Why do we sleep? What is the biologic al role of dreams? How are dreams generated, and why do so many of them feature anxiety, fear, and sexual desire? Here I review some experiments that address these questions, summarize our current knowledge in the area, and propose some new and speculat ive ideas.

Sleep

Adult humans spend about a third of their lives sleeping. Newborn babies spend about double that amount of time in this state. However, we are not the only lazy beings: sleep is conserved (retained in evolut ion) in all homeothermic (constant body temperature) vertebrate mammals. Sleep is also one of the strongest drives which guides animal -- including human--behavior: it is extraordinarily difficult to avoid sleeping for periods longer than 24 hours. Force d sleep deprivation can kill a rat. Additionally, sleep -- and especially REM, or dream sleep-- displays unique brain processes, including very active ones, unlike what the resting sensation it produces would suggest. All of these reasons suggest that s leep serves one or more important functions and make its study one of the most attractive subjects of contemporary psychology.

Physiological aspects

One interesting question about sleep concerns the mechanism of sleep activati on. Although the processes of sleep activation are not clearly defined, they probably involve a circadian rhythm-dependent mechanism, triggered by light, and an effect dependent on the time since the individual last slept.

Sleep consists of two very different alternating stages. These can be differentiated by means of an electroen-cephalogram (EEG), an electrooculogram (EOG), and an electromyogram (EMG), which measure the activity of brain waves, eye movement, and muscle tone, respectively. The f irst stage, termed NREM (non-rapid eye movement), is characterized by high-amplitude, low-frequency waves registered in the EEG, rolling, uncoordinated, and slow movement of the eyes, and passively decreased muscle tone. REM (rapid eye movement), in cont rast, is characterized by high-frequency, low-amplitude, more irregular waves in the EEG, rapid and coordinated movement in the EOG, and a very weak EMG, with muscle tone actively inhibited.

NREM can in turn be divided into four stages, characterized by a progressive increase in sensory thresholds, an increase in EEG wave amplitude, and a decrease in EEG wave frequency.¹ NREM always precedes REM in the adult, and is longer and/or deeper if the waking period preceding it was long or contai ned vigorous exercise.^1,2 This is valid for both day to day comparisons in an individual and for species to species comparisons of the average length of the period. Most animals have three or more cycles of NREM-REM sleep per night. In human s, the average is four to five REM bouts of 90-100 minutes each. The time span of the cycle is a function of animal and brain size, varying among species and individuals.¹

NREM and REM also differ in somatic activity. During NR EM, there are decreases in blood pressure, heart rate and respiratory rate, while REM sleep is associated with activation and increase in the irregularity of these functions, along with penile erection and clitoral engorgement.¹ Another strikin g feature of REM is the lack of thermal regulation--one is awakened if one's temperature falls too low.¹

REM sleep

J. Allan Hobson, Professor of Psychology at Harvard Medical Schoolc omments: "...the experience of dreams should have motivated scientists to look for evidence of brain activation during sleep." Yet REM was not discovered until 1953, and even then scientists were greatly surprised to find such increases in br ain activity during sleep.¹² There is strong evidence that both REM and NREM have important functions, but it is REM sleep which has attracted the most attention. This is due partly to the paradoxical nature of this state: while it is the deep est of sleep states in the sense that sensory thresholds are highest (sensory stimulation must be of higher intensity to wake one up during REM as opposed to NREM) and muscle tone is completely inhibited, it is also the most active of sleep states regardi ng brain activity. Additionally, we know that dreams occur mostly during REM sleep. Our desire to understand and interpret our dreams also pushes us to explore REM sleep

Dreams

When awakened during REM sleep, humans typically giv e detailed reports of dream activity, even if the subjects do not usually recall dreams in the morning.¹ This rarely happens if the subjects are awakened during NREM sleep, five minutes after the end of the last REM period.

Dreams are a "c onscious" experience, a psychological consequence of the physiological state of REM sleep. The word "conscious" is used in this context because subjects awakened during REM sleep are aware of their dreams. Although Hobson assumes a formal isomorphism be tween the conscious experience and the physiological, an isomorphism between the psychological effect of dreaming and physiological REM sleep may not be appropriate if what we usually call dreaming is only one of the manifestations of the REM state.

This state is characterized by five main features: hallucination (perception of objects with no sensory input), delusion (self-deception, belief in the truthfulness or reality of an imaginary experience), emotional intensification, amnesia and cognitive a bnormalities.¹⁰ Any model of dreaming has to explain these.

Hallucination can be accounted for by the generation of internal signals by the brain stem. These internal bursts of activity randomly activate the recall of certain memories an d produce corresponding perception. Delusion may be attributable to the brain's isolation from external reference points. With no sensory input to tell the brain that hallucinations are unreal, it may have no way of recognizing them as imaginary. As a subject with his eyes blinded in a silent room can imagine objects and still realize he is just imagining, however, delusion might well involve the turning off of some mechanism. Amnesia and cognitive abnormalities like "jumping" from one thought to anot her can be explained by "mode-switching," which Hobson describes as resulting from the turning off of aminergic neurons during REM sleep.¹⁰ These neurons are known to have regulatory roles and influence attention (the capacity to maintain an o rganized chain of thought). The switching off of these neurons could thus cause one thought to lead to another totally unrelated one, unlike a waking state in which a stream of thought is an organized sequential chain of related events.

If dreams are caused by the random activation of visual memories, the fact that dreams are story-like and not a succession of random visual images requires an interpretative machinery during dreaming (or during recollection) to knit a more or less coherent story of the dream. Here I disagree with, or at least wish to add to, Hobson's activation-synthesis model of dreaming. Hobson postulates that "the development of perceptions during dreaming is progressive, proceeding from primordial visual stimuli to perceptual imag ery."¹³ While images in a dream may be formed in a sequential fashion, part by part, as Kosslyn's model predicts,⁴ I would like to suggest that brain activation during dreaming occurs at the level of "whole-concept" memories (e.g. e pisodic memories) and that the visual sensation is secondary.

The distinction I wish to make is between Hobson's statement that images are generated sequentially from more primitive components and that which asserts that the original activation that causes the generation of the whole image is at that component level. Steven Kosslyn has shown that if a letter of the alphabet is presented to a subject visually, the subject's formation of a mental image of the letter occurrs segment by segment (li ne by line). This does not mean that the original activation (in Kosslyn's study motivated by visual input, in dreams motivated by internal activity) is at the level of the visual parts constituting the final image, however.

It is difficult to beli eve that when Kosslyn asks subjects if the upper-case version of a lower-case letter presented visually has any curves, there is a direct and sequential internal visualization of each of the lines composing the upper-case letter without a prior stimulatio n of a memory site for the concept of that letter. It is easier to imagine that seeing the lines composing the lower-case letter recalls the concept of this lower-case letter, that this in turn revives the concept of the corresponding upper case letter, and that this finally activates the memory for its component segments to form the image. I say this because there is little relationship between the lower-case letter and the lines composing the upper-case version without a concept of the upper-case lett er itself, as a whole.

In the same way, because we see entire objects in dreams and usually not just isolated lines (and in dreams people are not just random collections of arms and legs), I propose that the "calling-up" of an object in a dream is an evocation of the whole concept. The concept memory then activates the recollection and visualization of the component parts that form the image corresponding to the concept.

My model, like Hobson's, requires a high-level interpreter of dreams to con nect concepts appearing in a story, but this interpreter would have an easier job building a coherent story from whole images or remembrances than from raw primitive visual stimuli. Alternatively, random internal bursts of activity from the brain stem co uld initiate each dream, but subsequently each image could lead to the next without further intervention from these random bursts, so that an interpreter is not necessary any more than in normal thought.

Episodic memories are those which relate to inc idents in the life of the individual, as opposed to semantic memories, which involve meanings or facts unrelated to the individual's own history. I have not found any such study in the literature. An interesting paradigm to study which memories are acti vated by internal signals from the brain stem might be the study of the dreams of an episodic-amnesic patient.

Function of sleep

REM has only been found in mammals and in young birds, while NREM sleep is also found in reptile s, suggesting a more primitive role for NREM and a relationship between REM and the advanced cortex of mammals, and to a lesser degree, of birds.¹ While there is as yet no well-established function for sleep, most explanations proposed fall in to one of the following categories: behavioral, metabolic, developmental, or informational.

Behavioral advantages

Behavioral explanations for sleep rely on viewing sleep as a behavior which has adaptive advantages for the species. Observations which support such arguments include:

--Sleep protects organisms which can not see well in the dark and would run greater risks if active at night.

--Animals that need to regulate their internal temperature would be better off sl eeping in a warm place without the need for increased metabolism to fight the cold of night.

--A behavior which tends to bring sexual mates together in a nest every night will foster reproduction and an environment which is good for nurtur-ing the youn g.

Behavioral studies of sleep in animals has not been thoroughly explored. The development of remote portable monitoring systems should provide a useful tool of study.²

Metabolic explanations

Metabolic account s of sleep function rely on experiments that show the increase of NREM sleep after prolonged exercise or wakefulness and argue that slow-wave sleep saves energy. Because aminergic neurons are turned off or slowed down during REM, supporters of metabolic theories argue that sleep might allow these neurons to replenish stock of neurotransmitters which they may use up during periods of attention.² Yet because of the great activity of the brain during REM, and because the number of neurons inacti ve during NREM is relatively low (only a few percent), sleep as neural rest, the most intuitive of all apparent sleep functions, seems an inadequate notion.¹

Developmental role

The enormous amount of time a developing animal spends in REM has led researchers to ascribe a developmental role for this state. Some have postulated that organized activity of neural mechanisms might be necessary before testing them in the real world. Or, the periodic stimulation of the cor tex in a semi-random and unspecific manner may guarantee the maintenance of circuits vital for survival but rarely activated by external stimuli (e.g. emergency defense)^.2

Informational function

Th e informational approach to sleep function asserts that sleep is important in learning and consolidating memory,³ forgetting potentially parasitic memories (memories that are recalled when they should not be, e.g. with random activation of the neural network),⁵ and comparing new memories with older memories or genetic instructions.^2,6 Perhaps because of its relationship with high-level functions, it is this apprach that has been explored most widely. I will devote the ne xt section to various attempts to explain the function of sleep and REM at the informational level.

A role for dreams in learning and memory

There is considerable evidence that REM sleep is necessary for learning to take place. F or example, Karni & Sagi have established that changes in the plasticity of specific neuronal loci which underlie perceptual learning may occur during sleep.³ In their experiments, they trained human subjects for a visual recognition task and recorded their progress on several sessions each day during several days. Their work showed that except for a fast, rapidly saturating learning curve on the very first learning session, a very stable performance was observed within any given session. Slow learning took place between sessions on consecutive days, but not between sessions on the same day, suggesting that sleep might play a role in the learning process. Such learning persisted after several years, even without further practice. Karni & amp; Sagi also showed that the process of consolidation occurs during normal sleep and is dependent on the integrity of REM sleep.⁶ These experiments suggest that there is a direct involvement of REM sleep in long-term learning processes.

< p> Dreams to Forget

Crick and Mitchison have proposed that the function of dream sleep is to remove certain undesirable modes of interaction between cells in the cerebral cortex which could otherwise turn parasitic.⁵ They sug gested that the mechanism involved in such erasing is a reverse learning mechanism, whereby associations are weakened, rather than strengthened, by their occurrence in the dream. Because it is an active process, this is not the same as normal forgetting< p> Crick and Mitchison's proposal is based mainly on simulations with neural computer networks, which have many characteristic features which are believed to operate in the brain. Experiments with neural networks show that the networks become overloaded if an attempt is made to store too many associations in them. In that case, the net is likely to produce bizarre associations (`fantasy'), it may tend to recur in the same outcome whatever the input (`obsession'), and it may respond to inappropriate inp ut signals which would normally elicit no response (`hallucinations').

Crick and Mitchison suggest that the mammalian brain is likely to become overloaded during adult learning and especially during development, when new connections are being created as a consequence of growth. These connections are probably not final and are likely to be fine-tuned by experience, given that these organisms are capable of large amounts of learning. They suggest that sleep offers a mechanism by which meaningless mem ories can be 'unlearned' before they become obsessions which are activated by random internal spurs of electrical activity. By meaningless I mean not unimportant memories, but fantasy created by an overloading of the network. Hopfield et al. have shownt hat a mechanism which down-regulates associations evoked by random input patterns, shows improvement in the functioning of simple neural networks⁷ by reducing the frequency of spurious memories and equalizing the access to real ones.

Cri ck and Mitchison postulate that REM sleep is such a mechanism. According to their proposal, dreaming makes the associations in the dream less likely to occur in the future. They base their proposal on the following facts:

-- During REM sleep, the pontine reticular formation in the brain stem produces the so-called PGO waves, which stimulate the cortex in a semi-random way.

-- Most dreams are only remembered if the dreamer wakes up, so the function of dreams is probably not connected to th eir conscious consequences.

-- During REM sleep the brain is isolated from its normal input, so that internal activation is the only source of cortical stimulation.

-- Mammals that do not have REM have larger brains than would be expected comparat ively, suggesting that REM may contribute to reduce overloading and thus reduce necessary brain size.⁹

This theory predicts that defects in the ability to incur REM sleep might cause fantasy, hallucinations or obse ssion. Although some evidence has been found to support this, it indicates correlation rather than causality.¹⁰

It is interesting to note that David Hartley suggested the essence of Crick and Mitchison's idea in 1801: "The wilderness of our dreams seems to be of singular use to us, by interrupting and breaking the course of our associations. For, if we were always awake, some accidental associations would be so much cemented by continuance, as that nothing could afterw ards disjoin them; which would be madness. . ."

Sex, fear and anxiety

Crick and Mitchison's theory does not explain the enormous incidence in dreams of situations related to fear, anxiety or sexual desire.

There might be an a dvantage in enhancing the memory of situations which induced fear, for the individual would avoid a repeat of the dangerous situation. Events in the future which cause anxiety are usually important events for which we do not feel sufficiently prepared. Remembering them well could be of help. Sex is without doubt the most important drive in natural selection terms, so that a recreation in the mind of potentially successful sexual encounters could enhance fitness. Emotions are the link between cognitive information which the animal acquires during its lifetime and the biological drives which are hard-wired into its genome and which make complicated behaviors which favor the survivor. In this way, situations with possibly important implications for the individual are assigned emotions by the central nervous system. It is thus important that such information be assigned a privileged position in memory. Dreams could play a role in the maintenance of this privileged set of data.

A priori, this could seem to imply a complex cognitive-level mechanism to select dreams which contained fear, anxiety or sexual arousal. However, this is not necessary and is not supported by the evidence for apparently random bursts of activity of the brain stem. The sele ction of emotionally arousing memory contents in dreams could be done in a more general and unspecific manner by the activation of emotional centers in the limbic system and of "startle" networks in the brain stem.¹⁰ In contrast to Hobson's pr oposal which describes an "intensification of emotion," I propose that there is selective recall of emotion-arousing memories, suggesting there is a distinctive feature in the way these memories are stored: there is some label of the memories as emotion-a rousing. This is not surprising, since the recall of events which caused the emotion when they happened usually causes emotion itself. Rather than emotion being added to the dreams, I suggest that the activation of emotional centers causes the recall of memories which were already emotional when they were stored. I base my proposal on the fact that dreams are usually related to waking-life emotional situations (concerns, fears, sexual desires), and not with non-emotional or absolutely fictional (non-rel ated to waking-life) situations that are "artificially colored" with emotion. In addition, I can find a function for the first, while I cannot say the same of the second. And, as Hobson admits, mechanisms as complicated as these are unlikely to have evo lved if they do not have a selective advantage.¹¹

This proposal will remain speculative until experiments have been performed to confirm or deny it. Such tests could include the use of drugs that inhibit the functioning of the sy stems underlying emotion and verifying whether dreaming is reduced or if the topics of dreams change significantly, and the search for a lesion in rat brains - or a human disorder - that will impair the subjects' ability for the recall or storage of emot ion-associated conditioned responses (e.g. anger or fear) without disabling the ability to display emotional behavior itself or the learning or recalling abilities of non-emotion related memories.

Crick and Mitchison suggest that the prevalence of emo tionally arousing themes in dreams may be a characteristic only of dreams that are remembered, because they awake the subject due to the anxiety associated with them. In these cases, the learning process reverts from reverse learning to positive learning, and so their recurrence could be explained. This possibility could be tested by observing whether the dreams reported by subjects "artificially" awakened during an experiment show a decrease in the proportion of "anxiety-causing" themes.

In the selec tion of memories for dreams, it is obvious that other factors, apart from emotionality, intervene: one such example is the age of the memory. This is functionally advantageous too: recent events which aroused emotion are more likely to happen again in the near future than distant ones, and so intensifying their memory is likely to be more useful to the individual.

Implications for Freudian theory of dreams

Where does all of this leave Freud's view of dreams? The notion of a `conce alment' of wishes is unnecessary and unjustifiable by the evidence. However, the history, wishes and fears of each individual are likely to influence his/her dreams both by making different memories available to the randomly activating mechanism, and by d iffering the subject's "knitting" of these into a dream. Thus dreams may continue to be regarded as useful windows into the dreamer's mind, offering an unusual possibility of scrutiny while "the master of the school is away and the boys are in an uproar," (Thomas Paine, in An Essay on Dreams (1795)).

Although many advances have been made in our comprehension of the phenomena of sleep and dreams, the field is still in its youth. Although many of the ideas set forth are speculative, I hope that i n expressing them I have managed to convey my fascination for the subject. If, as a result, any one reader is motivated enough so as to show through experiment that one of them is wrong -- or right --I will have achieved my goal.

Ale jandro Backer was born in Buenos Aires, Argentina, in 1972. He studied Biology at the University of Buenos Aires (1991-1993), and then at the Massachusetts Institute of Technology, where he is currently a junior. He has worked in the Immunogenetics Lab of the Hospital of the University of Buenos Aires (1991-1993), and currently does research in the molecular biology of mammalian development in the lab of David Housman, at MIT's Center for Cancer Research.

Editors: John Bade, Jenn Mor azes, Pam Kirschner

References

1. Hobson, J. A., States of Brain and Mind, 106-109 (Birkhaeuser Boston, Boston, 1987).

2. Hobson, J. A., States of Brain and Mind, 110-111 (Birkh aeuser Boston, Boston, 1987).

3. Karni, A. & Sagi, D., Nature 365, 250-252 (1993).

4. Kosslyn, S. M., Science 240, 1621-1626 (1988).

5. Crick, F. & Mitchison, G., Nature 304, 111-114 (1983).

6. Karni, A. & Sagi, D., Neuroscience Ab stracts 18, 387 (1992), as reported in (3).

7. Hopfield, J. J. et al., Nature 304, 158-159 (1983).

8. Hicks, R. et al., Bulletin of the Psychonomic Society 26(1), 59- 60 (1988).

9. Crick, F. & Mitchison, G., Journal of Mind and Behavior 7(2-3 ), 229-249 (1986).

10. Hobson, J. A., The dreaming brain, 203-214 (Basic Books, New York, 1988).

11. Ibid, 288.

12. Ibid, 286.

13. Ibid, 218.