Effects of Locus Coeruleus Activation on n-Back Performance and Frontoparietal Activity

Working memory comprises an array of cognitive functions that enables conscious organisms to maintain and manipulate sensory information over brief periods (Baddeley, 1992). It is an essential component of learning, decision-making, and correlates with intelligence (Li et al., 2021; Schneider & McGrew, 2018; Unsworth, Fukuda, Awh, & Vogel, 2014). Working memory is typically measured by modality-specific tasks that present a set of information that must be retained for a few seconds and potentially manipulated (e.g., retaining a telephone number long enough to dial it). Tasks such as the n-back are commonly used in event-related fMRI to identify neural correlates of working memory function. During task administration, participants monitor a series of sequentially presented stimuli—typically letters or shapes—and determine if the current stimulus matches a previous letter in the sequence (Kirchner, 1958). This involves several component processes such as selective attention, active maintenance, updating, and interference mitigation. Neuroimaging studies have shown that engagement in working memory tasks such as the n-back or other executive function tasks elicits a significant increase in the BOLD hemodynamic responses within a network of interconnected cortical regions such as the premotor cortex, pFC, frontal poles, and posterior parietal cortex (PC) (Yaple, Stevens, & Arsalidou, 2019; Tomasi, Chang, Caparelli, & Ernst, 2007; Owen, McMillan, Laird, & Bullmore, 2005; Linden et al., 2003). Commonly referred to as the frontoparietal network, or task-positive network, activation within these regions is associated with goal-oriented cognition and behavior (Marek & Dosenbach, 2018).

A benefit of utilizing the n-back task is its capacity to parametrically manipulate working memory load. This is accomplished by increasing the length of the n-back sequences that must be retained and manipulated. Committing longer sequences to memory stresses working memory capacity and active maintenance, making a “three-back” task significantly more difficult than a “one-back.” This is evident in the respective accuracy score associated with each task—successful recall reliably decreases as working memory load increases. Functional neuroimaging studies of the n-back task reveal a dynamic relationship between the relative change of hemodynamic responses in frontoparietal regions and working memory load. Activation generally increases from easy to moderate levels of n-back difficulty (Wang et al., 2019; Braver et al., 1997) and then decreases near suprathreshold levels of capacity. These load-dependent responses have been identified in several regions of the frontoparietal network including the dorsolateral prefrontal cortex (DLPFC), PC, and SMA (Lamichhane, Westbrook, Cole, & Braver, 2020; Van Snellenberg et al., 2015; Cole, Yarkoni, Repovs, Anticevic, & Braver, 2012; Nyberg, Dahlin, Stigsdotter Neely, & Bäckman, 2009; Todd & Marois, 2004; Linden et al., 2003). This Gaussian trend in BOLD activation is typically interpreted as a change in task-related processing. In some cases, a significant decrease of hemodynamic responses in executive function areas (such as the DLPFC) may represent task disengagement (Callicott et al., 1999), which could reasonably emerge if task difficulty exceeds capacity.

The correlation between BOLD activation and task-related processing is a basic premise for theoretical frameworks such as the Compensation Related Utilization of Neural Circuits hypothesis. At a fixed level of difficulty, individuals with impaired working memory function demonstrate increased compensatory activation to achieve the same level of performance (Kang, Wang, & Malvaso, 2022; Nagel et al., 2011; Cappell, Gmeindl, & Reuter-Lorenz, 2010; Reuter-Lorenz & Cappell, 2008). Previous work by Nyberg and colleagues (2009) supported this hypothesis by comparing high- and low-performance individuals as determined by a median split of individual scores on the difficulty 3-back task. Despite decreased performance, the older and low performance groups showed significantly higher activations in the dorsal frontal cortex during the 1-back and 2-back tasks, respectively. A similar experimental approach employed by Jaeggi and colleagues (2007) yielded a similar finding—the high-performance group achieved significantly higher scores at lower levels of activation in several cortical regions including the DLPFC. These studies, in conjunction with a larger body of literature, suggests that changes in BOLD activation and related capacity-constrained responses predict individual differences in performance on a working memory task (Schneider-Garces et al., 2010).

One of the potential mechanisms responsible for individual differences in frontoparietal BOLD activation is the locus coeruleus noradrenergic (LC/NE) system. This brainstem nucleus has extensive connections to several regions of the frontoparietal network and is the primary source of noradrenaline in the mammalian cortex (Waterhouse, Predale, Plummer, Jensen, & Chandler, 2022; Chandler, 2016; Austin & Takaori, 1976). As a modulator of cortical excitability, NE signaling from LC neurons has a measurable impact on sensory processing (Kuo et al., 2017; Devilbiss, Page, & Waterhouse, 2006; Berridge & Waterhouse, 2003). Coordinated output from LC neurons could, in theory, enhance cognitive processing during working memory tasks (Bär et al., 2016; Mather, Clewett, Sakaki, & Harley, 2016; Agay, Yechiam, Carmel, & Levkovitz, 2014).

The LC/NE system has a significant impact on working memory processes in a broad range of contexts. Studies have explored how LC activity modulates memory encoding, consolidation, and retrieval, particularly in relation to emotional or salient stimuli (Clewett, Huang, Velasco, Lee, & Mather, 2018; Barsegyan, McGaugh, & Roozendaal, 2014; Sterpenich et al., 2006). Age-related memory impairments have also been linked with LC tissue integrity (Dahl et al., 2019). In extreme cases, damage to the LC is proposed to contribute to pathological memory loss in diseases such as Alzheimer (Gargano, Olabiyi, Palmisano, Zimmer, & Bilkei-Gorzo, 2023; Betts et al., 2019). According to theoretical models, the LC is capable of affecting working memory through NE signaling on cortical targets. This effectively increases the signal-to-noise ratio of prioritized cognitive functions. As such, NE activity of the LC is likely to have an impact on higher order cognitive functions involved in working memory tasks such as sustained attention, decision-making, and sensory processing (McBurney-Lin, Lu, Zuo, & Yang, 2019; Mather et al., 2016; Aston-Jones & Cohen, 2005).

The aim of this study was to determine if capacity-constrained responses in the frontoparietal network, and associated working memory capacity, rely on output from the LC. We hypothesized that LC activation would correspond to changes in frontoparietal activation, thereby enabling increased frontoparietal activation when working memory load was high. Using fMRI, we tested this hypothesis by first measuring LC activation during a parametrically designed verbal n-back task. Regression analyses were then applied to determine if LC activation correlated with successful recall at various levels of working memory load. Groups of high and low LC output were determined based on activation maps and contrasted to assess whether this corresponded to differences in cortical activation. Finally, we conducted an exploratory functional connectivity analysis using the LC as a seed region to determine if functional integration between the LC and frontoparietal network correlated with differences in working memory performance.

Twenty-one healthy individuals were recruited from the Tucson, Arizona, metropolitan area (mean age = 23.43 ± 4.69 years, 13 female participants). Sample size was based on power analysis, funding availability, and successful adherence to study requirements. All participants were right-handed, primarily English speaking, and free of neurologic conditions (determined by online screening). Actiwatches and sleep diary data were collected for 1 week before the scheduled date of the MRI scan. Actigraphy was cross-referenced with sleep diaries to ensure maintenance of a typical sleep–wake schedule. Participants were excluded if they met more than one of the following criteria: average sleep duration less than 7 hr, sleep efficiency less than 85%, or average sleep intervals occurring outside of a 10:00 p.m. to 9:00 a.m. timeframe. fMRI sessions occurred between 11:00 a.m. and 2:00 p.m. to control for circadian effects on arousal and cognition.

This study was conducted in accordance with the ethical standards outlined by the University of Arizona's institutional review board for the protection of human participants and the U.S. Army Human Research Protections Office. All participants provided informed consent before participating in any study procedures. Compensation was provided for their time and effort in accordance with the university's compensation guidelines. In addition, participants were assured of their right to withdraw from the study at any time without penalty or consequence. The confidentiality and privacy of all participants were strictly maintained throughout the study, and all data were anonymized to ensure confidentiality.

A power analysis was conducted based on effect size from a previous publications specifying LC fMRI activation related to successful memory retrieval (Sterpenich et al., 2006). With a desired statistical power of 0.80 and a significance level of 0.05, the analysis indicated that a minimum of 17 participants would be necessary. Consequently, based on funding constrains, our final sample size included 21 participants. This projects sufficient power of .89 to detect LC activation involved in memory recall. This analysis was conducted using the pwr package in R (Champely et al., 2020).

The n-back task is widely used to assess verbal working memory. Administration involves presenting a sequence of letters briefly displayed on a screen, followed by a response period. During this period, participants press a button to indicate if the current letter matches a letter seen previously by n iterations (i.e., 0-back, 1-back, or 2-back). Responses were automatically scored in E-prime as either a hit, correct rejection, false alarm, or no response. Raw scores were then converted into Critical Success Index (CSI) scores, which measure the probability of generating a correct response specifically to matching stimuli. This approach provides a more accurate measure of recall, as using total accuracy would be mathematically biased toward correctly rejected stimuli.

Working memory load was modulated at four levels of difficulty—0-back, 1-back, 2-back, and 3-back. Task progression included an interleaved pattern of crosshair fixation, followed by an n-back trial containing a sequence of 16 letters and 16 response periods. At the beginning of each n-back, an instruction slide was presented indicating the condition for a correct matching sequence. Letter stimulus duration (.5 sec) and response intervals (1.75 sec) were kept constant to ensure the same number of trial administrations per individual. Each block lasted 42 sec. Letter sequences were initially created through randomization pulled from a lexicon of following possibilities: B, T, G, P, V. Sets were then created from randomized letter sequences to ensure a fixed number of correct matching conditions in each trial. Because of randomization, some blocks contain lure trials at a rate consistent with chance. n-Back blocks were randomized for each participant to control for time-on-task effects. In total, each participant completed 12 n-back blocks—three iterations at each level of difficulty.

Before entering the scanner, participants were given instructions and practice trials were administered until performance scores reflected adequate comprehension (p > .50). Participants viewed stimuli presented on a screen via a 32-channel head coil equipped with a display mirror. Responses were logged using an MRI-compatible button box.

MRI scans were acquired with a 3-T Skyra scanner (Siemens). Anatomical scans consisted of a T1-weighted 3-D magnetization prepared rapid gradient echo sequence (repetition time = 2.15 sec, echo time = 2.71 msec, flip angle = 12°, n-slice = 256, voxel = .73 × .73 × .73 mm). Functional scans were collected using a T2 weighted interleaved sequence (repetition time = 2 sec, echo time = 36 msec, flip angle = 12°, voxel = 2 × 2 × 2 mm). Data preparation and analyses were conducted in MATLAB 2016b using SPM12 (Wellcome center for Neuroimaging https://www.fil.ion.ucl.ac.uk/spm). Preprocessing of fMRI data consisted of slice timing correction, realignment to the mean functional image, co-registration between structural and functional scans, tissue segmentation, and normalization to Montreal Neurological Institute (MNI) space according to forward deformation fields. Functional images were then smoothed using a Gaussian kernel (FWHM = 6 mm). Kernel size was chosen to maintain a balance between noise reduction and fine detail preservation; a larger kernel could confound activation within the LC contributed from anatomically distinct regions, whereas a smaller kernel could significantly decrease the signal-to-noise ratio (Triana, Glerean, Saramäki, & Korhonen, 2020). A 6-mm kernel has also been shown as an effective metric for spatial smoothing in assessing LC activation (Grueschow, Kleim, & Ruff, 2022; Suttkus, Schumann, de la Cruz, & Bär, 2021; Murphy, O'Connell, O'Sullivan, Robertson, & Balsters, 2014). Potential outlier scans were identified using the Artifact Detection Tool identifying framewise displacement above 1.0 mm or global BOLD signal changes above 3 SDs (Power et al., 2014; Whitfield-Gabrieli, Nieto-Castanon, & Ghosh, 2011). First-level analyses applied a classic generalized linear model approach in which the canonical hemodynamic response function was fit to task-related block design. Random effects analyses (Friston & Ashburner, 2007) were conducted to quantify group-level differences in activation at each incremental level of n-back difficulty relative to the 0-back condition (control). Individual performance metrics were subsequently included as primary covariates for regression analyses.

LC activation during the working memory task was assessed using a 10-mm ROI created with MarsBar (Brett, Anton, Valabregue, & Poline, 2002) to encapsulate MNI coordinates previously reported to have a high probability of containing LC neurons (Keren, Lozar, Harris, Morgan, & Eckert, 2009). This mask was applied to random effects analyses to determine potential LC activation at each level of the n-back relative to the control condition. The weighted average of BOLD signal contrast of the ROI, or eigenvariates, were extracted at nearest LC coordinates to capture the dominant temporal fluctuations within that region for each participant. These values were then utilized in a multiple regression approach to subsequently determine if differences in LC activation predict CSI within this region.

To account for potential confounded activation in the LC region produced from pulsatile movement of cerebrospinal fluid (CSF), an additional regression analysis was conducted to measure potential overlap between BOLD activation and CSF artifact. Although pulsatile movement from CSF flow elicits relatively small movement within the LC region of the brainstem (Soellinger, Ryf, Boesiger, & Kozerke, 2007; Enzmann & Pelc, 1992), this should still be considered. A segmented anatomical mask of CSF fluid, provided from SPM12 segmentation, was applied to participant's functional images for localization of the fourth ventricle. Raw BOLD signal within the CSF fluid of the fourth ventricle was extracted. Principal component decomposition was then applied to the voxel cluster. The first principal component of extracted CSF signal was included as a covariate in first-level analyses. Group-level influence of CSF was determined and compared with activation analyses.

Seed-based connectivity maps were calculated in the function connectivity toolbox CONN, following the default preprocessing pipeline, and estimated/visualized using the 164 HPC-ICA networks (Nieto-Castanon, 2022; Nieto-Castanon & Whitfield-Gabrieli, 2022) and Harvard–Oxford atlas ROIs (Desikan et al., 2006). Nuisance covariates were created and included as individual-level covariates including functional confounding timeseries effects of CSF (COMPCOR) (Behzadi, Restom, Liau, & Liu, 2007), white matter, outlier scans, and movement parameters with first-order derivatives (Friston, Williams, Howard, Frackowiak, & Turner, 1996). Functional connectivity strength was represented by Fisher-transformed bivariate correlation coefficients from a weighted general linear model (Nieto-Castanon, 2020), defined separately for each pair of seed and target areas, modeling the association between their BOLD signal timeseries. To compensate for possible transient magnetization effects at the beginning of each run, individual scans were weighted by a step function convolved with the SPM12 canonical hemodynamic response function and rectified.

Group-level analyses were performed using a general linear model (Nieto-Castanon, 2020). For each individual voxel, a separate generalized linear model was estimated, with first-level connectivity measures at this voxel as dependent variables (one independent sample per participant and one measurement per task or experimental condition, if applicable) and groups or other subject-level identifiers as independent variables. Voxel-level hypotheses were evaluated using multivariate parametric statistics with random-effects across participants and sample covariance estimation across multiple measurements. Inferences were performed at the level of individual clusters (groups of contiguous voxels). Cluster-level inferences were based on parametric statistics from Gaussian random field theory (Worsley et al., 1996). Results were thresholded using a combination of a cluster-forming p < .005 voxel-level threshold, and a familywise corrected p-FWE < 0.05 cluster-size threshold (Nieto-Castanon, 2020; Chumbley, Worsley, Flandin, & Friston, 2010).

CSI was chosen as the metric for evaluating working memory performance owing to its ability to capture rates of successful recall (CSI = hits/(hits + misses+ false alarms)). Traditional metrics such as total accuracy were not included as these are biased toward correct rejection of nonmatching stimuli, rather than identification of n-back sequences. Group-level performance at incremental levels of the n-back are shown in Figure 1. As expected, increasing task difficulty was associated with a significant decrease in performance, F(2, 21) = 43.25, p < .0001. Post hoc analyses indicate that group-level working memory capacity was exceeded in the 3-back condition. Out of 10 possible correct matching conditions, random chance would predict a 95% confidence interval of (5 ± σ_x̅ *T). The average number of correct responses in the 3-back condition is not significantly different from chance—95% CI [4.61, 5.73]. Comparatively, the 2-back condition appears to be near working memory capacity—95% CI [6.72, 8.42].

We conducted random effects analyses at each level of task difficulty to assess changes in activation relative to the 0-back condition, thereby examining the impact of working memory load rather than baseline resting states. To specifically investigate hemodynamic responses associated with the LC, we defined a ROI in the pontine tegmentum region of the brainstem. The ROI parameters were based on coordinates identified in previous research as having the highest probability of containing LC neurons (Keren et al., 2009). Significant BOLD activation within the ROI was observed across all task levels (Figure 2). Notably, hemodynamic responses exhibited a Gaussian trend at various task difficulties, with peak activation occurring near the threshold of working memory capacity in the 2-back condition, followed by a decrease in activation in the 3-back condition. Activation for reported values are not significantly confounded by CSF.

Regression analyses were conducted for each task to measure if BOLD activation within the ROI predicted CSI as a measurement of working memory performance while controlling for age and gender (Figure 3A). Several voxels within the LC regions showed a significant positive correlation with CSI scores in the 2-back condition (p_FWE < .05). Max eigenvariate values were extracted from the ROI at MNI coordinates previously reported to have a high probability of containing LC neurons (MNI:4, −36, −24; Keren et al., 2009). Functional responses strongly predicted CSI in the 2-back condition (p < .001, β = .71, R_adj² = .47, d = 2.71; Figure 3B). No significant correlations were found in the 1-back or 3-back conditions (p_FWE > .1). Activation for reported values are not significantly confounded by CSF flow.

To investigate whether differences in LC activity influence task-related activation in the frontoparietal network, we included LC contrast eigenvariates contrasted with the control condition from matching n-back blocks as a covariate of interest in each random effects analysis. In the 1-back condition, relative increases in LC eigenvariate values corresponded with potentiated responses in bilateral parietal regions (L: p_fdr < .001, d = 4.20, β = .70; R: p_fdr < .001, d = 3.19, β = .62), bilateral DLPFC (L: p_fdr = .002, d = 2.93, β = .76; R: p_fdr < .001, d = 2.55, β = .61), and anterior cingulate cortex (p_fdr = .001, d = 2.28, β = .98. In the 2-back condition, similar potentiated responses were observed in the parietal lobes (L: p_fdr < .001, d = 3.23, β = 1.07; R: p_fdr < .001, d = 4.32, β = .95) and left DLPFC (p_fdr = .0001, d = 2.47, β = 1.03). However, the 3-back condition yielded no significant responses in the parietal lobes, although potentiated responses were noted in the right DLPFC (p_fdr < .001, d = 2.75, β = 1.103) and ACC (p_fdr < .0001, d = 3.24; β = 1.14). These results illustrate changes in task-related activation within the frontoparietal network and significantly correspond with an extracted metric of hemodynamic responses within the LC region, suggesting a modulatory role of the LC in task performance (Figure 4).

Seed-based correlational analyses were subsequently conducted to show functional connectivity with the LC at the incremental levels of working memory load. This approach aims to build on the previous analysis showing a direct association between cortical-LC integration and performance. CSI was included as the primary covariate of interest as a metric of recall performance. Regions highlighted in blue show areas with significantly altered functional connectivity with the LC that correlated with higher CSI scores. Significant results were found in the 2- and 3-back conditions, whereas no significant changes in functional connectivity were associated with performance in the 1-back task. In the 2-back condition, improved performance was significantly associated with increased negative functional connectivity between the LC and the bilateral postcentral gyrus (Brodmann's area [BA] 2 and 3, cluster p_fdr < .02) as well as a small portion of the precuneus (BA 5, cluster p_fdr < .02; Figure 5). In the 3-back condition, improved performance was associated with more negative functional connectivity between the LC and several regions including: bilateral mid temporal lobes (BAs 20 and 21, cluster p_fdr < .01), right retrosubicular area (BA 48 cluster p_fdr < .001), left angular gyrus (BA 39, cluster p_fdr < .02), and mid-anterior cingulate (BA 24, cluster p_fdr < .02).

We tested whether LC activation was associated with load-dependent cortical responses during a parametrically modulated working memory task. Incremental manipulation of n-back difficulty demonstrated LC activation to be load dependent and capacity constrained. BOLD signal within the LC region increased with higher working memory demands but decreased once working memory capacity was exceeded. This rise and fall trend in activation bears a striking similarity to capacity-constrained responses previously demonstrated in the PC, SMA, and DLPFC. Considering that the LC receives input from prefrontal regions, it is likely executive functions may be modulating LC responses to correspond with cognitive demand (Arnsten, 2011; Jodo, Chiang, & Aston-Jones, 1998; Sara & Hervé-Minvielle, 1995).

Several theories of LC/NE function postulate optimal LC activation modulates cortical gain, thereby prioritizing task-related processes (Unsworth & Robinson, 2017; Mather et al., 2016; Aston-Jones & Cohen, 2005). According to this perspective, there should be a clear relationship between LC activation and working memory capacity. We tested this by correlating successful working memory recall (CSI) with LC activation. BOLD responses within the LC region correlated positively with task performance in the 2-back condition (p < .001, β = .71), but not in the 1-back or 3-back condition. Even at liberal significance thresholds (p < .10), this correlation was not evident in any other task. As such, optimization effects appear to be demand specific. Animal research suggests task-related optimization is typically associated with the phasic activity of LC neurons (Clayton, Rajkowski, Cohen, & Aston-Jones, 2004; Aston-Jones, Rajkowski, & Cohen, 1999; Usher, Cohen, Servan-Schreiber, Rajkowski, & Aston-Jones, 1999; Aston-Jones, Rajkowski, Kubiak, & Alexinsky, 1994). Intracranial recordings show improved performance on executive function tasks, and focused attention (Vazey, Moorman, & Aston-Jones, 2018) is accompanied by a transition from tonic LC activity—observed in mind-wandering—to phasic activity associated with task optimization. Building on this notion, fMRI research suggests significant increases in task performance and LC BOLD signal represents a similar transition between tonic and phasic modes of activity (Minzenberg, Watrous, Yoon, Ursu, & Carter, 2008; See also Astafiev, Snyder, Shulman, & Corbetta, 2010). Although neural activity is not directly evident with fMRI, it is likely that the conjunction of n-back optimization and significant change in hemodynamic response is consistent with phasic LC activity.

The specificity of task optimization poses additional implications for the potential cognitive mechanisms involved. The adaptive gain hypothesis argues that optimal activation of the LC is contingent on “task-utility,” which occurs when there is a perceived benefit in outcome. This assertion is supported by the fact that evoked potentials in the LC respond robustly to reward-based cues or changes in reward contingency (Bouret & Richmond, 2015; Bouret & Sara, 2004; Aston-Jones et al., 1994). In other words, motivation appears to be a cognitive promotor of optimal LC activation (Aston-Jones & Cohen, 2005). In the context of the current study, “task utility” poses a logical explanation of why certain conditions failed to elicit a significant correlation between LC activity and performance. The 1-back condition, for example, is essentially a pattern recognition task in which half of the participants received a perfect score. Assuming that attentional capacity is modulated according to situational demand, the simple 1-back task would likely require minimal cognitive resources to adequately complete. The same principle applies to the 3-back condition—excessive task difficulty presumably discourages full attentional investment, as this could be considered a waste of cognitive resources. Even if participants are highly motivated to complete the task, if the condition is beyond their capability, they may not perceive any benefit in maximizing their attentional capacity. This can lead to disengagement or a redirection of their attention to alternative problem-solving strategies (Van Snellenberg et al., 2015). In the latter case, this would likely direct cognition away from sensory processing. Therefore, the requirements of task utility are not met in either condition, which also fail to draw a significant correlation between LC activation and performance. The 2-back condition, however, provides a balance between participant capability and difficulty. Increasing attentional capacity in this task is likely to produce a measurable performance increase. This interpretation is consistent with the changes in LC BOLD activation and its correlation with task performance.

Although current theory portrays the LC as a modulator of cortical gain, it is unclear whether this effect occurs globally or within specific regions. Theoretical models such as the Arousal Based Competition model (Mather & Sutherland, 2011) suggest prioritization of certain cognitive functions involves competition between active regions. If true, NE effects on cortical activation may vary depending on the context. To investigate whether this applies to verbal working memory, BOLD signal LC eigenvariates were regressed on cortical activation to reveal potential correlation with hemodynamic response in working memory areas. As predicted, the degree of activation in the LC region correlated with increased hemodynamic responses within regions of the DLPFC and parietal cortices for each contrast. Both areas are associated with cognitive functions integral to n-back performance. Capacity limits of visual short-term memory have been associated with differences in fMRI activity in the parietal cortices (Xu & Chun, 2006; Todd & Marois, 2004). In addition, the pFC has been consistently identified as a primary locus of working memory capacity, owing to executive role in several working memory processes such as active maintenance, attention, and interference mitigation (Barbey, Koenigs, & Grafman, 2013; Miller et al., 2001; Barch et al., 1997). These results provide further elucidation of a potential mechanism between LC activation and working memory: Increased LC activation may sustain higher activation in regions associated with task-engagement/active maintenance and visual short-term memory, potentially facilitating improved performance.

The nature of the functional integration between the frontoparietal network and LC has, to our knowledge, not previously been studied in humans. To elucidate a potential mechanism of optimal LC activation, we conducted a seed-to-voxel functional connectivity analysis using performance as a continuous regressor for each level of working memory load. Results revealed a consistent trend of increasing negative functional connectivity in the 2-back and 3-back tasks. Those who performed well on the either task demonstrated greater anticorrelated connectivity between the LC and several cortical regions depending on task difficulty. One hypothesis explaining this result is top–down regulation of LC activity. Animal research shows that cortical regions such as the pFC can provide inhibitory input to the LC as a mechanism to elicit phasic bursts of activity associated with task optimization (Jodo et al., 1998; Sara & Hervé-Minvielle, 1995). It is possible that enhanced negative functional connectivity with these cortical regions may represent a similar mechanism, whereby integrated inhibitory input facilitates phasic activity within the LC region.

An alternative explanation for the increased negative functional connectivity between the LC and cortex is that NE suppression enhances network efficiency. By inhibiting cognitive functions not directly essential for task completion, this suppression may improve network efficiency, a phenomenon previously linked to enhanced performance on the n-back task (Stanley et al., 2015). This hypothesis is supported by findings from our functional connectivity analysis, which indicate that successful n-back performance correlates with regions exhibiting heightened negative functional connectivity. BAs 2, 3, and 5 are associated with functions such as tactile sensory processing, proprioception, and spatial discrimination, which are not directly pertinent to n-back completion. Even under challenging task conditions, cognitive functions like error monitoring (BA 24), language comprehension (BA 39), higher-order visual processing (BA 20 and 21), or spatial navigation (BA 48) are unlikely to significantly contribute to n-back task performance compared with the potential benefits of prioritizing cognitive resources toward sensory processing and working memory. It should also be noted that the combination of enhanced activation and inhibition of functionally segregated regions is consistent with theoretical assertions made by the arousal-based competition models of NE function: NE tone can potentiate activation in prioritized regions while inhibiting lower priority regions (Mather et al., 2016; Lee, Sakaki, Cheng, Velasco, & Mather, 2014; Salgado, Köhr, & Treviño, 2012). This suggests that enhanced negative functional connectivity between the LC and cortical/subcortical regions occurs when NE activity is simultaneously increased in task-active areas while being inhibited in regions not directly involved in successful task completion. The regression analysis provides partial evidence supporting this interpretation, showing enhanced activation in the parietal cortices and DLPFC that correlates with the BOLD signal in the LC region.

In conclusion, this study provides a characterization of LC activation during a parametrically designed working memory task. Hemodynamic responses in the LC are both load dependent and capacity constrained. This suggests that executive input may modulate task-related processing via NE output from the LC. Activation within the LC region predicts individual differences in working memory performance and is associated with increased hemodynamic responses in the superior parietal lobe and DLPFC. Functional connectivity analyses suggest that task optimization also involves anticorrelated responses between LC and cortical regions, potentially enhancing network efficiency resulting in improved memory recall. These findings align with previous evidence demonstrating a transition to phasic activity within the LC region, although further evidence is needed to fully support this hypothesis in the context of fMRI.

Corresponding author: David Negelspach, Psychiatry, University of Arizona Medical Center - University Campus, 1625 North Campbell Avenue, Tucson, AZ, or via e-mail: [email protected].

De-identified preprocessed fMRI data may be shared upon request from the corresponding author.

David Negelspach: Conceptualization; Data curation; Formal analysis; Visualization; Writing—Original draft; Writing—Review & editing. Anna Alkozei: Funding acquisition; Investigation; Methodology; Writing—Review & editing. Alisa Huskey: Formal analysis; Writing—Review & editing. William D. S. Killgore: Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Supervision; Validation; Writing—Review & editing.

This research was supported by the Congressionally Directed Medical Research Program (CDMRP) Discovery Award and the U.S. Army Medical Research Acquisition Activity, grant award number: W81XWH1910074.

Retrospective analysis of the citations in every article published in this journal from 2010 to 2021 reveals a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .407, W(oman)/M = .32, M/W = .115, and W/W = .159, the comparable proportions for the articles that these authorship teams cited were M/M = .549, W/M = .257, M/W = .109, and W/W = .085 (Postle and Fulvio, JoCN, 34:1, pp. 1–3). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article's gender citation balance.