Influence of Perceptual Load on Attentional Orienting in Post-Stroke Fatigue: A Study of Auditory Evoked Potentials

Objective Increasing perceptual load alters behavioral outcomes in post-stroke fatigue (PSF). While the effect of perceptual load on top-down attentional processing is known, here we investigate if increasing perceptual load modulates bottom-up attentional processing in a fatigue dependent manner. Methods In this cross-sectional observational study, in 29 first-time stroke survivors with no clinical depression, an auditory oddball task consisting of target, standard, and novel tones was performed in conditions of low and high perceptual load. Electroencephalography was used to measure auditory evoked potentials. Perceived effort was rated using the visual analog scale at regular intervals during the experiment. Fatigue was measured using the fatigue severity scale. The effect of fatigue and perceptual load on behavior (response time, accuracy, and effort rating) and auditory evoked potentials (amplitude and latency) was examined using mixed model ananlysis of variances (ANOVA). Results Response time was prolonged with greater perceptual load and fatigue. There was no effect of load or fatigue on accuracy. Greater effort was reported with higher perceptual load both in high and low fatigue. p300a amplitude of auditory evoked potentials (AEP) for novel stimuli was attenuated in high fatigue with increasing load when compared to low fatigue. Latency of p300a was longer in low fatigue with increasing load when compared to high fatigue. There were no effects on p300b components, with smaller N100 in high load conditions. Interpretation High fatigue specific modulation of p300a component of AEP with increasing load is indicative of distractor driven alteration in orienting response, suggestive of compromise in bottom-up selective attention in PSF.


Introduction
2][3][4] Despite high prevalence of fatigue after stroke, the underlying neural pathophysiology is poorly understood.While it is commonly agreed that fatigue is a direct consequence of the stroke, 5 neither is there a systematic association with lesion characteristics, 2 nor is stroke induced structural dysconnectivity 6 implicated in the origins of PSF.It is thought to be a result of maladaptive plasticity within neural networks, particularly those networks that subserve sensory processing. 7,8eightened effort perception has been proposed as a cause of fatigue, 7,9 with cognitive dysfunction underpinning altered effort perception.1][12][13] An in-depth analysis of the driving factors of poor performance in selective attention tasks indicate a problem with stimulus encoding and/or action execution rather than longer decision time. 13ecent results from our lab suggest stimulus encoding, specifically distractor stimuli, are linked to high levels of fatigue. 14We observed that with greater task related cognitive load (working memory demand), poorer distractor suppression was seen in high fatigue.Such poor distractor suppression increases perceptual load, but did not influence representation of top-down attentional set (anticipated target stimuli).However, its effect on processing of unanticipated novel stimuli (bottom up attention) remains unknown.Greater perceptual load has previously been linked to diminished bottom-up processing both in healthy humans, 15 and in those with neurological disorders 16 indicative of reduced capacity to process unexpected sensory stimuli, in line with the load theory of perceptual processing which states that the capacity to process perceptual stimuli is finite, and greater load results in attenuation of some stimuli at the expense of other more task-relevant stimuli.In PSF, with poor distractor suppression resulting in greater perceptual load, an attenuated bottom-up attentional response is anticipated. 17Such attenuated bottomup response has implications for one's ability to update behavior using new information and an inability to integrate new information reduces behavioral flexibility.Patient self-report of having to plan activities well in advance and the fatigue inducing nature of unexpected changes to plan are in line with poor behavioral flexibility.Understanding the basis of poor behavioral flexibility will allow one to develop mechanistically informed therapeutic and management strategies for PSF.
Bottom-up attention is investigated using a 3-tone auditory oddball paradigm which evokes event-related potentials (ERPs) recorded using electroencephalography (EEG).The fundamental premise of this paradigm is that when one is tasked with attending to a target tone presented amongst non-target tones, a third, non-target novel tone when presented, despite not requiring a response, involuntarily elicits a response, commonly named an orienting response, indicative of bottom-up processing.Such a response is reflected in the latency and amplitude components of p300a element of ERP, with p300b component reflective of response to an anticipated target tone, a measure of top-down attention and earlier components such as N100 thought to represent perceptual processing of sensory information. 18The effect of perceptual load on the orienting response will be investigated by introducing a "noise" condition where the task is performed in the presence of background noise.Background noise at an amplitude between 55 and 84 dB has been shown to affect the various components of the ERP and thereby increasing the perceptual load of the task. 19Additionally, markers of topdown attention and early sensory processing components will also be investigated to confirm if previously observed lack of effect of poor distractor suppression on top-down attention is replicable, and to identify if early sensory processing is compromised in PSF.
In summary, here we investigate if self-reported fatigue levels in chronic stroke survivors is associated with diminished orienting response driven by perceptual load in a 3-tone auditory oddball task.We test the hypothesis "Chronic stroke survivors with high fatigue exhibit attenuated orienting response when compared to low fatigue counterparts in the presence, but not in the absence of distractor stimuli."

Participants
This was a cross-sectional observational study approved by the London Bromley Research Ethics Committee (REC reference number: 16/LO/0714).Twenty-nine stroke survivors were recruited via the Clinical Research Network from the University College NHS Trust Hospital (UCLH), a departmental Stroke Database and from the community.All stroke survivors were screened prior to the study based on the following criteria: first-time ischemic or hemorrhagic stroke; stroke occurred at least 3 months prior to the study; no clinical diagnosis of any other neurological disorder; physically well recovered following their stroke defined as grip strength and manual dexterity of the affected hand being at least 60% of the unaffected hand assessed using a hand-held dynamometer and the 9-hole peg test (NHPT) respectively; not taking anti-depressants or any other medication that has a direct effect on the central nervous system; not clinically depressed with depression scores ≤ 11 assessed using the Hospital Anxiety and Depression Scale (HADS). 20The above criteria for minimally impaired has previously been used to define minimal impairment in chronic stroke survivors, 21 with comparison with the unaffected side contingent upon the unaffected side retaining full function as described by self-report.Following written informed consent in accordance with the Declaration of Helsinki, 29 stroke survivors participated in the study (Table 1).

Fatigue
Two measures of fatigue were captured at the start of the study, trait, and state fatigue.Trait fatigue represents the experience and impact of fatigue on day to day living for a pre-determined time leading up to the day of testing, whereas state fatigue characterizes fatigue at a given moment in time.Trait fatigue was quantified using the FSS-7, a 7-item questionnaire that rates fatigue over a 1-week period preceding the date of testing. 22State fatigue was quantified using a visual analog scale (VAS) ranging from 0 to 10 in steps of 1 (Not at all tired to extremely tired).A score of ≥5, out of a total of 7 in FSS-7 was necessary to be included in the high fatigue group, and ≤3 was necessary to be included in the low fatigue group.We excluded those with mid-range fatigue (3.1-4.9) as previous investigations in our lab indicated that the self-reported fatigue levels in the mid-range tends to be associated with highly variable underlying neurophysiological markers.Therefore, to ensure robust and definitive demarcation between groups, we chose to exclude the mid-range fatigue.While most previous studies use 4 as a threshold cut-off for high fatigue, the higher threshold of 5 for high fatigue has been previously recommended. 23

Stimuli and Procedure
Participants were seated 70 cm from the monitor (Dell U240 24″ monitor, at a screen resolution of 1280 × 768) and made their responses using a standard USB keyboard.The experiment was controlled using the Psychophysics Toolbox for Matlab, 24,25 running on a Windows computer.A 3-stimulus auditory oddball paradigm was used to elicit N100, P300a, and P300b ERPs both in the presence and absence of noise (Figure 1).Stimuli consisted of 1200 binaural, 80 dB tones of 150 ms duration presented to the participants through headphones (Sennheiser, HD 569), divided into 12 blocks (100 stimuli per block) lasting approximately 4 minutes each.In 50% of the blocks (6 of 12), participants performed the oddball task with no noise, while in the remaining 50% of the blocks, participants performed the oddball task in the presence of noise.The "noise" was an ecologically valid recording of chatter in the café (babble) played at 65 dB.The order of block presentation was counterbalanced across participants.Twelve percent of the stimuli (144 in total, 12 per block) were target tones (1.5 kHz tone), 12% of the stimuli (144 in total, 12 per block) were "novel" sounds (a cricket sound and a sneeze), and 76% were standard tones (1 kHz tone), with an inter-stimulus interval varying between 1.8 and 2.2 seconds.These parameters have been shown to effectively evoke ERPs in previous studies. 26,27articipants were instructed to press a button on the keyboard with the index finger of their right hand in response to   target tones only, while keeping their eyes on a fixation cross.This allowed us to have a measure of response time (time from auditory stimulus onset to button press) and accuracy (% correct of button press) for each participant.
Participants were also asked to rate the effort required to complete a block of the task, on a VAS ranging from 0 (very easy) to 10 (very hard).Before the start of the experiment participants were given the following written instructions that were presented on the screen in in front of them: "You will hear a sequence of tones that we describe as standard tones (example of standard tone).Every so often you will hear a tone of a different frequency that we call the target tone.The tone will sound like this (example of target tone).Every time you hear the target tone press the button on the keyboard in front of you with the index finger of your right hand."Following these instructions participants had to complete practice trials (15 correct trials) in the absence of background noise.After the initial set of practice trials, participants were given the following set of instructions: "In some blocks you will have to perform the same task in the presence of background noise.The background noise will sound like this (example of background noise).Once again, every time you hear the target tone (example of target tone), press the button on the keyboard in front of you with the index finger of your right hand."Participants had to complete another set of practice trials (15 correct trials), in the presence of background noise.Finally, participants were given the following instructions: "In some instances, you will hear a different sound that does not sound like the standard or target tone, for example the sound of a bell.In these instances, please do not press the button.The experiment is divided into various blocks lasting approximately 4 minutes, with a short break between each block.After each block you will have to rate the effort required to complete the block on a scale of 0 to 10, with 0 being very easy and a score of 10 indicating very hard.Please keep your eyes open and focused on the fixation cross throughout the experiment."

EEG Acquisition
Whole-scalp EEG data was recorded using a 64-channel cap array (ActiCap, Herrsching, Germany) and a BrainAmp EEG amplifier system (BrainProducts, Gilching, Germany).The 64 electrodes were positioned on the cap in accordance with the 10 to 20 international EEG electrode array.During online recordings, channels FCz and AFz were used as the reference and ground, respectively.Impedances were kept below 10 kΩ throughout the recording.The EEG signal was sampled at 1 kHz and visualized online using the BrainVision Recorder Software (BrainVision Recorder, Version 1.21.0102Brain Products GmbH, Gilching, Germany).Event markers were sent from the stimulus presentation PC to the BrainAmp amplifier via the TriggerBox which allows one to send accurate triggers via a USB port with millisecond precision.

EEG Analysis
EEG data were pre-processed and analyzed offline using EEGLAB, 28 ERPLAB, 29 and customized MATLAB scripts (Mathworks, Inc., MA, USA).EEG data was down-sampled to 250 Hz and subsequently band-pass filtered between 0.5 and 30 Hz with a zero phase-shift IIR Butterworth filter (24 dB/Oct).Noisy channels were identified and removed using automated procedures.To identify and remove ocular movements and blink artefacts from the EEG data, an independent component analysis (ICA) implemented within EEGLAB was used.The components were visually inspected and those containing ocular movements or blink artifacts were removed.The previously removed channels were then interpolated back into the dataset and finally, the EEG data was re-referenced against the grand average of all scalp electrodes.

Event-Related Potentials
The pre-processed EEG data was segmented into epochs of −200 to 800 ms time locked to the auditory stimulus onset and baseline corrected using the 200 ms pre-stimulus period.Individual epochs were inspected using a 200 ms sliding time window in steps of 100 ms across the entire length of the epoch for voltages exceeding ±100 µV.These trials were subsequently excluded from the analysis (2.5% ± 3.4) of trials.All artefact-free epochs were then averaged for each of the 3 conditions (Standard, Target, and Novel) and filtered using a low-pass IIR Butterworth filter of 12 Hz.The mean number of trials remaining was comparable between groups for each condition (Table 2).
The mismatch negativity (MMN) wave was estimated by subtracting the grand average of the standard tones from the grand average of the novel tones (MMNa) and from the grand average of the target tones (MMNb) in each participant.The N100 amplitude was defined as the instantaneous peak negative amplitude between 50 and 200 ms from the auditory stimulus onset at electrode Cz across the 3 conditions; the P300a amplitude was defined as the instantaneous peak positive amplitude between 250 and 450 ms from the auditory stimulus onset at electrode CPz in MMNa wave; the P300b amplitude was defined as the instantaneous peak positive amplitude between 280 and 650 ms from the auditory stimulus onset at electrode Pz in the MMNb wave.The latency of each peak across the 3 ERPs was also recorded for statistical analysis.These 3 midline locations were chosen as ERP responses were largest at these electrode sites.

Statistical Analysis
All statistical analyses were performed using R with the use of the rstatix package. 30,31Spearman rank correlations were used to identify the association between trait fatigue (FSS-7) and all continuous demographic measures (age, grip strength, NHPT, HADS-Depression, HADS-Anxiety, and Time Post-Stroke), while Wilcoxon rank sum tests were used to identify the association between trait fatigue and all categorical demographic measures (sex, hemisphere affected, type of stroke, and vascular territory affected).
The distribution of the dependent variable was assessed using the Shapiro-Wilk's test of normality, while homogeneity of variances was assessed using the Levene's test.As all data was normally distributed, group differences in the behavioral (response time, accuracy, and effort) and ERP data (amplitude and latency) were examined using a mixed analysis of variance with group (Low Fatigue and High Fatigue) as the between subject factor and noise (Off/On) as the within subject factor.Differences were considered statistically significant at the level of P < .05.Generalized eta squared (η 2 ) was reported for the group effect sizes.Bonferroni corrected pairwise t-tests were used to assess simple main effects in the post-hoc analysis.

Behavior
The response time (RT) results are shown in Figure 2A.As the RT data was not normally distributed, the data were log transformed.The log transformed response time data met the requirements of parametric statistical tests and were normally distributed with a homogeneity of variance.The resulting log transformed data were entered into a mixed ANOVA with noise (Off/On) as the withinsubject factor and fatigue (Low Fatigue/High Fatigue) as the between subject factor.There was a main effect of noise (F (1,27) = 8.12, P = .01,η 2 = .01)and a main effect of fatigue (F (1,27) = 5.40, P = .03,η 2 = .16)but no interaction between noise and fatigue (F (1,27) = 0.17, P = .69,η 2 < .01).Pairwise t-tests showed a significant difference in RT between the 2 fatigue groups in the noise off (P = .025)and the noise on (P = .036)conditions, with those in the low fatigue group having faster response times than those in the high fatigue group.

P300a Amplitude and Latency
Grand average waveforms and topographic maps for P300a are shown in Figure 3.The mixed ANOVA with the amplitude of the P300a as the dependent variable revealed a main effect of noise (F (1,27) = 37.18, P < .01,η 2 = .12),no effect of fatigue (F (1,27) = 3.36, P = .08,η 2 = .10),and a significant interaction between noise and fatigue (F (1,27) = 8.50, P < .01,η 2 = .03).Pairwise t-test showed that there was a significant difference in the amplitude of the P300a between the 2 fatigue groups in the noise on (P = .02)but not in the noise off (P = .45)conditions (Figure 4A), with those in the high fatigue group having a smaller P300a amplitude than those with low fatigue in the noise on condition.To further examine how the presence of noise differentially modulates the amplitude of the P300a across the 2 fatigue groups, the change in amplitude between the noise off and on conditions was calculated and compared between the 2 fatigue groups using a t-test (t-statistic = −2.75,df = 17.8,P = .01;Figure 4B).The mixed ANOVA with the latency of P300a as the dependent variable (Figure 4C) revealed a main effect of noise (F (1,27) = 13.80,P < .01,η 2 = .09),no effect of fatigue (F (1,27) = .12,P = .74,η 2 < .01),and a significant interaction between noise and fatigue (F (1,27) = 4.59, P = .04,η 2 = .03).Pairwise t-tests showed a significant difference in the latency of the P300a between the 2 noise conditions in the low fatigue group (t-statistic = −4.87,df = 15, P < .01)but not in the high fatigue group (t-statistic = −0.95,df = 12, P = .36).The low fatigue group had a shorter latency response in the noise off condition compared to the noise on condition.

N100 Amplitude and Latency
The mixed ANOVA with the amplitude of the N100 as the dependent variable (Figure 5C) had an additional within-subject factor of Condition (Standard, Target, and Novel) to examine whether the amplitude of the N100 was differentially modulated across the different conditions.The mixed ANOVA revealed no effect of condition (F (2,54) = 0.69, P = .44,η 2 = .01).The amplitude of the N100 was therefore averaged across the 3 conditions, and a mixed ANOVA was re-computed with noise as the only within-subject factor.The mixed ANOVA revealed a main effect of noise (F (1,27) = 42.06,P < .01,η 2 = .12)with a smaller amplitude (less negative) in the noise on condition compared to the noise off condition, no effect of fatigue (F (1,27) = 0.14, P = .71,η 2 < .01),and no significant interaction between noise and fatigue (F (1,27) = 1.17,P = .21,η 2 < .01).The mixed ANOVA with the latency of the N100 as the dependent variable (Figure 5D) revealed an effect of noise (F (1,27) = 5.28, P = .03,η 2 = .06)with a shorter latency in the noise off condition compared to the noise on condition, no effect of fatigue (F (1,27) = 0.06, P = .81,η 2 < .01),and no significant interaction between noise and fatigue (F (1,27) = 0.83, P = .37,η 2 = .01).

Discussion
In this study we tested the hypothesis "Chronic stroke survivors with high fatigue exhibit attenuated orienting response when compared to low fatigue counterparts in the presence, but not in the absence of distractor stimuli" and evidence from 29 chronic stroke survivors with minimal impairment, and no depression, confirms the predictions of the hypothesis.We show that fatigue severity scale score is predictive of a greater reduction in orienting response (P300a) with increasing perceptual load.An inverse relationship between fatigue and change in latency of orienting response was seen, with increase in perceptual load associated with longer latency of response in low fatigue compared to high fatigue.No effect of load or fatigue was observed on P300b while amplitude of N100 was reduced with increasing load but not fatigue.We also show that perceptual load prolongs behavioral response times irrespective of fatigue levels, and higher fatigue is associated with slower response times irrespective of perceptual load.Accuracy was unaffected by perceptual load and fatigue, however effort required to perform the task, as indicated by self-report, was higher with greater perceptual load and not affected by fatigue.
The main finding of this study is a fatigue dependent amplitude and latency modulation of P300a response with increase in task-irrelevant perceptual load.A lack of baseline difference in the orienting response between high and low fatigue shows that bottom-up processing is similar across groups in the absence of "noise."However, increasing perceptual load by task irrelevant background noise results in a significant reduction in the orienting response only in high fatigue and not in low fatigue, highlighting the importance of perceptual load in the experience of fatigue.While this study did not directly measure noise encoding, our previous results demonstrating poor distractor suppression with increasing perceptual load suggests that noise related alteration in orienting response is likely driven by poor noise (distractor) suppression. 14In healthy humans, fatigue inducing paradigms also result in poor orienting response only when the paradigm inducing fatigue has high perceptual load with low perceptual load paradigms having no effect on orienting response. 15In Parkinson's disease and traumatic brain injury, fatigue is associated with attenuated p300a while not affecting p300b components, 16,32 indicating similar neural processes might underlie fatigue across different neurological disorders.In disorders such as autism spectrum disorder and schizophrenia, changes in attentional responses, specifically bottom-up processing deficits manifest as a reduced capacity to meaningfully engage with the environment. 33,34Individuals with fatigue also severely restrict their interaction with the environment, and what was previously attributed to reduced motivation due to fatigue, might be driven by abnormal sensory processing.
With increasing perceptual load, the latency of p300a lengthened in low fatigue but not in high fatigue.Latency of p300a reflects stimulus evaluation and attention allocation time, 18 and with increasing perceptual load one expects a lengthening of latency as seen in low fatigue.However, in high fatigue there was no effect of load on latency.On closer examination, we see that average latency in low load condition is similar to that of high load condition in low fatigue suggesting that in high fatigue irrespective of load it takes longer to evaluate stimuli and allocate attention.Fatigue in other diseases such as Parkinson's disease 16 and multiple sclerosis 35 are also related to longer p300a latency, further highlighting the commonalities of fatigue across disorders.
The reduction of N100 amplitude, a marker of early sensory processing, which diminishes with increasing perceptual load, 36 demonstrates the noise condition succeeded in increasing perceptual load as expected.However, there was no effect of fatigue indicating that early sensory processing of auditory stimuli is not compromised in high fatigue.While a lack of difference between groups might indicate a true lack of altered early sensory processing, the increase in load may also not have been sufficient to observe a difference between groups.Chronic stroke survivors with high fatigue report being overwhelmed by sensory stimuli, therefore one might expect an alteration in early sensory processing as seen in other conditions where sensory overload is a significant problem such as autism spectrum disorder. 37here were no load specific behavioral consequences that were exclusive to the high fatigue group as would be expected from the load specific attenuation of orienting response in the high fatigue group.Both speed and accuracy of response, and task related effort were heightened with increasing perceptual load across both groups.A lack of behavioral effect could either be a result of the task difficulty not reaching a threshold for the underlying neural processing changes to reflect on behavior, or, in keeping with previous results, 21,38 such altered sensory processing exclusively informs "sensory awareness" without influencing motor output.

Strengths and Limitations
Despite relatively small number of participants which maybe considered a limitation of this study, this welldefined and homogenous groups of stroke survivors with primary fatigue (no overlapping comorbidities) and minimal impairment suggest the results of this study are likely a true indication of fatigue related processes and not driven by uncontrolled variables.A possible source of variability, which cannot be assessed in this study, could be the initial nature of the insult (ischemic vs hemorrhagic stroke) or the different locations of the stroke, however, numerous previous studies have indicated that initial stroke characteristics do not have a bearing on the incidence or severity of PSF. 6,39n summary, we show that processing of novel stimuli in the presence of background noise is significantly altered in PSF, indicating a problem with attentional orienting processes, one that selectively contributes to the feeling of fatigue without affecting performance.This evidence provides a physiological basis for self-reported fatigue-related difficulties of processing several streams of sensory information.These results pave the way for future research studies to investigate common fatigue mechanisms across diseases where sensory processing is compromised and develop new therapeutic interventions.
Abbreviations: NHPT, 9-hole peg test; SDMT, symbol digit modalities test; HADS, Hospital Anxiety and Depression Scale; ACA, anterior cerebral artery; MCA, middle cerebral artery; PCA, posterior cerebral artery.The median and interquartile range is shown for continuous variables, while the count is shown for categorical variables.a Median (25%,75%); n.

Figure 1 .
Figure 1.Illustration of the task design.Target (TGT) tones at 1.5 kHz requiring a requiring a response via a keyboard and novel tones were presented amongst a sequence of standard (STD) tones at 1 kHz.

Figure 2 .
Figure 2. Behavioral results.Box plots indicating the median and inter-quartile range with individual data points for the low fatigue group in yellow and the high fatigue group in green for response time (A), accuracy (B) and effort rating (C) across the 2 background noise conditions (Off and On).Significant differences between the groups are indicated with asterisks (*P < .05).

Figure 3 .
Figure 3. Grand average ERP of the P300a at electrode CPz and topographical plot for the high (A) and low (B) fatigue groups.The noise off condition is in red and the noise on condition is in blue.The topographical plots illustrate the difference between the 2 noise conditions.The grey box indicates the time window within which the P300a was measured.

Figure 4 .
Figure 4. P300a results.Box plots indicating the median and inter-quartile range with individual data points for the low fatigue group in yellow and the high fatigue group in green for the amplitude (A) and the difference in amplitude across the 2 noise conditions (B) for each fatigue group of the P300a.Box plots with individual data points for the latency of the P300a across the different noise conditions, with noise off in red and noise on in blue, is shown in panel C. Significant differences between the groups are indicated with asterisks (*P < .05).

Table 1 .
Demographics of the High and Low Fatigue Groups That Took Part in the Study.

Table 2 .
Number of Trials for Each Condition Across the 2 Fatigue Groups.Note.The mean number of trials and the standard deviation across each condition, as well as the P-value for the difference between the 2 groups is shown.
a Mean (SD).b Wilcoxon rank sum test.