Age-related changes of deep-brain neurophysiological activity

Abstract

Cognitive decline with age is associated with brain atrophy and reduced brain activations, but the underlying neurophysiological mechanisms are unclear, especially in deeper brain structures primarily affected by healthy aging or neurodegenerative processes. Here, we characterize time-resolved, resting-state magnetoencephalography activity of the hippocampus and subcortical brain regions in a large cohort of healthy young (20–30 years) and older (70–80 years) volunteers from the Cam-CAN (Cambridge Centre for Ageing and Neuroscience) open repository. The data show age-related changes in both rhythmic and arrhythmic signal strength in multiple deeper brain regions, including the hippocampus, striatum, and thalamus. We observe a slowing of neural activity across deeper brain regions, with increased delta and reduced gamma activity, which echoes previous reports of cortical slowing. We also report reduced occipito-parietal alpha peak associated with increased theta-band activity in the hippocampus, an effect that may reflect compensatory processes as theta activity, and slope of arrhythmic activity were more strongly expressed when short-term memory performances were preserved. Overall, this study advances the understanding of the biological nature of inter-individual variability in aging. The data provide new insight into how hippocampus and subcortical neurophysiological activity evolve with biological age, and highlight frequency-specific effects associated with cognitive decline versus cognitive maintenance.

Introduction

The course of healthy aging is associated with preserved daily life autonomy, although the efficiency of cognitive functions such as memory and executive control diminishes (Hinault and Lemaire 2020). Age-related cognitive decline is heterogeneous across individuals as brain functions are affected differentially across the population from the same age group (Reuter-Lorenz and Park 2014; Jauny et al. 2022b). Older individuals, relative to young adults, show variable amounts of decline of cognitive performance, brain atrophy and reduced brain activations (Spreng and Turner 2019). The hippocampus and subcortical structures have been extensively studied in cognition research (Olsen et al. 2013; Bourgeois et al. 2020). Their early involvement in age-related neurodegenerative pathologies is also now better understood (Gulyaeva 2019). Yet, age-related changes of rhythmic and arrhythmic neurophysiological activity have been seldom investigated in these brain regions.

Task-free, spontaneous fluctuations of brain activity at rest have long been considered as unwanted background noise, yet recent works highlight their potential as useful indicators of individual cognitive performance(Uddin 2020; Waschke et al. 2021; Wiesman et al. 2021). Some of the rich dynamical features of magnetoencephalography and electroencephalography (M/EEG; Buzsáki 2019; Hinault et al. 2019b) activity change with biological age (Cheng et al. 2015; Courtney and Hinault 2021). Indeed, the aging brain expresses increased slower activity below 4 Hz (delta frequency band), negatively associated with cognitive performance (Kumral et al. 2020; Jauny et al. 2022a). Recent tools have emphasized the distinction between rhythmic (oscillatory periodic activity, that can be specifically defined over limited frequency bands) and arrhythmic (non-oscillatory, aperiodic broadband components often associated with background neurophysiological signal) signal components in electrophysiology (Voytek et al. 2015; Donoghue et al. 2020). The magnitude of arrhythmic activity increases with aging (Merkin et al. 2021), a possible expression of increased neural noise associated with cognitive decline (Thuwal et al. 2021). However, most of these age-related electrophysiological observations so far are from neocortical activity.

Here, we sought to identify specific deep-brain neurophysiological signal features associated with the heterogeneity of cognitive aging. A growing body of work demonstrates MEG activity stemming from deeper structures (Coffey et al. 2016; Müller et al. 2019; Gorina-Careta et al. 2021; Samuelsson et al. 2021). We retrieved MEG data from younger and older individuals from the Cam-CAN (“Cambridge Centre for Ageing and Neuroscience”; Shafto et al. 2014; Taylor et al. 2017) repository to investigate time-resolved resting-state neurophysiological fluctuations in the hippocampus, striatum, and thalamus. Because these structures are critical for short-term and working memory functions and show age-related changes (O’Shea et al. 2016; Valdés Hernández et al. 2020), we expected cognitive changes to be associated with slower, reduced, or more variable neurophysiological activity in older adults. We also hypothesized that preserved cognitive performances in healthy older adults would be associated with increased deep-brain activity.

Methods

Participants’ characteristics

The structural MRI and resting-state MEG data from the Cam-CAN repository (Shafto et al. 2014; Taylor et al. 2017; were available from http://www.mrc-cbu.cam.ac.uk/datasets/camcan/). All participants gave written informed consent for the study. The research protocol has been conducted in compliance with the Helsinki Declaration, and has been approved by the Cambridgeshire Research Ethics Committee (reference: 10/H0308/50). The retrieved sample consisted of 47 young adults (20–30 years, 30 females) and 47 older adults (65–75 years, 30 females; Table 1). Details on the demographic and behavioral data are available online (https://camcan-archive.mrc-cbu.cam.ac.uk/dataaccess/). All older adults scored within normal range at the mini-mental state evaluation (MMSE score > 27; Folstein et al. 1975). No participants reported a history of neurological or cognitive disorders, traumatic brain injury, or major psychiatric disorders.

Table 1.

Demographic and cognitive characteristics [average (standard deviation)] of the study participants retrieved from the Cam-CAN dataset.

Variables	Young adults	Older adults	F	P
N	47	47	—	—
Females/males	30/17	30/17	—	—
Age in years	26 (2.0)	73 (2.7)	—	—
Years of education	16 (2.8)	13 (4.5)	12.45	0.001
Mini-mental state evaluation (MMSE)	29.51 (0.9)	28.43 (1.2)	25.65	<0.001
Visual short-term memory (accuracy)	0.49 (0.1)	0.42 (0.1)	21.19	<0.001
Individual alpha frequency (IAF, in Hz)	10.10 (0.9)	9.21 (0.7)	24.73	<0.001

Behavioral tasks

Because deep-brain regions are essential to short-term memory and working memory performance (McNab and Klingberg 2008), we investigated age-related changes in the visual short-term memory (VSTM) task. In this task, participants were briefly presented 1–4 colored discs on a computer screen and asked to recall the color of the target disc at a cued location. Participants reported their delayed response on a color wheel using a touchscreen input (further detail concerning the task is available online; https://camcan-archive.mrc-cbu.cam.ac.uk/dataaccess/).

Neuroimaging data

The MRI data consisted of T1-weighted image volumes (field of view: 256 × 240 × 192 mm, 1 × 1 × 1 mm voxel size, repetition time: 2,250 ms, echo time: 900 ms, and flip angle: 9°). MEG data consisted of ~9 min of eyes-closed resting-state recordings acquired with a 306-channel Elekta Neuromag Vectorview MEG system (102 magnetometers and 204 planar gradiometers), with 1-kHz sampling rate, and a 0.03–330 Hz online band-pass filter. Registration of MEG and MRI used digitized anatomical landmarks (i.e. fiducial points; nasion and left/right preauricular points; and additional scalp points). The electro-occulogram (EOG) and electrocardiogram (ECG) were recorded to capture eye movements and heartbeats, respectively.

MEG data analysis

The retrieved data were already partly preprocessed using the temporal signal-space separation approach (tSSS): 0.98 correlation, 10-s window; bad channel correction: ON; motion correction: OFF; and 50 Hz + harmonics (mains) notch. We performed further artifact detection and attenuation (on continuous data), band-pass filtering (0.3–100-Hz FIR filter (filter type: Kaiser; filter order: 3,626), 60-dB stopband attenuation, also on continuous data), and source estimation using “Brainstorm” (Tadel et al. 2011), all with default parameters, unless noted below. Remaining physiological artifacts (e.g. eye blinks and saccades) were identified and removed with bespoke signal-space projections (Uusitalo and Ilmoniemi 1997). For each participant, we identified defective sensors (M = 0.02 in young adults and M = 0.17 in older adults) and defined individual alpha peak frequencies (IAF) from power-spectrum density (PSD) estimates (Welch’s method, 1-s sliding window length, with 50% overlap) across the duration of the recordings. We defined IAF as the average frequency of the PSD spectral peaks in the alpha frequency range (7–13 Hz) observed over occipito-parietal gradiometers and magnetometers. No difference in IAF was observed between sensor types (P > 0.08). To account for individual brain atrophy and inform MEG source imaging (Baillet 2017), we used “Freesurfer” (Fischl 2012) to produce segmentations of head tissues, including cortex and subcortical structures, and to compute the intracranial volume from each participant’s MRI.

We modeled MEG brain source activity using elementary volume current dipoles (15,000 elementary source locations across the entire brain without orientation constraint; Baillet et al. 2001). To that end, we used the Desikan/Kiliany atlas (“aparc” + “aseg” segmentation; Desikan et al. 2006). Both gray and white matter structures were imported from Freesurfer’s individual cortical parcellation, and sources were distributed across the entire head volume. We then used the overlapping-sphere approach to MEG forward head modeling (Baillet et al. 2001). The distributed source model was defined across the entire brain volume (Attal and Schwartz 2013; Recasens et al. 2018), and outer gray matter surface was used. Because the orientation of major cell assemblies in deeper brain structures is not as homogenous as forpyramidal cells of the cortex, we have opted to relax the surface location and orientation constraints in our source modeling analyses.

We could not explicitly account for environmental and instrumental noise, as empty-room recordings were not available. We used a noninformative prior as a substitute of the noise statistics, which assumed unit and independent noise variance across sensors. We derived MEG imaging kernels for each individual using the dynamic Statistical Parametric Mapping approach (dSPM; Dale et al. 2000; Hauk et al. 2011) to estimate the source time-series of each region of interest (ROI; left and right hippocampus, thalamus, nucleus accumbens, caudate, and putamen). To determine the specificity of deep-brain effects, we performed control analyses on a subset of cortical regions of the “aparc” atlas (see Supplementary Fig. S1), which were previously reported as showing changes in neurophysiological activity associated with age-related short-term and working memory performance alterations (e.g. Hinault et al. 2020).

We performed time-frequency decompositions of MEG source time-series using the Hilbert transform in frequency bands of interest. The width of each frequency band was based on the surface IAF value of each participant (Toppi et al. 2018): delta (IAF-8/IAF-6), theta (IAF-6/IAF-2), alpha (IAF-2/IAF + 2), beta (IAF + 2/IAF + 14), low-gamma (IAF + 15/IAF + 30), and high-gamma (IAF + 31/IAF + 90). We used the first principal component of the time-series within each subcortical ROI as a summary statistic of distributed neurophysiological activity, which reduces cross-talk between regions (Sato et al. 2018). Signals were normalized for group comparisons through spectrum normalization (i.e. the power of each frequency bin is expressed as a ratio to the total power of the spectrum).

We used “specparam” to perform a parametric decomposition of the power-spectrum of each ROI into rhythmic and arrhythmic components (Donoghue et al. 2020). The model was fitted in fixed mode (default parameter in Brainstorm), without a knee parameter. The resulting arrhythmic components comprise 2 characteristic parameters, the offset and the slope of its 1/f decay (Merkin et al. 2021). More specifically, the slope accounts for the steepness of the decay of arrhythmic signal power with increasing frequencies, from the offset value of signal power at the lowest accessible frequency bin. The offset describes the broadband offset of the spectrum.

We used nonparametric inferential statistics based on permutation tests, false discovery rate (FDR) corrected (N = 10,000; Maris and Oostenveld 2007) and correlation analyses, also FDR-corrected. To limit the number of comparisons, only regions and frequencies showing significant between-group differences in permutation tests (P < 0.05, FDR correction across regions of interest, time, and frequency dimensions) were further considered for possible association with cognitive performance.

Results

Behavioral differences between age groups

The younger and older adults’ groups had similar biological sex ratios and showed normal general cognitive performance (above the MMSE > 27/30 cutoff; see Table 1).

Between-group differences in neurophysiological activity

IAF, recorded over parieto-occipital sensors, was lower in older adults relative to younger adults (Table 1). We found steeper slopes of arrhythmic activity in older adults relative to younger adults, in all ROIs (M_all = 0.78 (S.D._all = 0.25), relative to 0.56(0.15), Ps < 0.001). Delta-band activity was larger (0.16(0.09), relative to 0.08(0.03), Ps < 0.001), with reduced gamma-band activity (0.13 (0.03), relative to 0.15 (0.02), Ps < 0.005), in older adults (Fig. 1A and B), in line with the slowing of the dominant activity reported at the sensor/cortical level. Larger offsets in older relative to younger adults were also observed in bilateral striatum (0.97 (0.051), 0.59 (0.64), and Ps < 0.003, respectively) and hippocampus (1.1 (0.7), relative to 0.7 (0.6), Ps < 0.007; Fig. 2A and B). Importantly, in the right hippocampus only, theta-band activity was stronger in older adults (0.13 (0.02), relative to 0.11 (0.02), P < 0.003). Control analyses replicated previously reported effects of cortical increases of delta-band activity and reduced gamma-band activity at rest across cortical regions. Increased theta-band activity, however, was not significant in cortical regions (see Supplementary Fig. S1), suggesting that the changes observed in the hippocampus are specific to that structure. Additional analyses including individuals’ level of education revealed that partialling out the effect of education did not change the effect of age on neurophysiological activity.

Fig. 1.

Bilaterally in all tested ROIs of older adults relative to younger adults: A) Delta-band activity was stronger; B) Gamma-band activity was reduced. The plots show the average regional activity of left and right homologous structures.

Fig. 2.

A) Parametrization of the group average power spectral density in young and older adults in the right hippocampus region, showing the aperiodic slope and offset across the frequency range. B) Steeper slopes (P < 0.001) and larger offsets (P = 0.007) of the aperiodic spectral component were observed in older adults relatively to young adults, from resting-state activity of the right hippocampus. C) Negative association between the individual alpha peak frequency and power of theta-band neurophysiological activity of the right hippocampus in older adults. D) Positive association between theta slope of neurophysiological activity in the right hippocampus and VSTM task performances of older adults.

With the exception of the correlation between slope and offset arrhythmic parameters observed in both groups (Ps < 0.001), no correlation between neurophysiological activity measures was observed in young adults. In older adults, a lower occipito-parietal IAF was associated with stronger deep-brain delta-band activity (all P < 0.009) in all ROIs (Fig. 2C and D). Importantly, a lower surface IAF was also associated with a stronger right hippocampus theta activity (r = −0.34, P = 0.021).

Associations of deep-brain neurophysiological activity with cognition in older adults

No correlation between neurophysiological activity measures and VTSM or MMSE scores (Ps > 0.3) was observed in young adults. Stronger right hippocampus theta activity was associated with higher VSTM performance in older adults (r = 0.49, P < 0.001). We also found a positive association between the slopes of arrhythmic activity in the right hippocampus and VSTM performances (r = 0.62, P < 0.001; Fig. 2E). Partialling out the effect of did not change this correlation. No correlation involving the MMSE score was observed.

Discussion

We believe this study is first to report aging effects on human deep-brain neurophysiological activity. We used recent methodological advances in source modeling of resting-state M/EEG signals (Samuelsson et al. 2021), and spectral parametrization of rhythmic and arrhythmic brain neurophysiological activity (Donoghue et al. 2020). Overall, older adults showed stronger and more temporally-variable neurophysiological activity in lower frequency bands (Fig. 3). These effects were reversed in higher frequency bands, in line with a slowing of the dominant activity with advance age. Changes were associated with declined short-term memory performance; however, we also found that theta-band signal strength in the right hippocampus were positively associated with VSTM performance.

Fig. 3.

Summary of the reported frequency-specific and aperiodic effects found in older adults relative to young adults, across deep-brain regions of interest.

Our study replicates and extends previous findings of decreased IAF measured from scalp (Scally et al. 2018), and of cortical slowing across the typical frequency bands of electrophysiology with age (Courtney and Hinault 2021; Wiesman et al. 2022). Indeed, in older adults, stronger signal power was observed over the lower frequency bands of the neurophysiological spectrum, with concurrent reduced power over higher frequencies. Our present results point at possible deeper brain origins of such overall slowing of brain activity, and are in line with functional magnetic resonance imaging (fMRI) findings of age-related subcortical changes (Daugherty et al. 2015). Our data show that with aging, subcortical signal is reduced in older adults for the fastest frequency bands. Symmetrically, power was more pronounced in the slower delta frequency band, in association with reduced IAF. These effects have been previously associated with decreased cognitive performances (Rossiter et al. 2014; Jauny et al. 2022a).

Our study disentangles age-related effects on rhythmic versus arrhythmic deeper brain activity. Changes in arrhythmic cortical signal have been reported (Voytek et al. 2015; Merkin et al. 2021; Thuwal et al. 2021), and discussed as reflecting larger amounts of neural noise in brain communications with advancing age. Age-related brain noise is possibly related to alterations of the excitation/inhibition balance in neural circuits (Voytek and Knight 2015; Donoghue et al. 2020), and would impact information processing and the efficiency of cognitive processes (Merkin et al. 2021). At the cortical or the sensor level, both slope and offset have been shown to be reduced in older compared with younger participants (Thuwal et al. 2021). These changes have mainly been observed in frontal regions (Cesnaite et al. 2021). Noisier communications could impact information processing and the efficiency of cognitive processes (Merkin et al. 2021). Here, both slope and offset of subcortical rhythmic activity were larger in older adults relative to younger. Differences in the observed shift in power spectra between cortical and deeper regions in older adults could reflect the differential effect of aging on structures. Indeed, previous work revealed a linear structural atrophy of cortical regions with aging, whereas a curvilinear evolution was observed for deeper regions such as the hippocampus (Walhovd et al. 2005; Fjell et al. 2013).

The hippocampus generates theta activity (e.g. Goutagny et al. 2009), in association with memory processes such as, encoding (Fell et al. 2011), short-term and working memory (Axmacher et al. 2010). Here, we found that the spectral slope of hippocampus activity is positively associated with VSTM performance, suggesting that hippocampal activity is positively associated with short-term memory performance. Closely linked to the hippocampal network (Herweg et al. 2016), theta activity in the thalamus has previously been associated with the maintenance and encoding of task-relevant information (Sweeney-Reed et al. 2016), and could also contribute to preserved performance with advancing age.

Our data show that preserved cognitive performance with advancing age is associated with the strength of hippocampus theta-band rhythmic activity. This aspect is in line with previous fMRI work that showed an association between preserved cognitive performance in older adults and higher hippocampal activity (e.g. Lister and Barnes 2009). These theta-band changes were not observed in control cortical regions, suggesting that these changes (but not changes in other frequency bands) are hippocampal-specific. We anticipate that these results contribute to future investigations of the interindividual variability in cognitive performances in aging (Cabeza et al. 2018; Hinault et al. 2019a). We emphasize though that hyper-activity is not always of a compensatory nature (Hillary and Grafman 2017) and may be indicative of subsequent cognitive decline. Future longitudinal studies will need to clarify these aspects.

We also discuss some limitations to the present study. Courtney and Hinault (2021) and Finn (2021) recently discussed how resting-state activity is less directly associated with cognitive functioning than task-related activity. Moreover, the Cam-CAN cognitive tests did not specifically target executive functions that are associated with frequency-specific brain activity (Hinault et al. 2020, 2021), and a more detailed cognitive investigation could help identify additional associations with cognitive performance. Analyses of task-related activity, involving for example memory or reward-learning tasks (Geddes et al. 2018), could more efficiently identify specific patterns in each region. Regarding methodological aspects, the absence of empty-room recordings may have limited the accuracy of source reconstruction analyses. We used a noninformative prior as a substitute of the noise statistics, which assumed unit and independent noise variance across sensors, in line with previous work investigating hippocampus and subcortical sources (Attal and Schwartz 2013; Gauthier et al. 2020; Piastra et al. 2021). Nevertheless, recent work with simultaneous MEG and intracerebral recording (Pizzo et al. 2019; López-Madrona et al. 2022), showed the reliability of source reconstruction of hippocampus and subcortical regions’ signal. Finally, the investigation of age-related changes in deep-brain activity would benefit from the investigation of longitudinal brain changes, which was not possible with the present dataset.

Recent fMRI work has highlighted the major role of hippocampus and subcortical brain activity in higher-order cognitive functions (Chiu and Egner 2019; Bourgeois et al. 2020). Our results reveal that aging effects on these regions are characterized by changes in both rhythmic and arrhythmic signal strength, with an overall slowing of deeper neural activity. Individual differences were also observed, with specific increased theta-band activity in the right hippocampus associated with preserved short-term memory performances. Relative to healthy aging, the deep-brain regions investigated here are further impaired in age-related pathologies (Gulyaeva 2019), which we anticipate may be associated with further alterations of their neurophysiological activity. The precise investigation of their respective activity could lead to the identification of new markers of the heterogeneity of cognitive aging and cognitive decline.

Funding

Data collection and sharing for this project were partly provided by the Cambridge Centre for Aging and Neuroscience (Cam-CAN). Cam-CAN funding was provided by the UK Biotechnology and Biological Sciences Research Council (grant number BB/H008217/1), together with support from the UK Medical Research Council and University of Cambridge, United Kingdom. S.B. is supported by a NSERC Discovery grant, the Healthy Brains for Healthy Lives initiative of McGill University under the Canada First Research Excellence Fund, and the CIHR Canada Research Chair of Neural Dynamics of Brain Systems.

Conflict of interest statement: The authors declare no competing financial interests.