Adverse and compensatory neurophysiological slowing in Parkinson’s disease

Abstract

Patients with Parkinson’s disease (PD) exhibit multifaceted changes in neurophysiological brain activity, hypothesized to represent a global cortical slowing effect. Using task-free magnetoencephalography and extensive clinical assessments, we found that neurophysiological slowing in PD is differentially associated with motor and non-motor symptoms along a sagittal gradient over the cortical anatomy. In superior parietal regions, neurophysiological slowing reflects an adverse effect and scales with cognitive and motor impairments, while across the inferior frontal cortex, neurophysiological slowing is compatible with a compensatory role. This adverse-to-compensatory gradient is sensitive to individual clinical profiles, such as drug regimens and laterality of symptoms; it is also aligned with the topography of neurotransmitter and transporter systems relevant to PD. We conclude that neurophysiological slowing in patients with PD signals both deleterious and protective mechanisms of the disease, from posterior to anterior regions across the cortex, respectively, with functional and clinical relevance to motor and cognitive symptoms.

Introduction

Parkinson’s disease (PD) is the second most common neurodegenerative disorder worldwide (Feigin et al., 2019). It is characterized by hallmark declines in motor functions (Lang and Lozano, 1998) and debilitating declines in cognitive abilities (Park and Stacy, 2009). The neuropathological process of PD involves the progressive degeneration of dopaminergic neurons and glial cells in the substantia nigra pars compacta, leading to dysfunctional dopamine (DA) signaling along the nigrostriatal pathway (Lang and Lozano, 1998). Changes in cortical signaling are also well-documented in PD, but their functional interpretation rests on apparently conflicting accounts of compensatory versus adverse effects (Boon et al., 2019; Boon et al., 2020; Geraedts et al., 2018; Heinrichs-Graham et al., 2014; Helmich et al., 2012; Hirschmann et al., 2013; Litvak et al., 2011; McCarthy et al., 2011; Moran et al., 2011; Morita et al., 2011; Oswal et al., 2013; Pollok et al., 2013; Simon et al., 2022; Stoffers et al., 2008a; Stoffers et al., 2008b; Wiesman et al., 2023b). Here, we qualify as compensatory any neurophysiological signal changes relative to healthy controls measured in patients with milder clinical outcomes. The presence of compensatory and deleterious effects in the same patient are not necessarily mutually exclusive. Complex patterns of neurophysiological changes may indicate both dysfunction and adaptive compensation processes at play in patients. They may also manifest effects of pharmacotherapies and clinical interventions to remediate PD symptoms. A more nuanced understanding of adverse versus compensatory effects of neurophysiological changes is therefore needed to inform and advance interventions, such as targeted neuromodulation strategies, to ameliorate symptoms and enhance protective functional capabilities in patients with PD (Benninger et al., 2010; Brys et al., 2016; Cantello et al., 2002; Chou et al., 2015; Del Felice et al., 2019; Elahi et al., 2009; Fregni et al., 2006; Pereira et al., 2013; Teo et al., 2017).

Neurophysiological indicators of PD pathophysiology include frequency-specific components of the rich spectrum of brain electrophysiology. Notably, beta-band (15 – 30 Hz) activity is hypersynchronous across the cortico-basal ganglia circuit in patients with PD(Brown, 2003; Cassidy et al., 2002; Hammond et al., 2007; Heinrichs-Graham et al., 2014; Hirschmann et al., 2011; Hutchison et al., 2004), relates to severity of motor dysfunction, and can be normalized by common therapeutics (Giannicola et al., 2010; Kühn et al., 2008; Little and Brown, 2014; Quinn et al., 2015; Weinberger et al., 2006). In the cortex of patients with PD, decades of electrophysiological studies have demonstrated a stereotyped pattern of frequency-defined neural changes relative to healthy adults, including both increased activity in low-frequency bands (e.g., delta [2 – 4 Hz] and theta [5 – 7 Hz]) and concurrent decreased power in high-frequency bands (e.g., alpha [8 – 12 Hz] and beta) (Krösche et al., 2023; Morita et al., 2011; Soikkeli et al., 1991; Stoffers et al., 2007; Vardy et al., 2011). This has led to a hypothesized slowing of brain activity in patients with PD (Morita et al., 2011; Soikkeli et al., 1991; Stoffers et al., 2007; Vardy et al., 2011), but it remains unclear how this multi-frequency neurophysiological effect relates to clinical features of the disease, and whether any such relationships are of a deleterious or compensatory nature. These multi-spectral deviations from healthy levels also comprise rhythmic and/or arrhythmic components (Donoghue et al., 2020a; Donoghue et al., 2020b; Donoghue et al., 2021; Ostlund et al., 2021), for which clinical interpretation and significance for frequency-specific neuromodulation therapies remain to be established in PD (Hirschmann et al., 2022; Kim et al., 2022; Krösche et al., 2023; Oswal et al., 2021; Vinding et al., 2021; Wiesman et al., 2023b; Zhang et al., 2022).

In the present study, we measured neurophysiological slowing (Wiesman et al., 2022b) (Figure 1A) from task-free magnetoencephalography (MEG) data in patients with PD (N = 79) and a matched group of healthy older adults (N = 65). We fitted a linear model to PD-deviations in neurophysiological activity as a function of frequency, to map cortical slowing (i.e., represented by negative linear slopes; Figure 1A, right) in each patient with PD. We related these neurophysiological slowing effects in the patient group to detailed clinical and neuropsychological indicators of individual motor and cognitive deficits. We hypothesized that cortical slowing would be consistently more pronounced in patients with greater clinical impairments. We instead discovered that this association varies in a topographically-structured manner across the cortex, expressing a progressive change from compensation to impairment along the posterior to anterior sagittal direction (Figure 1B). We determined the clinical and neurochemical nature of this sagittal gradient effect by investigating its sensitivity to the clinical features and neurotransmitter receptor densities that are salient in PD. Finally, we provide evidence that cortical slowing also affects frequency-specific functional connectivity between brain regions, which indicates a slowing of neurophysiological signaling across the long-range brain networks of patients with PD.

Figure 1. Sagittal gradient of neurophysiological slowing in Parkinson’s disease.

(A) Neural slowing computation: Source-imaged magnetoencephalography (MEG) data are frequency-transformed and the resulting vertex-wise power spectral densities (PSD) are parameterized into rhythmic (i.e., periodic) and arrhythmic (i.e., aperiodic) features using specparam. The PSDs are then binned over typical frequency bands (i.e., delta: 2–4 Hz; theta: 5–7 Hz; alpha: 8–12 Hz; beta: 15–29 Hz) and each spectrally- and spatially-resolved power estimate of neurophysiological signal per patient is normalized to the mean and standard deviation of the same measures in the healthy control group. For each patient and at each spatial location, a linear model is then fit to these latter spectral deviations across frequencies, with the resulting slope capturing the relative slowing (i.e., negative slope values) of brain activity relative to healthy levels. This procedure is performed at each cortical vertex, resulting in a spatially resolved map of neurophysiological slowing per patient. (B) Spatial gradient analysis: Cortical surfaces are first smoothed to reduce the impact of gyrification on the estimation of spatial gradient effects. Per each vertex location, neurophysiological slowing values are separately correlated with motor (i.e., UPDRS-III scores) and cognitive (i.e., sign-reversed neuropsychological scores averaged over cognitive domains) impairments, beyond the effects of age. These partial correlation maps are then linearly scaled using the Fisher-transform and summed per vertex across participants, resulting in a single summary cortical map of the nature and strength of the relationships between neurophysiological slowing and clinical motor/cognitive impairments. A linear multiple regression is then fit to these data and the beta weights extracted, with each of the cardinal axes (X: left – right; Y: posterior – anterior; Z: inferior – superior) represented as a predictor. The neurophysiological slowing data are then randomly permuted across patients and the partial correlation and spatial multiple regression steps repeated 1,000 times, with the resulting beta weights extracted and used to derive empirical null distributions per each predictor. To test for the effect of binary clinical factors on these gradients, the same procedure is performed within each binarized patient subgroup, with the difference in beta weights between the two subgroups used as the statistic of interest.

Materials and Methods

Participants

The Research Ethics Board at the Montreal Neurological Institute reviewed and approved this study. Written informed consent was obtained from every participant following detailed description of the study, and all research protocols complied with the Declaration of Helsinki. Exclusionary criteria for all participants included current neurological (other than PD) or psychiatric disorder; MEG contraindications; and unusable MEG or demographic data. All participants completed the same MEG protocols with the same instrument at the same site.

Patients with mild to moderate (Hoehn and Yahr scale: 1 – 3) idiopathic PD were enrolled in the Quebec Parkinson Network (QPN; https://rpq-qpn.ca/) (Gan-Or et al., 2020) initiative, which comprises extensive clinical, neuroimaging, neuropsychological, and biological profiling of each participant. At the time of data preprocessing for the current study (March 2021), a total of 90 patients with PD in the QPN cohort had completed MEG data collection. From these participants, 10 were excluded due to major neurological or psychiatric disorder (other than PD; e.g., epilepsy, bipolar disorder, neuropathy) and one due to missing head coil position data, compromising MEG source imaging. A final sample of 79 participants with PD remained and fulfilled the inclusion criteria. All patients with PD were prescribed a stable dosage of antiparkinsonian medication with satisfactory clinical response prior to study enrollment. Patients were instructed to take their medication as prescribed before research visits, and thus all data were collected in the practically-defined “ON” state.

Neuroimaging data from 65 healthy older adults were collated from the QPN (N = 10), PREVENT-AD (N = 40) (Tremblay-Mercier et al., 2021) and Open MEG Archive (OMEGA; N = 15)(Niso et al., 2016) data repositories to serve as a comparison group for the patients with PD. These participants were selected so that their demographic characteristics, including age (Mann-Whitney U test; W = 2349.50, p = .382), self-reported sex (chi-squared test; χ² = 0.65, p = .422), handedness (chi-squared test; χ² = 0.25, p = .883), and highest level of education (Mann-Whitney U test; W =2502.50, p = .444), did not significantly differ from those of the patient group. Group demographic summary statistics and comparisons, as well as clinical summary statistics for the patient group, are provided in Table S1.

Clinical & Neuropsychological Testing

Standard clinical assessments were available for most of the patients with PD, including data regarding gross motor impairment (Unified Parkinson’s Disease Rating Scale – part III [UPDRS-III]; N = 61) (Goetz et al., 2008), general cognitive function (Montreal Cognitive Assessment [MoCA]; N = 70)(Nasreddine et al., 2005), disease staging (Hoehn & Yahr scale; N = 57) (Goetz et al., 2004; Hoehn and Yahr, 1998), symptom onset asymmetry (N = 66), use of dopamine agonists (N = 66), and subjective cognitive complaints (N = 55).

The patients were also asked to complete a series of detailed neuropsychological tests, with a final sample of 69 participants with PD providing useable data. These tests concerned five domains of cognitive function impacted in PD: attention (Digit Span – Forward, Backward, and Sequencing; Trail Making Test Part A), executive function (Trail Making Test Part B; Stroop Test – Colors, Words, and Interference; Brixton Spatial Anticipation Test), memory (Hopkins Verbal Learning Test-Revised [HVLT-R] – Learning Trials 1-3, Immediate and Delayed Recall; Rey Complex Figure Test [RCFT] – Immediate and Delayed Recall), language (Semantic Verbal Fluency – Animals & Actions; Phonemic Verbal Fluency – F, A & S; Boston Naming Test), and visuospatial function (Clock Drawing Test – Verbal Command & Copy Command; RCFT – Copy). To utilize as much available data as possible, missing values were excluded pairwise from analysis per each test. Negatively-scored test values were sign-inverted, the data for each individual test were standardized to the mean and standard deviation of the available sample, and these z-scores were then averaged within each domain listed above to derive domain-specific metrics of cognitive function. To corroborate the statistical independence of these domain composite scores, we computed a ratio of z-scores in the patient group representing the mean of all pairwise relationships (i.e., linearly-scaled Pearson correlation coefficients) amongst intra-domain tests, divided by the mean of all relationships with inter-domain tests. All domains had a ratio of Z_intra/Z_inter > 1.50, and the mean Z_intra/Z_inter ratio over all domains was 2.12 (SD = 0.30). This indicates that these domains were about twice more internally- than externally related on average. The mean across all five domains was also computed for each patient to represent general cognitive function. Importantly, as some participants were missing data on one or more tests within each domain, we verified that none of the domain scores were related to the number of tests used for their computation across individuals (attention: r = .04, p = .734; memory: r = −.19, p = .124; visuospatial function: no missing data; executive function: r = −.19, p = .117; language: r = −.08, p = .539; global function: r = −.14, p = .239; all BF₁₀’s < 0.50). Demographically-corrected neuropsychological data were not available for this study, therefore, demographic factors significantly covarying with cognitive domain scores were included as nuisance covariates in all relevant statistical models. No significant impact of self-reported sex, highest level of education, nor handedness was found on any of the neuropsychological domain scores (all p’s > .20). In contrast, age was a moderate-to-strong predictor of neuropsychological testing performance (memory: r = −.34, p = .004; attention: r = −.22, p = .068; visuospatial function: r = −.54, p < .001; executive function: r = −.43, p < .001; language: r = −.26, p = .030). Accordingly, all statistical analyses utilizing these neuropsychological data included age as a nuisance covariate. Importantly, the patients who reported subjective cognitive complaints did not exhibit significantly worse cognitive performance (mean across all domain scores) than those who did not report such complaints (t(50) = −1.58, p = .120).

Magnetoencephalography Data Collection and Analyses

Eyes-open resting-state MEG data were collected from each participant using a 275-channel whole-head CTF system (Port Coquitlam, British Columbia, Canada) at a sampling rate of 2400 Hz and with an antialiasing filter with a 600 Hz cut-off. Noise-cancellation was applied using CTF’s software-based built-in third-order spatial gradient noise filters. Recordings lasted a minimum of 5 min (Wiesman et al., 2022a) and were conducted with participants in the seated position as they fixated on a centrally-presented crosshair. The participants were monitored during data acquisition via real-time audio-video feeds from inside the MEG shielded room, and continuous head position was recorded during all sessions.

MEG preprocessing was performed with Brainstorm (Tadel et al., 2011) unless otherwise specified, with default parameters and following good-practice guidelines (Gross et al., 2013). The data were bandpass filtered between 1–200 Hz to reduce slow-wave drift and high-frequency noise, and notch filters were applied at the line-in frequency and harmonics (i.e., 60, 120 & 180 Hz). Signal space projectors (SSPs) were derived around cardiac and eye-blink events detected from ECG and EOG channels using the automated procedure available in Brainstorm (Niso et al., 2019), reviewed and manually-corrected where necessary, and applied to the data. Additional SSPs were also used to attenuate stereotyped artifacts on an individual basis. Artifact-reduced MEG data were then epoched into non-overlapping 6-second blocks and downsampled to 600 Hz. Data segments still containing major artifacts (e.g., SQUID jumps) were excluded for each session on the basis of the union of two standardized thresholds of ± 3 median absolute deviations from the median: one for signal amplitude and one for its numerical gradient. An average of 79.72 (SD = 13.82) epochs were used for further analysis (patients: 83.78 [SD = 7.24]; controls: 74.77 [SD = 17.82]) and the percentage of epochs rejected did not differ between the groups (p = .123). Empty-room recordings lasting at least 2 minutes were collected on or near the same day as the data recordings and were processed using the same pipeline, with the exception of the artifact SSPs, to model environmental noise statistics for source analysis.

MEG data were coregistered to each individual’s segmented T1-weighted MRI (Freesurfer recon-all) (Fischl, 2012) using approximately 100 digitized head points. For participants without useable MRI data (N = 14 patients with PD; N = 3 healthy older adults), a quasi-individualized anatomy was created and coregistered with Brainstorm to the MEG data, by warping the default Freesurfer anatomy to the participant’s head digitization points and anatomical landmarks (Tadel et al., 2019). Source imaging was performed per epoch using individually-fitted overlapping-spheres head models (15,000 cortical vertices, with current flows of unconstrained orientation) and dynamic statistical parametric mapping (dSPM) (Dale et al., 2000). Noise covariance estimated from the previously-mentioned empty-room recordings were used for the computation of the dSPM maps.

Analyses of Cortical & Functional Connectivity Slowing

Cortical slowing was assessed per patient with PD using a previously-validated method (Wiesman et al., 2022b), implemented as a linear model of spectral power deviations (in z-scored units) from healthy participants as a function of frequency (in Hz), resulting in units of ‘z/Hz’ (Figure 1A). This model is continuously-scaled, spatially-resolved, and unbiased by the natural differences in neurophysiological signal amplitude observed as a function of frequency. We computed vertex-wise estimates of power spectral density (PSD) from the source-imaged MEG data using Welch’s method (3-s time windows with 50% overlap) and normalized the resulting PSD estimates to the total power of the frequency spectrum. These PSD data were next averaged over all artifact-free 6-second epochs for each participant, and the PSD root-mean-squares across the three unconstrained current flow orientations at each cortical vertex location was projected onto a template cortical surface (FSAverage) for comparison across participants.

To disentangle the slowing effects due to rhythmic versus arrhythmic cortical activity in patients with PD, we parameterized the PSDs with specparam (Brainstorm Matlab version; frequency range = 2–40 Hz; Gaussian peak model; peak width limits = 0.5 –12 Hz; maximum n peaks = 3; minimum peak height = 3 dB; proximity threshold = 2 standard deviations of the largest peak; fixed aperiodic; no guess weight) (Donoghue et al., 2020b) and extracted the exponent of arrhythmic spectral components. The arrhythmic components of the power spectra were the aperiodic outputs of specparam and the rhythmic (i.e., aperiodic-corrected) spectra were derived by subtracting these arrhythmic components from the original PSDs. The PSD components were then averaged over canonical frequency bands (delta: 2–4 Hz; theta: 5–7 Hz; alpha: 8–12 Hz; beta: 15–29 Hz) (Niso et al., 2019). For each PD participant, the resulting PSD maps of spectrally-resolved estimates of neurophysiological signal power were normalized per frequency band to the mean and standard deviation of the comparable maps from the control group, resulting in cortical maps of PD spectral deviations from healthy variants. We then fit a linear model across the four frequency bands per each participant and cortical vertex location using the polyfit function in Matlab and extracted the model slope. This procedure yielded cortical maps of linear trends in spectral neurophysiological deviations per patient with PD. In these maps, cortical locations with flat slopes indicate locations of no substantial spectral change with respect to expected healthy variants, while more negative slopes indicate locations of stronger cortical slowing effects. To ensure that the choice of frequency-band limits did not bias our results, we repeated this procedure for the non-parameterized and rhythmic slowing models using all linearly-spaced frequency bins of the PSD estimates within the relevant frequency range (2 – 29 Hz * 1/3 Hz resolution = 82 samples).

We used a leave-one-out model comparison approach to examine the relative importance of each frequency-band on each relationship between rhythmic slowing and clinical features. For this, we recomputed the rhythmic slowing measure four times, each time leaving one frequency out of the model. These “one-frequency-missing” models were then regressed separately on the relevant clinical feature before we derived their respective Akaike information criterion (AIC) scores. The AIC of the original “full-frequency” model was then subtracted from each “one-frequency-missing” model, resulting in a ΔAIC metric, where higher values represent a greater contribution of the missing frequency to the overall effect. Meaningful model information was assessed using a standard threshold of ΔAIC > 2.

We also used the MEG source maps to investigate the potential for slowing of inter-regional functional connectivity in patients with PD. We extracted the first principal component from the three elementary source time series at each vertex location in each participant’s native space, and derived whole-cortex functional connectivity maps, using the cortical location with the strongest overall slowing effect (back-transformed into each participant’s native space) as the seed. We used orthogonalized amplitude envelope correlations (AEC) (Bruns et al., 2000; Colclough et al., 2015) as the connectivity measure, based on the same frequency-band definitions used for the previously-described cortical slowing derivations. We estimated connectivity over each epoch and averaged the resulting AEC estimates across epochs, yielding a single AEC map per participant and frequency band. We projected these individual AEC maps onto the same cortical surface template (FSAverage) for group analyses, and used the previously-described procedure to derive spatially-resolved maps of functional connectivity slowing per patient with PD (see Supplementary Results and Figure S10).

Testing of Cortical Clinical-Gradient Effects

To ensure that gyrification of the pial surface did not bias our estimation of absolute distance between neighboring cortical locations, we applied a smoothing kernel to the template surface coordinate matrix using the tess_smooth function in Brainstorm (100% smoothing; i.e., smoothing factor of 1 with 46 iterations). We then used a two-step procedure to test for spatial gradients in the relationships between clinical impairments and neurophysiological slowing along the cortical surface (Figure 1B). We first modeled linear relationships at each cortical location between neurophysiological slowing and both motor (i.e., UPDRS-III) and cognitive (i.e., mean cognitive domain scores) impairments, beyond the effects of age, using the partialcorr function in Matlab. The resulting Pearson correlation coefficient values were then normalized using the Fisher transform (i.e., the inverse hyperbolic tangent; using the atanh function in Matlab), the neuropsychology correlations were sign-reversed for comparability with those computed from the UPDRS-III scores, and the two were summed at each location to generate cortical maps of the association between neurophysiological slowing and clinical impairments. In the second step, we fit a multiple regression model to these data using the regress function in Matlab, with the summed Fisher-transformed correlation coefficients as the dependent variable and the three cardinal axes of the template brain space (i.e., X: left – right, Y: posterior – anterior, and Z: inferior – superior) as the independent predictors. The unstandardized beta weights for each predictor were extracted from this model, representing the absolute change in the slowing–clinical impairment relationship (i.e., sum(atanh[r])) per unit distance (i.e., meters) across the cortex. Where relevant, we also performed post-hoc testing separately for the motor (i.e., UPDRS-III) and cognitive (i.e., mean cognitive domain scores) impairment data using the same procedure. A non-parametric permutation approach that includes the high spatial autocorrelation of the neurophysiological slowing maps was used to derive a null distribution and determine the statistical significance of these effects (see Methods: Statistical Analyses).

Contributions of Periodic Features to Rhythmic Slowing Relationships

To determine which features of periodic neural activity contributed to rhythmic slowing effects, we extracted the periodic outputs of specparam (i.e., center frequency, bandwidth, and amplitude) for the most prominent peak found by the algorithm within the 2 – 29 Hz frequency range, at each cortical location and for each participant. When no such peak was found at a given cortical location and for a given participant, this participant was excluded from statistical analyses pairwise. For the main effect of slowing, these periodic values were extracted from the peak location (i.e., vertex) of the effect and included in a linear regression model alongside rhythmic slowing data extracted from the same location, with the form:

$${\mathit{rhythmic\; slowing}}^{\mathbf{~}}{\mathit{\beta }}_0+{\mathit{\beta }}_{\mathit{peak\; center\; frequency}}+{\mathit{\beta }}_{\mathit{peak\; bandwidth}}+{\mathit{\beta }}_{\mathit{peak\; amplitude}}+\mathit{\epsilon }$$

For the relationship between rhythmic slowing and cognitive function, each periodic feature was extracted from the peak location of the effect, and used to predict cognitive function in similar models to the original:

$${\mathit{periodic\; feature}}^{~}{\mathit{\beta }}_0+{\mathit{\beta }}_{\mathit{attention}}+{\mathit{\beta }}_{\mathit{memory}}+{\mathit{\beta }}_{\mathit{executive\; function}}+{\mathit{\beta }}_{\mathit{visuospatial\; function}}+{\mathit{\beta }}_{\mathit{language}}+{\mathit{\beta }}_{age}+\mathit{\epsilon }$$

Co-localization with Normative Atlases of Neurotransmitter Receptor Density and Functional Gradients

To determine the neurochemical systems that contribute to the observed cortical clinical-gradient effects, we adapted the two-step procedure described above, substituting as predictors region-wise estimates of normative neurotransmitter receptor/transporter density (Hansen et al., 2022; Markello et al., 2022b) for the cardinal spatial axis data. Mean cortical receptor distribution maps of 16 different receptors and transporters from 6 neurotransmitter systems were computed as in previous work (Hansen et al., 2022) and parcellated using the Desikan-Killiany atlas (Desikan et al., 2006). These included dopamine (D1: 13 adults, [11C]SCH23390 PET; D2: 92, [11C]FLB-457, DAT: 174, [123I]-FP-CIT), serotonin (5-HT1a: 36, [11C]WAY-100635; 5-HT1b: 88, [11C]P943; 5-HT2a: 29, [11C]Cimbi-36; 5-HT4: 59, [11C]SB207145; 5-HT6: 30, [11C]GSK215083; 5-HTT: 100, [11C]DASB), acetylcholine (α4β2: 30, [18F]flubatine; M1: 24, [11C]LSN3172176; VAChT: 30, [18F]FEOBV), GABA (GABAa: 16, [11C]flumazenil), glutamate (NMDA: 29, [18F]GE-179; mGluR5: 123, [11C]ABP688), and norepinephrine (NET: 77, [11C]MRB). In addition, to test the importance of total synapse density we extracted a similar map of synaptic vesicle glycoprotein 2A (76, [11C]UCB-J) (Hansen et al., 2022). To examine spatial congruence with a sensory-association axis of human brain function, we extracted the principal functional gradient reported by Margulies et al.(Margulies et al., 2016). To facilitate comparison of the clinical-gradient effects to these normative maps, we parcellated the source-imaged MEG PSDs using the mean within each region of the same atlas(Desikan et al., 2006), and recomputed the neurophysiological slowing metric per atlas region. These slowing values were then related to cognitive and motor function (controlling for effects of age), normalized (and, for the cognitive relationships, sign-reversed), and summed using the same procedure described for the clinical-gradient analysis (step 1). To enable comparisons across neurotransmitter systems, the density and neurophysiological slowing data were each standardized (i.e., z-scored) across cortical regions. Linear regressions were then used to derive standardized beta-weights representing the co-localization of the cortical clinical-gradient effect with each normative neurotransmitter map (step 2).

Symptom Profile Clustering

To examine potential effects of symptom variability across participants, we used kmeans (Matlab 2020b) to cluster participants based on shared patterns of variance across their UPDRS-III subscores (data available for N = 56). Sub-score data were z-scored within each participant (to control for absolute symptom severity) prior to clustering, and k was determined by selecting the number of clusters that produced the highest mean silhouette value (k = 2; mean silhouette = 0.47). Inspection of the cluster centroids revealed that the largest contributors to cluster distance were resting tremor (k₁ = 0.70, k₂ = 1.87) and rigidity (k₁ = 2.37, k₂ = −0.14), which aligns with extant literature (Zaidel et al., 2009). The binary cluster assignments generated with this approach were added post hoc to all major statistical contrasts (i.e., neurophysiological slowing effects, relationships between neurophysiological slowing and clinical features, and the primary clinical-spatial gradient effect) as a moderating variable. No significant moderating effect of symptom profile was found on any of these relationships.

Testing for Potential Confounds

In addition to our stringent data cleaning procedures (see Magnetoencephalography Data Collection and Analyses), we investigated possible confound effects due to the level of participant head motion, eye movements, and heart-rate variability. We extracted the root-sum-squared of the head-position indicator, EOG, and ECG channel time series, and averaged these values over channels and recording blocks within each data type to derive single measures of head motion, eye movement, and heart rate variability, respectively, per participant. Alongside age, disease duration (in years since diagnosis), and the number of epochs used for analysis per participant, these derivations were included as nuisance covariates in post hoc statistical models to examine the robustness of the initial effect(s) of interest against potential confounds. Note that disease duration was included among the nuisance covariates to separate the differential effects of disease severity from time spent living with the disease. Additionally, neither head motion (p = .627), eye movements (p = .152), heart rate variability (p = .283), nor percent of epochs rejected (p = .123) significantly differed between patients and controls.

We also tested whether the participant’s head size and distance from the MEG sensor array may have biased our findings. Per each participant, the mean Euclidean distance between all cortical locations and a posterior reference sensor (MZO03) was used to represent head position, and the mean Euclidean distance between the three fiducial points (i.e., the left and right preauricular points and the nasion) was used as a proxy of head size. There was a small but statistically significant difference in head position between patients and healthy controls (Welch’s test; t(141.86) = 2.96, p = .004), such that patients were on average 4.8 mm farther away from the back of the sensor array. Importantly, however, the participants’ head position did not significantly covary with the strength of neurophysiological slowing at the location of the strongest effect (r = −.02, p = .84). There was no significant difference in head size between patients and healthy controls (Welch’s test; t(132.80) = 0.86, p = .393) or relationship between head size and the strength of neurophysiological slowing at the location of the strongest effect (r = .08, p = .46). Therefore, we conclude that the primary findings regarding the neurophysiological slowing effect (i.e., the one-sample tests of this effect against zero and the linear relationships between it and clinical features) are unlikely to be biased by head position or head size.

Statistical Analyses

Participants with missing data were excluded pairwise per model. A threshold of p < .05 was used to indicate statistical significance, and all tests were performed two-tailed unless otherwise specified.

We derived statistical comparisons across the cortical maps produced, covarying out the effect of age, using SPM12. We tested the main effects of neurophysiological slowing with one-sample tests against a null hypothesis of zero (controlling for age). We formulated the relationships between slowing and cognitive scores as multiple regressions involving all five neuropsychological domain scores per neurophysiological slowing model:

$${\mathit{slowing}}^{~}{\mathit{\beta }}_0+{\mathit{\beta }}_{attention}+{\mathit{\beta }}_{memory}+{\mathit{\beta }}_{\mathit{executive\; function}}+{\mathit{\beta }}_{\mathit{visuospatial\; function}}+{\mathit{\beta }}_{language}+{\mathit{\beta }}_{age}+\mathit{\epsilon }$$

Initial tests used parametric general linear models, with secondary corrections of the resulting F-contrasts for multiple comparisons across cortical locations with Threshold-Free Cluster Enhancement (TFCE; E = 1.0, H = 2.0; 5,000 permutations) (Smith and Nichols, 2009). We applied a final cluster-wise threshold of p_FWE < .05 to determine statistical significance, and used the TFCE clusters at this threshold to mask the original statistical values (i.e., vertex-wise F values) for visualization. We performed a secondary correction of these cluster p_FWE values across related tests (i.e., cognitive domain scores) using the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995). The corrected relationships are reported in Figures 3-4. For completeness, we also show the results from the first-level TFCE correction in Supplementary Materials. In each cluster, we extracted data from the cortical location exhibiting the strongest statistical relationship (i.e., the “peak vertex”) for subsequent analysis (e.g., for testing of potential confounds) and visualization. Where appropriate, linear models were fit to these extracted data using the lm function in R (Team, 2017).

A non-parametric permutation approach was used to determine the statistical significance of the spatial gradient and neurotransmitter co-localization effects, wherein the pairings of the patient cortical slowing maps with the clinical data were randomly permuted (using the randperm function in Matlab) and then used to compute the partial correlations (step 1) and regressions (step 2). This process was repeated 1,000 times, and the resulting beta-weights were extracted to generate null distributions for each predictor. The original beta coefficients were then compared with their respective null distributions to generate non-parametric p-values. To test for significant subgroup effects on the spatial gradients based on binary clinical factors (i.e., subjective cognitive complaints, symptom onset laterality, and dopamine agonist use), we implemented the same two-step approach, using instead the difference in beta-weights (from step 2) between clinical groups as the statistic of interest in independent models. Importantly, as the associations between neighboring vertices of the cortical slowing maps were not permuted, this approach retains the spatial autocorrelation of these maps in the null model. We also consider this approach superior to autocorrelation-preserving spin-tests for our purposes, as it generates null distributions via permutations of participant-level data, rather than from group-level effect estimates.

Data & Code Availability

Data used in the preparation of this work are available through the Clinical Biospecimen Imaging and Genetic (C-BIG) repository (https://www.mcgill.ca/neuro/open-science/c-big-repository) (Gan-Or et al., 2020), the PREVENT-AD open resource (https://openpreventad.loris.ca/) (Tremblay-Mercier et al., 2021), and the OMEGA repository (https://www.mcgill.ca/bic/resources/omega) (Niso et al., 2016). Normative neurotransmitter density data are available from neuromaps (https://github.com/netneurolab/neuromaps) (Markello et al., 2022a). Code for MEG preprocessing and the neurophysiological slowing and spatial gradient analyses is available at https://github.com/aiwiesman/QPN_Slowing. Rejection of epochs containing artifacts was performed with the ArtifactScanTool (https://github.com/nichrishayes/ArtifactScanTool).

Results

Slowing of Rhythmic and Arrhythmic Neurophysiological Activity Relates Differentially to Clinical Impairments

The patients with Parkinson’s disease recruited for the present study exhibited slowing of cortical neurophysiological activity, as indicated by increased power compared to healthy controls at low frequencies, an inversion of this effect at ^~11 Hz, and decreased power in higher frequencies (Figure 2A). The strongest slowing effects were observed in bilateral parieto-occipital cortices (TFCE; p_FWE < .001; peak vertex = x: 50, y: −77, z: 1). The magnitude of this slowing effect was related to cognitive abilities (Figure 3), including to language abilities in bilateral prefrontal and temporal regions (TFCE; p_FWE = .012; peak vertex = x: 4, y: 38, z: 11), attention in bilateral inferior frontal and somatomotor cortices (TFCE; p_FWE = .030; peak vertex = x: 50, y: −13, z: 52), and visuospatial function in bilateral anterior temporal regions (TFCE; p_FWE = .014; peak vertex = x: −49, y: −8, z: −44).

Figure 2. Neurophysiological slowing in Parkinson’s disease.

(A) Cortical maps indicate clusters of neurophysiological slowing (computed on the non-parameterized data) in patients with Parkinson’s disease (PD) after spatial multiple comparisons correction with Threshold-Free Cluster Enhancement (p_FWE < .05). Power spectra to the bottom left illustrate the underlying data from the cortical vertex exhibiting the strongest effect of neurophysiological slowing. The horizontal colored lines show the frequency bandwidths used for binning the power spectra. The plot to the bottom right shows the individual patient spectral deviations at this same cortical vertex for each frequency band, with the light grey lines-of-best fit indicating individual neurophysiological slowing slopes, and the overlaid black line and blue shaded area representing the overall group effect and 95% confidence intervals, respectively. These individual and mean neurophysiological slowing effects are also represented as single dots in the scatterplot to the top right. (B) Plots similar to those in (A), but with neurophysiological slowing computed using the rhythmic (i.e., periodic) component of the parameterized spectra. (C) Plots similar to those in (A-B), but with neurophysiological slowing computed using the arrhythmic (i.e., aperiodic) component of the parameterized spectra.

Figure 3. Neurophysiological slowing associated with cognitive impairments in Parkinson’s disease.

Cortical maps showing where neurophysiological slowing was associated with cognitive function in patients with PD, after first-level correction with Threshold-Free Cluster Enhancement (p_FWE < .05)and second-level FDR correction across related tests (p_FDR < .05). The scatterplots of the residuals illustrate the nature and strength of this relationship at the cortical vertex exhibiting the strongest effect, with lines-of-best-fit, 95% confidence intervals, and R² values overlaid. The scatterplots are meant to emphasize the nature of the effect and allow visual inspection for outliers, not to assess the magnitude of the effects reported.

Both arrhythmic (TFCE; p_FWE < .001; peak vertex = x: 45, y: −81, z: 7; Figure 2B) and rhythmic (TFCE; p_FWE < .001; peak vertex = x: 43, y: −71, z: 31; Figure 2C) neurophysiological generators contributed to the slowing effect. Arrhythmic slowing was stronger than rhythmic slowing in bilateral inferior frontal regions (TFCE; p_FWE = .018; peak vertex = x: 9, y: 40, z: −5; Figure S1), and no clusters were identified where rhythmic slowing was significantly stronger than arrhythmic. Arrhythmic cortical slowing was associated with motor impairments in bilateral prefrontal and temporal cortices (i.e., UPDRS-III scores; TFCE; p_FWE = .028; peak vertex = x: −41, y: 43, z: 19; Figure 4). Arrhythmic slowing was also associated with domain-specific abilities in language in distributed frontal, temporal, and occipital areas (TFCE; p_FWE = .013; peak vertex = x: 49, y: −4, z: −9), attention in right superior parietal cortex (TFCE; p_FWE = .043; peak vertex = x: 23, y: −54, z: 68), and executive function in right fusiform/lingual cortex (TFCE; p_FWE = .047; peak vertex = x: 14, y: −54, z: −7), but these relationships to cognition (Figure S2) were not significant following secondary FDR correction.

Figure 4. Arrhythmic neurophysiological slowing associated with motor impairments in Parkinson’s disease.

Cortical maps of neurophysiological slowing associated with motor function in patients with PD, obtained from first-level correction with Threshold-Free Cluster Enhancement (p_FWE < .05). The scatterplot of residuals illustrates the nature and strength of this relationship at the cortical vertex exhibiting the strongest effect, with the line-of-best-fit, 95% confidence interval, and R² value overlaid. The scatterplots are meant to emphasize the nature of the effect and allow visual inspection for outliers, not to assess the magnitude of the effects reported.

Rhythmic neurophysiological slowing covaried only with attention abilities in right inferior frontal cortex (TFCE; p_FWE = .039; peak vertex = x: 50, y: 32, z: −13; Figure S3), but again this relationship was not significant following secondary FDR correction.

Regional neurophysiological slowing across the Yeo 7-networks atlas(Yeo et al., 2011) indicated pronounced slowing in visual and dorsal attention networks, and weaker slowing in somato-motor, ventral attention, and fronto-parietal networks (Figure S1). All the reported slowing effects and relationships to clinical metrics included age as a nuisance covariate, and remained significant (all p’s < .005) after inclusion of confounds in the respective linear models, such as head motion, eye movements, heart rate variability, disease duration, and the number of epochs used per participant for analysis. All relationships involving rhythmic activity (i.e., the non-parameterized and rhythmic slowing models) replicated when using neurophysiological slowing values derived from all linearly-spaced PSD estimates in the 2 – 29 Hz range (all p’s < .003; Figure S4A), indicating that the use of typically defined frequency bands in the slowing model was not a source of bias. Further, discrete features of the peaks modeled by specparam accounted for the main effect of rhythmic slowing (full model: R² = .50, p < .001; frequency: β = .03, p < .001; amplitude: β = .09, p = .024; bandwidth: β = .01, p = .013), indicating that these effects were not an artifactual outcome of the modeling approach.

Associations Between Neurophysiological Slowing and Clinical Impairments Exhibit a Spatial Gradient Across the Cortex

We observed that the nature of the relationships between cortical slowing and clinical impairments changed systematically across the sagittal plane of the cortex, with more posterior relationships generally indicating impairment (i.e., greater slowing associated with worse cognitive outcomes) and more anterior relationships indicating compensation. To test this effect empirically, we developed and implemented a new non-parametric method that preserves the spatial autocorrelation of the neurophysiological slowing maps (see Methods: Testing of Cortical Clinical-Gradient Effects and Figure 1B) and found evidence of significant posterior–anterior (1,000 permutations; b = 3.57, p = .002) and superior–inferior (1,000 permutations; b = −5.64, p = .004) spatial gradients, such that stronger slowing in superior parietal cortices related to worse clinical impairments, while greater slowing in inferior frontal regions was associated with relatively preserved motor and cognitive functions (Figure 5A). These regional slowing effects were not significantly related within individuals: participants with compensatory slowing of neurophysiology in inferior frontal cortex did not necessarily exhibit an adverse slowing effect in superior parietal cortex (r = −.05, p = .636).

Figure 5. Anatomical gradient of clinical effects of neurophysiological slowing in Parkinson’s disease.

(A) Cortical maps of the nature and strength of the relationship between neurophysiological slowing and clinical impairments (i.e., partial correlations linearly-scaled and summed across motor and cognitive domains) along the cortex of patients with Parkinson’s disease, with lower values indicating a more pathological relationship (i.e., greater slowing predicting worse clinical deficits) and higher values indicating a possible compensatory effect. Grey vectors plotted along the cardinal anatomical axes show the unstandardized beta weights from a multiple regression of the neurophysiological slowing – clinical impairment relationships on the relevant anatomical coordinates (X: left – right; Y: posterior – anterior; Z: inferior – superior), and indicate the magnitude and direction of the significant anatomical gradient effects. Overlaid p-values were obtained from non-parametric permutations and indicate statistical significance per each axis of the gradient effect. The blue vector indicates the magnitude and direction of the overall significant anatomical gradient effect. (B) Cortical maps of the nature and strength of the neurophysiological slowing – clinical impairment relationships across the cortex of patients with Parkinson’s disease. Here neurophysiological slowing was assessed from the rhythmic (left) and arrhythmic (right) components of the parameterized spectra. The anatomical gradient effects observed in the non-parameterized neurophysiological slowing data (panel A) did not differ qualitatively between the rhythmic and arrhythmic models.

This spatial gradient effect did not differ between the rhythmic and arrhythmic slowing models (1,000 permutations; posterior – anterior: p = .848; superior – inferior: p > .999; Figure 5B), included age as a nuisance covariate, and remained significant after correction for confounds (i.e., head motion, eye movements, heart rate variability, and disease duration; 1,000 permutations; posterior–anterior: p < .001; superior–inferior: p = .014; Figure S5). Both gradient effects were also replicated when using neurophysiological slowing values derived from all linearly-spaced PSD estimates in the 2–29 Hz range (1,000 permutations; posterior–anterior: p = .022; superior–inferior: p < .001; Figure S4B).

The adverse-to-compensatory gradient was also significantly modulated by meaningful clinical factors of PD (Figure 6). Patients who reported subjective cognitive complaints exhibited a weaker posterior–anterior gradient effect (1,000 permutations; Δb = −7.90, p = .010; Figure 6A). We also observed a similar effect of dopamine agonist use, with a weaker posterior–anterior gradient effect in patients taking a regimen of dopamine agonists (1,000 permutations; Δb = −5.56, p = .048; Figure 6B). Further, we discovered that the left–right asymmetry of the sagittal gradient differed significantly based on the laterality of symptom onset (1,000 permutations; Δb = −3.35, p = .006; Figure 6C), such that left-onset patients exhibited a bias toward compensatory effects of cortical slowing in the left hemisphere, while we observed the mirrored effect in right-onset patients. The dopamine agonist (1,000 permutations; p = .042; Figure S6) and symptom laterality (1,000 permutations; p = .032; Figure S7) effects were specific to the model considering only motor impairments, while the effect of subjective cognitive complaints was specific to cognitive abilities (1,000 permutations; p = .006; Figure S8; post-hoc testing of significant clinical subgroup differences in gradient effects).

Figure 6. The anatomical gradients of clinical effects of neurophysiological slowing in Parkinson’s disease are clinically meaningful.

Cortical maps of differences in the nature and strength of relationships between neurophysiological slowing and clinical impairments in patients with Parkinson’s disease, as a function of binary clinical features, including (A) subjective cognitive complaints, (B) drug regimen including dopamine agonists, and (C) laterality of initial symptom onset. Purple vectors plotted along the cardinal spatial axes show the differences in unstandardized multiple regression beta weights between the two clinical feature subgroups of the neurophysiological slowing – clinical impairment relationships on the relevant spatial coordinates (X: left – right; Y: posterior – anterior; Z: inferior – superior). Overlaid p-values were derived from non-parametric permutations and indicate statistical significance per each axis of the difference in the gradient effect. The blue and red vectors indicate the magnitude and direction of the overall anatomical gradient effects per each clinical feature subgroup.

Clinical Effects of Cortical Slowing Selectively Co-Localize with Cortical Neurotransmitter System Densities and Sensory-Association Functional Organization

To test for topographical similarities between the observed adverse-to-compensatory gradient and normative distributions of neurochemical systems across the cortex, we adapted the non-parametric method described above to neuromaps, a collection of normative atlases of cortical neurotransmitter systems (Markello et al., 2022a). We found that the relationship between neurophysiological slowing and clinical impairments co-localized selectively with normative densities of dopamine, serotonin, GABA, and norepinephrine systems, but not with the acetylcholine and glutamate systems, nor with overall synaptic density (Figures 7 and S9). Specifically, all three measures of dopaminergic density related positively to the topography of the sagittal gradient effect (1,000 permutations; D1: β = 0.38, p_FDR < .001; D2: β = 0.40, p_FDR = .024; DAT: β = 0.24, p_FDR = .024), such that regions with higher dopamine receptor/transporter density in health exhibited a compensatory effect of slowing in patients with PD. We found similar positive associations for four of the six tested serotonergic density measures (1,000 permutations; 5-HT1a: β = 0.49, p_FDR = .023; 5-HT2a: β = 0.30, p_FDR = .017; 5-HT4: β = 0.43, p_FDR = .024; 5-HTT: β = 0.35, p_FDR = .024). In contrast, both GABAergic (1,000 permutations; GABAa: β = −0.34, p_FDR = .038) and noradrenergic (1,000 permutations; NET: β = −0.53, p_FDR = .038) densities related negatively to the gradient effect, such that regions with higher healthy receptor/transporter density exhibited a stronger pathological effect of cortical slowing in PD. The sagittal gradient effect also co-localized with a sensory-association axis of human brain function derived from BOLD fMRI (Margulies et al., 2016) (1,000 permutations; β = 0.04, p < .001).

Figure 7. The clinical relationships of neurophysiological slowing are topographically aligned with selective neurotransmitter system densities and sensory-association organization.

(A) The vector heatmap indicates the strength (standardized β) and statistical significance(*P_FDR < .05, **PFDR < .005) of the topographical alignment between neurophysiological slowing relationships to clinical outcomes and neuromap values, and emphasizes the concordance with the dopamine (D1, D2, and DAT), serotonin (5-HT1a, 5-HT1b, 5-HT2a, 5-HT4, 5-HT6, 5-HTT), acetylcholine (α4β2, M1, VAChT), GABA (GABAa), glutamate (NMDA, mGluR5), norepinephrine (NET) cortical systems, and synapse density (glycoprotein). (B) The parcellated cortical maps show the region-wise relationship between neurophysiological slowing and clinical impairments in patients with Parkinson’s disease (i.e., partial correlations linearly-scaled and summed across motor and cognitive domains, z-scored across brain regions; top) and the sensory-association axis reported by Margulies and colleagues⁵⁵ (bottom). The scatterplot on the right indicates the nature and strength of colocalization between these maps and the regions exhibiting the five highest and five lowest residuals labeled (p < .001; 1,000 permutations; the line indicates the best linear fit).

Discussion

A vast body of research has demonstrated that brain activity shifts from faster to slower dynamics in patients with neurodegenerative disorders (Berendse et al., 2000; Boon et al., 2019; Bosboom et al., 2006; Dauwels et al., 2011; de Haan et al., 2008; Engels et al., 2017; Fernández et al., 2002; Geraedts et al., 2018; Huang et al., 2000; Morita et al., 2011; Osipova et al., 2005; Penttilä et al., 1985; Schreiter-Gasser et al., 1993; Soikkeli et al., 1991; Stoffers et al., 2007; Vardy et al., 2011; Wiesman et al., 2023b; Wiesman et al., 2022b). The cortical mapping and relevance of these slowing manifestations have been related to meaningful clinical features only recently (Krösche et al., 2023; Wiesman et al., 2023b; Wiesman et al., 2022b). Here, we provide a systematic approach to measure rhythmic and arrhythmic manifestations of neurophysiological slowing in patients with Parkinson’s disease. We find that patients with PD do exhibit broad neurophysiological slowing effects across posterior parietal, temporal, occipital, and inferior frontal cortices. This effect concerns both the rhythmic and arrhythmic components of the neurophysiological spectrum. Further, we show that the magnitude of this slowing effect is associated with individual cognitive and motor functions. Most notably, slowing across bilateral fronto-temporal cortical regions is related to better language abilities, while slowing in an ensemble of right-lateralized inferior frontal, somato-motor, and superior parietal regions is associated with worse attention scores.

Rhythmic and arrhythmic slowing are differentially related to these observations, although several of the effects were not significant according to a stringent secondary FDR correction. The relationship with language abilities was only recapitulated with arrhythmic slowing, while the relationship with attention was found in both the rhythmic and arrhythmic analyses, but with differing anatomical definitions. Arrhythmic slowing related to worse attention in right superior parietal regions, while rhythmic slowing exhibited the same association in right inferior frontal cortex. We also found relationships involving arrhythmic slowing that were not detected in the cortical slowing measures computed using the non-parameterized spectra, including a robust association between arrhythmic cortical slowing and better motor function (i.e., lower UPDRS-III scores) in bilateral prefrontal and anterior temporal cortices. Together, these results highlight not only the potential clinical relevance of cortical slowing in patients with PD, but also the insight gained by analyzing the respective effects of rhythmic and arrhythmic spectral features on neurophysiological slowing across patient populations.

The linear relationships with clinical features did not always co-localize with the main effects of neurophysiological slowing. This may indicate that regions outside of the main slowing effect exhibit more clinically-meaningful variability at this stage of the Parkinson’s disease continuum. Indeed, seven of the nine regional peaks exhibiting a relationship to clinical features also presented with a main slowing effect when clinical variability was controlled (i.e., covaried from the model). However, caution is warranted when interpreting how these effects overlap. Cluster-based corrections are recognized to be relatively conservative (Huang and Zhang, 2017; Pernet et al., 2015) and do not indicate the spatial limits of effects (Sassenhagen and Draschkow, 2019). It is thus recommended to report on the local peaks of such clusters, rather than their spatial extent. Additionally, and despite this interpretational limitation, the higher-order sagittal gradient effect described below incorporates clinical features and neurophysiological slowing into a holistic model, and thus supersedes relationships to individual clinical features as our primary finding of interest.

A salient finding of our study is that the nature of the association of neurophysiological activity with clinical outcomes varies across the cardinal sagittal axis of the brain. This novel finding of an adverse-to-compensatory gradient highlights apparently contradictory clinical manifestations of neurophysiological slowing: in superior and posterior cortices, slowing relates to worse clinical conditions (i.e., higher UPDRS-III and lower neuropsychological scores); in inferior and anterior regions, slowing is associated with better motor and cognitive abilities. We also note that the nature of these clinical-slowing relationships aligns along the sensory-association axis of task-free brain activity (Margulies et al., 2016), such that slowing of cortical activity in sensory regions relates to worse cognitive and motor impairments, while slowing in association cortices relates to better abilities. In Parkinson’s disease, the patterns of slower neurophysiological activity in higher-order brain regions may therefore have a compensatory effect, enabling a shift towards coarser, but more stable, temporal segmentation of information processing. In contrast, cortical slowing in primary sensory regions would indicate a genuine loss of function. We also found that the strength of this gradient topography is affected by key clinical factors: it is reduced by a dopamine agonist regimen, and stronger in patients with no subjective cognitive complaints. Further, although we did not observe a left–right anatomical gradient in patients with PD, we found a marked difference in the gradient effect along this axis when patients were sorted according to the laterality of their initial symptom onset: cortical slowing is related to a compensation effect on clinical symptoms when expressed in the less-affected hemisphere.

We identify four neurochemical systems as candidate contributors to the new adverse-to-compensatory gradient effect. Brain regions with higher normative dopamine and serotonin and lower GABA and norepinephrine densities tend to exhibit a compensatory effect of cortical slowing. These four neurotransmitter systems are impacted by PD (Scatton et al., 1983; van Nuland et al., 2020). In particular, reduced cortical dopamine is a strong predictor of cognitive dysfunction in PD (Aarsland et al., 2021). In combination with our finding that the sagittal gradient of cortical slowing is normalized by dopaminergic agonists, we interpret the dopaminergic co-localization of this effect as further evidence that neurophysiological compensation in PD is associated with frontal dopamine dysfunction. We note that these higher-order alignments of neurochemical systems with the clinical-neurophysiological gradient effect are not necessarily comparable to simpler relationships that may exist between cortical neurotransmitter systems and PD-related neurophysiological change (i.e., without considering clinical function). For example, our recent work in this area (Wiesman et al., 2023a) indicates that several of the neurochemical systems identified herein, including GABA, norepinephrine, and serotonin, are also relevant for lower-order multispectral neurophysiological changes. In contrast, dopaminergic systems appear to only be related to clinically-relevant cortical slowing, while cholinergic systems are instead related more prominently to broadband increases in neurophysiological power that are not considered in this study.

We emphasize that the manipulation of primary dopamine medications (i.e., levodopa) was not possible in the patient cohort, warranting caution when inferring the causal nature of these effects. All of the patients included in this study underwent data collection “ON” their usual regimen of dopaminergic medications. Previous work has indicated that levodopa does not substantively affect the multi-spectral slowing of neurophysiology that we investigated here (Stoffers et al., 2007). We also observed a normalizing of the clinical-neurophysiological gradient by dopamine agonists, suggesting a different origin than secondary medication effects. These suppositions require further testing in future empirical studies, because the reported effects may be modulated differently by levodopa than by dopamine agonists. Future research is also required to relate the reported patterns of cortical slowing to neuropathological changes in deep brain structures known to degenerate in PD (e.g., substantia nigra, locus coeruleus, and nucleus basalis of Meynert). These deeper structures have distinct neurochemical profiles, with implications for cortical neurophysiology, and thus could further inform the neurochemical bases of cortical slowing effects. In principle, spectrally-resolved neurophysiological measurements in these deeper structures, alongside measures of their structural degeneration and mapping of cortical slowing, would help address these questions, but such measurements remain challenging to perform non-invasively.

Taken together, these results suggest that the clinical significance of aberrant neurophysiological activity in PD depends on its cortical expression and neurochemical basis: the same neurophysiological patterns indicating impairment in one brain region can indicate compensation elsewhere on the cortex. These findings may account for the highly variable clinical outcomes of anatomically-targeted rhythmic modulation of frequency-specific neurophysiological activity using e.g., transcranial magnetic stimulation (Benninger et al., 2010; Cantello et al., 2002; Chou et al., 2015; Del Felice et al., 2019; Elahi et al., 2009; Fregni et al., 2006; Teo et al., 2017). In fact, the primary motor cortices were often targeted in those studies, which our results identify as a point of anatomical inflection along an adverse-to-compensatory gradient. Future studies to ameliorate cognitive and motor symptoms in patients with PD should deliver cortical stimulation in an anatomically selective manner, with the objective of normalizing slowing in posterior parietal cortices, and/or enhancing cortical slowing over inferior frontal regions. Ideally, spatial targeting of neuromodulation should be patient specific: our results hint at the notion that distinct neurostimulation protocols may be advised depending on the personalized profile of patients, including the laterality of symptom onset, prescription of dopaminergic agonists, and/or presence of subjective cognitive complaints.

To our knowledge, the present study is first to report both rhythmic and arrhythmic contributors to neurophysiological slowing effects in any patient population. Separating these contributions allowed us to detect some relationships to motor function and cognition that were not significant when using non-parameterized neurophysiological spectra. This provides evidence that rhythmic and arrhythmic slowing effects are at least partially distinct. We also anticipate these findings will impact ongoing research on rhythmic neuromodulation for the treatment of patients with neurodegenerative disorders. Although we report clinically-relevant rhythmic components of cortical slowing, we also find that shifts in the arrhythmic spectra are associated with clinical features. We foresee that future research will investigate whether the cortical slowing effects reported previously in other patient groups (Boord et al., 2008; Doesburg et al., 2011; Wiesman et al., 2022b) exhibit similar distinctions between arrhythmic and rhythmic components of neurophysiological signals. How the present model of cortical neurophysiological slowing may overlap with and differ from other methods of quantifying this effect is another question for future research. For example, recent work using the center of mass of the neurophysiological power spectra has shown a form of slowing of rhythmic brain activity in patients with atypical variants of Parkinson’s disease (Krösche etal., 2023). Such derivations, standardized with respect to healthy control data, may be an interesting alternative or complementary approach to the slope metric used herein.

We find that cortical slowing effects in PD are not confined to the amplitude of neurophysiological activity. Slowing also manifests in frequency-specific inter-regional connectivity (Supplementary Results and Figure S10). These effects may indicate a shift towards slower, more stable channels of neurophysiological communication in PD: a hypothesis that future research will need to adjudicate. The connectivity slowing effects are widely distributed across the cortex, and may define clinically meaningful measures of cortical slowing in other patient groups.

In sum, the present study shows that patients with Parkinson’s disease are affected by broad neurophysiological slowing across the cortex, albeit with distinctive associations to clinical outcomes depending on the locus of the slowed activity. Cortical slowing is associated with worse motor function and cognition in superior parietal regions, but transitions to a compensatory effect along a superior–inferior and posterior–anterior anatomical gradient towards inferior prefrontal regions. This sagittal gradient effect and its clinical implications can inform evidence-based targeting of neuromodulation therapies, particularly methods which are efficacious at the level of the cortex such as transcranial electrical and magnetic stimulation. We also demonstrate proof-of-concept for slowing of frequency-defined cortico-cortical functional connectivity in PD.

Highlights.

• Patients with Parkinson’s disease exhibit slowing of cortical neurophysiology

• This slowing comprises changes in both rhythmic and arrhythmic brain function

• Slowing in frontal cortex relates to preserved motor and cognitive abilities

• Cortical slowing in parietal regions relates to more pronounced clinical impairment

• This adverse-to-compensatory gradient mirrors individual clinical profiles