Working memory and its localisation in the brain based on neuroimaging studies

The second part of the twentieth century became a prominent period for cognitive science research on memory processes. Important evidence regarding the functionality and anatomy of human memory as a dichotomous system came from neuropsychological studies of brain‑damaged patients. Further neuropsychological studies introduced new cognitive tasks aimed at examining the processes of learning and remembering new information, which led to new discoveries and a broader understanding of the memory system as a highly complex system.

In 1974, English psychologist Alan Baddeley, together with his colleague Graham Hitch, proposed a multicomponent model of working memory (WM), which evolved from the theory of short‑term memory that was then understood as a unitary system. At that time, researchers separated the working memory system into three components: the central executive, the phonological loop, and the visuospatial sketchpad. Years of research later, taking into account the multifunctionality of information processing and the “limited capacities of the visuospatial sketchpad and phonological subsystems” (Baddeley, 2012), Baddeley suggested that a fourth component should be added to his multicomponent model, allowing all WM components to interact with long‑term memory, which was named the episodic buffer (Baddeley, 2000).

Since Baddeley’s WM model has been accepted and implemented in research in cognitive science, working memory, as a storage system for visual, spatial and auditory information and a manipulation system for such cognitive processes as attention, reasoning, problem‑solving, language and learning, has been extensively investigated to determine which areas of the brain are active during particular cognitive processes. A great contribution to understanding how articulatory information flows in the brain, which neural circuits are active during visual and spatial tasks, and which neural pathways are highlighted when information is processed within WM and interacts with long‑term memory has been made through the use of neuroimaging techniques such as positron emission tomography (PET), a scan that allows activation in the brain to be monitored through “introducing radioactive substance into bloodstreams” (Baddeley, Eysenck, & Anderson, 2015). Another method is functional magnetic resonance imaging (fMRI), which uses magnetic fields and radio waves, and creates detailed images of brain areas through measurement of blood flow (BOLD).

One of the first noteworthy studies that supported Baddeley’s multicomponent model of working memory, in which PET was applied, was conducted by Paulesu, Frith and Frackowiak in London (1993). The researchers measured regional cerebral blood flow (rCBF) in healthy participants during a cognitive task that required the use of the subvocal rehearsal system and phonological store (Paulesu et al., 1993). Differences in rCBF clearly showed that Broca’s area is linked with the subvocal system, whereas the left supramarginal gyrus is associated with the phonological store. The researchers assumed that even without direct auditory stimulation, the processing of visually presented letters passes through the subvocal rehearsal system to the articulatory loop after an earlier transformation into a phonological code. Employment of the subvocal rehearsal system by “rhyming decisions” was documented during the second experiment of Paulesu et al. (1993), in which, however, the phonological store was not engaged at any point during the task.

Detailed images of active brain structures obtained using PET during the two experimental tasks revealed that the areas of the brain associated with the articulatory loop are “Brodmann’s area (BA) 44, superior temporal gyri (BA 22/42), supramarginal gyri (BA 40) and insulae” (Paulesu et al., 1993). BA 40 has been assumed, through comparison of the magnitude of activation in both the phonological store and “rhyming” tasks, to be undoubtedly associated with the phonological store and thus with short‑term memory, which is strongly related to working memory processes. It is very important to mention that these results support previous studies of brain‑damaged patients, in which BA 40, a part of the left inferior parietal lobule, was found to be correlated with the phonological store. Additionally, the researchers found that activation of Broca’s area (BA 44), which is involved in speech production, occurred in both experimental tasks, which was important for understanding the processes of the rehearsal system. The PET study of Paulesu et al. (1993) also showed activation foci in the supplementary motor area (SMA) and cerebellum, which were concluded to be areas involved in language planning and execution.

With reference to Baddeley’s multicomponent model of working memory, the theory that the phonological store and the subvocal rehearsal system of the phonological loop are separate systems (Smith & Jonides, 1997) was supported by the study described above. Another very influential study that had a significant impact on supporting Baddeley’s model of memory was conducted by Edward Smith and John Jonides at the University of Michigan in 1997. Their results provided strong evidence that verbal, spatial and object information use different working‑memory systems and that their information processes are located in distinct areas of the brain.

Starting with a general analysis of the active parts of the brain during cognitive tasks, Smith and Jonides found the right hemisphere to be correlated with spatial memory, whereas the left hemisphere was clearly associated with verbal memory. After meticulous analysis of the images, the researchers stated that particular areas in the brain are involved in spatial information processing, namely the posterior parietal cortex and the anterior occipital cortex. Once again, the results firmly supported findings from studies of brain‑damaged patients, for instance the study by McCarthy and Warrington (1990), who concluded that the parietal region is linked with spatial memory processing.

Smith and Jonides (1997) stated that their results clearly support discoveries from previous neuroimaging studies that showed the occipital region to be associated with the preservation of visual information (Kosslyn, Alpert, Thompson, Maljkovic, Weise, Chabris, Hamilton, Rauch, & Buonanno, 1993; Kosslyn, Thompson, Kim, & Alpert, 1995). Another notable finding by Smith and Jonides (1997), which confirmed previous results from studies of patients with brain lesions regarding the location in the brain of WM components associated with verbal storage and rehearsal, concerned the posterior parietal cortex in the left hemisphere, where two areas were activated during verbal task performance. The scientists also concluded that during their verbal task, activation of frontal parts of the brain, including Broca’s area, could be observed, which again supported findings from earlier brain‑damage studies. Object recognition was assumed to involve the inferotemporal cortex, as well as the posterior parietal and premotor regions, which, as described above, are active during verbal task performance.

Many studies in which the main aim was to determine the location of information‑processing systems using neuroimaging techniques, such as the working memory studies by Petrides et al. (1993) and Cohen et al. (1994), routinely showed activation in the dorsolateral prefrontal cortex. Activation foci of particular importance during spatial tasks have been found in the right hemisphere, whereas the left hemisphere has been clearly associated with verbal processes (Smith & Jonides, 1997). A neuroimaging technique that allows more precise analysis of brain images and was used by Smith, Jonides, Marshuetz and Koeppe (1998) is functional magnetic resonance imaging (fMRI). Activation foci in brain areas were monitored during a memory‑load task with retention intervals of a few seconds, using echoplanar imaging. Activation of the dorsolateral prefrontal cortex was clearly observed during the memory‑load task, which was initially assumed to be a pure storage task. The researchers argued that although the task was a storage task, an executive function of attention, namely detecting information switching within the task, was one of the determinants of a dual‑task situation and could therefore explain the activation of the dorsolateral prefrontal cortex as an employment of the central executive (Baddeley, 2012). Overall, it has been stated that when executive processes are required during performance on a cognitive task, activation foci occur in the dorsolateral prefrontal cortex.

Although studies of information processes in working memory in healthy adults provide crucial findings for understanding localisation in the brain of areas that are active and required to perform particular cognitive tasks, very important implications can be found by focusing on neurodevelopmental changes in working memory and cognitive control functions. In the first decade of the twenty‑first century, many studies involving neuroimaging techniques, mostly using fMRI, examined changes in brain activation across the lifespan. These changes create a neuroanatomical map of working‑memory processes, which include short‑term storage of visual, spatial and auditory information. Neurodevelopmental neuroimaging studies, as well as neuropsychological studies in healthy adults, use highly targeted cognitive tasks that can test different assumptions regarding the cognitive anatomy of the brain. Paradigms used to determine developmental differences in working memory include interference suppression, selective attention, task‑switching, manipulation, response inhibition and selection, and error processing. Among the tasks used are 2‑back visuospatial working‑memory (VSWM) and single‑item VSWM tasks, the oculomotor delayed‑response (ODR) task, object working‑memory tasks, conjunction search, attention network tasks, VSWM tasks including a distractor task, set‑shift tasks, go/no‑go tasks, stop‑signal tasks, and the widely used Stroop task.

A very good example of a study in which many active brain areas were observed during task performance, thereby contributing to the cognitive map of the brain, is the fMRI working‑memory study by Kwon et al. (2002). This study, which involved a 2‑back VSWM task, showed increased activation in the lateral prefrontal cortex (LPFC) and posterior parietal cortex (PPC), with particular significance in the dorsolateral prefrontal cortex (DLPFC), ventrolateral prefrontal cortex (VLPFC), premotor cortex (PMC), angular gyri, intraparietal sulcus (IPS) and superior parietal lobule (SPL). Taking into account the findings from this study and from Olsen (2003), as well as the results of Klingberg et al. (2002), a correlation was found between increased anatomical activity in fronto‑parietal white matter and blood‑oxygen‑level‑dependent (BOLD) activation in the superior frontal cortex, intraparietal cortex and visuospatial working‑memory capacity. Edin et al. (2007) considered these findings highly important for the possibility of improving VSWM (Rumsey & Ernst, 2009).

In another neurodevelopmental study using fMRI, Booth et al. (2003) found that activation in the brain during a conjunction task was associated with the anterior cingulate cortex, and that the thalamus was linked to attention. Konrad et al. (2005), for the purposes of their study, used an attention network task that indicated many changes over childhood and adolescence. The study showed increased activation in the right middle frontal gyrus, which is associated with alerting, increased activity in the right temporoparietal junction, decreased activation in the superior frontal gyrus, which was assumed to be linked to reorienting, increased activity in the DLPFC, and, very importantly, decreased activation in the superior temporal gyrus, which in turn has been associated with selective attention (Rumsey & Ernst, 2009).

Interesting results came from the study by Crone et al. (2006), who used an object working‑memory task in three groups of participants aged 8–12, 13–17 and 18–25 years. Activation in the VLPFC, DLPFC and superior parietal regions correlated with increased accuracy in task performance. The conclusion was that working‑memory circuitry strengthens across the lifespan and can be modulated by cognitive training (Rumsey & Ernst, 2009).

In terms of neurodevelopmental changes, Gogtay et al. (2004) produced very convincing results regarding the regional pattern of grey‑matter loss in the brain, with early reduction in the orbitofrontal cortex (OFC) and ventrolateral PFC, followed by reduction in the dorsolateral PFC. Silvia Bunge and Eveline Crone (2009) argued that individual differences in maturation of prefrontal cortex subregions across the human lifespan affect the rate of development of particular cognitive control processes (Bunge & Zelazo, 2006). Furthermore, as Fuster (1997) assumed, interactions between the prefrontal cortex and parietal cortex play a crucial role in working‑memory processes. Improvements in cognitive control, as part of the central executive of WM, mainly arise from neural circuit connections between the prefrontal and parietal cortices. Due to the changes in grey‑matter volume described above, activation foci in brain regions during cognitive tasks may vary and, in some cases, may not occur at all (Smith & Jonides, 1997). Thus, research on the function of working memory from a neurodevelopmental perspective is very important for further studies in healthy individuals and brain‑damaged patients.

The research of Luna et al. (2001), who used an antisaccade paradigm (AS task) together with fMRI to test differences in activation in distinct brain regions, provided novel findings for the anatomical map of the brain. The AS task (an eye‑movement task) is associated with visuospatial working memory and thus, according to Baddeley (2012), is associated with the visuospatial sketchpad, a component of working memory in Baddeley’s model. In their neurodevelopmental study, Luna and colleagues demonstrated that across the human lifespan, activation foci in the supramarginal gyrus (Brodmann area 40) decrease, whereas activation increases in the basal ganglia, which are associated with procedural learning. Activation also increases in the intraparietal sulcus (IPS), which has been linked to visual‑attention processes. Activation foci were observed in the superior colliculus in the midbrain, which has been linked to visual information processing. The frontal eye fields (FEF), located in Brodmann area 8 (BA 8) in the frontal cortex and assumed to be involved in the control of visual attention, also showed increased activation in Luna’s task. Control of visual attention has been concluded to be related to the central executive, according to Baddeley’s theory (1974). Luna’s study showed that activation in the dorsolateral prefrontal cortex and lateral cerebellum peaks during adolescence.

The dorsolateral prefrontal cortex has been assumed to play an important role in cognitive processes related to the central executive, especially in decision‑making, planning and information manipulation in working memory. Language processing in working memory has been associated with the inferior frontal gyrus (IFG) in Brodmann areas 44, 45 and 47, where activation was shown in the fMRI study by Tamm et al. (2002). The researchers showed that across the human lifespan, activation in the SFG/MFG areas decreases in the left hemisphere. Other research has shown parietal and temporal cortices to be related to inhibition, as well as the ventrolateral prefrontal cortex to be linked to interference; both mechanisms are associated with learning.

Summarising all the findings from neuroimaging studies that have been strongly oriented towards investigating the localisation of working‑memory processes, it can be assumed that verbal working‑memory processes are predominantly left‑lateralised in the inferior frontal cortex, where Broca’s area forms part of the system, whereas visuospatial working‑memory processes are assumed to take place in the dorsal prefrontal cortex, where other cognitive processes, such as attention and decision‑making of the central executive, also occur. Object processes in working memory are linked to the ventrolateral prefrontal cortex (Wager & Smith, 2003). Smith and Jonides have provided strong arguments in favour of the effectiveness of using PET to investigate cognitive functions. The main advantage of employing positron emission tomography (PET) and functional magnetic resonance imaging in research on working memory is that they yield detailed images of activation foci in brain areas responsible for particular cognitive functions. The multicomponent model of working memory has been supported by many studies in which PET and fMRI have been applied as methods for detailed exploration of the human brain (Smith & Jonides, 1997; Rumsey & Ernst, 2009; Paulesu et al., 1993).