Differential neuronal encoding of novelty, familiarity and recency in regions of the anterior temporal lobe

Transcription

1 Neuropharmacology 37 (1998) Differential neuronal encoding of novelty, familiarity and recency in regions of the anterior temporal lobe J.-Z. Xiang, M.W. Brown * Department of Anatomy, Uni ersity of Bristol, Bristol BS8 1TD, UK Accepted 25 February 1998 Abstract Activity of 2072 neurones was recorded in the anterior temporal lobe in area TE, perirhinal cortex, entorhinal cortex and hippocampus during performance of a visual recognition task by monkeys. In area TE, perirhinal cortex and entorhinal cortex, 454 neurones (38% of the 1162 visually responsive neurones) responded differentially on the basis of the relative familiarity or recency of presentation of the stimuli; in the hippocampus only one (3% of its 40 visually responsive neurones) did so. The differentially responsive neurones were classified into those signalling information concerning the recency (19%), familiarity (37%) or novelty (38%) of stimuli. For 98% of these neurones a decreased response signalled that stimuli had occurred previously: no large response increments were observed. The mean differential latency of each of these types of neurone was shorter ( 75 ms) in area TE than in the other areas. Examples of each of these types of neurone with memory spans of 24 h were found in each region. The mean memory span of recency neurones was significantly longer in perirhinal cortex than area TE. For familiarity neurones a significant mean response decrement took 4 8 min to develop, indicating a slow underlying plastic change, in contrast to the rapid change seen for recency and novelty neurones. The implications of these results are discussed in relation to the neuronal basis of recognition memory Elsevier Science Ltd. All rights reserved. Keywords: Memory; Recognition memory; Visual response; Perirhinal cortex; Hippocampus; Inferior temporal cortex 1. Introduction Cooling and ablation studies have established that the perirhinal cortex within the anterior temporal lobe of monkeys is essential for visual recognition memory (Horel et al., 1987; Gaffan and Murray, 1992; Meunier et al., 1993; Suzuki et al., 1993; Eacott et al., 1994; Meunier et al., 1996; Murray, 1996). Electrophysiological recordings from single neurones in behaviourally trained monkeys have demonstrated that certain neurones in perirhinal and adjacent cortex signal information of potential importance to recognition memory concerning the prior occurrence of visual stimuli (Brown et al., 1987; Riches et al., 1991; Eskandar et al., 1992; Fahy et al., 1993; Li et al., 1993; Miller et al., 1993; Sobotka and Ringo, 1993; Miller and Desimone, 1994; Brown, 1996; Brown and Xiang, 1998). Indeed, it has been established (Fahy et al., 1993) that there is * Corresponding author. Tel.: ; fax: ; M.W.Brown@bris.ac.uk. separable neuronal encoding of information concerning a stimulus s relative familiarity (whether a stimulus is unfamiliar or highly familiar) and recency of occurrence (whether or not a stimulus was last encountered recently). Thus the responses of recency neurones encode whether or not a particular stimulus has been seen recently regardless of whether it is relatively familiar or is unfamiliar. In contrast, the responses of familiarity neurones encode whether a particular stimulus is highly familiar or relatively unfamiliar whether or not the stimulus has been seen recently (Fahy et al., 1993). Nevertheless, relatively little is known of the properties of these neurones with responses that encode information concerning the prior occurrence of stimuli. Importantly, although a few examples of neurones whose responses demonstrate that they have access to information held in memory for a period (the memory span) of more than 24 h, it is not known whether such neurones are common or rare. Do such neurones include recency neurones as well as familiarity neurones? Are there other categories of neuronal response sig /98/$ Elsevier Science Ltd. All rights reserved. PII: S (98)

2 658 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) nalling information of potential use to recognition memory? This is possible because not all neurones that changed response when stimuli were repeated could be categorised as recency or familiarity neurones in the study of Fahy et al. (1993). Moreover, that study did not explore the responses of neurones to familiar stimuli across different delays during performance of a serial recognition memory task, nor did it explore how the responses of familiarity neurones developed as stimuli became familiar through being shown repeatedly. This study set out to provide information on these issues. Characteristically, signalling that a stimulus is familiar or has been seen recently is by a reduction in the neurone s response compared to that to the first presentation of a novel stimulus. Previous studies have suggested that increments rather than such decrements in neuronal responses are rare (Riches et al., 1991; Fahy et al., 1993; Miller et al., 1993; Sobotka and Ringo, 1993). In the present study particular attention was paid to this matter because of its importance to theories of the generation of new representations (Amari, 1989; Rolls, 1995). Another incompletely explored topic concerns the location and extent of the anatomical distribution of the neurones signalling the prior occurrence of stimuli within the anterior and medial temporal lobe. In particular, are there differences in the incidence or properties of such neurones in area TE of the anterior inferior temporal cortex compared to those in the perirhinal cortex? Are such neurones also common in entorhinal cortex or is their distribution more limited, as previously suggested (Fahy et al., 1993)? Further, there is controversy concerning the incidence of such neurones in the hippocampus. Claims (Brown et al., 1987; Riches et al., 1991) that such neurones are not to be found in the hippocampus have been challenged (Rolls et al., 1989, 1993). It was therefore decided to compare the incidence of such neurones in the rhinal cortices and area TE with that in the hippocampus, using the same stimuli within the same experiment. This study used a serial recognition memory task (Gaffan, 1974) as this task requires an animal to remember the occurrence of more than one stimulus at a time even when these stimuli may not re-occur until many other stimuli have been seen. Thus this task does not allow the use of attentive or short-term memory mechanisms for its solution. Certain findings of this study have been published in abstract form (Xiang and Brown, 1997a). mulatta) weighing 7 and 9 kg. Procedures were performed in accordance with the UK Home Office Licensing regulations and animal welfare was overseen by a veterinary officer. Many methodological details have been published previously (Fahy et al., 1993) and will be mentioned only briefly here Beha ioural training The behavioural tasks were of two types: serial recognition and conditional visual discrimination. During training and recording sessions the animal was seated in a primate chair in an illuminated light-proof, sound-attenuating cubicle. A video monitor (20 27 cm) was placed 22 cm in front of the animal. A touch screen (Microvitec), modified to give an enhanced speed for detecting responses, was fixed in front of the monitor. Once trained, the animals performed the tasks at greater than 90% accuracy for more than 1000 trials per day. For the serial recognition task (Gaffan, 1974) the animal was taught to discriminate novel or unfamiliar stimuli from familiar or recently seen stimuli. A single stimulus (picture) appeared on each trial. When a stimulus appeared on the screen, the correct behavioural response was a left touch for a novel stimulus and a right touch for its repetition or a familiar stimulus. Each trial (Fig. 1) started with a cueing light (1.5 s, C), a dim red neon bulb situated centrally at the top of the screen. After the cueing light had been on for 1 s, a picture (S) filling the screen was shown for 2 s. Touches (T) to the correct side between 0.5 and 2.5 s following the onset of the stimulus were rewarded with approximately 0.3 ml fruit juice (R). Other responses were counted as errors and the lights in the cubicle were dimmed for 3 s. Successive trials were separated by a randomly variable interval of 3 5 s. A computer (Viglen 486 PC) initiated the sequences, controlled a videodisk player (Lasermax LDP1500, Sony) and governed all the other features of the behavioural tasks. The synchronised pulses of the various video signals 2. Materials and methods 2.1. Subjects Data were obtained from two monkeys (Macaca Fig. 1. Trial events. C: cue light on; S: stimulus on; T: the period within which a behavioural response is required; R: reward. The reward was given as soon as a correct response was made. The dashed lines before and after the trial indicate the intertrial interval (3 5 s).

3 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) for control purposes and will be reported only briefly, as pertinent. For this task a stimulus was one of four geometric triplets (Fig. 2). The task involved a conditional rule in that whether a touch to the triangle or the square was correct depended on the orientation of the central shape. The task employed the same stimulus presentation timings and behavioural responses as the serial recognition task. Different types of trial were arranged into pseudorandom sequences of trials in length. Fig. 2. Details of the behavioural tasks. A. Serial recognition memory task using pictures of naturalistic scenes or objects. S1 S3 are examples of different types of sequences of trials. Each letter represents a particular picture; in S1 the upper-case letters are novel stimuli and the lower-case letters familiar stimuli. Sequence types S2 and 3 contained repeats of the same novel (S2) or different novel (S3) stimuli. B. Conditional visual discrimination task using triplets of geometric shapes. For each of the four types of trial containing a specific triplet, the animals needed to touch the left (L) or right (R) side of the screen as indicated to get a reward. were all locked to each other, and the display on an video monitor (Cub 653, Microvitec) was gated on and off by a video switch, so that the pictures were presented as complete still frames starting at a known time. The stimuli were complex, videodisk pictures of abstract or naturalistic scenes from commercially available videodisks, or digitised images of 3-dimensional objects, pre-selected for salience. A subset of over 40 pictures (familiar stimuli) were shown to the animal each day so that they were highly familiar to the animal. Other pictures (novel stimuli) were used only twice a day and not again for at least 2 months; some pictures had been encountered rarely if at all by the animal. In a sequence, various numbers of other novel and highly familiar stimuli intervened between the first and the subsequent appearances of each particular stimulus. Several different pseudorandomly balanced sequences of trials length were used so that the animal could not predict what would appear on any given trial (Fig. 2). Sequences (S1) were constructed so that the repetition of a novel stimulus occurred after 0, 2, 4, 8, 16, 32 or 64 intervening trials. On some trials stimuli seen the preceding day were re-shown so as to investigate memory storage over even longer intervals ( 24 h). Two other types of sequence were also used. To explore further the effects of stimulus repetition, in sequences of type S2 the same novel stimulus was shown on up to five successive trials before another novel stimulus appeared. In sequences of type S3 the first and second presentations of a novel stimulus were separated by trials on which one to four other novel stimuli were presented. The conditional visual discrimination task was used 2.3. Neuronal recording When an animal could perform the behavioural tasks at over 90% correct, it was anaesthetised and prepared for neuronal recording using aseptic techniques (Fahy et al., 1993). Postoperative analgesia was provided by buprenorphine. Two weeks were allowed for recovery before recording commenced. Neuronal activity was recorded through a moveable microelectrode (Elgiloy mm in diameter). The amplification, monitoring and display of neuronal potentials were conventional. At the start of recording from each new site, a few video pictures were shown to the animal as a means of screening for visual responsiveness, using audio monitoring. Only sites that were judged by this means to be visually responsive were further investigated. The activity of simultaneously recorded single neurons was analogue-to-digital converted (at 17 khz) and discriminated from multi-neuronal activity by on-line spike-sorting software (1401 Plus interface and Spike2, CED, Cambridge) using a template-matching algorithm (see Zhu et al., 1995; Nicol et al., 1998). Up to eight separable spike trains could be recorded at the same time. Peristimulus-time histograms, rasters and counts of action potentials for the separated spike trains were displayed on-line and were stored in the computer for further, off-line analysis. The animal s hand and eye movements were monitored using a video camera and were stored on video tape together with the stimulus presentations using a video mixer (Videomat VM2E). Eye position was subsequently determined by measuring pupil position at 200 ms intervals from stimulus onset and converted into direction of gaze relative to the centre of the screen. The animals maintained their fixation during the period when the data were collected and there was no consistent change in saccadic eye movements preceding or at the onset of the presentations of the stimuli Data analysis All neurones recorded for at least one sequence of trials were subjected to analysis. Analyses used only data from correctly performed trials; those on which

4 660 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) the animal failed to fixate on the stimuli or made errors were excluded. There were too few error trials for these to be statistically analysed. Each neurone was first analysed to determine whether it was visually responsive. The mean firing rate in the 0.5 s immediately following stimulus onset was compared to that during the 3 s period before cue onset for each trial. Visually responsive neurones were those for which the mean change in firing rate across all the trials for which the neurone was recorded was significant (paired t-test, P 0.05). These neurones were further analysed to determine whether they responded differentially to the different categories of visual stimuli. An analysis of variance (ANOVA, P 0.05) with factors repeat (first or subsequent presentation), relative familiarity (novel or familiar stimulus) and period (time period after stimulus onset) was performed across all trials on the change in firing rate from that in the 3 s pre-cue period to that in the two 0.25 s periods immediately following stimulus onset on each trial. Neurones for which the interaction between the factors repeat and relative familiarity was significant, or for which both factors were significant, and for which the response to the first presentations of novel stimuli differed significantly from their second presentations and from first or second presentations of familiar stimuli were categorised as novelty neurones. Neurones for which the factor repeat, but not the factor relative familiarity nor the interaction between the two factors was significant were categorised as recency neurones. Neurones for which the factor relative familiarity, but not the factor repeat nor the interaction between the two factors was significant were categorised as familiarity neurones. Neurones for which there was a significant interaction between period and repeat or period and relative familiarity were categorised as recency or familiarity neurones respectively. Neurones for which there was a significant three-way interaction between period, repeat and relative familiarity were categorised as differentially responsive but were not counted as belonging to any of these types: their numbers were not large ( 7%). Differences in incidence of different categories of responsive neurones in different areas were established using two-way ANOVA based on a generalised linear model (GLIM) assuming a binomial error distribution (Baker and Nelder, 1978; Aitkin et al., 1989). The variations (deviances) associated with the factors monkey, area, and hemisphere were determined and are given as BEM 2 values. For every analysis the residual deviance was less than that expected by chance, i.e. the model gave an adequate fit to the data. The different characteristics of the different types of neuronal responses (latencies, memory spans) across the different areas were analysed by analyses of variance with repeat measures (Systat for Windows: Statistics, Version 5, Systat, Evanston, IL). These analyses included the factors: animal, area, type (recency, novelty or familiarity neurone), and time or interval. The factors time and interval and any interactions involving them were treated as within neurone (repeated) measures, other factors and interactions as between neurone measures. As appropriate, time or interval was treated as a covariate (unless otherwise stated by using the rank order rather than the absolute values). Only results that were consistent across monkeys are reported. All tests were two-tailed and used a significance level of P=0.05. Probability values for repeated measures analyses of variance were adjusted for any failures of compound symmetry by using the more conservative values based on Huynh-Feldt statistics (Systat) Histological localisation The depth of each neurone was noted at the time of its recording. At the end of certain recordings microlesions were made by passing a DC current (10 60 A for s) through the microelectrode at known positions near responsive cells. 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. Perfusion and histological processing of the brain were by standard techniques. Coronal sections were cut at 50 m on a freezing microtome and were stained with cresyl violet. The microlesions were identified by the Prussian blue reaction. Using this information the positions of the recorded neurones were marked onto line-drawings of brain coronal sections (every 1 mm apart); corrections were made for X-ray expansion and tissue shrinkage. The boundaries of the various areas followed Lorente de Nó (1934) and Burwell et al. (1995). 3. Results 3.1. Visual responsi eness in the serial recognition task Neurones were recorded between A22 and A8 (Snider and Lee, 1963; see also Fig. 3) in four hemispheres of two monkeys from the entorhinal and perirhinal cortices, area TE of inferior temporal cortex, and the hippocampus (i.e. chiefly from subfield CA3, the dentate gyrus and the subiculum), during the performance of the behavioural tasks. The monkeys used both hands to touch the response screen and no obvious preference for hand usage was observed. No differences in responsiveness between the right and left hemispheres were found. Results from left and right hemispheres have therefore been combined. Of 2072 neurones recorded during the serial recognition task, 1162 (57%) were visually responsive (Table

5 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 3. The locations of the differentially responsive neurones. A. A lateral view of a macaque brain. The distances are numbered posterior (P) to the outline of the sphenoid bone (Aggleton and Passingham, 1981), which is approximately 20 mm anterior to the interaural line. B. A line drawing of coronal section at level 4P indicating brain regions. amts: anterior medial tempral sulcus; rhs: rhinal sulcus: sts: superior temporal sulcus. C. Drawings of sections at levels indicated in A. The locations of the recency ( ), familiarity ( ), and novelty ( ) neurones are shown in the left, middle and right columns respectively. 1). However, in the rhinal cortices and area TE the proportion of the neurones that were visually responsive was significantly higher than in the hippocampus (63% c.f. 14%); see Table 1 and legend for details. There was no significant difference between the proportions of visually responsive neurones found in area TE, perirhinal cortex and entorhinal cortex. In these three areas all responses were excitatory. Across the population of visually responsive neurones in these three areas, the mean firing rate in the 0.5 s following stimulus onset was more than twice ( ) that during the pre-cue period. In the hippocampus the mean response was significantly smaller (mean change in firing rate ). Responses of individual neurones were not the same for all the tested stimuli, but the reported results are independent of this aspect of the neurones responsiveness and it will not be further considered in this paper Incidence of differential responses Stimuli seen many times previously by the animal will be termed familiar. Stimuli that have been seen infrequently if at all at the start of a particular recording session will be termed novel, even when such stimuli are shown for a second time in a particular session. A total of 455 neurones (39% of the 1162 visually responsive neurones) responded differently depending on the relative familiarity or repetition of stimuli. The locations of these neurones are illustrated in Fig. 3. The incidence of such differentially responsive neurones was very much lower in the hippocampus than in the remaining three regions (3% c.f. 38%; see Table 1 legend for details); indeed, the hippocampal incidence might be explained by chance (expected mean chance incidence 5%; significance of change for single differential hippocampal neurone: P=0.04). There was no significant difference in the incidence of such neurones between area TE, the perirhinal cortex and the entorhinal cortex. As only one such neurone was found in the hippocampus the remaining analyses concentrate on neurones in the other three areas. Almost all the response changes were in the direction that the response to familiar stimuli compared to novel stimuli or to second presentations compared to first presentations entailed a respose reduction. Only eight (1%) of 1122 visually responsive neurones (2% of the 454 differential neurones) in rhinal cortex and area TE showed increased activity and none of these increases was significant at the 1% level. Thus the incidence of response increments is significantly (binomial test, P= ) less than would be expected by chance (5%) and there are no neurones in this large sample that had very large increases in response.

6 662 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Table 1 Incidence of recognition-related visual responses in the rhinal cortices, area TE and the hippocampus Responses: Visual-differential Visual Total Recency Familiarity Novelty Subtotal V(n) V/T (%) T(n) D(n) D/V (%) D/T (%) D (n) D/V (%) D/T (%) D (n) D/V (%) D/T (%) D (n) D/V (%) D/T (%) Regions Entorhinal cortex Perirhinal cortex Area TE Hippocampus Total D, differentially responsive neurones; V, visually responsive neurones; T, total neurones recorded; n, number of neurones. The incidence of visually responsive neurones (V) differed significantly between the four regions (BEM 2 =467.1, df=3, P ). In particular the incidence was significantly lower in the hippocampus than in the other three regions (BEM 2 =442.6, df=1, P ), but incidence did not differ significantly amongst the other three regions. Also the magnitude of the mean response was significantly lower in the hippocampus than that for the other areas (independent t-test with non-equal variance: t 58 =2.39, P=0.03). Similarly, the incidence of differentially responsive neurones (D) as a proportion of visually responsive neurones (V) differed significantly between the regions (BEM 2 =29.0, df=3, P 0.001). In particular it was significantly lower in the hippocampus than in the other three regions (BEM 2 =28.7, df=1; P 0.001), but did not differ significantly amongst the other three regions. The incidence of differentially responsive neurones in each of these three regions (though not in the hippocampus) was significantly (binomial test, P 0.001) above chance (5%).

7 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Table 2 Population analyses of the neuronal responses to visual stimuli in the rhinal cortices and area TE Types of neurone: Recency Familiarity Novelty Types of visual stimuli: n First Repeat % n Novel Familiar % n N1 N2 % F1 % F2 % Regions: Entorhinal cortex a b cd c c 49 Perirhinal cortex a b cd c c 33 Area TE a b cd c c 39 Total a b cd c c 39 Data are shown as the ratio of the firing rate (mean S.E.M.) in the post-stimulus period (0.5 s) compared to that (1.0) in the pre-cue period (3 s). Also shown is the mean percentage (%) change in response from that for novel first stimuli to that for the other types of stimuli. N1, novel first; N2, novel repeat; F1, familiar first; F2, familiar repeat; n, numbers of the visual differential neurones recorded in the region. Superscript letters refer to comparisons between response magnitudes (paired t-tests, P 0.01). a Recency neurones, first versus repeat. b Familiarity neurones, novel versus familiar. c Novelty neurones, N1 versus N2, F1 or F2. d N2 versus F1 or F2.

8 664 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 3. The locations of the differentially responsive neurones. Top left: a lateral view of a macaque brain. The distances are numbered posterior (P) to outline of the sphenoid bone (Aggleton and Passingham, 1981), which is approximately 20 mm anterior to the interaural line. Right: line drawings of representative coronal sections taken at levels indicated by vertical lines on the lateral view. The locations of the recency ( ), familiarity ( ), and novelty ( ) neurones are shown in columns A, B and C, respectively. amts: anterior medial temporal sulcus; rhs: rhinal sulcus; sts: superior temporal sulcus. Most (93%) of the 454 differentially responsive neurones could be classified as differing significantly (analysis of variance) in response depending on: the recency of presentation but not the relative familiarity of stimuli (recency neurones; see e.g. Fig. 4); the relative familiarity of stimuli but not their recency of presentation (familiarity neurones; see e.g. Fig. 5); or both recency and familiarity information such that the response to the first presentations of novel stimuli differed from the response to their second presentations or to the response to familiar stimuli (novelty neurones; see e.g. Fig. 6). There were 85 (19% of the differentially responsive neurones) recency neurones, 167 (37%) familiarity neurones, and 171 (38%) novelty neurones. See Fig. 3 for the locations of these neurones. The incidence of these different types of neurone did not vary significantly between the three areas. Population analyses of the mean changes in firing rate on presentation of stimuli on the different types of trial for the three different types of neurone are given in Table 2. The mean response decrement on stimulus repetition is 52% for recency neurones. Mean responses to familiar stimuli are 39% less than those to novel stimuli for familiarity neurones. For novelty neurones the mean decrement on repetition of a novel stimulus is 66% while presentations of familiar stimuli evoke a response 37% less than that to the first presentations of novel stimuli. Typically, the response of novelty neurones reduced in magnitude when a novel stimulus was repeated while, in contrast, the duration of the response for familiar stimuli was much shorter ( 1 s for all 171 neurones and 200 ms for 58 of them) than for novel stimuli (2 s); see e.g. Fig. 6. As a population, novelty neurones displayed no significant mean response decrement on the repetition of familiar stimuli. For details of the mean decrements for the three types of neurone across the three regions see Table Differential latencies The mean latency to respond to stimuli for the populations of the different types of differentially responsive neurones was analysed for the different types of trials; see Fig. 7 and Table 3 and legends for statistical details. For each type of neurone, the shortest mean differential latency ( 75 ms) for the populations was found in area TE. The corresponding latencies in perirhinal cortex were longer at 105 ms for recency neurones and 135 ms for familiarity neurones. They were longer again in entorhinal cortex (Table 3). For novelty neurones the mean perirhinal latency for novel first versus novel second trials was 105 ms, but for

9 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 5. Responses of a familiarity neurone to visual stimuli during the serial recognition task. The neuronal responses to novel (left) and familiar (right) stimuli were shown as peristimulus histograms above rasters (conventions as for Fig. 4). Note that the responses to the novel stimuli were significantly larger in magnitude than to the familiar stimuli regardless of whether the stimulus was appearing for the first or second time in that session, i.e. the response signalled the relative familiarity of the stimulus but not whether it had been seen recently. novel first versus familiar first trials it was the same as in area TE ( 75 ms); the entorhinal latency was longer ( 135 ms) in each case. Thus on average across the recorded populations, changes for recency and familiarity neurones occurred earlier in area TE than in perirhinal cortex or entorhinal cortex. Accordingly, these data raise the possibility that the corresponding changes in rhinal cortex might be passive reflections of changes transmitted from area TE. For the population of novelty neurones information concerning the familiarity of stimuli was available in perirhinal cortex as early as in area TE; however, information concerning the recency of presentation of novel stimuli was available earlier in area TE than in perirhinal cortex Length of memory The memory span of a neurone may be defined as the longest interval following initial presentation of stimuli for which re-presentation of the stimuli results in a significant change in activity (Fahy et al., 1993). Mean memory spans were compared for the three different types of differentially responsive neurone in the three different regions (area TE, perirhinal and entorhinal cortex). The responses of differentially responsive neurones recorded during task sequence S1 were averaged for first presentations of novel stimuli and for their repeat presentations after differing numbers of intervening trials. Population means for the three different types of differentially responsive neurones in the three different areas were constructed; see Fig. 8 and legend for details. As may be anticipated from Fig. 8, the pattern of response decrement across the intervals differed significantly for the different types of neurone (analysis of variance with repeat measures: interval by type interaction: F 16,1144 =5.46; P 0.001). The patterns of response decrement for each of the types of neurone were therefore analysed separately. For the recency neurones the response pattern in the three regions differed significantly (Fig. 8 and legend). The average population response showed decrements in response that reduced at different rates as the time between the first and subsequent presentations increased. In particular, in perirhinal cortex the population memory span was greater than it was in area TE: the response decrement was significantly greater in perirhinal cortex than in area TE at all intervals 16 intervening trials (2 4 min). There was no overall significant difference in the population response decrements between entorhinal cortex and either perirhinal cortex or area TE. The population response decrement was still significant at 24 h in perirhinal cortex, but not in either entorhinal cortex or area TE. However, it should be noted that there were individual examples of neurones with memory spans of 24 h in all three areas, though they were far more common in perirhinal cortex

10 666 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 6. Responses of a novelty neurone to visual stimuli during the serial recognition task. The neuronal responses to the first (top) and second (bottom) presentations of novel (left) and familiar (right) stimuli were shown as peristimulus histograms above rasters (conventions as for Fig. 4). Note that the firing rate was significantly smaller to the second presentations than to the first presentations of novel stimuli, whereas the maximum firing rate was only slightly less but the firing duration was greatly shortened for presentations of familiar stimuli. Thus this neurone responded best to first presentations of novel stimuli and its responses signalled both the relative familiarity of a stimulus and whether it had been seen recently. than in the other areas. Thus memory spans of 24 h were found for 2/13 (15%) recency neurones in area TE, 15/19 (79%) in perirhinal cortex and 3/11 (27%) in entorhinal cortex. For the novelty neurones there was no consistent difference between the response pattern in the three regions. Thus these data did not differentiate between the responses of novelty neurones in area TE, perirhinal cortex and entorhinal cortex. The average population response across the three regions showed reducing decrements in response as the time between the first and subsequent presentations increased, i.e. memory decreased with the passage of time for these neurones. Significant decrements in response were found at all intervals from 0 to 64 intervening trials, but not at 24 h. Although the mean memory span for the whole population of recorded novelty neurones was less than 24 h. It should not be concluded that this restricted memory span applied to all such neurones. Thus memory spans of 24 h were found for 1/16 (6%) novelty neurones in area TE, 3/17 (18%) in perirhinal cortex and 6/24 (25%) in entorhinal cortex. Moreover, Fig. 9 illustrates the large mean response decrement at 24 h for the half of the population (28/57) of novelty neurones with the largest response decrements at 24 h. For the familiarity neurones there was no consistent difference between the response pattern in the three regions. Thus these data did not differentiate between the responses of familiarity neurones in area TE, perirhinal cortex and entorhinal cortex. The average population response across the three regions showed a significant monotonic decrease with increasing interval (Fig. 8 and legend). Thus the response decrement increased, i.e. memory increased, with the passage of time, being greatest at 24 h. The decrement in response between first and second presentations was first significant after 32 intervening trials (4 8 min), being also significant after 64 trials and 24 h. This result indicates that a single presentation of a stimulus after 4 8 min and two presentations of a stimulus after an interval of one day are sufficient for the population of familiarity neurones to respond as though these stimuli were familiar rather than novel. Individual examples of familiarity neurones with memory spans of 24 h were common in all three areas. Thus memory spans of 24 h were found for 19/21 (90%) neurones in area TE, 22/23 (96%) in perirhinal cortex and 14/16 (87%) in entorhinal cortex.

11 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 7. Differential latencies of the populations of recency, familiarity and novelty neurones in the entorhinal and perirhinal cortices and area TE. Plotted are the cumulated counts (30 ms bins, but smoothed for clarity) of action potentials from stimulus onset (time zero) for the different types of trial. The latencies were calculated only for the sample of differentially responsive neurones recorded for more than 50 trials. Statistical details: The mean latency to respond to stimuli for the populations of the different types of differentially responsive neurones was analysed by subjecting the cumulative spike counts from the time of stimulus onset for each neurone in successive 30 ms epochs for each of the different types of trials to an analysis of variance with repeat measures: there was a significant interaction between the factors for area, type of neurone and type of trial (F 6,662 =5.27; P 0.001), and no significant interaction involving the factor for animals. The data for each type of neurone was therefore analysed separately for each area. The differences in cumulative counts between first and second presentations for recency neurones, between novel and familiar stimuli for familiarity neurones, and for both these types of trial for novelty neurones were analysed for successive time bins after stimulus onset across the populations of neurones. The mean time of the earliest bin for which the difference was significant (t-tests, P 0.05) was taken as the mean latency for the population of neurones. In every case all subsequent bins were also significant. Moreover, the change for the ms bin differed significantly between area TE and perirhinal cortex, except for novel first vs. familiar first trials for novelty neurones (see also Table 3) Multiple repetitions of stimuli In sequence type S2 initially novel stimuli were repeated up to five times on successive trials (i.e. within a period of 2 min). As illustrated in Fig. 10, such repetition resulted in only a small, non-significant reduction in the mean response of the whole population of familiarity neurones. Thus multiple repetitions of an initially novel stimulus do not result in the rapid development of a response decrement in a population of familiarity neurones. For recency and novelty neurones there was an immediate and highly significant response decrement for the second appearance. Unexpectedly, this decrement did not become greater with further repetition. There were no significant differences between the three areas in these results. No change in mean response across the recorded populations was found when five different novel stimuli were presented on five successive trials (sequence type S3) Multiple differential recordings More than one differentially responsive neurone or a differentially responsive and a non-differentially responsive neurone were recorded simultaneously on many occasions (e.g. a triplet of a novelty, a familiarity and a recency neurone on eight occasions; a pair of differently differential neurones on 46 occasions); see for example Fig. 11. Such differences in patterns of response to the same stimuli for simultaneously recorded neurones exclude generalised modulations in arousal or attention or peripheral visual changes as explanations for their response differences.

12 668 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Table 3 Mean differential latencies (ms) by area and response type Area TE Perirhinal cortex Entorhinal cortex Recency First versus second Familiarity Novel versus familiar Novelty Novel first versus novel second Novel first versus familiar first The values are times (in ms) from stimulus onset to the mean of the first bin for which there was a significant difference between the types of trial in a given area for a given type of neuronal response. The width of the bins was 30 ms. The mean difference in counts (expressed as the normalised % change for each neurone) between the types of trial for the ms bin in area TE was compared (independent t-tests for groups with unequal variance) to the corresponding mean difference in perirhinal cortex for each type of response: recency neurones, t 41 =2.61, P=0.013; familiarity neurones t 60 =7.47, P 0.001; novelty neurones-n1 versus N2-t 40 =4.06, P and-n1 versus F1-t 59 =0.89, P 0.1. Thus population latencies were significantly shorter in area TE than in perirhinal cortex for each type of response, except for novel first versus familiar first trials for novelty neurones Neuronal responses in the conditional isual discrimination task The activity of 342 neurones was recorded during performance of the conditional visual discrimination task; 228 (67%) of the neurones were visually responsive. The responses of 114 neurones (50% of those visually responsive) differed significantly (analysis of variance) for the different stimulus configurations. In particular, there were only 21 neurones (9% of the visually responsive neurones) that responded differently on trials requiring a touch to the left rather than the right side of the screen. Moreover, none of these neurones was also differentially responsive in the serial recognition task. Indeed, no neurone was differentially responsive in both tasks. Thus the differential responses in the serial recognition task are not explicable solely on the basis of the direction of the animal s behavioural response. 4. Discussion The results advance understanding of the neuronal encoding of information of potential importance to recognition memory (judgement of prior occurrence) in several ways. (i) A third commonly occurring type of neuronal response, namely that of novelty neurones has been discovered to add to those of recency and familiarity neurones (Fahy et al., 1993). Novelty neurones respond significantly differently (typically more strongly) to the first presentations of novel stimuli than to their subsequent presentations or to presentations of highly familiar stimuli. (ii) It has been established that each of the three types of neurone are found commonly in area TE, perirhinal cortex and entorhinal cortex. A previous report (Fahy et al., 1993) had suggested a more restricted anatomical distribution for recency and familiarity neurones. (iii) It has been confirmed within a single experiment that these types of neurones (now additionally including independent assessments of recency, familiarity and novelty neurones) are uncommon in the hippocampus (Brown et al., 1987; Riches et al., 1991). (iv) The memory spans of exemplars of each of these types of neurones have been demonstrated to be at least 24 h. (v) The mean memory span of recency neurones recorded in perirhinal cortex has been shown to be significantly longer than that for area TE. This finding suggests that the differential responses of recency neurones in perirhinal cortex are more than mere passive reflections of those in area TE. (vi) The differential latencies of populations of recency, familiarity and novelty neurones have been compared across the different areas. The mean latencies of recency and familiarity neurones are significantly shorter in area TE than in perirhinal cortex, and in perirhinal cortex than in entorhinal cortex. This finding makes it probable that such differential responses in area TE are not mere passive reflections of those in perirhinal cortex. Accordingly, both area TE and perirhinal cortex may play roles in the generation of long lasting decremental response changes. The results neither exclude nor provide evidence that entorhinal cortex does so. (vii) The reduced response of familiarity neurones to a familiar compared to an unfamiliar stimulus appears to be dependent on the passage of time rather than purely on multiple repetitions of a stimulus. Indeed, for the recorded population of these neurones memory for the prior occurrence of a novel stimulus improved with the passage of time (from 4 8 min to 24 h). (viii) The average rate of development of the differential response for familiarity neurones is in marked contrast to that for novelty and recency neurones where the change occurs as soon as an unfamiliar stimulus is repeated, and then gradually declines. Accordingly, the plastic process underlying the development of the differential

13 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 8. Mean population responses across different intervals for the three types of differentially responsive neurone in the three areas. Across the intervals investigated, memory across the population increases with time elapsed for familiarity neurones, but decreases for novelty neurones and for recency neurones. The sample comprised all the recorded differentially and decrementally responsive neurones for which there were data at all intervals. * Significant difference between mean response to first and subsequent presentations within an area and type of neurone (paired t-test; P 0.05, uncorrected for multiple comparisons). Statistical details are from analyses of variance with repeat measures. For none of the analyses was there a significant interaction between animal and other factors, i.e. there was no evidence of a different pattern of responses between the animals. Recency neurones: Using the rank order of the number of intervening trials from the first to the subsequent presentations (interval) as a covariate, there was a significant interaction between region and interval (F 2,37 =3.80; P=0.03); thus the rate of change (slope) of the response decrement with interval varied between the regions. In particular, the rate of change (slope) was significantly less in perirhinal cortex than in area TE (F 1,28 =7.63; P=0.01). Data for the different intervals were therefore analysed separately. Significant differences (F tests, P 0.05) between the three areas were found for all intervals 16 intervening trials. In particular, in perirhinal cortex the population memory span was greater than it was in area TE: the response decrement was significantly (Tukey tests) greater in perirhinal cortex than in area TE at 16 (P=0.03), at 32 (P=0.04) and 64 (P 0.001) intervening trials and at 24 h (P=0.04). Familiarity neurones: There was no significant interaction between region and interval nor effect of region, thus the regions did not differ. However, there was a significant effect of interval (F 8,432 =4.37; P 0.001); indeed, there was a significant monotonic decrease in response (increase in decrement) with increasing interval (F 1,54 =18.49; P 0.001). The difference in response between first and subsequent presentations was significant (paired t-tests, Bonferroni-corrected P-values) at intervals of 32 (P 0.05) and 64 (P 0.01) intervening trials and at 24 h (P 0.001). Novelty neurones: There was no significant interaction between region and interval nor effect of region, thus the regions did not differ. However, there was a significant effect of interval (F 8,408 =10.67; P 0.001). The difference in response between first and subsequent presentations was significant (Bonferroni-corrected t-values; P 0.01) at all intervals up to 64 intervening trials but not at 24 h. response of familiarity neurones must have a much slower time course of expression than that for recency or novelty neurones. (ix) It has been confirmed that the incidence of incremental rather than decremental responses on repetition of an unfamiliar stimulus is very low (less than chance expectation). Much evidence that the differential responses are related to stimulus repetition per se and are not generated artefactually has been presented previously (Riches et al., 1991; Fahy et al., 1993; Miller et al., 1993; Sobotka and Ringo, 1993; Brown, 1996; Brown and Xiang, 1998). As previously, the differences cannot be explained as due to: differences in levels of alertness or attention (types of trials were intermingled and not predictable by the animal), differing reward values of the stimuli (all trials were equally rewarded), differences

14 670 J.-Z. Xiang, M.W. Brown / Neuropharmacology 37 (1998) Fig. 9. Comparison of the responses of recency, familiarity and novelty neurones to visual stimuli presented yesterday and today. The animals were shown the same file of visual stimuli after an interval of 24 h. Data were analysed for the 28 neurones of each type of differential response with the most highly significant decrements. Changes in neuronal firing rate (%) were relative to the pre-cue level (spontaneous activity, SA=0%). The data are shown as means S.E.M.. * P (paired t-test). in behavioural response (no differentially responsive neurones in the serial recognition task were also differentially responsive in the conditional discrimination task requiring the same behavioural response) or eye movements (analysis of critical recordings has demonstrated that neuronal response changes occur before eye movement changes). Additionally, in this study there were many examples of more than one differentially responsive neurone or a differentially responsive and a non-differentially responsive neurone being recorded simultaneously. Such response differences between simultaneously recorded neurones cannot be produced by a non-specific, generalised change in alertness. Further, consider the situation when a recency neurone and a familiarity neurone were recorded simultaneously (19 such pairs were observed). When an unfamiliar stimulus was seen again, the response of the recency neurone decremented while the familiarity neurone continued to respond strongly to the second presentation. In contrast, when a familiar stimulus was first presented, the recency neurone responded strongly while the familiarity neurone did not. This co-occurring dissociation of responsiveness to presentations of the same stimuli cannot be explained as a result of a generalised change in attention to the stimuli. It has been argued previously that, on the contrary, the differential neuronal response changes provide a basis for producing changes in attention or eye movements (Fahy et al., 1993; Miller et al., 1993). The results confirm that the change in response with repetition or increasing familiarity is overwhelmingly a reduction. The incidence of increases in response was less than expected by chance. Further, none of the observed increases in response was large in magnitude or was unexpectedly low in probability of occurrence. This result is in apparent conflict with certain ideas concerning the setting up of new representations in anterior inferior temporal cortex. It could be expected that experiencing novel stimuli would result in the setting up of new representations. Setting up a new representation of a stimulus is often assumed to result in an enhanced responsiveness to that stimulus of at least some members of a neuronal assembly, though for information storage capacity to be maximised, the proportion of synapses (and therefore, probably, neurones) undergoing modification should be small (Marr, 1971; Amari, 1989; Rolls, 1995). Nonetheless, even given such sparse encoding, there was no evidence of the predicted incremental neuronal responses during the performance of a recognition memory task employing many unfamiliar stimuli. There are two possible reasons for this unexpected result. Firstly, the representations and hence the increased responses could occur elsewhere than in anterior inferior temporal and entorhinal cortex, though the perceptual processing capacities of anterior inferior temporal cortex would seem to make it an obvious place for such representations to be formed. Secondly, perhaps it is not necessary to set up a new