Complex span tasks and hippocampal recruitment during working memory

Abstract

The working memory (WM) system is vital to performing everyday functions that require attentive, non-automatic processing of information. However, its interaction with long term memory (LTM) is highly debated. Here, we used fMRI to examine whether a popular complex WM span task, thought to force the displacement of to-be-remembered items in the focus of attention to LTM, recruited medial temporal regions typically associated with LTM functioning to a greater extent and in a different manner than traditional neuroimaging WM tasks during WM encoding and maintenance. fMRI scans were acquired while participants performed the operation span (OSPAN) task and an arithmetic task. Results indicated that performance of both tasks resulted in significant activation in regions typically associated with WM function. More importantly, significant bilateral activation was observed in the hippocampus, suggesting it is recruited during WM encoding and maintenance. Right posterior hippocampus activation was greater during OSPAN than arithmetic. Persitimulus graphs indicate a possible specialization of function for bilateral posterior hippocampus and greater involvement of the left for WM performance. Recall time-course activity within this region hints at LTM involvement during complex span.

Introduction

Working memory (WM) is thought of as a system in which information currently in the focus of attention can be maintained and manipulated. It is also seen as a gateway through which sensory information can enter into long term memory (LTM) or through which information can be recruited from LTM into the focus of attention (Atkinson and Shiffrin, 1968, Baddeley and Hitch, 1974, Cowan, 1988, Engle et al., 1999a, Unsworth and Engle, 2007b). A properly functioning WM system enables an individual to keep attention on a desired goal while preventing other environmental or cognitive stimuli from interfering with the completion of the desired goal. Furthermore, WM is crucial when attempting to override our automatic responses through a set of cognitively salient processes (Unsworth and Engle, 2007b), thus making it critical for the performance of a variety of everyday tasks.

Atkinson and Shiffrin, 1968, Baddeley and Hitch, 1974 espoused the idea that the process of WM is that by which information is at some point stored in a location, LTM, from where it can later be retrieved by another system, short term memory (STM). The information is then manipulated, updated, and maintained in accordance with the aim of the present goal state. Of importance is also the idea that the information held in STM does not have to be retrieved from LTM, but maybe newly acquired information that has been linked with other information in LTM. Linking, or relational encoding, is necessary in order to attach meaning to the newly acquired information.

Many of the current discussions on WM have emphasized the concept of capacity limits. To describe this concept, Cowan, 1988, Cowan, 1999, Cowan, 2005 embedded processes model examines three states of memory: the information residing in LTM, recently perceived or accessed information that is in an easily accessible (activated) state in LTM, and a sub-portion of that information which we are consciously aware of, known as the focus of attention. WM capacity differences are believed to arise from the ability to keep the focus of attention on the task at hand while suppressing interference from environmental stimuli or irrelevant cognitions caused by the activation of other memory traces in LTM. Much in the same way, Unsworth and Engle, 2006, Unsworth and Engle, 2007b suggest that differences in WM capacity arise from an individual's ability to actively maintain information in primary memory (i.e., the focus of attention) while also performing a controlled search of the information residing in secondary memory (i.e., LTM).

Several tasks have traditionally been used to assess WM capacity. For example, the typical digit span task assesses capacity by determining the maximum length of numbers that an individual can serially recall. More involved tests, such as the digits backwards and letter-number sequencing tasks, assess capacity while also requiring the ability of mental double-tracking, meaning that memorizing and reversing/ordering operations must be performed simultaneously (Lezak et al., 2004, pp. 359–363). Other tasks like Daneman and Carpenter's (1980) reading span task and Turner and Engle's (1989) operation span (OSPAN) task are complex working memory span (CWMS) tasks that require the participant to engage in a processing activity that is irrelevant to the to-be-remembered information. They involve encoding, maintenance, storage, and processing of various types of information. Proper performance on CWMS tasks requires a high degree of executive attentional-control (Conway et al., 2003, Engle et al., 1999a, Kane et al., 2007a) to switch between tasks and maintain attention on the task at hand. Of critical importance, the irrelevant task is often thought to force the to-be-remembered information to be temporarily displaced from the focus. If the to-be-remembered information is properly encoded, it may be placed into and retrieved from LTM as required (Kane et al., 2007a). The displacement of task relevant information from the focus occurs because the irrelevant task usually requires controlled, effortful processing (Conway and Engle, 1996, Engle et al., 1999b); it is of sufficiently high cognitive load that it may occupy the whole of the focus of attention, thereby displacing any information which exceeds the individual's immediate WM capacity (Bunting, 2006, McCabe, 2008). It is this type of complex processing, and the resultant interactions of items coming into and going out of the focus, that make CWMS tasks invaluable for the cognitive study and neuroimaging of WM.

Functional neuroimaging experiments of WM have typically used tasks such as the Sternberg (1966) or the n-back (Gevins et al., 1990); we will refer to these types of tasks as traditional neuroimaging WM (TNWM) tasks. During the Sternberg task subjects are presented with a set of stimuli and are asked if the target stimulus matches any of the stimuli presented in the previous set. The n-back task is more complex in that there is a continuous presentation of stimuli; on target trials subjects are asked either if the target stimulus matches a stimulus presented n trials back (usually 1 to 3) or to identify how many trials back the target stimulus was shown. Generally speaking, most neuroimaging investigations of WM have associated a core of regions with WM functioning.

The prefrontal cortex (PFC) is believed to be integral to WM and executive control (D'Esposito et al., 2000, Owen et al., 2005, Wager and Smith, 2003). Sub-sections of the PFC, such as the ventrolateral prefrontal cortex (VLPFC) and dorsolateral prefrontal cortex (DLPFC), have been said to be involved in object and spatial domain specific processing (Courtney et al., 1998, Smith and Jonides, 1999), respectively. However, an extensive review of the neuroimaging literature by Wager and Smith (2003) indicated that PFC sub-regions were not so much domain specific as they were process specific. The DLPFC being involved in executive processes, such as attentional control or the monitoring of complex information (Cabeza and Nyberg, 2000), while the VLPFC is involved in storage-related processes such as the maintenance of spatial information (Toepper et al., 2010) or the rehearsal of verbal information (Cabeza and Nyberg, 2000). Bor et al. (2006) showed activation of the VLPFC during a task where spatial information had to be kept online without aid of a spatial strategy; when spatial strategies for remembering the target stimuli were given, activation was only exhibited in the DLPFC. Further involvement of the VLPFC in storage processes has been demonstrated in proactive interference tasks where the left inferior frontal gyrus has shown significant activation during a recent negative condition (Jonides et al., 2000, Jonides and Nee, 2006). This activity has been shown to occur specifically during the retrieval stage of the recent negative condition (D'Esposito et al., 1999) and is linked to the resolution of interference. A more recent review by Blumenfeld and Ranganath (2007) has further indicated the VLPFC's role in the resolution of interference by noting that it is consistently recruited when controlled selection of items is required. The DLPFC's roles in executive processing are further confirmed by demonstrating it is highly recruited for the organizational processing of information. Blumenfeld and Ranganath (2007) have also summarized the roles of VLPFC and DLPFC in LTM, suggesting VLPFC supports the formation of LTMs through the process of controlled item selection, while the DLPFC aids in building associations between items in LTM and those in the focus of attention. Another key region in the frontal lobes, the anterior cingulate cortex (ACC), is also believed to be necessary for proper WM function and is thought to be involved in conflict monitoring and error detection (Bernstein et al., 1995, Botvinick et al., 2001, MacDonald et al., 2000). Both of these are attentional control processes and as such the ACC is believed to be critical to cognitive control (Smith and Jonides, 1999, Osaka et al., 2003). Furthermore, Kaneda and Osaka (2008) suggest that the ACC may play a greater role in executive functioning than the DLPFC.

The parietal lobes are thought to function as associative centers and be involved in higher level cognitive processes. They are also believed to be crucial to WM processes and serve as storage regions (Hamidi et al., 2008, Postle, 2006, Postle and D'Esposito, 1999, Rowe et al., 2000, Srimal and Curtis, 2008). More specifically, the superior parietal lobule (SPL) and the precuneus (Brodmann area 7) are believed to be crucial in maintaining and organizing items held in the WM store (Wager and Smith, 2003, Wendelken et al., 2008), while the supramarginal gyrus (part of the inferior parietal lobule) is thought to retrieve the temporal ordering of items that have been displaced from the focus of attention through serial scanning (Öztekin et al., 2008). Parietal cortex is also thought to select the appropriate response for a specific stimulus, known as stimulus-response mapping (Corbetta and Shulman, 2002, Miller, 2000, Miller and Cohen, 2001).

The medial temporal lobes (MTL) have traditionally been associated with the encoding and maintenance of declarative LTMs. For example, Scoville and Milner's (1957) classic study demonstrated that bilateral lesions to the hippocampal formation produced an extremely detrimental impact on the retention of new memories. More recently, neuroimaging studies have challenged this limited conception of MTL regions by demonstrating hippocampal recruitment during various types of WM tasks, and specifically, during the maintenance phases of some of these tasks. Öztekin et al. (2009) found the hippocampus was active during item recognition trials of a serial position task and that activity increased for earlier items rather than the last item on a judgment of recency task. Using neurosurgically implanted electroencephalograph (EEG) electrodes, Axmacher et al. (2007) detected significant load dependent negative DC potential shifts and increases in synchronized gamma band activity in the rhinal cortex during the maintenance of multiple items during a visual Sternberg task. The negative DC shift likely indicating membrane potential depolarization and increased firing and/or synaptic activation of rhinal cortex neurons, while synchronized gamma band activity further indicated recruitment of the rhinal cortex. Van Vugt et al. (2010) furthered these findings by demonstrating a local load dependent gamma oscillatory power increase in the hippocampus during Sternberg task maintenance. Additionally, this increase was greater for non-verbal items (faces) than for verbal items (letters).

Even though item recognition tasks such as the n-back and the Sternberg have proven to be in valuable in dissociating many of the brain regions involved in WM functioning, there are reasons to explore the use of CWMS tasks in neuroimaging settings. For example, the n-back has been shown to account for variability in general fluid intelligence (Gf), but it has done so only under a 3-back condition, and this variance in Gf is separate than that accounted for by WM span (Kane et al., 2007b). WM span, as measured by a CWMS task though, has been shown to account for up to half the variability in Gf (Conway et al., 2003, Kane et al., 2005). A CWMS task like the OSPAN has been shown to possess high levels of reliability and internal consistency as compared to other measures of WM capacity (Klein and Fiss, 1999). An automated version of the OSPAN has also demonstrated high levels of reliability and internal consistency, and shown high levels of correlation with other measures of WM (Unsworth et al., 2005). Moreover, more recent work has demonstrated correlations between complex WM span tasks and traditional measures of LTM (Unsworth et al., 2009, Unsworth, 2010) at sub-span levels, unlike TNWM tasks. Such correlations make the case for the use of CWMS tasks in neuroimaging WM research, especially when attempting to decipher the possible interplay of WM and LTM.

Unfortunately, CWMS tasks have seen little use in neuroimaging studies. In one of the few, Kondo et al. (2004) found the OSPAN elicited activation in regions usually activated during the n-back (e.g., PFC, ACC, and SPL), while the high-span group also exhibited significant activation in the inferior temporal cortex. Kondo et al. (2004) was limited, however, in that they mainly restricted their analysis to the functional connectivity differences of the cingulo-frontal network between high-span and low-span individuals. Recently Chein et al. (2010) examined domain general mechanisms during encoding and maintenance, and examined MTL activity during recall for verbal and spatial complex span tasks. They found activity in areas typically associated with WM during encoding and maintenance, and found the posterior hippocampus and immediately inferior portion of the parahippocampal gyrus were involved in the recall portion of the task. However, they did not specifically examine MTL recruitment during encoding and maintenance. Therefore, a more in depth neuroimaging exploration of CWMS tasks and the roles of MTL ROIs during encoding and maintenance is warranted.

In this study we aimed to elucidate whether significant differences exist between the neural resources required for the performance of CWMS and TNWM tasks. More precisely, we wanted to determine if the encoding and maintenance phase of a CWMS task results in significantly greater recruitment of areas typically associated with LTM functioning than might occur during a TNWM task. As previously stated, Axmacher et al., 2007, Van Vugt et al., 2010 demonstrated hippocampal activity during a Sternberg task maintenance, a TNWM task. To examine this, we compared functional magnetic resonance (fMRI) activity observed during the OSPAN task (complex span; letter span + equation verification) with that of an arithmetic task (traditional type of neuroimaging WM task; equation verification).

The goal of the OSPAN task is to recall the to-be-remembered items (letters) in serial order. During the OSPAN, equation verification is presented as the irrelevant, cognitively demanding task. As a result, participants should often not have sufficient cognitive resources available to rehearse the to-be-remembered letters while performing equation verification. In other words, the equation verification should occupy the focus of attention causing a displacement of the to-be remembered letters from the focus to LTM. If the to-be-remembered items have been displaced from the focus and properly stored in LTM, they can later be retrieved from LTM as needed. By contrasting the OSPAN and Arithmetic conditions we hoped to control for the common activation patterns resulting from equation verification in order to demonstrate that the OSPAN task forces recruitment of regions associated with LTM binding and retrieval, and that recruitment occurs during the encoding and maintenance phase of the task.

We hypothesized that 1) the OSPAN encoding and maintenance phase and Arithmetic task would yield activation in regions commonly associated with WM and the resolution of interference during on-going retrieval, such as VLPFC, DLPFC, ACC, SPL, and inferior parietal lobule (IPL), 2) activations in these regions would be greater during the OSPAN since CWMS tasks should require greater executive control than a typical neuroimaging WM task, 3) due to the nature of the OSPAN, activation would also be evidenced in areas typically associated with LTM binding and retrieval, specifically the hippocampus, and 4) the nature of CWMS tasks would be sufficiently different from TNWM tasks, resulting in unique patterns of activity within LTM associated regions such as the hippocampus. This would be the first instance where such activity would be demonstrated for a CWMS task during maintenance and encoding. We also aimed to explore what pattern of brain activity during a CWMS task is correlated with correct and incorrect recall. In other words, we explored what patterns of brain activation are associated with WM capacity and proper storage and retrieval. Edin et al. (2009) indicates we may find correct recall is associated with heightened DLPFC activity which modulates parietal activation.