Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Feb 1.
Published in final edited form as: J Neurosci Res. 2025 Feb;103(2):e70021. doi: 10.1002/jnr.70021

A tale of two n-backs: Diverging associations of dorsolateral prefrontal cortex activation with n-back task performance

Philip N Tubiolo 1,2,3, John C Williams 1,2,4, Jared X Van Snellenberg 1,2,5
PMCID: PMC11913012  NIHMSID: NIHMS2051802  PMID: 39902779

Abstract

In studying the neural correlates of working memory (WM) ability via functional magnetic resonance imaging (fMRI) in health and disease, it is relatively uncommon for investigators to report associations between brain activation and measures of task performance. Additionally, how the choice of WM task impacts observed activation-performance relationships is poorly understood. We sought to illustrate the impact of WM task on brain-behavior correlations using two large, publicly available datasets. We conducted between-participants analyses of task-based fMRI data from two publicly available datasets: the Human Connectome Project (HCP; n = 866) and the Queensland Twin Imaging (QTIM) Study (n = 459). Participants performed two distinct variations of the n-back WM task with different stimuli, timings, and response paradigms. Associations between brain activation ([2-back − 0-back] contrast) and task performance (2-back % correct) were investigated separately in each dataset, as well as across datasets, within the dorsolateral prefrontal cortex (dlPFC), medial prefrontal cortex, and whole cortex. Global patterns of activation to task were similar in both datasets. However, opposite associations between activation and task performance were observed in bilateral pre-supplementary motor area and left middle frontal gyrus. Within the dlPFC, HCP participants exhibited a significantly greater activation-performance relationship in bilateral middle frontal gyrus relative to QTIM Study participants. The observation of diverging activation-performance relationships between two large datasets performing variations of the n-back task serves as a critical reminder for investigators to exercise caution when selecting WM tasks and interpreting neural activation in response to a WM task.

Keywords: Working memory, n-back, fMRI, Dorsolateral prefrontal cortex, Brain-behavior relationships, Neural correlate

Graphical Abstract

graphic file with name nihms-2051802-f0008.jpg

In healthy participants from two public, task-based fMRI datasets performing variations of the n-back working memory task, we demonstrate diverging associations between brain activation and task performance despite observing similar task-evoked activation patterns. This suggests that working memory tasks are not interchangeable when studying “neural correlates” of working memory ability.

1. Introduction

Deficits in working memory (WM) are a hallmark of many neurological and psychiatric diseases, and understanding the neural mechanisms of these impairments is critical for the development of novel treatments and diagnostics. The field of psychiatry has thus embraced the study of WM in psychiatric patients, perhaps most notably in patients with schizophrenia. Differences between patients and control participants in blood-oxygen-level-dependent (BOLD) activation of the dorsolateral prefrontal cortex (dlPFC) or other brain regions activated by WM tasks are commonly observed, and this difference is presumed to reflect a neurobiological abnormality that has relevance to the patient deficit in task performance (Anticevic et al., 2013; Carter et al., 1998; Jiang et al., 2015; Perlstein et al., 2001; Perlstein et al., 2003; Wu & Jiang, 2019). However, as has been argued elsewhere (Weinberger & Radulescu, 2016), patient-control differences in brain imaging outcome measures do not necessarily reflect a true difference in the underlying neurobiology.

One approach that can demonstrate whether a patient-control difference in BOLD activation has neurobiological relevance is to demonstrate a between-participants correlation or other association (e.g., using linear regression) between activation and the phenomenon it is putatively related to (e.g., WM task performance). Indeed, while the term “neural correlate” is widely used in the literature to refer to a task-induced change in BOLD signal, such findings do not imply an actual correlation between activation and task performance, and reports of individual-differences associations between task-induced activation and WM task performance are relatively uncommon. After nearly three decades and hundreds of studies, only a small handful have reported an association between greater activation of dlPFC and WM task performance (Cole et al., 2012; Hakun & Johnson, 2017; Lamichhane et al., 2020; Nagel et al., 2011; Satterthwaite et al., 2013; Smucny et al., 2023; Suzuki et al., 2018; Wager et al., 2014), with a similar number reporting that the strength of deactivation of medial prefrontal cortex (mPFC) is also associated with better WM task performance (Anticevic et al., 2010; Eryilmaz et al., 2016; Van Snellenberg et al., 2016; Wager et al., 2014; Whitfield-Gabrieli et al., 2009; Williams et al., 2023).

In contrast, many studies report significant WM task-evoked activation in dlPFC without reporting an association with task performance (Cohen et al., 1994; Forsyth et al., 2014; Melrose et al., 2020; Nitchie et al., 2024; Stein et al., 2021; Tachibana et al., 2012). Evidently, the reporting of brain-behavior correlations in studies of WM is not systematic across the literature, and a consensus on the specific contribution of dlPFC to working memory ability in healthy controls remains elusive. For example, some evidence points to an “Inverted-U” or non-linear relationship between WM load and dlPFC activation that could complicate this relationship (Callicott et al., 2003; Jansma et al., 2004; Manoach, 2003; Van Snellenberg et al., 2016), and some studies have found that functional connectivity within the fronto-parietal control network may have relevance for task performance (Braun et al., 2015; Cassidy et al., 2016; Van Snellenberg et al., 2015; Voigt et al., 2023). Nevertheless, a clearer understanding of brain-behavior relationships, especially in dlPFC, is arguably a “missing link” in the literature that would permit a clearer interpretation of the functional significance of patient-control differences in task activation. In the absence of a well-characterized brain-behavior relationship between activation of a brain region and performance on the task, it is difficult to know whether any abnormalities in brain activation that are observed in a patient population have relevance to their cognitive deficits.

Critically, while a majority of WM fMRI studies use n-back tasks, the n-back has two major variants that share relatively few task properties (see Figure 1). These two tasks appear to have been invented independently—and more than once (Kane & Conway, 2016). The first was developed in the 1950s in an unpublished dissertation, and then made its way into the broader literature (Kay, 1953; Kirchner, 1958; Mackworth, 1959; Singleton, 1978; Welford & Cambridge. University. Psychological Laboratory. Nuffield Research Unit into Problems of Ageing. [from old catalog], 1958) and becoming one of the first tasks used in an fMRI study of patients with schizophrenia, at the National Institute of Mental Health (NIMH) in Daniel Weinberger’s group (Callicott et al., 2000; Callicott et al., 1999; Callicott et al., 1998). We term this a continuous delayed response n-back (NB-CDR), as it requires participants to delay their response for n trials, such that on each trial they make the behavioral response corresponding to the stimulus that was presented n trials previously (see Figure 1, top).

Figure 1.

Figure 1.

Open in a new tab

Top: Schematic of an NB-CDR 2-back trial. Stimuli typically consist of both numbers and spatial locations, and on each trial participants must make the response corresponding to the stimulus displayed n trials previously. This example reflects a sample 2-back trial of the QTIM NB-CDR task. Bottom: Schematic of an NB-DMS 2-back trial, including a recent-probe lure trial. Participants must indicate whether each presented stimulus matches the stimulus presented n trials previously. In this example, stimuli are images of places, as used in the Human Connectome Project n-back task.

The second, and most widely used, n-back was developed in the 1960s (Moore & Ross, 1963; Ross, 1966a, 1966b), but received relatively little attention (Kane & Conway, 2016). It appears to have then been independently re-invented by Alan Gevins for use in event-related potential (ERP) research in 1990 (Gevins et al., 1990; Kane & Conway, 2016), initiating its now widespread use in cognitive neuroscience and related fields (Awh et al., 1996; Casey et al., 1995; Cohen et al., 1994; Gevins & Cutillo, 1993; Gevins et al., 1990; Jonides et al., 1997; Kane & Conway, 2016; Schumacher et al., 1996). This task involves a series of sequentially presented stimuli, and participants indicate whenever a stimulus matches the stimulus presented n trials previously; we term this a delayed-match-to-sample n-back (NB-DMS; see Figure 1, bottom).Critically, to our knowledge, no study has ever used both tasks together, nor attempted to determine empirically which cognitive processes or brain activation patterns are shared, or distinct, between the two tasks. Notably, the correlation between dlPFC activation and WM task performance has primarily been demonstrated in NB-DMS tasks (Cole et al., 2012; Hakun & Johnson, 2017; Lamichhane et al., 2020; Nagel et al., 2011; Satterthwaite et al., 2013) and a SIRP task with distractors (Wager et al., 2014), raising the possibility that such a relationship might not exist in other WM tasks.

Consequently, we sought to utilize data from two large public datasets, the Human Connectome Project (HCP) 1200 Subjects Release (Barch et al., 2013; M. F. Glasser et al., 2016; Glasser et al., 2013; Hodge et al., 2016; Smith et al., 2013; Uğurbil et al., 2013; Van Essen et al., 2013) and the Queensland Twin Imaging (QTIM) Study (Blokland et al., 2008; Blokland et al., 2011; Strike, 2023; Strike et al., 2019), to investigate the correlation between neural activation to a WM task and in-scanner performance on that task. Furthermore, we sought to determine whether two versions of the n-back task, the NB-CDR and NB-DMS, elicit the same brain-behavior correlations, or whether important differences exist between these tasks in how the brain subserves task performance.

2. Materials and Methods

2.1. Participants

We utilized task-based fMRI data from 1,325 participants across two publicly available datasets: the HCP 1200 Subjects Release (n=866; Barch et al., 2013; M. F. Glasser et al., 2016; Glasser et al., 2013; Hodge et al., 2016; Smith et al., 2013; Uğurbil et al., 2013; Van Essen et al., 2013) and the QTIM Study (n=459; Blokland et al., 2008; Blokland et al., 2011; Strike, 2023; Strike et al., 2019). Table 1 contains a summary of differences in demographics, data acquisition, and task design between datasets. Subject IDs used in each dataset can be found in Tables S1 and S2.

Table 1.

Summary of all major differences in demographics, fMRI acquisition parameters, and N-Back task design between the Human Connectome Project (HCP) and Queensland Twin Imaging (QTIM) Study datasets.

Dataset
HCP QTIM
Demographics N 866 459
Mean Age (σ) 28.66 (3.74) 22.21 (2.90)
Biological Sex 472 F / 394 M 276 F / 183 M
Handedness 785 R / 78 L / 3 NP 459 R / 0 L
fMRI Acquisition Scanner Siemens Connectome Skyra 3T Bruker MedSpec 4T
TR (ms) 720 2100
TE (ms) 33.1 30
Flip Angle (°) 52 90
FOV (mm) 208 × 180 230 × 230
Voxel Size (mm) 2.0 × 2.0 × 2.0 3.6 × 3.6 × 3.0
Slice Gap (mm) 0 0.6
Multiband Factor 8 1
# Volumes 405 127
Matrix Size 104 × 90 64 × 64
Task Design # Runs 2 1
Run Length (min:sec) 4:52 4:27
Task Conditions 0-back, 2-back 0-back, 2-back
# Task Block per Run 8 (4 blocks/condition) 16 (8 blocks/condition)
# Trials per Block 10 16
Block Duration (s) 25 16
Trial Duration (s) 2.5 1
Stimulus Presentation Time (s) 2 0.2
Interstimulus Interval (s) 0.5 0.8
Stimulus Pool Faces, Tools, Places, Body Parts Numbers 1–4 in diamond configuration
# Potential Responses 2 4
Pre-Block Cue Duration (s) 2.5 0
Fixation Blocks? 4/run (15 s each) None
Open in a new tab

HCP = Human Connectome Project; QTIM = Queensland Twin Imaging; TR = Repetition Time; TE = Echo Time; FOV = Field of View; F = Female; M = Male; R = Right; L = Left; NP = Not Provided

2.1.1. Queensland Twin Imaging (QTIM) Study

Unprocessed structural MRI and task-based fMRI data from 592 unrelated participants of at least 18 years of age from the QTIM Study (Blokland et al., 2008; Blokland et al., 2011; Strike, 2023; Strike et al., 2019) was obtained from openneuro.org (RRID:SCR_005031; accession number ds004169; Markiewicz et al., 2021). Participants performed a NB-CDR task described in detail below (see below, 2.2. Task Procedures). From this original sample, an additional 133 participants were excluded due to poor data quality (as determined by rigorous visual quality control evaluations following preprocessing; see below, 2.4. fMRI Preprocessing) or poor task performance. Task performance thresholds were determined as the proportion of correct trials on each task condition needed for each individual subject to score above chance at 95% confidence according to a binomial cumulative distribution function (CDF) with a probability of success of 0.25. The resulting thresholds for proportion correct were 0.3215 and 0.3214 for 0-back and 2-back conditions, respectively. These exclusion criteria yielded a final sample of 459 participants.

2.1.2. Human Connectome Project (HCP)

A second cohort of preprocessed structural and task-based fMRI data was obtained from the HCP 1200 Subjects Release (Barch et al., 2013; M. F. Glasser et al., 2016; Glasser et al., 2013; Hodge et al., 2016; Smith et al., 2013; Uğurbil et al., 2013; Van Essen et al., 2013) comprising 866 participants performing a visual NB-DMS task (Barch et al., 2013). Selected participants had 100% task completion, no QC issues A (anatomical anomalies), B (segmentation and surface QC), and C (some data acquired during periods of head coil instability), at least two resting-state runs with at least 7.5 min of data between them (this was used as a judge of quality for MSMAll registration; Robinson et al., 2018; Robinson et al., 2014), and sufficient accuracy on both task conditions as determined by a binomial CDF with a success probability of 0.5 (proportion correct of 0.587 and 0.609 for 0-back and 2-back trials, respectively). Related subjects were permitted in this sample given that complete family information (Mother ID, Father ID, and Zygosity) was present. For each participant, the Structural Preprocessed and Working Memory Task fMRI Analysis – Grayordinates, 4mm Smoothing packages were obtained from ConnectomeDB (RRID:SCR_004830; Hodge et al., 2016).

2.2. Task Procedures

Participants in both datasets performed variations of a n-back task with different stimuli, timing, and response paradigms. Major differences between tasks are described in the Task Design section of Table 1, and examples of task response paradigms are shown in Figure 1. In the QTIM Study NB-CDR task, participants were presented with random sequences of the numbers 1–4 arranged in a diamond formation and responded using a response box with four buttons in the same diamond formation (Figure 1, top; Blokland et al., 2008; Blokland et al., 2011). For the 0-back condition, participants were asked to respond with the currently presented stimulus, while for the 2-back condition participants responded with the stimulus that was presented two stimuli prior.

In the HCP NB-DMS task (Barch et al., 2013), stimuli consisted of individually presented pictures of faces, places, tools, or body parts. In the 0-back condition, participants were presented with a ‘target’ stimulus in the beginning of the block; they were asked to respond ‘target’ on a two-button response device to any subsequent presentation of that stimulus, and ‘non-target’ everywhere else. In the 2-back condition, participants responded ‘target’ when the current stimulus matched the stimulus presented two stimuli prior (Figure 1, bottom).

For both tasks, performance was assessed via proportion correct and median reaction time (RT) for each task condition and across both conditions.

2.3. fMRI Procedures

Acquisition parameters for each dataset are detailed in Table 1 and described in detail elsewhere (Barch et al., 2013; Blokland et al., 2011; Uğurbil et al., 2013). There are several major acquisition differences between the two datasets. Notably, HCP data was collected using simultaneous multi-slice (multiband) acquisition (Feinberg et al., 2010; Moeller et al., 2010), while QTIM Study data was collected using single-slice (single-band) acquisition. The use of multiband acceleration in HCP allowed for a shorter TR and smaller voxels without slice gaps. Another notable difference is the disparity in data quantity, where HCP participants performed two task runs, equating to over twice the duration and six times the number of volumes acquired for each QTIM Study participant.

2.4. fMRI Preprocessing

HCP data was obtained in its preprocessed form (Hodge et al., 2016) and was preprocessed using the HCP Minimal Preprocessing Pipelines, as described elsewhere (Glasser et al., 2013). Structural MRI and fMRI data from the QTIM Study were preprocessed using fMRIPrep 22.0.1 (see Supplementary Methods; Esteban et al., 2020; Esteban et al., 2018; Esteban, 2022). Briefly, volumetric timeseries were spatially normalized to MNI152NLin2009cAsym template space, and left and right cortical surfaces were then extracted in fsaverage space. These surfaces were resampled to CIFTI format (*.dtseries.nii; Glasser et al., 2013), with 32,492 vertices per surface, to be used for all subsequent analyses. The first five timepoints of each run in the QTIM Study dataset were removed to ensure steady-state tissue magnetization. Each surface timeseries was then normalized to mean of 100 (Van Snellenberg et al., 2015) and smoothed with a 4mm full width at half-maximum Gaussian kernel. Each participant’s structural surfaces were obtained in fsaverage space and resampled to FreeSurfer left-right symmetric space with 32,492 vertices (fs_LR_32k) to match the geometry of their surface timeseries, as well as data from HCP.

2.5. Within-Participant Modeling

Within-participant modeling in the HCP dataset was performed as described elsewhere (Barch et al., 2013). Cortical [2-back − 0-back] contrast maps (x[subjectID]_tfmri_wm_level2_2bk_0bk_hp200_s4_msmall) were extracted in CIFTI format. The QTIM Study NB-CDR task was modeled using a block design with separate regressors of interest for 0-back trials, 2-back trials, and motor responses. Nuisance regressors included six rigid-body motion parameters (along with their squares, derivatives, and squared derivatives), cerebrospinal fluid signal, and spike regressors to identify timepoints at which excessive motion occurred. Spike regressors were determined based on a run-adaptive, generalized extreme value DVARS (GEV-DV; Afyouni & Nichols, 2018; Power et al., 2014; Smyser et al., 2011; Williams et al., 2022) parameter dG of 17, which flagged 2.961% of timepoints study-wide as high-motion. Regressors of interest were convolved with a three-parameter hemodynamic response function (HRF) with time and dispersion derivatives, as in prior work (Van Snellenberg et al., 2016; Van Snellenberg et al., 2015), while nuisance regressors were not. Activation at each cortical grayordinate was quantified as the total area under the curve of the three-parameter HRF with a 2–9 second window (Van Snellenberg et al., 2016; Van Snellenberg et al., 2015). Cortical [2-back − 0-back] contrasts were then calculated for each participant.

2.6. Regions of Interest

ROI masks of bilateral dlPFC and mPFC were identified using the parcellations (Gordon333.32k_fs_LR) developed by Gordon and associates (Gordon et al., 2016) derived from resting-state correlations. The dlPFC was taken to be regions of the Gordon parcellation communities “Dorsal Attention Network” and “Frontoparietal” localized in the PFC, while the Gordon parcellation community “Default Mode Network” was limited to just the mPFC. The bilateral dlPFC ROIs comprised the following parcels from the Gordon parcellation: 168, 240, 272, 273, 276, 277, 319, 320, 327, 328, 236, 250, 271, 275, 74, 106, 107, 110, 113, 155, 7, 78, 108, 109, 148, 149. The bilateral mPFC ROIs comprised the following parcels: 325, 323, 322, 184, 279, 278, 25, 116, 117, 150, 152. Cortical masks of the dlPFC and mPFC were then dilated by 5 mm to fill empty space between parcels and eroded by 3 mm to remove portions of the dilated ROI falling outside the outer boundary of the chosen parcels. Dilation and erosion were performed using metric-dilate and metric-erode, respectively, in HCP Connectome Workbench v1.4.2 (RRID:SCR_008750; Van Essen et al., 2011).

2.7. Between-Participants Analysis Within Datasets

Within each dataset, across-participant analyses of [2-back − 0-back] contrasts were performed using Permutation Analysis of Linear Models (PALM) alpha119 (RRID:SCR_017029; Winkler et al., 2014; Winkler et al., 2016) with 20,000 permutations or sign-flips using familywise error rate (FWER) correction (Holmes et al., 2016) and threshold-free cluster enhancement (TFCE; Smith & Nichols, 2009). For all between-participants analyses including participants from the HCP dataset, multi-level exchangeability blocks (Winkler et al., 2015) were defined using the MATLAB (RRID:SCR_001622) code hcp2blocks.m provided in the PALM User Guide (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM/ExchangeabilityBlocks). Blocks were defined using Mother IDs, Father IDs, and zygosity information such that families were shuffled as whole blocks, and participants within each family were only permuted amongst each other. Exchangeability blocks were not used in the QTIM Study dataset, since only unrelated participants were selected. All resulting t-statistic maps were thresholded (FWER-corrected p < 0.05 over contrasts (Alberton et al., 2020) and Dunn–Šidák corrected (Šidák, 1967) across hemispheres).

The first analysis identified regions of significant [2-back − 0-back] activation and deactivation with a design matrix containing only an intercept. This analysis was performed across the entire cortical surface.

The second analysis determined regions with significant associations between task performance, measured as 2-back proportion correct, and [2-back − 0-back] activation, using a design matrix containing predictors for performance, age, gender, and performance*age and performance*gender interactions. All predictors were mean-centered prior to the calculation of interaction terms. This analysis was performed separately within the dlPFC and mPFC masks (see above, 2.6. Regions of Interest), and across the entire cortical surface.

For characterization of significant grayordinates, the peak statistically significant t-statistic within each cortical parcel in the Cole-Anticevic Brain Network Parcellation (CAB-NP; Matthew F. Glasser et al., 2016; Ji et al., 2019), as well as the percentage of significant grayordinates within each cortical parcel, were calculated.

2.8. Between-Participants Analysis Across Datasets

A between-dataset analysis of [2-back − 0-back] contrasts was also performed to identify differences in performance-activation relationships that are associated with which variation of the n-back task was performed. This was achieved using PALM using identical methods as described previously (see above, 2.7. Between-Participants Analysis Within Datasets). The design matrix contained predictors for dataset membership, task performance, age, gender, and interaction terms for performance*age, performance*gender, dataset*performance, dataset*age, dataset*gender, dataset*performance*age, and dataset*performance*gender. The main predictor of interest was the dataset*performance interaction. Task performance was measured as 2-back proportion correct and was z-scored within each dataset. All other predictors were mean-centered prior to calculation of interaction terms. This analysis was performed within dlPFC and mPFC ROIs, and across all cortical grayordinates.

In addition to investigating BOLD associations with task performance, we performed a between-dataset analysis of differences in [2-back − 0-back] activation, using a design matrix containing predictors for dataset membership, age, and gender. For this analysis, exchangeability blocks were regenerated to allow for the exchange of participants across datasets (as opposed to being constrained within datasets, as in the previous between-dataset analysis of activation-performance relationship differences).

3. Results

3.1. Task Performance

Violin plots depicting proportion correct and median RT for each task condition (as well as for the overall task) in each dataset are shown in Figure 2. The median RTs across participants for the 0-back condition, 2-back condition, and the overall task in the QTIM Study dataset were 412.500 ms, 229.000 ms, and 359.000 ms, respectively. The median RTs across participants for the 0-back condition, 2-back condition, and the overall task in the HCP dataset were 741.906 ms, 957.656 ms, and 846.970 ms, respectively.

Figure 2.

Figure 2.

Open in a new tab

Proportion of correct responses (A) and median reaction time (B) for the 0-back condition, 2-back condition, and overall N-Back task in the Queensland Twin Imaging (QTIM) and Human Connectome Project (HCP) datasets. Central dashed lines in each violin indicate sample medians, while lower and upper dotted lines indicate 25th and 75th percentiles, respectively.

The median proportion of correct responses across participants for the 0-back condition, 2-back condition, and the overall task in the QTIM Study dataset were 0.891, 0.750, and 0.766, respectively. The median task accuracy across participants for the 0-back condition, 2-back condition, and the overall task in the HCP dataset were 0.941, 0.862, and 0.896, respectively.

3.2. Between-Participants Analysis Within Datasets

Cortical regions exhibiting significant [2-back − 0-back] activation and deactivation in each dataset are shown in Figure 3A and Figure 3B (peak significant t-statistics in CAB-NP parcels shown in Figure S1 and Figure S2). Both datasets exhibited significant activation in bilateral dlPFC, ventrolateral PFC, pre-supplementary motor area (SMA)/dorsal anterior cingulate, SMA, premotor cortex, anterior insula, intraparietal sulcus (IPS), precuneus, temporoparietal junction, and lateral temporal lobe. Significant deactivation was observed in bilateral mPFC, insular cortex, primary motor and somatosensory cortices, superior temporal gyrus, anterior temporal lobe and posterior cingulate cortex. Differing activation patterns were observed in primary and secondary visual cortex, with the QTIM Study dataset exhibiting deactivation in the cuneus and activation in lateral occipital lobe, and the opposite pattern in the HCP dataset.

Figure 3.

Figure 3.

Open in a new tab

Regions across all cortical greyordinates with significant [2-back − 0-back] contrast activation (warm colors) and deactivation (cool colors) in the Queensland Twin Imaging (QTIM; A) and Human Connectome Project (HCP; B) datasets. C) Regions across all greyordinates with significant differences in [2-back − 0-back] contrast activation between datasets. Significant greyordinates were determined via permutation testing in Permutation Analysis of Linear Models (PALM) using 20,000 permutations, threshold-free cluster enhancement (TFCE), and family-wise error rate (FWER) correction over greyordinates and contrasts (FWER corrected, p<0.05). BALSA Scene IDs (Study ID: 0wrMq) associated with this figure are as follows: Fig 3A_QTIMSignificantActivation; Fig 3B_HCPSignificantActivation;

QTIMminusHCP_taskActivationDifferences.

Significant activation-performance relationships across all cortical grayordinates are shown in Figure 4 (peak significant t-statistics in CAB-NP parcels shown in Figure S3 and Figure S4). The QTIM Study dataset only exhibited a positive relationship between activation and 2-back performance in a small region of the left postcentral gyrus. Negative activation-performance relationships were observed in the left middle frontal gyrus and pre-SMA. In the HCP dataset, positive activation-performance relationships were observed in bilateral dlPFC (including middle frontal gyrus), supplementary and premotor cortex, anterior poles, pre-SMA, anterior insula, middle temporal gyrus, precuneus, cuneus and calcarine fissure, right temporoparietal junction, and IPS. Negative performance-activation relationships were exhibited by mPFC, central sulcus, posterior cingulate cortex, and temporal poles.

Figure 4.

Figure 4.

Open in a new tab

Regions across all cortical greyordinates with significantly positive (warm colors) and negative (cool colors) associations between [2-back − 0-back] contrast activation and 2-back accuracy in the Queensland Twin Imaging (QTIM; A) and Human Connectome Project (HCP; B) datasets. Results reflect a whole-brain analysis. Significant greyordinates were determined via permutation testing in Permutation Analysis of Linear Models (PALM) using 20,000 permutations, threshold-free cluster enhancement (TFCE), and family-wise error rate (FWER) correction over greyordinates and contrasts (FWER corrected, p<0.05). BALSA Scene IDs (Study ID: 0wrMq) associated with this figure are as follows:

Fig 4A_QTIM_PerformanceActivation_WholeCortex;

Fig 4B_HCP_PerformanceActivation_WholeCortex.

Bilateral dlPFC grayordinates with significant associations between [2-back − 0-back] contrast activation and 2-back accuracy in each dataset are shown in Figure 5 (peak significant t-statistics in CAB-NP parcels shown in Figure S5 and Figure S6). The QTIM Study dataset only exhibited a positive activation-performance relationship in a small inferior region of the right precentral gyrus, and a negative relationship in the left orbitofrontal cortex (OFC) and a small region of the left middle frontal gyrus. In contrast, the HCP dataset exhibited widespread positive activation-performance relationships in a large majority of the bilateral dlPFC mask, centered on the middle frontal gyrus. A negative relationship was observed in the right ventrolateral PFC.

Figure 5.

Figure 5.

Open in a new tab

Regions with significantly positive (warm colors) and negative (cool colors) associations between [2-back − 0-back] contrast activation and 2-back accuracy in the Queensland Twin Imaging (QTIM; A) and Human Connectome Project (HCP; B) datasets. Significant greyordinates were determined within a bilateral dorsolateral prefrontal cortex (dlPFC) mask (grey underlay) via permutation testing in Permutation Analysis of Linear Models (PALM) using 20,000 permutations, threshold-free cluster enhancement (TFCE), and family-wise error rate (FWER) correction over greyordinates and contrasts (FWER corrected, p<0.05). BALSA Scene IDs (Study ID: 0wrMq) associated with this figure are as follows: Fig 5A_QTIM_PerformanceActivation_DLPFC; Fig 5B_HCP_PerformanceActivation_DLPFC.

Grayordinates within the bilateral mPFC mask exhibiting significant activation-performance relationships are shown in Figure 6 (peak significant t-statistics in CAB-NP parcels are shown in Figure S7 and Figure S8). The QTIM Study dataset exhibited negative relationships in the anterior portion of the bilateral pre-SMA, whereas the HCP dataset showed positive relationship in this region.

Figure 6.

Figure 6.

Open in a new tab

Regions with significantly positive (warm colors) and negative (cool colors) associations between [2-back − 0-back] contrast activation and 2-back accuracy in the Queensland Twin Imaging (QTIM; A) and Human Connectome Project (HCP; B) datasets. Significant greyordinates were determined within a bilateral medial prefrontal cortex (mPFC) mask (grey underlay) via permutation testing in Permutation Analysis of Linear Models (PALM) using 20,000 permutations, threshold-free cluster enhancement (TFCE), and family-wise error rate (FWER) correction over greyordinates and contrasts (FWER corrected, p<0.05). No positive associations were observed in the QTIM Study dataset. BALSA Scene IDs (Study ID: 0wrMq) associated with this figure are as follows: Fig 6A_QTIM_PerformanceActivation_MPFC; Fig 6B_HCP_PerformanceActivation_MPFC.

3.3. Between-Participants Analysis Across Datasets

Regions showing significant differences in [2-back − 0-back] contrast activation between datasets are shown in Figure 3C (peak significant t-statistics in CAB-NP parcels shown in Figure S9). The HCP dataset showed significantly greater activation in bilateral dlPFC, pre-SMA, premotor cortex, frontal poles, anterior insula, precuneus, and IPS. The Queensland dataset showed greater activation in a small region of right secondary visual cortex.

Cortical regions with significantly different activation-performance relationships between datasets, both within the bilateral dlPFC mask and across all cortical grayordinates, are shown in Figure 7 (peak significant t-statistics in CAB-NP parcels shown in Figure S10). No significantly greater activation-performance relationships in the QTIM Study dataset over the HCP dataset were observed. Within the dlPFC mask, greater relationships in the HCP dataset over the QTIM Study dataset were observed on the bilateral middle frontal gyrus, in the location which the strongest positive activation-performance relationships were observed in the HCP dataset alone (see Figure 5B). Across all cortical grayordinates, greater activation-performance relationships in the HCP dataset were observed in the right precuneus and bilateral superior frontal sulcus.

Figure 7.

Figure 7.

Open in a new tab

Regions with significantly lower associations between [2-back − 0-back] contrast activation and 2-back accuracy in the Queensland Twin Imaging (QTIM) dataset than the Human Connectome Project (HCP) dataset. Significant greyordinates were determined both within a bilateral dorsolateral prefrontal cortex (dlPFC) mask (A; grey underlay) and in a whole-cortex analysis (B) via permutation testing in Permutation Analysis of Linear Models (PALM) using 20,000 permutations, threshold-free cluster enhancement (TFCE), and family-wise error rate (FWER) correction over greyordinates and contrasts (FWER corrected, p<0.05). No significantly greater activation-performance associations in QTIM over HCP were identified, and thus they are not shown. BALSA Scene IDs (Study ID: 0wrMq) associated with this figure are as follows: Fig 7A_Combined_DLPFC_PerformanceActivationDifferences; Fig 7B_Combined_WholeCortex_PerformanceActivationDifferences.

4. Discussion

The results presented here demonstrate a striking dissociation between BOLD estimates of neural activation and the correlation between BOLD activation and WM task performance across two versions of the n-back task, the NB-CDR and NB-DMS. While both tasks result in highly similar patterns of activation in the same core network of brain regions commonly observed to be active during WM task performance (along with deactivation of the default-mode network), they differ dramatically in the association between brain activation and task performance. In HCP NB-DMS task data, activation of bilateral dlPFC, frontal poles, anterior insula, pre-SMA, SMA, IPS and superior parietal lobe, middle temporal gyrus, and left visual cortex were associated with improved task performance, while deactivation of mPFC, posterior cingulate, somatomotor regions, lateral OFC, and posterior insula was also associated with improved performance. However, these associations were largely absent in the QTIM NB-CDR task, with activation in a portion of left dlPFC and pre-SMA instead demonstrating a negative association with task performance. A direct comparison of these brain-behavior correlations between the two datasets revealed significant differences in bilateral dlPFC and right superior parietal lobe.

These findings highlight at least two major issues in the study of cognition and WM in both healthy and psychiatric populations. First, it is clear that the diverse array of tasks referred to as “WM tasks” in the literature should not be assumed to be interchangeable, even when they elicit activation of the same broad network of brain regions during their performance, and even when they share the same name (e.g., n-back). While we here characterize brain-behavior relationships in large datasets using two commonly used WM tasks, these relationships in other WM tasks (such as the commonly used SIRP) remain unknown, and could potentially even differ from the results presented here in different versions of the same tasks (e.g., letter NB-DMS tasks are extremely common in the literature, but use very different stimuli than the HCP NB-DMS used here).

Second, investigators should resist the temptation to assume that increased or decreased BOLD activation in a region that is commonly activated (or deactivated) during the performance of a cognitive task is reflective of successful task performance without directly demonstrating that that is the case. This is a particular issue in neuroimaging studies of psychiatric populations, in which increased or decreased activation by a patient population with cognitive deficits is commonly assumed to reflect a “failure to engage task networks” (in the case of decreased activation) or “inefficient cortical function” (in the case of increased activation), and such differences are broadly assumed to reflect pathology regardless of whether increased or decreased activation is observed. As demonstrated here, reduced activation in left dlPFC by a patient group could mean entirely different things depending on whether it was observed in an NB-DMS or NB-CDR task.

Unfortunately, it is impossible to pinpoint which characteristics of the NB-CDR and NB-DMS tasks, or potentially even characteristics of the participant samples or imaging acquisition procedures, led to the observed dissociation between these tasks in the relationship between BOLD activation and task performance. The HCP study was conducted on a slightly older sample of adults with a higher proportion of male participants and substantial inclusion of left-handed and ambidextrous individuals, and was undertaken using an aggressive multiband acceleration technique that led to much shorter TRs and higher spatial resolution (Uğurbil et al., 2013). In addition to superior data quality in the HCP study, there was a substantial difference in sample size between the two datasets, both in number of participants and volumes acquired per participant. While this could theoretically lead to differences in detected brain-behavior associations, a sample of 459 individuals in the QTIM study should possess sufficient statistical power to detect these associations if they are present. Further, while both tasks evoke similar patterns of BOLD activation, the HCP NB-DMS task clearly evoked stronger activation of task-positive brain regions than the QTIM study NB-CDR task (see Figure 3C).

While specific task effects in this analysis cannot be fully disentangled from differences due to varied acquisition parameters and data quality between datasets, the fact that task-positive regions are the only regions displaying significant activation differences raises concerns regarding the functional equivalence of the tasks. Therefore, we view differences in the WM task used in each of the two studies as a more likely cause of the observed difference in brain-behavior correlations between the two studies, as differences in sample and acquisition paradigm are unlikely to have produced associations with task performance in the opposite direction in some brain regions between the two datasets. In addition to major differences in the stimuli used (faces, tools, places, and body parts in the HCP study, and the numbers 1–4 combined with 4 spatial locations in the QTIM study), the QTIM study used a very rapid stimulus presentation (200 ms compared to 2000 ms in the HCP study) alongside a much shorter total trial length (1000 ms compared to 2500 ms) that could have impacted the effectiveness of any cognitive strategies used to perform the task, and thereby resulted in differences in how BOLD activation is related to task performance across individuals.

Another critical difference between the NB-CDR and NB-DMS tasks that could have led to the results reported here is that in the NB-CDR, the correct response can be prepared in advance for all task loads of 1-back or greater. An effect of this can be clearly seen in the long tails of the RT data for the NB-CDR presented in Figure 2, which fall below the lower limits of human RT for some participants. (We note that the long tail in RTs in the 0-back condition was driven by three participants with poor—though still above-chance—performance on that condition, and appear to have been caused by incorrect responses that were actually late responses to the preceding trial.) That is, because the NB-CDR requires participants to simply delay their response by n trials, rather than make a match/non-match discrimination on each trial based on information presented n trials previously, participants who have successfully encoded and maintained the stimulus can make an accurate response without even processing the stimulus presented in the current trial. When combined with fixed task presentation timing (i.e., a fixed duration and delay for each trial), participants can plan and initiate an accurate motor response before the current trial is even presented, resulting in extremely rapid RTs. These features of the NB-CDR likely also interact with the fact that each stimulus has a 1:1 mapping to a response button, raising the possibility that participants could be preparing a sequence of motor responses that are then initiated at a fixed pace. Indeed, this set of features of the NB-CDR could be responsible for the small region of the postcentral gyrus that showed a positive association with task performance in the QTIM dataset, as well as the negative association in dlPFC and pre-SMA. That is, it may have been more optimal in the QTIM NB-CDR to use lower-order somatomotor processing to maintain prepared motor responses than it was to actively maintain a higher-order representation of prior stimuli and only initiate a response once the current stimulus is presented. This interpretation is also consistent with our finding of significantly lower activation in dlPFC and other task-positive regions in the QTIM dataset compared to HCP. Although speculative, we argue that this is the most probable explanation for the striking differences in brain-behavior associations between the NB-CDR and NB-DMS tasks observed here.

In summary, the results presented here underscore a critical consideration for the design of studies of WM, as well as for the interpretation of BOLD findings resulting from WM task-based fMRI studies of health and disease. The finding of similar patterns of task-evoked BOLD activation, yet diverging regional associations between activation and task performance, in two large datasets performing different n-back tasks demonstrates that investigators must exercise caution when choosing WM tasks and interpreting findings as “neural correlates” of task performance. Clearly, substantial additional work in characterizing brain-behavior relationships across the full range of WM tasks used in the literature is warranted, ideally with multiple tasks used in the same sample of participants to more clearly demonstrate which tasks or task variables (such as stimulus type or duration) result in differences in observed brain-behavior relationships, without contamination from differences in sample composition or fMRI acquisition paradigms. Additionally, we would urge researchers to systematically test for, and report, correlations between BOLD activation and task performance in their samples, in both healthy and patient samples, to develop a larger literature on these associations across the WM literature, and to provide critical context for the interpretation of patient-control differences in brain activation measures. Indeed, identifying brain regions whose activation is associated with improved performance on specific cognitive tasks could provide novel clues as to the neural basis of cognitive deficits in psychiatric patient populations with impaired cognition and disrupted activation in these regions, potentially leading to biomarker development or novel targets for pharmacological or neurostimulation interventions.

Supplementary Material

Supinfo1
Supinfo2

Significance Statement.

Working memory (WM) deficits are a central feature of many neurological and psychiatric illnesses and are closely linked to poor functional outcomes. Understanding the neural mechanisms of WM is essential for developing novel treatments and diagnostics. fMRI-based studies of WM often report task-evoked brain activation but fail to report associations between brain activation and behavioral outcomes, such as task performance. Additionally, it is commonly assumed that various behavioral paradigms claiming to measure WM are interchangeable, despite substantial differences between them. Here, we demonstrate in two large, public datasets that WM tasks are not interchangeable and may evoke different brain-behavior relationships.

5. Acknowledgements

5.1. Assistance

The authors would like to acknowledge the computing resources and technical assistance provided by Stony Brook Medicine Research Computing, with substantial support from Allen Zawada and James Xikis, as well as Stony Brook Research Computing and Cyberinfrastructure and the Institute for Advanced Computational Science at Stony Brook University for access to the high-performance SeaWulf computing system (National Science Foundation Award Nos. 1531492 and 2215987, and matching funds from the Empire State Development’s Division of Science, Technology and Innovation program contract C210148), with notable support from Fırat Coşkun, Daniel Wood, and David Carlson. The authors would also like to acknowledge Elizabeth Chan for assisting in data visualization.

5.2. Funding Information

Research reported in this publication was supported by the National Institute of Mental Health of the National Institutes of Health (NIH) (Award Nos. R01MH120293 [to JXVS], and F30MH122136 [to JCW]). JCW was also supported by the Stony Brook University Medical Scientist Training Program (Award No. T32GM008444; Principal Investigator: Dr. Michael A. Frohman) and a Research Supplement to Promote Diversity in Health-Related Research (3R01MH120293–04S1). PNT was supported by a Stony Brook University Department of Biomedical Engineering Graduate Assistance in Areas of National Need Fellowship (P200A210006) and the Stony Brook University Scholars in Biomedical Sciences Program (Award No. T32GM148331; PI: Dr. Styliani-Anna [Stella] E. Tsirka). The Stony Brook University high-performance SeaWulf computing system was supported by National Science Foundation (NSF) Award Nos. 1531492 and 2215987 (PI: Dr. Robert Harrison), and matching funds from the Empire State Development’s Division of Science, Technology and Innovation (NYSTAR) program contract C210148. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the NIH, the NSF, or NYSTAR.

Footnotes

6.

Disclosures

The authors report no biomedical financial interests or potential conflicts of interest. A preprint of this manuscript has been submitted to bioRxiv (DOI: 10.1101/2024.05.23.595597).

5.3. Data Availability

Publicly available HCP data can be obtained from ConnectomeDB (https://db.humanconnectome.org/). QTIM study data can be obtained from openneuro.org (accession number ds004169).

Thresholded t-statistic maps, as well as peak significant t-statistics and percentage of significant grayordinates within each CAB-NP parcel, from all results described in Sections 3.1. and 3.2. are provided via BALSA (https://balsa.wustl.edu/study/0wrMq). Specific scene IDs are noted in figure captions where applicable.

References

  1. Afyouni S, & Nichols TE (2018). Insight and inference for DVARS. Neuroimage, 172, 291–312. 10.1016/j.neuroimage.2017.12.098 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Alberton BAV, Nichols TE, Gamba HR, & Winkler AM (2020). Multiple testing correction over contrasts for brain imaging. NeuroImage, 216. 10.1016/j.neuroimage.2020.116760 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Anticevic A, Repovs G, & Barch DM (2013). Working memory encoding and maintenance deficits in schizophrenia: neural evidence for activation and deactivation abnormalities. Schizophr Bull, 39(1), 168–178. 10.1093/schbul/sbr107 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Anticevic A, Repovs G, Shulman GL, & Barch DM (2010). When less is more: TPJ and default network deactivation during encoding predicts working memory performance. Neuroimage, 49(3), 2638–2648. 10.1016/j.neuroimage.2009.11.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Awh E, Jonides J, Smith EE, Schumacher EH, Koeppe RA, & Katz S. (1996). Dissociation of Storage and Rehearsal in Verbal Working Memory: Evidence From Positron Emission Tomography. Psychol Sci, 7(1), 25–31. 10.1111/j.1467-9280.1996.tb00662.x [DOI] [Google Scholar]
  6. Barch DM, Burgess GC, Harms MP, Petersen SE, Schlaggar BL, Corbetta M, Glasser MF, Curtiss S, Dixit S, Feldt C, Nolan D, Bryant E, Hartley T, Footer O, Bjork JM, Poldrack R, Smith S, Johansen-Berg H, Snyder AZ, & Van Essen DC (2013). Function in the human connectome: Task-fMRI and individual differences in behavior. NeuroImage, 80, 169–189. 10.1016/j.neuroimage.2013.05.033 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Blokland GAM, McMahon KL, Hoffman J, Zhu G, Meredith M, Martin NG, Thompson PM, de Zubicaray GI, & Wright MJ (2008). Quantifying the heritability of task-related brain activation and performance during the N-back working memory task: A twin fMRI study. Biological Psychology, 79(1), 70–79. 10.1016/j.biopsycho.2008.03.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Blokland GAM, McMahon KL, Thompson PM, Martin NG, de Zubicaray GI, & Wright MJ (2011). Heritability of Working Memory Brain Activation. Journal of Neuroscience, 31(30), 10882–10890. 10.1523/jneurosci.5334-10.2011 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Braun U, Schäfer A, Walter H, Erk S, Romanczuk-Seiferth N, Haddad L, Schweiger JI, Grimm O, Heinz A, Tost H, Meyer-Lindenberg A, & Bassett DS (2015). Dynamic reconfiguration of frontal brain networks during executive cognition in humans. Proceedings of the National Academy of Sciences, 112(37), 11678–11683. 10.1073/pnas.1422487112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Callicott JH, Bertolino A, Mattay VS, Langheim FJ, Duyn J, Coppola R, Goldberg TE, & Weinberger DR (2000). Physiological dysfunction of the dorsolateral prefrontal cortex in schizophrenia revisited. Cereb Cortex, 10(11), 1078–1092. http://www.ncbi.nlm.nih.gov/pubmed/11053229 [DOI] [PubMed] [Google Scholar]
  11. Callicott JH, Mattay VS, Bertolino A, Finn K, Coppola R, Frank JA, Goldberg TE, & Weinberger DR (1999). Physiological characteristics of capacity constraints in working memory as revealed by functional MRI [Clinical Trial]. Cereb Cortex, 9(1), 20–26. http://www.ncbi.nlm.nih.gov/pubmed/10022492 [DOI] [PubMed] [Google Scholar]
  12. Callicott JH, Mattay VS, Verchinski BA, Marenco S, Egan MF, & Weinberger DR (2003). Complexity of prefrontal cortical dysfunction in schizophrenia: more than up or down. Am J Psychiatry, 160(12), 2209–2215. 10.1176/appi.ajp.160.12.2209 [DOI] [PubMed] [Google Scholar]
  13. Callicott JH, Ramsey NF, Tallen K, Bertolino A, Knable MB, Coppola R, Goldberg T, van Gelderen P, Mattay VS, Frank JA, Moonen CTW, & Weinberger DR (1998). Functional magnetic resonance imaging brain mapping in psychiatry: Methodological issues illustrated in a study of working memory in schizophrenia. Neuropsychopharmacology, 18, 186–196. [DOI] [PubMed] [Google Scholar]
  14. Carter CS, Perlstein W, Ganguli R, Brar J, Mintun M, & Cohen JD (1998). Functional hypofrontality and working memory dysfunction in schizophrenia. Am J Psychiatry, 155(9), 1285–1287. 10.1176/ajp.155.9.1285 [DOI] [PubMed] [Google Scholar]
  15. Casey BJ, Cohen JD, Jezzard P, Turner R, Noll DC, Trainor RJ, Giedd J, Kaysen D, Hertz-Pannier L, & Rapoport JL (1995). Activation of prefrontal cortex in children during a nonspatial working memory task with functional MRI. Neuroimage, 2(3), 221–229. 10.1006/nimg.1995.1029 [DOI] [PubMed] [Google Scholar]
  16. Cassidy CM, Van Snellenberg JX, Benavides C, Slifstein M, Wang Z, Moore H, Abi-Dargham A, & Horga G. (2016). Dynamic Connectivity between Brain Networks Supports Working Memory: Relationships to Dopamine Release and Schizophrenia. J Neurosci, 36(15), 4377–4388. 10.1523/JNEUROSCI.3296-15.2016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Cohen JD, Forman SD, Braver TS, Casey BJ, Servan-Schreiber D, & Noll DC (1994). Activation of the prefrontal cortex in a nonspatial working memory task with functional MRI. Hum Brain Mapp, 1(4), 293–304. 10.1002/hbm.460010407 [DOI] [PubMed] [Google Scholar]
  18. Cole MW, Yarkoni T, Repovs G, Anticevic A, & Braver TS (2012). Global connectivity of prefrontal cortex predicts cognitive control and intelligence. J Neurosci, 32(26), 8988–8999. 10.1523/JNEUROSCI.0536-12.2012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Eryilmaz H, Tanner AS, Ho NF, Nitenson AZ, Silverstein NJ, Petruzzi LJ, Goff DC, Manoach DS, & Roffman JL (2016). Disrupted Working Memory Circuitry in Schizophrenia: Disentangling fMRI Markers of Core Pathology vs Other Aspects of Impaired Performance. Neuropsychopharmacology, 41(9), 2411–2420. 10.1038/npp.2016.55 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Esteban O, Ciric R, Finc K, Blair RW, Markiewicz CJ, Moodie CA, Kent JD, Goncalves M, DuPre E, Gomez DEP, Ye Z, Salo T, Valabregue R, Amlien IK, Liem F, Jacoby N, Stojic H, Cieslak M, Urchs S,…Gorgolewski KJ. (2020). Analysis of task-based functional MRI data preprocessed with fMRIPrep. Nat Protoc, 15(7), 2186–2202. 10.1038/s41596-020-0327-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Esteban O, Markiewicz CJ, Blair RW, Moodie CA, Isik AI, Erramuzpe A, Kent JD, Goncalves M, DuPre E, Snyder M, Oya H, Ghosh SS, Wright J, Durnez J, Poldrack RA, & Gorgolewski KJ (2018). fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature Methods, 16(1), 111–116. 10.1038/s41592-018-0235-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Esteban O, Markiewicz CJ, Goncalves M, Provins C, Kent JD, DuPre E, Salo T, Ciric R, Pinsard B, Blair RW, Poldrack RA, & Gorgolewski KJ (2022). fMRIPrep: a robust preprocessing pipeline for functional MRI (22.0.1). In: Zenodo. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Feinberg DA, Moeller S, Smith SM, Auerbach E, Ramanna S, Gunther M, Glasser MF, Miller KL, Ugurbil K, & Yacoub E. (2010). Multiplexed echo planar imaging for sub-second whole brain FMRI and fast diffusion imaging. PLoS One, 5(12), e15710. 10.1371/journal.pone.0015710 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Forsyth JK, McEwen SC, Gee DG, Bearden CE, Addington J, Goodyear B, Cadenhead KS, Mirzakhanian H, Cornblatt BA, Olvet DM, Mathalon DH, McGlashan TH, Perkins DO, Belger A, Seidman LJ, Thermenos HW, Tsuang MT, van Erp TGM, Walker EF,…Cannon TD. (2014). Reliability of functional magnetic resonance imaging activation during working memory in a multi-site study: Analysis from the North American Prodrome Longitudinal Study. NeuroImage, 97, 41–52. 10.1016/j.neuroimage.2014.04.027 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Gevins A, & Cutillo B. (1993). Spatiotemporal dynamics of component processes in human working memory. Electroencephalogr Clin Neurophysiol, 87(3), 128–143. 10.1016/0013-4694(93)90119-g [DOI] [PubMed] [Google Scholar]
  26. Gevins AS, Bressler SL, Cutillo BA, Illes J, Miller JC, Stern J, & Jex HR (1990). Effects of prolonged mental work on functional brain topography. Electroencephalogr Clin Neurophysiol, 76(4), 339–350. 10.1016/0013-4694(90)90035-i [DOI] [PubMed] [Google Scholar]
  27. Glasser MF, Coalson TS, Robinson EC, Hacker CD, Harwell J, Yacoub E, Ugurbil K, Andersson J, Beckmann CF, Jenkinson M, Smith SM, & Van Essen DC (2016). A multi-modal parcellation of human cerebral cortex. Nature, 536(7615), 171–178. 10.1038/nature18933 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Glasser MF, Smith SM, Marcus DS, Andersson JL, Auerbach EJ, Behrens TE, Coalson TS, Harms MP, Jenkinson M, Moeller S, Robinson EC, Sotiropoulos SN, Xu J, Yacoub E, Ugurbil K, & Van Essen DC (2016). The Human Connectome Project’s neuroimaging approach. Nat Neurosci, 19(9), 1175–1187. 10.1038/nn.4361 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Glasser MF, Sotiropoulos SN, Wilson JA, Coalson TS, Fischl B, Andersson JL, Xu J, Jbabdi S, Webster M, Polimeni JR, Van Essen DC, & Jenkinson M. (2013). The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage, 80, 105–124. 10.1016/j.neuroimage.2013.04.127 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Gordon EM, Laumann TO, Adeyemo B, Huckins JF, Kelley WM, & Petersen SE (2016). Generation and Evaluation of a Cortical Area Parcellation from Resting-State Correlations. Cerebral Cortex, 26(1), 288–303. 10.1093/cercor/bhu239 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Hakun JG, & Johnson NF (2017). Dynamic range of frontoparietal functional modulation is associated with working memory capacity limitations in older adults. Brain Cogn, 118, 128–136. 10.1016/j.bandc.2017.08.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Hodge MR, Horton W, Brown T, Herrick R, Olsen T, Hileman ME, McKay M, Archie KA, Cler E, Harms MP, Burgess GC, Glasser MF, Elam JS, Curtiss SW, Barch DM, Oostenveld R, Larson-Prior LJ, Ugurbil K, Van Essen DC, & Marcus DS (2016). ConnectomeDB—Sharing human brain connectivity data. NeuroImage, 124, 1102–1107. 10.1016/j.neuroimage.2015.04.046 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Holmes AP, Blair RC, Watson JDG, & Ford I. (2016). Nonparametric Analysis of Statistic Images from Functional Mapping Experiments. Journal of Cerebral Blood Flow & Metabolism, 16(1), 7–22. 10.1097/00004647-199601000-00002 [DOI] [PubMed] [Google Scholar]
  34. Jansma JM, Ramsey NF, van der Wee NJ, & Kahn RS (2004). Working memory capacity in schizophrenia: a parametric fMRI study [Research Support, Non-U.S. Gov’t]. Schizophr Res, 68(2–3), 159–171. 10.1016/S0920-9964(03)00127-0 [DOI] [PubMed] [Google Scholar]
  35. Ji JL, Spronk M, Kulkarni K, Repovš G, Anticevic A, & Cole MW (2019). Mapping the human brain’s cortical-subcortical functional network organization. NeuroImage, 185, 35–57. 10.1016/j.neuroimage.2018.10.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Jiang S, Yan H, Chen Q, Tian L, Lu T, Tan HY, Yan J, & Zhang D. (2015). Cerebral Inefficient Activation in Schizophrenia Patients and Their Unaffected Parents during the N-Back Working Memory Task: A Family fMRI Study. PLoS One, 10(8), e0135468. 10.1371/journal.pone.0135468 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Jonides J, Schumacher EH, Smith EE, Lauber EJ, Awh E, Minoshima S, & Koeppe RA (1997). Verbal Working Memory Load Affects Regional Brain Activation as Measured by PET. J Cogn Neurosci, 9(4), 462–475. 10.1162/jocn.1997.9.4.462 [DOI] [PubMed] [Google Scholar]
  38. Kane M, & Conway A. (2016). The invention of n-back: An extremely brief history. The Winnower. 10.15200/winn.146722.26397 [DOI] [Google Scholar]
  39. Kay H. (1953). Experimental studies of adult learning. Unpublished doctoral dissertation, Cambridge University. [Google Scholar]
  40. Kirchner WK (1958). Age differences in short-term retention of rapidly changing information. J Exp Psychol, 55(4), 352–358. 10.1037/h0043688 [DOI] [PubMed] [Google Scholar]
  41. Lamichhane B, Westbrook A, Cole MW, & Braver TS (2020). Exploring brain-behavior relationships in the N-back task. Neuroimage, 212, 116683. 10.1016/j.neuroimage.2020.116683 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Mackworth JF (1959). Paced memorizing in a continuous task. J Exp Psychol, 58, 206–211. 10.1037/h0049090 [DOI] [PubMed] [Google Scholar]
  43. Manoach DS (2003). Prefrontal cortex dysfunction during working memory performance in schizophrenia: reconciling discrepant findings. Schizophr Res, 60(2–3), 285–298. 10.1016/s0920-9964(02)00294-3 [DOI] [PubMed] [Google Scholar]
  44. Markiewicz CJ, Gorgolewski KJ, Feingold F, Blair R, Halchenko YO, Miller E, Hardcastle N, Wexler J, Esteban O, Goncavles M, Jwa A, & Poldrack R. (2021). The OpenNeuro resource for sharing of neuroscience data. eLife, 10. 10.7554/eLife.71774 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Melrose RJ, Zahniser E, Wilkins SS, Veliz J, Hasratian AS, Sultzer DL, & Jimenez AM (2020). Prefrontal working memory activity predicts episodic memory performance: A neuroimaging study. Behavioural Brain Research, 379. 10.1016/j.bbr.2019.112307 [DOI] [PubMed] [Google Scholar]
  46. Moeller S, Yacoub E, Olman CA, Auerbach E, Strupp J, Harel N, & Ugurbil K. (2010). Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel imaging with application to high spatial and temporal whole-brain fMRI. Magn Reson Med, 63(5), 1144–1153. 10.1002/mrm.22361 [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Moore ME, & Ross BM (1963). Context effects in running memory. Psychological Reports, 12(2), 451–465. 10.2466/pr0.1963.12.2.451 [DOI] [Google Scholar]
  48. Nagel IE, Preuschhof C, Li SC, Nyberg L, Backman L, Lindenberger U, & Heekeren HR (2011). Load modulation of BOLD response and connectivity predicts working memory performance in younger and older adults. J Cogn Neurosci, 23(8), 2030–2045. 10.1162/jocn.2010.21560 [DOI] [PubMed] [Google Scholar]
  49. Nitchie F, Casalvera A, Teferi M, Patel M, Lynch KG, Makhoul W, Sheline YI, & Balderston NL (2024). The maintenance of complex visual scenes in working memory may require activation of working memory manipulation circuits in the dlPFC: A preliminary report. Mental Health Science. 10.1002/mhs2.61 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Perlstein WM, Carter CS, Noll DC, & Cohen JD (2001). Relation of prefrontal cortex dysfunction to working memory and symptoms in schizophrenia. American Journal of Psychiatry, 158(7), 1105–1113. [DOI] [PubMed] [Google Scholar]
  51. Perlstein WM, Dixit NK, Carter CS, Noll DC, & Cohen JD (2003). Prefrontal cortex dysfunction mediates deficits in working memory and prepotent responding in schizophrenia. Biol Psychiatry, 53(1), 25–38. http://www.ncbi.nlm.nih.gov/pubmed/12513942 [DOI] [PubMed] [Google Scholar]
  52. Power JD, Mitra A, Laumann TO, Snyder AZ, Schlaggar BL, & Petersen SE (2014). Methods to detect, characterize, and remove motion artifact in resting state fMRI. NeuroImage, 84, 320–341. 10.1016/j.neuroimage.2013.08.048 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Robinson EC, Garcia K, Glasser MF, Chen Z, Coalson TS, Makropoulos A, Bozek J, Wright R, Schuh A, Webster M, Hutter J, Price A, Cordero Grande L, Hughes E, Tusor N, Bayly PV, Van Essen DC, Smith SM, Edwards AD,…Rueckert, D. (2018). Multimodal surface matching with higher-order smoothness constraints. NeuroImage, 167, 453–465. 10.1016/j.neuroimage.2017.10.037 [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Robinson EC, Jbabdi S, Glasser MF, Andersson J, Burgess GC, Harms MP, Smith SM, Van Essen DC, & Jenkinson M. (2014). MSM: A new flexible framework for Multimodal Surface Matching. NeuroImage, 100, 414–426. 10.1016/j.neuroimage.2014.05.069 [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Ross BM (1966a). Serial-order effects in two-channel running memory. Percept Mot Skills, 23(3), 1099–1107. 10.2466/pms.1966.23.3f.1099 [DOI] [PubMed] [Google Scholar]
  56. Ross BM (1966b). Serial order as a unique source of error in running memory. Percept Mot Skills, 23(1), 195–209. 10.2466/pms.1966.23.1.195 [DOI] [PubMed] [Google Scholar]
  57. Satterthwaite TD, Wolf DH, Erus G, Ruparel K, Elliott MA, Gennatas ED, Hopson R, Jackson C, Prabhakaran K, Bilker WB, Calkins ME, Loughead J, Smith A, Roalf DR, Hakonarson H, Verma R, Davatzikos C, Gur RC, & Gur RE (2013). Functional maturation of the executive system during adolescence. J Neurosci, 33(41), 16249–16261. 10.1523/JNEUROSCI.2345-13.2013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Schumacher EH, Lauber E, Awh E, Jonides J, Smith EE, & Koeppe RA (1996). PET evidence for an amodal verbal working memory system. Neuroimage, 3(2), 79–88. 10.1006/nimg.1996.0009 [DOI] [PubMed] [Google Scholar]
  59. Šidák Z. (1967). Rectangular Confidence Regions for the Means of Multivariate Normal Distributions. Journal of the American Statistical Association, 62(318), 626–633. 10.1080/01621459.1967.10482935 [DOI] [Google Scholar]
  60. Singleton WT (1978). Laboratory Studies of Skill. In Singleton WT (Ed.), The analysis of practical skills (pp. 16–43). Springer Netherlands. 10.1007/978-94-011-6188-6_2 [DOI] [Google Scholar]
  61. Smith S, & Nichols T. (2009). Threshold-free cluster enhancement: Addressing problems of smoothing, threshold dependence and localisation in cluster inference. NeuroImage, 44(1), 83–98. 10.1016/j.neuroimage.2008.03.061 [DOI] [PubMed] [Google Scholar]
  62. Smith SM, Beckmann CF, Andersson J, Auerbach EJ, Bijsterbosch J, Douaud G, Duff E, Feinberg DA, Griffanti L, Harms MP, Kelly M, Laumann T, Miller KL, Moeller S, Petersen S, Power J, Salimi-Khorshidi G, Snyder AZ, Vu AT,…Consortium, W. U.-M. H. (2013). Resting-state fMRI in the Human Connectome Project. Neuroimage, 80, 144–168. 10.1016/j.neuroimage.2013.05.039 [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Smucny J, Hanks TD, Lesh TA, & Carter CS (2023). Altered Associations Between Task Performance and Dorsolateral Prefrontal Cortex Activation During Cognitive Control in Schizophrenia. Biol Psychiatry Cogn Neurosci Neuroimaging, 8(10), 1050–1057. 10.1016/j.bpsc.2023.05.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Smyser CD, Snyder AZ, & Neil JJ (2011). Functional connectivity MRI in infants: exploration of the functional organization of the developing brain. Neuroimage, 56(3), 1437–1452. 10.1016/j.neuroimage.2011.02.073 [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Stein A, Iyer KK, Khetani AM, & Barlow KM (2021). Changes in working memory-related cortical responses following pediatric mild traumatic brain injury: A longitudinal fMRI study. Journal of Concussion, 5. 10.1177/20597002211006541 [DOI] [Google Scholar]
  66. Strike LT, Blokland Gabriella A.M., Hansell Narelle K., Nicholas G., Toga Arthur W., Thompson Paul M., de Zubicaray Greig I., McMahon Katie L., Wright Margaret J. (2023). Queensland Twin IMaging (QTIM). In: OpenNeuro. [Google Scholar]
  67. Strike LT, Hansell NK, Couvy-Duchesne B, Thompson PM, de Zubicaray GI, McMahon KL, & Wright MJ (2019). Genetic Complexity of Cortical Structure: Differences in Genetic and Environmental Factors Influencing Cortical Surface Area and Thickness. Cerebral Cortex, 29(3), 952–962. 10.1093/cercor/bhy002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Suzuki M, Kawagoe T, Nishiguchi S, Abe N, Otsuka Y, Nakai R, Asano K, Yamada M, Yoshikawa S, & Sekiyama K. (2018). Neural Correlates of Working Memory Maintenance in Advanced Aging: Evidence From fMRI. Frontiers in Aging Neuroscience, 10. 10.3389/fnagi.2018.00358 [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Tachibana A, Noah J, Bronner S, Ono Y, Hirano Y, Niwa M, Watanabe K, & Onozuka M. (2012). Activation of dorsolateral prefrontal cortex in a dual neuropsychological screening test: An fMRI approach. Behavioral and Brain Functions, 8(1). 10.1186/1744-9081-8-26 [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Uğurbil K, Xu J, Auerbach EJ, Moeller S, Vu AT, Duarte-Carvajalino JM, Lenglet C, Wu X, Schmitter S, Van de Moortele PF, Strupp J, Sapiro G, De Martino F, Wang D, Harel N, Garwood M, Chen L, Feinberg DA, Smith SM,…Yacoub, E. (2013). Pushing spatial and temporal resolution for functional and diffusion MRI in the Human Connectome Project. NeuroImage, 80, 80–104. 10.1016/j.neuroimage.2013.05.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Van Essen DC, Curtiss SW, Laumann T, Jenkinson M, Prior F, Glasser MF, Hodge M, Olsen T, Harwell J, & Marcus DS (2011). Informatics and Data Mining Tools and Strategies for the Human Connectome Project. Frontiers in Neuroinformatics, 5. 10.3389/fninf.2011.00004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Van Essen DC, Smith SM, Barch DM, Behrens TEJ, Yacoub E, & Ugurbil K. (2013). The WU-Minn Human Connectome Project: An overview. NeuroImage, 80, 62–79. 10.1016/j.neuroimage.2013.05.041 [DOI] [PMC free article] [PubMed] [Google Scholar]
  73. Van Snellenberg JX, Girgis RR, Horga G, van de Giessen E, Slifstein M, Ojeil N, Weinstein JJ, Moore H, Lieberman JA, Shohamy D, Smith EE, & Abi-Dargham A. (2016). Mechanisms of Working Memory Impairment in Schizophrenia. Biol Psychiatry, 80(8), 617–626. 10.1016/j.biopsych.2016.02.017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  74. Van Snellenberg JX, Slifstein M, Read C, Weber J, Thompson JL, Wager TD, Shohamy D, Abi-Dargham A, & Smith EE (2015). Dynamic shifts in brain network activation during supracapacity working memory task performance. Hum Brain Mapp, 36(4), 1245–1264. 10.1002/hbm.22699 [DOI] [PMC free article] [PubMed] [Google Scholar]
  75. Voigt K, Liang EX, Misic B, Ward PGD, Egan GF, & Jamadar SD (2023). Metabolic and functional connectivity provide unique and complementary insights into cognition-connectome relationships. Cerebral Cortex, 33(4), 1476–1488. 10.1093/cercor/bhac150 [DOI] [PMC free article] [PubMed] [Google Scholar]
  76. Wager TD, Spicer J, Insler R, & Smith EE (2014). The neural bases of distracter-resistant working memory. Cogn Affect Behav Neurosci, 14(1), 90–105. 10.3758/s13415-013-0226-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  77. Weinberger DR, & Radulescu E. (2016). Finding the Elusive Psychiatric “Lesion” With 21st-Century Neuroanatomy: A Note of Caution. American Journal of Psychiatry, 173(1), 27–33. 10.1176/appi.ajp.2015.15060753 [DOI] [PubMed] [Google Scholar]
  78. Welford AT, & Cambridge. University. Psychological Laboratory. Nuffield Research Unit into Problems of Ageing. [from old catalog]. (1958). Ageing and human skill. Published for the Trustees of the Nuffield Foundation by the Oxford University Press. [Google Scholar]
  79. Whitfield-Gabrieli S, Thermenos HW, Milanovic S, Tsuang MT, Faraone SV, McCarley RW, Shenton ME, Green AI, Nieto-Castanon A, LaViolette P, Wojcik J, Gabrieli JD, & Seidman LJ (2009). Hyperactivity and hyperconnectivity of the default network in schizophrenia and in first-degree relatives of persons with schizophrenia. Proc Natl Acad Sci U S A, 106(4), 1279–1284. 10.1073/pnas.0809141106 [DOI] [PMC free article] [PubMed] [Google Scholar]
  80. Williams JC, Tubiolo PN, Luceno JR, & Van Snellenberg JX (2022). Advancing motion denoising of multiband resting-state functional connectivity fMRI data. NeuroImage, 249. 10.1016/j.neuroimage.2022.118907 [DOI] [PMC free article] [PubMed] [Google Scholar]
  81. Williams JC, Zheng ZJ, Tubiolo PN, Luceno JR, Gil RB, Girgis RR, Slifstein M, Abi-Dargham A, & Van Snellenberg JX (2023). Medial prefrontal cortex dysfunction mediates working memory deficits in patients with schizophrenia. Biological Psychiatry: Global Open Science, 3, 990–1002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  82. Winkler AM, Ridgway GR, Webster MA, Smith SM, & Nichols TE (2014). Permutation inference for the general linear model. NeuroImage, 92, 381–397. 10.1016/j.neuroimage.2014.01.060 [DOI] [PMC free article] [PubMed] [Google Scholar]
  83. Winkler AM, Webster MA, Brooks JC, Tracey I, Smith SM, & Nichols TE (2016). Non-parametric combination and related permutation tests for neuroimaging. Human Brain Mapping, 37(4), 1486–1511. 10.1002/hbm.23115 [DOI] [PMC free article] [PubMed] [Google Scholar]
  84. Winkler AM, Webster MA, Vidaurre D, Nichols TE, & Smith SM (2015). Multi-level block permutation. NeuroImage, 123, 253–268. 10.1016/j.neuroimage.2015.05.092 [DOI] [PMC free article] [PubMed] [Google Scholar]
  85. Wu D, & Jiang T. (2019). Schizophrenia-related abnormalities in the triple network: A meta-analysis of working memory studies. Brain Imaging and Behavior, 14, 971–980. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supinfo1
NIHMS2051802-supplement-Supinfo1.pdf (15.6MB, pdf)
Supinfo2
NIHMS2051802-supplement-Supinfo2.docx (1.7MB, docx)

Data Availability Statement

Publicly available HCP data can be obtained from ConnectomeDB (https://db.humanconnectome.org/). QTIM study data can be obtained from openneuro.org (accession number ds004169).

Thresholded t-statistic maps, as well as peak significant t-statistics and percentage of significant grayordinates within each CAB-NP parcel, from all results described in Sections 3.1. and 3.2. are provided via BALSA (https://balsa.wustl.edu/study/0wrMq). Specific scene IDs are noted in figure captions where applicable.

RESOURCES