Sustained attention to response task-related beta oscillations relate to performance and provide a functional biomarker in ALS

2.1.1. Recruitment

2.1.2. Inclusion criteria

2.1.3. Exclusion criteria

The experimental paradigm is identical to that described in detail in our previous SART study [3], briefly described here. Participants were seated at a distance of 1 ± 0.1 m in front of a monitor. Individual digits of one to nine were shown in random order for 250 ms with an interstimulus interval randomised between 1120 and 1220 ms during which the screen was black. Responses were captured by clicking the left button of a computer mouse with the right index finger. Four recording sessions of 252 trials, with the number 3 appearing in 11% of trials, were undertaken. The behavioural measures captured were: response time, overall response accuracy, NoGo accuracy (percentage of 3-digit stimuli followed by response omission), Go accuracy (percentage of non-3 digit stimuli followed by a response in the permitted time window) and anticipation rate (clicking less than 150 ms after a go stimulus). Participants were instructed to click every time they saw a number except for the number three, to equally prioritise speed and accuracy and to use only their finger to respond, to reduce EMG artefact in the EEG signal. Participants performed one practice round without performance being measured under supervision of the experimenter.

2.3.1. EEG

2.3.2. ALS functional rating scale revised (ALSFRS-R)

2.3.3. Cognitive and behavioural tests

2.3.4. Disease duration

2.4.1. EEG signal pre-processing

2.4.2. Time-frequency analysis

Each trial epoch consisted of 563 data points (i.e. 1100 ms of data recorded at 512 Hz) per channel. A random subset of Go trials followed by correct responses (a click) were chosen to match the number of NoGo trials followed by correct response (no click) for further analysis. Complex Morlet wavelets were chosen for time-frequency analysis as they have a sinusoidal basis with symmetric Gaussian envelopes and their width can be adjusted for the desired number of oscillations. To facilitate the application of the complex Morlet wavelet transform, for each channel, each epoch was data-padded with 563 repetitions of the first data point at the start, and 563 repetitions of the last data point at the end. As data had been low-pass filtered at 35 Hz during pre-processing for a previous time-domain analysis [3], calculation of wavelet moduli was performed for a range of 1–35 Hz. The complex coefficients of transform (W) were subsequently re-epoched by removing the padded data. Mean inter-trial phase variance (ITV) across epochs (e) was calculated per time point (t) and frequency (f) to provide a measure specifically of non-phase locked (i.e. not captured by our previous event related potential (ERP) analysis) oscillatory activity [13] as:

$$\begin{equation*}{\text{ITV}}\left( \,{f,t} \right) = \frac{{\sum\nolimits_1^{{N_{\text{e}}}} {\left| {{{({W_{f,t,e}} - {{\overline W}_{f,t}})}^2}} \right|} }}{{{N_{\text{e}}} - 1}}\end{equation*}$$

where $\overline W$ denotes mean value of W across all epochs and ${N_{\text{e}}}$ denotes the total number of epochs. ITV were calculated for the Fz, Cz and Pz channels. These electrodes were chosen due to their proximity to executive and motor function associated cortical domains. Baseline values for each frequency were calculated as mean ITV across the −200 ms to 0 ms window. Event related spectral perturbation (ERSP) values for each time and frequency was then calculated as:

$$\begin{equation*}{\text{ERSP}}\left(\, {f,t} \right) = 100 \times \frac{{{\text{IT}}{{\text{V}}_{f,t}} - {{\overline {{\text{ITV}}} }_{f,{\text{ baseline}}}}}}{{{{\overline {{\text{ITV}}} }_{f,{\text{ baseline}}}}}}\end{equation*}$$

for t = 1–900 ms. Time–frequency analysis was performed separately for Go and NoGo trials. Analysis of the difference between NoGo and Go trials was performed by subtraction thereafter.

2.5.1. Oscillation analysis

2.5.2. Effect sizes and correlations

2.5.3. Demographics

2.5.4. Selected measures for reporting the main findings

The same ALS patient and control datasets were employed as in our previous time-domain study [3], with one additional patient dataset analysed here. Therefore, a total of 24 ALS patients (3 female, age median [interquartile range]: 69 [59–72] years) and 33 controls (17 female, age median [interquartile range]: 64 [57–69] years) were included in this analysis. Groups were age-matched but not gender matched, as previous comparison of male and female controls for parameters of interest revealed no gender-related differences [3]. ANOVA found significant group effects for some ROI (described below) but no significant gender effects. The ALS cohort included three patients with a C9orf72 gene expansion, with the remainder of patients not carrying or not tested for this pathogenic expansion. Site of onset of disease symptoms was spinal in eighteen patients, bulbar in five patients, and thoracic in one patient. Disease duration and time since diagnosis median [interquartile range] of patients was 17.65 [10.15–23.92] months and 4.42 [2.98–9.71] respectively. ALSFRS-R, ECAS and CWIT scores collected within 90 d of EEG were available for 18, 15 and 9 patients respectively. Survival data were not analysed as only two patients were deceased at time of analysis.

Mean response time was 364 ms in controls and 375 ms in patients. Mean ± standard deviation trial accuracy (Go and NoGo) was 99.67 ± 0.48 and 78.31 ± 12.73% in controls and 98.96 ± 1.60 and 77.90 ± 12.53% in patients respectively. Control and patient accuracy and response time measures were not significantly different. Patients had significantly (p = 0.0042) greater anticipation errors than controls (patient mean ± standard deviation: 0.084 ± 0.14%, control: 0.01 ± 0.03%).

3.3.1. Go trials

During Go trials, theta-band (4–7 Hz) ERS (i.e. increase in power relative to baseline), alpha-band (8–12 Hz) ERD (i.e. decrease in power relative to baseline) and beta-band (13–30 Hz) ERD were present across the frontoparietal axis. Beta ERD was followed by ERS, predominantly in Fz and Cz. These ERSP patterns in healthy controls, as informed by sign rank statistical analysis, were examined further as ROIs for comparing ALS patients against controls and when performing correlation analyses in 100–700 ms, 250–600 ms, 200–500 ms and 650–850 ms post-stimulus time windows respectively (figure 1(A)).

Zoom In Zoom Out Reset image size

3.3.2. NoGo trials

3.3.3. The difference between NoGo and Go trials

NoGo trials differed from Go trials by greater theta band ERS over Fz and Cz, greater alpha ERD (i.e. greater event-related reduction in oscillatory power in NoGo trials relative to Go trials) and reduced slow beta (13–22 Hz) ERS (i.e. less event-related increase in oscillatory power in NoGo trials relative to Go trials) in all three electrodes. These ERSP differences in the control group were in the 150–450 ms, 500–900 ms and 600–850 ms post-stimulus time windows respectively (figure 2), and were defined as additional ROIs (for comparing ALS patients against controls and when performing correlation analyses).

Zoom In Zoom Out Reset image size

Analysis of the entire time-frequency plane identified significantly reduced beta-band ERD in Fz and Pz during Go and NoGo trials in ALS compared to controls (AUROC values illustrated in figure 3, effect sizes and AUROC values listed in table 1). These findings are in keeping the findings of ANOVA, which identified significant group effect (across electrodes) on beta-band power in this window (p values listed in table 1). This ANOVA also identified significant reduction in Go trial beta ERS (650–850 ms, p = 7.72 × 10⁻⁴) across electrodes, which was predominantly accounted for by Fz, the only individual electrode to show significant difference between groups at post hoc testing (Tukey's p = 0.048). No significant differences between patients and controls were found for the difference between NoGo and Go trial ERSP.

Zoom In Zoom Out Reset image size

Correlations between late beta-band ERS in Go (650–850 ms post stimulus) and NoGo (750–850 ms post stimulus) trials and task performance measures are summarised in table 2. Significant negative correlation was identified between response accuracy and late beta band ERS in Pz during Go trials (i.e. poorer accuracy with greater beta ERS) for controls but not patients.

Significant negative correlations between response time and beta-band late ERS in patients and the overall group (i.e. faster response time with greater beta ERS) across the electrodes of interest were present in controls alone as a trend but were not significant following multiple comparison correction. Theta band ERS in Pz during Go trials showed significant negative correlation with response time in the total group (p = 0.010, rho = −0.34) with similar trends when patients and controls were considered separately (patients: p = 0.055, rho = −0.39, controls: p = 0.052, rho = −0.34). Patient, but not control, response times also negatively correlated with Cz alpha band ERSP in NoGo trials (i.e. greater alpha ERD was associated slower response times, p = 0.013, rho = −0.5). Patient ECAS ALS-specific score was correlated with beta ERS (750–850 ms post stimulus, p = 0.0024, rho = 0.72) over Cz during NoGo trials (i.e. greater executive performance with greater beta ERS, figure 4). No significant correlations were identified for other ROIs, disease duration or ALSFRS-R or CWIT scores.

Zoom In Zoom Out Reset image size

Beginning at approximately 150 ms post-stimulus, alpha and beta band ERD were observed during both Go and NoGo trials across the frontoparietal axis, in addition to theta band ERS over the frontal lobe (in Fz and Cz), which was greater during correct response withholding. Beta ERD was followed by ERS, which was significantly reduced during correct response withholding.

Motor tasks evoke well-characterized movement-related beta desynchronization (βMRD), beginning in the second before movement onset and peaking during movement performance, followed by movement-related synchronization (βMRS) during movement termination [20, 21]. βMRD is associated with motor planning and execution, while βMRS reflects inhibition of the motor networks to terminate the motor program [22]. Like SART-related beta ERS observed here, βMRS is maximal in Cz and larger in Go trials than NoGo trials, and reaches significance at approximately 800 ms post-stimulus [22]. Similar to the SART, a Go/NoGo task not designed to test sustained attention is also found to elicit beta ERD/ERS during Go and NoGo trials, with Go trial ERS inversely correlating with response time [23]. These similarities indicate that the SART captures this measure of motor cortical activation and inactivation in addition to those of attention and response control.

Alpha ERD/ERS did not correlate with control task performance in this study, such that the ability to determine the role of this ERD in SART performance is limited. Alpha ERD is associated with thalamocortical network excitation [24] and release of the task-engaged cortical regions from inhibition [25]. This measure has been captured during a number of other attention and memory tasks, and is considered to reflect retrieval of task-relevant information from one's 'knowledge system' [25, 26].

Peri-movement alpha ERD is also observed in Go/NoGo tasks not designed to test attention or memory, and represents a general disinhibition of the motor networks to facilitate movement. In keeping with our observations, this movement-related alpha band ERD (αMRD) generally persists for longer than βMRD and does not typically rebound to synchronization [27]. However, αMRD is not found to be greater during NoGo trials of these tasks [23, 27, 28], as we observed here, while cognitive alpha ERD increases with task complexity. Therefore, this alpha change may not be a purely motor cortical phenomenon and requires further characterization by larger, source level studies to differentiate potential cognitive and motor underpinnings.

Theta band ERS showed significant correlation with SART response time in the overall group, with similar trends within the individual groups that were probably underpowered to detect this effect in each single group. An n-back task study, which also identified frontocentral-predominant theta ERS peaking approximately 250 ms after stimulus delivery, demonstrated association of this ERS with attention allocation, rather than working memory [29]. Further, frontal midline theta, a focal increase in theta power induced by numerous cognitive tasks and localized to the dorsal anterior cingulate and medial prefrontal cortex [30], reflects attentional processing [31]. This is consistent with the presence of theta ERS for all trial types, with greater magnitude during NoGo trials.

Utility of the SART as a test of sustained attention has been criticized due to the extent of performance variation within healthy populations, which has been attributed to the speed-accuracy trade-off [2]. Task performance requires sufficient working memory, attention, response inhibition and motor control, among other functions. Therefore, differences in performance measures such as response time and accuracy may reflect normal or abnormal differences in an array of cortical functions. However, this lack of specificity is advantageous when EEG is recorded simultaneously, as a battery of measures which individually interrogate each of these functions, differentiated by their spatial, temporal and frequency characteristics, can be measured from a single paradigm.

We have previously demonstrated how time-domain analysis of SART-EEG provides individual measures of response control and attention, facilitating identification of specific cognitive and motor functions affected in ALS [3]. Here we have extracted non-phase locked cortical oscillatory changes across time and frequency domains, capturing additional measures of specific network activity and communication that were lost through averaging in our previous analysis and which are not frequency-domain reflections of event related potentials. We have demonstrated that individuals who prioritize speed over accuracy display greater beta ERS during Go trials, potentially reflecting greater post-movement motor cortical inhibition in these individuals. While the causative relationship between this measure of motor cortical deactivation and task approach warrants further investigation, the specificity of this correlation to this ROI within the time-frequency plane facilitates separation of this variation in speed-accuracy trade-off observed in healthy cohorts [2] from other, pathological, changes in cortical networking captured at other times and frequencies, such as those we have identified in ALS.

ALS patients showed significant reduction in frontal and parietal beta ERD. βMRD elicited by motor preparation-specific paradigms [32] has previously been shown to be reduced in ALS. This has been proposed to reflect upper motor neurone degeneration, although motor function of the upper extremities does not correlate with βMRD [33, 34]. SART-elicited beta ERD similarly did not correlate with task performance, in alignment with the lack of difference in task response time or accuracy between patients and controls. The lack of significant change in Cz beta ERD in ALS patients observed here also indicates measurement of broader motor network dysfunction beyond the precentral gyrus, such as in the premotor, supplementary motor and posterior parietal cortices. Therefore, while the similarity of this ERD to βMRD suggests that they are of similar (or the same) motor physiological basis, source localisation is required to clarify the specific generators of these oscillations.

Regardless of its physiological origin, the large effect size (Cohen's d = 1.12) and good discrimination (AUROC = 0.82) of ALS patients from controls by beta ERD highlights the need for further exploration of this promising measure as a biomarker of ALS and ALS subphenotypes. Further, as the symptoms of neurodegenerative diseases such as ALS can limit the duration of data collection sessions, the ability of SART to simultaneously elicit a number of distinct measures of motor, motor preparatory and executive function could maximise the efficiency with which cognitive and motor networks are interrogated in both research and clinical settings.

As a group, ALS patients also showed reduced Go trial beta ERS, predominantly over the prefrontal cortex. While the correlation between Go trial beta ERS and poorer response accuracy observed in controls was absent, strong correlations were observed between NoGo trial beta ERS over the motor cortex and executive performance in patients. This correlation, alongside existing literature (described above), may reflect ALS patients with sufficient executive function exerting increased prefrontal control over motor cortex activation to sustain task performance. Patient (but not control) response times were also longer in those with more central alpha ERD in NoGo trials. Together, these findings indicate that sustained performance in patients is achieved through balancing pathological dysfunction with compensatory engagement in cognitive and motor networks. These findings exemplify the utility of EEG in capturing cortical network (dys)function in disease with greater sensitivity and source specificity than task performance measures, which do not emerge until cognitive reserve and alternative neural networks can no longer compensate [35]. However, larger dataset collection is now required to perform comparisons of these measures between clinical, genetic and disease stage ALS subcohorts and facilitate further interrogation of the relationship between this cortical pathophysiology and cognitive and motor symptom severity with higher statistical power. Further, the pathological or compensatory roles of ERSP in ALS require further elucidation through longitudinal and source level analyses.

This analysis focussed on three electrodes of interest across the frontoparietal axis in order to simultaneously investigate previously unexamined SART-associated cortical oscillations and perform preliminary screening for potential ALS biomarkers. Spatial resolution of these findings is poor. Therefore, while they capture the activity of important generators of SART response across primary motor, pre-motor and supplementary motor areas (in the motor domain) and prefrontal and parietal generators (in the cognitive domain), our ability to attribute different ERSP to specific cortical regions is limited. Source-localised analysis will be needed in a future study to elucidate the sources of abnormalities in cognitive oscillations across the cortex, informed by a dense electrode montage, and potentially increase the discriminative ability of these measures in detecting ALS. Expansion of the datasets shall facilitate further interrogation of the relationships between these measures of cortical pathophysiology and disease stage, rate of progression and symptom severity.