Intersecting the divide between working memory and episodic memory


Figure 18.1 Illustration of implementation of the mixed blocked/event-related fMRI design.


exhibit sustained activation changes throughout an entire task. Item-related processes on the other hand implicate the cognitive operations that are triggered by the individual events within a task (e.g., stimulus presentation). As these processes are engaged in a stimulus-synchronous manner the brain activity they evoke should be transient in nature, and distinct from the prolonged changes in the neural responses that reflect state-related control processes and task-set maintenance.


Mixed fMRI designs combine standard event-related fMRI procedures that explicitly measure the temporally isolated and transient brain signals evoked by individual items, with epoch-related blocked procedures that index the average brain activity changes that occur throughout the duration of task performance as compared to the brain activity measured during control blocks. Importantly, mixed designs allow the simultaneous assessment of these temporally independent brain responses. Generally, mixed paradigms alternate task blocks of cognitive performance with control blocks comprising periods of some low-level baselines (e.g., resting) (see Figure 18.1). A critical feature of the mixed design is that the interstimulus-intervals (ISIs) that are interpersed between stimulus presentations must be of varying time duration or ‘jittered’ including some prolonged ISIs approximating 20 s, in order to keep the correlation between the state and item regressors reasonably low which is a prerequisite to separate the sustained and transient fMRI signals. To date several studies have successfully utilized mixed fMRI designs to measure sustained and transient brain activity separately (Donaldson et al2001; Braver et al2003).


Dividing lines and intersecting points: dissociable and common activation patterns for working memory and episodic memory


To elucidate the functional-anatomic relationship between working memory and episodic memory, we have conducted a mixed blocked/event-related fMRI experiment in which sustained and transient brain activity patterns were first separated within each of four tasks (working memory, (p.311)episodic memory, semantic memory and attention) and then compared across tasks (Marklund et al2007). By analyzing sustained state-related neural responses and transient stimulus-locked neural responses separately across tasks we were able to demonstrate patterns of regional activation commonalities that differed as to the temporal signature of the underlying neural modulation. In keeping with previous cross-function imaging studies that incorporated tasks of working memory and episodic memory for direct within-study comparisons (Braver et al2001; Cabeza et al2002; Nyberg et al2002; Ranganath et al2003), common activations were found in a distributed network of brain areas, with the most salient points of intersection occurring within the frontal lobes. However, prior studies investigating these two memory functions or ‘systems’ together have focused exclusively either on comparisons based on the neural activity triggered by distinct types of events (event-related designs), or comparisons based on the average relative change in neural activity evoked by the total number of trials (including the ISIs) within running blocks of continuous task performance (blocked designs). The neural mechanisms and different component processes evoked during performance of cognitive tasks are arguably represented throughout a spectrum of different time scales. Investigations using either a blocked or event-related design cannot provide measures of item-specific transient responses or temporally extended state responses, respectively, because the former confounds the two types of responses, while the latter selectively index transient changes. Mixed fMRI designs such as the one utilized here are unique because they allow researchers to dissociate and concurrently measure transient item-related responses induced on a trial by trial basis within tasks and item-independent sustained neural activity that occur throughout entire task blocks (Donaldson et al2001).


In the next sections we consider the implications of overlapping PFC activity between working memory and episodic memory with respect to underlying temporal profiles and the degree to which commonalities further overlap with the activation patterns associated with semantic memory and attention/vigilance, respectively. We start by considering the PFC regions that exhibited common sustained activity in working memory and episodic memory and tentatively propose hypotheses as to the candidate component processes that may be attributable to each of the implicated areas. The results of our mixed fMRI study will also be considered with respect to other relevant brain imaging studies, neuropsychological studies and electrophysiological findings.


Overlapping sustained neural activity: common control processes related to maintenance of an attentive state and task-set representation


Compared to a low-level resting baseline, sustained activity that was common for working memory and episodic memory was manifested within a network of primarily frontal and parietal cortices. The areas of commonality within PFC included bilateral ventrolateral PFC (BA 47/45), anterior PFC (i.e., frontopolar cortex; BA 10), and a midline portion of frontal cortex encompassing parts of dorsal ACC (BA 32) and the pre-supplementary motor area (pre-SMA) (BA 6) (for an example, see Figure 18.2). Beyond PFC, both tasks additionally exhibited shared activation with a sustained temporal profile in the left parietal cortex (BA 40/39), and left temporal cortex (BA 21). The sustained activation pattern shared by working memory and episodic memory involved three out of four PFC regions that were identified in a previous study by Nyberg and colleagues (2003) to show common recruitment in two large-scale PET experiments that together comprised multiple tasks indexing working memory, episodic memory and semantic memory, respectively. These common frontal lobe regions involved left VLPFC, frontopolar cortex and the dorsal ACC, all of which (in particular VLPFC and ACC, and to a lesser degree frontopolar cortex) have been (p.312)



                   Intersecting the divide between working memory and episodic memoryEvidence from sustained and transient brain activity patterns

Figure 18.2 Illustration of a common frontal sustained activation increase. See also color plate 6.

highlighted in a number of comprehensive reviews of across-study similarities in PFC activity (Cabeza and Nyberg 2000; Christoff and Gabrieli 2000; Duncan and Owen 2000; Fletcher and Henson 2001; Ramnani and Owen 2004), which further corroborate the generality of their involvement in a wide spectrum of memory tasks and other cognitive challenges. Despite extensive research, it remains unclear how to best characterize the precise functional contribution of these regions, although several candidate processes have thus far been proposed. Nevertheless, many prior accounts regarding the nature of the component processes mediated by these commonly recruited regions (i.e., theories that posit item-related transient processes) cannot account for the current data, as our analysis clearly demonstrated item-independent sustained activity within these areas, and predominantly no transient neural activity changes.


In the overall task comparison both the right VLPFC (BA 47) and medial PFC/ACC (BA 6/32) regions were found to yield a similar state-related effect across all four tasks relative to a low-level baseline (resting while maintaining the gaze upon a small circle in the center of the visual field), which suggested a generic role in basic attentional processes such as maintaining an attentive state throughout task performance and an enhanced level of alertness in comparison with baseline. The right lateralization of the VLPFC response associated with attention converge with prior imaging studies that have found right hemisphere PFC regions to be engaged in vigilance tasks (Pardo et al1991). Another closely related interpretation would be that these areas may play a role in the apprehension of upcoming events, a feature pertaining to all tasks. Such a notion concurs with the consistent findings of activation in right VLPFC/anterior insula and medial PFC in association with task preparation and attention in studies that have employed different kinds of cueing paradigms (Brass and von Cramon 2002; Luks et al2002; Curtis et al2004). All of these processes may be partly or entirely subserved by a common attention network (Pashler et al2001), and ACC has for a long time been proposed to play a key role in an anterior attentional system (Posner and Petersen 1990). In contrast to the regions in right VLPFC and ACC, the frontopolar (p.313) regions (BA 10) which was found to exhibit common sustained activity in working memory and episodic memory elicited no significant response during attention/vigilance. In the literature, the frontopolar area has been intimately linked with episodic memory, and more specifically, the instantiation and maintenance of retrieval mode, which reflects a specific cognitive state or task-set that supports recovery of episodic event information (Tulving 1983; Lepage et al2000). Although this account fits well with the sustained temporal signature of the observed frontopolar activity, it cannot explain our findings of parallel effects in both working memory and semantic memory. The common sustained effect obtained across all memory tasks (but not attention/ vigilance), clearly indicate that this region of anterior PFC may play a more general role in state-related control than endorsed by the retrieval mode hypothesis, which putatively correspond to what in the literature has been referred to as task-set maintenance or context representation (Braver et al1999; Allport and Wylie 2000). Some authors have considered such control processes to correspond to a procedural working memory: however, in our view such accounts merely encompasses a subset of the high-level control processes that subserve task-set maintenance. Task-set or ‘task-set configuration’ refers to the task goal, strategic and procedural directives and constraints, stimulus-response mappings, and an abstract model of the appropriate set of operational component processes to be utilized in order to achieve the goal. Importantly, task-set representations can be defined as schemas in its most abstract form. This implicates very fundamental control mechanisms that should be engaged in the service of performance on practically any type of experimental task, even very simple tasks, although the complexity of the task-set representation and required top-down modulations would differ. However, it should be emphasized that we posit that the task-set processes represented by the anterior-most section of PFC may diverge from those being exploited in very basic tasks contexts (e.g., attention/vigilance tasks). In congruence with the cognitive control theory put forward by Braver, Cohen and Barch (2002) in which context representations (i.e., task-sets) are contingent upon sustained neural implementation, the neural mechanisms of the frontopolar control processes presumed to subserve task-set maintenance in all memory tasks appear to operate throughout the duration of task performance as indicated by the sustained neural activity. Frontopolar cortex allegedly maintains a mental state of tonic intentionality and anticipation related to current and future goal states.


Theories concerning the brain mechanisms that underlie cognitive control related to representing and maintaining contextual information and the exertion of top-down modulation in the service of goal-directed behavior have typically focused on the DLPFC (BA 9/46) as the principal brain structure responsible for such higher-order control processes (Braver et al2002; Curtis and D’Esposito 2003). A primary influence to this strong emphasis on DLPFC in contemporary theories of cognitive control and working memory comes from experimental studies of nonhuman primates (e.g., Funahashi et al1993).


Human neuropsychological studies of patients with selective damage to the DLPFC have not always been able to establish a similarly strong link between DLPFC and working memory/ executive processes (Müller et al2002). These results would suggest that the working memory or executive processes (e.g., manipulation and monitoring) often specifically assigned to DLPFC in many theoretical models of PFC function may not be strictly dependent on this area. (A cautionary note: based on evidence of compensatory neuroplasticity suffice to functional reorganization in the human brain [Buckner et al1996] it could be argued that, since all DLPFC lesions were unilateral, critical operations subserved by the damaged area might be transferred contralaterally, whereby the operational responsibilities could be adopted by the intact homologous area in that hemisphere). Only lesions comprising both DLPFC and VLPFC were associated with impaired performance in all tasks (most severely in the two-back tasks), including simple one-back tasks. (p.314) Nevertheless, we do not question the critical role subserved by DLPFC in diverse executive control processes, although we posit that the specific control processes associated with task-set maintenance might be subserved by the frontopolar cortex.


As previously mentioned prior imaging studies have noted prominent activity in the right frontopolar cortex in tasks of episodic retrieval (Tulving et al1994b; Cabeza and Nyberg 2000; Lepage et al2000), seemingly independent of level of retrieval success, which served to inspire the attribution of this consistent pattern to the adoption and maintenance of retrieval mode (e.g., Lepage et al2000). However, as noted above, the retrieval mode hypothesis cannot account for the equivalent state effects associated with semantic categorization and working memory in the current study (Marklund et al2007). Prior imaging literature has reported instances of anterior PFC activation in working memory tasks (Braver and Bongiolatti 2002; Cabeza et al2002; Nyberg et al2002), the compliance of which appears to depend intrinsically on whether tasks entail subgoals embedded within working memory tasks or not (Braver and Bongiolatti 2002). Activations of frontopolar cortex have also been demonstrated in semantic monitoring (MacLeod et al1998), cognitive branching (Koechlin et al1999) and prospective memory (Burgess et al2003). Prior suggestive evidence for a frontopolar role in task-set maintenance was obtained in a recent event-related fMRI study by Sakai and Passingham (2003) that aimed to investigate preparatory neural activity evoked before the onset of upcoming working memory tasks. Frontopolar cortex was found to be activated in a sustained manner over a delay interval following the short presentation of a pre-task cue that merely informed participants of what specific task they were about to perform. This finding was taken to reflect the sustained engagement of frontopolar cortex to subserve the establishment and maintenance of task-set configurations and control processes related to task preparation (Sakai and Passingham 2003). In a mixed fMRI study that explored the temporal dynamics underlying task switching right frontopolar cortex demonstrated selective sustained activity during mixed-task blocks (i.e., not in single-task blocks) which was interpreted as reflecting augmented demand on cognitive control mechanisms to regulate the need for flexibility and keeping in mind multiple tasks (Braver et al2003). Finally, two previous mixed fMRI studies of episodic retrieval both reported sustained frontopolar activity, although a hemispheric asymmetry in the brain responses was revealed between the studies (Donaldson et al2001; Velanova et al2003). In one of these episodic memory investigations the anterior PFC showed left lateralized sustained recruitment in a simple item-recognition task (Donaldson et al2001), whereas in the other episodic study two discrete frontopolar sites elicited right lateralized sustained activity that modulated with increased amount of controlled processing required during retrieval (Velanova et al2003). None of the regions demonstrated any transient effects. Collectively, the current data and prior literature clearly indicate state-related processes, rather than item-related processes to encompass the functional contributions of the anteriormost portion of PFC (BA 10). Ramnani and Owen (2004) have proposed that anterior PFC is engaged in task context processing when there is demand to integrate the outcome of two or more different cognitive operations in a coordinated fashion to meet task requirements. However, this view cannot readily explain the general finding of sustained frontopolar activity in simple item recognition tasks and semantic categorization.


On the basis of current findings and prior data we propose that state-related control processes subserved by anterior PFC (BA 10) appear to be recruited in task situations that require the integration of a predefined type of internal representation not externally accessible (e.g., conceptual knowledge, episodic memoranda, working memory content), with some anticipated aspect of the immediate or intermediate future (e.g., upcoming item stimuli such as retrieval cues) via a discrete set of cognitive operators or computational maneuvers (e.g., classification, memory search, matching, updating, selection processes), in accordance with a given set of ‘if-then’ rules given by (p.315) the task instructions (i.e., task-set configuration including stimulus-to-response mappings). In other words, right frontopolar cortex is posited to play a critical role in task-set control processes that might be preferentially engaged in tasks that involve working memory, and episodic and semantic long-term memory. This would imply a putatively distinct conception of a particular kind of task-set configuration that might be referred to as a ‘general retrieval mode’.


Selective sustained neural activity: distinct processes related to active maintenance and monitoring within working memory


As would be expected, there was a set of prefrontal areas that exhibited sustained neural activity during delay-intervals in the two-back working memory task, in the absence of corresponding state-related effects in the episodic memory task. These regions involved the left mid-DLPFC (BA 9), right posterior DLPFC (BA 6) and the medial frontal cortex including dorsal ACC (BA 6/32). Conversely, no region was selectively engaged in association with maintenance of the task-set assumed to be uniquely devoted to retrieval mode (see above). It should be noted, though, that the apparent lack of process-specificity in the state-related activation pattern associated with the episodic retrieval task might be eloquently linked to the pertinent differences in processing load across the two functions. Taking into consideration system-specific neurocognitive component processes generally found to evoke sustained neural activity in respective task, the episodic task concur principally with retrieval mode related to the cognitive state of directed remembering of things past, whereas the two-back working memory task involve constant online processing related to active maintenance and continuous monitoring of the two most recently presented items. Hence, with respect to processing demands, the ISIs in the episodic retrieval task as compared to the two-back task must be considered relatively process vacant. Although state-related neural activity was found in regions classically associated with retrieval mode (e.g., right-lateralized frontopolar and VLPFC activations), they were also activated in the two-back task. Prior imaging studies of working memory have also reported significant activity in frontopolar areas, although not invariantly so (Braver and Bongiolatti 2002).


A quite unexpected result was the finding that the state-related pattern of sustained activity associated with semantic categorization showed such an extensive overlap with the activation pattern exhibited by working memory. Only the region of left hemispheric mid-DLPFC (BA 9) prevailed as selective to working memory after taking into account or disregarding sustained activity that was shared with the semantic memory task. Although the type of semantic task employed inherently involves working memory in order to retain the specific category instance given in the task cue (we used four different categories, the specific target category being presented in the task cue immediately before each of four task blocks) and that all non-target items belonged to a strongly related category (e.g., fruit vs vegetables), it was surprising that only the left DLPFC (BA 9) region turned out to be selectively engaged in the two-back task.


The functional contributions ascribed to this area appear closely linked to the type of control operations generally associated with the central executive (e.g., D’Esposito et al1998). Two major perspectives related to DLPFC functions can be distinguished in the imaging literature. One view holds that DLPFC is preferentially involved with the mechanisms underlying active maintenance of task-relevant information within working memory (e.g., Fuster 1995; Goldman-Rakic 1995), whereas the other view postulates a broader role of DLPFC in subserving multiple executive control processes that monitor, coordinate and manipulate/act upon representations held online within working memory while attributing active maintenance processes to the VLPFC (D’Esposito et al1998; Smith and Jonides 1997). Brain imaging studies specifically designed to dissociate the brain regions involved in maintenance versus manipulation processes (p.316) have reported inconsistent findings (Veltman et al2003). Successful two-back task performance requires the combined deployment of processes underlying maintenance and manipulation/ monitoring. Regarding the selective sustained activation that we observed, there are a number of distinct task properties of the two-back paradigm that differentiate it from the other tasks with respect to control processes expected to operate throughout the ISIs. Specific requirements involve the need to constantly hold in memory the two most recent items and keep track of which item was most recently presented. These specific task demands are assumed to engage processes of active maintenance and monitoring of item information (representations of item identity and their interrelational order), both of which generally engage the DLPFC (Owen et al1996).


Overlapping transient neural activity: shared control processes related to response selection and context-integrative item coding into transiently internalized representations



Similarities in transient item-related activity between working memory and episodic memory were assessed after first controlling for sensory and motor activity by subtracting out transient responses elicited by target detection in the attention/vigilance task. Overlapping transient activity was found in three frontal sites including left posterior DLPFC (BA 44/9), a small area of left VLPFC (BA 47), and medial PFC involving ACC/pre-SMA (BA 32/6). Other areas exhibiting shared transient engagement were located in left superior parietal cortex (BA 7) and medial cerebellum. In comparison with the item effects observed in the semantic memory task, it was found that all of the transiently activated regions that were shared between working memory and episodic memory also exhibited similar transient activity in the semantic memory task. The finding of shared involvement across all memory tasks in a region of left posterior DLPFC (approximating an area at the conjunction of BA 44/9) was of special interest (see Figure 18.3).



                   Intersecting the divide between working memory and episodic memoryEvidence from sustained and transient brain activity patterns

Figure 18.3 Illustration of a common transient activation increase. See also color plate 7.

(p.317) Noteworthy, this frontal area is identical to an area of commonality that was observed in a previous analysis assessing the prevalence of overlapping activation patterns across two PET experiments that together comprised nine different memory tasks (i.e., triplets of tasks were included for working memory, episodic memory and semantic memory) (Nyberg et al2003). However, since PET data (and blocked designs in general) do not permit inferences with respect to differential temporal properties of neural activity changes, this former analysis made detailed functional interpretations of the reported common DLPFC activity problematic.


As discussed in the prior section on overlapping sustained neural activity, several authors have considered DLPFC to represent an area of central importance in the service of different aspects of higher-order cognitive control and executive functions (Braver et al2002). Reminiscent of the functions ascribed to the ‘supervisory attentional system’ in Norman and Shallice’s model (1986), the DLPFC has been posited to influence the level of activation within multiple other brain systems via sustained (excitatory and/or inhibitory) top-down modulations to bolster context-appropriate processing pathways and goal-oriented behavior (Cohen and Servan-Schreiber 1992). Although certainly a most prevalent region of activation in functional brain imaging investigations of the brain systems engaged in demanding working memory tasks and other executive tasks, the precise role of DLPFC in executive processing remains controversial. In the view of DLPFC as a sort of ‘task-process coordinator’ that represents and maintains context-relevant information in the service of exerting top-down modulatory influence on the neural processing throughout the cortex to promote efficient and accurate task performance, this area should be expected to exhibit sustained activation (Curtis and D’Esposito 2003).



The response selection hypothesis


An opposing view holds that DLPFC is preferentially concerned with memory-guided short-term control processes that act in the service of response selection, rather than maintenance processes (Frith 2000). Several recent imaging studies have provided evidence that supports this claim (e.g. Schumacher and D’Esposito 2002). For example, in an fMRI study of spatial working memory, where the task was explicitly designed to assure that selection of the relevant information held online in working memory could not be carried out until the trial was terminated (i.e., with the onset of a test probe) (Rowe et al2000), DLPFC was found to exhibit significant transient activity in association with the response phase, while showing no sustained activity during the maintenance phase. Consequently, it has been argued that brain imaging findings of persistent delay period activity in DLPFC may be attributable to control processes related to response preparation that may occur only to the extent that the employed experimental task design allows for response selection to take place during delays, i.e., in advance of its execution (Curtis et al2004).


According to the selection account, DLPFC should be expected to evoke transient neural activity associated with control processes that may help to resolve response competition (e.g., evaluating the relevance of active internal representations with respect to task goal). The current data converge with this account by demonstrating common transient activity in left posterior DLPFC across working memory and episodic memory in the absence of concurrent sustained activity. We also note that the present activation foci in the left posterior DLPFC (two peaks corresponding to the x, y, z coordinates –46, 6, 30 and –44, 16, 26) are more caudal and ventral as compared to the mid-DLPFC (BA 46/9) area that has previously been proposed as a prime candidate for higher-order control processes linked with the central executive (Curtis and D’Esposito 2003). However, our general transient activation focus in posterior DLPFC is in complete concordance with the focus of a previous event-related fMRI study that also examined working memory and episodic memory together (Cabeza et al2002). This cross-study overlap of common activation (p.318) foci applied only to the retrieval phase of the working memory and episodic memory tasks, whereas no significant effect was found in relation to either the encoding or maintenance phase in the study by Cabeza and colleagues (2002). Hence, across two independent within-study comparisons of working memory and episodic memory, retrieval was associated with a topographically identical overlap in transient left posterior DLPFC activity. However, our mixed study additionally revealed a similar transient activation for the semantic categorization task, which together with previously reported data (e.g., Nyberg et al2003) further implicate a rather domain-independent functional role of this region (at least with respect to a variety of mnemonic retrieval demands).


Brain imaging of episodic retrieval has previously shown increased transient activity in the left posterior DLPFC as a function of the executive requirements associated with memory search (Cabeza et al2003; Velanova et al2003; Wheeler and Buckner 2003), which might entail auxiliary increases on selection control processes. For example, the study by Wheeler and Buckner (2003) demonstrated significantly greater activity in the left posterior DLPFC for retrieval trials involving items studied only once as compared to retrieval trials involving either extensively studied ‘overlearned’ items or ‘new’ items. The observed transient effects related to episodic memory retrieval tasks that require executive control during retrieval may conceivably be accounted for by the (response) selection hypothesis, in that more elaborative evaluation would be required before a memory judgment could be decided upon for items that only received shallow encoding (items studied once) as opposed to those receiving deeper encoding (items repetitively studied) (Craik and Lockhart 1972).


Returning to the item effects obtained in our study, it was found that the PFC regions commonly recruited across working memory and episodic retrieval also exhibited similar item-related activity during semantic categorization. Both the left posterior DLPFC (BA 44/9) and left VLPFC (BA 47) have in numerous previous studies been strongly linked with controlled retrieval and selection of information from semantic memory (Badre and Wagner 2002) and corresponding left PFC regions have been proposed to subserve semantic working memory processes (Gabrieli et al1998)). With respect to such ‘narrow’ and domain-specific interpretations, which are typically inferred from studies of single cognitive functions in isolation, our results point to the value of employing multiple tasks of different cognitive functions for cross-comparisons within the same ‘within-subjects’ study.


However, a more general account of left PFC function has been proposed by Thompson-Schill (2003), who argues that regions within the left VLPFC (approximating BA 47/45/44), although representing the most replicated area of activation in semantic tasks (together with inferior temporal cortex), do not serve controlled semantic processing per se, but may rather mediate a general control mechanism supporting selection processes independent of cognitive domain. This account fits well with the previously discussed response selection hypothesis.

Only gold members can continue reading. Log In or Register to continue

Stay updated, free articles. Join our Telegram channel

May 10, 2017 | Posted by in NEUROLOGY | Comments Off on Intersecting the divide between working memory and episodic memory

Full access? Get Clinical Tree

Get Clinical Tree app for offline access