Introduction

The ability to maintain relevant information in mind in the presence of interference or distracting information is critical for higher cognitive functions required in daily life. Working memory (WM) is the theoretical construct used to refer to this capacity to simultaneously maintain and process information over brief periods of time according to current task goals [1–3]. Studies in children and adolescents show that WM capacity plays a crucial role in the development of many cognitive activities (e.g., learning, reasoning, problem solving, language comprehension), and also predicts academic performance and achievement [4–6]. Moreover, WM is impaired in various developmental disorders, e.g. attention deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD) or specific language impairment (SLI), providing a crucial neuropsychological measure in several neuropsychiatric conditions and useful risk marker for cognitive development [7–9].

From a developmental point of view, WM capacity develops rapidly over childhood [10–13]. This is usually measured by the increase in the amount of information that can be retained and transformed using complex memory span tasks that require maintaining information for further recall while performing a concurrent activity [7]. An important component of WM maintenance, involving active verbal rehearsal and attentional refreshing, emerges around 7 years of age [14]. Evidence suggests that multiple mechanisms contribute to childhood development of WM, affecting all the processes involved in encoding, maintenance, and retrieval (e.g., increase in attentional capacity, process automatisation, increase in knowledge, mnemonic strategies, and so forth; see [15]).

In terms of neural substrates, development of WM ability parallels structural changes in frontal-parietal cortices affecting grey matter [16] and white matter [17]. Similar to neuroimaging findings in adult populations, this core network of fronto-parietal brain areas is consistently found to activate in children and adolescents, and is apparent as early as 5 years of age during different verbal and visuospatial tasks thought to evaluate WM functions [18–20]. One recent imaging study compared encoding and retrieval processes in a Sternberg item recognition paradigm with digits in children and adolescents from 9 to 19 years [21]. Encoding of digits activated the right prefrontal and parietal cortex, left motor areas, occipital cortex, and cerebellum; retrieval activated the left prefrontal and parietal cortex, right motor areas, as well as anterior and posterior cingulate cortex, and cerebellum. Other functional neuroimaging studies investigating WM in school-age children have used an n-back task in which a sequence of stimuli is presented to the participant who must indicate when the current stimulus matches the one from n steps earlier in the sequence (e.g., [22, 23]). Despite its popularity in fMRI studies, empirical evidence shows that the n-back task correlates weakly with WM span tasks, suggesting that it is unlikely that these two types of tasks reflect a single construct, and questionning the empirical validity of using n-back tasks (continuous-recognition or updating measures) as a WM task [24, 25]. Other tasks, such as the Steinberg item recognition paradigm (e.g., [12, 26]), have also been used to study WM in developmental populations. However, these tasks require the maintenance of information in short-term memory, but not the simulatenous maintenance and manipulation of information as the theorethical construct of WM specifies [3, 27]. Thus, very few developmental studies have explored the neural correlates of WM using tasks requiring not just maintenance, but also active manipulation of information [18, 19]. To our knowledge, brain activity associated with WM processes of maintenance during the simultaneous processing of a concurrent task and retrieval have not yet been studied in developmental fMRI studies.

Previous literature has identified the major challenges inherent in studying both typical and atypical development, including designing tasks that can be administered to individuals across a wide age range in both typical and atypically developing groups [28]. In this study, our aims were to design a novel WM paradigm that: i) is demanding of WM capacity but simple enough to be administered to both children and adolescents and both healthy and clinical paediatric populations (e.g., populations with mild intellectual difficulties), and for which brain activity could not be explained by difference in age or WM performance; ii) would enable investigation of neural substrates for encoding, maintenance and retrieval WM processes during fMRI; and could identify the effect of different concurrent processing tasks on maintenance and retrieval.

Among the paradigms appropriate for measuring the impact of concurrent processing on maintenance, the Brown-Peterson task is best suited to examine encoding, maintenance, and retrieval processes in WM. The original Brown-Peterson task requires participants to encode and retrieve a string of letters with a concurrent task (i.e., counting backward by three) interposed between encoding and subsequent retrieval [29, 30]. In opposition to the immediate serial recall paradigm, the concurrent task in Brown-Peterson paradigm impairs maintenance and thus retrieval of the encoded information. Here, we designed a novel task inspired from the Brown-Peterson paradigm in which children and adolescents had to maintain verbal information (letters) while performing a concurrent task involving either verbal (lexical decision) or visual (face decision) task appropriate for children and adolescents. This design allowed us to compare not only encoding and retrieval components of verbal WM during fMRI, but also to probe for neural substrates differentially modulated by the concurrent task, both within-domain (i.e. verbal distractors) and cross-domain (i.e. visual distractors). According to the influential model of Baddeley (1986; [31]), verbal and visuo-spatial maintenance and processing involve separate and domain-specific systems, a phonological loop for verbal information and a visuospatial sketchpad for visuospatial information. Thus, processing irrelevant verbal information should produce selective interference with verbal maintenance because verbal processing would mobilize the phonological loop, thus impeding the articulatory rehearsal process in charge of verbal maintenance. By contrast, processing visuospatial information should involve the domain-specific visuospatial sketchpad and should not have any effect on verbal maintenance.

To validate this novel paradigm, we applied it in children and adolescents aged 8 to 16 years. We expected that all would successfully complete our adapted Brown-Peterson fMRI paradigm, which tailors task difficulty to each participant according to their WM capacity. We predicted that distinct activation patterns would be elicited by the two concurrent tasks (i.e. within and cross-domain), not only during the maintenance interval, but also during the subsequent retrieval period. Based on Baddeley’s WM model [30], the nature of the concurrent task was expected to differentially impact verbal WM and thus modulate brain areas recruited during retrieval, despite the fact that identical verbal stimuli were encoded. Specifically, exposure to words vs faces during the maintenance interval should hamper vs favour the engagement of language-related regions in the left hemisphere during the subsequent retrieval phase.

Materials and methods

Material and design

Participants completed an adapted version of the Brown-Peterson paradigm [29, 30] implement during functional magnetic resonance imaging (fMRI). A mixed block and event-related design allowed separate examination of specific WM processes: encoding, maintenance and retrieval. The task required a combination of verbal storage and maintenance during either verbal (within-domain) or visual (cross-domain) concurrent tasks. Each active trial consisted of three active phases (Fig 1):

1) Encoding period.

Participants were presented with a series of single upper-case letters for further recall displayed sequentially in the middle of the screen at a rate of one letter per second. All consonants of the English alphabet were used as memory items except W, which is three-syllabic. Series of two and three letters were created for within-domain and cross-domain blocks in such a way that each letter appeared with the same frequency in both blocks. Participants were asked to maintain the letters in order of appearance.

2) Maintenance delay filled with a concurrent task.

During the maintenance delay of 6 seconds, a concurrent task required to process either verbal or non-verbal stimuli involving within- or cross-domain interference respectively.

The within-domain concurrent task was a lexical decision task. Two successive letter-strings were presented for 3 seconds each and required simple motor responses (i.e. press as quickly and as accurately as possible the left-most/green button if the letter-string was a word; or the right-most/red one if it was a non-word). Words were selected from the “Oxford Wordlist”, which is an Australian database of high frequency words in young children’s writing and reading development [35]. Among the 307 most frequently used words, only nouns were selected based on the following search terms: any gender, any location (urban or rural), any socioeconomical status, any text type (e.g., description, discussion, narrative) and appearing during the first three years of school (40% were within 1 to 100 most frequently used words; 35% were within 101 to 200 most frequently used words; 25% were within 201–307 most frequently used words). Non-words with orthographically existing onsets and bodies were selected from the “ACR Nonword database” [36]. Three to eight letter-strings (words and non-words) were displayed centrally on the screen. Words and non-words were equally often presented.

The cross-domain concurrent task was a face decision task. Two successive pictures were presented for 3 seconds each, requiring similar motor responses (i.e. press as quickly and as accurately as possible the left-most/green button if a real face was presented; or the right-most/red one if it was a scrambled face). Ten males and 10 females faces with a neutral expression were selected from the NimStim database [37], and converted into greyscale using Matlab R2013a (The MathWorks, 2012). Scrambled faces were created from the original faces using Matlab (size of square = 300, iterations = 2). Faces and scrambled faces were equally often presented.

3) Retrieval period.

At retrieval, one single upper-case letter was presented along with either one or two place-holders (for paradigm with 2 or 3 letters to remember, respectively) made of dashes with a question mark. Participants had to decide if the single letter matched the letter that was presented in that serial position during the encoding period by giving a simple motor responses, i.e. press as quickly and as accurately as possible the left-most/green button or the right-most/red one for positive and negative responses respectively. This was done to make sure that participants memorised both item and serial order information.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Adapted Brown-Peterson fMRI paradigm using within- and cross-domain concurrent tasks.

https://doi.org/10.1371/journal.pone.0179959.g001

In addition to the active condition, there was a baseline condition (no-concurrent task) in which participants were required to encode a single letter and recognise it after a short empty delay of 1 second. They were instructed to press as quickly and as accurately as possible the left-most/green button if the single letter was the same during encoding and retrieval; or the right-most/red one if it was a different letter.

For both the active and baseline conditions, a randomized inter-trial interval of 2000, 2500, or 3000 milliseconds was presented before the next trial. Three types of blocks of 10 trials each were created: two active blocks, one including the within-domain concurrent task and the other including the cross-domain concurrent task, and a third baseline block. The order of presentation of these three blocks was counterbalanced across participants and repeated twice for a total of six blocks of 10 trials. Within each block, half of the probes were positive (i.e., 5 trials required a “yes” response) and the position of positive and negative probes were randomized within each blocks.

Two important challenges of brain imaging studies examining cognitive development are that differences in both participant age and task performance may influence activation patterns. One concern is whether changes in neural activity reflect changes in functional maturation of the central nervous system, independently of behavioural efficiency, or whether they reflect changes in task performance consequent upon increasing age [22, 38]. For these reasons, in our paradigm, task difficulty was adapted to each participant by adapting the number of verbal items to remember. Based on pilot testing conducted outside the scanner, participants with a backward digit span of 5 or more were presented with the version of the paradigm with 3 letters to be remembered, and those with a backward digit span lower than 5 were presented with the version of the paradigm with 2 letters to be remember. In our sample, seven participants completed the 3-letters paradigm (age range = 10 to 15 years; M = 12.53; SD = 1.44) and nine participants completed the 2-letters paradigm (age range = 8 to 16 years; M = 11.93; SD = 2.78). All participants had a retrieval accuracy of 80% or more, which suggested that task difficulty was appropriate for each participant.

Results

Functional magnetic resonance imaging

Active letter encoding and retrieval vs. baseline.

To delineate brain regions generally recruited during WM, we first contrasted the active encoding period relative to the baseline encoding period, regardless of the domain of concurrent task during the maintenance interval. This showed activation in a widespread network, including bilateral visual areas in the occipital lobes, parahippocampal gyri, as well as left prefrontal regions, the caudate nucleus, and the cerebellum (Table 1). Likewise, we contrasted the active retrieval relative to the baseline retrieval period, regardless of concurrent conditions, which revealed a distributed pattern of activation encompassing mainly bilateral prefrontal cortices, but also temporal and parietal areas (Table 1). These data confirm that our working memory paradigm successfully engaged brain networks associated with visual stimulus processing and executive functions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. List of activations for active encoding and retrieval compared to baseline condition.

https://doi.org/10.1371/journal.pone.0179959.t001

Active letter encoding vs. letter retrieval.

We next sought to identify regions selectively recruited by distinct WM processes. Encoding, as compared to retrieval (during the active task), was associated with widespread activations bilaterally in the occipital and ventral temporal lobes (inferior occipital and fusiform gyri), as well as in medial frontal areas (supplementary motor area (SMA), middle cingulate gyrus) and precentral gyrus. Smaller activation foci were found in the insula (Fig 2 and Table 2). Conversely, the retrieval phase, compared to encoding, activated bilateral dorsolateral prefrontal areas (mainly inferior and middle, but also superior frontal gyri), as well as the anterior cingulate cortex (ACC), inferior parietal lobule (angular, supramarginal, and postcentral gyri), and lateral temporal areas (superior and middle temporal gyri).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Activation maps related to the contrasts encoding vs retrieval.

https://doi.org/10.1371/journal.pone.0179959.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. List of activations for contrasts of interest.

https://doi.org/10.1371/journal.pone.0179959.t002

Maintenance delay filled with a concurrent task (within-domain vs. cross-domain).

Comparing activations during the within-domain concurrent task (lexical decision task), relative to the cross-domain concurrent task (face decision task), revealed differential increases in the right middle frontal gyrus (Brodmann area 10) and medial fusiform cortex only (Table 2 and Fig 3). Conversely, the cross-domain concurrent task (face decision task) compared to within-domain concurrent task (lexical decision task) produced a more extensive pattern of activation, particularly in bilateral visual areas, including occipital and fusiform cortex overlapping with the fusiform face areas (FFA). Activations were also found in several frontal areas (left inferior and medial frontal gyri, SMA, right middle cingulate cortex, precentral gyrus), the temporo-parietal junction, left parahippocampal gyrus, and right pulvinar. Thus, the cross-domain concurrent task appeared to recruit a more widespread network than the within-domain concurrent task, even though behavioural data show that this could not be explained by task difficulty since accuracy did not significantly differ in the two concurrent tasks.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Activation map for the contrast within-domain vs cross-domain concurrent tasks (MRIcron reference slices).

Activations in yellow: within-domain concurrent task > cross-domain. Activations in red: cross-domain concurrent task > within-domain.

https://doi.org/10.1371/journal.pone.0179959.g003

Letter retrieval following within-domain vs. cross-domain concurrent tasks.

The most critical question concerning the WM system in our paradigm is whether the nature of the concurrent task during the maintenance interval may produce different degrees of competition and thus result in different neural substrates during retrieval. We therefore tested for brain regions that would be differentially activated during the retrieval period when following within-domain concurrent task (lexical decision) or when following cross-domain concurrent task (face decision). Greater increases following the within-domain concurrent task were found in visual areas, with large bilateral clusters in occipital cortices (bilateral middle and inferior occipital gyri, fusiform gyri, right cuneus and left calcarine). Conversely, greater increases were found after the cross-domain concurrent task in the left middle and superior temporal cortex, overlapping with usual location of phonological processing [48, 49], plus left calcarine gyrus and bilateral medial occipital cortex (Table 2 and Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Activation map for retrieval following within-domain vs cross-domain (MRIcron reference slices).

Activations in yellow: retrieval following within-domain concurrent task > cross-domain. Activations in red: retrieval following cross-domain concurrent task > within-domain. For illustration purpose, activations observed in retrieval following within-domain concurrent > cross-domain are represented with a threshold of p < .005 uncorrected for multiple comparisons.

https://doi.org/10.1371/journal.pone.0179959.g004

Age and retrieval accuracy-related activations.

Several functional ROIs were defined for each of the contrast of interest described above (marked with a star in Tables 1 and 2) and used for additional analyses to examine any modulation by individual characteristics of the participants. Parameter estimates (beta values) extracted and averaged across voxels from these ROIs were then submitted to Pearson’s correlation with age and WM retrieval accuracy. No significant correlation was found between encoding- or retrieval-related activation (relative to baseline activation) with neither age nor WM retrieval accuracy on the adapted Brown-Peterson paradigm for any of these ROIs. Table 3 summarizes these correlation coefficients.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Pearson’s correlations between activity of functional ROIs and (a) age or (b) retrieval accuracy.

https://doi.org/10.1371/journal.pone.0179959.t003

We also performed an exploratory whole-brain regression analysis in SPM using (a) age; and (b) WM retrieval accuracy for the main contrasts of interest as described above (encoding vs retrieval, within vs cross domain concurrent tasks). None of these analyses revealed any significant overlap with activations identified by the main contrasts of interest reported in Table 2 indicating that all effects reported above are largely independent of age (within the range of our sample) and WM retrieval accuracy.

Discussion

We report and validate an adapted Brown-Peterson fMRI paradigm that probes for the neural correlates of different WM processes, including encoding, maintenance and retrieval, as well as the effect of within- and cross-domain concurrent tasks during maintenance. Results indicate that this paradigm can be performed equally well by children and adolescents of different ages, with reliable results across different levels of performance. To our knowledge, this is the first study to propose a paradigm to delineate distinct patterns of brain activity for the different WM processes in children and adolescents. We provide the first exploratory results on brain activity related to encoding, maintenance, and retrieval WM processes in children and adolescents, and compare verbal WM in the presence of both verbal (within-domain) and visual (cross-domain) concurrent tasks.

As expected, our adapted Brown-Peterson paradigm was successfully completed with high accuracy in the MRI scanner by typically developing children as young as 8 years of age, indicating that it is suitable to examine WM processes in children and adolescents from 8 to 16 years of age. It is important to note that task difficulty was adapted to each participant’s WM capacity using a simple procedure (based on backward digit span performance, the participant completed the paradigm with two or three letters to remember), and we found no significant association between age or task performance and brain activation patterns. These findings indicate that our paradigm is well suited to examine brain systems associated with different WM capacities in different age groups. This may be an important advantage when comparing groups with different developmental trajectories, because previous studies show that WM-related activations may increase with age in parallel with changes in performance and improvements in WM capacity [12, 18].

Secondly, our imaging results demonstrate that, while distributed networks in frontal and visual areas activated in the context of the verbal WM paradigm used here (i.e. during the active conditions compared to the baseline), distinct neural substrates were selectively recruited during the encoding and retrieval periods. The verbal encoding period induced stronger activations in posterior and ventral brain regions, with large bilateral increases in occipital, as well as parahippocampal cortices. In contrast, the verbal retrieval period induced stronger activations in more anterior and dorsal regions, in particular in prefrontal and parietal areas, and to a lesser extent in lateral temporal areas.

The predominance of activity in visual cortex together with medial temporal lobe (parahippocampal gyrus) during encoding is consistent with the need to extract discriminative visual information from the to-be-remembered stimuli and store this information into short-term memory. On one hand, ventral occipito-temporal areas differentially engaged during encoding are crucial for perceptual shape analysis, especially for letters with a letter-sensitive activation in these regions [50, 51]. We did not find selective activations corresponding to the “visual word form area” but this region is typically responsive to letter-strings or words rather than isolated letters [52–54]. Moreover, we did not find language-related activation during verbal encoding, in particular Broca’s area which has been implicated in the subvocal rehearsal system [55]. However, language-related activation has been mainly found during encoding of words [56] and not during encoding of letters [57].On the other hand, the parahippocampal cortex is a key brain region at the interface between perception and memory, therefore likely to make an important contribution to efficient storage of visual information into WM [58].

As expected, predominant activity in frontal and parietal areas during retrieval is consistent with executive control and attentional focusing. The executive control system serves as an attention controller that allocates and coordinates attentional resources for cognitive tasks, such as retrieval of information encoded in working memory [59, 60]. Our findings accord with previous studies showing the involvement of frontal areas, especially prefrontal and anterior cingulate cortices, in the executive control required during WM demands [61, 62]. Focusing attention is crucial for efficient executive control [63] and recruits parietal regions [64], which were strongly implicated during the retrieval period in our study. In addition, WM retrieval of serial order is dissociable from the type of information contained in the item sequence [65] and also relies on activation in frontal and parietal activations [66].

Overall, our findings converge with those of van den Brosh and colleagues (2014), who reported a similar posterior and perceptual network during the encoding phase compared to a more anterior and executive network during the recall phase of a Sternberg item recognition paradigm (which did not include a distracting phase) in children and adolescents aged 9 to 19 years. However, these authors did not find any temporal or parahippocampal activations, possibly reflecting differences in the paradigm and material used (digits in their study vs. letters in ours). More generally, our findings of extensive fronto-parietal and visual activity during WM also dovetail with previous neuroimaging studies investigating brain systems associated with verbal WM in children and adolescents, across different kinds of verbal WM paradigms, such as the Steinberg item recognition task using letters [21, 67, 68] or n-back tasks using letters [69, 70].

Study hypotheses were supported by results revealing that brain activation patterns differ as a function of the nature of the concurrent task performed during the maintenance interval. Our design allowed us to compare the impact of within-domain (lexical decision task) versus cross-domain (face decision task) concurrent task processing during the maintenance period intervening between encoding and retrieval, while information stored in WM itself did not differ. A lexical decision task was expected to produce within-domain interference, as it involved verbal material resembling the to-be-remembered material (i.e. letters), while a face decision task was considered to induce cross-domain interference as it relied on non-verbal visual processes.

As predicted, the within-domain and cross-domain concurrent tasks evoked distinct brain activations when compared to each other. Localised and right-sided activations in the right frontal pole (Brodmann area 10) and medial fusiform gyrus were observed during the within-domain/lexical concurrent task, whereas the cross-domain/face concurrent task elicited much more distributed activations in occipital temporal extrastriate areas, but also left parahippocampal gyrus and fronto-parietal regions. These differences could not be attributed to task difficulty (since there were no significant difference in accuracy between the within-domain/lexical and the cross-domain/face decision task) but most likely reflect the different task demands and perhaps different strategies and processes applied during the maintenance interval. Since verbal information had to be held in WM, it might have produced stronger interference and greater conflict in resource allocation during the within-domain/lexical decision task than the cross-domain/face decision task, eventually affording less efficient engagement of task-specific networks in the former condition and hence lower accuracy. The involvement of the right frontal pole (Brodmann area 10), thought to organize an optimal use of cognitive resources and overcome potential impasses [71], may reflect this conflict in resource allocation and an increase in cognitive load during a verbal concurrent task. Such recruitment of attentional control mechanisms during interference appears consistent with the time-based resource-sharing model (TBRS; [72–74]). This model postulates the existence of attention-based mechanisms involved to maintain relevant verbal information when the capacity of the verbal-specific system (comparable to the phonological loop in Baddeley and Hitch’s model) is exceeded [75]. Alternatively, greater activation of visual and fronto-parietal areas as well as temporal regions, including parahippocampal gyrus, during the cross-domain/face decision task might reflect the dual process of face decision task and active maintenance of verbal information.

Critically, and in keeping with our hypotheses, the two concurrent tasks (within- and cross-domain) elicited distinct patterns of brain activity during the subsequent retrieval phase, despite the fact that identical stimuli were encoded, maintained and retrieved from WM. This indicates that partly different processes mediated retrieval after within- and cross-domain interference, and thus WM retrieval differed according to the nature of the preceding concurrent task. Large bilateral occipital activations were engaged during retrieval after the within-domain/lexical concurrent task, whereas only limited activity was observed in medial occipital cortex in addition to left superior and middle temporal cortex during retrieval after the cross-domain/face concurrent task. Interestingly, the latter cluster in temporal cortex overlapped with regions often reported in phonological tasks and associated with language networks [48, 49]. A plausible explanation for such difference would be that the maintenance of letters relied on a preferentially visual format when a concurrent verbal task had to be performed (i.e., within-domain concurrent task), hindering the use of the phonological loop for maintenance. On the other hand, the visual concurrent task may not prevent maintenance in the phonological loop, explaining a lesser involvement of visual cortex but conversely greater recruitment of language-related areas (left superior and middle temporal) during retrieval. These interpretations would accord with Baddeley and Hitch’s model previously mentioned, and the proposed effect of articulatory suppression on verbal WM [59, 76, 77].

The current study is not without limitations. The study sample size could be considered relatively small. We note, however, that it is comparable with previous studies exploring neural correlates of WM [12, 67, 78]. Even if our data showed no hint of any systematic modulation of brain activity patterns by age or retrieval accuracy, correlation and regression analysis performed here can be sensitive to small size. Nevertheless, by design, our procedure of tailoring task difficulty to each participant according to their WM capacity precisely aimed at avoiding age related effects and minimizing confounding effects due to individual differences in performance. We acknowledge that the lack of variability and the high retrieval accuracy resulting from this procedure may have limited the sensitivity of our study to activations modulated by age or other individual factors. Another limitation is that our paradigm did not test the reverse situation of verbal versus visual concurrent tasks on visual information held in WM. Examining both verbal and visuospatial WM in the presence of verbal and visuospatial interference could map more precisely how the different processes subserving verbal and visuospatial WM are influenced by different kinds of concurrent tasks.

Conclusions

Our study provides new insights into WM-related brain activity. We show a greater role of perceptual brain systems for encoding processes, and a fronto-parietal attentional network for retrieval processes. More critically, we show that a concurrent task during maintenance in WM produced distinct activations not only during the concurrent task itself, but also during subsequent retrieval. We conclude that the specific demands of the concurrent task affect the way memory items are maintained in WM, selective verbal interference resulting in greater reliance on visual cortex for retrieval, whereas visual interference leaves verbal systems of maintenance unaffected, hence resulting in the involvement of language-related areas in left temporal cortex for retrieval. These data accord with WM models postulating differentiated cognitive processes, with distinct neural substrates, according to the concurrent material interfering in verbal WM [59, 76, 77]. In addition we show that these activation patterns are robust across different ages and different WM capacities. More generally, our work validates a new WM paradigm derived from the Brown-Peterson task allowing us to probe for the neural correlates of different WM processes. Because the difficulty of the task was adapted to each participant and results were stable across age, this fMRI paradigm may be usefully applied in developmental populations with a wide age range and also feasible in clinical paediatric population (e.g., populations with mild intellectual difficulties).

Acknowledgments

This study was supported by Victorian Government’s Operational Infrastructure Support Program, the Murdoch Childrens Research Institute. VS was supported by the Swiss National Science Foundation Doc.CH scholarship. VA was supported by Australian National Health and Medical Research Council Senior Practitioner Fellowship. We gratefully thank the families who participated in this study.