(p.220) population vectors calculated by a population of DLPFC activities at the 180° trial of the R-ODR task, in which the visual cue was presented at the 180° direction and the direction of the correct saccade was the 90° direction. Population vectors were directed toward the 180° direction at the beginning of the delay period. However, population vectors began to rotate in the middle of the delay period, continued to rotate slowly from the 180° direction to the 90° direction during the late half of the delay period, and finally directed toward the 90° direction at the response period (see Figure 13.2d). This result indicates that the information represented by a population of DLPFC activities changes during the delay period. Fuster (1997) had proposed that the mediation of cross-temporal contingency is an important function of the DLPFC. He stated that the delay period is the period for the cross-temporal bridging of sensory-to-motor information transformation, which is a dynamic process for internal transfer of information as well as a process of cross-temporal matching. The present result supports the notion that the DLPFC plays a significant role for mediating the cross-temporal contingency. This result also suggests that delay-period activity observed in the DLPFC contributes significantly to the dynamic process related to internal information transfer.

Interactions among prefrontal neurons

Takeda and Funahashi (2006) showed the presence of the information flow from neurons having cue-period activity to neurons having oculomotor activity through neurons having delay-period activity in the DLPFC. While monkeys performed ODR tasks, cue- (C), delay- (D), or response-period (R) activities or their combinations (CD, CR, DR, CDR) were observed in DLPFC neurons. They characterized each neuron based on which task-related activity was exhibited and which information (cue location or saccade direction) each task-related activity represented, then classified these neurons into nine groups (C, Dcue, Dsac, CDcue, DcueRcue, DsacRsac, DcueRsac, CDcueRcue and CDcueRsac). Preferred directions were similar between cue- and delay-period activities in groups of CDcue, CDcueRcue and CDcueRsac, all of which represented visual information, indicating that the directional selectivity of delay-period activity was affected by the directional selectivity of cue-period activity for each neuron. Preferred directions were also similar between delay- and response-period activities in groups of DcueRcue, CDcueRcue and DsacRsac, indicating that the directional selectivity of delay-period activity affects the directional selectivity of response-period activity in these neurons. By the comparison of temporal profiles of delay-period activity among these groups, they found (1) cue-period activity of C, CDcue and CDcueRcue groups might contribute to the initiation and the maintenance of delay-period activity of CDcue, CDcueRcue, Dcue and DcueRcue groups, and (2) saccade-related activity of DsacRsac could be affected by delay-period activity of Dsac and DsacRsac groups. Thus, while monkeys performed the ODR tasks, the information flow from neurons having cue-period activity to neurons having oculomotor activity through neurons having delay-period activity is present in the DLPFC. Information processing could be performed during this information flow process.

Dynamic and flexible changes in functional interaction among temporary information-storage processes would be a key component to understand neuronal mechanisms of information processing. In the DLPFC, Wilson et al. (1994) showed that types of responses (excitatory or inhibitory) of pyramidal and nonpyramidal neurons were often inverted (e.g., pyramidal neurons exhibited excitatory responses while nonpyramidal neurons exhibited inhibitory responses) and that the timing of excitatory and inhibitory responses appear to be phased. This result suggests the presence of functional interactions between pyramidal and nonpyramidal neurons. Rao et al. (1999) used a cross-correlation analysis to examine functional interactions between pyramidal and nonpyramidal neurons in the DLPFC and found inhibitory interactions between pyramidal (p.221) neurons and adjacent nonpyramidal interneurons. Funahashi and Inoue (2000) also applied a cross-correlation analysis to simultaneously isolated pairs of prefrontal activities during ODR performances and examined functional interactions between task-related DLPFC neurons. When both neurons of examined pairs exhibited delay-period activity, these neurons tended to show excitatory interactions and to show similar directional preferences. Interactions between task-related neurons with different directional preferences increased with the progress of the trial. Excitatory interactions among neurons having delay-period activity were frequently observed in the DLPFC, suggesting that the information stored in temporary information-storage processes may be transformed from one form of representation to another or integrated to represent more complex information (e.g., delay-period activity representing a combination of visual cue positions and the temporal order of their presentation) by these interactions.

Each neuron changes the magnitude of activity depending on the trial conditions (e.g., difference in the position where the visual cue was presented or the direction where the saccade was directed), the temporal context of the trial and trial events. Therefore, the strength of functional interactions could be changed depending on the trial conditions or the temporal context of the trial. In fact, the height of the peak of the cross-correlogram changed depending on the cue conditions. This modulation of the peak height in the cross-correlogram was observed in most of the neuron pairs examined in the DLPFC. These results suggest that, although many DLPFC neurons have functional interaction with neurons exhibiting same or different kinds of task-related activity, the strength of the interaction may change dynamically depending on the conditions of the trials. Thus, dynamic and flexible changes of functional interactions among temporary information-storage processes could be an important element in understanding neural mechanisms of information processing.

Two working memory systems

A process-based model of working memory in the DLPFC was explained in the previous section. However, a similar system is also necessary in the brain areas where sensory or motor information are processed. Temporary maintenance of information and processing information could be ubiquitous processes for any kind of information processing in the brain. In fact, tonic sustained delay-period activity has been observed not only in the DLPFC, but also in the frontal eye field (Funahashi et al. 1989; Lawrence et al. 2005; Roesch and Olson 2005), the inferior temporal cortex (Fuster 1990; Fuster and Jervey 1982; Miller et al. 1993; Miyashita and Chang 1988; Naya et al. 1996), the posterior parietal cortex (Chafee and Goldman-Rakic 1998; Murata et al. 1996; Pesaran et al. 2002; Sereno and Amador 2006; Zhang and Barash 2004), and the premotor cortex (Crammond and Kalaska 2000; Kurata and Wise 1988; Obayashi et al. 2003; Weinrich and Wise 1982;). Thus, the process-based model of working memory that we proposed is not just for the model to explain functions of the DLPFC but may be a ubiquitous system in the brain.

The specific feature of working memory system operating in the DLPFC would be that this system is not for information processing of specific purpose, such as only for visual information processing, only for control of limb movements, or only for speech comprehension, but rather is general purpose and is an essential element for a variety of information processing that the DLPFC participates in. The DLPFC is located in the best position to monitor the functional state of the posterior association cortices and to control their functions by sending command signals (or top-down signals), since the DLPFC has heavy anatomical connections, mostly reciprocal connections, with the posterior association cortices. On the other hand, working memory systems must be present in the modality-specific cortical areas, such as the sensory system, as well as the motor system. In these modality-specific cortical areas, temporary maintenance and processing (p.222) of information are also essential processes. Therefore, working memory systems can be divided into two types: a general-purpose working memory system and a modality-specific working memory system.

A modality-specific working memory system is a mechanism for temporary maintenance and processing of certain types of information, such as sensory, linguistic, or motor information. Modality-specific working memory systems will be present in cortical and subcortical areas which participate mostly in sensory and motor information processing. On the other hand, a general-purpose working memory system is a mechanism to maintain or process a variety of information including the behavioral context to perform various cognitive tasks or control signals (e.g., top-down signals) to monitor or regulate operations of other cortical and subcortical areas. The contents of a general-purpose working memory system would affect or regulate the contents of modality-specific working memory systems and lead information processing operated in modality-specific working memory system to necessary directions.

Cohen and others (Cohen and Servan-Schreiber 1992; Cohen et al. 1996; Miller and Cohen 2001) explained prefrontal executive functions by computational architectures with the connectionist framework. They hypothesized that the DLPFC represents the context information, which corresponds to the goal representation in the computational architecture of the production system. Context information can be defined as the information necessary to mediate an appropriate behavioral response, and includes a set of task instructions, a specific prior stimulus, or the result of processing a sequence of prior stimuli (Cohen et al. 1996). To accomplish a particular goal, some brain region needs to maintain an internal representation of the goal, to suppress unnecessary behaviors and to temporally coordinate series of behaviors. Cohen et al. (1996) considered the DLPFC as this cortical region. In this regard, prefrontal executive functions that Cohen and others proposed are similar to the functions of the general-purpose working memory system.

Interactions between the DLPFC and other cortical areas

To understand how the DLPFC acts as the general-purpose working memory system, it is important to consider dynamic and flexible functional interactions between the DLPFC and other cortical and subcortical areas. Corticocortical and corticosubcortical anatomical connections with the DLPFC have been examined extensively – see reviews by Fuster (1997), Petrides (1994), Petrides and Pandya (2002). The DLPFC has reciprocal connections with the posterior parietal cortex, the inferior temporal cortex, the superior temporal polysensory areas, the anterior cingulate, the retro-splenial cortex and the parahippocampal gyrus. The DLPFC also has strong reciprocal connections with the mediodorsal nucleus of the thalamus (Ray and Price 1993; Siwek and Pandya 1991). In addition, the DLPFC has connections with the frontal eye field, the presupplementary motor area, the premotor cortex, and the caudate nucleus (Miyachi et al. 2005). Morris et al. (1999) reported that the mid-dorsolateral prefrontal cortex (areas 46 and 9) projects to the retrosplenial area 30 and the posterior presubiculum, and suggested that this system could be the anatomical substrate of functional interaction between the DLPFC and the hippocampus. Thus, the DLPFC has anatomical connections with various cortical and subcortical areas, especially sensory and motor association areas and the limbic brain areas. Most of these connections are reciprocal. These observations indicate that the DLPFC can receive various kinds of information including sensory, motor, and emotional information as well as information stored in long-term memory, and is located in a good position to monitor both internal and external environment. These observations also indicate that the DLPFC can send various kinds of information and signals to various brain areas, and is again located in a good position to coordinate information processing operated in various brain regions and direct these operations to a particular goal. Strong reciprocal (p.223) connections between the DLPFC and various brain areas are the anatomical basis of dynamic and flexible interactions between the general-purpose working memory system and the modality-specific working memory systems.

Noninvasive brain imaging studies have also revealed dynamic and flexible functional interactions between the DLPFC and other cortical areas. n-back tasks have often been used as working memory tasks. Bilateral DLPFC has been shown to be activated in visuospatial n-back tasks (Cohen et al. 1997; Smith and Jonides 1997, 1998). The magnitude of the activation in the DLPFC depended on the memory load required to perform the task (Callicott et al. 1999; Carlson et al. 1998). However, working memory tasks like n-back tasks are composed of numerous elementary neural processes. Based on the analysis using evoked potentials, McEvoy et al. (1998) showed that, although some of these neural processes were affected by the type of information (verbal or spatial), the neural processes associated with frontal and parietal slow waves were affected not by the type of information, but by the amount of information being retained. McEvoy et al. (1998) suggested that working memory is operating by dynamic cortical networks of information-specific processes (e.g., modality-specific working memory system) and nonspecific, capacity-limited, higher-order processes (e.g., general-purpose working memory system).

Dynamic cortical networks including the DLPFC have been shown directly by correlation analyses in functional brain imaging studies as well as by covariance and correlation analyses of event-related potentials. Using a working memory task of faces, McIntosh et al. (1996) examined functional interactions among different cortical regions by an anatomy-based covariance structural equation model. As a result, they found that different functional interactions among different cortical areas were observed in different task conditions (e.g., different delay lengths) and that a top-down interaction from the DLPFC to the temporal and occipital cortices was observed in a long delay condition. McIntosh et al. (1999) also showed that only aware subjects exhibited learning-related modulations of brain activation in the left DLPFC during an associative learning task and that a correlation analysis of this modulation revealed significant functional connectivity among bilateral DLPFC, the sensory association cortices and the cerebellum. In addition, evidence for homotopic interaction or coactivation of the bilateral DLPFC has been reported (Klingberg et al. 1997; McGuire et al. 1991). Klingberg et al. (1997) reported the increase of bilateral coactivation of the DLPFC and the inferior parietal cortex in association with the increase of task difficulty. They also found that the inferior parietal cortex was often coactivated with the DLPFC in working memory tasks and tasks involving planning. Further, Huettel et al. (2005) showed that the degree of uncertainty on decision making influenced the magnitude of activation of both prefrontal and parietal cortices. Collette et al. (2005) showed activation of both the prefrontal and the inferior parietal cortex in dual-task performance. Thus, all of these results indicate the importance of functional interactions as well as coactivation between the DLPFC and other cortical and subcortical areas when subjects perform cognitive tasks. These results also indicate that functional interactions and coactivation between the DLPFC and other cortical and subcortical areas are dynamically and flexibly modulated depending on the context of the cognitive task.

Evidence of top-down modulatory effects by the DLPFC have also been shown. Using visual associative memory tasks, Hasegawa et al. (1998) showed that retrieval of visual images from long-term memory was under the executive control of the DLPFC by an experiment using partial split-brain monkeys. In addition, by neurophysiological experiments using monkeys having resection of the corpus callosum, Tomita et al. (1999) showed that top-down signals from the DLPFC regulated retrieval of associative memory in inferior temporal neurons. These observations indicate that the DLPFC can control information processing in other cortical areas by sending top-down signals (Miyashita 2005). Lumer and Rees (1999) showed by fMRI that covariation of activation was observed in multiple extrastriate areas, the parietal cortex, and the prefrontal (p.224) cortex in bistable viewing conditions. Coordinated activation of these areas was not linked to either sensory or motor events, but rather reflected perceptual events or internal changes of perception. Therefore, they concluded that functional interactions between visual areas and the prefrontal cortex are important for conscious vision. On the other hand, either enhancement or suppression of the visual response was observed in inferior temporal neurons when two visual stimuli (one preferred and the other nonpreferred) were presented simultaneously in the neuron’s receptive field. Chelazzi et al. (1998) proposed that this phenomenon can be explained by a ‘biased competition’ model of attention. Based on this model, simultaneously presented objects in the visual receptive field would compete with others for representation within the cortex. They suggested that this competition could be biased by ‘top-down’ feedback signals from brain structures involved in working memory. Based on the results by Hasegawa et al. (1998), Tomita et al. (1999), and Lumer and Rees (1999), it would be reasonable to think that top-down signals from the DLPFC bias the magnitude of the activity in the visual system. These results indicate the presence and importance of top-down modulatory effects by the DLPFC to perform cognitive tasks. These results also suggest that top-down modulatory effects by the DLPFC affect a wide variety of cognitive activities including sensory perception, directed attention, episodic memory encoding and retrieval through dynamic and flexible functional interactions. This could be the typical function of the DLPFC as the general-purpose working memory system.

Conclusions

In this manuscript, we proposed a general-purpose working memory system to explain how the DLPFC performs executive control. This general-purpose working memory system can (1) represent and process various kinds of information, (2) access various systems (e.g., modality-specific working memory systems) to obtain any kind of necessary information and monitor their operations, and (3) send appropriate information and signals (e.g., top-down signals) to various systems to control their operations. To represent and process various kinds of information in the general-purpose working memory system, this system needs to integrate following four processes: a selection process, a temporary information-storage process, an output process and modulatory signals. We also proposed the presence of extensive functional interactions among temporary information-storage processes for information processing. The dynamic and flexible nature of these interactions is an essential feature of the general-purpose working memory system. In addition, dynamic and flexible interactions between the general-purpose working memory system and the modality-specific working memory systems are also essential to consider the function of the general-purpose working memory system as well as the functions of the DLPFC. However, it is not known how the selection process selects necessary information from a variety of information in a variety of brain areas, how the DLPFC creates appropriate top-down signals, how the DLPFC sends appropriate information and top-down signals to the exact systems or exact brain areas to control their operations, and how the DLPFC can manage a variety of functions efficiently. Therefore, we need further experiments to solve these questions and to understand functions of the DLPFC as the general-purpose working memory system.

Stay updated, free articles. Join our Telegram channel

Join