The Journal of Neuroscience, November 1990, fO(11): 3479-3493

Amygdalar Interaction with the Mediodorsal of the and the Ventromedial in Stimulus-Reward Associative Learning in the Monkey

David Gaffan’ and Elisabeth A. Murray* ‘Department of Experimental Psychology, Oxford University, Oxford OX1 3UD, England, and *Laboratory of Neuropsychology, National Institute of Mental Health, Bethesda, Maryland 20892

Cynomolgus monkeys (Macaca fascicularis) were assessed One kind of associative learning in which the amygdala is of for their ability to associate visual stimuli with food reward. central importance is the associationof a sensorystimulus with They learned a series of new 2-choice visual discriminations the incentive value of food reward. In visual-discrimination between colored patterns displayed on a monitor screen. learning for food reward, the role of the amygdala varies ac- The feedback for correct choice was the delivery of food. In cording to the associative structure of the task. If visual dis- order to promote associative learning between the visual criminative stimuli are associateddirectly with the incentive stimuli and the incentive value of the food reward, reward value of the food reward, efficient learning dependson an in- delivery was not accompanied by any distinctive visual feed- trahemispheric interaction between the amygdala and the vi- back on the display screen. The rate of learning new prob- sual-associationcortex ipsilateral to it (E. A. Gaffan et al., 1988). lems was assessed before and after surgery in a total of 16 But when visual discriminative stimuli are associatedwith the monkeys. Three groups of 3 monkeys received bilaterally incentive value indirectly, via the mediation of a direct asso- symmetrical ablations in either the amygdala, the medio- ciation with an auditory or visual secondary reinforcer, visual dorsal nucleus of the thalamus, or the ventromedial pre- interaction with the amygdala is less important (Gaffan and frontal cortex. All these groups showed a severe postop- Harrison, 1987; Gaffan et al., 1989). erative learning impairment. Seven other animals were given In the presentexperiment, we examined a task in which visual a unilateral ablation in 1 of those 3 structures and a second discriminative stimuli are associateddirectly with the incentive unilateral ablation, in each case contralateral to and different value of the food reward. Our aim was to determine which from the first, in order to produce 2 groups in which a putative structures, apart from the visual-associationcortex, the amyg- amygdalo-thalamo-prefrontal pathway had been discon- dala interacts with in this task. E. A. Gaffan et al. (1988) showed nected by crossed unilateral lesions. One disconnection that this type of learning is severely impaired by crosseduni- group had lesions in the amygdala and ventromedial pre- lateral lesions of the amygdala in one hemisphereand of the frontal cortex; the other had lesions in the amygdala and the visual-association cortex in the other hemisphere.This com- mediodorsal nucleus of the thalamus. The disconnection bination of lesions disconnectsthe intrahemispheric pathway groups showed a significant impairment, but the effect of from the visual-associationcortex to the amygdalaby destroying the disconnection surgeries was significantly milder than the its origin in one hemisphereand its destination in the other. effect of any of the 3 bilaterally symmetrical lesions. There- The results suggestedthat visual information and incentive- fore, symmetrical bilateral lesions in either the amygdala, value information converge in the amygdala and are associated the mediodorsal nucleus, or the ventromedial prefrontal cor- together there. If so, then the product of this associativeprocess tex produce similar impairments in the present task, implying (the information, retrieved from , that a certain visual that these structures are functionally related to each other; stimulus has been paired in the past with food) needsto be but the relatively mild effect of disconnecting these struc- relayed to further projection targets in order to elicit ultimately tures from each other argues against the hypothesis that a motor response,namely, an instrumental discriminative re- they are serial stages in a single, tightly linked functional sponseto that stimulus. Possiblestages of this further pathway pathway. can be tested by disconnection experiments. Disconnection of the amygdalafrom further projection targetsin the hypothetical pathway should disrupt the present type of visual associative- learning task in the same way as does disconnection of the Received Jan. 17, 1990; revised June 6, 1990; accepted June 8, 1990. amygdala from visual-associationcortex. The research was supported by the Medical Research Council and by NATO The amygdala and adjacent periallocortical regions send a Grant 0 184/87. We thank S. Watkins for assistance in training the monkeys. Send direct ipsilateral projection to the medial part of the mediodorsal reprint request to either D. Gaffan, Department ofExperimental Psychology, South Parks Road, Oxford OX1 3UD, England, or E. A. Murray, Laboratory of Neu- thalamus (Porrino et al., 1981; Aggleton and Mishkin, 1984; ropsychology, National Institute of Mental Health, Building 9, Room IN-107, Russchenet al., 1987; Gower, 1989). The medial part of the Bethesda, MD 20892. Correspondence should be addressed to D. Gaffan, Department of Experimental mediodorsalthalamus, in turn, projects heavily to the ipsilateral Psychology, South Parks Road, Oxford OX1 3UD, England. ventromedial prefrontal cortex (Tobias, 1975; Goldman-Rakic Copyright 0 1990 Society for Neuroscience 0270-6474/90/l 13479-15$03.00/O and Porrino, 1985; Giguere and Goldman-Rakic, 1988).Bache- 3480 Gaffan and Murray * Amygdala-MD-Frontal: Is It a System?

valier and Mishkin (1986) have argued from behavioral results VF/A, n = 4) or in the mediodorsal thalamus and the amygdala (group in recognition memory tasks that “the medial temporal, medial MD/A, n = 3). All of the animals in the disconnection groups were first given one unilateral lesion alone and were tested behaviorally after this thalamic and ventromedial prefrontal regions of the brain are first surgery (Post 1) in order to verify that a single unilateral lesion was each apparently part of an integrated memory circuit” (p 260). without effect in the present task. They were then given a seeond lesion, In the present task, therefore, a simple and attractive hypothesis contralateral to the first, and a further behavioral test (Post2) was ad- is that information and control flows from the visual discrim- ministered. All 3 animals in group MD/A had the lesion in the medio- inative stimulus, associated in memory with food reward, to the dorsal thalamus first and the lesion in the amygdala second. Among group VF/A, 2 animals(Dl and D2) had the ventromedialprefrontal instrumental discriminative response along a tightly linked lesionfirst, whilethe other2 (D3 andD4) hadthe amygdalalesion first. pathway through the visual-association cortex, amygdala, thal- Finally, groupVF/A wassubjected to forebraincommissurotomy and amus. and prefrontal cortex. If this hvoothesis is true. then weregiven a third behavioraltest (Post3). bilaterally symmetrical lesionsin any ofthese structuresshould Surgical methods. The surgical.opera&ons were carried out under sterileconditions. Barbiturate was infused through an intravenouscan- disrupt this type of learning. Furthermore, crossedunilateral nulaat a rate sufficientto maintaindeep anesthesia. The ablationswere lesionsof the amygdala and ventromedial prefrontal cortex, or madeby aspirationunder visual guidance with the aid of an operating of the amygdala and mediodorsal thalamus, should disrupt microscope. The tissue was closed in layers with silk sutures. A pro- learning as effectively as bilateral lesionsin each of thesestruc- phylactic dose of antibiotic was administered at the end of the surgery. tures, becausethey should disconnect the pathway of the flow Postoperative recovery was in all cases uneventful. The surgical method for making amygdala lesions was identical to in the sameway asthe crossedunilateral lesionsof the amygdala that described by Gaffan and Harrison (1987). The medial temporal and visual association cortex did in the experiment by E. A. lobe was exposed by retracting the frontal lobe over the orbit. The pia Gaffan et al. (1988). mater was cauterized in an area approximately 5 mm in diameter on Therefore, we compared the learning disability produced by the medial surface of the temporal lobe medial and superior to the rhinal sulcus. The amygdala and the periamygdaloid cortex medial to the bilateral lesions with the learning disability produced by dis- amygdala were then ablated by aspiration through the defect in the pia connection of the hypothesized pathway. One disconnection mater. of the temporal stem, and the lateral ventricle and group had crossedunilateral lesions of the ventromedial pre- anterior surface of the hippocampus, were visible boundaries of the frontal cortex and the amygdala; in addition, this group was ablation posteriorly, laterally, and inferiorly. subsequentlytested, after commissurotomy, for the To produce an ablation in the ventromedial frontal cortex, a bone flap was raised over the frontal lobe, and the dura mater was cut in a possibility that their residual learning ability was mediated by Y shape to expose the whole of the dorsolateral surface of the frontal interhemispheric pathways. Another disconnection group had lobe. The lobe was then retracted from the orbit in order to expose its crossedunilateral lesionsof the amygdalaand mediodorsalthal- inferior surface. The pia mater was cauterized along the medial bank amus. These 2 disconnection groups were compared with 3 of the lateral orbital sulcus, and this line of cautery was extended pos- teriorly to the lateral sulcus and anteriorly to the lip of the inferior groups having bilaterally symmetrical lesionsin the amygdala, convexity. The line of cautery was further continued first along the lateral mediodorsalthalamus, or ventromedial prefrontal cortex. The sulcus in a medial direction, then along the cortex adjacent to the ol- lesionsof the ventromedial prefrontal cortex were designedto factory tract at the most medial part of the inferior convexity, and finally remove the part of the prefrontal cortex that receivesthe heav- along the lip of the inferior convexity to meet the original cautery along iest projection from the medial part of the mediodorsalnucleus the line of the lateral orbital sulcus. The cortex and pia mater bounded by this continuous line of cautery was then removed by aspiration. of the thalamus. Finally, the lesion was extended by aspiration into the cortex in the In addition to the projection relayed via the mediodorsal inferior part of the medial wall of the hemisphere, the intention being thalamus, the ventromedial prefrontal cortex receives a direct to extend the lesion to the level of the rostra1 sulcus, which, however, projection from the amygdala(Porrino et al., 1981; Amaral and was not prominent in most cases. The dura mater was sewn, and the wound was closed in layers. Price, 1984).If this direct projection is of functional importance For the ablations in the mediodorsal thalamus, the method employed in the present task, crossedunilateral lesionsof amygdala and was similar to that described by Aggleton and Mishkin (1983). The right ventromedial prefrontal cortex should have a greatereffect than hemisphere was exposed and retracted over the splenium of the corpus crossedunilateral lesionsof the amygdalaand mediodorsalthal- callosum. The splenium was cut over the midline with a glass aspirator. amus.Similarly, bilateral ventromedial prefrontal lesionsshould The membrane covering the thalamus was cut in the midline by applying cautery in order to expose the posterior commissure, the third ventricle, have a greatereffect than bilateral mediodorsalthalamic lesions. and the posterior 5 mm of the midline thalamus. Next, the most pos- The present experiment allowed these subsidiary comparisons terior 5 mm of the massa intermedia of the thalamus was cut in the to be made. midline with a glass aspirator. Tissue adjacent to the cut massa inter- media was removed by aspiration with a metal aspirator, either bilat- erally (group MD) or on 1 side only (group MD/A). Approximately 2 Materials and Methods mm of tissue lateral to the midline was removed. The lesion extended Subjects. Sixteen experimentally naive male cynomolgus monkeys (Ma- 5 mm anterior from the posterior end of the thalamus. caca fkscicularis) wereused in the experiment.At the time of surgery For forebrain commissurotomy in group VF/A, the methods were (see below) their averaged weight was 4.4 kg. Before beginning the similar to those previously described (Gaffan and Harrison, 1987). The experimental task described below, they were first trained to touch corpus callosum and the anterior commissure were sectioned in the patterns displayed on the television screen. This preliminary training, midline. including magazine training and autoshaping, was identical to that de- Histology. The animals were deeply anesthetized, then perfused through scribed by Gaffan et al. (1984, p 202). the heart with saline followed by formol-saline solution, The brains Surgical groups. Upon completing preoperative training as described were blocked in the coronal stereotaxic plane posterior to the lunate below, the animals were divided into 5 surgical groups. Nine animals, sulcus, then removed from the head, photographed, and allowed to sink constituting 3 groups of 3 animals each, were given bilaterally sym- in a sucrose formalin solution. The brains were cut in 50-&m sections metrical lesions. One group had bilateral amygdalectomy (group A), on a freezing microtome. Every fifth section was retained and stained another had bilateral lesions in the ventromedial prefrontal cortex (group with Cresyl violet. The series of sections from each animal was examined VF), and a third had bilateral lesions of the medial portion of the microscopically. rnediodorsal thalamus (group MD). The remaining 2 groups were dis- Group VF. The lesions in this group were as intended. The cortex on connection groups in which crossed unilateral lesions were produced the orbital surface extending medially from the lateral orbital sulcus either in the ventromedial prefrontal cortex and the amygdala (group was removed bilaterally. The lateral bank of the lateral orbital sulcus The Journal of Neuroscience, November 1990. IO(11) 3491

Figure 1. Five sections, 2 mm apart, through the lesions in animal VF3. Figure 2. Three sections, 2 mm apart, through the lesions in animal A3. The Journal of Neuroscience, November 1990, fO(11) 3483 and the inferior convexity of the frontal lobe and its dorsolateral surface and VF. In addition, the corpus callosum and anterior commissure were were intact. On the medial wall of the hemisphere, the most inferior 3- sectioned in all cases. A series of sections from a representative animal 5 mm of cortex was ablated, but the rest was intact. A series of sections is shown in Figure 8. from a representative animal in this group is shown in Figure 1. Apparatus and stimuli. The computer-controlled apparatus, described Group A. The bilateral amygdalectomies in this group were similar in detail by Gaffan et al. (1984) displayed pairs of complex patterns to the unilateral amygdalectomies in the study by E. A. Gaffan et al. against a plain gray background on a television screen. The stimuli were (1988). The amygdala removals were complete except for small parts presented 90 mm to the left and right of the screen’s center. The monkey of the posterior dorsal amygdaloid complex in some animals. A series reached out through the bars of a transport cage to touch stimuli on the of sections from a representative animal in this group is shown in screen, Infrared beams crossing the surface of the screen detected the Figure 2. animal’s touches. Each of the 2 visual stimuli that constituted a dis- Group MD. The terminology we have adopted for describing the crimination problem was made of a small shape superimposed on a posterior thalamus is taken from Olszewski (1952), modified in accor- larger shape (approximate height, 20 mm and 40 mm, respectively). dance with Burton and Jones (1976) and Asanuma et al. (1983). The Each of the shapes was taken at random from a predefined set of 127 term rhomboid nucleus (Berman and’Jones, 1982) is used to refer’to the symbols and was displayed in a color chosen at random from 255 collection of areas designated by Olszewski ( 19 5 2) as parts of the central possibilities. The random-number generator that determined stimuli for nucleus (e.g., superior central nucleus, densocellular central nucleus, each problem and their left-right position at every trial was seeded with intermediate central nucleus). The region Olszewski refers to as the the session number, so the stimulus sequence was unique to each session, inferior central nucleus is termed the central medial nucleus (see Jones, but was the same in a given numbered session for all monkeys. 1985). Food rewards were dispensed into a bowl just in front of the lower In order to facilitate identification of nuclear boundaries in the ani- edge of the screen. The rewards were either peanuts, sugar-coated puffed mals with bilateral MD lesions, sections adjacent to those stained with rice, or banana-flavored pellets, according to the preference of the mon- Cresyl violet were fiber stained by a modified Gallyas method. To aid key. The only visible illumination in the experimental cubicle came the reconstruction ofthe lesions in this group, we first prepared drawings from the television screen. The background color of the screen, against of a series of 8 sections at l-mm intervals taken through the posterior which stimuli were displayed, was a medium gray that shed enough thalamus of a cynomolgus monkey in which the posterior thalamus was light by which to see. An infrared light source allowed the monkey to intact. (This animal had received a fomix transection in the course of be watched over closed-circuit television. a different experiment.) The brain of this animal had been processed in Task. Each daily session had 5 new visual-discrimination problems the same manner as those of the animals with bilateral MD lesions in presented serially. For each new problem, a positive and a negative the present experiment. The lesions in the MD group were illustrated stimulus were chosen at random from the large population of available by plotting them onto the drawings of the intact posterior thalamus. stimuli (see Apparatus, above). The problem was presented for 20 con- Only 6 sections are illustrated (Fig. 3), because no significant amount secutive trials. The intertrial interval was 10 set throughout. For each of damage in the MD animals was observed in the most anterior and trial within a problem, the 2 stimuli were presented simultaneously, most posterior of the 8 sections used. Drawings of the 6 intact sections occupying the left and right positions at random, and the animal chose are shown in Figure 3 next to photographs of the Nissl-stained material 1 stimulus by touching it. The stimulus not chosen then immediately from the animal on which they are based. disappeared from the screen, while the chosen stimulus blinked off once Figure 4 shows photomicrographs of Nissl-stained coronal sections (20 msec), then on again for 0.5 sec. During the 0.5-set redisplay of the through the lesion in animal MD1 at 2 of the levels (numbers 4 and 5) chosen stimulus, a food reward was dispensed if the chosen stimulus that are illustrated in Figure 3. The reconstruction ofthis animal’s lesion, was the positive one. plotted onto our drawings of the intact thalamus, is shown in Figure 5 Trainingschedule. After initial shaping, as described in Subjects, above, (top panel). Figure 5 (lower panel) and Figure 6 (top panel) show the the animals were given 45 preoperative sessions (225 new discrimina- lesions in animals MD2 and MD3, reconstructed in the same way. tion problems) of the task described above. Following each surgery, 20 All 3 animals in the MD group had extensive bilateral lesions in the sessions (100 new problems) of the same task were given. magnocellular medial region of the mediodorsal nucleus (MDmc). The largest lesion was in animal MD3, with only slight sparing of the most Results caudal portion. The lesions in MD1 and MD2 left a small amount of the rostra1 MDmc intact and spared the most caudal portion. The par- To measure the effect of surgery, speed of learning was assessed vocellular lateral portion of the mediodorsal nucleus was damaged in for each animal in the last 5 sessions (25 problems) of each all cases, but the damage was minor except in the right hemisphere of postoperative behavioral test and compared with that animal’s animal MD2, where an infarction resulted in more extensive gliosis and cell loss. Damage outside the MDmc was similar in the 3 animals: speed of learning in the last 5 sessions (25 problems) given Among the intralaminar thalamic nuclei, there was extensive damage preoperatively. Data from the early postoperative sessionsgave to the rhomboid nucleus and the central medial nucleus, but the para- qualitatively similar results to those obtained from the last 5 fascicular and center median nuclei were damaged only in their medial sessions,but with greater variance among individual animals parts. Other nuclei of the midline thalamus, including the parataenial within groups. The method we have adopted, that of giving 20 and reuniens, were also damaged, and the paraventricular nucleus of the was removed for the entire anteroposterior extent of postoperative sessionsand basingour conclusionsprimarily on the lesions. The anterior thalamus appeared normal in each of the 3 the last 5 of the postoperative sessions,is the same as that brains. All 3 animals had the sagittal section of the corpus callosum and adopted in several recent similar studies(e.g., Gaffan and Har- hippocampal commissure overlying the posterior thalamus, and some rison, 1987; E. A. Gaffan et al., 1988; Gaffan et al., 1989). We gliosis was observed in the fimbria and fomix. have generally observed that the first few postoperative sessions Group MD/A. The unilateral amygdala lesions in this group were similar to those made bilaterally in group A. The unilateral MD lesions can give highly variable results,and this is understandablegiven in this group were reconstructed in the same way as in the MD group the variable reactionsof individual animals to a prolongedbreak (described above). Figures 6 and 7 show the individuals’ unilateral in daily behavioral testing, with or without surgery. lesions. Damage to the midline thalamic nuclei and to the corpus cal- Figure 9 showsdata from an individual animal in order to losum was similar in the MD/A group to that observed in the MD group, but the MDmc was left intact unilaterally in the MD/A group. illustrate how the efficiency of learning was measured.When Group VF/A. The unilateral amygdala and ventromedial prefrontal errors (choicesof the negative stimulus) are plotted againsttrial lesions in this group were similar to those made bilaterally in groups A position within a problem and averaged over 25 problems, it

Figure 3. Six sections, 1 mm apart, through the intact posterior thalamus of a monkey not taking part in the present experiment. Next to each section is the drawing of that section, on which the lesions were reconstructed in groups MD and MD/A (see Figs. 5-7). For definitions of abbreviations, see Appendix. Scale bars, 2 mm. \CalYdLEx- 7

1 5 I

6\ t I : 3486 Gaffan and Murray l Amygdala-MD-Frontal: Is It a System?

Table 2. Results from the animals with disconnection surgery

Group Animal Pre Post 1 Post2 Post3 W/A VF/Al 6.5 9.9 8.8 7.2 VF/A2 5.1 10.5 15.2 19.6 VF/A3 12.8 12.8 25.1 21.1 VP/A4 8.2 6.3 17.9 20.0 Mean 8.3 9.9 16.8 17.0 MD/A MD/Al 14.9 13.7 19.8 - MD/A2 10.9 13.3 19.4 - MD/A3 9.9 9.7 13.1 - Mean 11.9 12.2 17.4 - Each cell shows percent error in T2-20 of 25 new discrimination problems.

can be seenthat the animal beginseach problem at chanceon average (50 nercent errors). then reducesthe rate of errors as learn&g ‘proceeds.Thus the error rate in trials 2-20 (T2-20) falls quickly if learningis efficient, but slowly if learningis poor. Therefore, the average rate of errors in T2-20 can be taken as an index of the animal’s speedof learning. Figure 9 showsthe animal’slearning curve beforeand after operationand alsoshows, on the samescale, the averagepercent error for T2-20 pre- and postoperatively. Finally, the figure showsthe difference (D) be- tween the pre- and postoperative T2-20 averages.This differ- encewas the measureof eachanimal’s deficit. The deficit is the decline in the animal’s speedof learning as a result of surgery. Tables 1 and 2 show the mean percent error for T2-20 for each animal at each surgicalstage of the experiment. Figure 10 showsthe deficits for the groups and the individuals. Hence, in the caseof animal A3, whose data is shown in Figure 9, the averagesshown in Figure 9 as the bars labeled T2-20 Pre and Post appear in Table 1 as the percentageerror rates 7.6 and 22.3, and the difference between these 2 error rates, 14.7, ap- pearsas the bar labeled D in Figure 9 and as the lowest of the 3 dots in group A in Figure 10. Figure 4. Two sections, 1 mm apart, through the lesion in animal All the animals with bilaterally symmetrical lesions(groups MDl. Scale bar, 2 mm. A, MD, and VF) showedsubstantial postoperative impairments. Among the animals in the disconnection groups (MD/A and VF/A), it should first be noted that a singleunilateral lesion did not substantially affect learning rates(Table 2, Post1). However, the addition of-a crossedasymmetrical lesion resulted in con- Table 1. Results from the animals with bilateral symmetrical lesions sistent deficits (Post2). Figure 10 showsthe disconnectiondef- icits (Post2 minus Pre), as well as the deficits of the groupswith Group Animal Pre Post bilaterally symmetrical lesions. A Al 20.0 35.4 To assessthe statistical significanceof the comparisonsshown A2 6.9 25.7 in Figure 10, we computed designedcomparisons using the A3 1.6 22.3 pooled within-group error term with 11 degreesof freedom. Mean 11.5 21.8 Overall, the effect of the bilaterally symmetrical surgerieswas more severethan the effect of the disconnection surgeries(I; = MD MD1 1.6 16.8 9.237; df = 1,ll; p < 0.02). There wasno significant difference MD2 19.8 37.1 among the 3 different bilaterally symmetrical surgeries(F < 1; MD3 11.8 48.2 df = 2,ll) and no significant difference between the 2 discon- Mean 13.1 34.0 nection surgeries(F < 1; df = 1,ll). The bilaterally symmetrical VF VFl 9.5 28.6 surgeries,on average,produced a significantpositive deficit, that VF2 10.3 20.4 is, significantly greater than 0 (F = 59.788; df = 1,ll; p < O.Ol), VF3 9.1 21.8 and the disconnectionsurgeries, on average,similarly produced Mean 9.6 25.6 a significant positive deficit (F = 7.656; df = 1,ll; p < 0.05). Each cell under Pre and Post shows percent error in T2-20 of 25 new dixrimination Therefore, we conclude that both bilaterally symmetrical sur- problems. geriesand disconnection surgeriesproduced a deficit, that the The Journal of Neuroscience, November 1990, fO(11) 3487

MD1

Figure 5. Reconstructions of the le- sions in animals MD1 (upperpanel) and MD2 (lowerpanel). The 6 levels shown are 1 mm apart. For definitions of ab- MD2 breviations, see Appendix.

bilaterally symmetrical deficit was more severe than the dis- within groups could be explained by variation in the extent of connection deficit, and that there wasno evidence of differences the lesions. In the MD group, it is noticeable that the animal in severity betweenthe different kinds of bilaterally symmetrical with the largest deficit, MD3, also had the largest lesion (see surgery or betweenthe different kinds of disconnection surgery. Fig. 6). In group VF, the animal with the smallestdeficit, VF2, The Post3 scoresin Table 2 show that the addition of fore- wasnot found to have any lessdamage than the other 2 animals, brain commissurotomy did not exacerbate the impairment of whose scoreswere similar to each other. In the A and MD/A the VF/A group. groups, there was little variation among individual animals’ Within each surgicalgroup, individual animals’ deficit scores behavioral results.In the VF/Agroup, the animal with the small- were compared with their histological results in order to assess est deficit had the full intended removal (Fig. 8). Therefore, the possibility that behavioral variation between individuals there is little evidence that variations in lesionswithin groups 3488 Gaffan and Murray * Amygdala-MD-Frontal: Is It a System?

MD3

Figure 6. Reconstructions of the le- sions in animals MD3 (upperpanel) and MD/Al (lower oanel). The 6 levels shown are 1 m&apart. For definitions of abbreviations, see Appendix. MD/Al contributed to behavioral variation among individuals, apart most salient feedback to indicate that a choice wascorrect was from the suggestivecase of animal MD3. the delivery of food to a site spatially separatefrom the dis- criminative stimulus. Thus, the conditions of the present task Discussion were similar to those in the experiment by E. A. Gaffan et al. Bilateral amygdalectomy produced a substantialand consistent (1988), where an impairment was observed in a task that pro- deficit in the present task. This effect is in accordancewith the vided no local differential visual feedback for correct choices, analysisof amygdalarfunction in visual discrimination learning but were different from those in the experiment by Gaffan et al. put forward by Gaffan and Harrison (1987), E. A. Gaffan et al. (1989), where no impairment was observed in a task that did (1988), and Gaffan et al. (1989). In the present task, there was provide local visual feedback for correct choices. The present no localvisual feedbackon the monitor screento indicate whether resultsfrom bilateral amygdalectomy thus support the idea that the animal’s choice had been correct or incorrect. Instead, the the importance of the amygdalafor visual-discrimination leam- The Journal of Neuroscience, November 1990, fO(11) 3489

MD/A2

Figure 7. Reconstructions of the le- sions in animals MD/A2 (upper panel) and MD/A3 (lower DanelI The 6 levels shown are 1 mm apart.

Figure 8. Thirteen sections, 2 mm apart, through the lesions in animal VFI Al. The Journal of Neuroscience, November 1990, fO(11) 3491

40 I- O-0 PRE . -3 POST

30 I- :: 30- I- t zo- s - l- 2 IO- k 20 U . : LL - O 1 : . TRIAL T2-20 10I- :. . . Figure 9. Resultsfrom an individual animal (A3) illustratehow the : data wereanalyzed. The 2 learning curves on the left showwithin- . problemlearning in the last 25 problemslearned preoperatively and the last25 learned postoperatively by this animal.On the right, the bars - tl I- . labeledPre and Post showthe averageerror rate in T2-20 of the 2 . A MD G MD/A WA learningcurves on the left. The bar labeledD showsthe differencebe- l-d tweenthe Pre and Postbars. For eachindividual in the experiment, Figure 10. sThe postoperative deficit (D in Fig. 9) is shownfor the 3 this quantity (D) wastaken as a measureof the animal’spostoperative groupswith bilaterallesions and the 2 groupswith disconnectionsur- deficitto heused in comparisonsbetween the differentoperated groups. gery. The bars showthe groupaverages and the dots showthe scoresof individual animals.

Murray’s (1990) analysis,as well as with Gaffan and Harrison’s (1987). erally symmetrical lesions,on the other hand, had a single sur- Like bilateral amygdalectomy, bilateral posterior thalamic le- gical operation and postoperative test. The possibility should sions(group MD) produced a severeimpairment in the present be considered,therefore, that this difference in the scheduleof task. Our resultsfrom unilateral posteriorthalamic lesions,which operations and behavioral tests is the causeof the relatively had no consistenteffect (Table 2) allow the effectsof the surgery mild impairment in the disconnection groups as compared to in the MD group to be attributed with confidence to their bi- the groups with bilaterally symmetrical lesions.Several results lateral lesion in the MDmc. The unilateral lesionsproduced as from other comparable studies suggestthat 2-stage surgery is much damageas the bilateral lesionsin structures outside the not sufficient to explain the mild effect of the disconnectionsin mediodorsalnucleus, including the midline thalamic nuclei and the present experiment. In an unpublished experiment still in the corpus callosum. progress,we have examined the effectsof bilaterally symmetrical Bilateral ventromedial prefrontal lesions(group VF) also had ablations in the amygdala, made in 2 stagesof surgery with a severe effect upon the present task, comparable to that of behavioral testing after each stage, on the sametask as in the bilateral amygdalectomy or bilateral MD lesions. Taken to- present experiment. Results from the first animal to complete gether, therefore, our results from the 3 groups with bilaterally this procedure show a deficit score (calculated as for the data symmetrical ablations are consistentboth with recent analyses presentedin Fig. 10)of 30 points. This preliminary result strong- of amygdalar function and with the proposition that the amyg- ly suggeststhat the effect of a bilaterally symmetrical lesionmade dala, mediodorsai nucleus,and ventromedial prefrontal cortex in 2 stagesis no lesssevere than that of a bilaterally symmetrical are functionally related to eachother. However, the resultsfrom lesion made in 1 stage. Furthermore, in experiments discon- the disconnectiongroups appear to contradict the idea (outlined necting the amygdala from the sensoryassociation cortex, very in the introductory remarks)that these3 structuresform a tightly severe effects of disconnection have been observed following linked functional pathway. Both of the disconnection groups 2-stagelesions (Gaffan and Harrison 1987; E. A. Gaffan et al., had only mild impairments, significantly lesssevere than those 1988). E. A. Gaffan et al., studying a task similar to the present groupswith bilaterally symmetrical lesions. task, examined learning during trials 1-14 of a seriesof visual- It is unlikely that the mild effectsin the disconnection groups discrimination problems. Choice accuracy during trials 2-14 can be explained by unintended sparing of tissue in their ab- showed an average error rate of 11.7% before disconnection lations. For example, animal A3 (bilateral amygdalectomy; see surgery and an average of 33.1% after disconnection surgery Fig. 2) had a lesscomplete amygdalectomy in eachhemisphere (theseare averagescomputed from the data shown in Fig. 2 of than the unilateral amygdalectomy in animal VF/Al (discon- E. A. Gaffan et al., 1988). Therefore, the deficit score in that nection of amygdala from ventromedial prefrontal cortex; see experiment, computed in a manner similar to that of the deficit Fig. 8) but animal A3 had a severe impairment similar to that scoresin the present experiment (Fig. lo), was,on average,2 1.4; of the others with bilateral amygdalectomy, while animal VF/ (the difference between 33.1 and 11.7). That deficit score is Al was virtually unimpaired (seeTables 1, 2). Similarly, the similar to those obtained from bilaterally symmetrical lesions unilateral lesionsof the mediodorsal nucleusshowed the same in the present study. If the amygdala was as intimately linked range in lateral and anteroposterior extent as did the bilateral with the posterior thalamus and prefrontal cortex as it is with lesions(Figs. 5-7). sensory-associationcortex, one should expect a similar severity The groupswith disconnection lesionshad 2 stagesof surgery, of impairment from the disconnection groups in the present with behavioral testing after each stage.The groups with bilat- experiment. 3492 Gaffan and Murray * Amygdala-MD-Frontal: Is It a System?

It is also unlikely that the mild effects in the disconnection reward associative learning cannot be wholly explained by the groups can be explained by interhemispheric transfer of infor- fact that that lesion interrupts the flow of information from the mation. For example, it could be suggested that an intact amyg- amygdala to the ventromedial prefrontal cortex. The output of dala in the left hemisphere could transfer information to an information from the amygdala in this task must therefore be intact ventromedial prefrontal cortex in the right hemisphere partially directed along a pathway independent of the ventro- by some route that would remain intact even after the left ven- medial prefrontal cortex. This pattern of results shows that a tromedial prefrontal cortex and the right amygdala had been mnemonic function such as stimulus-reward associative leam- removed. One such possible route would be via interhemispher- ing is not necessarily subserved by a single pathway of infor- ic heterotopic cortico-cortical connections, but this possibility mation flow through the serial, tightly linked stages of a single is ruled out by our finding that the effects of disconnection were memory system, but can instead be subserved by multiple, par- not exacerbated by forebrain commissurotomy. Another pos- tially independent, and partially overlapping pathways. sible route would be by interhemispheric subcortical connec- tions. Although there is some evidence that the amygdala pro- Appendix jects into the contralateral as well as the ipsilateral thalamus AD Anterodorsal nucleus (Aggleton and Mishkin, 1984), the contralateral projection is AM Anteromedial nucleus slight and may well pass through the area that was ablated in AV Anteroventral nucleus our unilateral thalamic lesions. Therefore, this route also ap- Ca Cauclatenucleus CeM Central medial nucleus pears an unlikely explanation of our results. The anatomical CL Central lateral nucleus evidence reviewed in the introductory remarks, supporting the CM Center median nucleus idea of a tightly linked functional pathway from the amygdala CSL Central superior lateral nucleus to the ventromedial prefrontal cortex via the thalamus, indicates Fx Fomix a very heavy predominance of ipsilateral connections at every Hl Lateral habenular nucleus Hm Medial habenular nucleus stage in the pathway. Furthermore, the powerful effects of dis- Hpt Habenulopeduncular tract connecting the visual-association cortex from the amygdala, LD Lateral dorsal nucleus strongly contrasting with the results from the present discon- Li Limitans nucleus nection groups, were obtained without any commissurotomy LP Lateral posterior nucleus MD Mediodorsal nucleus (E. A. Gaffan et al., 1988). MDdc Mediodorsal nucleus, densocellular portion For all these reasons, we conclude that the mild effects of MDmc Mediodorsal nucleus, magnocellular portion disconnection surgery in the present groups disconfirm the sim- MDmf Mediodorsal nucleus, multiformis portion ple hypothesis we outlined in the introductory remarks. This MDpc Mediodorsal nucleus, parvocellular portion was the hypothesis that memory information retrieved by the MG Medial geniculate complex Mm Mamillary complex interaction of the visual cortex with the amygdala is passed on Mtt Mamillothalamic tract from the amygdala to the prefrontal cortex via the thalamus in Pa Paraventxicular nucleus a serially organized functional pathway. One alternative pos- PC Paracentral nucleus sibility to be investigated in future experiments is that this type Pf Parafascicular nucleus Pla Anterior pulvinar nucleus of memory information is passed on directly from the amygdala Pli Inferior pulvinar nucleus not only to the structures we have considered here, but also Pll Lateral pulvinar nucleus independently to some other area, such as the ventral . PO Posterior nucleus Another possibility is that a severe impairment would be pro- Pt Parataenial nucleus duced by doubly disconnecting the amygdala both from the R Reticular nucleus Re Reuniens nucleus mediodorsal thalamus and from the ventromedial prefrontal Rhomboid nucleus cortex by ablating the amygdala in one hemisphere and both L% Suprageniculate nucleus the mediodorsal thalamus and the ventromedial prefrontal cor- Sm Stria medullaris tex in the other hemisphere. However, even if this double dis- SN Substantia nigra St Stiia terminalis connection were to produce an impairment equal in severity to Sub that of the animals with bilateral symmetrical ablations, that VA result would not support the hypothesis of a single serial path- VLa Ventral lateral nucleus, anterior portion way. Rather, it would suggest that 2 parallel pathways from the VLP Ventral lateral nucleus, posterior portion amygdala, to the mediodorsal thalamus and to the ventromedial VMb Ventral medial nucleus, basal portion VMP Ventral medial nucleus, principal portion prefrontal cortex, were partially independent of each other. VP1 Ventral posterior inferior nucleus From the data presently available, it can be concluded that VPL Ventral posterior lateral nucleus the functional relationships between the amygdala, thalamus, VPM Ventral posterior medial nucleus and prefrontal cortex are of a more diffuse nature than that Zi envisaged in the hypothesis of a single serial pathway. We have seen that the effects of bilateral mediodorsal thalamic lesions References or of bilateral ventromedial prefrontal lesions upon stimulus- Aggleton JP, Mishkin M (1983) Visual recognition impairment reward associative learning cannot be wholly explained by the following medial thalamic lesions in monkeys. Neuropsychologia 2 1: fact that those lesions interrupt the pathway to those structures 189-197. from the amygdala. These structures must therefore participate Aggleton JP, Mishlcin M (1984) Projections of the amygdala to the thalamus in the cynomolgus monkey. J Comp Neurol 22256-68. in functions that contribute to the efficiency of stimulus-reward Amaral DG, Price JL (1984) Amygdalo-cortical projections in the associative learning but are independent of input from the amyg- monkey (Mmzca ficiculuris). J CompNeural 230:465496. dala. Equally, the effect of bilateral amygdalectomy in stimulus- AsanumaC, ThachWT, JonesEG (1983) Cytoarchitectonicdelinea- The Journal of Neuroscience, November 1990, IO(11) 3493

tion of the ventral lateral thalamic region in the monkey. Brain Res Giguere M, Goldman-Rakic PS (1988) Mediodorsal nucleus: areal, Rev 5:219-235. laminar and tangential distribution of afferents and efferents in the Bachevalier J, Mishkin M (1986) Visual recognition impairment fol- frontal lobe of rhesus monkeys. J Comp Neuro1277: 195-2 13. lows ventromedial but not dorsolateral prefrontal lesions in monkeys. Goldman-Rakic PS, Porrino LJ (1985) The primate mediodorsal (MD) Behav Brain Res 20~249-26 1. nucleus and its projection to the frontal lobe. J Comp Neurol 242: Berman AL, Jones EC (1982) The thalamus and basal telencephalon 535-560. of the cat. Madison: University of Wisconsin, Gower EC (1989) Efferent projections from limbic cortex of the tem- Burton H, Jones EG (1976) The posterior thalamic region and its poral pole to the magnocellular medial dorsal nucleus in the rhesus cortical projection in New World and Old World monkeys. J Comp monkey. J Comp Neurol280:343-358. Neurol 168249-302. Jones EG (1985) The thalamus. New York: Plenum. Gaffan D, Harrison S (1987) Amygdalectomy and disconnection in Murray EA (1990) Representational memory in nonhuman primates. visual learning for auditory secondary reinforcement by monkeys. J In: Neurobiology of comparative cognition (Kesner RT, Olton, DS, Neurosci 7:2285-2292. eds), pp 127-155. Hillsdale, NJ: Erlbaum. Gaffan D, Saunders RC, Gaffan EA, Harrison S, Shields C, Owen MJ Murray EA, Mishkin M ( 1985) Amygdalectomy impairs crossmodal ( 1984) Effects of fomix transection upon associative memory in mon- association in monkeys. Science 228:604-606. keys: role of the hippocampus in learned action. Q J Exp Psycho1 Olszewski J (1952) The thalamus of the Macaca mulutta. Basel: S 36B:173-221. Karger. Gaffan D, Gaffan EA, Harrison S (1989) Visual-visual associative Porrino LJ, Crane AM, Goldman-Rakic PS (198 1) Direct and indirect memory and reward-association learning in monkeys: the role of the pathways from the amygdala to the frontal lobe in rhesus monkey. J amygdala. J Neurosci 9:558-564. Comp Neurol 198:121-136. Gaffan, EA, Gaffan D, Harrison S (1988) Disconnection of the amyg- Russchen FI, Amaral DG, Price JL (1987) The afferent input to the dala from visual association cortex impairs visual reward-association magnocellular division of the mediodorsal thalamic nucleus in the learning in monkeys. J Neurosci 8:3 144-3 150. monkey, Mucucu fusciculuris. J Comp Neurol 256: 175-2 10. Gallyas F (1979) Silver staining of myelin by means of physical de- Tobias TJ (1975) Afferents to prefrontal cortex from the thalamic velopment. Neurological Res 1:203-209. mediodorsal nucleus in the rhesus monkey. Brain Res 83: 19 l-2 12.