(Work on Title)Visual Learning and the Necessary Nap

1

Naps prevent perceptual deterioration and facilitate learning in local

networks of human visual cortex

A thesis presented

By

Sara Carole Mednick

To

The Department of Psychology

In partial fulfillment of the requirements

For the degree of

Doctor of Philosophy

In the subject of Cognition, Brain, Behavior

Harvard University

Cambridge, Massachusetts

May 23, 2003
INTRODUCTION

Humans and other animals are constantly and rapidly learning from the environment that surrounds them. We learn spatial layouts of rooms and neighborhoods, auditory patterns in music and language, detailed configurations of faces, and much more, all without conscious effort or much apparent attention. We learn an extraordinary amount of information, but not all information is learned in the same way, or by the same mechanisms, or brain areas.

One of the main goals of cognitive neuroscience is to understand how we learn. To this end, researchers have recently investigated the effect of sleep on learning and found that a wide range of learning is dependent on post-training, nocturnal sleep, including visual perceptual, auditory and motor learning. In fact, some studies find that improvement on a visual texture discrimination task occurs only after subjects have slept for at least six hours (Stickgold, 2000a). Findings such as these open the door to new areas of research and invite new questions. One such question is: why do people nap? and do daytime naps have a measurable benefit for learning similar to nocturnal sleep?

In this dissertation, the effect of napping on within-day repeated testing on a texture discrimination task is investigated. I examine whether napping is an effective tool for within-day maintenance of optimal performance on a visual perceptual task; whether naps, similar to nocturnal sleep, can facilitate learning; I compare the benefits of napping to nocturnal sleep. These investigations on napping occur against the backdrop of a second investigation into performance deterioration as a result of repeated, within-day testing on a texture discrimination task.

Overview

The most important new findings in a field can sometimes originate outside the field. These findings usually bring together areas of research not previously considered similar, as well as incorporate techniques not common to the field. Sometimes these new findings can bring about a paradigmatic shift in thinking and scientific approach. Such a shift may be occurring in the field of memory research with the study of the effects of sleep on learning (for review, (Maquet, 2001).

Traditional methods for studying learning have focused on what we learn (faces, spatial layouts, and grammatical rules) in order to understand memory representation, as well as to study who learns what (amnesics, lesion patients, or people with schizophrenia). One goal of these studies is to understand where in the brain such representation is stored. Understanding what and who has been important for laying the ground work for the study of memory and learning, such as the development of the multiple memory system model, which separates declarative from non-declarative memories by their relative dependence on the hippocampal formation. This ground work helps make it possible to begin examining the mechanism of learning and memory.

Recent developments in research on sleep and memory contribute to our understanding of mechanisms. The approach of this research has been to ask not just when, but also under what circumstances does learning occur. Examples of circumstances that influence learning are: circadian rhythm, sleep deprivation and/or amount of sleep. Results from this research have demonstrated that some learning occurs slowly and depends on sleep, while other memories are more rapidly formed and do not rely on sleep. Understanding the contribution of sleep to learning illuminates the biological processes and neurophysiological changes (e.g. protein synthesis, alterations in neurotransmitters and electrical brain waves) that may be important and necessary for learning to occur.

Questions may also now be asked that consider new aspects of the relationship of sleep and learning, such as what happens to performance with and without sleep, the quality and quantity of sleep that is necessary for learning, the type of learning that requires sleep and the type of sleep that contributes to learning. In my dissertation, these are the questions I will be asking.

A brief overview of learning and memory research

Since the time of Plato and Aristotle memory has been a widely studied area of psychology. Experimental studies of memory began in the early 1900s, when Ebbinghaus began his now famous studies of his own memory by tracking how many rehearsals he needed to recall a list of nonsense syllables, methods that are still used in modern memory tests. The birth of the multiple memory system began with Broadbent’s (1958) information processing approach which explicitly contained a limited capacity, short-term memory store where information could either get rehearsed and become part of a long-term memory store or else decay into forgetfulness.

In the mid-fifties doctors removed the hippocampal complex and parts of the temporal lobe bilaterally of an epileptic patient named H.M. and thus accidentally revolutionized the neuroscience of memory research (Scoville, 1957; Milner, 1962). The removal of the hippocampal complex helped H.M.’s epilepsy, but also permanently damaged his memory so that he could not remember new experiences (anterograde amnesia), and he experienced memory loss for events that occurred up to a few years before his surgery. The study of amnesic patients like H.M. revealed a role of the medial temporal lobe (includes the hippocampus, the dentate gyrus and subiculum, as well as complementary areas of the entorhinal cortex, perirhinal cortex and parahippocampal gyrus) in memory acquisition as well as in long term memory consolidation (Weiskrantz and Warrington 1979; Schacter 1985).

Amnesic patients, such as H.M., are living proof of another natural division in memory between explicit and implicit memory. These patients, with severe medial temporal lobe damage, evidence marked deficits in explicit memory tasks that require conscious encoding and retrieval of items, but are unimpaired on implicit memory tests, which access memory for items that have not been consciously encoded and are, therefore, not available for explicit recall. A memory model had been constructed that consisted of a hippocampal-dependent, declarative (explicit) memory system and a non-hippocampal-dependent, non-declarative (implicit) memory system, both of which are supported by a wide range of brain areas (Schacter and Tulving 1994; Squire 1994).

The standard model of declarative memory includes the subtype of episodic memory (comprising knowledge of personal events or episodes), and the subtype of semantic knowledge (comprising knowledge of “facts” about the world: name of the first president, public events and personalities, as well as conceptual knowledge of words, grammars and objects).

All of the other memories that do not require conscious acquisition and recall are part of a collection of non-declarative and procedural memories. These memories includes information acquired during skill learning (including motor skills, perceptual skills, and cognitive skills) and habit formation, simple classical conditioning, as well as priming and non-associative learning. None of these memory categories are thought to depend on the hippocampus, as amnesic patients are unimpaired in these areas. Procedural learning can be found across all sensory modalities, visual (perceptual learning), auditory (music learning and classical conditioning), tactile (pain conditioning), gustatory (taste aversion), and smell (animal studies of maze learning), as well as language (grammar rule learning), and motor learning (sequence learning).

Until recently, the study of memory and learning has been dominated by the methodologies of declarative memory research and the functions of both declarative and non-declarative memory were evaluated similarly. Subjects were exposed to a stimulus once, followed by a test of retention, either overtly or covertly. Or alternatively, a stimulus was repeatedly shown and the number of exposures necessary for consolidation of the stimulus was calculated, either by explicit or implicit response. If learning was achieved, researchers could then examine deterioration of the memory representation from interference or decay.

Recently, a new approach to examining the mechanism of learning has broken the mold of prior memory research. These studies investigate whether other factors may contribute to learning such as slowly developing, off-line processes that produce long-lasting behavioral changes. These studies have examined experience-dependent plasticity of the cortex (Gilbert, 2001), the effect of electrical stimulation of the brain (McGaugh et al, 1979), activation of genes associated with plasticity (Herrera et al, 1996), neural development (Frank et al, 2001), and the effect of sleep on sensory-motor learning (Walker, 2002; Karni, 1991; Stickgold, 2000a). As this dissertation focuses on visual perceptual processing and sleep, the next section will give an overview of visual perceptual learning.

Visual perception and learning

Research in visual perception has formed the basis for much of the work in learning and memory, from early studies of visual short-term memory by Sperling (Sperling 1960) to present work in perceptual learning. Over the past twenty years a special kind of learning has been identified and studied; this learning shows distinct signs of isolated plasticity of early visual areas that may or may not depend on higher areas of processing. This has been termed perceptual learning, as it does not appear to entail higher cognitive processing, but instead shows improved performance on very basic psychophysical measures, such as discrimination of orientation, spatial frequency, motion signal and vernier acuity.

Perceptual learning is characterized as being highly rigid, in that most studies find that improved performance on one task does not generalize to other similar tasks (Gilbert et al. 2001; Ahissar and Hochstein 1997). Learning has been shown to be specific to the trained stimulus (e.g. oriented lines, spatial frequency) (Fiorentini and Berardi 1980; Crist, Kapadia et al. 1997), the retinotopic area of the visual cortex (Fahle, Edelman et al. 1995; Crist, Li et al. 2001), and, in some cases, the trained eye (Fahle, Edelman et al. 1995; Schwartz 2002; Karni et al, 1991). The processes underlying this learning presumably involve mechanisms of experience-dependent cortical plasticity occurring in primary visual cortex (Zohary, Celebrini et al. 1994). Learning has also been shown to generalize to untrained stimuli in circumstances where a wide range of stimuli are trained (Ahissar and Hochstein1997; Liu, 2000). Many models have been proposed to explain the specificity question in perceptual learning. The rich mapping of the visual cortex allows for the marriage of both structure and function in the modeling of perceptual learning.

Wiring of the visual cortex

Orientation selectivity is a remarkable property of neurons in the visual cortex which are thought to provide the detection of local bars and edges in processed visual images and encodes their orientations (Hubel and Wiesel 1974). According to the concept of columnar organization, neighboring neurons in the visual cortex have similar orientation tunings and comprise an orientation column(Hubel and Wiesel 1974). Columns close together generally have similar, but not identical, orientation preferences, and distant columns generally have more dissimilar preferences. For orientation preferences, the arrangement of cells forms an orientation map of the retinal input (Blasdel 1992). Each location on the retina is mapped to a region on the orientation map (i.e., retinotopic specificity), with each possible orientation at that location represented by different orientation selective cells. The global layout of the orientation map (and consequently the orientation preferences of the individual neurons in the map) is formed with experience during early development (Hubel and Wiesel 1968; Blakemore, 1970; Hubel and Wiesel 1974; Blakemore, 1975).

An important addition to the structure and function of the visual cortex is a network of extensive, long-range lateral connections between neurons in neighboring columns with similar preferences (Gilbert et al, 1983; Gilbert et al, 1990). This is a network of connectivity formed by the axons of cortical pyramidal cells. The lateral connectivity is not uniform or genetically determined, but develops based on visual experience (Katz et al, 1992; Burkhalter 1993; Rubin, 1997; Dalva et al, 1994). These connections are initially widespread, but develop into clustered patches at approximately the same time as the orientation maps form. The lateral connections are far more numerous than the afferents and they are believed to have a substantial influence on cortical activity.

An important factor in understanding plasticity underlying perceptual learning in the visual cortex is to understand the temporal characteristics of the learning. Depending on the task, plasticity underlying learning can occur both over relatively short periods of training, as in the “eureka effect” (seconds to minutes) (Ahissar and Hochstein 1993; Fahle, Edelman et al. 1995;Rubin, 1997), as well as over periods of days (Karni and Sagi 1993; Stickgold, Whidbee et al. 2000b). This slow improvement occurs even in the absence of continued practice, indicating that some off-line processing must occur. Though many models have been proposed, the actual mechanisms of visual cortical plasticity involved in both the fast and slow improvement remain almost completely unknown. Studies that investigate not only what is learned, but also examine under which circumstances fast and slow learning occur, have found evidence that some of the slow learning is sleep-dependent (Karni and Sagi 1993; Karni, Tanne et al. 1994; Karni, Weisberg et al. 1995; Gais, Plihal et al. 2000; Stickgold, James et al. 2000a; Stickgold, Whidbee et al. 2000b; Stickgold, Hobson et al. 2001).

Sleep-dependent learning and plasticity

Sleep-dependent learning has been demonstrated across a wide range of sensory and motor tasks (for review see, Maquet 2001), some of the clearest evidence of sleep-dependent learning comes from studies of a texture discrimination task originally developed by Karni and Sagi (1991). Karni and colleagues have shown that improvement is only evident several hours after training, and that improvement can develop overnight, although only when rapid eye movement (REM) sleep is allowed (Karni, Tanne et al. 1994). (A further discussion of sleep architecture can be found in the next subsection.) Karni et al. subsequently showed in an fMRI study that training across days - weeks could lead to enlarged regions of activation in the primary and secondary visual cortex (Karni, Weisberg et al. 1995). This finding was later replicated by Schwartz et al. (2002) over a 12hr period.

In a series of subsequent studies, Stickgold and colleagues demonstrated that improvement in performance on the texture discrimination task can be achieved only after a full night's sleep (2000a,b). Without a post-training night of sleep, no learning occurs even after subjects are allowed two nights of recovery sleep. This consolidation process continues beyond the first post-training night without further training, such that texture discrimination performance tested four days after the initial testing is the same in a group tested once a day for four days as a group tested only once on the first day and once on the fourth day (2000b). By examining the effect of particular sleep stages on learning, Stickgold noted a relationship between overnight improvement and both deep slow wave sleep (SWS) and rapid eye movement (REM) sleep, specifically improvement correlated with the product of the amount of SWS in the early part of the night and REM in the last part of the night (2000a). Stickgold et al proposed a two-step model for sleep-dependent learning, in which SWS and REM have independent and sequential roles in the process of consolidation. As such, it explains why a full night of sleep (6 to 8 hours) is required for optimal consolidation of post-training learning. The next section briefly reviews sleep structure.

The structure of nocturnal sleep.

Sleep is a highly structured set of processes separated into five phases, each demonstrating stereotypic electrical activity, neuro-chemical expressions, and enhancement and depression of varying brain regions. The five phases, stage 1, 2, 3, 4 , and Rapid Eye Movement (REM), progress in a cycle from stage 1 through stage four and then back up to REM sleep (Figure 1).We spend almost 50 percent of our total sleep time in stage 2, about 20 percent in REM sleep, and the remaining 30 percent in the other stages. Infants, by contrast, spend about half of their sleep time in REM sleep. The duration of an entire cycle lasts for 90-110 minutes. The period of deepest sleep, slow wave sleep (SWS), is composed of stages 3 & 4. In stage 3, extremely slow brain waves called delta waves begin to appear, interspersed with smaller, faster waves. By stage 4, the brain produces delta waves almost exclusively. The beginning of the night is characterized by a larger proportion of SWS. As the night progresses the amount of SWS decreases and there is a corresponding increase in REM sleep. Consequently, the morning period is rich in REM sleep. REM sleep, in contrast to SWS, is a lighter sleep accompanied by rapid irregular shallow breathing, rapid jerking eye movements, increases in heart rate, as well as limb muscle paralysis.

Neuromodulator fluctuations occur across different sleep stages. Brainstem systems that control the REM-NREM (non-REM) cycle include the noandrenergic (NE) locus coerlus, the serotonergic (5-HT) dorsal Raphe nucleus, and the cholinergic (ACh) nuclei of the dorsolateral pons (Hobson 1975). Whereas NREM is characterized by decreases in all three neuromodulators compared with waking, ACh levels in REM are equal to or higher (Karnetani, 1990) than during wake, and levels of NE and 5-HT drop to zero. Changes in both ACh and 5-HT have been proposed as mediators of memory consolidation (Hasselmo 1999; Graves, Pack et al. 2001). Table 1 shows physiological correlates of the different sleep stages.

Neurophysiological basis of sleep-dependent learning