The Function of Dream Sleep
Total Page:16
File Type:pdf, Size:1020Kb
p&mJRJz VOL.30 I4 JULY 1983 III COMMENTARY The function of dream sleep Francis Crick* & Graemi Mitchison* Wepropose that thefunction of dreamsleep (more proper& rapid-&yemovemen tor REM sleep)LT to removecertain undesirablemodes of interactionin networksof cells in the cerebralcortex. We postulatethat this is done in REMsleep by a revem learningmechan ism(see also p . 158), so that the trace in the brain of the unconscipusdream is weakenedrather, than strengthened,by the dream. MANKIND has always been fascinated by is explained in more detail below. Without to go into large-amplitude instabilities5 . dreams. As might be expected, there have it we believe that the mammalian cortex heenmany attempts to assign a purpose Or could not perform so well. Neuronal networks significance to them. Although we dream We first describe our ideas about the cor- Now, if one asks what functions such richly for one or two hours every night, we do not tex followed by a brief account of neural interconnected assemblies of cells could remember most of our dreams. Earlier networks. Next we outline what is known serve, one attractive possibility is that they thinkers, such as Freud, did not know this. about REM sleep. (For general accounts, could store associations6-R. To see this, Modern theories (not reviewed here in see refs 1,2.) We then describe our suppose an ‘event’ is represented by the ac- detail) have usually proposed that sleep postulated mechanism and how it might be tivity of a subset of cells in a cell assembly. and dreams save energy or have various tested. Finally.we discuss various implica- If all the cells involved in that event form restorative functions, either to replenish tions of our ideas. mutual synapses, then when part of that the brain biochemically in some way, or to event is encountered again these synapses reclassifyor reorder the information stored The cortex can causethe regeneration of the activity in in it. The cortex consists of two separate sheets the entire subset. Sleepis of several kinds. Dream sleep, or of neural tissue, one on each side of the Much exploratory theoretical work has rapid eye movement (REM) sleep, is head. The neocortex, which has a been done on such networks of cells (for an predominantly found in viviparous mam- characteristic layered structure, is found introduction seerefs 6-8). In these models, mals and birds. It seems to be associated only in mammals (see ref. 3 for recent information is stored in the strengths of the with homeothermy (a constant internal survey), although a somewhat analogous many synapsesand sometimes in the firing temperature) and the possession of an structure, the wulst, is found in birds. If thresholds of cells as well. Although the ex- appreciable neocortex or its equivalent. allowance is made for body weight, it is act behaviour naturally depends on the It is not unimportant because of the ap- larger in primates than in most other mam- details of the particular model, certain preciable amount of time we spend in this mals and larger in man than in other general properties can emerge even from Peculiar state. primates. It makes up a substantial fraction relatively simple models. The associations We propose here a new explanation for of the human brain. which are stored are not assigned specific the function of REM sleep. The basis of Different areas of the cortex perform locations for each item, as in a digital com- our theory is the assumption that in different functions, some being mainly puter. Instead the information is: (1) viviparous mammals the cortical system associated with vision, touch and so on, Distributed: this implies that a particular (the cerebral cortex and some of its’ while others appear to process more com- piece of information is distributed over associatedsubcortical structures) can be plex information not associated with a very many synapses. (2) Robust: this im- regarded as a network of interconnected single sensory mode. The exact function of plies that the information will not be totally Cellswhich can support a great variety of the neocortex is unknown but it appears to lost if a few synapsesare added or remov- modesof mutual excitation. Such a system be closely associatedwith higher mental ac- ed. (3) Superimposed: this implies that one is likely to be subject to unwanted or tivities. It seemslikely that it has evolved to synapse is involved in storing several ‘Parasitic’ modes of behaviour, which arise perform in a rather special way. distinct pieces of information. asit is disturbed either by the growth of the In examining the neuroanatomy of the A properly designed net can be trained brain or by the modifications produced by neocortex one is struck by the very large (meaning that the strengths of the synapses experience. We propose that such modes number of axon collaterals (this is not true, can be adjusted) so that given an input (a are detected and suppressed by a special for instance, of the thalamus). In any area pattern of axonal firings) it can produce the mechanism which operates during REM of the cortex the great majority of synapses appropriate output (another pattern of ax- sleepand has the character of an active pro- come from axons originating locally and onal firings). It is found that certain cesswhich is, loosely speaking, the op- running within it. There is also evidence general properties will often emerge. (1) posite of learning. We call this ‘reverse that the majority of the synapsesin the cor- Completion: given only part of the input learning* or ‘unlearning’. This mechanism, tex are excitatory in their action. This sug- (as a clue) it can produce fairly exactly the which is not the sameas normal forgetting, gests a capacity for self-excitatory modes whole of the appropriate output (examples of behaviour in the cortex. And indeed, in aregiven in ref. 7). In computer jargon, the *TheSalkInstitute 10010NorthTorreyPinesRoad, La JOlkX California 92037. USA. Present address (GM.): various conditions, such as epilepsy, memory is ‘content addressable’. (2) MRC Laboratory of Molecular Biology and the migraine and certain kinds of drug-induced Classification: given an input which is KmWh Craik Laboratory, Cambridge CBZ 3EG. UK. hallucination*, parts of the cortex appear related toseveral of its associations, it may NATURE VOL304 14 JUI Y I%3 112 COMMENTARY produce an output which combines many REM periods may of the muscles of the Jouvet “, have postulated a ‘dream state of the common features of its normal out- sleeping animal, especially its head and generator’ which lies mainly in the pontine puts. neck muscles, are more relaxed than in reticular formation (the question of which A major difficulty with all nets of this non-REM sleep. Its cortex, as judged by exact cell groups are involved is controver. general type is that they become overload- the electroencephalogram (EEG) and by sial). It produces the so-called PGO waves. ed if an attempt is made to store the rapid movement of the eyes beneath They propose that the activity of such cells simuhaneously too many. different pat- closed lids, appears to be very active and in is the cause of both rapid eye movements terns or associations of patterns, or if the a state similar to the waking state. On the and the periodic intrusion of new subject stored patterns have too large an overlap. other hand, the monoamine neurones in matter into hallucinoid dreams. Our pro- This is becauseof the superimposed nature the brain stem, especially those in the locus posals are based on this idea. of the storage. How the net will behave coeruleus, raphe and peribranchial nuclei, In summary, the evidence suggests that when overloaded depends on the exact reduce their firing rates in REM sleep to on- in REM sleep the brain is isolated from its structure of the net, but certain patterns of ly a few per cent of the corresponding rate normal input and output channels and that behaviour are likely to emerge: (1) The net in the waking state9. it is very active, this activity being pro- may produce many far-fetched or bizarre Another major difference between REM moted by rather nonspecific signals from associations (‘fantasy’). (2) The net may and non-REM sleep lies in the dreams the brain stem and reflected in the uncon- tend to produce the same state, or one of a associated with them. For most people the scious equivalent of dreaming, which only small set of states, whatever the input few dreams found in non-REM sleep tend reachesnormal consciousnessif the sleeper (‘obsession’). (3) Certain kinds of nets, to -have a rather thought-like character. awakes. particularly those which feed back on During REM sleep, on the other hand, themselves, may respond to inappropriate dreams occur more frequently and usually The postulated mechanism input signals which would normally evoke have a perceptual vividness and the illogical We need a mechanism which will tune the no response from the net.(‘hallucination’). episodic character with which we are all cortical system, in the sense of removing It is against this background of rather familiar. A human adult usually spends a parasitic modes which arise after the tentative and idealized theory that our pro- total of 1% to 2 hours each night in REM system has been disturbed either by growth posals must be judged. sleep, spread over several periods. The of the brain (when new connections are If the cortex were hard-wired during em- evidence suggeststhat most of the dreams constantly being made) or by the modifica- bryogenesis to an exactly predetermined during these REM periods do not reach tions produced by experience. The pattern of synaptic connections, the normal consciousness, dreams being mechanism we propose is based on the burden of eliminating parasitic modes in remembered only if the sleeper awakes more or less random stimulation of the cortical nets would have to be undertaken whiledreaming.