Exp.erimental Brain Research 9 Springer-Verlag 1993

Transcription

1 Exp Brain Res (1993) 96:457M72 Exp.erimental Brain Research 9 Springer-Verlag 1993 Neuronal activity related to visual recognition memory: long-term memory and the encoding of recency and familiarity information in the primate anterior and medial inferior temporal and rhinal cortex F.L. Fahy, I.P. Riches, M.W. Brown Department of Anatomy, University of Bristol, School of Medical Sciences, University Walk, Bristol, BS8 1TD, UK Received: 27 December 1992 / Accepted: 12 May 1993 Abstract. Recordings of the activity of 2705 single neurones were made in entorhinal and perirhinal cortex, area TG of the temporal lobe, and the inferior temporal cortex both during monkeys' performance of a serial recognition memory task using complex pictures and when monkeys were shown objects. Responses of 120 (9.7%) of the visually responsive neurones recorded were significantly smaller to the second than to the first presentations of unfamiliar stimuli. The incidence of such responses was highest in perirhinal cortex plus areas TEl and TE2 of the temporal lobe, intermediate in lateral entorhinal cortex and areas TE3 and TG, and lowest in other parts of entorhinal and inferior temporal cortex. Response decrements were maintained across 20 or more intervening presentations of other stimuli for the majority of the neutones tested. Responses of 43 (14.4%) of the visually responsive neurones tested were significantly greater to unfamiliar than to highly familiar stimuli. Such differential responses were found only in lateral entorhinal and perirhinal cortex plus areas TG, TEl, TE2 and TE3. For 6 neurones the response difference was significant even when the familiar stimuli had not been seen for more than 24 h: such neurones demonstrate access to information stored in long-term memory for more than 24 h. Seven familiarity neurones signalled information concerning the relative familiarity of stimuli but not information concerning how recently they were last seen; 58 recency neurones signalled information concerning the recency of presentation of stimuli, but not their relative familiarity. Thus certain neurones demonstrate the separable encoding of recency and familiarity information. Neurones signalling information of use for recognition memory are found in cortex close to the rhinal sulcus where lesions result in major deficits in the performance of recognition memory tasks. The conjunction of these findings provides strong evidence for the importance of these neurones and this cortex for processes (recency and familiarity discrimination) necessary for recognition and working memory. The possible relation of the neuronal responses to priming memory is also discussed. Correspondence to: M.W. Brown Key words: Memory: visual recognition or working memory - Medial temporal lobe - Perirhinal cortex - Hippocampal formation - Monkey - Single unit recording Introduction Recognition memory requires both the identification of stimuli and judgement about their prior occurrence. Judgements concerning the prior occurrence of stimuli may be made on the basis of various different types of information. These types of information include how long ago a stimulus was last encountered (recency), whether it is novel or has been often encountered before by the observer (familiarity), and in what circumstances it has been previously experienced (contextual cues). Thus, for example, it is possible to judge whether a stimulus is very familiar or relatively unfamiliar independently of the judgement of how long ago it was last seen. Judgements about the prior occurrence of stimuli are crucial to the performance of both recognition memory and working memory (in contrast to reference memory tasks (Olton et al. 1979). This report concerns the neuronal encoding of recency and familiarity information. The responses of neurones of the anterior and medial inferior temporal and rhinal cortex decrease on stimulus repetition (Brown et al. 1987; Riches et al. 1991b). The tendency to response decrement with stimulus repetition is so strong that it is evident in population measures of the response of neurones in the region (Riches et al. 1991b; Miller et al. 1991b). For individual neurones the response to unfamiliar stimuli that have not been seen for some time is substantially stronger than to familiar stimuli or to stimuli that have been seen recently. Further, such differences in response persist even when presentations of other stimuli have occurred between the first and the subsequent presentations of stimuli, i.e. the neurones demonstrate access to information held in memory even when the animal's attention has been distracted (Brown

2 458 et al. 1987; Riches et al. 1991b). Thus neurones in this medial temporal cortex signal information of use for visual recognition or working memory concerning the recency and/or familiarity of stimuli. In addition, the decremental responses provide information of potential use for priming memory (the facilitation of performance by a prior related experience). Impairment of recognition memory is a prominent feature of the loss of memory that follows lesions of the medial temporal lobes in humans (see for reviews: Parkin 1987; Squire 1987; Mayes 1988; Weiskrantz 1990; Squire and Zola-Morgan 1991), including the damage of early Alzheimer's disease (Pearson et al. 1985; Van Hoesen and Hyman 1990). Patients demonstrate impaired long-term memory, forgetting material once they have been distracted ("post-distractional amnesia"). A deficit in recognition memory can be demonstrated in monkeys as well as humans by using delayed matching or non-matching to sample tasks (Mishkin 1982; Zola-Morgan and Squire 1985a; Aggleton et al. 1988; Squire et al. 1988). However, it is only recently that major deficits in the performance of such tasks have been produced with relatively circumscribed lesions (Zola-Morgan et al. 1989; Murray 1991; Gaffan and Murray 1992), and that it has been demonstrated that such deficits are not due to an impairment in visual discrimination learning (Gaffan and Murray 1992). These recent lesion experiments are in agreement with earlier experiments employing cooling (Fuster et al. 1981; Horel et al. 1987). The recent studies have demonstrated the importance of the cortex surrounding the rhinal sulcus for performance of recognition memory tasks, and so have brought the lesion and cooling studies into close agreement with the electrophysiological findings. The conjunction of the electrophysiological and the lesion data provides powerful evidence for the importance of this cortical region to the performance of recognition memory tasks. This paper concerns the response capabilities of neurones in this and neighbouring regions. The objectives of the experiments reported here were: (i) to establish the boundaries of the region within which are found neurones that signal information related to recognition memory, (2) to record neuronal responses during the performance of an explicit recognition memory task that is similar to one used to study amnesia in both humans and monkeys (Gaffan 1974, 1992), and (3) to investigate the types and quality of information that the neuronal responses encode. The findings demonstrate that recency and familiarity information are separably signalled by the activity of neurones in a circumscribed region of the medial temporal cortex, and that certain neurones have access to information held in memory for more than 24 h. Abstracts of parts of this work have already appeared (Riches et al. 1990, 1991a). Materials and methods Subjects and behavioural tasks All animals were maintained under the supervision of a qualified veterinary officer. Three monkeys (Macaca mulatta) weighing kg were trained to perform a visual discrimination task and a serial visual recognition memory task, based on Gaffan's (1974) running recognition task. In these two tasks the monkey was required to make either a left or a right reach, contingent upon the stimulus that appeared on a video monitor. The stimuli were complex videodisc pictures (described more fully below). Each animal was first trained to perform the visual discrimination task using a pair of stimuli. Discrimination of several such pairs was taught before training began on the recognition memory task. The recognition task was taught by presenting novel stimuli which the animal had to discriminate from repeatedly used, highly familiar stimuli. The animal was then taught that a second presentation of a novel stimulus should be treated in the same way as the familiar stimuli. Gradually, the number of trials intervening between the first and repeat presentations of stimuli was increased until the serial task could be performed at greater than 90% accuracy ("90% correct"). Once trained an animal would perform more than 200 trials per day. In both tasks trials started with a cuing stimulus: the illumination for 0.5 s of a dim red neon light situated centrally at the bottom of the screen. After 0.5 s, a picture was displayed on the video monitor for 2 s. Reaches to the correct side made between 0.1 and 5.0 s after the offset of the stimulus were rewarded with fruit juice (approximately 0.5 ml). Other responses were counted as errors and the lights were dimmed for a brief time (between 0.2 and 10 s) in the light-proof, sound-attenuating cubicle within which the animal was trained. A variable interval of 1-3 s separated the successive trials. In the serial recognition memory task first and repeat presentations of infrequently seen stimuli occurred in a balanced, pseudorandom series, with varying numbers of intervening trials between the first and the subsequent appearances of each stimulus. The correct response was a left reach to the first and a right reach to a subsequent presentation of a stimulus. Several pseudorandom series, each of 100 trials, were used so that the series as well as the stimuli within it changed from day to day. A set of stimuli were used only once on a given day. The stimuli were complex pictures pre-selected for salience by the experimenters from the many thousands on commercially available videodiscs (University of London Pilot Disk; Spectrum, Stories from Hans Christian Andersen; Sony, Allianz; Valvo, Die FarbbildrShre - unser Fenster zur Welt; ITVA, Preistr~iger Video-Wettbewerb 1984; Lufthansa, Neue Verkaufstechniken). These pictures were a mixture of abstract or naturalistic scenes and animate or inanimate objects. After a picture had been seen on one day it was not used again for at least 2 weeks; some pictures had been encountered rarely if at all by the animal. A videodisc player (Lasermax LDP1500, Sony) was under the control of the computer (BBC Model B, Acorn) that governed all the other features of the behavioural tasks. The synchronisation pulses of the various video signals were all locked (Masterlock) to each other, 9 and the display on the colour video monitor (Cub 653, Microvitec) was gated on and off by a video switch, so that the pictures were presented as complete frames starting at a known time. The monkey was seated in a primate chair during training and recording sessions. The video monitor was 22 cm in front of the animal; its screen was 20 cm high and 27 cm wide; the images filled the screen. A touch screen (Microvitec), modified to give an enhanced speed for detecting responses, was placed in front of the monitor. Touches to either the left or the right of the screen that were more than 5.5 cm lateral to its midline were registered as behavioural responses. Two cameras separately monitored the animal's hand and eye movements. Eye movements were video-recorded together with the stimulus presentations using a video mixer (Videomat VM2E). Both trials on which the monkey failed to fixate the stimuli and error trials were excluded from analysis. Eye position was subsequently determined by measuring pupil position, using computer graphics, at 200 ms intervals from stimulus onset and converted into direction of gaze relative to that when the animal was looking straight ahead. When the monkey was not performing the visual discrimination or serial recognition memory tasks, the video monitor and touch screen were removed to allow three-dimensional objects or two-dimensional pictures to be shown to the animal through the resultant

3 459 hole and against a white background. This was done at the start of recording from each new neurone as a means of screening both for visual responsiveness and for activity changes based upon stimulus repetition or familiarity. Some thousands of objects and pictures of widely varying colours, sizes and shapes were used. Two sets of over 50 of these stimuli were shown to the animal frequently so that they were highly familiar to the animal. The two sets were shown on alternate days so that familiar stimuli that had not been seen for 24 h or more were available. Other objects were seen infrequently, with the recycling time being at least 4 weeks. The unfamiliar objects that had been seen the previous day were also kept to hand. As well as objects the animal was shown and then fed either pieces of fruit or fruit juice in a syringe. A pseudorandom order of presentation was used so that the animal could not predict what would appear on any given trial. No behavioural response was required, but the animal's fixation was monitored. Control trials for this procedure involved displaying objects in front of a colour camera (VCC3950, Sanyo); these appeared on the video monitor, under the control of the computer via the video switch in the same way as for the videodisc pictures. Neuronal recording When a monkey could perform the behavioural tasks at greater than 90% correct, it was anaesthetised and prepared using aseptic techniques for the recording of single neurones by fixing a chamber made of titanium to the skull. The chamber was surrounded with sheets of woven carbon-fibre fabric to provide an interface that was compatible with the skin. Postoperative analgesia was provided by buprenorphine. Two weeks were allowed for recovery from the operation before recordings were commenced using moveable tungsten microelectrodes (Riches et al. 1991b). The amplification, monitoring and display of neuronal potentials were conventional. A pulse-shape discriminator separated potentials from simultaneously recorded neurones (Brown and Leendertz 1979). A computer was used to process (BBC Model B, Acorn) the times of occurrence (to the nearest millisecond) of potentials of well-isolated single neurones in relation to the onset of stimulus presentations. A pulse was sent to this computer indicating the onset time of each stimulus presentation either by the computer controlling the behavioural task or by the experimenter initiating the presentation of an object. Both on- and off-line raster displays, peristimulus-time histograms and counts of action potentials were pt'oduced. Data analysis The major objective of these experiments was to find neurones with activity possibly related to recognition memory. The criteria for determining neuronal responsiveness were chosen accordingly. A neuron was considered not visually responsive if no audible change in firing rate occurred in response to at least one of the first four objects shown to the monkey. To be considered visually responsive such changes in firing rate had either to occur for repeated presentations of at least one object or for the first presentations of at least four different objects. Only neurones that were visually responsive under these criteria were further investigated. Tests of responses were not continued for more than eight different stimuli for visually responsive neurones that demonstrated no consistent change in response when stimuli were repeated or for which there was no consistent difference in the magnitude of response to unfamiliar and familiar stimuli: such neurones were categorised as not showing recognition-related responses. To determine statistically whether the activity of visually responsive neurones was related to stimulus familiarity or recency of occurrence, counts of spikes were made for 1 s after stimulus onset in the serial recognition task and for 2 s after stimulus onset when objects were shown. Spike counts during first and second presentations of stimuli were compared by paired t- tests. Spike counts for familiar and unfamiliar stimuli were corn- pared using two-sample t-tests. The proportions of responsive neurones in the various brain regions were subjected to analyses of variance based on a generalised linear model assuming a binomial error distribution (GLIM statistical package; Baker and Nelder 1978; Lindsey 1992). The variation (deviance) associated with the factors monkey, area, and side of brain were isolated as chi-squared values identified in the text as binomial error model (BEM) chisquared (Z2) values - by calculating the change in deviance when one of the factors was removed from a model containing them all. Only results that were consistent across monkeys are reported. All tests were two-tailed and used a significance level of P = 0.05 unless otherwise stated. Histological localisation Perfusion and histological processing were by standard techniques. Coronal sections were cut at 50 gm on a freezing microtome. The depth of each neurone was noted at the time of its recording. Anterior-posterior and lateral x-ray photographs were taken at the end of each electrode penetration to show the position of the electrode in situ in relation to both skull landmarks and fixed, reference electrodes. During the recordings microlesions were also made at known positions near responsive cells. Using the above information the positions of the recorded neurones were plotted onto large scale tracings of magnified, cresyl violet stained coronal sections of the brain; corrections were employed for x-ray expansion and tissue shrinkage. The boundaries of the various areas were entered on the tracings made every 1 mm through the medial temporal lobe (see Fig. 1). The definition of perirhinal and entorhinal cortex was that of Insausti et al. (1987); the limits of other areas followed Seltzer and Pandya (1978), Baylis et al. (1987) and von Bonin and Bailey (1947). Comparing this paper with that of Riches et al. (1991b), lateral entorhinal cortex approximates to their prorhinal cortex, and perirhinal cortex includes parts of their TEl. Results The activity of 2705 neurones was recorded from the cortex of the medial temporal lobe in both hemispheres of two monkeys and the left hemisphere of another. The cortical areas sampled were the entorhinal, the perirhinal, TG, TE, and the fundus of the superior temporal sulcus (STS). Of the recorded neurones 1236 (45.7%) were visually responsive. As reported previously (Riches et al. 1991b), the visually responsive neurones typically responded better to certain stimuli than to others, and in some cases responded to only a small subset of the stimuli tried. It is therefore probable that the true incidence of visually responsive neurones exceeds the figure quoted. Hence all statistical analyses used the total number of recorded neurones as the denominator. No significant difference between the left and right hemispheres was found for any of the reported measures, even taking account of the hand preferences of the individual monkeys. Moreover, there were no significant interactions between the factors side and area, nor between side or area and monkey, for any of the analyses quoted. The data for left and right sides and across monkeys have therefore been pooled in the results reported below. Decremental responses upon stimulus repetition Incidence. The responses of 120 of the recorded neurones (9.7% of those visually responsive) were maximal to the

4 460 Fig. 1. A Planes of section anterior (A) or posterior (P) to the sphenoid bone, drawn on a lateral view of the monkey brain. B-D Example sections illustrating the boundaries (arrows or dashed lines) between the various anatomical areas. Sections taken 2, 4 and 6 mm posterior to sphenoid, respectively. Ec, El, Em,entorhinal area: caudal, lateral and medial parts; P, perirhinal cortex; TEl, TEL TE3, TEa, TErn, parts of area TE. Scale bar 5mm first presentations of visual stimuli that had not been seen recently and significantly (paired t-tests, P < 0.05) smaller when these stimuli were seen again. Examples of these decremental responses may be seen in Figs. 2 and 3. The locations of the neurones are shown in Fig. 4. Control procedures (performance of the visual discrimination task and video monitoring of the hands and eyes) indicated that such response decrements could not be due to differences in the behavioural responses made by the animal, or differences in eye movements or pupil diameter: see Figs. 2 and 3, and Discussion for further details. For no neurone was the response significantly greater to subsequent than to first presentations of stimuli. The variation in incidence of the decremental responses across different anatomical areas is given in Table 1. Of the decremental responses 49 were to stimuli presented during performance of the serial recognition memory task and 97 were to the sight of objects presented with no required behavioural contingency; 26 neurones had decremental responses in both situations. For responses to the sight of objects there was significant variation in the incidence of decremental responses between the various areas (BEM Z 2 = 46.0, df = 9, P < 0.001). Further analysis revealed that the incidence of decremental re- Fig. 2 A-D. Example of response decrement with stimulus repetition for a neurone (9:23.8) recorded in perirhinal cortex. A The activity during the serial recognition task. Responses to the first (upper histogram and rasters) and repeat (lower histogram and rasters) presentations often different, unfamiliar stimuli. In the rasters each line is a single trial with dots representing the times of occurrence of action potentials. For each trial the cue onset is at time zero and the time the monkey touched the screen is marked by a P. For first presentations the touch was to the left of the screen and for repeat presentations it was to the right. Note the significantly (P < 0.001) smaller responses on trials with the second (repeat) presentations of the stimuli. Both the latency of the response and the differential,latency of responses to the first compared with the repeat presentations were 140 ms. The prestimulus firing rate of the neurone was not less for the repeat trials than for the first presentation trials; the difference in response is not attributable to a change in the background firing rate of the neurone. B The activity of the same neurone during the visual discrimination task. Activity for trials upon which were shown the stimulus requiring a touch to the left of the screen (upper histogram and rasters) and that requiring a touch to the right of the screen (lower histogram and rasters). There is no significant difference between the two types of trial. These trials provide control data indicating that the direction of the behavioural response required of the monkey does not account for the difference in activity between first and repeat trials in A. The two stimuli had last been seen some weeks previously. Note the slightly greater

5 461 g m 60' 40 2O Cue Stimulus 9:23.8 First O) 60 +~ 1-4O 20 Cue Stimulus Left I1~' t,',',",',,'h,l, ',, ~, ',' ','" ',",, " 'p ' ', 11] 1' %t.,,.,,l,.f,v'l -.t[. II" I' 1~, I,m... t,,[ ~ l 'l I' '.,,~',l '1~'1 I ~,, [,~,[,,,,,, I 9 = m e ' ' ',, 9 ~,,,.Ih 1% i 11 P I,, t, 9 il, H, t,, P, ~... 'Jl... ':',,,, l u sec,,,..I,.,,,,,q.,, i,, t,,',,, ' ", ",, 'C', ',' "l i '''~'t,,l.",',,''',, ' ~,,,',,,,...,,r,,,,,,,,,,,,,, ~,,,,,,,,,, ',:":",'"' i : ': ;2 ' :'"""' ih '"" I t ':'y", i ~'r, ie i,.t,,,,m,,.,,t..,j I,P,,,. 80 t Cue g 09 8 o - - Stimulus Repeat sec,.,..,., t, I' '.,,, i, =,,, I..,,,F,,, t 9,.,.[.=,,,p.d..,....,. '",~',,'.','.,"~,"',,,''"d,,f-',',','.,~j..,,...,,~... J" 9.., p'.'".',...,,,,], l,,,i.,, P,, J,,,,,,l, t.,, s,,p.,,,,,,,,,,.,t,,.,, l.t..,,,,,.,,,,.,,,,,. I.,,..,, o 860).~Cue Stimulus Right.~_ +t I -- '-mr --. / i, Jil ] i~l i i i - - i i i,+ L,l, I,,k,, Jt+,,,L~,, L, I, +, I, I ~, t,, L L L + I sec,,,,,',,,,,,.;..~,,,.-.,1,,,,,,,,,,,.,.,,.,~ r, '' ",' ',."",,',', I",... ' ',~-"':-"',' l, l,, h, t,b't..,,,.,,,,,,, i. i I t,,., i,, 9 v,,,'... ",, ' '.',... :,,',',,:, ' ~i,,'" B Left Horizontal Eye Movements 0. 9 ~ -10_.8 Right oi+ o Time (sec) Vertical Eye Movements response on the first trial (first raster line) than for the subsequent presentations for each stimulus. C Response decrement to repeated presentations of objects for the same neurone. The mean (_+ SEM) response to first presentations of unfamiliar objects was significantly (P < 0.05) greater than that to their second presentations. In addition, the response to first presentations of unfamiliar objects was significantly (P < 0.01) greater than to first presentations of highly familiar objects even though the highly familiar objects had not been seen for > 15 rain. This neurone signalled a mixture of recency and familiarity information. SA, mean 4- SEM spontaneous activity. D Mean _+ SEM horizontal (upper graph) and vertical(lower graph) eye position for ten first presentations of unfamiliar (solid lines) and ten first presentations of familiar (broken lines) objects. Stimulus onset is at time zero. There are no significant differences in mean eye position or its variance between unfamiliar and familiar trials in the 2 s following stimulus onset. Accordingly, differences in neuronal activity related to the relative familiarity of stimuli are not readily explained by consistent differences in eye position (see also Discussion, paragraph 5) -10. Up "~ -15 ~ g "~ Down D " 014 " " 116 " 210 Time (sec)

6 462 80~ Cue Stimulus 14"51.9 First sec... i... t.9./,.,ij,,,, ~,,., ~,~,~.,,.,~,.,,,,,,6,~......,... J.,..~... ~{ ~,... ~a.,, ~, ~..., I,... I... Repeat D Pupil Diameter During the Recogriition Task " 190" 180" 170" 160" 150" Time (msec) iiiiiiiiiiiiii n sec m,,,,a[,, a ma,,r,.,,]. tm,dm~,,.,d,tln[~ ~,l,,a~',., tm~ I,,,,,,.,,I,t,,,,,7,:,....,h,.,..... e $, t,i, 111~ I.. I(I Hi 1,,~. 1,1.,,,*mlt lj,d, Ill,, t~.~.,,d h.,,q.,,, ~,, d,u.,,,., ~di?,=,. ~,, * E O B ~ Left 40, 2 ~ 20, ro "S O, "~ -20, Right Up 10. [o Cumulative ~-, ---- Counts 14: First _--- ~ Repeat 200 4;0 6()0msec N -20.,~- ~ First g -3o..~ Down 0.0 C Horizontal Eye Movements Time (sec) Vertical Eye Movements Time (sec) First Repeat Fig. 3A-E. Example of response decrement with stimulus repetition for a neurone (14:51.9) recorded in lateral entorhinal cortex. A The activity during the serial recognition task. Responses to the first (upper histogram and rasters) and repeat (lower histogram and rasters) presentations of ten different unfamiliar stimuli. Note the significantly (P < 0.001) smaller responses on trials with the second (repeat) presentations of the stimuli: There was no significant regression between the size of the response decrement and the number of stimuli that intervened between the first and the repeat presentations the memory span of this neurone was > 40 intervening stimuli. Conventions as for Fig. 2. B Differential latency of response. The cumulative number of action potentials from the stimulus onset (time zero) is plotted for the first and repeat presentations of stimuli for the same trials shown in A. The latency of visual response is approximately 150 ms; the differential latency is approximately 260 ms. C. Mean 4- SEM horizontal (upper graph) and vertical (lower graph) eye position for the same first (solid lines) and repeat (broken lines) trials of the serial recognition task as in A. Stimulus onset is at time zero. There are no significant differences in mean eye position or its variance between first and repeat trials in the 600 ms following stimulus onset. For these trials there is significantly (P < 0.001) less neuronal activity to the repeat than to the first presentations by 400 ms after stimulus onset. Accordingly, the difference in neuronal activity cannot readily be explicable as due to differences in eye position. D Mean + SEM diameter of pupil of eye for the same trials - first presentations (solid lines) and second presentations (broken lines) - of the serial recognition task as in A and B. Stimulus onset is at time zero. There are no significant differences in mean pupil diameter or its variance ~oetween first and repeat trials in the 600 ms following stimulus onset. Accordingly, the difference in neuronal activity cannot readily be explicable as due to differences in pupil diameter. E Response decrement to repeated presentations of objects for the same neurone (14:51.9). The mean ( + SEM) response to first presentations of objects was significantly greater than that to their second presentations whether the objects were unfamiliar (Today's Unfamiliar Objects, P < 0.01) or were familiar objects already seen that day (Today's Familiar Objects, P

7 463 sts Sphenoid sts rhs 2P - ~ k,--/// ~-- sts amts amts arnts 8P / sts Fig. 4. The locations of neurones for which responses to first presentations of stimuli were significantly greater than those to subsequent presentations. Each triangle marks a neurone with such a response in the serial recognition task; each diamond marks a neurone with such a response to the sight of objects; each dot marks a neurone with such a response both in the serial recognition task and to the sight of objects. The locations are plotted on to reconstructions of sections through the medial temporal lobe at the indicated distances anterior (A) or posterior (P) to the outline of the sphenoid bone as seen on lateral radiographs of the skull (Aggleton and Passingham 1981). The plane of the sphenoid was approximately 21 mm anterior to the interaural line. No distinction has been made between neurones recorded in the left or right hemispheres. amts, anterior medial temporal sulcus; rhs, rhinal sulcus; sts, superior temporal sulcus sponses was highest in perirhinal cortex plus areas TEl and TE2 (70/767, or 9.1%), was significantly lower in the lateral entorhinal cortex plus area TG and TE3 (25/ 401, or 6.2%), and was significantly least in the medial and caudal entorhinal cortex plus the remainder of the inferior temporal cortex (2/68, or 2.9%). No other differences in incidence reached significance. These results were broadly paralleled for the responses of neurones recorded in the task. The incidence of decremental responses varied between the various areas (BEM < 0.05). or were familiar objects not seen since the previous day (Yesterday's Familiar Objects, P < 0.05). In addition, the response to first presentations of unfamiliar objects was significantly greater than to first presentations of highly familiar objects, whether seen already that day (P < 0.001) or not seen since the previous day (P < 0.005), i.e. even when the objects had not been seen for more than 24 h. The mean response to first presentations of Yesterday's Unfamiliar Objects was less than that to first presentations of Today's Unfamiliar Objects, but not significantly so. This neurone signalled a mixture of recency and familiarity information and demonstrated access to familiarity information held in memory for more than 24 h. SA mean SEM spontaneous activity ~2 = 26.0, df = 9, P < 0.005). It was highest in lateral entorhinal and perirhinal cortex plus areas TEl, TE2, TE3 and TG (48 of 1168 visually responsive neurones) and was significantly lower in the remaining parts of the entorhinal and inferior temporal cortices (1/68). (These proportions for neurones recorded in the task are low because only neurones selected for their responsiveness and not all visually responsive neurones were so tested.) There was evidence for clustering of the neurones with decremental responses. Successively recorded neurones had decremental responses on 14 occasions; this is a significantly higher incidence of such pairs than would be expected by chance (binomial test, one tailed, P < 0.002). Of the visually responsive neurones, 141 were tested in the serial recognition task as well as with objects. Of those so tested, 43 had decremental responses: for 26 (60%) the decrement was significant in both situations; for 12 (28%) the decrement was significant only in the task and for 5 (12%) the decrement was significant only for objects. That some neurones did not respond in the same way in both tests was due to differences in responsiveness to the two types of stimuli (objects and pictures). Typically neurones were unresponsive to at least some of

8 464 Table 1. Incidence of recognition memory-related responses Region Type of response Decrement on repetition D(n) D/V(%) D/N(%) Unfamiliar >familiar Visual F(n) F/T(% i F/N(%) V(n) V/N(%) N Total recorded Entorhinal Caudal Medial Lateral Perirhinal Area TG Area TE m and a Fundus STS Total Data from the serial recognition memory task and the showing of objects have been pooled. Of the neurones for which responses to familiar stimuli were significantly less than to unfamiliar stimuli, the responses of 36 also showed significant decrement on stimulus repetition. D(n), the number of neurones with responses that decreased significantly on stimulus repetition. F(n), number of neurones for which responses to familiar stimuli were significantly less than to unfamiliar stimuli; T, the number of neurones tested with stimuli differing in familiarity; V(n), the number of visually responsive neurones; N, the total number recorded in each of the different areas; STS, superior temporal sulcus; TG and TE, areas of temporal lobe (yon Bonin and Bailey 1947; Baylis et al. 1987) the stimuli tested; any decrement including data for such stimuli might not attain statistical significance. Additionally, the responsiveness of the neurones may have been influenced by the animals' training in the serial recognition memory task. Length of memory. Response decrements usually persisted when the second appearance of stimuli was separated from the first by intervening presentations of other stimuli: examples are illustrated in Fig. 5. The maximum number of such intervening presentations for which the response was still significantly less than the initial response was termed the inter-trial memory span of the neurone. The inter-trial memory span was at least 5 for 69% (36/52) of the neurones tested; 60% (12/20) of the neuroues tested had a memory span of at least 20. The incidence of memory spans of varying lengths in the different areas is given in Table 2. There were differences in the incidence of the lengths of memory spans between the areas. The proportions of neurones with memory spans of less than 5 (10/ 72, or 14%) or less than 20 (15/72, or 21%) in the lateral entorhinal and perirhinal cortex plus area TG were significantly less (Z 2 = 9.82, P < and Z 2 = 4.81, P < 0.05, respectively) than for area TE plus the fundus of the superior temporal sulcus (14/34, or 41% in each case). However, there was no significant difference between the different areas in the incidence of neurones whose memory spans were 20 or more. At least one example of a neurone whose memory span was at least 20 was found in each of the areas TEl, TE2 and TE3; all these neurones were located anterior to a plane 10 mm behind the sphenoid (approximately 11 mm anterior to the interaural line). There was no significant rank correlation between the memory spans of the neurones and the anterior/posterior position (between 2 mm anterior and 10 mm posterior to the sphenoid) at which the neurones were recorded. The length of memory of neurones recorded during the serial recognition task was also assessed by regressing their response decrement on the logarithm of the number of intervening stimuli plus one (see for example Fig. 5A). Sufficient data were available to assess 28 neurones for which there was a significant response decrement. For only 6 (21%) neurones was there a significant slope. For only 2 of these (7% of the total) was the intercept less than 20 intervening stimuli: for 1 neurone in the fundus of the superior temporal sulcus it was less than 5 and for 1 in TE3 it was 10 intervening stimuli. Thus this analysis suggests that the memory spans of 26/28 (93%) of the neurones were long (20 or more intervening stimuli). The locations of these neurones were: 4 in lateral entorhinal and 5 in perirhinal cortex, 4 in TG, 9 in TEl, 2 in TE2 and 2 in TE3. The neurones in TE2 and TE3 were located anterior to a plane 5 mm behind the sphenoid. Familiarity Detection Incidence. There were sufficient data to establish whether the response to highly familiar objects differed significantly from that to unfamiliar objects for 298 of the visually responsive neurones; for 43 (14.4%) of these neutones responses to the first presentations of stimuli that had been seen rarely were significantly (two-sample t- tests, P < 0.05) greater than those to first presentations

9 t Response to novel 3(1 Repeated Presentations of Objects 14:56.16 No. of Intervening Stimuli > ~ 2(1 1Q 20- A 0 2 ; 20 ~o too No. of intervening stimuli (since first presentation) t SA 200 Fig. 5 A,B. Neuronal memory spans. A The mean _+ SEM responses (ordinate) of a neurone (9:69.7) recorded in TEl during performance of the serial recognition task to the second presentations of stimuli after differing numbers of other stimuli (abscissa, on logarithmic scale) have intervened since the original stimuli were first seen. The mean response to the first presentations of the stimuli (Response to novel) is indicated by the broken line. The magnitude of the response decrement signals information concerning the number of trials that have intervened since the stimulus was last seen. There is still a significant decrement in response even after >_ 20 intervening stimulus presentations; thus the neuronal memory span is at least 20 intervening presentations. The intercept of the regression B Time (mins) line would indicate a memory span of up to 120 intervening items. SA, mean _ SEM spontaneous activity. B The responses of a neurone (14:56.16) recorded in perirhinal cortex to successive presentations of seven different unfamiliar objects plotted against time after the initial presentation. For no object did the response to subsequent presentations recover to its value for the first presentation, even when > 0.5 h and > 30 presentations of other stimuli intervened between the successive presentations. The spontaneous firing rate was 4.9 _+ 1.1 spikes/s. The responses of this neurone were also significantly less to first presentations of familiar stimuli that had not been seen since the previous day than to first presentations of unfamiliar stimuli (not illustrated) Table 2. Neuronal memory spans Area Memory spans < 5 < 20 Total > 5 Total _> 20 tested tested n % n % n n % n n % Total tested n Lateral entorhinal Perirhinal TG TE TE TE TEa Fundus STS Total The number (n) of neurones with decremental responses for which the response decrement was no longer significant after fewer than 5 or fewer than 20 intervening stimuli, or remained significant after at least 5 or at least 20 intervening stimuli for each of the different areas. Neurones that do not appear as either passing or failing at either 5 or 20 intervening stimuli had memory spans that exceeded some smaller number of intervening stimuli, but for which there was insufficient evidence to establish whether their memory spans were greater or less than 5 or 20 STS, superior temporal sulcus; TG and TE, areas of temporal lobe (yon Bonin and Bailey 1947; Baylis et al. 1987) of highly familiar stimuli, even when the familiar stimuli had not been seen for some considerable time previously. Examples may be' seen in Figs. 2C and 3E. The locations of such neurones are shown in Fig. 6. In control trials there were no differences in mean eye position or its vari- ance during the presentation of familiar and unfamiliar stimuli (Fig. 2D). The incidence of neurones that responded on the basis of the relative familiarity of stimuli varied significantly between the different anatomical areas (BEM Z 2 = 23.1,

10 466 2A ~ ~ 4P ~~sts 2P Fig. 8P ~---.,'~ rhs J amts 6. The locations of neurones for which responses to first presentations of unfamiliar stimuli were significantly greater than those to first presentations of familiar stimuli. Each dot marks a neurone that so responded and also showed significant response decrement on stimulus repetition; each triangle marks a neurone that so responded but did not show significant response decrement on stimulus repetition. Conventions as for Fig. 4 df = 9, P < 0.01). The variation in incidence of these neurones followed the same division as that for the neurones with decremental responses: all (43/282) such neurones were located in the lateral entorhinal and perirhinal cortex plus areas TEl, TE2, TE3 and TG; none (0/16) were found in the surrounding areas (medial and caudal entorhinal cortex, areas TEa, TEm and the fundus of the superior temporal sulcus). All except two of these neurones were located anterior to a plane 5 mm posterior to the sphenoid. No neurones that responded significantly more to familiar than to unfamiliar stimuli were found. As for the neurones with decremental responses, there was evidence for clustering of the neurones that responded on the basis of the relative familiarity of stimuli. Successively recorded neurones had such responses on five occasions; this is a significantly higher incidence of such pairs than would be expected by chance (binomial test, one tailed, P < 0.001). Length of memory. For such neurones the temporal memory span was defined as the maximum length of time that had elapsed since the familiar stimuli which gave rise to the significantly reduced response had last been seen. The memory span was greater than 5 min for all the neurones tested, but was greater than 20 min for at least 12 of them. Further to test the memory spans of such neurones, sets of stimuli that had not been seen for at least 24 h were presented. Evidence for a memory span of at least 24 h was found in all six neurones tested (examples are shown in Figs. 3E and 7). For five neurones (one in TG, one in lateral entorhinal and three in perirhinal cortex) there was a significantly greater response to first presentations of unfamiliar stimuli than to first presentations of highly familiar stimuli even when these had not been seen for more than 24 h; for the sixth neurone (in TG) the memory span was at least 72 h. For two of these neurones (both in TG) there was also a significantly greater response to first presentations of unfamiliar stimuli that were novel or had not been seen for weeks than to unfamiliar stimuli that were last seen more than 24 h ago. Recency Versus Familiarity Detection The responses of 36 of the recorded neurones signalled a mixture of information concerning the relative familiarity of the stimuli and how much time had elapsed since they were last seen. However, for seven other neurones, all located in TG, responses were significantly smaller to first presentations of highly familiar than to unfamiliar stimuli, while there was no significant decrement with stimulus repetition. An example is shown in Fig. 8B. Such familiarity neurones signal information concerning the relative familiarity of stimuli, but not their recency of presentation. Contrastingly, 58 neurones were found whose responses decremented upon stimulus repetition but which demonstrated no significant difference in their responses to familiar or unfamiliar stimuli. Examples were found in all areas except the medial and caudal entorhinal cortex: one is shown in Fig. 8A. Such recency neurones signal information concerning when stimuli were last seen, but not their familiarity. Thus recency and

11 467 familiarity information are separably encoded at the single neurone level in the medial temporal cortex. A further example of the signalling of recency information is given by the responses of the neurone illustrated in Fig. 5A, where the size of the decrement in the serial recognition task is related to how many intervening stimuli have occurred since a stimulus was last seen. Another type of result also indicated a difference in the processing of recency information dependent upon the relative familiarity of stimuli. For three neurones (all in TG) there was a significant decrement on stimulus repetition to familiar stimuli, but not to unfamiliar stimuli. For two neurones (one in TG and one in TEl) there was a significant decrement on stimulus repetition to unfamiliar stimuli, but not to familiar stimuli, despite there being a significant response to the two types of stimuli. Thus recency information may be signalled by different individual neurones: (1) either together with or independently from familiarity information; and (2) either only for highly familiar stimuli or only for unfamiliar stimuli or, most commonly, for both. Discussion The results demonstrate that within the medial temporal cortex there are neurones signalling information to do with how long ago stimuli were last seen, i.e the recency of their last presentations, and their relative familiarity, i.e. whether the stimuli had been seen many times before or not. It has been shown for the first time that information about familiarity and recency is separably encoded at the single cell level. Thus there are recency neurones that signal recency but not familiarity information and familiarity neurones that signal familiarity but not recency information. These recency and familiarity neurones represent the extremes of the whole population of neurones signalling information about recency and/or familiarity; because the sensitivity of responses to recency or familiarity is not the same from one neurone to another, it is clear that recency and familiarity information could Fig. 7 A-C. Neuronal memory spans exceeding 24 h. A Difference in response to unfamiliar and highly familiar objects for a neurone (14:54.2) recorded in perirhinal cortex. The mean (+ SEM) response to first presentations of unfamiliar objects (Today's Unfamiliar Objects) was significantly greater than that to first presentations of highly familiar objects, whether they had been seen that day (Today's Familiar Objects, P < 0.005) or not for more than 24 h (Yesterday's Familiar Objects, P < 0.05). Thus this neurone demon- strates access to information held in memory for at least 24 h. For the unfamiliar objects the response was significantly greater to first than to second presentations (P < 0.005). B The arrow in perirhinal cortex marks a lesion made close to the site at which the neurone whose responses are shown in A was recorded. A, amygdala; rs, rhinal sulcus; sts, superior temporal sulcus. C Neurone (14: 72.8B),xecorded in TG, signalling familiarity information with a 72-h memory span. This neurone responded selectively to pictures of human or monkey faces. The mean (. SEM) response to first presentations of novel faces differed significantly from that to either familiar (P < 0.001) or rarely seen (P < 0.005) faces. The rare faces had been seen only two or three times previously, more than 72 h previously. The familiar faces also had not been seen for over 72 h. Thus this neurone demonstrates access to information held in memory for at least 72 h. The mean response to first presentations of novel faces was not significantly different to that to their second presentations, i.e. there was no consistent response decrement with stimulus repetition. Thus this neurone signalled familiarity but not recency information. SA, mean _+ SEM spontaneous activity

12 468 Fig. 8 A,B. Recency and familiarity neurones. A This recency neurone (9:19.1), recorded in TG, signals recency but not familiarity information. The mean (+ SEM) response to first presentations of familiar or unfamiliar objects was significantly (P < 0.05) greater than that to their second presentations. However, the mean response to first presentations of familiar stimuli did not differ significantly from that to unfamiliar stimuli, neither was there a significant difference in the decrement in response upon stimulus repetition. B This familiarity neurone (9:59.23), recorded in TG, signals familiarity but not recency information. The mean (_+ SEM) response to first presentations of highly familiar objects differed significantly (P <0.001) from that to unfamiliar objects. However, the mean response to first presentations of familiar stimuli was not significantly different to that to their second presentations, i.e. there was no response decrement on stimulus repetition. SA, mean + SEM spontaneous activity be readily distinguished by appropriate sampling of the activity of the whole ensemble of neurones. Hence overlapping but separable memory mechanisms may be found not only between grossly different types of memory (see, for example, Squire 1987; Tulving and Schacter 1990; Weiskrantz 1990), but also even within the single category of recognition memory. The results also demonstrate that the distribution of the neurones within the temporal lobe that signal information of use for the performance of recognition and working memory tasks is not uniform. The highest incidence of neurones with decremental responses was in the perirhinal and medial inferior temporal cortex (areas TEl and TE2), though such neurones were also found in the lateral entorhinal cortex, area TG and more lateral inferior temporal cortex. No such neurones were found in medial or caudal entorhinal cortex. The incidence of such responses for stimuli presented in the serial recognition task was broadly similar to that when objects were shown. There was evidence for the clustering of neurones with recognition memory related activity. Thus it is possible that such neurones may occur in cortical columns. The experiments establish that the memory spans of many of the neurones are long. For the majority of neurones with decremental responses the memory span exceeded the longest interval tested. There was a greater incidence of neurones with short memory spans in regions outside the lateral entorhinal and perirhinal cortex plus area TG. However, neurones with memory spans of at least 20 intervening items were found not only in the anterior and medial areas just listed, but also in areas TEl, TE2 and TE3 anterior to the plane 10 mm behind that of the sphenoid (i.e. approximately 11 mm anterior to the interaural line). Thus the region containing neurones with long memory spans extends further laterally (at least anteriorly) than had been established previously. The results present the first description of the distribution of neurones that respond significantly more to unfamiliar than to highly familiar stimuli, i.e. neurones that may signal information concerning the relative familiarity of stimuli. Such neurones were found in the lateral entorhinal and perirhinal cortex plus areas TG, TEl, TE2 and TE3, but not in other parts of the entorhinal or inferior temporal cortex. All but two of these neurones were located anterior to a plane 5 mm behind that of the sphenoid. Examples of neurones with memory spans of more than 24 h (> 72 h in one case) were found in area TG and in the lateral entorhinal and perirhinal cortex. These memory spans greatly exceed any previously described (Riches et al. 1991b). Familiarity neurones were not found outside area TG. Control procedures and behavioural monitoring establish that the observed response decrements upon stimulus repetition are not artifactual. The differences in response on first and repeat trials in the serial recognition task could not be due to: (1) reward differences, because both types of trial were equally rewarded; (2) differences in the monkey's behavioural responses, because the same left or right reach was required in the visual discrimination task and did not produce the same change in neuronal activity and, moreover, such responses had to be delayed until after the stimulus had been displayed; (3) gross changes in arousal or attentiveness, because the monkey's performance of the task was consistently accurate across large blocks of trials within which he could not predict whether a novel or familiar stimulus would appear on any given trial (the few error trials were excluded from the analysis); or (4) differences in the monkey's eye movements or pupillary diameter, because there

13 469 were differences in neuronal activity before such ocular changes had occurred (trials on which the animal failed to fixate the stimuli were excluded from the analysis). Studies using the subscleral search coil technique have also indicated that eye movements do not change upon stimulus repetition during the 1st s of display (F.A.W. Wilson and P.S. Goldman-Rakic, unpublished observations). Further, recent studies using an anaesthetised monkey or requiring constant visual fixation have also found response decrement on stimulus repetition (Miller et al. 1991a, b). Similarly, no changes in eye movements were found that could explain the difference in neuronal response to familiar compared with unfamiliar stimuli (and such changes in neuronal response have also been reported with constant fixation (Miller et al. 1991b). Moreover, the incidence of response differences with stimulus repetition or relative familiarity was not constant from area to area as might be expected if the differences were caused by some change at the periphery or in general arousal. The present study has shown that changes in neuronal activity are present before ocular changes; thus it is possible that the responses of these neurones contribute to the production of ocular changes in response to stimuli of differing familiarity. Previous studies have failed to find evidence of long memory spans for neurones recorded in the more posterior and lateral parts of inferior temporal cortex (Baylis and Rolls 1987; Riches et al. 1991b). These regions provide afferent input to the medial and anterior cortex (Van Hoesen and Pandya 1975). It has thus seemed unlikely that the responses signalling recency or familiarity information in the more anterior and medial cortex are simply passive reflections of changes earlier in the visual pathway. While it remains the case that no neurones sensitive to stimulus familiarity have been found outside the anterior and medial temporal cortex, in the present experiments a few neurones sensitive to stimulus recency with memory spans of 20 or more were found in areas TE2 and TE3 - though even these were in the anterior part of these areas. This finding raises the possibility that recordings under appropriate conditions in more posterior visual cortical areas might reveal that the region containing neurones signalling recency information for more than very brief intervals is more extensive than currently assumed. The mechanism underlying the response decrements is unknown. For the recency neurones a process allied to habituation is possible; however, this process must be capable of producing a large (typically 50%) decrement in response for the first repetition of a stimulus even if its second appearance does not occur until after many presentations of other items and the stimuli are being used by the animal to obtain reward. Simple neuronal fatigue cannot explain the decrements because the neurones were capable of demonstrating large response decrements on stimulus repetition while still responding vigorously to other stimuli that had not been seen recently. Response differences due to the relative familiarity of stimuli may arise by another mechanism in at least some neurones. In such neurones response decrement on stimulus repetition is very weak but the response to stimuli that have been seen only a few times many hours previously is nevertheless greatly reduced (see Fig. 7B). Response decrements have now been found using a variety of classes of "visual stimuli: three-dimensional objects, and two-dimensional geometric stimuli and complex pictures (see the present study and Brown et al. 1987; Riches et al. 1991b). The responsiveness of individual neurones is not necessarily the same to each of these classes of stimuli; recent behavioural experiments have also demonstrated that there are likely to be differences in the processing of complex pictures (scenes) and individual items (Gaffan 1992). The present study has shown for the first time that neuronal response decrements occur across trials for stimuli the animal is explicitly using to solve a recognition memory task. Although possibly influenced by training, response decrement upon stimulus repetition is likely to be an endogenous property of at least some of the neurones in the medial temporal lobe, because decrements occur to repeated presentations of three-dimensional objects for which there has been no behavioural training and which require no behavioural response. The incidence of responses that changed in relation to stimulus repetition may have been underestimated in the present experiments. Most neurones did not respond to all stimuli tried, so that the ideal stimuli for a given cell may not have been used. Moreover, the criteria employed to select a cell for full study may have excluded those showing weak effects. There was a tendency towards response decrement for many more neurones than those for which such decrement was proved to be significant. A recent, briefly reported study (Miller et al. 1991b) found a higher proportion of visually responsive neurones that decreased with stimulus repetition than that in the present experiments (though there were no decrements from trial to trial for their sample stimulus presentations). However, that study employed a different behavioural paradigm, mainly highly familiar (and less complex) stimuli, a higher rate of stimulus presentation, and many more repetitions. Miller et al. (1991a) have shown that the responses of a far higher proportion of inferior temporal neurones show decrement with stimulus repetition if the rate of repetition is high. It is also probable that different pools of neurones are recruited according to the precise computational requirements of the task. Indeed, the responsiveness of inferior temporal neurones is known to be influenced by the training history of the animal (Fuster and Uyeda 1971; Miyashita 1988). The fact that no neurones were found with responses that became significantly greater with stimulus repetition or that responded more strongly to familiar than unfamiliar stimuli makes it extremely improbable that the observed changes were due to chance. Two reasons may be advanced for why there is usually a decrement rather than an increment in responsiveness as items become more familiar. Firstly, in normal every day life most encountered items will be familiar: it reduces energy requirements for neuronal activity to reduce rather than to increase for such stimuli. Secondly, on average, novel items will require more neuronal processing and are likely to evoke a greater behavioural reaction than items that

14 470 have been encountered recently or are familiar. Thus the findings accord with the general principle of processing within the nervous system that where possible neurones signal change rather than constancy. Functional considerations Decremental responses have also been described for neutones in the amygdala (Nishijo et al. 1988; Riches et al. 1991b; Wilson and Rolls 1993) and hippocampal formation (Rolls et al. 1989; though see Heit et al. 1990; Riches et al. 1991b), but the memory spans are limited and no neurones responding on the basis of the relative familiarity of stimuli have been found. However, long memory spans (i.e. exceeding the maximum tested) have been found for neurones in the medial thalamus (Fahy et al. 1991, 1993) and basal forebrain (Rolls et al. 1982; Wilson and Rolls 1990), regions where lesions also interfere with the performance of recognition memory tasks (Aggleton and Mishkin 1983; Aigner et al. 1987). Neurones sensitive to stimulus familiarity have been described in the medial thalamus, but not in the basal forebrain. It remains to be established whether these types of responses occur in other closely anatomically related regions such as the medial prefrontal cortex, medial septal nuclei or mamillary bodies. Further, although there are differences in the types of responses encountered in the various regions studied, the direction(s) of flow of information related to recognition memory between these highly interconnected structures is unknown (Fahy et al. 1993). There are strong grounds for believing that the neuronal responses signalling recency and/or familiarity information are of importance for the performance of certain types of recognition and working memory tasks. The decremental responses signal the type of information necessary for the solution of tasks that require recency discrimination. For instance, short memory spans should be valuable in working memory tasks where remembering the occurrence of stimulus items across many trials or from day to day is not advantageous. The necessary responses are indeed found during the performance of recognition memory tasks and, across the whole population of cells, the neuronal memory spans can be of sufficient length to solve most such tasks. Moreover, the neurones that signal such information occur in those regions where lesions cause profound deficits in the performance of these tasks (Mishkin 1982; Aggleton and Mishkin 1983; Zola-Morgan and Squire 1985b; Aigner et al. 1987; Zola-Morgan et al. 1989; Wilson and Rolls 1990; Fahy et al. 1991, 1993; Murray 1991; Riches et al. 1991b; Gaffan and Murray 1992). Furthermore, a considerable body of data has been obtained concerning neuronal responses in these critical regions (particularly in the medial temporal cortex) with the discovery of only one other type of neuronal signal that could possibly be appropriate for the solution of such tasks. This other type of signal is a sustained increase in firing during the delay period of a delayed matching task (Fuster and Jervey 1981). Such sustained activity is unlikely to provide a basis for the solution of tasks where other stimuli intervene between sam- ple and choice, or where the future need for the information is not predictable. Similarly, responses related to the familiarity of stimuli are found in the medial temporal lobe and medial thalamus where lesions cause recognition memory deficits. Such familiarity information may be used to solve certain recognition memory tasks and may also be of use in tasks with both working and reference memory components. Nevertheless, recency and familiarity information is not adequate to solve all recognition memory tasks; the important role of contextual encoding in recognition memory has been discussed elsewhere (Brown 1982, 1990; Mayes et al. 1985; Brown and Brown 1990; Riches et al. 1991b). A further function for which the information signalled by the decremental responses may be used is priming memory, i.e. the facilitation of performance by a prior related event. (Responses related to the relative familiarity of stimuli seem unlikely to be of such importance to priming.) Human amnesic patients of temporal lobe as well as medial diencephalic aetiology display essentially intact priming (Shimamura 1986; Tulving and Schacter 1990). However, the present results demonstrate that there are neurones with decremental responses and long (> 20) memory spans in areas outside the usual region of temporal lobe damage (hippocampal formation, amygdala and rhinal cortex) in amnesic patients (Parkin 1987; Squire 1987; Mayes 1988). Priming and recognition memory can be doubly dissociated (Brown et al. 1989; Tulving and Schacter 1990), but this dissociation has not been shown for recency and priming. Features of the behavioural paradigm used by Miller at al. (1991b) approach those of tasks used to demonstrate lexical decision priming in human subjects: many inferior temporal neurones demonstrated decremental responses in their paradigm. If decremental responses do underlie priming, it remains to be explained how a decrease in neuronal responsiveness leads to a facilitation in performance. In conclusion, the findings of the present experiments in conjunction with those of recent studies that have selectively lesioned the cortex surrounding the rhinal sulcus (Murray 1991; Gaffan and Murray 1992) provide powerful evidence for the importance of the cortex around the rhinal sulcus to recognition memory processes related to determination of the recency of presentation or familiarity of visual stimuli. It has been shown that memory spans of the neurones of the region with responses that signal such information are of sufficient length to allow solution of many such tasks. However, other regions may be important for other functions necessary to recognition memory (e.g. the hippocampal formation for encoding contextual informatiov), and the neurones may provide information for other types of memory as well as recognition (e.g. priming). The experiments have shown that there are also neurones with long memory spans signalling information about the recency and familiarity of stimuli in areas TE and TG, areas outside the cortex immediately adjacent to the rhinal sulcus. Such neuronal responses could provide a substrate for the remaining capacities in recognition and priming memory of human amnesic patients with temporal lobe lesions that do not extend to these areas.

15 471 Acknowledgements. This work was supported by the Wellcome Trust and the Anatomical Society of Great Britain and Ireland. The authors wish to thank Mr R. Chambers, Mr J.A. Leendertz and Mrs A.M. Somerset for their assistance with the experiments. References Aggleton JP, Mishkin M (1983) Memory impairments following restricted medial thalamic lesions in monkeys. Exp Brain Res 52: Aggleton, JP, Passingham RE (1981) Stereotaxic surgery under x- ray guidance in the rhesus monkey, with special reference to the amygdala. Exp Brain Res 44: Aggleton, JP, Nicol RM, Huston AE, Fairbairn AF (1988) The performance of amnesic subjects on tests of experimental amnesia in animals: delayed matching to sample and concurrent learning. Neuropsychologia 26:272 Aigner TG, Mitchell S J, Aggleton JP, DeLong MR, Struble RG, Price DL, Wenk GL, Mishkin M (1987) Effects of scopolamine and physostigmine on recognition memory in monkeys with ibotenic-acid lesions of the nucleus basalis of Meynert. Psychopharmacology 9 Springer Verlag. 92: Baker RJ, Nelder JA (1978) The GLIM (Generalised Linear Interactive Modelling) system. Royal Statistical Society (Numerical Algorithm Group, 7 Banbury Rd, Oxford) Baylis GC, Rolls ET (1987) Responses of neurones in the inferior temporal cortex in short term and serial recognition memory tasks. Exp Brain Res 65: Baylis GC, Rolls ET, Leonard CM (1987) Functional subdivisions of the temporal lobe neocortex. J Neurosci 7: Bonin G yon, Bailey P (1947) The neocortex of Macaca mulatta. University of Illinois Press, Urbana Brown MW (1982) Effect of context on the responses of single units recorded from the hippocampal region of behaviourally trained monkeys. In: Ajmone-Marsan C, Matthies H, (eds), Neuronal plasticity and memory formation. (IBRO monograph series, vol 9) Raven, New York, pp Brown MW (1990) Why does the cortex have a hippocampus? In: Gabriel M, Moore J (eds), Learning and computational neuroscience: foundations of adaptive networks. MIT, New York, pp Brown J, Brown MW (1990) The effects of repeating a recognition test in lorazepam-induced amnesia: evidence for impaired contextual memory as a cause of amnesia. Q J Exp Psychol [B] 42A: Brown MW, Leendertz JA (1979) A pulse-shape discriminator for action potentials. J Physiol (Lond) 298:17-18P Brown MW, Wilson FAW, Riches IP (1987) Neuronal evidence that inferomedial temporal cortex is more important than hippocampus in certain processes underlying recognition memory. Brain Res 409: Brown MW, Brown J, Bowes JB (1989) Absence of priming coupled with substantially preserved recognition in lorazepam-induced amnesia. Q J Exp Psychol [B] 41A: 59%617 Fahy FL, Riches IP, Brown MW (1991) Neuronal evidence of the involvement of the primate medial thalamus in recognition memory. Eur J Neurosci [Suppl ]4:88 Fahy FL, Riches IP, Brown MW (1993) Neuronal signals of importance to the performance of visual recognition memory tasks: evidence from recordings of single neurones in the medial thalamus of primates. Prog Brain Res 95: Fuster JM, Jervey JP (1981) Inferotemporal neurons distinguish and retain behaviorally relevant features of visual stimuli. Science 212: Fuster JM, Uyeda AA (1971) Reactivity of limbic neurons of the monkey to appetitive and aversive signals. Electroencephalogr Clin Neurophysiol 30: Fuster JM, Bauer RH, Jervey JP (1981) Effects of cooling inferotemporal cortex on performance of visual memory tasks. Exp Neurol 71 : Gaffan D (1974) Recognition impaired and association intact in the memory of monkeys after transection of the fornix. J Comp Physiol Psychol 80: Gaffan D (1992) Amnesia for complex naturalistic scenes and for objects following fornix transection in the rhesus monkey. Eur J Neurosci 4: Gaffan D, Murray EA (1992) Monkeys (Macaca fascicularis) with rhinal cortex ablations succeed in object discrimination learning despite 24-hour intertrial intervals and fail at matching to sample despite double sample presentations. Behav Neurosci 106:30-38 Heit G, Smith ME, Halgren E (1990) Neuronal activity in the human medial temporal lobe during recognition memory. Brain 113: Horel JA, Pytko-Joiner DE, Voytko M, Salsbury K (1987) The performance of visual tasks while segments of the inferotemporal cortex are suppressed by cold. Behav Brain Res 23:29-42 Insausti R, Amaral DG, Cowan WM (1987) The entorhinal cortex of the monkey. III. Subcortical afferents. J comp Neurol 261: Lindsey JK (1992) The analysis of stochastic processes using GLIM. Springer, Heidelberg New York Berlin Mayes AR (1988) Human organic memory disorders. University Press, Cambridge Mayes AR, Meudell PR, Picketing A (1985) Is organic amnesia caused by a selective deficit in remembering contextual information? Cortex 21: Miller EK, Gochin PM, Gross CG (1991a) Habituation-like decrease in the responses of neurons in inferior temporal cortex of the macaque. Vis Neurosci 7: Miller EK, Li L, Desimone R (1991b) A neuronal mechanism for working and recognition memory in inferior temporal cortex. Science 254: Mishkin M (1982) A memory system in the monkey. Philos Trans R Soc Lond [Biol] 298:85-95 Miyashita Y (1988) Neuronal correlate of visual associative longterm memory in the primate temporal cortex. Nature 335: Murray EA (1991) Medial temporal lobe structures contributing to recognition memory: the amygdaloid complex versus the rhinal cortex. In: Aggleton JP (ed) The amygdala: neurobiological aspects of emotion, memory and mental dysfunction. Wiley-Liss, New York, pp Nishijo H, Ono T, Nishino H (1988) Single neuron responses in amygdala of alert monkey during complex sensory stimulation with affective significance. J Neurosci 8:357~3583 Olton DS, Becker JT, Handelmann GE (1979) Hippocampus, space and memory. Behav Brain Sci 2: Parkin AJ (1987) Memory and amnesia: an introduction, Blackwell, Oxford Pearson RCA, Esiri MM, Hiorns RW, Wilcock GK, Powell TPS (1985) Anatomical correlates of the distribution of the pathological changes in the neocortex in Alzheimer's disease. Proc Natl Acad Sci USA 82: Riches IP, Brown MW, Wilson FAW (1990) Neurones ofinferomedial temporal cortex display evidence of memory for the previous occurrence of visual stimuli. Perception 19: Riches IP, Fahy FL, Brown MW (1991a) Involvement of the primate medial temporal lobe in recognition memory: neuronal evidence of long-term retention and discrimination of familiarity and recency. Eur J Neurosci [Suppll4:89 Riches IP, Wilson FAW, Brown MW (1991b) The effects of visual stimulation and memory on neurons of the hippocampal formation and the neighboring parahippocampal gyrus and inferior temporal cortex of the primate. J Neurosci 11: Rolls ET, Perrett DI, Caan AW, Wilson FAW (1982) Neuronal responses related to visual recognition. Brain 105: Rolls ET, Miyashita Y, Cahusac PMB, Kesner RP, Niki H, Feigenbaum JD, Bach L (1989) Hippocampal neurons in the monkey with activity related to the place in which a stimulus is shown. J Neurosci 9:

16 472 Seltzer B, Pandya DN (1978) Afferent cortical connections and architectonics of the superior temporal sulcus and surrounding cortex in the rhesus monkey. Brain Res 149:1~4 Shimamura AP (1986) Priming effects in amnesia: evidence for a dissociable memory function. Q J Exp Psychol [B] 38A: Squire LR (1987) Memory and brain. Oxford University Press, New York Squire LR, Zola-Morgan S (1991) The medial temporal lobe memory system. Science 253: Squire LR, Zola-Morgan S, Chen KS (1988) Human amnesia and animal models of amnesia: performance of amnesic patients on tests designed for the monkey. Behav Neurosci 102: Tulving E, Schacter DL (1990) Priming and human memory systems Science 247: Van Hoesen GW, Hyman BT (1990) Hippocampal formation - anatomy and the patterns of pathology in Alzheimer's disease. Prog Brain Res 83: Van Hoesen GW, Pandya DN (1975) Some connections of the entorhinal (area 28) and perirhinal (area 35) cortices of the rhesus monkey. 1. Temporal lobe afferents. Brain Res 95:1-24 Weiskrantz L (1990) Problems of learning and memory - one or multiple memory systems. Philos Trans R Soc Lond [Biol] 329: Wilson FAW, Roils ET (1990) Neuronal responses related to the novelty and familiarity of visual stimuli in the substantia innominata, diagonal band of Broca and periventricular region of the primate basal forebrain. Exp Brain Res 80: Wilson FAW, Rolls ET (1993) The effects of stimulus novelty and familiarity on neuronal activity in the amygdala of monkeys performing recognition memory tasks. Exp Brain Res 93: Zola-Morgan S, Squire LR(1985a) Medial temporal lesions in monkeys impair memory on a variety of tasks sensitive to human amnesia. Behav Neurosci 99: Zola-Morgan S, Squire LR (1985b) Amnesia in monkeys after lesions of the mediodorsal nucleus of the thalamus. Ann Neurol 17: Zola-Morgan S, Squire LR, Amaral DG, Suzuki WA (1989) Lesions of perirhinal and parahippocampal cortex that spare the amygdala and hippocampal formation produce severe memory impairment. J Neurosci 9: