Abstract

Many aspects of the complex relationship between working memory (WM) and long-term memory (LTM) remain unclear. Here, we manipulated task demands on a brief delayed-recognition paradigm to reveal behavioral and neural dissociations between these systems. Variations from a Baseline task included 3 challenges: increased delay duration, distraction during maintenance, and more closely matched memory probes, which were presented in behavioral experiments and during functional magnetic resonance imaging. Each of the challenges resulted in a significant decline in WM accuracy, and interestingly, a concurrent improvement in incidental LTM. Neural data revealed that, in task blocks, when participants anticipated, and then experienced, increased demands, they engaged medial temporal lobe (MTL) regions more during both the encoding and delay periods. Overall, these results indicate that distinct memory systems are recruited based on anticipated demands of a memory task, and MTL involvement underlies the observed dissociation between WM and LTM performance.

Introduction

Theories and models of memory have undergone many transformations since the seminal work of Atkinson and Shiffrin (1971), which described a framework in which all memories must pass through short-term memory in order to enter long-term stores. In addition, terminology for the storage of information over different time scales has been reconceptualized many times. Here, we consider the theoretical constructs of working memory (WM), defined as the temporary, short-term storage and/or manipulation of information over a brief delay (seconds) (Baddeley and Hitch 1974; Baddeley 2000), and long-term memory (LTM) as the consolidation of information for recall at a later time (beyond minutes). Studies involving brain injury have indicated that WM and LTM performance are often dissociable, such that damage to the medial temporal lobe (MTL), specifically the hippocampus, causes anterograde amnesia, a severe LTM deficit (Milner 1966), while WM may be intact in these individuals (Drachman and Arbit 1966; Milner 1966). The dissociation between WM and LTM has been supported by human neuroimaging research that has shown the MTL to be involved in LTM (Squire 1992), whereas the prefrontal cortex (PFC) is involved in the maintenance and manipulation processes of WM (D'Esposito et al. 2000; D'Esposito 2007). However, recent research has suggested that it is difficult to clearly delineate the boundary of memory subtypes, such as when WM ends and LTM begins, and how they interact [reviewed in Jonides et al. (2008)]. Specifically, there is accumulating evidence that the MTL is also engaged in WM tasks (Ranganath et al. 2003,, 2005; Ranganath and Blumenfeld 2005; Axmacher et al. 2007,, 2008) and the PFC is active during LTM encoding and retrieval (Ranganath et al. 2005; Blumenfeld and Ranganath 2007; Wais et al. 2010,, 2012). The finding that both the PFC and MTL are engaged in WM and LTM processes suggests that these 2 memory systems are not entirely independent of one another (Ranganath and Blumenfeld 2005).

If WM and LTM processes overlap, the question arises as to what factors mediate the dominant memory system that is engaged during an interaction with our environment when the goal is to retain information. One approach to assess this is to evaluate whether memory systems are differentially recruited to support performance when one of the systems is overloaded. Previous studies have revealed that there are several experimental manipulations of the delayed-recognition paradigm, frequently used to assess WM, which increase task demands and result in diminished WM performance. For example, WM accuracy declines when memory load is increased (Rissman et al. 2009), when delay periods are lengthened (Curtis and D'Esposito 2003), and when distraction is introduced (Clapp et al. 2010; Clapp and Gazzaley 2012). These task manipulations are interpreted as challenging WM maintenance processes required to actively hold information in mind. Results from 2 studies converge to suggest that such manipulation of task demands results in tradeoffs between PFC and MTL engagement (Rissman et al. 2008; Shrager et al. 2008). However, it has not yet been documented if task demands also drive a tradeoff in WM and LTM performance, and if this occurs in the setting of a concurrent neural dissociation.

The aim of the current behavioral and functional magnetic resonance imaging (fMRI) experiments was to assess the presence, and neural basis, of a dissociation between WM and incidental LTM performance in response to a series of “Challenge” manipulations of a delayed-recognition task. These were presented in separate blocks, so participants could anticipate the task demands prior to each trial, and involved: 1) Manipulating the length of delay between memory encoding and probe, 2) adding distraction during the delay period, and 3) increasing the difficulty of discrimination required at the memory probe by closely matching encoded and probed stimulus features (Fig. 1).

Figure 1.

fMRI experimental paradigm. Each experiment consisted of 2 blocks for each of the 5 tasks. The Passive View task (not displayed) was similar to the Baseline task, but displayed an arrow as the probe stimulus. Participants were instructed before each block which task they were going to perform. Note: an additional task (Short Delay, not displayed) was included in the behavioral study.

Figure 1.

fMRI experimental paradigm. Each experiment consisted of 2 blocks for each of the 5 tasks. The Passive View task (not displayed) was similar to the Baseline task, but displayed an arrow as the probe stimulus. Participants were instructed before each block which task they were going to perform. Note: an additional task (Short Delay, not displayed) was included in the behavioral study.

Materials and Methods

Study 1: Behavioral Experiments

Participants

Forty-three young healthy adults (ages 18–29, M = 22.5, 29 males) with normal or corrected-to-normal vision volunteered, gave consent, and were monetarily compensated to participate in the study. Two different behavioral experiments utilizing face delayed-recognition tasks were performed (unique participants in each experiment) to determine if manipulations of the demands of the task result in a tradeoff between WM and LTM performance. The first experiment involved a distraction manipulation and the second experiment involved delay period duration manipulation.

Stimuli

The stimuli consisted of grayscale images of faces and were novel across all tasks, across all blocks, and across all trials of the experiment. The stimuli consisted of a variety of neutral-expression male and female faces across a large age range. Hair and ears were removed digitally, and a blur was applied along the contours of the face as to remove any potential nonface-specific cues. All images were 225 pixels wide and 300 pixels tall (14 × 18 cm) and were presented foveally, subtending 3° of visual angle from fixation.

Tasks

Common across both experiments, participants were shown a face for 800 ms (encode), and after a delay (time period from offset of encode stimulus to onset of probe stimulus) they were shown a second face and then indicated with a button press whether it was a match or nonmatch (Fig. 1). Participants were instructed to respond as quickly as possible without sacrificing accuracy, yielding the measure of WM accuracy. The “Baseline” task in both experiments utilized a delay period of 7.2 s and serves a comparator for the manipulations. Each experiment consisted of 2 blocks of each task with 25 trials per block resulting in 50 trials per task. Blocks were presented pseudorandomly so that each task was run once before any task was run twice, and the 2 middle blocks were never the same. All task blocks were counterbalanced across participants. Participants were instructed before each block which task they were going to perform. About 60% of the trials were nonmatching. This distribution was selected to have a near 50% probability, yet highlighting the differences between the Baseline and Complementary Pairing tasks during the fMRI study (detailed below), which only differed based on nonmatched stimuli.

Experiment 1: Distraction Manipulation

In this experiment (20 participants), there were 2 tasks (4 total blocks). The first was the Baseline task and the other was a distraction task, which included an additional face presented for 1 s in the middle of the 7.2-s delay that was completely irrelevant to the delayed-recognition task. Participants were instructed to ignore the distracting face to the best of their ability without diverting their gaze or closing their eyes. They were instructed to foveate on the continuously present fixation cross in the center of the screen.

Experiment 2: Delay Duration Manipulation

In this experiment (23 participants), there were 3 tasks (6 total blocks), which varied the length of the delay period both shorter and longer than the duration of the delay in the Baseline task: delay periods were 3.2 s (short delay), 7.2 s (Baseline), or 18.2 s (long delay).

Incidental Long-Term Memory

Fifteen minutes after the completion of the delayed-recognition task, a surprise incidental LTM test was administered where participants viewed 30 encode face stimuli from each task along with the same number of novel face stimuli that were not presented during the primary experiment. No encode stimuli were included that were also a match during the main experiment (i.e., no stimulus in the post-experiment test was viewed more than once in the main experiment). All stimuli (both novel stimuli and stimuli from the experiment) were randomly ordered, and the participants were asked to give a confidence judgment reflecting the degree to which they believed each image was old or new (1 = definitely new, 2 = probably new, 3 = probably old, and 4 = definitely old). This section of the experiment was self-paced. The summary statistic used for this task took each participant's average rating for each task and subtracted their average rating for novel stimuli from each of the task averages, thus creating an LTM index that was independent of response biases.

Study 2: fMRI Experiment

Participants

Twenty-two young healthy adults (ages 18–31, M = 23.85, 11 males) with normal or corrected-to-normal vision volunteered, gave consent, and were monetarily compensated to participate in the study. Participants were prescreened to ensure that the sample did not include anyone with psychological or neurological disorders, brain injuries, or under the influence of medication known to affect cognitive state. Two participants were removed from the analyses due to significant drop-out of blood oxygen level-dependent (BOLD) signal.

Stimuli

Same as described in Study 1.

Tasks

Similar to Study 1, this experiment uses a delayed-recognition task involving grayscale face images, and includes the same general task manipulations of distraction and delay period duration manipulation. However, the short delay was not included because it would not allow adequate time for the BOLD response to summate. The final manipulation involved stimulus discriminability during recognition based on the similarity between encode and probe stimuli. The specifics of the tasks were as follows (Fig. 1).

Baseline

The delayed-recognition task that serves as the baseline had a delay period of 7.2 s and minimal steps were taken in matching the encoded and probe stimuli. We will refer to this as basic pairing, and involves matching race, gender, and age.

Long Delay

Basic pairings were used along with a long delay (18.2 s) in this task.

Distractor

Basic parings were used along with the Baseline delay duration (7.2 s), but the addition of an irrelevant, novel face appearing for 1 s in the middle of the delay.

Complementary Pairing

Closely paired encoded and probe images were used along with the Baseline delay (7.2 s) in this task. Complementary Pairings between the features of the encoded and probe stimuli were generated manually based on multiple parameters of the images including race, gender, age, head orientation, luminance, image resolution, and salient facial features. To ensure the Complementary Paired stimuli were more similar than the stimuli used in the Baseline task, an independent group of 5 healthy young adults (aged 23–37 years, M = 29.8, 3 males) assessed 40 paired stimuli that were randomly selected from the Baseline and Complementary Pairing tasks (20 from each task). On each trial, the 2 stimuli were presented sequentially for 1 s each with a 1-s interstimulus interval. Participants were instructed to respond to the second stimulus as to how similar it was to the first stimulus based on a 4-point Likert scale (1 = very dissimilar, 2 = somewhat dissimilar, 3 = somewhat similar, and 4 = very similar). All 5 participants verified that the Complementary Paired stimuli were assessed to be more similar (M = 2.94, SEM = 0.21) than the Baseline task stimuli (M = 1.87, SEM = 0.16; t(4) = 12.44, P < 0.0005).

Passive View

This task was included to serve as a metric for each participant's visual evoked response to viewing the stimuli. The stimulus and timing were similar to the Baseline task; however, the probe stimulus was a left or right pointing arrow indicating which button the participant should press. Participants were instructed to passively view the face stimuli without trying to remember them and respond to the probe arrow as quickly and accurately as possible.

The collective term for all tasks, except Baseline and Passive Views, will be referred to as “Challenge” tasks. Each experimental session comprised 2 blocks of the 5 tasks (15 trials/block; 10 total blocks; 30 total trials per task), pseudorandomly presented so that each task was run once before any task was run twice, and the 2 middle blocks were never the same. All task blocks were counterbalanced across participants. Participants were instructed before each block which task they were going to perform.

Incidental LTM

Fifteen minutes after completing the experimental tasks, a surprise LTM test was administered outside of the scanner as described in Study 1 with 144 faces, including 18 encode stimuli from each task except passive view (72 encode stimuli) and 72 novel stimuli.

fMRI Acquisition and Processing

All images were acquired on a Siemens 3-T Trio Magnetom. Images were collected with a 2-s time repetition (TR) and 1.78 × 1.78 × 3.5 mm voxel size. For our functional data, we collected 33 3.0 mm oblique axial T2*-weighted gradient-echo slices (TR = 2000 ms, 25 ms  time echo, 90° flip angle, and 250 mm2 field of view in a 128 × 128 matrix). Images were corrected for slice acquisition timing, motion artifacts, and spatially filtered using a 5-mm full-width at half-maximum Gaussian smoothing kernel. Data were analyzed using SPM5 in participant-native space. Group whole-brain maps were calculated from MNI-normalized data. In addition, high-resolution structural (T1-magnetization-prepared rapid gradient-echo) images were collected.

Data Analysis

fMRI Univariate Analysis

BOLD responses were modeled as epochs of neural activity by using 3 separate regressors for the encode, delay, and probe periods, which were convolved with the canonical hemodynamic response function (HRF). All epochs were modeled with the SPM5 canonical HRF convolved with a boxcar function with a duration of one TR. The onsets of temporally adjacent covariates were spaced at least 3.6 s apart to minimize the contamination of residual activity and auto-correlation (Zarahn et al. 1997). Thus, delay regressor onsets were placed at the midpoint of the delay period, whereas the encode and probe regressors were placed at the onset of the respective stimuli. Additionally, 6 motion regressors were included to control for possible participant movement (X, Y, Z, roll, pitch, yaw). All trials (correct and incorrect) were modeled with the same regressor, though trials where participants failed to respond to the probe were modeled with a separate regressor and excluded from the final analysis. Comparisons between combined Challenge tasks and Baseline were appropriately weighted to render a valid contrast (i.e., Long Delay + Distractor + Complementary Pairing tasks − 3 × Baseline task).

Region-of-Interest Localization

A separate functional localizer task was conducted prior to the main fMRI tasks, which was used to identify the fusiform face area (FFA), a face-selective area in the visual association cortex (Kanwisher et al. 1997). In this task, participants performed a 1-back task during 16-s blocks of face stimuli (5 blocks) and scene stimuli (5 blocks). Blocks of face and scene stimuli were separated by 16 s of rest and presented in an alternating order. Participants were instructed to indicate when a match (1-back) occurred within a block with a simple key press. Face and scene stimuli regressors consisted of a boxcar function extending across the 16-s block of stimuli and were convolved with the SPM5 canonical HRF. Face and scene stimuli regressors were then contrasted to generate SPM[T] images, and from these contrasts, region of interests (ROIs) were defined. A face-selective ROI (FFA) was then identified as the cluster of 35 contiguous voxels with the highest t-value within the right fusiform gyrus of each participant. The decision of the ROI voxel extent was based on the methodology of similar studies (Leppanen et al. 2004; Fischer et al. 2005; Clapp et al. 2010; Rota et al. 2009; Clapp and Gazzaley 2012) and was chosen in order to achieve a reasonable balance between regional specificity (diminished by the use of a larger cluster) and susceptibility to noise (a problem with smaller ROIs). The right FFA has been shown to be most strongly activated by faces (Bentin et al. 1996; Kanwisher et al. 1997).

fMRI Functional Connectivity Analysis

A new general linear model (GLM) design matrix was constructed in which each stage (encode, delay, and probe) of each trial was coded with a unique covariate, consistent with previous work (Clapp et al. 2010, 2011; Bollinger et al. 2010; Wais et al. 2010; Zanto et al. 2010; Chadick and Gazzaley 2011; Zanto et al. 2011). The placement and duration of the regressors were the same as that used in the univariate analysis. However, whole-brain maps of functional connectivity were generated by extracting mean β-values for each stage of every trial from each participant's FFA ROI (averaged over the ROI voxels) and correlating these β-values across all trials (correct and incorrect, except no response trials) with every voxel in a whole-brain analysis (Gazzaley et al. 2004; Rissman et al. 2004). This was done separately for each condition.

Previous work has suggested that, due to excessive suprathreshold voxels caused by local auto-correlations (Clapp et al. 2010) among other factors (such as vasculature or machine noise), there exists a possibility that the resultant correlation maps may not be amenable to parametric testing. To circumvent this issue, a nonparametric permutation test was employed to generate whole-brain contrast maps between each Challenge task and the Baseline task, separately. This involved conducting an iterative sign-inversion procedure that calculated significance over the course of 10 000 iterations (Nichols and Holmes 2002; Bollinger et al. 2010; Bollinger et al. 2011).

Our primary interest was to identify patterns that were consistent across like contrasts. To that end, Challenge > Baseline contrasts and Baseline > Challenge contrast maps were created by separately conjoining the component contrasts for each task stage (e.g., Challenge > Baseline = Long Delay > Baseline and Distractor > Baseline and Complementary Pairings > Baseline), in accordance with previously established conjunction methodology using Fisher's method (Fisher 1950; Lazar et al. 2002; Dolcos et al. 2008; Giovanello et al. 2010; St Jacques et al. 2010). Briefly, Fisher's method combines multiple test statistics (P-values) into one-test statistic using the formula: TF=2log(P), where P is a set of P-values with k elements, log denotes the natural logarithm, and TF reflects Fisher's conjoined test statistic that is compared with a χ2 distribution with 2k degrees of freedom. Each of the 6 (2 conjunctions × 3 task stages) statistical maps used for the conjunction analysis were subjected to a cluster extent-based correction for multiple comparisons. To match the thresholding procedure of the univariate analysis, the pooled significance threshold of the conjoined statistical maps was equal to or surpassed an elemental alpha of P < 0.005, computed using a modified version of Fisher's method of estimating conjoint significance. The modification entailed a requirement for each voxel of the component images to reach individual significance, thus ensuring that outliers would not skew the resulting conjunctions.

Correction for Multiple Comparisons

Where applicable, behavioral and neural data were corrected for multiple comparisons. Corrections for behavioral data utilized a false discovery rate method (Benjamini and Hochberg 1995), whereas neural data were corrected through a Monte Carlo simulation that was employed using AlphaSim from the AFNI toolbox (Cox 1996). Clusters were defined using the default linkage radius (nearest-neighbor, 6-connected voxel neighborhood) and a voxel-wise alpha of P ≤ 0.005. Corrections were mask-delimited based on a priori hypotheses regarding regional involvement derived from previous research (i.e., small volume correction) (Worsley et al. 1996; Brett et al. 2003; Larmar et al. 2004; Viard et al. 2007). Masks were created using the software package WFU_pickatlas (Lancaster et al. 2000; Maldjian et al. 2003). Specifically, masks of the MTL, PFC, and all noncerebellar gray matter (GM) were utilized. The MTL mask included bilateral hippocampus and parahippocampus (Supplementary Fig. 1A), whereas the PFC mask included all GM within and anterior to the precental gyrus (Supplementary Fig. 1B). The correction employing the GM mask will be referred to as a whole-brain correction since, for the purposes of fMRI data interpretation, our hypotheses and interest lay solely within this tissue class.

Results

Study 1: Behavioral Experiments

In Experiment 1, WM accuracy dropped significantly when participants were faced with distracting stimuli presented in the middle of the delay period relative to performance on a Baseline task (Baseline—96%, Distractor—93%; t(19) = 2.77, P < 0.05). In contrast, the incidental LTM test revealed that participants remembered the encoded stimuli better in the Distraction task (Baseline—0.013, Distractor—0.25; t(19) = 2.39, P < 0.05; Fig. 2A).

Figure 2.

Behavioral results. Study 1—Behavioral experiments: (A) Baseline versus Distraction, (B) Baseline versus Short and Long Delays. Study 2—fMRI experiment: (C) Baseline versus Distraction, Long Delay and Tight Pairing. Gray asterisks represent LTM comparisons with P < 0.05; black asterisks represent WM accuracy comparisons with P < 0.05.

Figure 2.

Behavioral results. Study 1—Behavioral experiments: (A) Baseline versus Distraction, (B) Baseline versus Short and Long Delays. Study 2—fMRI experiment: (C) Baseline versus Distraction, Long Delay and Tight Pairing. Gray asterisks represent LTM comparisons with P < 0.05; black asterisks represent WM accuracy comparisons with P < 0.05.

In Experiment 2, WM accuracy was influenced by the length of the delay period, such that participants did best on the short delay (91%), followed by the medium delay (Baseline, 88%), and then the long delay (83%) exhibited the least accurate WM performance (see Fig. 2B for individual comparisons: Short Delay > Medium Delay and Long Delay, Medium Delay > Long Delay; all 3 comparisons t(22) > 2.43, P < 0.05). Similarly, response times were fastest following the short delay (M = 819 ms, SEM = 53 ms), slower following the medium delay (M = 881 ms, SEM = 50 ms), and slowest after the long delay (M = 970 ms, SEM = 55 ms; all 3 comparisons t(22) > 4.76, P < 0.01). Thus, accuracy and response time data both show WM performance declines with an increasing delay period. However, incidental LTM performance showed the opposite pattern, such that encode stimuli were remembered best in the long delay task (0.51), followed by the medium delay task (0.41) and then the short delay task (0.29) exhibited the least accurate LTM performance (see Fig. 2B for individual comparisons: Long Delay > Medium Delay and Short Delay, Medium Delay > Short Delay; all 3 comparisons t(22) > 2.41, P < 0.05).

Taken together, these 2 experiments demonstrate a tradeoff between WM and LTM performance as a function of delayed-recognition task demands. With increasing demands, whether due to longer delay periods or the presence of a distraction, there was a decline in WM accuracy concomitant with an increase in LTM performance. Thus, these results provide novel evidence of a dissociation between WM and LTM performance based on task demands, and establish the utility of these task manipulations to examine the neural basis of this dissociation in Study 2.

Study 2: fMRI Experiment

Behavioral Performance

The WM performance results in the fMRI experiment revealed that, as predicted and consistent with Study 1, participants performed best on the Baseline task, such that accuracy for all Challenge tasks (Long-Delay, Distraction, and Complementary Pairing—a new challenge in Study 2) was significantly reduced relative to Baseline levels (i.e., WM accuracy during Baseline task > Long-Delay, Distraction, and Complementary Pairing; all 3 comparisons t(21) > 2.26, P < 0.05; Fig. 2C). Of note, there were no significant differences between any of the Challenge tasks (i.e., Long-Delay = Distraction = Complementary Pairing; all 3 comparisons t(21) < 1, P > 0.05). Similar to accuracy data, response times during the Baseline task (M = 1021 ms, SEM = 54 ms) were faster than those for all 3 Challenge tasks (all 3 comparisons t(21) > 2.45, P < 0.05). Moreover, no differences were observed in response times between the Complementary Pairing (M = 1066 ms, SEM = 48 ms) and Distraction (M = 1073 ms, SEM = 45 ms) tasks (t(21) < 1, P > 0.05), whereas the Long-Delay task (M = 1148 ms, SEM = 57 ms) resulted in slowed response times compared with all other tasks (all 3 comparisons t(21) > 2.45, P < 0.05). Although it is unclear why the Long Delay resulted in slowed responses compared with the other Challenge tasks, the response time and accuracy data both show that, compared with the Baseline task, WM performance declines when participants are challenged.

In contrast, incidental LTM of the encode faces presented in all the Challenge tasks was significantly better than that of the Baseline task (i.e., LTM accuracy during Baseline task < Long Delay, Distraction, and Complementary Pairing; all 3 comparisons t(21) > 2.14, P < 0.05). The only difference in LTM between Challenge tasks was that encoded stimuli from the Complementary Pairing task were recognized better than those from the Long-Delay task (t(21) = 2.89, P < 0.05). Although it is unclear why these 2 Challenge tasks exhibited a LTM difference and not a WM accuracy difference (unlike the behavioral studies), this may be a product of the distracting fMRI environment altering WM performance or possibly due to the differential sample sizes between the behavioral studies (50 trials) and the fMRI study (30 trials). It could be argued that different LTM performance between these Challenge tasks reflects different levels of task difficulty. However, the only goal of participants in performing these tasks was to remember information over the short term, and thus the assessment of WM accuracy, which did not differ between these Challenge tasks, suggests that they were balanced in terms of difficulty. Nonetheless, the behavioral results from the fMRI experiment were overall consistent with the memory performance results of the 2 behavioral experiments, confirming a dissociation between WM and LTM performance with shifting task demands.

Univariate fMRI

To investigate the neural underpinnings of the memory performance dissociation described above, we examined BOLD univariate measures from the encode and delay stage of the delayed-recognition tasks separately. Since the Challenge tasks were all different from Baseline and largely comparable in difficulty as indicated by performance measures, analysis of the neural data combined the 3 task manipulations and collectively compared them with the Baseline task (see Materials and Methods for details of our approach). We focused our analysis on the PFC and the MTL since these areas are well known to play an important role in WM and LTM processes (Ranganath et al. 2003; Ranganath and Blumenfeld 2005; Blumenfeld and Ranganath 2006; Axmacher et al. 2007, 2008; Berryhill and Olson 2008a, 2008b). Regions within these search areas with significantly different activity patterns for planned contrasts will be described below [multiple comparisons addressed using small volume corrections (Cox 1996); univariate encoding and delay data shown in Table 1]. For a complete list of brain regions that revealed differential activation between the Challenge and Baseline tasks from a whole-brain analysis, see Supplementary Tables.

Table 1

Univariate activity

Stage Contrast Lobe Region label BA Volume (mm3X Y Z 
Encode Challenge > Baseline Frontal Middle frontal gyrus 216 30 16 55 
Temporal Hippocampus 35 256 −23 −9 −21 
Parahippocampal gyrus 30 200 18 −8 −28 
Parahippocampal gyrus 30 104 20 −26 −14 
Baseline > Challenge Frontal Supplementary motor 23 224 −4 −9 50 
Frontal orbital gyrus 47 176 27 26 −16 
Insula 48 144 38 −8 −4 
Inferior frontal gyrus 47 128 −46 35 
Delay Challenge > Baseline Frontal Superior frontal gyrus 192 20 24 58 
Inferior frontal orbital 47 144 45 35 −9 
Middle frontal orbital 11 120 −31 53 −2 
Precentral gyrus 120 57 43 
Baseline > Challenge Frontal Anterior cingulate gyrus 24 336 10 40 
Precentral gyrus 216 −41 −15 62 
Precentral gyrus 128 −25 −10 60 
Anterior cingulate gyrus 24 120 46 
Middle frontal gyrus 46 120 −33 34 39 
Stage Contrast Lobe Region label BA Volume (mm3X Y Z 
Encode Challenge > Baseline Frontal Middle frontal gyrus 216 30 16 55 
Temporal Hippocampus 35 256 −23 −9 −21 
Parahippocampal gyrus 30 200 18 −8 −28 
Parahippocampal gyrus 30 104 20 −26 −14 
Baseline > Challenge Frontal Supplementary motor 23 224 −4 −9 50 
Frontal orbital gyrus 47 176 27 26 −16 
Insula 48 144 38 −8 −4 
Inferior frontal gyrus 47 128 −46 35 
Delay Challenge > Baseline Frontal Superior frontal gyrus 192 20 24 58 
Inferior frontal orbital 47 144 45 35 −9 
Middle frontal orbital 11 120 −31 53 −2 
Precentral gyrus 120 57 43 
Baseline > Challenge Frontal Anterior cingulate gyrus 24 336 10 40 
Precentral gyrus 216 −41 −15 62 
Precentral gyrus 128 −25 −10 60 
Anterior cingulate gyrus 24 120 46 
Middle frontal gyrus 46 120 −33 34 39 
Encode Period Activity

During the encode period, we observed widespread differential activation throughout the brain between the Baseline and Challenge tasks (Supplementary Table 1). Importantly, greater MTL activation (specifically, in the left hippocampus; −23 −9 −21 MNI and right parahippocampus; 19 −25 10 MNI) was observed for the Challenge tasks in comparison with the Baseline task (Fig. 3A, P < 0.05). In addition, a region within the PFC, the right middle frontal gyrus (MFG; 30 16 55 MNI), also exhibited greater activity in the Challenge tasks compared with the Baseline task (P < 0.05). In the reverse comparison, the inferior frontal gyrus showed more activation in the Baseline task compared with the Challenge tasks (P < 0.05), whereas no areas of the MTL exhibited this pattern. Of note, the MTL was significantly more active during the Baseline task compared with either a passive view task or the intrinsic GLM baseline (P < 0.05).

Figure 3.

Univariate BOLD activity. (A) During the encode period, MTL activity was increased in the Challenge tasks compared with the Baseline task (yellow box). (B) During the delay period, Challenge and Baseline tasks utilize differential PFC regions (Table 1).

Figure 3.

Univariate BOLD activity. (A) During the encode period, MTL activity was increased in the Challenge tasks compared with the Baseline task (yellow box). (B) During the delay period, Challenge and Baseline tasks utilize differential PFC regions (Table 1).

Delay Period Activity

Similar to encode period activity, we observed widespread differential activation throughout the brain between the Baseline and Challenge tasks during the delay period (Supplementary Table 2). Notably, there were several areas in both the MTL and PFC that exhibited more activation in the Challenge tasks compared with the Baseline (Table 1). In assessing areas that were more active in the Baseline task, PFC regions in the MFG, precentral, and anterior cingulate gyrus (P < 0.05) were more active compared with the Challenge tasks, with no MTL regions exhibiting greater Baseline activity (Fig. 3B). Of note, activity in the parietal cortex exhibited a similar relationship as the PFC, such that the Challenge and Baseline tasks displayed differential activity profiles that did not clearly dissociate with regard to task demands (Supplementary Table 2).

In summary, during the encoding and maintenance processing stages of the delayed-recognition tasks, several PFC regions were more active during the Baseline task, whereas others were more active during the Challenge tasks, but the MTL was more active only during Challenge tasks versus Baseline (i.e., hippocampus in encoding, and parahippocampal gyrus in encoding and delay).

fMRI Functional Connectivity

To assess network differences associated with face processing across tasks, we used the beta-series functional connectivity approach (Gazzaley et al. 2004; Rissman et al. 2004) to compute MTL and PFC functional connectivity with a face-selective visual cortical seed region (i.e., FFA; Kanwisher et al. 1997). During the encode period, we observed widespread functional connectivity between the FFA and the rest of the brain, which differed between the Baseline and Challenge tasks (Supplementary Table 3). Importantly, FFA functional connectivity with multiple subregions of the MTL (i.e., parahippocampal gyrus) was greater for Challenge tasks compared with Baseline (P < 0.05, corrected; Fig. 4A), whereas there were no MTL regions that exhibited greater functional connectivity with the FFA in the Baseline task (Table 2). Additionally, FFA functional connectivity with the PFC was more widespread during Challenge tasks (10 PFC regions exhibited Challenge > Baseline, see Table 2) compared with the Baseline task (3 PFC regions exhibited Baseline > Challenge, see Table 2). Interrogation of the delay period revealed a similar pattern, such that functional connectivity between the FFA and the rest of the brain was widespread and differed between the Baseline and Challenge tasks (Supplementary Table 4). Notably, functional connectivity between the FFA and MTL regions was greater in the Challenge tasks, whereas there were no MTL regions that exhibited greater functional connectivity with the FFA in the Baseline task (Fig. 4B). Additionally, FFA functional connectivity with the PFC was more widespread during Challenge tasks (12 PFC regions exhibited Challenge > Baseline, see Table 2) compared with the Baseline task (2 PFC regions exhibited Baseline > Challenge, see Table 2). Of note, functional connectivity between the FFA and parietal cortex exhibited a similar relationship as the PFC, such that the Challenge and Baseline tasks displayed differential connectivity profiles (Supplementary Tables 3 and 4).

Table 2

Functional connectivity

Stage Contrast Lobe Region label BA Volume (mm3X Y Z 
Encode Challenge > Baseline Frontal Anterior cingulate gyrus 24/32 1568 −7 11 40 
Supplementary motor area 968 11 55 
Superior frontal gyrus 488 11 −4 67 
Precentral gyrus 432 −27 61 
Precentral gyrus 4/6 400 19 −18 67 
Superior frontal gyrus 288 −18 68 
Middle frontal gyrus 6/9 272 −37 18 58 
Precentral gyrus 9/44 272 48 29 
Precentral gyrus 240 −53 38 
Precentral gyrus 9/48 216 −63 12 
Temporal Parahippocampal gyrus 36 1008 27 −28 
Parahippocampal gyrus 30 968 −28 −20 −22 
Parahippocampal gyrus 20 688 26 −18 −23 
Baseline > Challenge Frontal Subcallosal gyrus 11 296 28 −12 
Superior frontal gyrus 240 33 65 
Insula 13/48 216 38 10 
Delay Challenge > Baseline Frontal Precentral gyrus 6968 34 −2 52 
Medial frontal gyrus 1288 −4 −15 58 
Precentral gyrus 3/4 712 44 −14 49 
Superior frontal gyrus 672 −1 15 58 
Inferior frontal orbital 47 480 −48 22 −1 
Anterior cingulate 24 384 −3 41 
Central opercular 48 368 −49 −2 
Superior frontal gyrus 10/46 320 33 60 21 
Anterior cingulate 24 312 −9 14 36 
Middle frontal gyrus 280 32 64 
Superior frontal gyrus 232 −10 69 
Paracingulate gyrus 24/32 208 −9 13 50 
Temporal Hippocampus 35 1064 −26 −20 −22 
Parahippocampal gyrus 30 640 24 −4 −36 
Parahippocampal gyrus 30 600 −24 −2 −28 
Parahippocampal gyrus 30 408 22 −22 −28 
Baseline > Challenge Frontal Precentral gyrus 48 448 −45 −8 29 
Middle frontal gyrus 46 208 −39 27 43 
Stage Contrast Lobe Region label BA Volume (mm3X Y Z 
Encode Challenge > Baseline Frontal Anterior cingulate gyrus 24/32 1568 −7 11 40 
Supplementary motor area 968 11 55 
Superior frontal gyrus 488 11 −4 67 
Precentral gyrus 432 −27 61 
Precentral gyrus 4/6 400 19 −18 67 
Superior frontal gyrus 288 −18 68 
Middle frontal gyrus 6/9 272 −37 18 58 
Precentral gyrus 9/44 272 48 29 
Precentral gyrus 240 −53 38 
Precentral gyrus 9/48 216 −63 12 
Temporal Parahippocampal gyrus 36 1008 27 −28 
Parahippocampal gyrus 30 968 −28 −20 −22 
Parahippocampal gyrus 20 688 26 −18 −23 
Baseline > Challenge Frontal Subcallosal gyrus 11 296 28 −12 
Superior frontal gyrus 240 33 65 
Insula 13/48 216 38 10 
Delay Challenge > Baseline Frontal Precentral gyrus 6968 34 −2 52 
Medial frontal gyrus 1288 −4 −15 58 
Precentral gyrus 3/4 712 44 −14 49 
Superior frontal gyrus 672 −1 15 58 
Inferior frontal orbital 47 480 −48 22 −1 
Anterior cingulate 24 384 −3 41 
Central opercular 48 368 −49 −2 
Superior frontal gyrus 10/46 320 33 60 21 
Anterior cingulate 24 312 −9 14 36 
Middle frontal gyrus 280 32 64 
Superior frontal gyrus 232 −10 69 
Paracingulate gyrus 24/32 208 −9 13 50 
Temporal Hippocampus 35 1064 −26 −20 −22 
Parahippocampal gyrus 30 640 24 −4 −36 
Parahippocampal gyrus 30 600 −24 −2 −28 
Parahippocampal gyrus 30 408 22 −22 −28 
Baseline > Challenge Frontal Precentral gyrus 48 448 −45 −8 29 
Middle frontal gyrus 46 208 −39 27 43 
Figure 4.

Functional connectivity data. Functional connectivity between the FFA and MTL increased in the Challenge tasks compared with the Baseline task during (A) the encoding period as well as during (B) the delay period (yellow boxes). Challenge tasks and Baseline task exhibit differential functional connectivity between FFA and PFC, although much more widespread during Challenge tasks (Table 2).

Figure 4.

Functional connectivity data. Functional connectivity between the FFA and MTL increased in the Challenge tasks compared with the Baseline task during (A) the encoding period as well as during (B) the delay period (yellow boxes). Challenge tasks and Baseline task exhibit differential functional connectivity between FFA and PFC, although much more widespread during Challenge tasks (Table 2).

To explore a neural–behavioral relationship, MTL regions that exhibited significant univariate activity or functional connectivity to the FFA were correlated with LTM scores across participants. However, neither the univariate data nor the functional connectivity data from the MTL exhibited a significant correlation with LTM performance. Therefore, it is likely that other factors in addition to MTL recruitment also contribute to the change in LTM performance based on the expected task demands.

Overall, the functional connectivity data revealed the same pattern as the univariate data, and support the conclusion that the dissociation between WM and LTM was associated with greater recruitment of MTL networks in the encoding and delay periods of delayed-recognition tasks in the setting of increased task demands.

Discussion

This study provides neural and behavioral evidence of a dissociation between the engagement of WM and LTM systems based on the demands of a brief duration memory task. Specifically, we demonstrate and replicate using several manipulations of a face delayed-recognition task that increased demands (i.e., longer delay period, distraction during delay period, and similarity of encode and probe stimuli) all result in lower WM accuracy, but higher LTM recognition of the same stimuli (a push-pull pattern was documented in the behavioral experiments). These results indicate that distinct memory systems are differentially involved when individuals engage in delayed-recognition tasks of varying difficulty. It is important to note that task manipulations were presented in blocks, allowing participants to anticipate the challenge of each task prior to their engagement (rather than unexpectantly being presented with a distractor, long delay, or a difficult pairing). The data suggest that expectations of task demands in an experimental block underlie our findings. This is most clearly highlighted by noting that the encode stage was perceptually identical for all tasks and so the only factor that differed during encoding between tasks was the participant's expectation of what they would encounter later in the trial. Therefore, it could only be expectation of task demands that drove the increased MTL recruitment during the encode stage of the Challenge tasks.

Our interpretation of these findings is that the anticipation of increased difficulty in accomplishing WM task goals leads to a shift in the engaged neural circuitry in an attempt to preserve high-level performance. This shift, however, is clearly not entirely effective in maintaining WM performance, as documented by diminished WM accuracy in the challenge conditions, but it does result in the unintended consequence of stronger LTM encoding (note: unintended in that there were no LTM goals). The fMRI results support this interpretation and inform the mechanistic basis of the behavioral dissociation by revealing that MTL regions exhibit increased activity and functional connectivity with visual cortical areas during both the encoding and delay periods of the more challenging tasks relative to the Baseline task. This finding is consistent with the well-documented role of the MTL in LTM consolidation processes (Alvarez and Squire 1994). Moreover, these findings support research indicating that hippocampus-dependent WM is beneficial for LTM encoding (Axmacher et al. 2009) and that hippocampal activity prior to stimulus presentation predicts subsequent LTM performance (Park and Rugg 2010). Thus, this study reveals that expectations of difficulty in performing a brief duration memory task influence the neural networks engaged, resulting in a dissociation between WM and LTM performance.

We observe that the MTL is more active and exhibits greater functional connectivity with stimulus-selective visual cortical areas in the Challenge tasks during encoding and delay periods, with no examples of the reverse pattern (see Figs 3 and 4, yellow boxes). This dissociation suggests MTL engagement as a putative underlying neural correlate of the observed behavioral dissociation. In contrast, PFC activity and functional connectivity does not clearly dissociate across tasks, that is, different regions of the PFC are more engaged by both Challenge and Baseline tasks. And so, we do not have a clear neural indicator of disrupted PFC-mediated WM maintenance processes in the Challenge tasks, as we hypothesized. We suspect this is due to different PFC subregions being engaged by the many control processes demanded by these tasks, thus masking simple interpretations of distinct contributions of PFC regions to WM processes. This complexity may help explain why the FFA exhibited more widespread functional connectivity to the PFC during the Challenge tasks. Additionally, consistent with the parietal lobe previously being shown to be involved in both WM and LTM (Berryhill and Olson 2008a, 2008b; Uncapher and Wagner 2009; Cabeza et al. 2011, 2012; Uncapher et al. 2011;), the pattern of activation in these regions also did not consistently dissociate with regard to task demands.

The data suggest that the MTL was more engaged when simple maintenance was not anticipated to be sufficient for task performance, as in the Challenge scenarios when delay durations were long, when distraction interrupted the WM maintenance period, and when there was a need for high levels of details of a stored representation. This raises the question as to why the hippocampus was engaged in the setting of increased WM demands. One possibility is suggested by previous findings that show the hippocampus to be critical in forming relationships between items and representation binding (Cohen and Eichenbaum 1993; Cohen et al. 1997; Davachi and Wagner 2002; Preston et al. 2004). Perhaps, when the opportunity for rote maintenance (e.g., visualization) is diminished, the act of encoding associations between face features (e.g., the spatial relations between eyes and nose of a face) makes subsequent recognition of the face probe more effective (Morris et al. 1977). If participants were driven to store relational data in an attempt to optimize their memory in the setting of these challenges (even if subconsciously), then MTL engagement used to accomplish this act might lead to stronger subsequent LTM as an unintended consequence, as there were no LTM goals. This is not to say that the LTM system was accidentally engaged or that it does not operate independent of the WM system. Rather, knowledge of increased task difficulty resulted in a shift in the engaged neural circuitry in an attempt to preserve high-level performance. Although the current data cannot speak to whether this was a conscious effort to “try harder” during the Challenge tasks, it is certainly plausible. Additionally, it should be noted that these data do not support the idea of an antagonistic relationship between WM and LTM, as decreased WM performance was not correlated with increased LTM performance across participants.

The neural results of the current study are in contrast with findings that MTL was equally active during episodic retrieval and WM tasks (Cabeza et al. 2002), suggesting a similar role for WM and LTM processes. Yet, our data corroborate other research showing increased MTL activity when the WM maintenance period is elongated, resulting in lower WM accuracy (Talmi et al. 2005). The discrepancy between these studies may lie with task difficulty, in that Cabeza et al. (2002) observed >90% accuracy on both WM and episodic retrieval, whereas Talmi et al. (2005) observed <75% accuracy for long WM delays and >90% accuracy for short WM delays. Notably, our current results do show elevated MTL activity during the Baseline task compared with rest; however, MTL activity was increased during Challenge tasks compared with the Baseline task concomitant with decreased WM and increased LTM performance. Thus, task demands, or anticipation thereof, appear to alter the balance of how WM and LTM systems are utilized and may help reconcile disparate results regarding the role of MTL in WM and LTM processes.

The current results are also consistent with evidence from patient populations that show that the MTL is necessary in situations when simple memory maintenance is not possible. Shrager et al. (2008) investigated WM performance in both controls and amnesics and found that both groups were able to maintain face stimuli over delays of 2 and 7 s. However, when the duration of the delay increased to 14 s, the amnesics exhibited a much larger drop in WM performance. These results were interpreted as revealing that amnesics were able to rely on non-MTL-dependent active maintenance processes at short delays, and that there was a reliance on the MTL system in the longer delay tasks. This is entirely consistent with our findings that increased duration of the delay drives reliance on the MTL. Other research with amnesics also supports our results that longer delay durations elicit greater reliance on the MTL (Ranganath and Blumenfeld 2005).

Of interest, another study using a face delayed-recognition task revealed that amnesics were impaired in memory performance when delays were only 7 s (Nichols et al. 2006). Moreover, it was noted by the authors that the stimuli “could not be solved using easily verbalizable features.” This description is reminiscent of the rigorous manner in which we manually paired stimuli in the Complementary Pairings task, which exhibited increased engagement of the MTL over baseline. Additionally, Nichols et al. (2006) reported that no WM differences existed between amnesics and controls with a 1-s delay interval. They also postulated that LTM processes are more relevant at the longer delay. This statement is consistent with our hypothesis that increasing difficulty by manipulating delay duration results in a situation in which rote maintenance is not sufficient and thus the MTL is recruited. Similarly, another study found that increasing demands in a face delayed-recognition task by increasing the memory load resulted in decreased WM performance and an increase in functional connectivity between the MTL and the visual association cortex (i.e., FFA) during the delay period (Rissman et al. 2008); yet another manipulation that increased delayed-recognition task demands and resulted in MTL network engagement.

It is possible that increased attentional demands increased vasodilation, thereby influencing measures of functional connectivity. This could reflect an increase in arousal and help explain why more neural regions exhibited functional connectivity to the FFA during the Challenge tasks. However, it is unlikely that this reflects a general (non-specific) increase in arousal, as many regions exhibited greater functional connectivity during the Baseline task and even more regions did not dissociate between the task demands. Thus, mean FFA BOLD activity changes (and corresponding vascular state) did not influence functional connectivity measures throughout the brain in a homogeneous manner, as would be expected from a general increase in arousal with a global increase in vasodilation.

It could be argued that, due to the close proximity of the GLM regressors for the encoding and delay periods, auto-correlation could bias the results to be similar across the 2 task stages. Thus, we chose not to directly contrast the encoding and delay periods, nor do we draw strong conclusions about specific neural regions and their role in distinct task stages. Nonetheless, we report data from both the encoding and delay periods because the results showed minimal overlap between task stages in terms of the regions identified as being significantly modulated by task demands. Specifically, univariate BOLD activity exhibited <1% overlap between regions listed in Supplementary Table 1 (encoding activity) and Supplementary Table 2 (delay activity). Similarly, functional connectivity data exhibited only 5% overlap between regions listed in Supplementary Table 3 (encoding) and Supplementary Table 4 (delay). Thus, any confounds introduced by auto-correlations were negligible and we consider results from the 2 task stages as supporting our overall conclusions.

In the current study, manipulations of task demands drive how different memory systems are engaged, and this in turn differentially influences WM and LTM performance. Our results indicate that when participants anticipate a challenging task in which simple WM maintenance is not possible or sufficient, the MTL is more engaged during encoding and maintenance periods. This neural pattern serves to explain the behavioral findings of enhanced LTM in the setting of a task that exhibits diminished WM accuracy. These findings offer another level to Baddeley's revised theory of memory that introduced an episodic buffer, referred to as transitive memory, which aids in the crystallization of items from WM into LTM (Baddeley 2000). Our results suggest that engagement of the episodic buffer is weighted by task demands, such that increased difficulty in active WM maintenance results in the episodic buffer shunting information to an LTM system.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/

Funding

This project was supported by the National Institutes of Health Grant R01-AG30395 (A.G.) and a University of California Office of the President Presidential Postdoctoral Fellowship (W.C.C.).

Notes

Conflict of Interest: None declared.

References

Alvarez
P
Squire
LR
.
1994
.
Memory consolidation and the medial temporal lobe: a simple network model
.
Proc Natl Acad Sci USA
 .
91
:
7041
7045
.
Atkinson
RC
Shiffrin
RM
.
1971
.
The control of short-term memory
.
Sci Am
 .
225
:
82
90
.
Axmacher
N
Elger
CE
Fell
J
.
2009
.
Working memory-related hippocampal deactivation interferes with long-term memory formation
.
J Neurosci
 .
29
:
1052
1060
.
Axmacher
N
Mormann
F
Fernandez
G
Cohen
MX
Elger
CE
Fell
J
.
2007
.
Sustained neural activity patterns during working memory in the human medial temporal lobe
.
J Neurosci
 .
27
:
7807
7816
.
Axmacher
N
Schmitz
DP
Weinreich
I
Elger
CE
Fell
J
.
2008
.
Interaction of working memory and long-term memory in the medial temporal lobe
.
Cereb Cortex
 .
18
:
2868
2878
.
Baddeley
A
.
2000
.
The episodic buffer: a new component of working memory?
Trends Cogn Sci
 .
4
:
417
423
.
Baddeley
AD
Hitch
GJ
.
1974
.
Working memory
. In:
Bower
GA
, editor.
Recent advances in learning and motivation
 .
New York
:
Academic
, p.
47
89
.
Benjamini
Y
Hochberg
Y
.
1995
.
Controlling the false discovery rate—a practical and powerful approach to multiple testing
.
J R Stat Soc Series B Methodol
 .
57
:
289
300
.
Bentin
S
Allison
T
Puce
A
Perez
E
McCarthy
G
.
1996
.
Electrophysiological studies of face perception in humans
.
J Cogn Neurosci
 .
8
:
551
565
.
Berryhill
ME
Olson
IR
.
2008a
.
Is the posterior parietal lobe involved in working memory retrieval? Evidence from patients with bilateral parietal lobe damage
.
Neuropsychologia
 .
46
:
1775
1786
.
Berryhill
ME
Olson
IR
.
2008b
.
The right parietal lobe is critical for visual working memory
.
Neuropsychologia
 .
46
:
1767
1774
.
Blumenfeld
RS
Ranganath
C
.
2007
.
Prefrontal cortex and long-term memory encoding: an integrative review of findings from neuropsychology and neuroimaging
.
Neuroscientist
 .
13
:
280
291
.
Blumenfeld
RS
Ranganath
C
.
2006
.
Prefrontal cortex promotes long-term memory formation through its role in working memory
.
J Neurosci
 .
26
:
916
925
.
Bollinger
J
Rubens
MT
Masangkay
E
Kalkstein
J
Gazzaley
A
.
2011
.
An expectation-based memory deficit in aging
.
Neuropsychologia
 .
49
:
1466
1475
.
Bollinger
J
Rubens
MT
Zanto
TP
Gazzaley
A
.
2010
.
Expectation-driven changes in cortical functional connectivity influence working memory and long-term memory performance
.
J Neurosci
 .
30
:
14399
14410
.
Brett
M
Penny
W
Kiebel
S
.
2003
.
Introduction to random field theory
. In:
Frackowiak
SJ
Friston
KJ
Dolan
RJ
Frith
CD
Price
CJ
Ashburner
JT
Zeki
S
Penny
WD
, editors.
Human brain function
 .
London
:
Academic Press
.
Cabeza
R
Ciaramelli
E
Moscovitch
M
.
2012
.
Cognitive contributions of the ventral parietal cortex: an integrative theoretical account
.
Trends Cogn Sci
 .
16
:
338
352
.
Cabeza
R
Dolcos
F
Graham
R
Nyberg
L
.
2002
.
Similarities and differences in the neural correlates of episodic memory retreival and working memory
.
NeuroImage
 .
16
:
317
330
.
Cabeza
R
Mazuz
YS
Stokes
J
Kragel
JE
Woldorff
MG
Ciaramelli
E
Olson
IR
Moscovitch
M
.
2011
.
Overlapping parietal activity in memory and perception: evidence for the attention to memory model
.
J Cogn Neurosci
 .
23
:
3209
3217
.
Chadick
JZ
Gazzaley
A
.
2011
.
Differential coupling of visual cortex with default or frontal-parietal network based on goals
.
Nat Neurosci
 .
14
:
830
832
.
Clapp
WC
Gazzaley
A
.
2012
.
Distinct mechanisms for the impact of distraction and interruption on working memory in aging
.
Neurobiol Aging
 .
33
:
134
148
.
Clapp
WC
Rubens
MT
Gazzaley
A
.
2010
.
Mechanisms of working memory disruption by external interference
.
Cereb Cortex
 .
20
:
859
872
.
Clapp
WC
Rubens
MT
Sabharwal
J
Gazzaley
A
.
2011
.
Deficit in switching between functional brain networks underlies the impact of multitasking on working memory in older adults
.
Proc Natl Acad Sci USA
 .
108
:
7212
7217
.
Cohen
NJ
Eichenbaum
H
.
1993
.
Memory, amnesia, and the hippocampal system
 .
Cambridge
(
MA
):
MIT Press
.
Cohen
NJ
Poldrack
RA
Eichenbaum
H
.
1997
.
Memory for items and memory for relations in the procedural/declarative memory framework
.
Memory
 .
5
:
131
178
.
Cox
RW
.
1996
.
AFNI: software for analysis and visualization of functional magnetic resonance neuroimages
.
Comput Biomed Res
 .
29
:
162
173
.
Curtis
CE
D'Esposito
M
.
2003
.
Persistent activity in the prefrontal cortex during working memory
.
Trends Cogn Sci
 .
7
:
415
423
.
D'Esposito
M
.
2007
.
From cognitive to neural models of working memory
.
Philos Trans R Soc Lond B Biol Sci
 .
362
:
761
772
.
D'Esposito
M
Postle
BR
Rypma
B
.
2000
.
Prefrontal cortical contributions to working memory: evidence from event-related fMRI studies
.
Exp Brain Res
 .
133
:
3
11
.
Davachi
L
Wagner
AD
.
2002
.
Hippocampal contributions to episodic encoding: insights from relational and item-based learning
.
J Neurophysiol
 .
88
:
982
990
.
Dolcos
F
Diaz-Granados
P
Wang
L
Mccarthy
G
.
2008
.
Opposing influences of emotional and non-emotional distracters upon sustained prefrontal cortex activity during a delayed-response working memory task
.
Neuropsychologia
 .
46
:
326
335
.
Drachman
DA
Arbit
J
.
1966
.
Memory and the hippocampal complex. II. Is memory a multiple process?
Arch Neurol
 .
15
:
52
61
.
Fischer
H
Sandblom
J
Gavazzeni
J
Fransson
P
Wright
CI
Backman
L
.
2005
.
Age-differential patterns of brain activation during perception of angry faces
.
Neurosci Lett
 .
386
:
99
104
.
Fisher
RA
.
1950
.
Statistical methods for research workers
 .
London
:
Oliver and Boyd
.
Gazzaley
A
Rissman
J
D'Esposito
M
.
2004
.
Functional connectivity during working memory maintenance
.
Cogn Affect Behav Neurosci
 .
4
:
580
599
.
Giovanello
KS
Kensinger
EA
Wong
AT
Schacter
DL
.
2010
.
Age-related neural changes during memory conjunction errors
.
J Cogn Neurosci
 .
22
:
1348
1361
.
Jonides
J
Lewis
RL
Nee
DE
Lustig
CA
Berman
MG
Moore
KS
.
2008
.
The mind and brain of short term memory
.
Annu Rev Psychol
 .
59
:
193
224
.
Kanwisher
N
McDermott
J
Chun
MM
.
1997
.
The fusiform face area: a module in human extrastriate cortex specialized for face perception
.
J Neurosci
 .
17
:
4302
4311
.
Lancaster
JL
Woldorff
MG
Parsons
LM
Liotti
M
Freitas
CS
Rainey
L
Kochunov
PV
Nickerson
D
Mikiten
SA
Fox
PT
.
2000
.
Automated Talairach atlas labels for functional brain mapping
.
Hum Brain Mapp
 .
10
:
120
131
.
Larmar
M
Yousem
DM
SM
R
.
2004
.
Age differences in orbitofrontal activation: an fMRI investigation of delayed match and nonmatch to sample
.
Neuroimage
 .
21
:
1368
1376
.
Lazar
NA
Luna
B
Sweeney
JA
Eddy
WF
.
2002
.
Combining brains: a survey of methods for statistical pooling of information
.
Neuroimage
 .
16
:
538
550
.
Leppanen
JM
Milders
M
Bell
JS
Terriere
E
Hietanen
JK
.
2004
.
Depression biases the recognition of emotionally neutral faces
.
Psychiatry Res
 .
128
:
123
133
.
Maldjian
JA
Laurienti
PJ
Kraft
RA
Burdette
JH
.
2003
.
An automated method for neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets
.
Neuroimage
 .
19
:
1233
1239
.
Milner
B
.
1966
.
Amnesia following operation on the temporal lobes
 .
London
:
Butterworths
.
Morris
CD
Bransford
JD
Franks
JJ
.
1977
.
Levels of processing versus transfer appropriate processing
.
J Verbal Learn Verbal Behav
 .
16
:
519
533
.
Nichols
EA
Kao
YC
Verfaellie
M
Gabrieli
JD
.
2006
.
Working memory and long-term memory for faces: evidence from fMRI and global amnesia for involvement of the medial temporal lobes
.
Hippocampus
 .
16
:
604
616
.
Nichols
TE
Holmes
AP
.
2002
.
Nonparametric permutation tests for functional neuroimaging: a primer with examples
.
Hum Brain Mapp
 .
15
:
1
25
.
Park
H
Rugg
MD
.
2010
.
Prestimulus hippocampal activity predicts later recollection
.
Hippocampus
 .
20
:
24
28
.
Preston
AR
Shrager
Y
Dudukovic
NM
Gabrieli
JD
.
2004
.
Hippocampal contribution to the novel use of relational information in declarative memory
.
Hippocampus
 .
14
:
148
152
.
Ranganath
C
Blumenfeld
RS
.
2005
.
Doubts about double dissociations between short- and long-term memory
.
Trends Cogn Sci
 .
9
:
374
380
.
Ranganath
C
Cohen
MX
Brozinsky
CJ
.
2005
.
Working memory maintenance contributes to long-term memory formation: neural and behavioral evidence
.
J Cogn Neurosci
 .
17
:
994
1010
.
Ranganath
C
Johnson
MK
D'Esposito
M
.
2003
.
Prefrontal activity associated with working memory and episodic long-term memory
.
Neuropsychologia
 .
41
:
378
389
.
Rissman
J
Gazzaley
A
D'Esposito
M
.
2004
.
Measuring functional connectivity during distinct stages of a cognitive task
.
Neuroimage
 .
23
:
752
763
.
Rissman
J
Gazzaley
A
D'Esposito
M
.
2008
.
Dynamic adjustments in prefrontal, hippocampal, and inferior temporal interactions with increasing visual working memory load
.
Cereb Cortex
 .
18
:
1618
1629
.
Rissman
J
Gazzaley
A
D'Esposito
M
.
2009
.
The effect of non-visual working memory load on top-down modulation of visual processing
.
Neuropsychologia
 .
47
:
1637
1646
.
Rota
G
Sitaram
R
Veit
R
Erb
M
Weiskopf
N
Dogil
G
Birbaumer
N
.
2009
.
Self-regulation of regional cortical activity using real-time fMRI: the right inferior frontal gyrus and linguistic processing
.
Hum Brain Mapp
 .
30
:
1605
1614
.
Shrager
Y
Levy
DA
Hopkins
RO
Squire
LR
.
2008
.
Working memory and the organization of brain systems
.
J Neurosci
 .
28
:
4818
4822
.
Squire
LR
.
1992
.
Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans
.
Psychol Rev
 .
99
:
195
231
.
St Jacques
P
Dolcos
F
Cabeza
R
.
2010
.
Effects of aging on functional connectivity of the amygdala during negative evaluation: a network analysis of fMRI data
.
Neurobiol Aging
 .
31
:
315
327
.
Talmi
D
Grady
CL
Goshen-Gottstein
Y
Moscovitch
M
.
2005
.
Neuroimaging the serial position curve
.
Psychol Sci
 .
16
:
716
723
.
Uncapher
MR
Hutchinson
JB
Wagner
AD
.
2011
.
Dissociable effects of top-down and bottom-up attention during episodic encoding
.
J Neurosci
 .
31
:
12613
12628
.
Uncapher
MR
Wagner
AD
.
2009
.
Posterior parietal cortex and episodic encoding: insights from fMRI subsequent memory effects and dual-attention theory
.
Neurobiol Learn Mem
 .
91
:
139
154
.
Viard
A
Piolino
P
Desgranges
B
Chetelat
G
Lebreton
K
Landeau
B
Young
A
De La Sayette
V
Eustache
F
.
2007
.
Hippocampal activation for autobiographical memories over the entire lifetime in healthy aged subjects: an fMRI study
.
Cereb Cortex
 .
17
:
2453
2467
.
Wais
PE
Kim
OY
Gazzaley
A
.
2012
.
Distractibility during episodic retrieval is exacerbated by perturbation of left ventrolateral prefrontal cortex
.
Cereb Cortex
 .
22
:
717
724
.
Wais
PE
Rubens
MT
Boccanfuso
J
Gazzaley
A
.
2010
.
Neural mechanisms underlying the impact of visual distraction on retrieval of long-term memory
.
J Neurosci
 .
30
:
8541
8550
.
Worsley
KJ
Marrett
S
Neelin
P
Vandal
AC
Friston
KJ
Evans
AC
.
1996
.
A unified statistical approach for determining significant signals in images of cerebral activation human brain mapping
.
Hum Brain Mapp
 .
4
:
58
73
.
Zanto
TP
Rubens
MT
Bollinger
J
Gazzaley
A
.
2010
.
Top-down modulation of visual feature processing: the role of the inferior frontal junction
.
Neuroimage
 .
53
:
736
745
.
Zanto
TP
Rubens
MT
Thangavel
A
Gazzaley
A
.
2011
.
Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory
.
Nat Neurosci
 .
14
:
656
661
.
Zarahn
E
Aguirre
GK
D'Esposito
M
.
1997
.
Empirical analyses of BOLD fMRI statistics. I. Spatially unsmoothed data collected under null-hypothesis conditions
.
Neuroimage
 .
5
:
179
197
.

Author notes

T. P. Zanto, W. C. Clapp, and M. T. Rubens contributed equally.