Waking and dreaming consciousness: neurobiological and functional considerations - PubMed Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Review
. 2012 Jul;98(1):82-98.
doi: 10.1016/j.pneurobio.2012.05.003. Epub 2012 May 15.

Waking and dreaming consciousness: neurobiological and functional considerations

J A Hobson  1 , K J Friston
Affiliations
Review

Waking and dreaming consciousness: neurobiological and functional considerations

J A Hobson et al. Prog Neurobiol. 2012 Jul.

Abstract

This paper presents a theoretical review of rapid eye movement sleep with a special focus on pontine-geniculate-occipital waves and what they might tell us about the functional anatomy of sleep and consciousness. In particular, we review established ideas about the nature and purpose of sleep in terms of protoconsciousness and free energy minimization. By combining these theoretical perspectives, we discover answers to some fundamental questions about sleep: for example, why is homeothermy suspended during sleep? Why is sleep necessary? Why are we not surprised by our dreams? What is the role of synaptic regression in sleep? The imperatives for sleep that emerge also allow us to speculate about the functional role of PGO waves and make some empirical predictions that can, in principle, be tested using recent advances in the modeling of electrophysiological data.

PubMed Disclaimer

Figures

Fig. 1
Fig. 1
(a) Standard sleep polygraphic measurements. These traces show 90−100 min cycles of rapid eye movement (REM) and non-rapid eye movement (NREM) sleep. The traces show cycles for three subjects, where the blue lines indicate periods of REM sleep. Reports of dreaming are most common from sleep onset stage I (when dreams tend to be fragmentary), late-night stage II (when dreams tend to be thought-like) and REM (when they tend to be long, vivid, hallucinatory and bizarre). Deep phases of sleep (III and IV) occur in the first half of the night, whereas lighter stages (stages I and II) predominate in the second half. (b) The states of waking and sleep. These states have behavioral, polygraphic and psychological correlates that appear to be orchestrated by a control system in the pontine brainstem. In this panel, the neuronal clock that controls these states is depicted as a reciprocal interaction between inhibitory aminergic neurons and excitatory cholinergic neurons: aminergic activity is highest during waking, declines during NREM sleep and is lowest during REM sleep; whereas cholinergic activity shows the reverse pattern. Changes in sleep phase occur whenever the two activity curves cross; these are also the times when major postural shifts occur. The motor immobility during sleep depends on two different mechanisms: disfacilitation during stages I–IV of NREM sleep and inhibition of motor systems during REM sleep. The motor inhibition during REM sleep prevents motor commands from being executed, so that we do not act out our dreams. (c) Human sleep and age. The preponderance of rapid eye movement (REM) sleep in the last trimester of pregnancy and the first year of life decreases progressively as waking time increases. Note that NREM sleep time, like waking time, increases after birth. Despite its early decline, REM sleep continues to occupy approximately 1.5 h per day throughout life. This suggests that its strongest contribution is during neurodevelopment but that it subsequently plays an indispensable role in adulthood. (d) The evolution of REM sleep. Birds and mammals evolved separately after branching off from the ancestral tree. Both birds and mammals are homoeothermic, and both have appreciable cognitive competence. With respect to the enhancement of cognitive skills by REM, it is significant that both birds and mammals are capable of problem solving and both can communicate verbally. (e) AIM model. This panel illustrates normal transitions within the AIM state-space from waking to NREM and then to REM sleep. The x-axis represents A (for activation), the y-axis represents M (for modulation) and the z-axis represents I (for input–output gating). Waking, NREM sleep and REM sleep occupy distinct loci in this space. Waking and REM sleep have high activation but different I and M values. Thus, in REM sleep, the brain is both off-line and chemically differentiated compared with the waking brain. NREM sleep is positioned in the centre of the space because it is intermediate in all quantitative respects between waking and REM sleep.
Fig. 2
Fig. 2
(a) PGO waves and their relation to REM sleep and eye-movements. (A): NREM-REM transition showing PGO waves in LGB (types I and II). During transition periods from NREM to REM sleep, biphasic (PGO) waves in LGB first appear as large single events (type I waves). Waves become clustered with decreasing amplitude (type II waves) as signs of REM sleep become more prominent: atonia (EMG), desynchronization of cortical EEG (Cx), hippocampal- (HIP), and REMs (EOG). (b) Side-to-side alternation of primary waves: Once a REM period is established, predominant PGO wave amplitudes alternate from one geniculate to the other, according to lateral direction of eye movements. When there is rightward movement of the eyes (EOG-R), the corresponding PGO wave cluster is larger in right LGB (dots) than in left. Conversely when there is leftward movement (EOG-L) waves are larger in left LGB (dots). (c) The neuronal firing of a PGO burst cell is shown in the top trace. The PGO waves of the ipsilateral and contralateral geniculate bodies are shown below. It can be seen that the ipsilateral PGO waves are larger in amplitude. In (d) the brain is schematically depicted to reveal eye movement direction. PGO waves form in the two geniculate bodies and PGO burst cell activity in the pons. When the eyes move ipsilaterally (left panel), the cell fires a cluster of spikes prior to the eye movement and prior to the PGO waves. When the eyes move in the opposite direction (right panel), the burst cell is silent and the contralateral PGO wave is twice the amplitude of its ipsilateral counterpart.
Fig. 3
Fig. 3
This is a schematic summarizing a generative model of sensory data as a probabilistic graphical model, where the arrows denote statistical dependencies. This is just a formal way of writing down various quantities in a model and how they depend on each other. A generative model can be regarded as a prescription of how to generate a virtual reality and specifies the sorts of quantities required: Some quantities are variables that depend upon time, whereas others specify the causal architecture of the model. These are usually considered to be real valued parameters θ∈ℝ of equations describing the motion of hidden states x(t)∈ℝ and the mapping from hidden states to sensory states s(t)∈ℝ (see Fig. 4). The text in the figure describes the nature of these quantities and how they may be encoded with biophysical or internal brain states. Hidden states correspond to states of the world that generate sensory data; for example, the motion of a visual object and the nature of ambient light that conspire to produce some visual impressions. The switching variables η∈{0,1} at the top can be regarded as priors or constraints on the parameters that determine whether a particular connection or causal dependency among states exists or not.
Fig. 4
Fig. 4
This schematic details a neuronal architecture that optimizes the conditional or posterior expectations about hidden variables in hierarchical models of sensory input of the sort illustrated in Fig. 3. These schemes are based on minimizing the free energy in Box 1 using a gradient descent and can be regarded as a generalization of predictive coding. The particular example here focuses on the PGO system: It shows the putative cells of origin of forward driving connections that convey prediction errors from a lower area to a higher area (red arrows) and nonlinear backward connections (black arrows) that construct predictions (Mumford, 1992; Friston, 2008). These predictions try to explain (cancel) prediction-error in lower levels. In these schemes, the sources of forward and backward connections are superficial and deep pyramidal cells (triangles), respectively, where units representing predictions and prediction error are drawn in black and red, respectively. If we assume that synaptic activity encodes posterior predictions about states, then perceptual inference can be formulated as a gradient descent on free energy: this provides the differential equations shown on the right. Under Gaussian assumptions, these posterior expectations can be expressed compactly in terms of precision weighted prediction-errors: (ξx(i),ξv(i)) on the motion of hidden states and causes at the ith level of the cortical hierarchy. Here, we have supplemented hidden states with hidden causes that, in hierarchical models, link hierarchical levels. The ensuing equations suggest two neuronal populations that exchange messages; with state-units (black) encoding conditional predictions (μ˜x(i),μ˜v(i)) and error-units (red) encoding prediction-error. In hierarchical models, error-units receive messages from the state-units in the same level and the level above; whereas state-units are driven by error-units in the same level and the level below. These provide bottom-up messages that drive conditional expectations towards better predictions to explain away prediction-error. Top-down predictions correspond to g(μ˜x(i),μ˜v(i),θ) and are specified by the generative model, while the dynamics of hidden states are described by the equations of motion f(μ˜x(i),μ˜v(i),θ). This scheme suggests the only connections that link levels are forward connections conveying prediction errors to state-units and reciprocal backward connections that mediate predictions. Note that the prediction errors that are passed forward are weighted by their conditional precisions, (μγ(i)), that we have associated with the activity of aminergic and cholinergic neuromodulatory systems. Technically, the scheme in this figure corresponds to generalized predictive coding because it is a function of generalized variables, which are denoted by a ∼ such that every variable is represented in generalized coordinates of motion: for example, μ˜=(μ,μ′,μ″,…). See (Friston, 2008) for further details. In this schematic, occipital cortex sends top-down predictions to visual cortex, which then projects to the lateral geniculate body. However, occipital cortex also sends proprioceptive predictions to the pontine nuclei, which are then passed to the oculomotor system to cause movement through classical reflexes. Predictions from the pontine nuclei are also passed to the lateral geniculate body. These predictions can be thought of as corollary discharge. Every top-down prediction is reciprocated with a bottom-up prediction error to ensure predictions are constrained by sensory information.
Fig. 5
Fig. 5
This figure shows how the reciprocal interactions between cholinergic and aminergic systems entrain perceptual processes during the sleep wake cycle. Aminergic modulation plays a critical role in gating or enabling precise sensory information to drive action and perception. When this modulatory gating suppresses sensory information, the brain's optimization processes are no longer informed by precise sensory prediction errors and change quantitatively. The implicit optimization processes during waking (upper panel) and sleep (lower panel) are based on free energy minimization, using the generative models described in the previous two figures. The free energy principle requires all internal brain states encoding posterior beliefs or conditional expectations to minimize free energy. The particular updates shown in this figure are based upon a standard variational Bayesian procedure (Beal, 2003) that allow us unpack the various processes involved. For every quantity in the generative model (see Fig. 3) there is a corresponding update. These updates are computed using conditional expectations denoted by EQ[⋅]. Here, L(s)=−ln⁡Pa(s|x,γ,θ) is sensory surprise and is just sensory prediction error times its precision. Sensory surprise is effectively turned off during sleep because the precision of sensory prediction errors is reduced. Neurobiologically, we assume this is mediated by circadian fluctuations in aminergic neurotransmission (pink circles). The text describes, briefly, the biophysical and neurobiological processes that can be associated with each of the updates. These are largely the same in sleep and wake, with the exception of action that depends exclusively on sensory surprise (that is absent during sleep). We have included priors governing the presence or absence of a particular connection in these updates. Minimizing the free energy of these priors is formally identical to model selection using the Savage-Dickey density ratio: see (Friston and Penny, 2011) for details. Effectively, this removes or prunes redundant model parameters (synaptic connections) to reduce model complexity and minimize free energy. The optimization of the parameters can be seen in the same light, because these effectively minimize the complexity of empirical priors on hidden states.
Fig. 6
Fig. 6
This figure illustrates, schematically, the functional anatomy of visually guided eye movements during waking (left) and REM sleep (right). The upper panels summarize the implicit functional differences in terms of active inference. During wakefulness, top-down predictions about the proprioceptive and exteroceptive consequences of eye movements are sent to pontine and visual centers, respectively. The former elicit eye movements through classical reflex arcs (to suppress proprioceptive prediction error), while the latter anticipate the changes in retinal input. In sleep, there is a selective loss of precision on visual prediction errors. All this means is that the brain thinks its predictions in the visual domain are perfect, because they do not need correcting. This allows for perception without sensation; that is, dreaming. The lower panels show the implicit functional anatomy based on previous figures: this uses a simplified network that comprises the lateral geniculate body (LGB), early visual or striate cortex, occipital cortex (that stands in for all high-level cortical areas) and the pontine nuclei controlling eye movements (the cranial nerve nuclei and paramedian pontine reticular formation). Each component of the network is drawn using the principal output cell populations; for example, superficial (dark red) and deep pyramidal cells (black) for forward and backward connections in the cortex Cholinergic (blue) and aminergic (pink) projections control the postsynaptic sensitivity of superficial pyramidal cells that report prediction error and send forward connections. Aminergic projections have been deployed here such that they selectively gate early visual cells in the lateral geniculate body and visual cortex. During sleep, these cells are effectively silenced (denoted by open triangles in the right panel), restricting dynamics to the pontine, geniculate and occipital structures. It is this restriction we associate with the difference between PGO waves in sleep and visually evoked responses associated with orienting saccades during wakefulness. As in Fig. 4, forward connections conveying prediction errors are shown in dark red, while backward connections from state-units that furnish predictions are shown in black. LC: Locus Coeruleus and NBM: Nucleus Basalis of Meynert.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Ashby W.R. Principles of the self-organizing dynamic system. Journal of General Psychology. 1947;37:125–128. - PubMed
    1. Aston-Jones G., Chiang C., Alexinsky T. Discharge of noradrenergic locus coeruleus neurons in behaving rats and monkeys suggests a role in vigilance. Progress in Brain Research. 1991;88:501–520. - PubMed
    1. Barlow H.B. Inductive inference, coding, perception, and language. Perception. 1974;3:123–134. - PubMed
    1. Beal, M.J., 2003. Variational algorithms for approximate bayesian inference. PhD. Thesis. University College London.
    1. Bernard C. Charles C Thomas; Springfield (IL): 1974. Lectures on the Phenomena Common to Animals and Plants.

Publication types