Pupil-linked arousal, cortical activity, and cognition in Alzheimer's disease

Abstract

Arousal dysfunction contributes to impairments seen in Alzheimer's disease. However, the nature and degree of this dysfunction have not been studied in detail. We investigated changes in tonic and phasic arousal using simultaneous pupillometry-EEG, relating these changes to locus coeruleus integrity, a key arousal nucleus. Forty Alzheimer's disease participants and 30 controls underwent neuropsychological testing using the Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-Cog), MRI designed to show contrast in the locus coeruleus as a measure of integrity and simultaneous pupillometry-EEG during 5 min of eyes-open resting-state. Pupillometry-EEG was then also applied during an oddball task which included a passive session and sessions in which responses to target stimuli were required, to test the effect of salience. Alzheimer's disease had lower locus coeruleus integrity (b = -0.26, P = 0.02) and lower peak alpha frequency (tonic arousal) (b = -1.09, P < 0.001). Both were related to ADAS-Cog. There was a very strong relationship between pupil size and both periodic and aperiodic EEG power. Cortical slowing in Alzheimer's disease affected this relationship, particularly at low frequencies. During the attentionally demanding oddball task, Alzheimer's disease participants' behavioural performance was impaired, with reduced accuracy and slower and more variable reaction times. They also had reduced pupil responses to salient stimuli (phasic arousal) (estimate = -0.19, P < 0.001). EEG and pupil measures of pre-stimulus tonic arousal were strongly correlated and predicted behavioural responses in both groups. Arousal fluctuations at rest and in response to stimuli are abnormal in Alzheimer's disease as measured by combined pupillometry and EEG. Salient stimuli that require a behavioural response are accompanied by a phasic increase in arousal, demonstrated by pupil dilation to oddball stimuli. This response is slower and of smaller magnitude in Alzheimer's disease patients. Cortical slowing (reduced peak alpha frequency) is seen in Alzheimer's disease, and this is modulated by arousal level and relates to overall cognition. Pupil-linked arousal responses and alpha EEG fluctuations are tightly coupled, but cortical slowing in Alzheimer's disease influences this coupling. The tools used here to measure neurophysiological arousal level have potential in understanding the nature of arousal system dysfunction in Alzheimer's disease at the group level. These tools may also be used as biomarkers at the individual level in order to target patients most likely to benefit from arousal-modulating medications.

Keywords: attention; neuromodulation; noradrenaline; oddball; pupils.

Graphical Abstract

Figure 1

Oddball task paradigm. Schematic detailing the oddball task paradigm, consisting of three sessions: ‘passive’, ‘frequent’ and ‘oddball’. ISI, inter-stimulus interval.

Figure 2

Relationship between locus coeruleus contrast, EEG slowing and cognition in Alzheimer’s disease. (A) Violin plot comparing locus coeruleus contrast between groups calculated as a ratio between the contrast in the locus coeruleus and a reference region in the pons (b = −0.26, P = 0.02). (B) Relationship between locus coeruleus contrast and ADAS-Cog total score (b = −0.01, P = 0.02). (C) Spectra of normalized power separated by group showing a shift to lower frequency of the alpha peak in AD. (D) Violin plots showing a reduced peak alpha frequency in AD (b = −1.09, P < 0.001). (E) Scatter plot of AD individuals’ peak alpha frequency against their ADAS-Cog score (b = −4.37, P = 0.03). (F–H) Violin plots comparing log power measured in arbitrary units (au), in the delta, alpha and beta bands, respectively, during the resting-state, between the two groups (all non-significant). (I) Histograms showing distribution of pupil size data for all individuals overlayed, demonstrating no difference in the shape of the distributions between subjects in the AD group (pink) and the control group (blue). (J) Violin plots showing no significant difference between groups in the number of spontaneous pupil dilations during resting-state. For B, C and E, shaded areas represent standard error of the mean. All statistics shown are results of linear regression including age and sex as covariates. Plotted line in scatter plots represents best fit through the raw data. HC, healthy control; AD, Alzheimer’s disease; ADAS-Cog, Alzheimer’s Disease Assessment Scale–Cognitive Subscale. *P < 0.05; ***P < 0.001.

Figure 3

Pupil size relates to EEG power and alpha frequency during resting-state. (A) Pearson correlation values (β) with pupil size across all channels, sorted from anterior (top) to posterior (bottom), and for lags ranging from −3 to +3 s, averaged within three frequency bands of interest: delta (1–4 Hz) (left), alpha (6–10 Hz) (centre) and beta (12–30 Hz) (right), split by group (HC above, AD below). Negative lags are indicative of EEG preceding pupil, and vice versa. Positive correlations are depicted with red and negative with blue. Point of 0 s lag marked with solid vertical black line, −1 s lag with dashed line and −2 s lag with dotted line. Bottom row shows regions highlighted in grey where HC β > AD β (P < 0.05). (B) Example distribution of pupil data from one participant, arranged by size, binned into deciles (each decile in a different colour). (C) Representative snapshot of pupil time course with deciles indicated by colour. (D) Peak alpha frequency (centre of gravity), as a function of pupil size, as above, split by group (HC above in blue, AD below in pink). (E) Z-scored spectral power, averaged across the same three frequency bands, as a function of pupil size, split by group (HC above in blue, AD below in pink). Top plots surrounded by solid box represent point indicated with a solid line in A at time of 0 s lag between pupils and EEG power. Middle plots surrounded by dashed box represent point indicated with a dashed line in A at time of −1 s lag. Bottom plots surrounded by dotted box represent point indicated with a dotted line in A at time of −2 s lag. For D and E, the data points represent data for 10 deciles which is averaged at the decile level within each subject and then across subjects within each group. Adjusted R-squared and P-values for each correlation found in Supplementary Table 1 or 2. Significant correlations (P < 0.05) indicated with best fit line. Linear or quadratic line used according to that with lower Bayesian information criterion. AD, Alzheimer’s disease; HC, healthy control.

Figure 4

Pupil size relates to aperiodic EEG power during resting-state. (A) Spectra of log aperiodic power in arbitrary units (au) separated by group. (B) Violin plots showing the exponent of the aperiodic power spectra. There was no significant difference between groups when accounting for age and sex. (C) Exponent of aperiodic power as a function of pupil size (at −1 s lag), shown for all participants (grey) (R² = 0.84, P < 0.001) and split by group [HC (blue) R² = 0.51, P = 0.01; AD (pink): R² = 0.96, P < 0.001]. The data points represent data for 10 deciles which is averaged at the decile level within each subject and then across subjects within each group. Best fit line, adjusted R-squared and P-values on plot refer to grey scatter points. AD, Alzheimer’s disease; HC, healthy control.

Figure 5

Pre-stimulus pupil size and EEG power relates to pupil dilation response and reaction times. (A) AD group (pink): top row shows pupil dilation responses to correct oddball target trials only, split into quintiles by baseline pupil size. Second row shows baseline (0.5 s pre-stimulus) relative alpha EEG power (rank converted) (left) (HC: b = −0.24, P = 0.004; AD: b = −0.17, P = 0.02), RTs (centre) (HC: b = 0.19, P = 0.02; AD P > 0.05) and CoV of RTs (right) (HC: P > 0.05; AD: b = 0.16, P = 0.01), both rank converted, as a function of baseline pupil size quintile, for all correct target trials (frequent and oddball). (B) As in A but for controls (blue). (C) AD group (pink) and (D) HC group (blue): pupil dilation responses (left) (HC and AD: P < 0.05), RT (HC: b = 0.32, P < 0.001; AD: b = 0.17, P = 0.02) and CoV of RTs (HC and AD: P < 0.05), as a function of baseline alpha power. Significant relationships (P < 0.05) indicated with best fit line; linear in light grey and quadratic in dark grey. *P < 0.05; **P < 0.01; ***P < 0.001. AD, Alzheimer’s disease; HC, healthy control; RT, reaction time; CoV, coefficient of variation.

Figure 6

Smaller, slower pupil dilation and slower and more variable reaction times, in Alzheimer’s disease. (A–G) Group-average pupillary response curves for the seven trial types, showing pupil size from 0.5 s pre-stimulus to 2 s post-stimulus, normalized to a 0.5 s pre-stimulus baseline. Stimulus onset marked by vertical dashed line. Incorrect trials and those with incomplete data following blink interpolation excluded. Periods of significant difference between the lines (t-test) indicated by black bar. AD in pink and HC in blue. Insets show box plots comparing group-average pupil dilation responses (average baseline-corrected pupil size in 2 s post-stimulus period) with significant differences indicated where applicable. (H) Scatter plot showing relationship between ADAS-Cog total score and pupil dilation in response to oddball targets in AD group only (b = −0.01, P = 0.03). (I) Violin plot showing group difference in accuracy (percentage of trials correct from frequent and oddball sessions) (b = 4.37, P = 0.03). (J/K) Violin plots showing group differences in reaction times from all target trials (oddball and frequent) (reaction times: b = −0.07, P = 0.02; CoV: b = −0.10, P < 0.001). (L) Scatter plot showing relationship between pupil change (dilation to oddball target) and reaction time (all target trials), split by group. (M) Scatter plot showing relationship between ADAS-Cog total score and reaction time (all target trials) in AD group only (b = 0.01, P = 0.02). Significance displayed in I–K are results of linear regression including age and sex as covariates. Plotted line in scatter plots represents best fit through the raw data. Shaded area in line and scatter graphs = standard error of the mean. *P < 0.05; **P < 0.01; ***P < 0.001. AD, Alzheimer’s disease; HC, healthy control; ADAS-Cog, Alzheimer’s Disease Assessment Scale–Cognitive Subscale; CoV, coefficient of variation.