Callosal Window Between Prefrontal Cortices: Cognitive Interaction to Retrieve Long-Term Memory

See allHide authors and affiliations

Science  07 Aug 1998:
Vol. 281, Issue 5378, pp. 814-818
DOI: 10.1126/science.281.5378.814


A perceptual image can be recalled from memory without sensory stimulation. However, the neural origin of memory retrieval remains unsettled. To examine whether memory retrieval can be regulated by top-down processes originating from the prefrontal cortex, a visual associative memory task was introduced into the partial split-brain paradigm in monkeys. Long-term memory acquired through stimulus-stimulus association did not transfer via the anterior corpus callosum, a key part interconnecting prefrontal cortices. Nonetheless, when a visual cue was presented to one hemisphere, the anterior callosum could instruct the other hemisphere to retrieve the correct stimulus specified by the cue. Thus, although visual long-term memory is stored in the temporal cortex, memory retrieval is under the executive control of the prefrontal cortex.

The primate inferior temporal cortex, located at the final processing stage of visual object perception (1), plays an important role in recall as well as storage of visual memory; inferotemporal neurons can be dynamically activated by retrieval of visual long-term memory in monkeys (2), and electric stimulation of this region results in imagery recall in humans (3). The neural network that enables such imagery recall in cognition has not been established. A likely component is the prefrontal cortex, which has been implicated in executive processes such as planning, working memory, and memory retrieval (4, 5). A conventional approach by means of lesion to the prefrontal cortex often produces devastating cognitive impairments (4). On the other hand, the capacity for interhemispheric transfer through the anterior corpus callosum (CC), the callosal window between prefrontal cortices (6, 7), would positively highlight executive processes undertaken by the prefrontal cortex. So far, there has been little evidence for what transfers via the anterior CC (8, 9), whereas it has been established that posterior callosal fibers between sensory cortical areas (7) provide channels for communication in each modality (6, 8, 10, 11). In a clinical report of an epileptic patient who had undergone selective posterior callosotomy (12), although sensory stimuli lateralized to the nondominant right hemisphere could not be transferred for naming, semantic features of these stimuli somehow could be described by the expressive language system of the left hemisphere. This observation leads to a hypothesis that top-down processes originating from the prefrontal cortex can regulate retrieval of long-term memory from the modality-specific posterior association cortex, even in the absence of direct sensory input. To test this hypothesis, we examined in partial split-brain monkeys whether the prefrontal cortex can instruct, through the anterior CC, the contralateral hemisphere to retrieve long-term memory when sensory interaction between posterior cortical areas is prevented (Fig. 1A).

Figure 1

(A) Ventral view of a monkey brain illustrating the experimental design. The posterior CC and AC (both colored red), which interconnect occipito-temporal visual areas, were selectively transected (hatched), while the anterior CC (blue) between prefrontal cortices was left intact. This preparation allowed dissociation of mnemonic processes undertaken by prefrontal and posterior association cortices. A, anterior; P, posterior. (B) Schedule of two-stage commissurotomy in operated animals drawn to time scale. At the first operation, AC and the posterior CC were transected. At the second operation, the remaining anterior CC was transected. Monkeys were behaviorally tested in two stages: posterior-split stage [hatched, also shown in (A)] and full-split stage (filled).

Monkeys underwent two-stage scheduled commissurotomy (Fig. 1B) (13). In the first operation, we transected occipito-temporal visual commissural fibers (7)—the splenium (SP) of the CC and the anterior commissure (AC). At this posterior-split stage, we could evaluate communication between prefrontal cortices via the anterior CC. After behavioral testing, the second operation was performed to sever the anterior CC for full-split control experiments (see below). The extent of callosal lesions for individual animals, based on magnetic resonance imaging (MRI) and histological data (14, 15), is summarized (Fig. 2A). MRI after the first operation showed that, as intended, SP and AC were split and the anterior CC was left intact in all the posterior-split monkeys (Fig. 2B). Histological data obtained after the second operation ensured that the anterior CC as well as SP and AC were divided at the full-split stage (Fig. 2C). In sections stained for myelin, commissural lesions were evident and the residual transected fibers looked atrophic and demyelinated (Fig. 2D). Slight unintended lesions were found in the right anterior cingulate gyrus and area preoptica medialis. The fornix was bilaterally intact. Interhemispheric cortico-cortical connections were further analyzed with retrograde fluorescent tracers (15). After diamidino yellow (DY) was injected in anteroventral portions of unilateral inferotemporal cortex, no labeled neurons were detected in the contralateral homotopic areas in the posterior-split preparation (Fig. 2E), in marked contrast to the unoperated control (Fig. 2F). However, injections of fast blue (FB) in the lateral prefrontal areas at the posterior-split stage produced extensive contralateral labeling in the supragranular layers (Fig. 2G), which was absent after the full-split surgery (Fig. 2H). These results confirmed that commissural fibers between prefrontal cortices were connected and alive in posterior-split monkeys, whereas those between visual cortical areas were selectively disrupted.

Figure 2

(A) Extent of the first (hatched) and second (filled) commissurotomy in each operated animal. One posterior-split monkey (P1) did not undergo the second surgery and was used for tracer experiments (E and G). Three monkeys (PF1 to PF3) underwent two-stage commissurotomy (13). Numbers indicate coronal slice levels in (B) and (C). D, dorsal; V, ventral; A, anterior; P, posterior. Scale bar = 10 mm. (B) Representative MRIs from monkey PF3 during the interoperative period. Lesions of CC (arrowheads) and AC (arrow) are marked. Scale bar = 10 mm. (C) Line drawings of coronal sections from PF3 after perfusion (15). Scale bar = 10 mm. (D) Fiber-stained coronal sections, showing enlarged areas enclosed by red rectangles in (C). Scale bar = 2 mm. (E toH) Yellow (DY) and blue (FB) arrowheads in upper diagrams show injection sites of tracers in the left hemisphere (15). Small red squares (arrows) indicate the loci of dark-field fluorescent photomicrographs enlarged below. In the posterior-split monkey (P1), retrogradely labeled cells were not observed in the right inferotemporal areas (E) but were abundant in the right prefrontal areas (G). In the full-split monkey (PF1), prefrontal labeling was absent (H). Upper diagram, scale bars = 10 mm; lower photomicrograph, scale bar = 50 μm.

Two behavioral experiments were carried out. In the first experiment, we found that visual stimulus-stimulus association learning (16) did not transfer in posterior-split monkeys. In this task, the monkeys were required to memorize associations between arbitrarily assigned cue and choice pictures. On each trial, after a sequential presentation of cue and choice stimuli during fixation, the subject must select one of the choices specified by the cue with saccade (Fig. 3, A and B). In the intrahemispheric (INTRA) condition, the information necessary to recall the visual stimulus-stimulus association was lateralized to a single cerebral hemisphere (17). The monkeys were trained to reach criterion in one hemisphere and then tested in the opposite hemisphere until the criterion was reachieved (18). In the unoperated group, the experience of original learning in the first hemisphere facilitated relearning in the second hemisphere (Fig. 3C, left). However, there was no apparent learning improvement in the posterior-split group (Fig. 3C, right). Analysis of variance (ANOVA) revealed a significant interaction between monkey group and learning effect (F (1, 5) = 19.71; P < 0.007 ). In the unoperated controls, significantly fewer trials were needed in the second than in the first hemisphere [t = 4.62; degrees of freedom (df) = 2; P < 0.05, two-tailed t test], whereas in the posterior-split animals there was no significant difference (t = 0.67; df = 3; P > 0.5). As an index to estimate the effect of learning transfer, a percentage saving score was calculated for each stimulus set (19). The average saving scores for the unoperated controls were significantly above zero (t = 12.99; df = 2; P < 0.006), showing nearly perfect transfer (Fig. 3D). On the other hand, those for the posterior-split animals were not significantly different from zero (t = 0.27; df = 3; P > 0.8). Thus, long-term memory of visual stimulus-stimulus association learned in one hemisphere did not transfer to the other hemisphere via the anterior CC.

Figure 3

(A) Visual stimulus-stimulus association task for the INTRA conditions. One cue and subsequently two choice stimuli were presented to the same visual hemifield while the monkey maintained fixation. The animal must saccade to one of the choices instructed by the cue. Because the animal continued to fixate during the stimulus presentation and then saccaded straightforward to one of the targets (B), all the information necessary to recall the visual stimulus-stimulus association was exclusively lateralized to a single hemisphere in this condition (17). The monkey was trained to reach criterion in one hemisphere. When the performance reached criterion in the first hemisphere, then the identical stimulus set was tested in the second hemisphere for transfer of learning (18). (B) Horizontal (H) and vertical (V) eye positions of a posterior-split monkey aligned at the onset (“Fix” in upper traces) and offset (“Fix Off” in lower traces) of the fixation spot. Data sampling rate is 250 Hz. (C) Number of trials to criterion for the first and second hemisphere in normal (open bars, open symbols) and posterior-split (hatched bars, filled symbols) monkeys. Each symbol represents data from an individual animal, the average score for four stimulus sets. (D) Average saving scores for normal and posterior-split monkeys. Symbols are as in (C).

In the second experiment, an interhemispheric (INTER) version of a visual stimulus-stimulus association task (Fig. 4A) was introduced. In the INTER condition, choice stimuli were presented to the opposite side of the cue (16). Because the monkeys' fixation and saccade were just as accurate as in the INTRA condition (Fig. 4B), the cue and choice stimuli were received by separate cerebral hemispheres. The two hemispheres must then communicate with each other, moment to moment (20), to select the correct choice specified by the cue. Surprisingly, all of the posterior-split animals could successfully solve such an INTER task (21). The performance level attained by these monkeys was almost the same for the INTER and INTRA conditions (Fig. 4, C and D). To determine whether the INTER performance depended on cortical interaction through the anterior CC, the performance before and after the second full-split operation in the same animal was compared for each condition (21). There was a significant interaction between operative stage (posterior- or full-split) and hemispheric condition (INTER, left INTRA, right INTRA) (F (2, 10) = 52.79; P < 0.0001). In the INTER condition (Fig. 4C), performance after the full-split operation fell at chance and was significantly lower than at the posterior-split stage (t = 18.78; df = 2;P < 0.003). This ruled out the possibility that peripheral cuing strategy or subcortical commissural interaction (22) might account for the INTER performance. The drop in the INTER performance could not be attributable to surgical damage affecting general mnemonic ability, because the INTRA performance was not significantly altered after the second operation (left:t = 0.54; df = 2; P > 0.6; right:t = 0.72; df = 2; P > 0.5) (Fig. 4D). We conclude that the anterior CC is able to support cognitive interaction necessary to recall visual stimulus-stimulus association.

Figure 4

(A) Visual stimulus-stimulus association task for the INTER condition. The cue and choice stimuli were sequentially presented to separate hemispheres. (B) Eye trajectories during fixation and saccade for the INTER (left) and INTRA (right) conditions in a posterior-split monkey. The central square represents the 1° × 1° fixation window. Circles indicate locations where the choice stimuli were presented. H, horizontal; V, vertical. (C and D) Performance for the INTER (C) and INTRA (D) conditions at the posterior-split (hatched bars) and full-split (filled bars) stages. In posterior-split monkeys, the INTER performance was almost the same for the INTRA performance. After the full-split operation, the INTER performance fell at chance and the INTRA performance was not altered. Symbols represent data from individual animals.

The finding that visual long-term memory acquired through stimulus-stimulus association learning does not transfer between prefrontal cortices (Fig. 3, C and D) has two implications. First, this suggests that visual associative long-term memories are primarily stored in the inferotemporal cortex (2, 3, 23, 24) and that the prefrontal cortex cannot make up for this function. Therefore, deficits in visual associative learning observed after periarcuate lesions (25) would be ascribed not to loss of long-term memory but rather to dysfunction of executive processes. Second, consistent with previous behavioral, anatomical, and neurophysiological data (26), this result indicates that the visual image of cue or choice per se did not transfer in posterior-split monkeys. After transection of SP and AC, the monkeys were visually split. Nevertheless, cognitive unity could be maintained through the callosal window between prefrontal cortices. The nature of the signal carried by this callosal bundle should be characterized by further studies with the current paradigm.

The uncinate fascicle, bidirectional temporo-frontal cortical pathway is necessary for visual associative learning in monkeys (27). A distributed cortical network along this pathway, which also might be involved in object working memory (4,5), would regulate retrieval of visual long-term memory from the inferotemporal cortex (1–3, 23, 24) even in the absence of direct visual input. Consistent with this view, human functional neuroimaging studies have revealed that prefrontal areas are activated in various memory retrieval tasks (28). Thus, in primates, prefrontal and posterior association cortices should play essentially different roles in retrieval and storage of long-term memory.

  • * To whom correspondence should be addressed. E-mail: hasegawa{at}


View Abstract

Stay Connected to Science

Navigate This Article