Isometric handgrip exercise speeds working memory responses in younger and older adults

Abstract

Physiological arousal affects attention and memory, sometimes enhancing and other times impairing what we attend to and remember. In the present study, we investigated how changes in physiological arousal - induced through short bursts of isometric handgrip exercise - affected subsequent working memory performance. A sample of 57 younger (ages 18–29) and 56 older (ages 65–85) participants performed blocks of isometric handgrip exercise in which they periodically squeezed a therapy ball, alternating with blocks of an auditory working memory task. We found that, compared with those in a control group, participants who performed isometric handgrip had faster reaction times on the working memory task. Handgrip-speeded responses were observed for both younger and older participants, across working memory loads. Analysis of multimodal physiological responses indicated that physiological arousal increased during handgrip. Our findings suggest that performing short bouts of isometric handgrip exercise can improve processing speed, and they offer testable possibilities for the mechanism underlying handgrip’s effects on performance. The potential for acute isometric exercise to temporarily improve processing speed may be of particular relevance for older adults who show declines in processing speed and working memory.

Introduction

Physiological arousal affects attention, perception, and memory (Berridge & Waterhouse, 2003; Sara, 2009). These effects are mediated by the brain’s locus coeruleus (LC), a small nucleus in the brainstem that serves as the brain’s primary source of noradrenaline (Schwarz & Luo, 2015). Many studies testing arousal’s effects on cognition have employed manipulations with emotion- and/or stress-inducing components, with fewer studies testing how isolated effects of physiological arousal and LC activity affect cognitive performance. An influential model of LC function posits an inverted U-shaped relationship between tonic LC activity and cognitive performance, with performance peaking at moderate levels of tonic LC activity and being worse at high and low levels of tonic LC activity (Aston-Jones & Cohen, 2005).

Isometric handgrip exercise is one manipulation that affects arousal and noradrenergic activity, with recent studies reporting that performing short bouts of handgrip temporarily increases salivary alpha amylase levels, heart rate, systolic blood pressure, and blood flow velocity (Nielsen & Mather, 2015; Washio et al., 2021). Furthermore, this manipulation does not elicit concomitant activation of the hypothalamic-pituitary-adrenal axis that could lead to stress effects (Finke et al., 2018). Handgrip also increases pupil diameter, a non-invasive proxy of LC activity (Joshi et al., 2016; Murphy et al., 2014; Nielsen & Mather, 2015). Handgrip furthermore offers the possibility of testing questions about the laterality of pupillary responses to LC stimulation. Stimulating the LC elicits lateralized pupillary responses, wherein dilation in the ipsilateral pupil relative to the LC is greater than that of the contralateral pupil, and this lateralization effect is thought to arise from the lateralized nature of LC-pupil pathways (Liu et al., 2017).

Critically, there are also effects on arousal and cognition immediately after handgrip. We previously reported that while isometric handgrip increased concurrent pupil diameter, in the period after handgrip, it reduced tonic pupil diameter, enhanced phasic pupillary responses, and decreased reaction times during an auditory oddball task (Mather et al., 2020). Post-handgrip benefits on cognition have been reported in other domains, with isometric handgrip exercise enhancing subsequent memory retention (Nielson et al., 1996) and speeding reaction times on a subsequent go/no-go task (Washio et al., 2021). These results suggest that a brief period of static exercise may temporarily reduce subsequent tonic levels of arousal and LC activity to a point that optimizes task performance (Aston-Jones & Cohen, 2005).

In addition to lower tonic pupil diameter and enhanced phasic pupillary responses after handgrip, we found that handgrip led to greater engagement of the brain’s right frontoparietal network during a subsequent oddball task (Mather et al., 2020). Frontoparietal brain regions are densely innervated by noradrenergic projections from the LC (Pickel et al., 1974; Schwarz & Luo, 2015), and during arousal, phasic signals from the LC promote neural gain within frontoparietal brain regions, facilitating selective attention and task performance (Arnsten et al., 1996; Aston-Jones & Cohen, 2005; Corbetta et al., 2008). Therefore, changes in arousal may be particularly relevant for tasks that engage frontoparietal brain regions, including working memory, the short-term storage and maintenance of information in memory (Baddeley, 2003). Previous studies have indicated that stress can either impair, enhance, or not affect working memory (Duncko et al., 2009; Human et al., 2018; Schoofs et al., 2008; Schoofs et al., 2013), but it is unclear how non-stress related increases in arousal, such as those induced through isometric handgrip, affect working memory.

How arousal affects working memory may also change with age. Noradrenergic innervation of the frontal cortex declines in aging (Ishida et al., 2001). In addition, the sensitivity of noradrenergic receptors in the prefrontal cortex changes with age, with spatial working memory performance impairments observed at lower doses of an alpha-2-adrenergic antagonist for older compared to younger rats (Caetano et al., 2013). These changes may affect the extent to which handgrip-induced noradrenergic activity can affect performance on working memory tasks in older adults. Alpha-2-adrenergic agonists have been demonstrated to rescue memory deficits in aged rats (Arnsten et al., 1988; Arnsten & Goldman-Rakic, 1985), but whether arousal changes, such as those induced through handgrip, affect cognitive performance comparably in humans remains unclear. During a fear conditioning task, arousal increased functional connectivity between the LC and frontoparietal network less in older compared to younger adults (Lee et al., 2018). In addition, arousal increased the selectivity of attention and memory in older adults less than it did in younger adults (Durbin et al., 2018; Gallant et al., 2022). One explanation for these findings is that older adults have higher baseline levels of LC firing rates compared with younger adults, yielding a reduced dynamic range within which arousal can affect cognitive performance in aging (Mather, 2020). On the other hand, we found that isometric handgrip benefited younger and older women similarly in terms of frontoparietal network activation, raising the possibility that handgrip-induced arousal has similar effects on performance for younger and older adults (Mather et al., 2020).

In this study, we tested the effects of isometric handgrip on subsequent working memory performance in younger and older adults. We randomly assigned participants to one of two groups: a handgrip or a control group. Participants completed short bouts of isometric handgrip exercise (handgrip group), or a control task of equal duration (control group), alternating with rounds of an auditory n-back task to probe working memory performance. Based on reports of beneficial effects of isometric handgrip exercise on subsequent cognitive task performance (Mather et al., 2020; Washio et al., 2021), we predicted that both younger and older participants in the handgrip group would have higher accuracy and faster reaction times on the working memory task relative to those in the control group. Furthermore, we expected that levels of tonic arousal would be reduced and that phasic pupillary responses would be enhanced after handgrip.

Methods

Transparency and Openness

De-identified data, necessary materials, and code to reproduce all analyses are available at the links provided in the Author Note. The study design, hypotheses, and analysis plan were not preregistered.

Participants

The sample size for this project was determined through a power analysis conducted with G*Power (Version 3.1.9; Faul et al., 2007). This indicated that a sample size of 52 participants would provide 80% power to detect a between-subjects effect of size f = 0.33 in a 2×3 mixed-design ANOVA, with α = 0.05. We selected a moderate-to-large size estimate for the effects of handgrip on n-back performance, tonic pupil diameter, and phasic pupillary responses. This estimate was based on a previous report of medium-to-large effects of handgrip on task performance and pupil dynamics (Mather et al., 2020), as well as reports of medium-to-large effects of both stress and threat of shock on working memory performance (Bolton & Robison, 2017; Human et al., 2018; Schoofs et al., 2013). To account for possible exclusions, we therefore aimed for a sample size of 56 younger and 56 older participants.

Younger participants were recruited from the University of Southern California Psychology Subject Pool, and older participants were recruited from the Los Angeles and USC communities. Eligible participants were fluent English speakers with corrected or normal-to-corrected vision who were not taking psychoactive or beta-blocker medications and who did not self-identify as having any chronic illness or cognitive impairment. Prior to participation, older adults were screened for hearing impairment over the phone using the Hearing Handicap Inventory for the Elderly-Screening Version (Yueh et al., 2003); individuals who scored greater than 8 on this assessment, corresponding to a probability of hearing impairment greater than or equal to 50%, were not eligible for participation. Prior to taking part in the study, participants were asked to adhere to the following requirements: no consumption of alcohol, caffeine, or nicotine for 24 hours prior, no cardiovascular exercise for 24 hours prior, no prescription or over-the counter medications for 12 hours prior, awake for at least 3 hours prior, and no eating, brushing teeth, chewing gum, or eating mints for 1 hour prior. Compliance with these requirements was assessed before participants began the study.

A total of 57 younger and 56 older adults participated in the study. Of these, 2 participants (1 younger, 1 older) were excluded from all analyses after disclosing non-compliance with at least 3 pre-study requirements during the experiment. An additional 2 participants in the handgrip group were excluded from all analyses for not having greater mean integrated electromyogram (EMG) signal during the squeeze relative to the rest phases during handgrip runs (Supplementary Methods, Section 1). The sample included for analysis (n = 109) is described in Table 1. Of the 56 younger participants included for analysis, 17 identified as Asian, 5 as biracial, 5 as Black, 5 as “other/not specified,” and 24 as White; of the 53 older participants included for analysis, 8 identified as Asian, 3 as biracial, 12 as Black, 2 as “other/not specified,” and 28 as White. The study was approved by the University of Southern California Institutional Review Board, and all participants provided written, informed consent prior to experimentation. Younger participants received course credit for participation, and older participants received monetary compensation.

Table 1.

Sample characteristics

Age group	Group	N	Age, mean (SD)	Age, range	N (%) female	Edu, mean (SD)	Edu, range	MMSE, mean (SD)	MMSE, range
Younger	Handgrip	28	20.11 (1.73)	18–26	18 (0.64)	13.89 (1.13)	12–16	NA	NA
Younger	Control	28	20.21 (2.06)	18–29	13 (0.46)	13.82 (0.87)	13–16	NA	NA
Older	Handgrip	25	73.92 (5.28)	65–85	13 (0.52)	17.42 (2.62)	14–23	27.68 (2.19)	22–30
Older	Control	28	72.77 (5.58)	66–83	15 (0.54)	16.71 (1.46)	14–20	26.89 (1.95)	23–30

Note. Age and education (Edu) are expressed in years; MMSE = Mini-Mental State Exam.

Procedure

Study overview

The study consisted of a single experimental session lasting between 90 and 120 minutes (Figure 1A). Participants were randomly assigned to either a handgrip or a control group. Blinded to group assignment, participants were told that the purpose of the study was to measure the effects of muscle exertion on cognition. Baseline physiological signals were first recorded for an initial 4-minute resting period, during which participants sat with their feet resting flat on the ground and their palms facing down on a flat surface. Participants then completed runs of either an isometric handgrip protocol (handgrip group) or control protocol (control group), with each run followed by a run of an auditory working memory (n-back) task. Physiological recordings were performed continuously to measure changes in physiological arousal, and saliva samples were collected periodically to assess sympathetic arousal and stress. After the baseline period and after all n-back and handgrip runs, participants completed the Sustained Attention to Response Task; data are not reported here. After all tasks, participants also completed a forward and reverse digit span task as an assessment of working memory capacity. The digit span data are also not presented here.

Figure 1. Overview of experiment, handgrip and control protocol, and auditory n-back task.

Note. (A) After completing a 4-minute baseline resting period, participants completed alternating runs of a handgrip (or control) protocol and an auditory n-back task. (B) During the handgrip and control protocols, participants in the handgrip and control groups heard the same sounds through headphones but performed different tasks: upon hearing “LEFT” or “RIGHT” (squeeze phase), participants in the handgrip group squeezed a therapy ball with their left or right hand, respectively, whereas participants in the control group placed their left or right palm facing down on a flat surface. Upon hearing “BREAK” (rest phase), participants in both groups rested their hands. During each run of the protocol, participants squeezed (or overturned the palm of) each hand twice. (C) As a measure of working memory performance, participants completed an auditory n-back task in which they heard a series of digits through headphones and indicated with the keyboard when the digit they heard matched the digit they heard n digits before. Figure displays the 2-back load, but multiple working memory loads were tested (0-, 1-, 2-, and 3-back; older adults did not complete 3-back). Each n-back run consisted of multiple blocks, each testing a different working memory load. Each block consisted of 18 trials preceded by an 8-second fixation period and a display of the working memory load for that block.

Isometric handgrip protocol

The isometric handgrip protocol (Figure 1B) was based on a finding from our laboratory that a protocol consisting of alternating 18-second squeeze and 60-second rest phases elicited greater increases in salivary alpha amylase and pupil diameter compared to a protocol consisting of a constant 3-minute squeeze period (Nielsen & Mather, 2015). In the present study, each handgrip run consisted of a 10-second initial rest period, followed by 4 repetitions of the following sequence: an 18-second squeeze phase followed by a 30-second rest phase (the rest phase was shortened from 60s as in Nielsen et al. (2015) to reduce the overall duration of the experiment). During each run, participants in the handgrip group held two identical therapy balls, one in each hand, while resting their forearms on a flat surface. They heard sounds through headphones telling them to either squeeze with their left hand (“LEFT”), squeeze with their right hand (“RIGHT”), or to rest their hands in a resting position until the next sound (“BREAK”). Participants squeezed with each hand twice during a single run. To avoid handedness or laterality effects, participants used alternating hands such that they squeezed with each hand twice during a single run. The hand with which the first squeeze was performed was counterbalanced across runs within individual participants, as well as across participants.

Instead of handgrip, participants in the control group performed a task of equivalent length (Figure 1B). After an initial 10-second rest period, participants heard sounds through the headphones telling them to either rest their left palm facing down on the table (“LEFT”), rest their right palm facing down on the table (“RIGHT”), or to rest with their hands in a resting position (“BREAK”). The sounds used in the control protocol were identical to those used in the handgrip protocol. Counterbalancing of the hand used for palm placement was the same as for the handgrip group.

Auditory n-back task

After each handgrip (or control) run, participants performed 1 run of an auditory n-back task (Figure 1C). Within each n-back run, participants completed multiple blocks, each testing a different working memory load. The number of blocks per run differed by age group, with younger participants completing 4 blocks per run (0-, 1-, 2- and 3-back) and older participants completing 3 blocks per run (0-, 1- and 2-back). The 3-back load was not tested for older participants as we anticipated that it could be stress-inducing. Within each run, the order of n-back loads was randomized, and across runs, the order of n-back loads was never the same for an individual participant.

During the task, participants were instructed that they would hear a series of digits through headphones. They were instructed to press the J key when the number they heard matched the number they heard some number of trials before, referred to by the experimenter as the “delay.” They were instructed that this delay would vary across blocks of trials, and that they would be shown the delay on screen before starting each block. At the beginning of each block, before participants were shown the delay number, we included an 8-second resting period for the purpose of assessing tonic pupil diameter. The delay number was subsequently shown on screen for 5 seconds, followed by another 1-second rest, and then participants began the task.

Each block consisted of 18 trials, and the number of target trials - on which a response was expected - depended on working memory load. Placement of target trials within n-back blocks is detailed in the Supplementary Methods (Section 2). Prior to all n-back and handgrip runs, participants reviewed examples of each working memory load with the experimenter. Participants then practiced 1 block of each working memory load in ascending order (e.g., 0-, 1-, 2-, 3-back).

Physiological recordings

To assess physiological arousal and handgrip effort during the experiment, electrocardiogram (ECG) and forearm EMG signals were collected using a BIOPAC MP150 system (Goleta, CA) at a sampling rate of 2KHz. The ECG setup involved a standard Lead I configuration with disposable, pre-gelled Ag/AgCl electrodes; the positive electrode was placed near the lateral end of the left clavicle, the negative electrode near the lateral end of the right clavicle, and the reference electrode near the jugular notch (Biopac Systems, Inc., n.d.). ECG signals were transmitted to the MP150 system using a wireless BioNomadix transmitter (BN-RSPEC-T). EMG signals were collected from the right and left anterior forearm using an EMG100C module with leads connected to disposable, pre-gelled Ag/AgCl electrodes at the following locations: on the flexor digitorum radialis and distal to elbow joint (positive), on the flexor digitorum superficialis and medial to the positive electrode (negative), and just distal to wrist joint (ground). Integrated EMG signals for each arm were computed online from raw EMG signals. ECG and EMG signals were recorded in Acqknowledge (Version 5.0).

Pupil diameter was recorded continuously from both eyes using an SMI RED eyetracker at a sampling rate of 120 Hz. Throughout the experiment, participants rested their head in a chinrest placed at a fixed distance from the experimental computer screen. All tasks in which we planned to analyze pupil diameter involved only auditory stimuli, with a black fixation cross of 40×40 pixels in size displayed on a gray background throughout the tasks.

Salivary assays

To assess sympathetic arousal changes due to handgrip, salivary alpha amylase was assessed immediately before the first handgrip (or control) run and immediately after the first handgrip (or control) run. To ensure that neither the handgrip nor the n-back task elicited stress effects, salivary cortisol was assessed from samples taken immediately after the baseline resting period and after all handgrip and n-back runs. The first saliva sample was collected at least 10 minutes after participants arrived in the lab and consumed 8 ounces of water.

Samples were collected using Salimetrics, LLC (State College, PA) Oral Swabs. Swabs were frozen and shipped to Salimetrics’ SalivaLab (Carlsbad, CA). Samples were assayed using the Salimetrics Salivary Alpha-Amylase Assay Kit (Cat. No 1-1902) and the Salimetrics Salivary Cortisol Assay Kit (Cat. No 1-3002), both without modifications to the manufacturers’ protocol.

Analysis

Physiological data processing and analysis

Prior to preprocessing and analysis, physiological signals for each participant were split into 7 segments: baseline, handgrip runs 1–3, and n-back runs 1–3.

ECG.

ECG and EMG data from 10 participants were missing due to recording errors. Available ECG data segments were visually inspected for quality by two trained researchers. A total of 16 segments (0.02%) with noise or artifact such that QRS complexes were not detectable by either researcher were excluded from further analysis. A summary of excluded ECG segments is provided in the Supplementary Methods (Table S1). QRS detection on raw ECG segments was then performed using the jqrs algorithm as implemented in the Physionet Cardiovascular Signal Toolbox (Version 1.0.2; Vest et al., 2018; parameters described in Supplementary Methods, Section 4). At this stage, an additional 7 ECG segments were excluded from analysis because they contained too little high-quality data for QRS detection. Following r-peak delineation, we used the R package `RHRV` (Version 4.2.6; Rodriguez-Linares et al., 2020) to calculate an instantaneous heart rate signal from heartbeat positions in each segment. Instantaneous heart rate signals were filtered to eliminate outliers and spurious points using the default parameters in RHRV and then interpolated to a sampling rate of 4 Hz. Mean heart rate for events was calculated from the resulting interpolated signals. All heart rate values were baseline-corrected using mean heart rate from the initial baseline period.

To calculate a measure of sympathetic tone, we used the neuECG method described by Kusayama et al. (2020), implemented in MATLAB (Version R2021b). Specifically, we calculated the aSKNA measure, which increases during sympathetic-activating manipulations (Kusayama et al., 2020). In this approach, raw ECG segments deemed to be high-quality were high-pass filtered with a digital finite impulse response filter with a cutoff frequency of 500 Hz. Following full-wave rectification, we applied a leaky integrator with a time constant of 100 milliseconds and calculated the average of the resulting signal (aSKNA) for events of interest. All aSKNA values were baseline-corrected using mean aSKNA from the initial baseline period.

EMG.

For analysis of forearm muscle activity during handgrip, we took integrated left and right arm EMG signals from the handgrip (or control) segments, and computed the mean signal for each squeeze and rest phase. For squeeze phases, we used data from the relevant arm (depending on whether participants heard “LEFT” or “RIGHT”). For rest phases, we averaged signals over the left and right arms. Resulting mean values were baseline-corrected using the mean integrated EMG value across both forearms from the initial baseline period.

Eyetracking.

Eyetracking data from 6 participants were not recorded due to failure of the eyetracker to calibrate (n = 5) and being unable to position the participant in the chinrest (n = 1). Available eyetracking data segments were downsampled to 60 Hz and subsequently preprocessed using the R package `gazeR` (Version 0.1; Geller et al., 2020). First, samples marked as invalid by iView were marked as missing (NA). Blinks were identified using a velocity-based algorithm from the `saccades` R package (Version 0.2–1; von der Malsburg, 2019); in this approach, x- and y-coordinates were smoothed using a moving window average with window size 3 prior to detection and a tuning parameter (lambda) of 6 was applied. Pupil diameter values during blinks were treated as NA, and NAs were extended 100ms before and after blinks in order to remove spurious data points caused by the closing and opening of the eyes before and after blinks (Geller et al., 2020). Pupil diameter data were then linearly interpolated and smoothed using a 10-point moving average. The fraction of missing samples after blink detection and extension for each event of interest was calculated (events of interest included entire baseline segments, squeeze and rest periods during the handgrip protocol, and fixation periods and trials during the n-back task). Baseline segments with more than 70% of samples missing were excluded from relevant analyses; for the handgrip and n-back tasks, events with >50% missing values were excluded from relevant analyses (a summary of excluded events is provided in the Supplementary Methods, Section 5). Pupil diameter values during handgrip were baseline-corrected using mean pupil diameter from the 4-minute baseline period. Baseline correction for n-back segments is described below. For all analyses, left and right pupil diameter data were preprocessed and baseline-corrected separately, then averaged for analysis. If pupil diameter data were missing in one eye but not the other (which occurred in 0% of samples from the handgrip task and 0.22% of samples from the n-back task), values from the non-missing eye were used for averaging.

Statistical analysis

Prior to analysis of handgrip’s effects, we performed Welch’s t-tests to test for age group differences in mean pupil diameter, mean heart rate, and mean sympathetic tone as measured during the baseline period.

Effect of handgrip on concurrent physiological arousal.

To examine how handgrip affected concurrent physiological arousal, we calculated baseline-corrected, average measures of pupil diameter, heart rate, and sympathetic tone during the squeeze and rest phases of the handgrip protocol (for control group participants, we computed average measures for the corresponding control and rest phases). Mixed-design analyses of variance (ANOVAs) were performed to test the effects of group (handgrip/control), phase (squeeze/rest) and age group (younger/older) on each arousal measure. ANOVAs were supplemented with planned comparisons of squeeze-rest differences for each measure, group and age group. Intra-class correlation coefficients (ICCs) were also calculated to assess test-retest reliability of mean values of each arousal measure across runs of the handgrip task.

We then tested whether handgrip effort was associated with arousal increases during the handgrip protocol, for participants in the handgrip group only. Handgrip effort was calculated as the difference between each participant’s mean integrated EMG signal across all squeeze phases relative to that across all rest phases. Each arousal measure’s increase during handgrip was calculated as the difference between each participant’s mean value across squeeze phases relative to that across all rest phases. Pearson correlation analyses were used to test for relationships between handgrip effort and increases in each arousal measure. In a second step, to test for intraindividual associations between handgrip effort and increases in each arousal measure during handgrip, we computed squeeze-rest differences in each measure for each round of each handgrip run, and performed repeated measures correlation analyses using the `rmcorr` R package (Version 0.4.7; Bakdash & Marusich, 2017).

Effect of handgrip on laterality of pupil diameter response.

Since participants squeezed with alternating hands during the handgrip task, we tested whether pupil diameter was higher in the contralateral or in the ipsilateral eye relative to the squeezing hand. This entailed calculating mean pupil diameter for each eye (left/right), across all squeeze phases of the handgrip task, separately for phases in which the pupil in question was contralateral and ipsilateral relative to the squeezing hand. We then performed a mixed-design ANOVA to test the effects of eye laterality relative to squeezing hand (ipsilateral/contralerality), age group (younger/older), and handgrip group (handgrip/control) on pupil diameter. Planned comparisons of ipsilateral-contralateral differences in pupil diameter were performed for each group and age group.

Effect of handgrip on subsequent n-back performance.

N-back accuracy was computed for each participant and working memory (n-back) load as the percentage of target trials correctly identified as targets. 1 older participant was excluded from all n-back analyses for not making any correct responses on the n-back task. Each participant’s reaction times were trimmed to include only values falling within 2.5 standard deviations of their mean reaction time on target trials. The mean reaction time for each participant and working memory load, on target trials only, was calculated. Because significant deviations from normality were observed for aggregated accuracy data (younger: W = 0.829, p < .001; older: W = 0.858, p < .001) and reaction time data (younger: W = 0.943, p < .001; older: W = 0.980, p = .021), we used permuted mixed-design ANOVAs (Kherad-Pajouh & Renaud, 2015) to test the effects of group and working memory load (younger: 0-/1-/2-/3-back; older: 0-/1-/2-back) on accuracy and reaction time. Because younger and older participants completed different working memory loads, these analyses were carried out for each age group separately. Planned comparisons of each performance measure for the handgrip versus control group were performed for each age group and working memory load. In addition, ICCs were calculated to assess test-retest reliability of mean accuracy and reaction time across runs of the n-back task.

Effect of handgrip on subsequent salivary amylase and cortisol.

Next, we tested whether handgrip affected salivary alpha amylase levels, using a permuted mixed-design ANOVA with timepoint relative to the first handgrip run (pre-handgrip/post-handgrip) as a within-subjects factor, and group and age group as between-subjects factors. To confirm that the handgrip and n-back tasks did not elicit a stress response, we also tested for an interaction between timepoint (pre-baseline/after all handgrip and n-back runs), group and age group on salivary cortisol levels. Planned, pairwise comparisons of salivary alpha amylase and cortisol levels at each timepoint were performed for each group and age group separately. Outliers for salivary measures were identified for each age group separately using the mean absolute deviation-median rule (Wilcox, 2011) and removed prior to analysis. Permuted ANOVAs were used after Shapiro-Wilk tests indicated deviations from normality after outlier removal for salivary alpha amylase, W = 0.834, p < .001, and cortisol levels, W = 0.961, p < .001.

Effect of handgrip on subsequent physiological arousal.

We subsequently tested whether physiological arousal levels during the n-back task differed between the handgrip and control groups. As a measure of tonic pupil diameter, we calculated mean pupil diameter during the initial fixation period of each n-back block for each participant. As separate measures of tonic arousal, we calculated mean heart rate and sympathetic tone for each participant during each n-back block. All arousal measures were baseline-corrected using mean values from the initial baseline resting period. Previous work in our laboratory suggested that the effect of isometric handgrip on post-handgrip tonic arousal is time-dependent, diminishing as time after handgrip increases (Mather et al., 2020). To take this into account and also test for handgrip effects on post-handgrip tonic arousal, we performed mixed-design ANOVAs testing the effects of group and n-back block number relative to handgrip offset (1–4 for younger participants; 1–3 for older participants) on each tonic arousal measure. Because the number of blocks in each n-back run differed by age group, these analyses were performed for younger and older participants separately. We performed planned comparisons of each tonic arousal measure for the handgrip versus control group, for each block and age group. Test-retest reliability of each tonic arousal measure across runs of the n-back task was assessed with ICCs.

We then examined whether handgrip affected phasic pupillary responses during the n-back task. To assess phasic pupillary responses, we computed the maximum pupil diameter on each n-back trial. To avoid these values being contaminated by local fluctuations in tonic arousal, values were baseline-corrected using mean pupil diameter during the initial fixation period at the beginning of the respective n-back block. We also computed the onset time of the maximum pupil diameter within each trial. Then, to test whether phasic n-back pupillary responses or their onset times were affected by handgrip and whether these effects depended on working memory load, we performed mixed-design ANOVAs with group as a between-subjects factor and working memory load as a within-subjects factor, for each age group separately. To avoid the influence of trials in which participants may have been distracted, ANOVAs included only trials on which participants correctly responded or correctly withheld a response. Planned, pairwise comparisons of maximum pupil diameter and onset times for the handgrip versus control groups were performed for each working memory load and age group separately. Test-retest reliability of each phasic pupillary measure across runs of the n-back task was assessed using ICCs.

Statistical analyses were performed in R (Version 4.0.4; R Core Team, 2021). ANOVAs were performed with the `afex` R package (Version 1.0–1; Singmann et al., 2021) with Greenhouse-Geisser sphericity corrections automatically applied. Permuted ANOVAs were performed with the `permuco` package (Version 1.1.0; Frossard & Renaud, 2019) and 10,000 permutations. For parametric ANOVAs, comparisons of estimated marginal means were performed with the `emmeans` package (Version 1.7.0; Lenth, 2021). For permuted ANOVAs, a bootstrapping procedure with 1000 bootstrap samples was used for pairwise comparisons. For each set of pairwise comparisons, p-values were adjusted with a Bonferroni correction. ICCs were calculated with absolute-agreement, 2-way mixed effects models using the `irr` package (Version 0.84.1; Gamer et al., 2019). Preprocessing and analyses were performed using a 64-bit Linux kernel with version 5.13.0-7614.

Results

Handgrip increased concurrent physiological arousal

Arousal measures during the handgrip protocol are shown in Figure 2A. Each measure demonstrated good reliability across runs of the handgrip protocol (Supplementary Results, Table S4). Using a mixed-design ANOVA, we found a significant group x phase interaction (p < .001; Table 2A), as well as a significant group x age group x phase interaction on baseline-corrected pupil diameter during the handgrip protocol (p < .001; Table 2A)¹. The 3-way interaction reflected a larger group x phase interaction in younger compared with older participants, driven by a larger squeeze-rest difference in the handgrip versus the control group for younger participants but the opposite pattern for older participants (Supplementary Results, Table S5). Separate ANOVAs indicated significant group x phase interactions on baseline-corrected heart rate (p < .001; Table 2B) and baseline-corrected sympathetic tone (p < .001; Table 2C) during the handgrip protocol. These interactions were driven by each measure being significantly higher during squeeze relative to rest for the handgrip but not the control group, for both younger and older participants (Supplementary Results, Tables S6–7). There were no significant group x age group x phase interactions on either heart rate or sympathetic tone (ps >= .056).

Figure 2. Pupil diameter, heart rate and sympathetic tone and their associations with handgrip effort during the handgrip protocol.

Note. (A) Pupil diameter, heart rate and sympathetic tone are shown across the course of the handgrip protocol. During squeeze phases, participants in the handgrip group squeezed a therapy ball, whereas participants in the control protocol turned their hand from a supine to a prone position. Pupil diameter, heart rate and sympathetic tone values were baseline-corrected using mean values from a 4-minute baseline resting period at the beginning of the experiment. Error bars indicate standard errors of the mean. (B) Scatterplots depict associations between handgrip effort and increases in pupil diameter, heart rate, and sympathetic tone during the handgrip protocol. Handgrip effort was computed as the difference between each participant’s mean integrated EMG signal from the relevant arm during all squeeze phases and that across both forearms during all rest phases. Change in each arousal measure during handgrip was computed as the difference between each participant’s mean value during all squeeze phases and the mean value during all rest phases. 95% confidence bands are shown in gray. A and B reflect data averaged across all 3 runs of the handgrip protocol.

Table 2.

Results of mixed-design ANOVAs testing the effects of group, phase, and age group on pupil diameter (A), heart rate (B), and sympathetic tone (C) during the handgrip protocol

Effect	df	F	η2p	p
A. Pupil diameter
Group	1, 88	2.96	.033	.089
Age group	1, 88	0.00	<.001	.972
Group × Age group	1, 88	1.19	.013	.278
Phase	1, 88	69.33	.441	<.001
Group × Phase	1, 88	15.04	.146	<.001
Age group × Phase	1, 88	5.63	.060	.020
Group × Age group × Phase	1, 88	20.65	.190	<.001
B. Heart rate
Group	1, 92	2.87	.030	.093
Age group	1, 92	9.81	.096	.002
Group × Age group	1, 92	1.14	.012	.288
Phase	1, 92	41.39	.310	<.001
Group × Phase	1, 92	36.98	.287	<.001
Age group × Phase	1, 92	6.93	.070	.010
Group × Age group × Phase	1, 92	3.74	.039	.056
C. Sympathetic tone
Group	1, 92	0.05	<.001	.831
Age group	1, 92	2.01	.021	.159
Group × Age group	1, 92	0.03	<.001	.855
Phase	1, 92	11.54	.111	.001
Group × Phase	1, 92	22.47	.196	<.001
Age group × Phase	1, 92	0.80	.009	.372
Group × Age group × Phase	1, 92	1.42	.015	.236

Note. Pupil diameter, heart rate, and sympathetic tone values were baseline-corrected prior to analysis based on mean values from a 4-minute initial baseline period.

Mean integrated EMG signals, reflecting forearm muscle activity during squeeze and rest phases during the handgrip protocol, are visualized in the Supplementary Results (Figure S1). Associations between average handgrip effort and increases in each arousal measure during the handgrip protocol are depicted in Figure 2B. Across participants in the handgrip group, we found that handgrip effort was significantly correlated with increases in pupil diameter, r(41) = 0.414, p = .006, heart rate, r(41) = 0.431, p = .004, and sympathetic tone, r(41) = 0.405, p = .007. Repeated measures correlation analyses indicated that at an intraindividual level, handgrip effort was significantly correlated with increases in heart rate, but not pupil diameter or sympathetic tone (Supplementary Results, Section 4).

Handgrip had lateralized effects on concurrent pupil diameter

Mean pupil diameter during the squeeze phase of handgrip, separated by ipsilateral vs. contralateral pupil relative to squeezing hand, is shown in Figure 3. A mixed-design ANOVA indicated a significant group x age group x laterality interaction effect on pupil diameter (p = .018; Table 3), as well as significant group x laterality, age group x laterality, and laterality effects on pupil diameter during handgrip (ps <= .003; Table 3). Pairwise comparisons indicated that pupil diameter was greater on the contralateral vs. the ipsilateral side relative to the squeezing hand for the handgrip but not the control group, and this laterality effect was larger for older relative to younger participants (Supplementary Results, Table S8).

Figure 3. Pupil diameter in contralateral vs. ipsilateral eye relative to squeezing hand during handgrip protocol.

Note. Crossbars indicate standard errors of the mean.

Table 3.

Results of mixed-design ANOVA testing the effects of group, age group, and pupil laterality on mean pupil diameter during the handgrip protocol

Effect	df	F	η2p	p
Group	1, 90	6.43	.067	.013
Age group	1, 90	0.16	.002	.686
Group × Age group	1, 90	3.49	.037	.065
Laterality	1, 90	29.08	.244	<.001
Group × Laterality	1, 90	28.77	.242	<.001
Age group × Laterality	1, 90	9.55	.096	.003
Group × Age group × Laterality	1, 90	5.76	.060	.018

Note. Laterality refers to pupil laterality relative to the squeezing hand during handgrip. Pupil diameter values were baseline-corrected prior to analysis based on mean values from a 4-minute initial baseline period.

Handgrip speeded n-back reaction times but did not affect accuracy

Mean accuracy and reaction times on the n-back task are presented in Figure 4. Accuracy and reaction times demonstrated moderate-to-good reliability across runs of the n-back task (Supplementary Results, Table S9). A permuted ANOVA indicated an expected significant main effect of load on accuracy in both age groups (permuted ps < .001; Table 4A), but no significant main effects of group (younger: permuted p = .769; older: permuted p: = .730; Table 4A) or group x load interactions (younger: permuted p = .680; older: permuted p: = .351; Table 4A) on accuracy. However, examining reaction times on the n-back task, in addition to an expected significant main effect of working memory load for both age groups (permuted ps < .001; Table 4B), we found a significant main effect of group on reaction times for both younger and older adults (younger: permuted p = .046; older: permuted p = .017; Table 4B), which was driven by participants in the handgrip group having faster reaction times than those in the control group². There were no significant load x group interaction effects on reaction times in either age group (younger: permuted p = .838; older: permuted p = .274; Table 4B). Pairwise comparisons of accuracy and reaction times by group, for each working memory load and age group separately, are presented in the Supplementary Results (Tables S10–11).

Figure 4. Mean accuracy and reaction times on the n-back task.

Note. Mean accuracy (A) and reaction times (B) on the n-back task, for each working memory load. Crossbars indicate standard errors of the mean. Plots reflect data averaged across all 3 n-back task runs. Older participants did not complete the 3-back task.

Table 4.

Results ofpermuted, mixed-design ANOVAs testing the effects of group and working memory load on n-back accuracy (A) and reaction times (B) for younger and older participants

Age group	Effect	df	F	η2p	Parametric p	Permutation p
A. Accuracy
Younger	Group	1, 54	0.08	.002	.772	.769
Younger	Load	3, 162	77.62	.590	<.001	<.001
Younger	Group × Load	3, 162	0.50	.009	.683	.680
Older	Group	1, 50	0.13	.003	.724	.730
Older	Load	2, 100	59.08	.542	<.001	<.001
Older	Group × Load	2, 100	1.05	.021	.354	.351
B. Reaction times
Younger	Group	1, 54	4.29	.074	.043	.046
Younger	Load	3, 162	61.33	.532	<.001	<.001
Younger	Group × Load	3, 162	0.28	.005	.839	.838
Older	Group	1, 50	6.31	.112	.015	.017
Older	Load	2, 100	22.32	.309	<.001	<.001
Older	Group × Load	2, 100	1.33	.026	.269	.274

Note. ANOVAs were performed separately for each age group because older participants did not complete the 3-back working memory load.

Handgrip did not affect subsequent salivary alpha amylase or cortisol levels

Salivary alpha amylase levels immediately before and after the first handgrip run are presented in Figure 5A. Examining these values with a permuted mixed-design ANOVA, we did not find a significant timepoint x group (permuted p = .648; Table 5A) or timepoint x group x age group (permuted p = .953; Table 5A) interaction on salivary alpha amylase levels. Instead, there was a significant main effect of age group (permuted p < .001; Table 5A), driven by higher salivary alpha amylase levels in older relative to younger participants, as well as a significant main effect of timepoint (permuted p = .029; Table 5A), driven by lower levels after versus before the first handgrip run.

Figure 5. Salivary alpha amylase and cortisol levels during the experiment.

Note. Salivary alpha amylase levels immediately before and after the first handgrip run are shown in A. Salivary cortisol levels before the initial baseline resting period and following all handgrip and n-back task runs are shown in B. Crossbars indicate standard errors of the mean.

Table 5.

Results of permuted, mixed-design ANOVAs testing the effects of timepoint, group and age group on salivary alpha amylase (A) and cortisol (B) levels

Effect	df	F	η2p	Parametric p	Permutation p
A. Salivary alpha amylase
Group	1, 102	0.01	<.001	.909	.906
Age group	1, 102	31.03	.233	<.001	<.001
Group × Age group	1, 102	0.09	.001	.766	.769
Timepoint	1, 102	4.84	.045	.030	.029
Timepoint × Group	1, 102	0.21	.002	.644	.648
Timepoint × Age group	1, 102	0.59	.006	.445	.450
Timepoint × Group × Age group	1, 102	0.00	<.001	.952	.953
B. Salivary cortisol
Group	1, 102	1.82	.018	.180	.178
Age group	1, 102	27.09	.210	<.001	<.001
Group × Age group	1, 102	0.80	.008	.372	.369
Timepoint	1, 102	3.07	.029	.083	.082
Timepoint × Group	1, 102	0.27	.003	.602	.607
Timepoint × Age group	1, 102	4.93	.046	.029	.025
Timepoint × Group × Age group	1, 102	0.42	.004	.517	.514

Note. For (A), salivary alpha amylase levels were assessed at two timepoints: immediately prior to and immediately following the first handgrip run. For (B), salivary cortisol levels were assessed at two points: prior to the initial baseline resting period, and following all handgrip and n-back runs.

We also tested whether the tasks elicited a stress response in either group by assessing salivary cortisol before the baseline resting period and after all handgrip and n-back runs (Figure 5B). A permuted mixed-design ANOVA indicated there was no significant timepoint x group (permuted p = .607; Table 5B) or timepoint x group x age group (permuted p = .514; Table 5B) interaction on salivary cortisol levels. Instead, there was a main effect of age group, driven by higher salivary cortisol in younger compared to older participants (permuted p < .001; Table 5B), and a significant timepoint x age group interaction (permuted p = .025; Table 5B). This interaction effect was driven by greater decreases in cortisol for younger relative to older participants. Results of all pairwise comparisons of salivary alpha amylase and cortisol levels are presented in the Supplementary Results (Tables S12–13).

Handgrip did not affect subsequent tonic arousal

Next, we tested the effects of group and block relative to handgrip offset on tonic measures of arousal during the n-back task (Figure 6). Each measure of tonic arousal exhibited good reliability across runs of the n-back task (Supplementary Results, Table S14). There were no significant main effects of group on fixation pupil diameter (younger: p = .699; older: p = .079; Table 6A), and there were also no significant group x block interaction effects on fixation pupil diameter in either age group (younger: p = .079; older: p = .152; Table 6A). Planned pairwise comparisons indicated that for older participants, fixation pupil diameter was significantly lower in the handgrip relative to the control group for the first block after handgrip (p = .025; remaining comparisons described in Supplementary Results, Table S15). No handgrip-control differences in fixation pupil diameter were significant for younger participants in any block (p’s > .140; Supplementary Results, Table S15). We also found significant main effects of block relative to handgrip, driven by decreases in fixation pupil diameter as block number relative to handgrip offset increased (younger: p < .001; older: p < .001; Table 6A).

Figure 6. Pupil diameter during n-back fixation periods, heart rate during n-back blocks, and sympathetic tone during n-back blocks.

Note. Pupil diameter during n-back fixation periods is shown in A, heart rate during n-back blocks is shown in B, and sympathetic tone during n-back blocks is shown in C. Each measure is plotted by n-back block number relative to handgrip offset, with 1 reflecting the first n-back block after handgrip in a given run of the n-back task. Crossbars indicate standard errors of the mean. Plots reflect data averaged across all 3 runs of the n-back task.

Table 6.

Results of mixed-design ANOVAs testing the effects of group and n-back block number relative to handgrip offset on pupil diameter during n-back fixation periods (A), heart rate during n-back task blocks (B), and sympathetic tone during n-back task blocks (C)

Age group	Effect	df	F	η2p	p
A. Pupil diameter during n-back pre-block fixation periods
Younger	Group	1, 33	0.15	.005	.699
Younger	Block	2.54, 83.94	87.46	.726	<.001
Younger	Group × Block	2.54, 83.94	2.45	.069	.079
Older	Group	1, 26	3.34	.114	.079
Older	Block	1.71, 44.53	27.67	.516	<.001
Older	Group × Block	1.71, 44.53	2.01	.072	.152
B. Heart rate during n-back blocks
Younger	Group	1, 46	0.25	.005	.622
Younger	Block	1.77, 81.37	1.37	.029	.259
Younger	Group × Block	1.77, 81.37	0.62	.013	.523
Older	Group	1, 44	0.53	.012	.469
Older	Block	1.66, 72.95	0.22	.005	.764
Older	Group × Block	1.66, 72.95	1.86	.041	.169
C. Sympathetic tone during n-back blocks
Younger	Group	1, 47	1.15	.024	.289
Younger	Block	2.35, 110.24	8.23	.149	<.001
Younger	Group × Block	2.35, 110.24	0.11	.002	.926
Older	Group	1, 44	0.06	.001	.815
Older	Block	1.52, 66.88	3.12	.066	.064
Older	Group × Block	1.52, 66.88	3.10	.066	.065

Note. Because younger and older participants completed a different number of blocks in each n-back run, these analyses were performed for each age group separately. Pupil diameter, heart rate, and sympathetic tone values were baseline-corrected prior to analysis based on mean values from a 4-minute initial baseline period.

Testing the effects of group and block relative to handgrip on other measures of arousal during the n-back task, there were no significant main effects of group or significant group x block interactions on either heart rate or sympathetic tone (ps >= .065; Tables 6B and 6C). In younger participants, we found a significant main effect of block relative to handgrip offset on sympathetic tone (p < .001; Table 6C), which was driven by decreases in sympathetic tone with increasing block number relative to handgrip offset; this effect was not significant in older adults (p = .064; Table 6C). Results of planned, pairwise comparisons of heart rate and sympathetic tone are reported in the Supplementary Results (Tables S16–17).

Handgrip did not affect subsequent phasic pupillary responses

We then examined how handgrip affected phasic pupillary responses on n-back trials on which participants responded correctly. Pupil diameter time courses, maximum pupil diameter, and onset times of maximum pupil diameter during n-back trials are presented in Figure 7. Pupil time course data are presented for illustrative purposes only. As expected, across the handgrip and control groups, there were significant effects of working memory load on maximum pupil diameter for both younger and older participants (ps <= .001; Table 7A). There was a significant main effect of load on maximum pupil diameter onset time for younger (p = .011) but not for older participants (p = .101; Table 7B). We did not find significant main effects of group, or group x working memory load interaction effects, on either maximum pupil diameter (ps >= .120; Table 7A) or the onset time of maximum pupil diameter during n-back trials (ps >= .142; Table 7B) in either age group. Planned pairwise comparisons did not indicate significant handgrip-control differences in either pupil measure during any working memory load (comparisons detailed in Supplementary Results, Tables S18–19). In addition, intra-class correlation analyses indicated that maximum pupil diameter and onset time of maximum pupil diameter showed poor reliability across runs of the n-back task (Supplementary Results, Table S20).

Figure 7. Smoothed mean pupil diameter time courses, maximum pupil diameter during n-back trials, and onset time of maximum pupil diameter during n-back trials.

Note. Smoothed mean pupil diameter time courses on correct n-back trials are shown in A, maximum pupil diameter during n-back trials is shown in B, and onset time of maximum pupil diameter during n-back trials is shown in C. Correct trials were those in which participants made a correct response or correctly withheld a response, and incorrect trials were those in which participants incorrectly made a response or incorrectly withheld a response. Mean time courses and their 95% confidence bands in A were visualized by downsampling individual time courses to 10Hz and applying a loess smoothing function. Crossbars indicate standard errors of the mean in B and C. Plots reflect data averaged across all blocks and runs of the n-back task.

Table 7.

Results of mixed-design ANOVAs testing the effects of group and working memory load on maximum pupil diameter (A), and the onset time of maximum pupil diameter (B) during n-back trials

Age group	Effect	df	F	η2p	p
A. Maximum pupil diameter on n-back trials
Younger	Group	1, 39	0.68	.017	.415
Younger	Load	2.65, 103.26	53.31	.578	<.001
Younger	Group × Load	2.65, 103.26	1.49	.037	.224
Older	Group	1, 31	0.31	.010	.579
Older	Load	1.89, 58.71	17.28	.358	<.001
Older	Group × Load	1.89, 58.71	2.22	.067	.120
B. Onset time of maximum pupil diameter on n-back trials
Younger	Group	1, 39	2.25	.055	.142
Younger	Load	2.57, 100.1	4.21	.097	.011
Younger	Group × Load	2.57, 100.1	0.79	.020	.484
Older	Group	1, 31	0.37	.012	.546
Older	Load	1.72, 53.41	2.48	.074	.101
Older	Group × Load	1.72, 53.41	0.36	.011	.670

Note. Because older participants did not complete the 3-back load, these analyses were performed for each age group separately. Pupil diameter values during each trial were baseline-corrected prior to analysis based on the mean pupil diameter in the fixation period at the start of the respective block, before participants were aware of the working memory load tested on that block. ANOVAs included only trials for which participants made a correct response or correctly withheld a response.

Discussion

Decades of animal and pharmacological research have indicated that working memory depends on noradrenaline (Arnsten, 2000; Chamberlain et al., 2006; Robbins & Arnsten, 2009). Such findings suggest that behavioral states that vary in their noradrenergic activity levels should show corresponding differences in working memory, but there has been relatively little non-pharmacological work in humans linking working memory with fluctuations in activity of the noradrenergic system. In this study, we examined how isometric handgrip exercise, a manipulation known to engage the noradrenergic system, affected subsequent working memory performance in younger and older adults. In line with previous studies demonstrating beneficial effects of handgrip on cognitive performance (Mather et al., 2020; Nielson et al., 1996; Washio et al., 2021), we found that, compared with participants who performed a control procedure, participants who performed handgrip had faster reaction times on a subsequent n-back task that assessed working memory performance. Handgrip-speeded reaction times on the n-back task were observed for younger and older participants and across working memory loads, although handgrip did not affect accuracy in either age group. These results provide novel evidence that short bursts of isometric exercise can temporarily improve working memory reaction times in both younger and older adults.

Examining multiple measures of physiological arousal during handgrip (pupil diameter, heart rate and sympathetic tone), we found that handgrip increased concurrent arousal. This is consistent with previous studies demonstrating that handgrip temporarily increases pupil diameter, heart rate, systolic blood pressure, and blood flow velocity (Nielsen & Mather, 2015; Washio et al., 2021), with the coronary circulation effects of handgrip mediated by beta-adrenergic receptors (Prodel et al., 2021). Although the effects were in the same direction, overall we did not replicate our previous finding of post-handgrip reductions in tonic pupil diameter (Mather et al., 2020). Furthermore, we did not find effects of handgrip on other measures of subsequent tonic arousal (heart rate and sympathetic tone) beyond tonic pupil diameter, highlighting the need for further studies to confirm handgrip’s effects on subsequent arousal levels. However, older adults in the handgrip group did show reduced pupil diameter compared to those in the control group in the first n-back block after handgrip. This suggests that handgrip does not increase tonic noradrenergic activity after the active squeezing period, so any changes in working memory performance during that post-handgrip period are unlikely to be due to increased tonic noradrenergic activity. A post-handgrip reduction in tonic pupil diameter may be due to noradrenergic depletion, which has been documented to occur in animals following arousing events (Maynert & Levi, 1964; Shinba et al., 2010). One possibility is therefore that during the handgrip period, tonic LC activity is temporarily elevated, and in the period after handgrip, tonic LC activity is reduced to a point that optimizes working memory, in line with the U-shaped relationship between tonic noradrenergic activity and performance (Arnsten, 1998; Aston-Jones & Cohen, 2005). However, it remains to be determined whether the effects of handgrip on working memory performance are driven by increased noradrenergic activity during handgrip or reduced noradrenergic activity in the post-handgrip period. Further studies testing these potential mechanisms in humans are warranted.

Regardless of the mechanism at work, our findings suggest testable neural substrates of handgrip’s effects on performance. Working memory engages the brain’s prefrontal cortex (D’Esposito & Postle, 2015), and ascending projections from the LC reach a number of frontal and parietal brain regions (Pickel et al., 1974; Schwarz & Luo, 2015). Phasic release of noradrenaline from the LC promotes neural gain within frontoparietal brain regions, affecting selective attention and task performance (Arnsten et al., 1996; Aston-Jones & Cohen, 2005; Corbetta et al., 2008). Tonic-phasic tradeoffs in LC activity are also reflected in frontoparietal brain regions: For instance, spontaneous fluctuations in pupil diameter at rest, thought to reflect tonic LC activity, have been associated with activity in frontoparietal brain regions (Breeden et al., 2017; Schneider et al., 2016), and individuals with greater resting-state oscillatory activity in frontoparietal brain regions, reflecting tonic frontoparietal activity, had dampened phasic pupillary responses on a working memory task (Elman et al., 2017). We previously reported that participants who performed handgrip exhibited greater activity of frontoparietal brain regions, lower tonic pupil diameter, and better performance during a post-handgrip oddball task (Mather et al., 2020), implicating frontoparietal brain regions in handgrip’s effects. Future studies should test whether the benefits of isometric exercise for working memory are also mediated by LC-frontoparietal interactions.

That we observed handgrip-speeded working memory reaction times is especially relevant for older adults, since working memory and processing speed exhibit general declines in aging (Rypma & D’Esposito, 2000; Salthouse, 2000; Verhaeghen & Salthouse, 1997). Although one possibility was that due to elevated baseline levels of noradrenaline in aging (Mather, 2020), older participants would exhibit smaller effects of handgrip on working memory performance, we found comparable effects of handgrip on working memory reaction times in both age groups. This finding may furthermore reflect an important link between the noradrenergic system and the frontoparietal network in aging (Bachman et al., 2021; Robertson, 2013). One hypothesis is that a right-lateralized frontoparietal network facilitates the effects of cognitive reserve, an index of the brain’s protection against age-related decline (Robertson, 2014). In particular, the ability of the noradrenergic system to effectively modulate the frontoparietal network may be important for the maintenance of cognitive faculties such as attentional control and working memory in aging (Kennedy & Mather, 2019).

In addition to effects of handgrip on working memory reaction times, we found that handgrip had lateralized effects on pupil diameter during the squeeze phase of the handgrip protocol, with handgrip eliciting greater dilation in the contralateral vs. ipsilateral pupil relative to the squeezing hand. One potential explanation for this finding is that the LC contributes to localized blood flow in active cortical regions (Bekar et al., 2012), which could occur during handgrip in the hemisphere contralateral to the squeezing hand. Goal-directed prefrontal processes may thus stimulate left LC when the left motor cortex needs resources, in order to execute a right-handed squeeze. Stimulation of the contralateral LC relative to the squeezing hand could in turn elicit greater pupil dilation in the eye contralateral vs. ipsilateral to the squeezing hand. Furthermore, the interaction with age group that we observed may indicate age differences in how handgrip affects pupil diameter. Liu et al. (2017) reported that the LC’s influence on ipsilateral pupil dilation occurred via both sympathetic and parasympathetic pathways, but its influence on the contralateral LC occurred only through parasympathetic pathway. Therefore, it is possible that effects on pupil dilation in aging occur primarily via sympathetic pathways, whereas in younger adults, effects on pupil dilation also occur via parasympathetic pathways.

There are several limitations to note. First, the sample size in this study is somewhat small, and additional investigations to confirm these effects using larger samples of younger and older adults are warranted. As additional limitations of the sample, there were age differences in racial distribution, with younger participants including more individuals who identified as Asian and older participants including more individuals who identified as Black, and age differences in education level, with older participants being more highly educated than younger participants. In addition, a possible alternative explanation for handgrip’s effects on working memory reaction times is that performing handgrip primes faster finger-press responses on a subsequent task. We cannot rule out this explanation in the current study, therefore future studies testing other isolated manipulations of physiological arousal that do not involve hand use are warranted. We also note that we did not find effects of handgrip on working memory accuracy. This result is not inconsistent with previous findings that stress affects working memory reaction times but not accuracy (Duncko et al., 2009; Schoofs et al., 2013), but it raises the possibility that handgrip’s effects are on processing speed, which can manifest in faster reaction times across cognitive domains.

In addition, although we demonstrated that handgrip increased concurrent physiological arousal, we did not find group differences in salivary alpha amylase levels immediately before and after the first handgrip run. This is inconsistent with a previous report of handgrip increasing concurrent sympathetic arousal (Nielsen & Mather, 2015). One possibility is that because we did not measure salivary alpha amylase during the squeeze phase but rather after the fourth rest phase of the first handgrip run (approximately 30 seconds after the last squeeze event), group differences in levels due to handgrip could have disappeared by the time of measurement.

Furthermore, based on a previous finding that handgrip enhanced phasic pupillary responses on a subsequent auditory oddball task (Mather et al., 2020), we predicted that post-handgrip reductions in tonic pupil diameter would be coupled with enhanced phasic pupillary responses across working memory loads. Yet we did not find evidence for effects of handgrip on phasic pupillary responses. One possibility is that we may not have detected effects of handgrip due to the poor reliability of the phasic pupillary measures across runs. All other arousal measures demonstrated moderate-to-good reliability across task runs, but the phasic pupillary measures exhibited poor reliability, which could reflect fatigue across the course of the n-back task and/or low power due to exclusions of trial-level pupil data. Another explanation is that phasic pupillary responses may be more pronounced in an oddball task compared to the n-back task used here, in which working memory load may have complex influences on phasic pupillary responses. Additional studies testing whether handgrip affects phasic pupillary responses in the context of a taxed cognitive system are warranted.

Finally, we note the limitations of drawing inferences about LC activity based on pupil diameter. For one, pupil size is influenced not only by the noradrenergic system but also the cholinergic and serotonergic systems (Cazettes et al., 2021; Reimer et al., 2016). In addition, recent studies in mice reported that there is substantial context-related variability in the relationship between LC spiking and pupil diameter (Megemont et al., 2022; Yang et al., 2021), and in humans, it is not yet clear how baseline pupil diameter relates to task-evoked pupil diameter (Martin et al., 2022) or to another putative marker of LC activity, the P3b event-related brain potential (LoTemplio et al., 2021).

In conclusion, we tested how isometric handgrip, a manipulation that temporarily engages the noradrenergic system, affected working memory performance. We found that in both younger and older participants, handgrip decreased reaction times on a subsequent working memory task. Probing mechanisms of handgrip’s effects, we found that handgrip increased concurrent levels of pupil diameter, heart rate and sympathetic tone. Our results provide novel evidence that handgrip affects working memory in humans and offer testable possibilities for the mechanism underlying handgrip’s effects on performance. Furthermore, our findings suggest that isometric exercise represents a strategy to temporarily improve working memory in both younger and older adults.

Public Significance Statement.

This study suggests that isometric exercise is a strategy to temporarily improve working memory performance in younger and older adults. These findings may be of particular relevance for older adults who show declines in processing speed and working memory.