Acute psychosocial stress modulates neural and behavioral substrates of cognitive control

Abstract

Acute psychosocial stress affects learning, memory, and attention, but the evidence for the influence of stress on the neural processes supporting cognitive control remains mixed. We investigated how acute psychosocial stress influences performance and neural processing during the Go/NoGo task—an established cognitive control task. The experimental group underwent the Trier Social Stress Test (TSST) acute stress induction, whereas the control group completed personality questionnaires. Then, participants completed a functional magnetic resonance imaging (fMRI) Go/NoGo task, with self‐report, blood pressure and salivary cortisol measurements of induced stress taken intermittently throughout the experimental session. The TSST was successful in eliciting a stress response, as indicated by significant Stress > Control between‐group differences in subjective stress ratings and systolic blood pressure. We did not identify significant differences in cortisol levels, however. The stress induction also impacted subsequent Go/NoGo task performance, with participants who underwent the TSST making fewer commission errors on trials requiring the most inhibitory control (NoGo Green) relative to the control group, suggesting increased vigilance. Univariate analysis of fMRI task‐evoked brain activity revealed no differences between stress and control groups for any region. However, using multivariate pattern analysis, stress and control groups were reliably differentiated by activation patterns contrasting the most demanding NoGo trials (i.e., NoGo Green trials) versus baseline in the medial intraparietal area (mIPA, affiliated with the dorsal attention network) and subregions of the cerebellum (affiliated with the default mode network). These results align with prior reports linking the mIPA and the cerebellum to visuomotor coordination, a function central to cognitive control processes underlying goal‐directed behavior. This suggests that stressor‐induced hypervigilance may produce a facilitative effect on response inhibition which is represented neurally by the activation patterns of cognitive control regions.

We showed enhancements in performance on a cognitive control task following exposure to an acute psychosocial stressor. Additionally, using multivariate pattern analysis of fMRI data, we showed stress‐related differences in activation patterns within regions associated with visuomotor coordination—a skill central to successful cognitive control.

Practitioner Points.

1. Acute stress may facilitate cognitive control, thereby enhancing cognitive performance.

2. Stress‐related task performance may be represented in the activity of control‐related brain regions.

1. INTRODUCTION

Acute stress (hereafter “stress”) stems from the perception of situations or contexts as threatening and taxing to one's adaptive capacity (Lazarus, 1966), and is accompanied by a multilevel response comprising physiological, psychological, and behavioral dimensions. Physiologically, stress activates the body's regulatory and neuroendocrine systems, including the sympathetic–adrenal–medullary (SAM) and the hypothalamic–pituitary–adrenal (HPA) axes, which have complementary and often overlapping effects (Ulrich‐Lai & Herman, 2009). These systems and their hormonal outputs produce a coordinated physiological response that supports adaptive coping with stress and the maintenance of homeostasis. Shortly after a stressful event, increases in SAM activity induces the canonical “fight‐or‐flight” response, with effects including hypervigilance and elevations in levels of stress mediators (e.g., epinephrine and norepinephrine), heart rate, blood pressure, and peripheral vasoconstriction. Compared to the rapid, short‐lived SAM axis response, the HPA axis response is delayed and prolonged. Its primary action is the gradual release of glucocorticoids (e.g., cortisol) shortly after stress exposure, reaching peak levels about 15–20 min after stress exposure (Droste et al., 2008). In the immediate aftermath of stress, cortisol acts to restore energy reserves to promote recovery. On a longer timescale, it acts to downregulate SAM and HPA activity after the stressor has passed (Joëls et al., 2011; Joëls & Baram, 2009; Ulrich‐Lai & Herman, 2009).

These stress‐sensitive axes also interact at a brain systems level, exerting neuromodulatory effects that serve to promote adaptive coping. Specifically, it has been proposed that the stress response involves the salience network (SN), default mode network (DMN), and distributed cognitive control networks (CCNs) (Hermans et al., 2014; van Oort et al., 2017). According to a so‐called triple network model, dynamic stress‐related changes within and between these networks may underlie the cognitive‐behavioral effects of stress and may also have implications for stress‐related psychopathologies (B. Menon, 2019; V. Menon, 2011; van Oort et al., 2017). The SN, which includes the amygdala, dorsal anterior cingulate cortex, and anterior insula, is active in response to the processing and integration of “homeostatically relevant” stimuli (Seeley, 2019). The DMN includes the medial prefrontal cortex, posterior cingulate cortex, medial temporal lobe, and angular gyrus and is involved in internally‐focused, self‐referential mentation (e.g., rumination, mind‐wandering) (Andrews‐Hanna et al., 2014; Binder et al., 1999; Greicius et al., 2003; Raichle, 2015). The CCNs, by contrast, are involved in higher‐order cognitive control processes (e.g., conflict monitoring, attentional control) and subsume regions such as the dorsolateral prefrontal cortex, dorsomedial prefrontal cortex, frontal eye fields, and posterior parietal lobe (Berryhill et al., 2011; Casini & Ivry, 1999; Cole & Schneider, 2007; Hermans et al., 2014; Ridderinkhof et al., 2004; van Oort et al., 2017; Weissman et al., 2006). Interestingly, according to one perspective, the SN may be a subnetwork of one such network called the cingulo‐opercular network, pointing to the highly interrelated nature of these brain systems (Ji et al., 2019). The triple network model further suggests that stress is associated with increased SN (to promote threat‐related hypervigilance) and DMN activity (to support internally focused coping) (Dedovic et al., 2009; Soares et al., 2013; van Oort et al., 2017). However, there appears to be a nonlinear relationship between stress and activity in the CCNs (van Oort et al., 2017), perhaps underscoring the various stress‐related effects on multiple domains of cognition.

Stress is thought to influence cognition by biasing attention toward salient (i.e., stress‐related) information processing. Thus, cognitive resources are reallocated to cope with the current stressor (or additional potential stressors), thereby limiting resources necessary for goal‐directed behavior (Shields et al., 2016). This theoretical framework is supported by empirical evidence showing stress‐related shifts in attention allocation toward threats and away from presumably less salient information (Booth & Sharma, 2009; Dandeneau et al., 2007; Hancock, 1989). Stress also increases distraction, which in some cases may manifest as post‐event processing, or internally focused, self‐relevant intrusive thoughts about the stressful event (Rachman et al., 2000; Wells & Papageorgiou, 1995). These stress‐related attentional shifts are coupled with impairments in cognitive processes supporting goal‐directed behavior, including working memory (Al'Absi et al., 2002; Elzinga & Roelofs, 2005; Oei et al., 2006; Schoofs et al., 2008, 2009; Shields et al., 2016), cognitive flexibility (Alexander et al., 2007; Plessow et al., 2011; Shields et al., 2016), psychomotor vigilance (Hancock, 1989; Olver et al., 2014; Qian et al., 2015), divided attention (Vedhara et al., 2000), and set shifting (Orem et al., 2008; Plessow et al., 2012).

It has also been proposed that stress impacts cognitive control engagement via increased reliance on habitual behavior at the expense of more controlled behavior (Schwabe & Wolf, 2009). Thus, based on this perspective, a reasonable prediction would be that stress impairs processes like inhibition that are related to successful cognitive control. However, empirical evidence suggests that the effects of stress on inhibitory control processes are more nuanced. While most studies have shown stress‐related impairments in inhibition (Mueller et al., 2010; Roos et al., 2017; Sänger et al., 2014; Scholz et al., 2009; Starcke et al., 2016), others have shown no effect (McGrath et al., 2016) or even stress‐related enhancements (Schwabe et al., 2013). Critically, this effect appears to be modulated by acute cortisol reactivity, such that greater cortisol reactivity predicted enhanced inhibition (Schwabe et al., 2013; Shields et al., 2015). This suggests that acute stress may affect cognition through the actions of stress hormones.

Evidence from the human neuroimaging literature suggests that these cognitive‐behavioral effects of stress may map on to changes in brain activity. For example, Chang and colleagues showed that acute stress enhanced inhibitory control, and that better inhibitory control was associated with greater connectivity between the superior/middle frontal gyrus and the striatum among participants in the stress group (Chang et al., 2020). By contrast, chronic stress appears to be associated with impaired cognitive control. Liston and colleagues showed that chronically stressed participants show deficits in attentional control and disrupted patterns of functional connectivity within a frontoparietal CCN that includes that the dorsolateral prefrontal cortex (Liston et al., 2009). And Mueller and colleagues showed that adolescents who experienced early life stress showed deficits in cognitive control and increased activity in cognitive control regions, including the inferior frontal cortex and the striatum (Mueller et al., 2010). Event‐related potential studies in this area also show inconsistent results, with evidence for both stress‐related enhancements (Dierolf et al., 2018) and impairments (Jiang & Rau, 2017) in neural activity associated with inhibitory control.

The present preregistered study aimed to contribute to this mixed literature by examining the effects of acute psychosocial stress on the neural processes supporting performance during a Go/NoGo task. To induce a stress response, we used the Trier Social Stress Test (TSST) (Kirschbaum et al., 1993), which involves giving a speech to an “expert” with little preparation, followed by a serial subtraction task. The TSST is an effective stressor because (1) it is an environment of high pressure and low control, and (2) it presents a sense of salient social‐evaluative threat. It also reliably induces a physiological stress response (Dickerson & Kemeny, 2004). We also employed a novel Go/NoGo task that uses traffic lights as response cues to assess response inhibition (Ceceli et al., 2020). Optimal performance on this task requires cognitive control and sustained attention, and may thus prove to be quite difficult, given the attentional and behavioral stress‐related effects described above. Further, the traffic light response cues offer a unique opportunity to leverage preexisting habits, and, given the effects of stress on habitual behavior, may add an additional element of difficulty to the task.

Given that the effects of stress on cognitive control appear to be nuanced, we broadly hypothesized that acute stress would affect task performance on this Go/NoGo task, and that this would be reflected in changes in regions subsumed by brain networks implicated in the stress response (e.g., SN, DMN, CCNs). We expected that the effects of acute stress would be especially pronounced on trials requiring the most cognitive control, that is, those where the color‐action contingencies are incongruent with the real‐world associations (i.e., NoGo Green). Successful responding on these trials in particular requires inhibition of the habitual response to a cue that would normally warrant a response, that is, to withhold a response to a green traffic light, which has been strongly associated with a “Go” response in daily life. We also hypothesized that there would be a speed‐accuracy trade off on Go trials, especially those on which the color‐action contingencies are incongruent with the real‐world associations (i.e., Go Red). If acute stress impairs performance, we may see faster responding but worse performance accuracy during these specific trials. This may be due to hypervigilance to threats following stress exposure, which may translate into psychomotor vigilance and increased likelihood to respond (vs. withhold a response) when a cue appears without taking into account the specific rule for that response cue. On the other hand, if acute stress enhances performance, we may see slower responding but better performance accuracy, as attentional resources are focused on deliberate, accurate responses.

2. MATERIALS AND METHODS

2.1. Participants

This study was conducted in accordance with protocols approved by the Rutgers University Institutional Review Board. The data collection strategy, hypotheses, and analytic plan were preregistered prior to data collection through the Open Science Framework. Eighty‐two right‐handed participants were recruited from the Rutgers University‐Newark campus and the surrounding community. Eligibility to participate in the study was based on the following criteria: between 18 and 40 years old; for females, no pregnancy or use of hormonal contraceptives; absence of standard MRI contraindications (including presence of metal in or on the body and claustrophobia); and normal color vision. Although history of neurologic or psychiatric diagnosis was not part of these exclusion criteria, this information was collected during participant prescreening as way to identify participants who could potentially have an adverse reaction to the task or stress manipulation. Seven participants endorsed having such a diagnosis, including attention‐deficit/hyperactivity disorder and depression. This subset of the sample did not differ significantly by group or sex. One participant did not finish the study due to claustrophobia, and another did not perform the experiment due to a positive pregnancy result. Three participants were excluded from the analysis due to task performance below chance (i.e., <50%), one of whom also had insufficient cortisol data. The final analytic sample was 77 participants (48.1% female, mean age of 20.79 years [range = 18–29 years; SD = 2.41], 79.2% non‐White), with 39 participants in the Stress group and 38 participants in the Control group. In post hoc sensitivity analyses, the two participants who were excluded solely due to poor performance were added back to the analytic sample. When these participants were included, the patterns of statistical significance reported below remained similar. Effect sizes and p values differed negligibly, by less than a tenth of a point.

2.2. Procedure

To control for circadian fluctuations in cortisol, all participants completed the study between 1 and 5 p.m., as it is during this time interval that diurnal cortisol rhythms are relatively stable and acute changes are better able to be detected (Dickerson & Kemeny, 2004). The study procedures are outlined in Figure 1. Upon arrival, participants provided informed consent and rested for 15 min. After this resting period, baseline blood pressure and cortisol levels were measured. Then participants were asked about their college major (if applicable) and their dream career and given instructions for the stress manipulation, which they completed during a 16‐min interval. Blood pressure and cortisol measurements were taken immediately after the stress manipulation, and participants were prepared to enter the MRI scanner. Once in the scanner, participants completed two 8‐min runs of the Go/NoGo task with stimuli presented as congruent with real‐world associations (i.e., Go Green, NoGo Red). Participants provided subjective stress ratings and another cortisol measurement at about 40 min after the end of the stress manipulation, which is as close to the expected peak of induced stress as the protocol would allow. They then completed two 8‐min runs of the Go/NoGo task with stimuli presented as incongruent with real‐world (i.e., Go Red, NoGo Green), after which they provided another cortisol sample. Participants were then instructed to remain lying still and awake in the scanner as they underwent a 12‐min resting state sequence followed by acquisition of structural images. After exiting the scanner, final blood pressure and cortisol measures were taken, and participants completed a battery of questionnaires related to their task experience. Participants were then debriefed and compensated $50.

FIGURE 1.

Experimental timeline. Times are relative to stress onset (0 min).

2.3. Stress induction and measures

2.3.1. Stress induction

Participants were pseudorandomly assigned to complete either a stress induction (i.e., Stress group) or a neutral control task (i.e., Control group), such that there were equal numbers of males and females in each group. The experimenter probed all participants about their college major, if applicable, and dream career. Participants in the Stress group then underwent the TSST. They were told that after a 6‐min preparation period, they would deliver a 5‐min job interview style speech in front of a “speech expert” who would be evaluating their performance and recording with a video camera for review later. In reality, the speech expert was a female research assistant, and the video camera was nonfunctional. While each participant delivered the speech, the speech expert observed with a cold disposition and did not provide any nonverbal positive feedback (e.g., smiling, nodding, etc.). For participants who finished the speech early, the speech expert informed them of the time remaining and instructed them to continue until the time was up. Then, each participant was asked to do a 5‐min mental arithmetic task, which involved verbally counting backward by 13 from 2063. If the participant made an error, the speech expert instructed them to start over. The total time for the TSST was approximately 16 min. Participants in the Control group completed the Big Five Inventory personality questionnaire (John & Naumann, 2010) for the same amount of time. If participants finished early, they were asked to sit quietly until the experimenter returned with further instruction. Completion of this questionnaire served as a neutral control task because it was meant to evoke self‐reflection, like the TSST, without the added stressful social evaluation component.

2.3.2. Stress measures

To assess perceived stress levels in response to the stress induction, participants provided ratings on a 7‐point Likert scale once at the midpoint of the experimental task, about 57 min after stress onset (see timeline in Figure 1). To assess the physiological stress response, blood pressure and salivary cortisol were measured intermittently throughout the course of the experiment. Blood pressure was assessed using a blood pressure monitor attached to the upper right arm. Measurements were taken at three timepoints: baseline (BP1; 15 min before stress onset), after the stress manipulation (BP2; 16 min after stress onset), and at the end of the experiment (BP3; 106 min after stress onset).

Salivary cortisol samples were taken using SalivaBio Oral Swab collection devices (Salimetrics, Carlsbad, CA, United States). Measurements were taken at five timepoints: baseline (T0; 15 min before stress onset), after the stress manipulation (T1; 16 min after stress onset), during the experimental task (T2; 57 min after stress onset), after the experimental task (T3; 73 min after stress onset), and at the end of the experiment (T4; 103 min after stress onset). Participants were asked to refrain from eating or drinking (besides water) at least 1 h before the study to prevent sample contamination. During collection, participants were instructed to hold the collection apparatus under their tongue for 2 min to saturate it as much as possible. The saturated swab was then deposited into a swab storage tube. Samples were stored in a freezer at −20°C for up to 6 months before being shipped on dry ice for analysis. Samples were assayed at the Salimetrics' SalivaLab (Carlsbad, CA, United States) using the Salimetrics Cortisol Assay Kit (Cat. No. 1‐3002), without modifications to the manufacturers' protocol. The intra‐assay variability was 4.6% and the inter‐assay variability was 6.0%.

2.4. Response inhibition task

After participants completed the stress manipulation, they completed a brief 10‐trial practice round of the Go/NoGo task that they would later complete in the MRI scanner. In this task, which was adapted from Ceceli et al. (2020) and designed and presented using E‐Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA), participants were instructed to respond with a button press on Go trials or withhold a response on NoGo trials. The response cues were pictures of traffic lights, and the color‐action contingencies were presented as both congruent (i.e., NoGo Red, Go Green) and incongruent (i.e., Go Red, NoGo Green) with real‐world associations across four 120‐trial runs. Within each of the runs, there were 100 Go trials and 20 NoGo trials (5:1 Go/NoGo ratio), to increase task difficulty by biasing participants toward responding with a button‐press. On each trial, stimuli appeared on the screen for 500 ms, followed by a fixation cross. The inter‐trial intervals ranged in duration from 1000 to 4000 ms (see Figure 2). Although feedback on accuracy for each trial was provided during the practice round to aid comprehension, participants did not receive feedback during the main task. Pilot data (N = 100) revealed no significant between‐group differences with respect to presentation order, so congruent and incongruent runs were not counterbalanced. All participants completed the congruent runs first, followed by the incongruent runs. This presentation order was chosen to reduce the possibility of practice effects. More specifically, if participants were first presented with the presumably more difficult color‐action contingencies of the incongruent runs, they may have found the reversal of these contingencies in the subsequent congruent runs to be easier, and thus show enhanced task performance. To circumvent this issue, participants instead completed the congruent runs first to get acclimated to the task, then completed the incongruent runs, in which the reversal of contingencies would presumably be seen as more challenging.

FIGURE 2.

Go/NoGo task schematic. ITI = inter‐trial interval.

2.5. Self‐report measures

Participants also completed self‐report questionnaires about their cognitions during the task. Specifically, the Intrinsic Motivation Inventory (IMI) (Ryan & Deci, 2000) was used to assess their subjective task experience across multiple dimensions: interest/enjoyment (α = .895), perceived competence (α = .747), effort/importance (α = .82), and pressure/tension (α = .78). Scores were derived for each of these subscales, and higher scores indicate that a particular dimension had a greater impact on the participant's motivation levels during the task. Additionally, the Cognitive Interference Questionnaire (Sarason et al., 1986) was used to assess the degree to which participants got distracted during the task (α = .755). Higher scores on this scale indicate more task distraction. Participants were also probed about the degree to which they experienced mind‐wandering and their level of engagement during the task on a 7‐point Likert scale.

2.6. Functional magnetic resonance imaging data acquisition and preprocessing

The data acquisition and preprocessing approach, and thus the details to follow, were adapted from previous studies with a similar analytic approach (DiMenichi & Tricomi, 2017; Ito et al., 2020). Data were collected at the Rutgers University Brain Imaging Center (RUBIC), using a 3 T Siemens Trio MRI scanner. Whole‐brain multiband echo‐planar imaging (EPI) acquisitions were collected with a 32‐channel head coil with the following parameters: TR = 785 ms, TE = 37.5 ms, flip angle = 55 degrees, bandwidth 1924 Hz/px, in‐plane FOV read = 224 mm, 64 slices, 2.2 mm isotropic voxels, and a multiband acceleration factor of 8. Slices were obtained in an interleaved sequence in the anterior‐to‐posterior direction. In order to maximize our ability to capture the effects of stress on task performance, four task scans were collected, each 8 min and 12 s (615 TRs) long. A resting‐state scan was then collected over a duration of 15 min and 10 s (1147 TRs). Whole‐brain high‐resolution T1 structural images were acquired over a duration of 6 min and 38 s with the following parameters: FOV = 256 mm², TR = 2400 ms, TE = 2.31 ms, 8 mm isotropic voxels. Whole‐brain high‐resolution T2‐weighted structural images were acquired over a duration of 5 min and 34 s with the following parameters: FOV = 256 mm², TR = 3200 ms, TE = 566 ms, 8 mm isotropic voxels.

Task and resting‐state functional magnetic resonance imaging (fMRI) data were preprocessed using the Human Connectome Project (HCP) minimal preprocessing pipeline version 3.5.0 (Glasser et al., 2013). This pipeline includes anatomical reconstruction and segmentation, as well as functional EPI distortion correction (using acquired spin echo fieldmaps), motion realignment, volume‐to‐surface mapping, spatial normalization to a standard MNI surface template and multimodal cross‐subject surface alignment (Glasser et al., 2016). Minimal spatial smoothing (2 mm FWHM) was conducted using an HCP algorithm (Glasser et al., 2013) that constrained surface smoothing within the gray matter surface and volume smoothing within FreeSurfer atlas‐defined parcels (Fischl et al., 2002). Additional preprocessing was conducted on the output CIFTI 64k grayordinate standard space for the BOLD time series, which was first parcellated using a surface‐based atlas (i.e., one time series for each of the 360 cortical parcels) (Glasser et al., 2016). The Cole‐Anticevic Brain‐wide Network Partition (CAB‐NP) (Ji et al., 2019) network partition scheme was then used to parcellate subcortex into 358 regions, and to derive functional network affiliations for all cortical/subcortical parcels. The CAB‐NP is specifically designed to provide a data‐driven and independent basis for linking established subcortical regions to large‐scale functional networks (Ji et al., 2019). Nuisance regression was then performed on the demeaned and detrended BOLD regional time series, in line with best‐practice recommendations (Ciric et al., 2017). The nuisance regression model included the 6 standard motion parameters, their derivatives and quadratic expansions (24 motion regressors total). Physiological regressors were specified per the aCompCor procedure (Behzadi et al., 2007), with the first 5 principal components estimated within white matter and ventricular masks (5 white matter + 5 ventricle regressors, with their derivatives and quadratics, totaling 40 regressors). Altogether, this nuisance regression model had 64 nuisance parameters.

2.7. Statistical analysis

2.7.1. Behavioral analyses

For each stimulus type (i.e., Go Green, NoGo Red, NoGo Green, and Go Red), average accuracy was calculated by dividing the number of trials on which the participant responded correctly (i.e., responding with a button press on Go trial and withholding a response on NoGo trials) by the total number trials (200 total Go trials and 40 total NoGo trials). To test whether stress exposure affected task performance, we ran repeated‐measures ANOVAs on task accuracy for both Go trials and NoGo trials, each with group (stress or control) and sex (male or female) as between‐subjects factors and run (congruent or incongruent) as a within‐subjects factor.

d′ scores were also calculated for type of run (congruent and incongruent). This measure takes into account “hits” (correct Go trials), “misses” (errors of omission on Go trials), “false alarms” (FA; commission errors on No‐Go trials), and “correct rejections” (correct No‐Go trials). To estimate d′, the difference between the z transforms of the hit rate (H) and the FA rate was calculated, using the formula d′ = z(H)‐z(FA) (Macmillan & Creelman, 2004). The miss (M) rate was derived from the hit rate (M = 1‐H), and the FA rate was derived from the correct rejection (CR) rate (FA = 1‐CR). A higher d′ score indicates better task performance. Group differences in task performance, as measured by the d′ score, were assessed using the same repeated‐measures ANOVA described above.

To assess reversal‐related impairment associated with the task, a hierarchical regression analysis was run for both Go and NoGo trials, using a difference measure (i.e., incongruent–congruent accuracy) as the outcome variable. Model 1 included demographic covariates, age and sex. An additional covariate, neurologic or psychiatric disorder diagnostic history (dummy coded such that 1 indicated a diagnosis), was added into Model 2. Model 3 included the above variables and the main predictor of interest, group (i.e., Stress or Control). Each of these models met assumptions for homoscedasticity and normality. Tests for collinearity yielded negligible variance inflation factor (VIF) values (VIF <1.21 for all variables), indicating that the assumption for linearity was also met.

To examine how stress measures related to task performance and self‐report measures, bivariate correlations were computed for each group. To test whether these associations differed between groups, correlation coefficients were transformed using Fisher's Z transformation, then compared to test whether they differed significantly (Diedenhofen & Musch, 2015).

2.7.2. Psychophysiological analyses

To meet the requirement of homoscedasticity needed to conduct most statistical tests, the skewness of distribution of cortisol measures was determined for each of the five collection time points. Each timepoint had a negative skew, averaging at about 2.17. To normalize these data, the power transformation x′ = (x ^0.26–1/0.26) was performed, as this per recommendations for effective transformations for cortisol time courses (Miller & Plessow, 2013). After transforming the data, the average skew across timepoints was 0.80. Then, an area under the curve with respect to increase (AUC_i) analysis was run using the trapezoidal method (Pruessner et al., 2003), with T0 as the baseline cortisol level and T1–T4 as points in the analysis. To test whether the TSST effectively elicited a physiological stress response (i.e., an increase in cortisol levels), an independent samples t test was run on AUC_I values for stress versus control participants. To compare time‐dependent changes in cortisol levels between groups, a repeated‐measures ANOVA was run with group and sex as between‐subjects factors and timepoint as a within‐subjects factor.

2.7.3. fMRI analyses

As mentioned above, analyses of the fMRI data were focused on regions‐of‐interest identified by the 718‐parcel CAB‐NP network partition scheme (Ji et al., 2019), which assigns functional networks to the multimodal Glasser atlas (Glasser et al., 2016). The network labels facilitated functional inferences about significant regions (Figure 3). We took a whole‐brain approach because it is more comprehensive than an a priori network‐based approach, which confines the analyses to just a subset of ROIs. Standard univariate task general linear model (GLM) analyses were run to estimate task‐evoked activations from each of the color‐action contingencies. These GLMs were fit for each subject separately using the fully concatenated task dataset, which was concatenated across four runs. There were four regressors for trials on which participants responded correctly, one for each trial type (Go Green, NoGo Red, NoGo Green, Go Red), and one regressor of no interest for trials on which they responded incorrectly regardless of trial type (Incorrect). Task activation analyses were run to identify brain regions that elicited task‐related activity. Beta amplitude coefficients for each participant were derived for each parcel from the univariate GLM described above for two main contrasts: correct NoGo Green > baseline and correct NoGo > correct Go. The former contrast is commonly implemented in similar event‐related studies, as it most precisely identifies all regions that are associated with correctly withholding a response, and thus important for response inhibition (Simmonds et al., 2008). This latter contrast is likely to show the strongest control‐related differences in activation. It would be more difficult to isolate an incongruent > congruent effect (i.e., NoGo Green > NoGo Red) in the context of our task, since the imbalance between Go trials (83.3%) and NoGo trials (16.7%) was included intentionally to increase the oddball effect of the NoGo trials (which presumably require the most cognitive control). Nonetheless, we also ran the incongruent > congruent contrast, as well as a congruent > incongruent contrast. The beta coefficients from these contrasts were compared against zero across participants with a one sample t test. Multiple comparison correction was applied via the false discovery rate (FDR) method (Benjamini & Hochberg, 1995), indicated below as p _corr values. Group‐level task activation analyses for the sample contrasts were run in the same manner described above, separated out by Stress and Control groups. To assess between‐group differences, contrast values were compared between groups with an independent samples t test.

FIGURE 3.

The Cole‐Anticevic Brain‐wide Network Parcellation (CAB‐NP), which the network affiliations in this study are based on. This parcellation partitions brain regions into 718 (a) cortical and (b) subcortical parcels. Colors represent network affiliation (see right for labels). For more information, see Ji et al. (2019).

Given prior reports highlighting lower sensitivity in detecting task‐evoked activation effects for univariate GLM compared to multivariate pattern analysis (MVPA) approaches (Jimura & Poldrack, 2012; Mill et al., 2021; Spronk et al., 2021), we also implemented the latter. Specifically, we trained multivariate machine learning classifiers to discriminate between Stress and Control groups, on the basis of within‐region voxel activations (features) estimated for the NoGo Green > baseline contrast (the condition likely eliciting the highest cognitive control demands). We used an established minimum‐distance classification approach (Haxby et al., 2001; Mill et al., 2020, 2021; Mur et al., 2009; Spronk et al., 2021) with k‐fold cross‐validation, which outputs classification decisions for held‐out test subjects in each group based on the similarity (Pearson correlation) of those test subjects' regional activation patterns with the Stress and Control group averages (computed in the training set). Hence, for an exemplary test subject in the Stress group, the classifier would output a correct decision if the similarity of that subject's regional activation pattern with the Stress group average (computed excluding that test subject to prevent circularity) was greater than its similarity with the Control group average. After iterating over all cross‐validation loops, the accuracy of each regional classifier in distinguishing between the two groups was averaged and contrasted against chance 50% classification accuracy via binomial test to assess statistical significance. This process was repeated for all cortical and subcortical regions (718 total), with accuracy p values corrected for multiple regional comparisons via FDR (p < .05).

3. RESULTS

3.1. Stress measures reveal success of stress induction

Table 1 summarizes the behavioral descriptive statistics. Independent samples t tests revealed significant between‐group differences (i.e., Stress > Control) for both subjective stress ratings, t(75) = 9.60, p < .001, d = 2.18 (see Figure 4a and Figure S1a) and change in systolic blood pressure, t(75) = 4.85, p < .001, d = 1.11 (see Figure 4b and Figure S1b). Following the stress manipulation, participants in the stress group rated feeling more stressed and experienced a significant increase in systolic blood pressure from baseline. Bivariate correlations revealed that these two measures were moderately correlated, r(75) = .342, p = .002. Overall, these results suggest that the TSST successfully elicited a stress response.

TABLE 1.

Summary descriptive by stress induction group.

Variable	Stress (n = 39)		Control (n = 38)
	M	SD	M	SD
Stress rating	4.80	1.66	1.76	1.05
ΔSBP (mmHg)	3.49	8.65	−6.89	10.08
Cortisol level (nmol/L)
T0	2.24	0.85	2.22	1.02
T1	2.23	0.92	2.11	0.80
T2	2.37	1.08	1.95	2.25
T3	1.96	0.82	1.76	2.23
T4	1.70	0.72	1.54	2.37
Go/NoGo task accuracy (%)
Go Green	88.30	0.11	84.44	0.14
NoGo Red	83.89	0.13	81.58	0.15
NoGo Green	77.56	0.15	67.50	0.18
Go Red	86.76	0.13	82.61	0.16
Congruent d′ score	2.14	0.73	2.15	0.72
Incongruent d′ score	2.04	0.74	1.62	0.72
Go Green RT (ms)	322.61	40.83	306.89	40.93
Go Red RT (ms)	293.78	48.61	309.27	43.84
Mind wandering	4.90	1.37	4.30	1.49
Task engagement	4.96	1.18	4.88	1.48
CIQ	2.72	0.56	2.90	0.68
IMI
Interest/enjoyment	4.60	1.14	4.59	1.28
Perceived competence	4.88	0.74	4.79	0.97
Effort/importance	4.90	1.15	5.19	1.21
Pressure/tension	3.81	1.32	3.65	1.33

Abbreviations: ΔSBP, change in systolic blood pressure; CIQ, Cognitive Interference Questionnaire; IMI, Intrinsic Motivation Inventory.

FIGURE 4.

(a) Subjective stress ratings—Bars depict average subjective stress ratings for Control (light purple) and Stress (dark purple) groups. (b) Change in systolic blood pressure (SBP)—Bars depict average change in SBP from baseline following the stress manipulation for Control and Stress groups. (c) Cortisol levels—Lines represent cortisol levels at each timepoint (T0 through T4) for Control and Stress groups. Error bars on all plots represent the 95% confidence interval. T0 = 15 min before stress onset, T1 = 16 min after stress onset, T2 = 57 min after stress onset, T3 = 73 min after stress onset and T4 = 103 min after stress onset. ***p < .001; **p < .01; *p < .05.

The groups did not differ significantly in their cortisol levels, however. An independent samples t test revealed no significant between‐group differences (i.e., Stress > Control) in AUC_I values, t(75) = 1.14, p = .26, d = .26. In addition, there were no significant sex differences in AUC_I values, t(75) = −1.44, p = .07, d = .32. Repeated‐measures ANOVA results revealed a significant main effect of timepoint (F(4, 292) = 19.05, p < .001, η _p ² = .207), no significant main effect of group (F(1, 73) = .1.86, p = .177, η _p ² = .03), a significant main effect of sex (F(1, 73) = 18.36, p < .001, η _p ² = .20), and no significant interaction effects (timepoint × group: F(4, 292) = 1.35, p = .25, η _p ² = .02; timepoint × sex: F(4, 292) = 1.33, p = .26, η _p ² = .02; timepoint × group × sex: (F(4, 292) = 1.92, p = .15, η _p ² = .03)). Post hoc pairwise comparisons via the Tukey method revealed that cortisol levels differed significantly by timepoint: T0 was significantly higher than both T3 (t(73) = 3.06, p = .03) and T4 (t(73) = 4.67, p < .001); T1 was significantly higher than both T3 (t(73) = 3.27, p = .01) and T4 (t(73) = 5.03, p < .001); and T2 was significantly higher than both T3 (t(73) = 4.12, p < .001) and T4 (t(73) = 5.13, p < .001) (see Figure 4c). Further, males had higher cortisol levels at each timepoint than females, t(75) = −4.18, p < .001.

The above analyses were rerun after the sample was separated into cortisol responders, defined as those who had at least a 15.47% (or >1.5 nmol/L) increase in cortisol levels from baseline to peak, and non‐responders (Kirschbaum et al., 1993; Miller & Plessow, 2013). The independent samples t test and ANOVA run with the subset of 66 responders (33 stress, 33 control) yielded the same pattern of results. It is of note that 6 of the 11 non‐responders were in the stress group. The lack of group differences, particularly for the expected peak at T2, may be due to this sample being collected outside of the 10–30 min interval following stress exposure during which cortisol levels are expected to peak (Giles et al., 2014), as well as due to elevated baseline cortisol levels (e.g., due to stress of participating in an fMRI study), which resolved over the course of the experiment.

3.2. Stress enhanced behavioral task performance

The repeated‐measures ANOVA on task accuracy for NoGo trials revealed a significant main effect of run (F(1, 73) = 51.83, p < .001, η _p ² = .42), no significant main effect of group (F(1, 73) = 2.98, p = .09, η _p ² = .04), and a significant main effect of sex (F(1, 73) = 4.47, p = .04, η _p ² = .06). There was also a significant run × group interaction effect (F(1, 73) = 11.16, p = .001, η _p ² = .13), but no other significant interaction effects (run × sex: F(1, 73) = .26, p = .61, η _p ² = .004; run × group × sex: F(1, 73) = .12, p = .73, η _p ² = .002). Post hoc pairwise comparisons via the Tukey method revealed that participants in both groups performed better on NoGo Red trials than on NoGo Green trials (stress: t(73) = 2.75, p = .04, control: t(73) = 7.40, p < .001). However, participants in the stress group committed significantly fewer commission errors on NoGo Green trials compared to controls, t(73) = −2.70, p = .04 (see Figure 5 and Figure S2). Further, females performed better on NoGo trials than males, t(73) = 2.11, p = .04. For Go trials, there were no significant main or interaction effects. For d prime scores, there was a significant main effect of run (F(1, 73) = 11.51, p = .001, η _p ² = .14), no significant main effect of group (F(1, 73) = 2.16, p = .15, η _p ² = .03), and no significant main effect of sex (F(1, 73) = .002, p = .97, η _p ² = .0001). There was also a significant run × group interaction effect (F(1, 73) = 5.07, p = .03, η _p ² = .07), but no other significant interaction effects (run × sex: F(1, 73) = .62, p = .43, η _p ² = .008; run × group × sex: F(1, 73) = .54, p = .46, η _p ² = .007). Post hoc pairwise comparisons via the Tukey method revealed that participants in the stress group performed better on the incongruent trials (i.e., more correct rejections, less FAs) than those in the control group, t(73) = −2.09, p < .04. Further, among stress participants, performance on Go Red trials was negatively associated with subjective stress ratings, r(37) = .47, p = .02. Taken together, these results suggest that stress may have facilitated better performance on the task, especially on those trials requiring the most cognitive control.

FIGURE 5.

Go/NoGo task accuracy. Bars represent average performance accuracy on Go Green (dark green solid), Go Red (pink vertical crosshatch), NoGo Green (light green dots), and NoGo Red (red diagonal striped) trials for Control and Stress groups. Error bars depict the 95% confidence interval. ***p < .001; **p < .01; *p < .05. Dashed horizontal line reflects chance performance (50%).

3.3. Associations between Go performance and reaction time

Results from an independent samples t test revealed no significant between‐group differences for reaction time on neither Go Green (t(75) = 1.69, p = .10) nor Go Red trials (t(75) = 1.47, p = .15). However, for both congruent and incongruent Go trials, task accuracy was positively correlated with response time for both Go Green (r(75) = .88, p < .001) and Go Red (r(75) = .91, p < .001) trials. This pattern of results was also observed when the data were separated by group (r > .80 and p < .01 for both stress and control groups). These results suggest that there was a speed‐accuracy tradeoff, such that when participants responded faster (reflected by a smaller RT) on Go Trials, they were less accurate, regardless of stress exposure.

3.4. Reversal‐related impairment

The hierarchical regression model testing reversal‐related impairment (assessed using a difference score, i.e., incongruent–congruent accuracy) on Go trials revealed no significant reversal‐related impairment (Model 1: F(2, 74) = .21, p = .81, R ² = .01, R ² _adj = −.02; Model 2: F(3, 73) = .44, p = .72, R ² = .02, R ² _adj = −.02; Model 3: F(4, 72) = .40, p = 0.81, R ² = .022, R ² _adj = −.03) (see Table 2).

TABLE 2.

Summary of hierarchical regression model results predicting go reversal‐related impairment.

Model	B	SE		β	t
Model 1
Age	.002	.12		.04	.37
Sex	.01	.03		.05	.47
Model 2
Age	.003	.01		.06	.61
Sex	.02	.03		.09	.48
Diagnosis	−.05	.05		−.12	.34
Model 3
Age	.003	.01		.07	.56
Sex	.02	.03		.09	.74
Diagnosis	−.05	.05		−.13	−1.05
Group	−.01	.03		−.06	−.53

Model summary statistics
Model	R 2	R 2 adj	ΔR 2	ΔR 2 adj	F	ΔF
Model 1	.01	−.02			F(2, 74) = .21
Model 2	.02	−.02	.01	−.001	F(3, 73) = .44	.90
Model 3	.022	−.03	.002	−.01	F(4, 72) = .40	.28

The model for NoGo trials revealed that the covariates alone were not significant predictors (Model 1: F(2, 74) = .51, p = .60, R ² = .01, R ² _adj = −.01; Model 2: F(3, 73) = .64, p = .59, R ² = .03, R ² _adj = −.01). However, when the group regressor was added to Model 3, it explained an additional 15% of the variance in the reversal‐related impairment outcome: 𝛽 = .41, ΔR ² = .16, ΔR ² _adj = .15, p < .001. Additionally, the model as a whole significantly predicted reversal‐related impairment, F(4, 72) = 4.09, p < .01 (see Table 3). These results suggest that exposure to stress is associated with less reversal‐related impairment on the Go/NoGo task.

TABLE 3.

Summary of hierarchical regression model results predicting NoGo reversal‐related impairment.

Model	B	SE		β	t
Model 1
Age	.01	.01		.11	.90
Sex	.01	.03		.03	.29
Model 2
Age	.005	.01		.09	.74
Sex	.0002	.03		.001	.01
Diagnosis	.05	.05		.12	.96
Model 3
Age	.002	.01		.04	.38
Sex	−.004	.03		−.02	−.17
Diagnosis	.10	.05		.23*	1.95*
Group	.10	.03		.41***	3.75***

Model summary statistics
Model	R 2	R 2 adj	ΔR 2	ΔR 2 adj	F	ΔF
Model 1	.01	−.01			F(2, 74) = .51
Model 2	.03	−.01	.01	−.001	F(3, 73) = .64	.1.08
Model 3	.19	.14	.16	.15	F(4, 72) = 4.09***	14.09***

Note: Bold face indicates significance. *p < .05; **p < .01; ***p < .001.

3.5. Stress is associated with distraction, engagement, and motivation

Among participants in both groups, subjective stress rating was significantly associated with cognitive interference, stress: r(37) = .47, p = .001; control: r(36) = .32, p = .02. Among participants in the stress group, ratings were significantly associated with psychometrically assessed mind‐wandering (r(37) = .29, p = .04), engagement (r(37) = .43, p = .003), and intrinsic motivation, as measured by the enjoyment dimension of the IMI (r(37) = .29, p = .04). Additionally, in the stress group, stress‐related change in systolic blood pressure was significantly associated with mind‐wandering (r(37) = −.370, p = .010) and feelings of pressure or tension (r(37) = −.28, p = .04). Among participants in the control group, stress‐related change in systolic blood pressure was significantly associated with cognitive interference (r(36) = −.33, p = .02), intrinsic motivation (r(36) = −.29, p = .04), and perceived competence (r(36) = −.28, p = .04), but not mind‐wandering (r(36) = .12, p = .24). Of these associations, only the correlation coefficient for the association between stress‐related change in blood pressure and mind‐wandering was significantly different between the two groups, z(75) = 2.14, p = .03.

3.6. fMRI univariate GLM reveals cognitive control‐related activations

All activation patterns reported are significant at a level of p _corr < .05. For the correct NoGo Green > baseline contrast, task performance was associated with deactivation in regions affiliated with the DMN, including the posterior cingulate cortex (including the posterior cingulate gyrus), ventromedial prefrontal cortex, inferior parietal cortex, middle frontal gyrus, hippocampus, and right cerebellum (see Figure 6 and Table S1). NoGo Green task performance was also associated with increased activation in regions affiliated with the CCNs, including premotor cortex, frontal operculum, the anterior ventral insular area, the dorsolateral intraparietal area, the ventral visual complex, caudate nucleus, and left cerebellum.

FIGURE 6.

False discovery rate (FDR)‐corrected t statistic maps for correct NoGoGreen > baseline contrast for (a) cortical and (b) subcortical regions. Maps show increased activity during performance of NoGo Green trials. Patterns of deactivation were observed in regions affiliated with the default mode network, including medial prefrontal and rostral cingulate areas. Patterns of activation were observed in regions affiliated with cognitive control networks, including lateral prefrontal, anterior insula, and caudate areas. p _corr < .05.

For the correct NoGo > correct Go contrast, task performance was associated with deactivation in regions affiliated with the DMN, including the posterior cingulate cortex, left ventrolateral prefrontal cortex, left medial prefrontal cortex, posterior parietal cortex, medial temporal cortex, middle temporal gyrus, and caudate, as well as brainstem and cerebellar regions affiliated with the CCNs. NoGo task performance was also associated with increased activation in regions affiliated with the CCNs, including the right dorsolateral prefrontal cortex, inferior parietal cortex, intraparietal sulcus, dorsolateral intraparietal area, premotor areas, middle insula, and accumbens (see Figure 7 and Table S2).

FIGURE 7.

False discovery rate (FDR)‐corrected t statistic maps for correct NoGo > correct Go contrast for (a) cortical and (b) subcortical regions. Maps show increased activity during performance of NoGo trials relative to Go trials. Patterns of deactivation were observed in regions affiliated with the default mode network, including medial prefrontal, medial temporal, posterior cingulate areas, as well as the caudate. Patterns of activation were observed in regions affiliated with cognitive control networks, including lateral prefrontal areas and the accumbens. p _corr < .05.

However, there were no significant FDR‐corrected between‐group differences for either of these contrasts, p _corr > .05. Additionally, neither the NoGo Green > NoGo Red contrast nor the NoGo Red > NoGo Green contrast yielded significant results that survived FDR correction, p _corr > .05. This lack of between‐group differences are likely due to low sensitivity of univariate GLM analyses.

3.7. Multivariate machine learning classifiers reliably distinguish between Stress and Control groups based on regional activation patterns

MVPA successfully differentiated between Stress and Control groups on the basis of NoGo Green > baseline trial‐related activation (see Section 2.7.3 for details on this contrast). Regions include the left medial intraparietal area (mIPA; MNI coordinates: X = −23.59, Y = −64.90, Z = 50.44), which is affiliated with the dorsal attention network (classification accuracy, CA = 69.4%, p _corr = .045); the right brainstem (X = 4.60, Y = −46.90, Z = −53.10; CA = 72.2%, p _corr = .025); and the left cerebellum (X = −3.10, Y = −54.60, Z = −44), affiliated with the DMN (CA = 70.9%, p _corr = .019) (see Figure 8 and Table S3).

FIGURE 8.

False discovery rate (FDR)‐corrected classification accuracy (CA, in %) for NoGo Green > baseline contrast for (a) cortical and (b) subcortical regions. Multivariate pattern analysis (MVPA) successfully differentiated between stress and control groups for NoGo Green task activation in the medial intraparietal area (mIPA), the brainstem, and the cerebellum. p _corr < .05.

4. DISCUSSION

4.1. Effects of stress on performance

In the present study, we combined neuroimaging, behavioral, and psychophysiological methods to investigate the effects of acute psychosocial stress on the neural processes supporting performance on a Go/NoGo task. The stress induction successfully increased blood pressure and subjective stress ratings, and interestingly, acute stress improved accuracy on task trials requiring the most cognitive control. This finding is in contrast to previous work indicating that stress can negatively affect performance (e.g., DiMenichi & Tricomi, 2017; Qian et al., 2015). However, the finding is in line with other work that has demonstrated that stress can improve performance on tasks requiring cognitive control, especially response inhibition tasks, such as the Stroop (Chajut & Algom, 2003; Kofman et al., 2006) (Chajut & Algom, 2003; Kofman et al., 2006) and stop‐signal tasks (Schwabe et al., 2013). These tasks may have subtle differences in the processes they capture (i.e., stop‐signal tasks appear to assess automatic inhibition, while Go/NoGo assesses controlled inhibition (Raud et al., 2020)). Nevertheless, the fact that they appear to demonstrate the same effect shows that these tasks may involve some common underlying mechanism whereby stress facilitates cognitive control.

One commonality across experiments in which stress impairs or improves accuracy is that a speed/accuracy tradeoff is observed (DiMenichi et al., 2018; Scholz et al., 2009). This occurred in our experiment as well. Thus, when stress acts to increase vigilance to avoiding errors, it does so by causing slower, more careful responses. Commission errors on green/No‐Go trials in this task can be thought of as “slips of habit,” as they occur when previously acquired, well‐learned associations govern behavior, rather than newly learned associations (Ceceli et al., 2020). Stress has been shown to accentuate habitual behavior (Schwabe & Wolf, 2009), but it may be that when there is little time pressure and an emphasis on accurate performance, psychosocial stress may motivate more careful, goal‐directed behavior, at the expense of speed. If a more speeded version of this task had been used, as in previous work (Ceceli et al., 2020), it is possible that stress would have caused more habitual, rather than goal‐directed, responding.

Stress has been found to increase selectivity to task‐relevant attributes rather than task‐irrelevant attributes, at the expense of cognitive flexibility (Chajut & Algom, 2003; Plessow et al., 2011). In our study, self‐reported stress was associated with more task engagement among those in the stress group, which fits with the idea that stressed participants were more focused on the task. Yet, paradoxically, self‐reported stress was also associated with more cognitive interference. One possible explanation for this dichotomy is the suggestion that stress enhances response inhibition while impairing cognitive inhibition (Shields et al., 2016). Thus, stress may have aided engagement and performance in this task because it required response inhibition, while still causing cognitive intrusions in the form of mind‐wandering. This may also explain why stress is particularly likely to impair performance on tasks requiring sustained attention over a long period of time (DiMenichi et al., 2018; Plessow et al., 2011), as the effects of cognitive intrusions may eventually cause more errors in performance.

Another suggestion is that the effects of stress on performance follow an inverted U‐shaped curve, with lower performance for low or high levels of acute stress, and optimal performance for intermediate levels of stress (Salehi et al., 2010). Athletes, for example, regularly perform under stress, and experience an increase in cortisol during competition, especially during successful performance (Chennaoui et al., 2016; O'Donnell et al., 2018). Too much stress, however, can lead to decrements in performance, or “choking under pressure” (Beilock & Carr, 2000). It may be that the level of stress experienced by the stress group helped optimize performance on our task, as it may have increased task engagement and vigilance, whereas greater levels of stress may have had a more deleterious effect on performance.

4.2. Physiological effects of stress

Systolic blood pressure was significantly elevated for the stress group relative to the control group, providing physiological evidence to corroborate the increased subjective ratings of stress for the stress group. However, although participants rested for 15 min after arriving at the imaging center, cortisol levels did not increase from the baseline level sampled at T0, and the final two cortisol measurements were significantly lower than the baseline level. This suggests that anticipation of the fMRI experiment may itself have been a stressor, which may have increased cortisol levels prior to the experiment. These levels then resolved to what may be considered the true baseline (T3/T4) after the experimental task ended. Although blood pressure and subjective ratings of stress increased for the group who performed the TSST relative to the control group, the stress group did not display significantly greater cortisol levels. While it is possible that the stress of the TSST was not enough to significantly increase the participants' already elevated cortisol levels, it may be that we missed the peak of the cortisol increase, which usually occurs about 20 min after a stressor (Kirschbaum et al., 1993). It took longer for participants to enter the scanner than anticipated, so that our T2 sample was not acquired until approximately 40 min after the stressor. These results highlight some of the limitations of using cortisol measurements in an fMRI experiment and suggest that blood pressure and subjective ratings can provide additional useful measures of stress.

4.3. Neural effects of stress

Originally, we hypothesized that the effects of stress on cognitive performance might be reflected in the dynamics of the DMN and CCNs. As expected, across participants, task performance was associated with increases in activation in the CCNs and decreases in the DMN. The pattern of activity that differentiated between the stress and control groups encompassed one of the CCNs (the dorsal attention network), as well as other brain regions involved in visuomotor coordination that may also support successful task performance. Specifically, acute stress may have increased cognitive performance by increasing vigilance to the inhibition of a default response. The reversal‐related impairment when changing from the congruent Go Green to the incongruent NoGo Green task requirements was smaller for the stress group than the control group. In line with this interpretation, our MVPA results for the NoGo Green condition suggest differential activation patterns between groups in the mIPA and the cerebellum, brain regions involved in visuomotor coordination (Brown et al., 1993; Grefkes et al., 2004; Swaminathan et al., 2013; Voogd et al., 2012). The mIPA is affiliated with the dorsal attention network and is involved in goal‐directed selection of motor responses (Corbetta et al., 2008; Corbetta & Shulman, 2002; Grefkes et al., 2004; Swaminathan et al., 2013). The cerebellum contributes to executive control of voluntary actions, including response inhibition and performance monitoring (Brunamonti et al., 2014; Peterburs & Desmond, 2016). These functions of the mIPA and cerebellum are of central importance for accurate performance on our particular task. Therefore, activity in these regions in the stress group may have supported the enhanced performance observed in this group.

However, since we did not observe group differences in our univariate analyses, we cannot make strong claims about the directionality of activation differences between groups; thus, determining whether activation in the brain regions identified in our MVPA analysis was higher in the stress or the control group the directionality of is an important direction for future work. Another potential direction for future research would be to investigate whether individual differences in reversal‐related impairment could be predicted from the multivariate activation patterns of the regions identified in this work. This would lend further credence to the idea that these activation patterns support response inhibition and cognitive control.

4.4. Limitations

A major limitation of the current study is that certain logistical considerations affected our stress measures and thus may have limited our ability to capture between group differences and conclusions that can be drawn from the results. First, as mentioned above, the timing of the cortisol measurements was not well‐aligned with the expected biological fluctuations. Ideally, we could have allowed for a longer acclimation period for participants' baseline cortisol levels to stabilize. But due to unforeseen logistical factors (e.g., participant tardiness, available resources in the scanning center), this was not always a feasible option. We also could have obtained more cortisol samples at more frequent intervals to track stress‐related changes in cortisol levels over time more effectively. However, the logistics of collecting salivary cortisol samples without removing a participant from the MRI scanner are difficult to navigate and error‐prone, and there are currently no published standard best practices for studies of this nature. Thus, we decided to prioritize minimizing participant discomfort and motion to ensure the quality of the data obtained. Relatedly, although we restricted our sample to only females were not using hormonal contraceptives, we did not collect information about their menstrual cycle phase during the time of the experiment. This is an important factor to take into consideration in stress studies using cortisol measurement, given evidence suggesting that cortisol responsivity to acute stress varies based on menstrual cycle phase (Maki et al., 2015; Montero‐López et al., 2018). Finally, we only assessed subjective stress once after the stress induction, rather than collecting repeated assessments over the course of the experiment (including before stress exposure). Thus, it is unclear whether (and how) feelings of stress truly changed in response to the stressor, and how they changed with the physiological stress measures. An additional limitation is that the neural signal elicited from our Go/NoGo task might not entirely represented inhibitory control processes. According to one perspective, in fMRI studies using Go/NoGo tasks to assess response inhibition, the activity elicited by NoGo signals could mainly reflect attentional or working memory processes, rather than inhibition per se (Criaud & Boulinguez, 2013). While this is plausible, we were limited in that we didn't separate inhibitory control processes from working memory and attentional processes in our analyses. This is an avenue for future investigation.

5. CONCLUSION

Taken together, our results suggest that acute psychosocial stress may be capable of exerting a facilitative effect on cognitive control, and that these effects may be represented in the activation patterns of task‐associated brain regions. This study suggests that use of MVPA for between‐groups fMRI designs can provide additional insights beyond traditional GLM analyses into how the brain supports cognitive performance. While psychosocial stress produced readily apparent effects on subjective experience of stress, blood pressure, and task performance, the concomitant effects on brain function were more subtle and were elucidated with MVPA rather than simple univariate group differences in BOLD signal. Thus, while the neural processes supporting cognitive control may be largely similar when under stress and without stress, acute psychosocial stress may nevertheless create identifiable differences in neural patterns of activation that aid in task performance under duress.

CONFLICTS OF INTEREST

The authors have no conflicts of interest to report.

Supporting information

DATA S1: Supporting information.

Spencer, C. , Mill, R. D. , Bhanji, J. P. , Delgado, M. R. , Cole, M. W. , & Tricomi, E. (2024). Acute psychosocial stress modulates neural and behavioral substrates of cognitive control. Human Brain Mapping, 45(8), e26716. 10.1002/hbm.26716

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.