- Top of page
- TEXT OF REVIEW
- CONFLICT OF INTERESTS
Tumor necrosis factor alpha (TNF), interleukin-1 beta (IL1), and other cytokines are involved in non-rapid eye movement sleep (NREM) regulation under physiological and inflammatory conditions. Brain levels of IL1 and TNF increase with prolonged wakefulness. Injection of exogenous IL1 or TNF, mimicking sleep loss, induces sleepiness, excess sleep, fatigue, poor cognition, and enhanced sensitivity to pain. These symptoms characterize the syndrome associated with sleep loss. Extracellular ATP released during neuro- and glio-transmission, acting via purine P2 receptors on glia, releases IL1 and TNF. This extracellular ATP mechanism may provide an index of activity used by the brain to keep track of prior wakefulness. Prolonged wakefulness is associated with enhanced neuronal activity. TNF and IL1, in turn, act on neurons to change their intrinsic properties and sensitivities to neurotransmitters and neuromodulators such as adenosine and glutamate. Such actions change network input–output properties (i.e. state shift). State oscillations, for instance, occur within cortical columns and are responsive to TNF. Sleep is thus viewed as a local use-dependent process regulated in part by cytokines. Further, state oscillations are viewed as a fundamental process of any neuronal/glia network. To investigate these hypotheses we developed an in vitro neuronal/glia culture system exhibiting field potential oscillations and have mathematically modeled the local use-dependent view of sleep initiation. These views have profound implications for sleep pathologies and function.
TEXT OF REVIEW
- Top of page
- TEXT OF REVIEW
- CONFLICT OF INTERESTS
Cytokines such as tumor necrosis factor alpha (TNF) play a role in sleep regulation in health and disease.1,2 Hypothalamic and cerebral cortical levels of TNF mRNA or TNF protein have diurnal variations (2- and 10-fold, respectively) with higher levels associated with greater sleep propensity. Sleep loss is associated with enhanced brain TNF. Central or systemic TNF injection enhances sleep. Inhibition of TNF using the soluble TNF receptor, or anti-TNF antibodies, or a TNF siRNA reduces spontaneous sleep. Mice lacking the TNF 55 kD receptor experience less spontaneous sleep.3,4 Injection of TNF into sleep-regulatory circuits, for example the hypothalamus, promotes sleep. In normal humans, plasma levels of TNF co-vary with EEG slow-wave activity (SWA) and in multiple disease states plasma TNF increases in parallel with sleep propensity. Injection of exogenous TNF, thereby mimicking the enhanced levels of TNF observed with sleep loss, induces many of the symptoms associated with the syndrome induced by sleep loss, e.g. sleepiness, rebound sleep, fatigue, poor cognition and enhanced sensitivity to pain.
Downstream mechanisms of TNF-enhanced sleep include nitric oxide, adenosine, prostaglandins and activation of nuclear factor kappa B (NFkB).1,2 Neuronal use induces cortical neurons to express TNF (like Fos) and if applied directly to cortical columns TNF induces a functional sleep-like state within the column as determined from surface-evoked potentials5 (see below). Glutamate also induces TNF in neurons6 and TNF mechanistically promotes synaptic scaling.7 Unilateral application of TNF to the surface of the somatosensory cortex induces state-dependent enhancement of EEG SWA ipsilaterally, suggesting that regional sleep intensity is enhanced.8 Similar state-dependent enhancements of EEG SWA are also observed regionally after disproportionate stimulation of localized areas of the cortex, whether this is accomplished by learning paradigms or by afferent stimulation.1,2
Such data suggest that sleep is a local use-dependent process.1,2 However, how the brain cumulatively tracks prior activity has remained uninvestigated. We propose that a biochemical network operating in both neurons and glia, composed in part of extracellular ATP-induced release of TNF and other cytokines such as IL1 and brain-derived neurotrophic factor (BDNF) from glia, is responsible for regulating state oscillations within local networks. A key step in this cascading biochemical network is ATP co-released with neurotransmitters and released from glia in response to cell activity into the extracellular space. Because extracellular ATP levels are linked to cellular activity and ATP has a relatively short half-life, its local levels will dynamically change (in a fashion analogous to neurotransmitter spatial and temporal summation). The consequent extracellular ATP interacts with purine type 2 receptors (P2R) located on glia and other cells to release TNF, IL1 and BDNF. This step provides a translation of rapid dynamic activity-linked extracellular ATP levels into levels of longer-lived sleep-regulatory substances (SRSs). These cytokines interact with the post-synaptic neuron (possibly with the pre-synaptic neuron as well) to activate NFkB. NFkB is an enhancer element that promotes the production of receptors such as the adenosine A1 receptor and the gluR1 component of the AMPA receptor. Changed expression of these receptors on the neuron will change its sensitivity to adenosine and glutamate, respectively. Thus the sensitivity, or gain, of the post-synaptic neuron is scaled to the activity in the pre-synaptic neuron. Such processes are thought to be involved in long-term plasticity mechanisms and a role for TNF in synaptic scaling is well characterized. These processes also link the very rapid events occurring during neurotransmission to events involved in the longer-term regulation of sleep and neuronal sensitivity to various effector molecules. Because the sensitivity of the post-synaptic neuron is changed by the proposed mechanism, the same input will result in a different output signal from the network within which the scaling occurred. By definition this is a state change.
If neurotransmission-associated ATP release is involved in keeping track of past brain activity, then ATP agonists or antagonists should alter sleep. Preliminary data from our laboratory suggest that the ATP agonist, BzATP, enhances non-rapid eye movement sleep (NREMS) while the ATP antagonist, OxATP, inhibits NREMS.9 Further, other preliminary data indicate that P2R mRNAs (both P2X7 and P2Y1) have a diurnal rhythm in the brain and are altered by sleep loss, IL1 and TNF injection.10 Mice lacking the P2X7 receptor have attenuated sleep-rebound responses after sleep deprivation. Moreover, preliminary data from our laboratory indicate that mice lacking the ectonucleotidase, CD73, involved in ATP catabolism to adenosine, have excessive spontaneous NREMS. Such data suggest that the mechanism by which the brain keeps long-term track of prior activity (states) involves neurotransmission-released ATP that in turn releases brain cytokines and their subsequent actions on receptor expression and consequent cell sensitivity to excitatory and inhibitory signals. This also strongly suggests, because the ATP-neurotransmission mechanism is a local event involving autocrine and paracrine signaling within the neuronal assembly where the neurotransmission took place, that sleep is initiated locally and is, thus, fundamentally a local process.
There is direct evidence that local neuronal networks exhibit state oscillations. Data from the developmental and memory literatures suggest that local sleep is use dependent. Thus experimental interventions ranging from whisker stimulation in rats to unilateral somatosensory stimulation, arm immobilization, adroit learning paradigms in humans,1,2 and selective sensory deprivation of neonates indicate that localized changes in sleep EEG delta power or blood flow are enhanced if, during prior waking, the areas were activated.
Cortical columns are an example of a local network in brain capable of state oscillations. Cortical columns are densely interconnected neuronal assemblies thought to be the basic unit of information processing. Individual somatosensory cortical columns (whisker barrels) exhibit evoked response potentials (ERPs) that are a measure of input/output relationships. ERPs are larger in magnitude during sleep than during wakefulness. Individual columns can show characteristic high-amplitude ERPs during sleep while neighboring columns exhibit wake-like low-amplitude ERPs. Conversely, individual cortical columns can display wake-like states while neighboring columns show sleep-like responses when the rat is functionally and behaviorally asleep.11 These data suggest that cortical-column local state is to a degree autonomous from other columns although local column state usually corresponds with whole-animal state. The local sleep state occurs more frequently when the cortical column is stimulated intensively thereby exhibiting a use-dependency of state. The cortical columns are thus good examples of the basic unit of brain circuitry involved in sleep and sleep regulation. Finally, if TNF is applied locally to the cortex it enhances ERP amplitudes, suggesting that it induces the local sleep-like states in cortical columns.5
Cortical-column state has an impact on overt behavior in rats. In an experimental learning paradigm dependent upon sensory stimulation of a single whisker, if the corresponding whisker barrel was in the wake-like state, correct behavioral responses to whisker twitching were elicited. However, when the cortical column was in the sleep-like state, the rat made errors.
The top-down paradigm of sleep regulation (i.e. the imposition of sleep by so-called sleep centers such as the ventrolateral pre-optic hypothalamic area) requires intentional action from the specialized sleep/wake regulatory brain circuits to initiate and terminate whole-organism sleep. This raises unresolved questions as to how such purposeful action might itself be initiated. Our new paradigm of local, use-dependent sleep regulation avoids such infinite regresses. Within our paradigm, local sleep is a direct consequence of prior local neuro- and glio-transmission and whole-organism sleep is a bottom-up, self-organizing emergent property of the collective states of cortical columns throughout the brain.12 A role remains for specialized sleep centers and pathways, as they coordinate cortical columns to synchronize into single whole-brain vigilance states and coordinate overall brain states with other physiological systems in the body for niche adaptation. However, in our view, sleep initiation resides in the individual cortical columns and other neuronal assemblies.
The local, use-dependent paradigm for sleep that we are pursuing is fundamentally difficult to study in vivo, in that it requires comprehensive measurement at multiple precise locations within the brain. To supplement in vivo experimentation, we are developing cell-culture and analytical models for network state oscillations that hold promise for advancing understanding and prediction of sleep.
Our cell-culture model of network state oscillations is built on field potential (FP) measurements from mixed cultures of dissociated neurons and glial cells, recorded simultaneously at multiple network nodes. Multi-electrode array (MEA) culture dishes are used for multi-node data acquisition from dissociated cultures. MEA culture dishes are specially fabricated cell culture containers with multiple tiny electrodes protruding from the bottom of the dish. Somatosensory cortex from embryonic day 18 (E-18) rats were dissected, dissociated with enzymes and then grown in a cell culture incubator on MEA culture dishes. E-18 neurons form spontaneous connections in a few days forming a complicated in vitro network structure, and stay electrically active in culture for months13 (Fig. 1).
Figure 1. FP recording from a neuronal culture grown on an MEA culture dish shows oscillations. Neuronal cultures are grown on MEAs for multi-point electrical recordings. Fluorescence images of neurons on the MEA (a) recorded on day 7 in vitro show an intricate network structure with many connections. Black arrow in (a) points to an electrode used for the FP measurement. One such FP recording, filtered for 0.5 to 10 Hz, is shown in (b) (blue trace, total duration 125 min). Waxing and waning of the FP is obvious in the recording indicating a temporal dynamic behavior of the culture. Power spectra (c) are calculated from the data in (b) using analysis routines written in MATLAB. Power spectra (presented as an average over 5 min) also show temporal variation. The power spectra graph is color coded for different wave regions: 0.5 to 4 Hz (cyan to yellow), 4 to 7 Hz (orange), and >7 Hz (red). If no cells are grown on the MEA, electrical recordings are flat.
Download figure to PowerPoint
The new paradigm of local, use-dependent sleep regulation hypothesizes sleep as a fundamental property of any group of viable neurons forming a network. Any region of the network (node) oscillates between two states at a particular time: a sleep-like state and a wake-like state. The phase and modulation of this oscillation depend on the prior activity level of that network node. Synchronization of network state oscillations between multiple nodes determines the global network state of the culture.
Our data from this cell culture model system demonstrate that (i) state oscillations occur in dissociated cultures, (ii) activity in the slow-wave range shows diurnal variation, and (iii) two nodes in the network can go in and out of synchronization. Further sleep-related studies are underway that will demonstrate the effect of cytokines and other SRSs on network state oscillation and state synchronization using this in vitro model of sleep.
We are also pursuing analytical modeling of use-dependent networked sleep regulation, with the aim of developing a keen yet inexpensive tool for predicting sleep-state oscillations. To this end, we have developed a mathematical model for the network of neuronal groups governing sleep/wake states and performance.12 Global sleep regulation has been extensively modeled, as have the dynamics of neuronal groups or local networks or assemblies.14,15 Broadly, our work advances these efforts by making explicit that local activity (i.e. use) and network coupling among multiple assemblies guides the emergence of a global sleep/wake homeostatic state, and by representing in more detail some of the underlying mechanisms for sleep/wake regulation. Specifically, we have modeled the sleep/wake homeostasis at multiple time scales.12 Here, the shorter time scale (milliseconds to minutes) is concerned with the biochemical and bioelectrical mechanisms underlying sleep/wake homeostasis. These include the processes by which local activity, coupling of neighboring neuronal assemblies, and regulatory circuits modulate SRSs, as well as the mechanisms by which accumulated SRSs effect the functional changes associated with sleep–wake state changes. At the longer time scale (minutes to days), we view the neuronal assemblies throughout the brain as a network of activity integrators with associated functional states, which further interact through weak coupling and regulatory circuitry; we refer to this nonlinear model as the activity-integrator network. Our preliminary analyses of the activity-integrator network indicate, qualitatively, how the local dynamics of the neuronal assemblies can lead to the emergence of a global oscillating sleep/wake state through synchronization. We are currently advancing the mathematical modeling effort toward permitting quantitative prediction and design of sleep oscillation by pursuing network identification,16 uncertainty-modeling, and regulation-design tasks.
The posited sleep-regulatory mechanism results in the stabilization of cell sensitivity by changing receptor populations for inhibitory (adenosine) and excitatory (glutamate) molecules. This action occurs as a consequence of neuronal activity and is localized to the sites where activity-induced changes in synaptic efficacy and connectivity are occurring. Thus TNF and other SRSs are altered by activity, and in turn alter expression of the receptors involved in plasticity. Thus, our view of sleep mechanisms clearly links sleep and neural connectivity.
The sleep mechanisms outlined herein also provide insight into unconsciousness. During waking, input to individual neuronal assemblies induces environmentally adaptive outputs. With activation of networks, the consequent release of ATP and TNF would induce a new output in response to the same input. The new output would be different from the prior output that was adaptive and thus likely to be irrelevant to the environmentally driven input. It would be maladaptive if it manifested as cognitive or motor real-time events because behavior would not be coordinated in real-time to environmental inputs. There is thus an adaptive need to prevent the animal from behaving at such times. The local ATP–cytokine–adenosine sleep mechanism is consequently not only inseparable from the plasticity functions of sleep but they also provide the necessity for unconsciousness.