including NREM sleep slow waves (Braun et al., 1997; Murphy et al., 2009) and spindles (Schabus et al., 2007). The present findings reveal a novel role for the insula in direct sleep-wake control, including a previously unknown role in regulating REM sleep. The increased network coactivation of the AI in insomnia dovetails with the wake-promoting, sleep-suppressing role of the AI suggested in the present study. The AI has projections throughout the neuraxis that may enable circuit-level control of sleep: promoting and stabilizing wakefulness while suppressing NREM and REM sleep. While AI lesions might directly affect sleep-wake via intrinsic cortical sleep-regulatory cells (Morairty et al., 2013), lesions to the AI are small relative to the size of the entire cortex.
More likely, the AI regulates sleep, wakefulness, and patterns of behavior via efferent projections to other structures. The AI has several subcortical projections to regions implicated in direct sleep-wake control. For example, the AI may promote wakefulness via excitatory projections to brainstem regions such as the parabrachial nucleus and other potential regions involved in wakefulness, including the posterior lateral hypothalamus, amygdala, and mediodorsal thalamus (Saper, 1982; Allen et al., 1991). Consistent with this, lesions of the AI increase sleep and decrease wakefulness, although it is unclear if these changes are due to altered control of one or more of these subcortical structures. Similarly, projections to REM-regulating regions such as the ventrolateral periaqueductal gray (Saper, 1982; Yasui et al., 1991; Lu et al., 2006) or lateral hypothalamus may play a role in suppressing REM sleep. Lesions of the AI may disinhibit a NREM to REM switch, allowing more frequent transitions to REM and more REM bouts. Intriguingly, the changes in gross sleep and wake times following AI lesions resemble changes following lesions to the medial parabrachial nucleus (Lu et al., 2006). The role of other cortical areas, including nearby areas and areas targeted by AI projections, is unclear, but lesions to other cortical regions have different effects on sleep-wake behavior than lesions to the AI. For example, lesions to the medial prefrontal cortex increase REM sleep (Chang et al., 2014), while lesions of the anterior cingulate cortex decrease the wake-promoting response to novelty (Gompf et al., 2010). The heterogeneous regions of the cortex may have different roles in organizing behavior, depending on their circuitry and their response to different environments.
Beyond sleep-wake control, the AI contributes to the organization of activity over a 24-h period. In the seemingly irregular fluctuations of activity signals, there are robust, nonrandom underlying patterns, including long-range correlations of activity that persist over a wide range of time scales. This temporal correlation can be characterized by DFA, as the average fluctuation amplitude changes at different time scales following a power law distribution. This power law property implies the fluctuation is invariant in temporal domains, much as fractal patterns are invariant in spatial domains at different magnifications. The neural basis of these fractal-like self-similar properties is unknown, but these scale-invariant patterns are seen in humans as well as in rodents (Hu et al., 2007). However, one possibility is that multiple interacting oscillators within the brain form functional networks to control activity and produce fractal properties in signals of activity (Hu et al., 2007; Pittman-Polletta et al., 2013). Lesions to the AI disrupt these temporal properties, and both FFT and DFA analyses suggest degradation in ultradian rhythms but preservation of 24-h rhythms.
These temporal properties may help an animal sustain behavioral arousal by stabilizing wakefulness. Consistent with this interpretation, AI-lesioned rats have reduced wake bout duration. Interestingly, DFA analysis of sleep-wake transition data reveals increased regularity of transition structure at smaller time scales, with degraded fractal-like properties at longer time scales. This difference may reflect different measurements (EEG-based sleep-wake data vs. infrared-based movement) or different levels of regulation (short time-scale patterns vs. long time-scale patterns). Interestingly, lesions of the AI also shift the EEG, especially during NREM sleep and wakefulness, toward REM sleep. This may reflect the increased REM sleep drive in these animals, and the overall NREM EEG may contain more transition states between NREM and REM sleep, reflected by reduced delta power and increased theta power. The AI has also been implicated in the generation of low-frequency oscillations in sleep (Braun et al., 1997; Murphy et al., 2009), and the reduced delta and increased theta may reflect the loss of a cortical slow-wave generator.
Both FFT and DFA analyses implicate the AI in control of ultradian patterns of behavior, as AI lesions did not affect circadian patterns. Lesions of the master clock (SCN) drastically alter rhythms at both 24 h and at harmonic frequencies of 24 h, including the complete loss of correlations above time scales of 4 h (i.e., α2 ~0.5) (Hu et al., 2007). AI lesions, on the other hand, are limited in their effects across multiple time scales, with a peak at 12 h. AI lesions did not perturb the 24-h rhythm, although we did not test AI lesions in constant dark conditions, which is necessary to determine the extent of which the changes in these rhythms are dependent on external cues. Even in the frequencies with alterations due to the AI lesions, correlations partially remained in animals with AI lesions (i.e., α2 >0.5; Fig. 5C,D). We propose that correlation structures at different time scales are governed by different sets of neural circuits; the interaction of these circuits generates predictable and organized patterns of behavior. In this model, the SCN has a strong role coordinating and synchronizing these circuits, while individual circuits have a more limited role in modulating time-scale rhythms. Interestingly, while mapping sleep transitions using DFA analysis revealed group differences at smaller time scales, using motion-based sensors did not. An intriguing possibility is that the overall behavioral state (sleep vs. wake) and the extent of activity may both have intrinsic temporally invariant organization patterns. How this organization interacts with the sleep homeostat, such as in the response to sleep deprivation, will be an important future avenue of investigation.
The AI contributes to the organization of sleep, wakefulness, and activity, particularly at subcircadian time scales, but it is unclear if this is as an independent generator or interacting with suprachiasmatic control. The AI, and the cortex in general, may regulate sleep or locomotor activity using separate sets of circuits, or a common set of networked oscillators may influence a range of behaviors. The wide range of inputs and outputs of the AI, including connectivity with powerful wake-promoting regions such as the parabrachial nucleus (Saper, 1982; Allen et al., 1991), suggests a possible role for the AI in the oscillatory networks that regulate multiscale regulatory patterns of activity. Notably, these patterns are disrupted in neurodegenerative diseases, such as Alzheimer disease, with increasing disease severity leading to increased disruption in activity patterns (Hu et al., 2009, 2013) and sleep (Ju et al., 2014). The insula, like many other cortical regions, shows signs of degeneration in Alzheimer disease (Rombouts et al., 2000). The disruptions in fractal activity patterns and sleep in neurodegenerative diseases may stem from the loss of one or more of the oscillators such as the AI.
The authors acknowledge the expert technical assistance of Quan Ha and Xi Chen, as well as the contributions of acquisition hardware and software from Men-Tzung Lo and Yi-Chung Zhang. This work was funded by the Hilda and Preston Davis Foundation (M.C.C.) and National Institutes of Health (R00HL102241 and R01AG048108 to K.H.; NS062727 and NS061841 to J.L.).