62is-23 Memory and the Hippocampus:A Synthesis From Findings With Rats,Monkeys,and Humans Larry R.Squire This article c na h largely in a ampus and th dif t fu s The hipp us (t r with plicit and trasteditha hipp ampus6ilsasndhabt,inptcoodno sent a whole memory In recent years a co sus has been developing about the the medial temporal lobe was responsible for H.M.'s amnesia roleofther was not knowr ior in a selective way:Howe consi rable unc ans and nonhuman primates,the term hipp could e ion that the hi h brings to a hig ory in humans and nonhuman primates. dem nteg x1966 mnesia foll urgical rem d of the media obe bila ststhat,in ad on to the hippocampus proper dial tem al lobe (es dam ntal principle that m mory could be ory systen ere wa ond.the idea is developed that the the hippocan and underly Miler,97),precisely what da amage withir tant kind of ere memory ot simila tion that the i ith hin la-Mo an,Frank commentsoneaicerYeiong mpa function weref de the hi 2161 195
Psychological Review 1992, Vol. 99, No. 2,195-231 In the public domain Memory and the Hippocampus: A Synthesis From Findings With Rats, Monkeys, and Humans Larry R. Squire Veterans Affairs Medical Center, San Diego, California and University of California, San Diego This article considers the role of the hippocampus in memory function. A central thesis is that work with rats, monkeys, and humans—which has sometimes seemed to proceed independently in 3 separate literatures—is now largely in agreement about the function of the hippocampus and related structures. A biological perspective is presented, which proposes multiple memory systems with different functions and distinct anatomical organizations. The hippocampus (together with anatomically related structures) is essential for a specific kind of memory, here termed declarative memory (similar terms include explicit and relational). Declarative memory is contrasted with a heterogeneous collection of nondeclarative (implicit) memory abilities that do not require the hippocampus (skills and habits, simple conditioning, and the phenomenon of priming). The hippocampus is needed temporarily to bind together distributed sites in neocortex that together represent a whole memory. In recent years a consensus has been developing about the role of the mammalian hippocampal formation in learning and memory. The idea that the hippocampus is important for memory is not in itself new. What is new is that this idea is now supported by direct and compelling evidence for each of the three species that has been important to this work: rats, monkeys, and humans. In addition, there have been major gains in understanding exactly how the hippocampal formation is involved in memory. Not too many years ago, when the topic of memory and hippocampus was discussed in the context of research on humans and nonhuman primates, the term hippocampus could be used only tentatively. Elegant neuropsychological studies of the noted amnesic patient H. M. (Scoville & Milner, 1957) had demonstrated convincingly that memory depends on the integrity of the medial temporal lobe (Milner, 1966). H. M. developed severe amnesia following surgical removal of the medial temporal lobe bilaterally in an attempt to relieve severe epilepsy. Continuing study of H. M. (Corkin, 1984; Milner, 1972) established the fundamental principle that memory could be dissociated from other intellectual functions. However, the medial temporal lobe is a large region that includes the hippocampus, amygdala, and adjacent cortical areas. Although there was reason to believe that the posterior aspect of the lesion was especially critical, that is, the hippocampus and underlying cortex (Scoville & Milner, 1957), precisely what damage within This research was supported by the Medical Research Service of the Department of Veterans Affairs, National Institute of Mental Health Grant MH24600, the Office of Naval Research, and the McKnight Foundation. I thank Stephen Kosslyn, Stuart Zola-Morgan, Frank Haist, and Gail Musen for their helpful comments on earlier versions of this article. Correspondence concerning this article should be addressed to Larry R. Squire, Veterans Affairs Medical Center, 3350 La Jolla Village Drive, San Diego, California 92161. the medial temporal lobe was responsible for H. M.'s amnesia was not known. In the rat, which has been the most commonly used experimental animal for neurobehavioral studies, it was clear that the hippocampus proper was important for some function, because lesions placed within the hippocampus disrupted behavior in a selective way. However, until recently there has been considerable uncertainty about which tasks are the appropriate ones for detecting behavioral deficits and about how to interpret the deficits. The purpose of this article is threefold. First, recent evidence is summarized, which brings to a high level of certainty the conclusion that the hippocampus itself is important for memory in humans and nonhuman primates. Indeed, there is now good correspondence among the findings for all the commonly studied mammalian species. Furthermore, the recent evidence suggests that, in addition to the hippocampus proper, certain adjacent and anatomically related cortical structures in the medial temporal lobe (especially entorhinal, perirhinal, and parahippocampal cortex) also participate in memory functions. The components of the medial temporal lobe memory system can now be identified in broad outline. Second, the idea is developed that the role of the hippocampus (and related cortex) is narrower than once believed. The hippocampus is essential for a specific but important kind of memory—here termed declarative memory (other similar terms include explicit and relational memory). The first suggestion that the hippocampus is involved in only one kind of memory was developed by Hirsh (1974) on the basis of studies of rodents with hippocampal lesions. Subsequently, other hypotheses about hippocampal function were also presented that contained the idea that only a particular kind of memory is dependent on the hippocampus (Gaffan, 1974; O'Keefe & Nadel, 1978; Olton, Becker, & Handelmann, 1979). Eventually, considerable evidence for the idea that only one kind of memory is 195
196 LARRY R.SQUIRE 1982)e othe ely amnesic (Cohen,1984 Squire Indeed,in the absence of the hippo mpus.sev nitive function were rene dly evaluated and his m pairment was d ented.The only cognitive deficit that was the tion of R.B's brain after his deathn revealed a d that the role fthe hinr in memo is time limited.Amne patients.includ campal dal extent of the hippocar asonably b SRk然2S uire,Haist,& of retro cas thus showed that to th hippocampusi ent to cause and cl mesia in the monkey (2 gan Squi 199% rage is only t rary Memory is gradually reo pp learn ng M ris initial bl duces selective neuronal loss in the ed later,bilateral CAl damage and f the Medial foun e that the at and mor ey pro ing e ntial for ncxDerimcnta memory The strongly supported b and studies of the effects of selective lesions in rats. 89 Memory-Impaired Patients The ding rats are dise ussed more fully in a later s ovided val Additional confir ation for the idea that the human hippo 1987) have been thosc where the amnesia occurs against a back Robninneiomicalinomatioanvingpgticas pocampus was developed tha permits visualization of the hir rofound forgetful pocamp det nes a)and o Until re 89).con campus were demonstr ated in 4 pati en with circu be at neur path patients being studied.S case studies attribu the mbria. damage sizel.In contrast,the area of the e temp et al1985;Victor.Ang call.6 damage in patients with a selective m nory disorde and othe truct matter on firme ound.Patient B.became the severity of H.M.s memory impairment resulted from dam 97 at the age earuyo-Morg.quire. age be structu
196 LARRY R. SQUIRE affected by hippocampal damage accumulated in demonstrations of entirely intact learning and memory abilities in patients who were otherwise severely amnesic (Cohen, 1984; Squire, 1982). The important implication was that memory is not a single entity. Indeed, in the absence of the hippocampus, several other kinds of learning can still be accomplished, including the learning of skills and habits, simple conditioning, and the phenomenon of priming. Third, the idea is developed that the role of the hippocampus in memory is time limited. Amnesic patients, including patients with confirmed hippocampal damage, have difficulty recalling the recent past but can recall remote events as well as normal subjects (MacKinnon & Squire, 1989; Squire, Haist, & Shimamura, 1989). Recently, the significance of this observation has been illuminated by a prospective study of retrograde amnesia in the monkey (Zola-Morgan & Squire, 1990c). The findings indicate that the role of the hippocampal formation in memory storage is only temporary. Memory is gradually reorganized as time passes after learning. Memory is initially dependent on the hippocampus formation, but its role diminishes as a more permanent memory is gradually established elsewhere, probably in neocortex. Identification of the Components of the Medial Temporal Lobe Memory System Information about which structures and connections are important for memory (and that when damaged produce amnesia) comes from three sources: studies of neurological patients with circumscribed memory impairment, systematic experimental work with an animal model of human amnesia in the monkey, and studies of the effects of selective lesions in rats. Memory-Impaired Patients Cognitive studies of memory impairment have provided valuable information about the organization of memory functions (Baddeley, 1982; Cermak, 1982; Milner, 1972; Schacter, 1985; Squire, 1986; Weiskrantz, 1987). The most informative cases have been those where the amnesia occurs against a background of normal intellectual function and intact immediate memory. The hallmark of the disorder is profound forgetfulness for new material (anterograde amnesia) and some loss of previously acquired information (retrograde amnesia). Until recently, comparatively little was known about what neuropathological changes in the medial temporal lobe had occurred in the patients being studied. Several single-case studies attributed memory impairment to hippocampal damage (Cummings, Tomiyasu, Read, & Benson, 1984; DeJong, Itabashi, & Olson, 1968; Duyckaerts et al., 1985; Victor, Angevine, Mancall, & Fisher, 1961), but the assessment of memory functions ia these cases was often informal or incomplete. In addition, the damage was often not restricted to the hippocampus but extended into the amygdala, the parahippocampal gyrus, and other structures. The findings from a carefully studied single case have placed the matter on firmer ground. Patient R. B. became amnesic in 1978 at the age of 52 as the result of an ischemic event that occurred following open-heart surgery (Zola-Morgan, Squire, & Amaral, 1986). Ischemia (ISC) refers to a condition during which the blood supply to the brain is insufficient. In R. B.'s case, a tear occurred in the atrium of the heart. R. B. survived for 5 years after the ischemic event, during which time his cognitive functions were repeatedly evaluated and his memory impairment was documented. The only cognitive deficit that was noted was moderately severe memory impairment. Examination of R. B.'s brain after his death in 1983 revealed a lesion in the CA1 region of the hippocampus (Zola-Morgan et al., 1986, Figure 1). The lesion was bilateral and extended the full rostrocaudal extent of the hippocampus. There was some other minor pathology, but the only finding that could reasonably be associated with the amnesia was hippocampal damage. This case thus showed that damage limited to the hippocampus is sufficient to cause easily detectable and clinically significant memory impairment. Recently, another case has been reported of memory impairment associated with a bilateral lesion of the hippocampus (Victor & Agamanolis, 1990). The CA1 region of the hippocampus is especially vulnerable to ischemic damage. Thus, global ischemia in the rat also produces selective neuronal loss in the CA1 region together with memory impairment (Auer, Jensen, & Whishaw, 1989; Davis & Volpe, 1990). Also, as discussed later, bilateral CA1 damage and memory impairment can be found after global ischemia in the monkey (Zola-Morgan & Squire, 1990a; Zola-Morgan et al., in press). The findings from R. B., and the observed effects of global ischemia in the rat and monkey, provide compelling evidence that the hippocampus proper is essential for mammalian memory. The same conclusion is now strongly supported by findings of memory impairment in rats following surgical damage limited to the hippocampus when appropriate tasks are used (Barnes, 1988; Eichenbaum, Mathews, & Cohen, 1989; Olton et al., 1979; Sutherland & Rudy, 1989). The findings with rats are discussed more fully in a later section. Additional confirmation for the idea that the human hippocampus is important for memory has come from recent improvements in magnetic resonance (MR) imaging, which make it possible to obtain anatomical information in living patients (Figure 1). A high-resolution protocol for imaging human hippocampus was developed that permits visualization of the hippocampal formation in considerable detail (Press, Amaral, & Squire, 1989). Using this protocol, abnormalities in the hippocampus were demonstrated in 4 patients with circumscribed memory impairment (Squire, Amaral, & Press, 1990). Specifically, in the patients, the region of the hippocampus (defined as the fimbria, dentate gyrus, hippocampus proper, and subiculum) appeared markedly shrunken and atrophic (57% of normal size). In contrast, the area of the temporal lobe excluding the hippocampal region was normal. Thus, the MR technique has been able to provide direct visual evidence of hippocampal damage in patients with a selective memory disorder. One important finding was that neither patient R. B. nor the 4 other amnesic patients studied with MR imaging were as severely memory impaired as the well-studied surgical patient H. M. (Scoville & Milner, 1957). This observation suggests that the severity of H. M.'s memory impairment resulted from damage to medial temporal lobe structures other than or in addition to the region of the hippocampus itself. Recently, it has
MEMORY AND THE HIPPOCAMPUS 19 pane righ (Fron ganiz ead8pomtnmnainthtteatdeawuwiagnaninal man amnes I the monkev s the subicular complex Memory Impairment in Nonhuman Primate order to reach the hippo f th ampal gyrus.This more estricted lesion has b to mimic the surgical er to the hippocampus and+to the Monkeys t ol-Moran 198 ola-Morga squ 1086.7E o failed by hu Saunders,&Malamut.9:Squire&Zola-Morgan,9 for onofeasy object discriminations,eight-pairc era of r lays tested up to0s(Zoa-Morgan&Squire,199b).It is useful the sia in the ed frequently in earl on the an t pairs to be learned together,are among the tasks sensi d the sequence ially imp rtant in und ding patient H.M.s am (10 to 20 trials for normal monkeys).Concur ent discrimina esia.A lesion of the hippocampal formation isordinarily pro tion tasks are ones in which pairs of objects are pre
MEMORY AND THE HIPPOCAMPUS RR 197 Figure 1. Top left panel: Section through the hippocampus of a normal subject. Top right panel: Section through the hippocampus of amnesic patient R. B. showing damage to the CA1 region. Bottom left panel: Magnetic resonance scan of a normal subject (resolution = .625 mm). Several anatomical features of the hippocampal formation can be distinguished. Bottom right panel: Magnetic resonance scan of amnesic patient W H. using the same protocol. The hippocampal formation is markedly reduced in size. The calibration bars to the right represent 5 cm in 1-cm increments. (From "Memory: Organization of Brain Systems and Cognition" by L. R. Squire, S. Zola-Morgan, C. B. Cave, F. Haist, B. Musen, and W Suzuki, 1990, Cold Spring Harbor Symposium on Quantitative Biology, 55, p. 1012. Copyright 1990 by Cold Spring Harbor Laboratory. Reprinted by permission.) become possible to confirm this idea directly using an animal model of human amnesia in the monkey. Memory Impairment in Nonhuman Primates In 1978, it was reported that a large medial temporal lobe lesion in monkeys, which was intended to mimic the surgical lesion sustained by patient H. M., caused severe memory impairment (Mishkin, 1978). The lesion included the amygdala, the hippocampus (including the dentate gyrus and subiculum), and surrounding cortical regions (the H+A + lesion, Figure 2). Although more work was needed before the impairment was well understood (Mahut & Moss, 1984; Mishkin, Spiegler, Saunders, & Malamut, 1982; Squire & Zola-Morgan, 1983; for recent reviews, see Squire & Zola-Morgan, 1991; Zola-Morgan & Squire, 1990b), Mishkin's 1978 publication was the first in a new era of research on the anatomy of memory, and in this sense it signaled the successful development of an animal model of human amnesia in the nonhuman primate (Table 1). The H+A + lesion was used frequently in early work on the animal model of human amnesia. However, the effects of more limited lesions involving the hippocampal formation were also of interest, primarily because of the early suggestion (Scoville & Milner, 1957) that damage to the hippocampal region might be especially important in understanding patient H. M.'s amnesia. A lesion of the hippocampal formation is ordinarily produced by a direct surgical approach through the ventral surface of the brain that damages the hippocampus proper, the dentate gyrus, the subicular complex, together with the underlying cortex that is necessarily removed in order to reach the hippocampus, that is, the posterior entorhinal cortex and much of the parahippocampal gyrus. This more restricted lesion has been termed H+ , where H refers to the hippocampus and + to the underlying cortex (see Figure 2). Monkeys with the FT lesion are impaired on a variety of memory tasks (Mahut, Moss, & Zola-Morgan, 1981; Moss, Mahut, & Zola-Morgan, 1981; Zola-Morgan & Squire, 1986; ZolaMorgan, Squire, & Amaral, 1989a) that are also failed by human amnesic patients (Squire, Zola-Morgan, & Chen, 1988; see Table 2). The tasks used to demonstrate memory impairment include retention of easy object discriminations, eight-pair concurrent discrimination learning, and delayed response with delays tested up to 30 s (Zola-Morgan & Squire, 1990b). It is useful to emphasize that simple object-discrimination tasks and concurrent object-discrimination tasks, which require several object pairs to be learned together, are among the tasks sensitive to FT" lesions. Simple object-discrimination tasks are ones that present to the animal two easily distinguishable objects. A choice of the correct object is rewarded, and the sequence is repeated until animals choose the rewarded object consistently (10 to 20 trials for normal monkeys). Concurrent discrimination tasks are ones in which different pairs of objects are pre-
198 LARRY R.SQUIRE 鸿 TEO CBL PONS ng the com e in the perirhi ndicated all ofth cal In the fis ddle d N rk:New fs ht 1990 b by pe Memory S em"by I R.So nd S.Z 253.. 1380.Copyrieht 1991 by the steny the rewarded pair (several hundred n succeeding trial. Monkeys and lea eF in98 when the delay between presenta Mishkin valier,1984: ire Zola-Mo 1990).The s cores of both normal and operate d animals ar tain skill-ba orhabit-basda Mala (Mahut.Zola-MoanMoss 198 ce ofth a-Morg Squire,198 )The 1986Z01a Morga Squire,&Amaral,】 a than when mor The most widely used task sensitive has shkin,978).Also,the magnitude of the deficit,that s,the dif d by normal an nd test ng are done pos
198 A LARRY R. SQUIRE B UNIMOOAL AND POLYMOOAL ASSOCIATION AREAS (Frontal, temporal, and parietal lobes) Figure 2. Panel A: A ventral view of the left hemisphere of a monkey brain showing the components of the large H*A+ lesion that first established an animal model of human amnesia. H = hippocampus, A = amygdala; and + refers to the adjacent cortex underlying each structure [periamygdaloid cortex, perirhinal cortex, entorhinal cortex, and parahippocampal cortex]. The view shows the amygdala [square-shaped black-and-white plaid], the hippocampus [black], and the underlying cortical regions typically included in surgical ablations of these structures. Large dots = perirhinal cortex; horizontal lines = periamygdaloid cortex; diagonal lines = entorhinal cortex; fine dots = parahippocampal cortex. Panel B: A schematic view of the structures of the medial temporal lobe important for declarative memory. The entorhinal cortex is the major source of inputs to the hippocampus. Approximately two thirds of the input to entorhinal cortex originate in the perirhinal and parahippocampal cortices. The entorhinal cortex also receives other direct projections from orbital frontal cortex, cingulate cortex, insular cortex, and superior temporal gyrus. As indicated, all of these projections are reciprocal. In the figure, the area designated hippocampus includes dentate gyrus, the cell field of the hippocampus proper, and the subicular complex, sts = superior temporal sulcus; amis = anterior middle temporal sulcus; pmts = posterior middle temporal sulcus; ios = inferior occipital sulcus; CBL = cerebellum; OLF = olfactory bulb; OC = optic chiasm. (Panel A: From "Neuropsychological Investigations of Memory and Amnesia: Findings From Humans and Nonhuman Primates," p. 442, by S. Zola-Morgan and L. R. Squire, 1990, in A. Diamond, The Development and Neural Bases of Higher Cognitive Functions, New York: New York Academy of Sciences. Copyright 1990 by the New York Academy of Sciences. Reprinted by permission. Panel B: From "The Medial Temporal Lobe Memory System" by L. R. Squire and S. Zola-Morgan, 199\, Science 253, p. 1380. Copyright 1991 by the American Association for the Advancement of Science. Reprinted by permission.) sented successively (e.g., eight pairs of objects, five times each during 40 daily trials). Training continues until animals choose consistently the rewarded object of each pair (several hundred trials for normal monkeys). In contrast to these tasks, the learning of pattern discrimination tasks and learning of the 24-hr concurrent discrimination task (Malamut, Saunders, & Mishkin, 1984) are not affected by even larger removals within the medial temporal lobe (for discussions of this difference, see Mishkin, Malamut, & Bachevalier, 1984; Squire & Zola-Morgan, 1983). The point is that monkeys with H+ lesions can succeed at certain skill-based or habit-based tasks (Mishkin, Malamut, & Bachevalier, 1984; Zola-Morgan & Squire, 1984). The significance of this finding is developed later in the article. The most widely used task sensitive to H1 " lesions has been delayed nonmatching to sample. This task requires an animal to remember a single visual object across a delay (up to 10 min) and then to demonstrate recognition of the object at the end of the delay. Recognition is tested by presenting the animal with a two-choice test (the original object and a new one) and rewarding a choice of the new object. New pairs of objects are used on each succeeding trial. Monkeys with № lesions are impaired on delayed nonmatching to sample (Figure 3), as well as on the other tasks mentioned earlier. Performance is good when the delay between presentation of the sample object and the choice is short and it becomes poorer as the delay increases (Overman, Ormsby, & Mishkin, 1990). The scores of both normal and operated animals are lower when monkeys are trained and tested postoperatively (Mahut, Zola-Morgan, & Moss, 1982; Zola-Morgan & Squire, 1986; Zola-Morgan, Squire, & Amaral, 1989a) than when monkeys are first trained preoperatively and then tested postoperatively (Mishkin, 1978). Also, the magnitude of the deficit, that is, the difference between the scores obtained by normal and operated animals, appears numerically larger when training and testing are done postoperatively, as compared with preoperatively. However, signal-detection analysis suggests that the defi-
MEMORY AND THE HIPPOCAMPUS 199 Table 1 That Have Been Produced in Monkeys With Reference Selective of memory(decarative) skil-basme n:278eS9t18 diate memory is spared zola-M d L ad Am Diam of Sciences Adapted by permission. cisntheemocodiosaeaaaiCRnonl193817T86 etheen of the task and the abilitytoemere medial temporal together as the performance ceiling is approached.The tures are important for n mory func ions can be obt ed by ups of animais.For xamp in studies of monkeys with 、skills that oud assist in t s a dela ether the n emory imp ess sever medial temn al lohe el.Clo Amaral.Zola-Morgan.&Squire.199:Zola motorskills(Salmon,Zola-Morgan,&Squire,197),andit facil Morgan et al,in press).Ischemia was produced by 15-mi Table 2 Performance of Amnesic Patients and Monkeys With HA Lesions on the Same Tasks Amnesic patients Monkeys with H+A*lesions Te对 + Reference Delayed nonmatching to sample Mishkin,197:Zola-Morgan g.garconcurTentdiscrimination Squire et al.,1988:Oscar-Berman + Zola-Morgan&Squire,1985 omcmtmo Motor skill learning la-Morgar 8nrmtisgnindicatsnoimpair of the s for fullerd ons of
MEMORY AND THE HIPPOCAMPUS 199 Table 1 Characteristics of Human Amnesia That Have Been Produced in Monkeys With Large Bilateral Medial Temporal Lobe Removals Characteristic Reference Selective loss of one kind of memory (declarative) Sparing of skill-based memory Severity of memory impairment dependent on locus and extent of damage Immediate memory is spared Memory is impaired when the number of stimuli exceeds immediate memory capacity Distraction exacerbates the memory impairment Memory impairment is modality general Memory impairment can be enduring Malamut, Saunders, & Mishkin, 1984; ZolaMorgan & Squire, 1984, 1985 Malamut etal, 1984; Zola-Morgan & Squire, 1984 Mishkin, 1978; Zola-Morgan & Squire, 1985,1986 Mishkin, 1978; Overman, Ormsby, & Mishkin, 1990; Zola-Morgan & Squire, 1985 Zola-Morgan & Squire, 1985 Zola-Morgan & Squire, 1985 Murray & Mishkin, 1984 Zola-Morgan & Squire, 1985 Note. References are to representative studies and are not exhaustive. From "Neuropsychological Investigations of Memory and Amnesia: Findings From Humans and Nonhuman Primates" (p. 438) by S. Zola-Morgan and L. R. Squire, 1990, in A. Diamond, The Development and Neural Bases of Higher Cognitive Functions, New York: New York Academy of Sciences. Copyright 1990 by the New \fork Academy of Sciences. Adapted by permission. cits in these two conditions are equivalent (Ringo, 1988). Preoperative training improves postoperative performance of both normal and operated animals and brings their scores closer together as the performance ceiling is approached. The benefit of preoperative training for postoperative performance is probably due to postoperative savings of some preoperatively acquired information about the rules of the task as well as postoperative retention of certain skills that could assist in the task of remembering a new sample object across a delay (e.g., attention and immobility). Some of this information probably survives medial temporal lobe surgery, as has been shown directly for motor skills (Salmon, Zola-Morgan, & Squire, 1987), and it facilitates the relearning of the task and the ability to remember new objects across a delay. Useful information about which medial temporal lobe structures are important for memory functions can be obtained by comparing the severity of memory impairment in different groups of animals. For example, in studies of monkeys with ischemic damage to the hippocampus, one can ask whether a detectable memory impairment is produced in the monkey and whether the memory impairment, if detectable, is less severe or more severe than the impairment associated with the H4 " lesion (Rempel, Glower, Amaral, Zola-Morgan, & Squire, 1991; ZolaMorgan et al., in press). Ischemia was produced by 15-min Table 2 Performance of Amnesic Patients and Monkeys With H+A* Lesions on the Same Tasks Amnesic patients Test Reference Monkeys with H+A + lesions Reference Delayed nonmatching to sample + Retention of object + discrimination 8-pair concurrent discrimination + Object reward association + 24-hr concurrent discrimination + Motor skill learning - Pattern discrimination + Squire, Zola-Morgan, & Chen, 1988; + Oscar-Herman & Bonner, 1985 Squire, Zola-Morgan, & Chen, 1988 + Squire et al., 1988; Oscar-Berman & + Bonner, 1985 Squire et al., 1988 + Squire etal., 1988 Pursuit rotor task; Brooks & - Baddeley, 1976 Predicted outcome; not yet tested - Mishkin, 1978; Zola-Morgan & Squire, 1985 Zola-Morgan & Squire, 1985 Zola-Morgan & Squire, 1985 Phillips & Mishkin, 1984 Malamut, Saunders, & Mishkin, 1984 Lifesaver task; Zola-Morgan & Squire, 1984 Zola-Morgan & Squire, 1984 Note. References are to representative studies and are not exhaustive. Plus sign indicates impairment; minus sign indicates no impairment. Monkeys may approach the 24-hr concurrent-discrimination task and the pattern-discrimination task differently than humans approach these two tasks. Humans try simply to memorize which stimulus is correct and which is incorrect (i.e., using declarative memory). Monkeys gradually learn incrementally, perhaps by gradually strengthening associations or by "tuning in" relevant dimensions of the stimuli (for fuller discussion, see Zola-Morgan & Squire, 1984). From "Neuropsychological Investigations of Memory and Amnesia: Findings From Humans and Nonhuman Primates" (p. 446) by S. Zola-Morgan and L. R. Squire, 1990, in A. Diamond, The Development and Neural Bases of Higher Cognitive Functions, New York: New York Academy of Sciences. Copyright 1990 by the New \fork Academy of Sciences. Adapted by permission
200 LARRY R.SQUIRE A100 NMTS-1 B100 NMTS-2 A= 3 0 10m Delay (s Delay (s) Per (NMTS) ask h N s with h ent to th ala(H) h the fir the num (NMTS-2).The perfo e left pane may und was alw The HA Copyright 1991 by the American on for the Advancement of Sci ence.Reprinted by permis ca duced hy tid on Mor cted ainly to the CA field of the hippocam sional histologica damag coul be ete ted in the e additional isct a ffe ry fund thi na that this da was not de alth han the mpair at ed wi surgical esion of the able to illuminate this issue to some extent ed non to atch with ischem inareas important for mer mpaired to monkeys wit ation).monkeys in 06 all tha the H t as H surgi ns.Ac rdingly,it is impla vere produ using stere oordi ate establishe R B)had wide cha affecting memor and bi the un as not er,Zola gan, lated to memory function than is involvec M ith the performed nila to r gh the H lesion produ a co nsiderable deg the tas after a large ial lesions,taken together with the the H? lesion;Mi shkin,1978:Zola-Morgan Squire,1985
200 A 100 r 90 0 80 1 > S. 70 I so 50 LARRY R. SQUIRE NMTS-1 B I00r 90 80 70 60 50 NMTS-2 A=3 15 60 10nwi Delay (s) 15 60 10min Delay (s) Figure 3. Performance on the delayed-nonmatching-to-sample (NMTS) task by normal monkeys (N), monkeys with lesions of the amygdala (A), monkeys with damage to the hippocampal formation (H*), monkeys with conjoint lesions of the hippocampal formation and the amygdala (H*A), monkeys with large medial temporal lobe resections (H*A+ ), and monkeys with lesions of the hippocampal formation and the perirhinal cortex adjacent to the amygdala (H**). The numbers in the figure show the number of animals in each group. Performance was tested approximately 1 month after surgery (NMTS-1) and then again 1 to 2 years after surgery (NMTS-2). The performance curve for the H*+ group in the left panel may underestimate the memory deficit, because one animal in this group required a remedial procedure in which the sample object was always presented twice instead of once. The WA* group was tested only once. (From "The Medial Temporal Lobe Memory System" by L. R. Squire and S. Zola-Morgan, 1991, Science, 253, p. 1382. Copyright 1991 by the American Association for the Advancement of Science. Reprinted by permission.) bilateral carotid occlusion together with pharmacologically induced hypotension. Monkeys prepared in this way had cell loss restricted mainly to the CA1 field of the hippocampus and to somatostatin-containing cells in the dentate gyrus. Only minor, occasional histological damage could be detected in other brain regions. Monkeys with this lesion (ISC) also had impaired memory, although the impairment was less severe overall than the impairment associated with surgical lesions of the hippocampal formation (the H+ lesion). Specifically, on the delayed nonmatching to sample task monkeys with ischemic lesions were impaired to about the same degree as monkeys with H + lesions. However, on two other tasks (object discrimination and eight-pair concurrent discrimination), monkeys in the ISC group performed better (p = .06) than monkeys with H* lesions. It has also been possible to compare monkeys with ISC lesions to monkeys with selective lesions of the hippocampus, which were produced using stereotaxic coordinates established by MR imaging (the H lesion). The H lesion damaged the hippocampus, dentate gyrus, and subiculum, but spared the underlying cortex (Alvarez-Royo, Glower, Zola-Morgan, & Squire, 1991; Glower, Alvarez-Royo, Zola-Morgan, & Squire, 1991). Monkeys with the H lesion performed similarly to monkeys with ISC lesions across all the tasks and significantly better than JH* monkeys on two of the tasks. The findings for ISC lesions, taken together with the findings for H and H* lesions, show that even incomplete damage to the hippocampus is sufficient to produce detectable memory impairment in monkeys, just as it can in humans (patient R. B.). It has been difficult to rule out entirely the possibility that some additional ischemic damage affecting memory functions did occur in patient R. B. and that this damage was not detected in histological analysis. The results with ischemic monkeys are able to illuminate this issue to some extent. Specifically, if significant additional damage had occurred in the ISC monkeys in areas important for memory function, one would expect the memory impairment to have approximated more closely or even to have exceeded the memory impairment associated with surgical lesions of the hippocampal formation. However, the ISC lesion produced less severe memory impairment overall than the H+ surgical lesion and about the same level of impairment as H surgical lesions. Accordingly, it is implausible that the ischemic animals (and by analogy the ischemic patient R. B.) had widespread pathological change affecting memory that was not subsequently detected by histological examination. The pathology in the ischemic animals must involve less tissue in structures related to memory function than is involved in the № lesion itself. Although the H* lesion produces a considerable degree of memory impairment, the level of the deficit is unmistakably greater after a larger bilateral medial temporal lobe removal (the H+A + lesion; Mishkin, 1978; Zola-Morgan & Squire, 1985;
MEMORY AND THE HIPPOCAMPUS 201 Zola-Morgan et al 1989a).The finding that Ha*lesi plex but spared adiacent cortex (the a lesion).Monkeys with the A lesion performed normally on four tasks(delaye rrent diser sample. ed res e (Zola-Mo why H.M.is Squire,&Amaral,1989b).I s with Hor R.B.H.M.sustained an HA es but R.B sustained a monkey i ch tures are damaged in the HA'lesion but not i n th abgmpairedontear mory tasks but their impairmen he (Mishkin.1978:Murra &Mishkin.1985 ne did not im oair me Sau ory impair 2 ing cortex At the same These findings e severe r ve and th of the to the ight b the ala.This r supposing that it shoud sibility was tested dicetyintwodiferentwas. t of study to includ theant and much of the pe rhinal cortex (theH tion in th al cortex was re sibly cooled (Horel d nonmatching to sample,the n surg ere a had a le were prod e3)and ex(Horel,Pytko- ere than th imairmentfolov ing either Hor HA lesions rito m neuroe mic elt d memory impa an,1987 The ent rtex wa already known to be also as impai on le of the idea that the ampal gyrus provide ear important for mer was mo inal and parahippoca nal co for between the cnipandaa9ccoTeieume2。 cribed al and pa ippocam ith th natomi amygdala Substantial eto perirhina that spared the hippoc mpus,theam and the 1989)On the hasis of this obs le that d age to peri cortex ad ent to the other cortica nput to ent ninal cor to make more limited ns di cted at the eda r ks they tching t nle ob that con ala her to test the rol d foll nd H lesions.Histolog d. that on anato onty to the rahip ctions to the orbitofrontal cortex separate co s to is u the severe mem ory def on memo ofthe
MEMORY AND THE HIPPOCAMPUS 201 Zola-Morgan et al., 1989a). The finding that H*A+ lesions produce more severe memory impairment than H*" lesions (Figure 3) implies that some of the damage produced by the tPA"1 " lesion, which is not included in the H+ lesion, are important for memory functions. Herein lies the explanation for why H. M. is more amnesic than other amnesic study patients, including R. B. H. M. sustained an H+A + lesion, but R. B. sustained a lesion involving only a portion of the hippocampus. Considerable effort has been directed toward identifying which structures are damaged in the EPA"1 " lesion but not in the H* lesion. The early evidence on this point seemed to point to the amygdala (Mishkin, 1978; Murray & Mishkin, 1985; Saunders, Murray, & Mishkin, 1984). However, all these early studies were based on surgical groups in which the amygdala was removed together with underlying cortex. At the same time, the cytoarchitectonics and connectivity of the underlying cortex were incompletely understood, and there was little basis for supposing that it should contribute to memory function. The underlying cortex was only incidentally involved in these surgical procedures, not a target of study. An early hint that the underlying cortex might play a role in memory came from behavioral studies of monkeys in which anteroventral temporal cortex was reversibly cooled (Horel & Pytko, 1982). When surgical lesions of this area were produced, the most affected animal had a lesion that involved perirhinal cortex (Horel, Pytko-Joiner, Voytko, & Salsbury, 1987). The possibility of studying this cortical region improved considerably when the territory of the perirhinal cortex was denned by modern neuroanatomical methods, and its connectivity with the hippocampal formation was established (Insausti, Amaral, & Cowan, 1987). The entorhinal cortex was already known to be the source of the major afferent projection to the hippocampus and dentate gyms. The important newer finding was that the adjacent perirhinal and parahippocampal gyrus provide nearly two thirds of the cortical input to the entorhinal cortex. Perirhinal and parahippocampal cortex are therefore essential for the normal exchange of information between the neocortex and the hippocampal formation. With these anatomical facts in mind, we reexamined a group of H*A+ monkeys (Zola-Morgan, Squire, & Mishkin, 1982) that had been prepared with the standard surgical approach to the amygdala. Substantial damage to perirhinal cortex had occurred in all the animals (see Zola-Morgan, Squire, Amaral, & Suzuki, 1989). On the basis of this observation, it seems plausible that damage to perirhinal cortex adjacent to the amygdala would also have occurred during the similar surgical approach used to make more limited lesions directed at the amygdala itself or the amygdalofugal pathway (Bachevalier, Saunders, & Mishkin, 1985; Mishkin, 1978). In short, it appears that the lesions in the earlier studies that were intended to test the role of the amygdala had damaged cortex in addition to the amygdala that on anatomical grounds might be expected to have a role in memory function. It thus became important to determine experimentally the separate contributions to memory impairment of amygdala damage and damage to underlying cortex. To evaluate the effect of amygdala damage itself on memory, we developed a surgical procedure involving bilateral stereotaxic lesions, which damaged virtually all the components of the amygdaloid complex but spared adjacent cortex (the A lesion). Monkeys with the A lesion performed normally on four memory tasks (delayed nonmatching to sample, retention of object discriminations, concurrent discrimination, and delayed response (Zola-Morgan, Squire, & Amaral, 1989b). In contrast, monkeys with H+ or H*A+ lesions were impaired on all four tasks. We also evaluated monkeys who had bilateral lesions of the hippocampal formation (H*) made conjointly with circumscribed lesions of the amygdaloid complex (the H"A lesion). The H*A monkeys were also impaired on the four memory tasks, but their impairment was no greater than after H* lesions. Thus, amygdala damage alone did not impair memory; nor did it exacerbate the memory impairment associated with damage to the hippocampal formation (Figure 3). These findings suggested that more severe memory impairment observed after H*A+ lesions might be attributable to damage to the cortical regions that surround the amygdala. This possibility was tested directly in two different ways. First, monkeys were prepared with H*" lesions that were brought forward to include the anterior entorhinal cortex and much of the perirhinal cortex (the H1 " 1 " lesion). The intention in this group was to reproduce as much of the H+A+ lesion as possible but to leave the amygdala intact. On delayed nonmatching to sample, the impairment associated with H*4 ' lesions was nearly as severe as that following H"A+ lesions (Figure 3) and significantly more severe than the impairment following either H*" or H*A lesions (Squire & Zola-Morgan, 1991). Thus, lesions of the cortex surrounding the amygdala, but not lesions of the amygdala itself, exacerbated memory impairment in monkeys following lesions of the hippocampal formation. The H4 " 1 " monkeys were also as impaired as №A+ monkeys on a second task: object-discrimination learning. The second test of the idea that the cortical regions surrounding the amygdala are important for memory was motivated by current understanding of the anatomical connections of the hippocampus and adjacent cortex (Figure 2). In particular, as described earlier, perirhinal and parahippocampal cortex are major routes by which information is exchanged between the neocortex and the hippocampal formation. Accordingly, monkeys were prepared with bilateral lesions limited to the perirhinal cortex and parahippocampal gyrus (PRPH) that spared the hippocampus, the amygdala, and the entorhinal cortex (Zola-Morgan, Squire, Amaral, & Suzuki, 1989). The white matter underlying perirhinal cortex was also transected in an attempt to remove other cortical input to entorhinal cortex. These monkeys were severely impaired on the three memory tasks they were given (delayed nonmatching to sample, object discrimination, and concurrent discrimination). In general, the impairment following PRPH lesions was similar to that observed following tfA+ lesions and H1 " 1 " lesions. Histological analysis showed that, as intended, damage had occurred not only to the perirhinal and parahippocampal cortex but also to projections to the entorhinal cortex from orbitofrontal cortex, superior temporal gyrus, insula, and cingulate gyrus. It is unlikely that the severe memory deficit after either PRPH lesions or H^ lesions resulted from indirect effects of these lesions on the function of the amygdala. This possibility merits consideration because perirhinal cortex does originate direct projections to the amygdala (Amaral, 1987; Van Hoesen
202 LARRY R.SQUIRE 1981).However,there are two difficulties with the idea that x and thm ory Seco stud ies of d 1987 Van Ho 1982).The entorhi ven meme ry tests The findin was that eithe artial or co ical inpu and th d b endency to approach or touch stimul sobjects(亿ola-Morga oad. e of th to b ouble dissociat keys empha motional beavior. so hipp mpal dam ely all operat d gro obe.Ot er recent -ite e ited memory impair ever,unless the amygdala was 1989:Squire,Ojemann,Mic n,Peter taken to cther with the ing of intact me ory after circumscrib system influe primarily through its reciproca iated with large medial temporal lobe to the major efferent system of the hi am s)re ults from da to the hip al form the fornix and damage to the m not ae to the hippo ampus a and amygdala.One imp on mem ory using 8 th same tasks (Aggleton&Mishkin nt to he hir s is not sim rfindings do not des icribe thes of me on from tex to .bu ute sense。. t they make the impc tant p the H++l is less severe than after damage to the hipp campal than hus. pears tha info 4)The es adia Mishkin.983)s inal,perirhin and parahipp ampal cortex)app maged to amnesia (for review sec Zola-Mor ion This idea explains why mer ment can he in functions are the medial do the creased by lesions in structures adjacent to the hippocampus. the internal medullary lamina,and the mammillothalami UNIMODAL AND POLYMODA ASSOCIATIONAL AREAS FRONTAL LOBE TEMPORAL LOBE PARIETAL LOBE ity ofthe perirhir d the the fro SUB CAl ar L R. 990.Cold Harbor Laboratory.Reprinted by permis ion)
202 LARRY R. SQUIRE 1981). However, there are two difficulties with the idea that damage to these projections contributed to the memory impairment. First, removal of the amygdala itself had no effect on memory. Second, quantitative studies of emotional behavior have been carried out with the same operated groups that were given memory tests. The finding was that either partial or complete damage to the amygdala caused readily detectable changes in emotional behavior as evidenced by an abnormal tendency to approach or touch stimulus objects (Zola-Morgan, Squire, Alvarez-Royo, & Clower, 1991). Yet, monkeys with PRPH lesions exhibited normal emotional behavior. Indeed a double dissociation was found. Among six operated groups, the groups of monkeys with amygdala damage exhibited abnormal emotional behavior. Unless there was also hippocampal damage, memory was unaffected. Conversely, all operated groups with damage to the hippocampus or its associated cortex exhibited memory impairment. However, unless the amygdala was also damaged, emotional behavior was normal. The findings from PRPH lesions and H++ lesions, taken together with the finding of intact memory after circumscribed amygdala damage, strongly suggest that the severe memory impairment associated with large medial temporal lobe lesions (tTA* lesions) results from damage to the hippocampal formation and adjacent anatomically related cortex, not from conjoint damage to the hippocampus and amygdala. One important implication of these studies with monkeys is that the cortex adjacent to the hippocampus is not simply a conduit for funneling information from neocortex to the hippocampus. This conclusion follows from the finding that the PRPH lesion and the H4 " 1 " lesion produced a more severe memory impairment than the IF lesion and also from the finding that the H+ lesion produced a more severe impairment than the H lesion. Thus, it appears that information from neocortex need not reach the hippocampus itself for some memory storage to occur (Figure 4). The cortical structures adjacent to the hippocampus (entorhinal, perirhinal, and parahippocampal cortex) appear to participate with the hippocampus in a common memory function. This idea explains why memory impairment can be increased by lesions in structures adjacent to the hippocampus. These cortical structures in the medial temporal lobe are sites of convergent projections from widespread unimodal and polymodal association areas in neocortex, and these connections are reciprocal (Amaral, 1987; Van Hoesen, 1982). The entorhinal cortex itself (which projects directly to hippocampus) receives direct cortical input from a limited number of cortical areas. The perirhinal and parahippocampal cortices (and the other cortical regions that project to entorhinal cortex) receive information from and send information to a much broader extent of neocortex. Thus, the system as a whole is likely to be privy to much of the processing that occurs in neocortex. In summary, the findings from work with monkeys emphasize the importance for memory functions of the hippocampal formation and the surrounding cortex of the medial temporal lobe. Other recent work in monkeys and humans is consistent with this proposal (Friedman & Goldman-Rakic, 1988; George, Horel, Cirillo, 1989; Squire, Ojemann, Miezin, Petersen, Videen, & Raichle, in press; Van Hoesen & Damasio, 1987). It seems likely that the medial temporal lobe memory system influences memory primarily through its reciprocal projections with widespread areas of neocortex. In separate studies, damage to the major efferent system of the hippocampal formation, the fornix, and damage to the major diencephalic target of the fornix, the mammillary nuclei, had only mild effects on memory using the same tasks (Aggleton & Mishkin, 1985; Bachevalier et al., 1985; Zola-Morgan et al., 1989a). These latter findings do not describe the severity of memory impairment in any absolute sense, but they make the important point that the impairment after fornix section or mammillary nuclei lesions is less severe than after damage to the hippocampal formation. Another region of the brain that when damaged produces amnesia is the medial thalamus. Medial thalamic lesions in the monkey produce severe memory impairment (Aggleton & Mishkin, 1983). It is not yet entirely clear which thalamic nuclei must be damaged to cause amnesia (for a review, see Zola-Morgan & Squire, in press). The areas most often linked to memory functions are the medial dorsal nucleus, the anterior nucleus, the internal medullary lamina, and the mammillothalamic UNIMODAL AND POLYMODAL ASSOCIATIONAL AREAS FRONTAL LOBE TEMPORAL LOBE PARIETAL LOBE Figure 4. Schematic representation of the connectivity of the perirhinal and the parahippocampal cortices in the monkey brain. The width of the arrows corresponds to the relative proportion of cortical inputs arising from the areas indicated. EC - eatorhinal cortex; DC = dentate gyrus; SUB = subicular complex; CA3 and CA1 are fields of the hippocanpus proper. (From "Memory: Organization of Brain Systems and Cognition" by L. R. Squire, & Zo1» Mocgin, C B. Cave, F. Haist, G. Musen, and W Suzuki, 1990, Cold Spring Harbor Symposium on Quantitative Biology, 55, p. 1019. Copyright 1990 by Cold Spring Harbor Laboratory. Reprinted by permission.)
MEMORY AND THE HIPPOCAMPUS 203 on Mishkin.1983:Graff-Radford.Tranel.Var amygdala lesions were typically produced stereotaxically This en. Amaral, gan,Krit edure makes it possible to prod amygdala les sons with &Sc uri.1985).Most of th is a repo ppocampa that conjoint lesions of amygdala and h pus impa extent diencephalicand medial temp al lobe pat ogy in h ons of hi edala had no effect mans and mon t (for ntof view :Victor animals .This up to delays of 60s(A et al,1989).Al ching- ple task used with gion may belong to a tightly linked functional s d for choosing the arm of a Y maz hat damage able arm on the starting arm anda second arm that diff Memory Impairment in Rats ight expect from studies of mo keys that perf at ha pocamp: cen work with rats and work with humansan et al 1989a)rats with hippocampalles mance ona varietyofm mory tasks.T mcthaoeCcdwHhlhelesioasOaepOSbifg discrimination task that reouin d,is tha of this wor been to c st non uman prin ectso hist ved hy ces in studiesdirectedatthisbe us and the ar la in orx The rtains to there is noe e that ygdala le n rat contr re hippo ampal ons or les hn ngs from mo eys,amely that the ory functi and that the am ygdala isnot part of this functiona work in what obs the hip ent of affe on alor (Aggletc d&MeD by thed nce Io Imull,as expr 1990).One important feature of the studies with rats is that toward conditioned stimuli (Davis,1986:Gaffan&Harrison Task Hippocampus Amygdala Reference 1990 1986 oto 0.1980
MEMORY AND THE HIPPOCAMPUS 203 tract (Aggleton & Mishkin, 1983; Graff-Radford, Tranel, Van Hoesen, & Brandt, 1990; Squire, Amaral, Zola-Morgan, Kritchevsky, & Press, 1989; Victor, Adams, & Collins, 1989; von Cramon, Rebel, & Schuri, 1985). Most of these thalamic regions have anatomical connections to either the hippocampal formation or the perirhinal cortex. It is also unclear to what extent diencephalic and medial temporal lobe pathology in humans and monkeys might produce different patterns of memory impairment (for two points of view, see Parkin, 1984; Victor et al, 1989). Although one would expect that these brain regions make different contributions to normal memory, each region may belong to a tightly linked functional system such that damage to any component causes rather similar kinds of impairment. Memory Impairment in Rats During the past several years, many points of contact have developed between work with rats and work with humans and nonhuman primates. In the rat, hippocampal lesions or lesions of related structures (fornix or entorhinal cortex) impair performance on a wide variety of memory tasks. These include spatial memory tasks, odor-discrimination learning, timing tasks, and discrimination tasks that require learning relationships between stimuli. One major focus of this work has been to characterize the kind of learning and memory that is impaired (see next section). Another focus has been to compare the effects of hippocampal lesions with the effects of lesions in adjacent structures. Studies directed at this second objective have obtained two important findings. The first pertains to the roles of the hippocampus and the amygdala in memory. The second pertains to the separate contributions of the structures and connections within the hippocampal formation. First, at least seven examples can be identified where hippocampal lesions or lesions of anatomically related structures produce an effect on memory, but amygdala lesions produce no impairment (Table 3). In addition, three studies have found that adding an amygdala lesion to a lesion of the hippocampal system did not increase the deficit beyond what was observed following the hippocampal lesion alone (Aggleton, Hunt, & Rawlins, 1986, Experiment 2; Eichenbaum, Pagan, & Cohen, 1986; Sutherland & McDonald, 1990). One important feature of the studies with rats is that amygdala lesions were typically produced stereotaxically. This procedure makes it possible to produce amygdala lesions without damaging cortical areas surrounding the amygdala. An apparent exception to this pattern of findings is a report that conjoint lesions of amygdala and hippocampus impaired performance on a task of object recognition, whereas separate lesions of hippocampus or amygdala had no effect, even when animals had to retain information up to delays of 60 s (Aggieton, Blindt, & Rawlins, 1989). This task was a modified version of the visual delayed nonmatching-to-sample task used with monkeys. Rats were rewarded for choosing the arm of a Y maze that differed visually from the starting arm. Multiple, removable arms were used so that on each trial rats saw one arm identical to the starting arm and a second arm that differed from the starting arm along several dimensions. Although one might expect from studies of monkeys that performance on such a task should be measurably impaired by hippocampal lesions alone (Mahut et al., 1982; Mishkin, 1978; Zola-Morgan et al, 1989a), rats with hippocampal lesions performed this task normally. Because hippocampal lesions alone did not disrupt performance on this task, it is difficult to interpret the impairment that occurred with larger lesions. One possibility, which has scarcely been explored, is that rats might approach some variants of the delayed nonmatching-to-sample task with a different strategy than nonhuman primates. For example, Sutherland and Rudy (1989) pointed out that this task could in principle be solved by two fundamentally different strategies (also see the note to Table 2 for the importance of strategy differences in how monkeys and humans accomplish pattern-discrimination learning). In any case, there is no evidence that amygdala lesions in rats impair performance on tasks that are also impaired by hippocampal lesions. The findings in rats are therefore in agreement with the findings from monkeys, namely that the hippocampus and related structures participate in a particular kind of memory function and that the amygdala is not part of this functional system. Indeed, work in both monkeys and rats suggests that the amygdala is important for other functions, including the acquisition of conditioned fear and the establishment of affective significance for neutral stimuli, as expressed, for example, by the development of conditioned responses that are directed toward conditioned stimuli (Davis, 1986; Gaffan & Harrison, Table 3 Effects of Lesions of the Hippocampal System or the Amygdala on Memory Tasks in the Rat Task Hippocampus Amygdala Reference Water maze Odor discrimination Timing of events Learning cue relationships Spatial alternation Nonspatial alternation Radial maze Sutherland & McDonald, 1990 Eichenbaum, Pagan, & Cohen, 1986' Olton, Meek, & Church, 1987' Sutherland, McDonald, Hill, & Rudy 1989 Aggleton, Hunt, & Rawlins, 1986; Aggleton, Blindt, & Rawlins, 1989 Raffaele & Olton, 1988" Becker, Walker, & Olton, 1980* Note. Plus sign indicates impairment; minus sign indicates no impairment. • These lesions damaged the fornix rather than the hippocampus itself
204 LARRY R.SQUIRE ald,199).It is pos sible.too,that the amygdala has a mor cit that can be are selected role in Mish 197 SK 78 d on amyedal y ofskill learn ing that t ir th Cohen pa /980 ation a thetinmcofretricralIincicitingrc call in anke in c patients cam emerge from ed hippocamp ons can be increa by additiona as the su lum or the alveus (Jar ard,98;Morris.Schenk ions fus ts (Gra uced of da 8er1978 on)As de ribed the s me is also true in hu unders d as an example of ord prim (eg patient R.B.vs.H.M) and a large body of work has accumulate both norma Multiple Memory Systems kind of memory than the kind that is mura in ider ctures and cor ections that nak up the oral I m ce m of dife 07 ates in m in thisachi ing,1985 Weisk 1987 ght that ind of ocamp 9)w d greatly b htmensy5notasneeeatmyrconsisisoritpilepe see Hintzman 99 pal ing v n m yis no can be found ine and human of dev lopm tablished In 1962 the se (Bereson 1911:me and of day-t o-day im y(Ryle, mory for the practice and proce ning and to fo the uni ary nature of the res men tory Amne on should be made for m amnesia In addition the idea that the ipp angs of u on the ha ag Spe esions 9974H h1974 OKeefe&Nad Picture s and experimental animals.in 968 1970,19 others that foll ved m nd related cture mes good o at least better than m ,1980)in thes ing to mind or dec ected, 1976.Tbe some aksare simply casier than oth asks c r D rkwithhuman sub has be al T ed that ce ce are one' in normal subjects and amnesi nts alike. ever arative men es men
204 LARRY R. SQUIRE 1987; Gallagher, Graham, & Holland, 1990; Kesner, in press; LeDoux, 1987; Nachman & Ashe, 1974; Sutherland & McDonald, 1990). It is possible, too, that the amygdala has a more general role in forming associations between stimuli, for example, in making associations across modalities (Murray & Mishkin, 1985). However, when these ideas are based on amygdala lesions prepared by a direct surgical approach, the contribution of underlying cortical regions included in the amygdala lesions needs to be evaluated. The second important and anatomically relevant finding to emerge from work with rats is that the deficit associated with restricted hippocampal lesions can be increased by additional damage to anatomically related structures and fiber tracts such as the subiculum or the alveus (Jarrard, 1986; Morris, Schenk, Tweedie, & Jarrard, 1990). These findings are in agreement with the findings from monkeys that different levels of impairment can be produced depending on the extent of damage within the medial temporal lobe (e.g., the H+ lesion vs. the H++ lesion). As described earlier, the same is also true in humans (e.g., patient R. B. vs. H. M.). Multiple Memory Systems Progress in identifying the structures and connections that make up the medial temporal lobe memory system has been paralleled by gains in understanding how this system participates in memory functions. An important step in this achievement was the insight that the hippocampal formation is important for only a particular kind of memory. The implication was that memory is not a single entity but consists of multiple processes or systems. Converging evidence about the selective role of the hippocampal formation in memory is now available from rats, monkeys, and humans. It took time for the idea of multiple memory systems to become firmly established. In 1962, the severely impaired amnesic patient H. M. was reported to be capable of day-to-day improvement in a hand-eye coordination skill, despite having no memory for the practice sessions (Milner, 1962). Nevertheless, subsequent discussions of memory in general and amnesia in particular tended to set aside motor skill learning and to focus on the unitary nature of the rest of memory. Amnesia was considered to impair memory globally, with the recognition that an exception should be made for motor skills. Findings of unexpectedly good learning by amnesic patients on tasks not requiring motor skills were also reported many years ago. Specifically, patients performed well when the retention test provided partial information (e.g., fragments) about previously presented pictures or words (Milner, Corkin, & Teuber, 1968; Warrington & Weiskrantz, 1968, 1970, 1974, 1978). However, there were two reasons why these reports, and others that followed, did not lead to the idea of multiple memory systems. First, although the performance of amnesic patients was sometimes good, or at least better than might have been expected, it was often below normal levels. Accordingly, the data were open to a proportionality interpretation, namely, that some tasks are simply easier than other tasks or provide more sensitive measures of memory. It could therefore be argued that certain task conditions simpJy improve performance in normal subjects and amnesic patients alike. Second, even when amnesic patients appeared to perform normally, the data could be interpreted as evidence that amnesia is a retrieval deficit that can be reversed when the appropriate tasks are selected (Warrington & Weiskrantz, 1970; Weiskrantz, 1978). Subsequently, it was discovered that motor skills are just one example of a broader category of skill learning that is intact in amnesic patients (Cohen, 1984; Cohen & Squire, 1980; Squire, 1982). At the same time, the success of partial information at the time of retrieval in eliciting recall in amnesic patients came to be better understood (e.g., word stems like inc or mot— as cues for recently studied words). It turned out that only one kind of instruction yields normal performance (complete the stem to form the first word that comes to mind; Graf, Squire, & Mandler, 1984). With conventional memory instructions (use the stem as a cue to recall a recently presented word), normal subjects maintain their advantage over amnesic patients (Graf et al., 1984; Squire, Wetzel, & Slater, 1978). Intact performance by amnesic patients on such tasks, when indirect instructions are used, is now understood as an example of word priming, and a large body of work has accumulated with both normal subjects and amnesic patients in support of the idea that priming reflects a different kind of memory than the kind that is tapped in conventional memory experiments (Shimamura, 1986; Tulving & Schacter, 1990). The emergence of the idea that memory consists of different systems (Cohen, 1984; Moscovitch, 1982; Schacter, 1987; Squire, 1982; Tulving, 1985; Weiskrantz, 1987; Wickelgren, 1979) was influenced greatly by work with amnesic patients. In addition, experimental work with normal subjects was influential (for reviews, see Hintzman, 1990; Polster, Nadel, & Schacter, 1991; Richardson-Klavehn, & Bjork, 1988). Distinctions between kinds of memory can be found in earlier writings that reflect the traditions of developmental psychology (Bruner, 1969; memory with record and memory without record), psychology (Bergson, 1911; memory and habit), philosophy (Ryle, 1949; knowing how and knowing that), and artificial intelligence (Winograd, 1975; Winston, 1977; declarative and procedural). The tradition of work with amnesic patients explains why the idea of multiple memory systems led naturally to a consideration of what kind of memory depends on the integrity of the brain structures, including hippocampus, that are damaged in amnesia. In addition, the idea that the hippocampus might be involved in only one kind of memory appeared independently in the animal literature, on the basis of the selective effects of limbic lesions (Gaftan, 1974; Hirsch, 1974; O'Keefe & Nadel, 1978; Olton et al., 1979). The sections that follow suggest that the findings from humans and experimental animals, including rats and monkeys, are now in substantial agreement about the kind of memory that depends specifically on the hippocampus and related structures. This kind of memory has been termed declarative (Cohen & Squire, 1980) in the sense that one can bring to mind or declare the content of this kind of memory (for its earlier use in psychology, see Anderson, 1976). The term declarative was derived from work with human subjects and has been difficult to apply usefully to experimental animals. The problem is not that declarative memory seems to imply an ability to declare one's knowledge verbally. Indeed, declarative memory includes mem-
