Clinical severity in Parkinson’s disease is determined by decline in cortical compensation

Abstract Dopaminergic dysfunction in the basal ganglia, particularly in the posterior putamen, is often viewed as the primary pathological mechanism behind motor slowing (i.e. bradykinesia) in Parkinson’s disease. However, striatal dopamine loss fails to account for interindividual differences in motor phenotype and rate of decline, implying that the expression of motor symptoms depends on additional mechanisms, some of which may be compensatory in nature. Building on observations of increased motor-related activity in the parieto-premotor cortex of Parkinson patients, we tested the hypothesis that interindividual differences in clinical severity are determined by compensatory cortical mechanisms and not just by basal ganglia dysfunction. Using functional MRI, we measured variability in motor- and selection-related brain activity during a visuomotor task in 353 patients with Parkinson’s disease (≤5 years disease duration) and 60 healthy controls. In this task, we manipulated action selection demand by varying the number of possible actions that individuals could choose from. Clinical variability was characterized in two ways. First, patients were categorized into three previously validated, discrete clinical subtypes that are hypothesized to reflect distinct routes of α-synuclein propagation: diffuse-malignant (n = 42), intermediate (n = 128) or mild motor-predominant (n = 150). Second, we used the scores of bradykinesia severity and cognitive performance across the entire sample as continuous measures. Patients showed motor slowing (longer response times) and reduced motor-related activity in the basal ganglia compared with controls. However, basal ganglia activity did not differ between clinical subtypes and was not associated with clinical scores. This indicates a limited role for striatal dysfunction in shaping interindividual differences in clinical severity. Consistent with our hypothesis, we observed enhanced action selection-related activity in the parieto-premotor cortex of patients with a mild-motor predominant subtype, both compared to patients with a diffuse-malignant subtype and controls. Furthermore, increased parieto-premotor activity was related to lower bradykinesia severity and better cognitive performance, which points to a compensatory role. We conclude that parieto-premotor compensation, rather than basal ganglia dysfunction, shapes interindividual variability in symptom severity in Parkinson’s disease. Future interventions may focus on maintaining and enhancing compensatory cortical mechanisms, rather than only attempting to normalize basal ganglia dysfunction.

Behavior Disorder Screening Questionnaire. 9 Autonomic function was assessed with the Scales for Outcomes in Parkinson's disease. 10

Detailed description of the action selection task
Each trial began with the presentation of a fixation cross.After a random inter-stimulus interval of 2-4 seconds, the cross was replaced by a cue consisting of four circles, with each circle corresponding to a button on a response device.The circles were either filled or empty to indicate which responses were correct or incorrect, respectively.Participants were instructed to respond to cues by pressing a single button corresponding to a single filled circle.If multiple circles were highlighted, then participants were instructed to choose which circle to respond to.Only one response was allowed per trial.Each cue included either one, two, or three filled circles, thereby varying the number of response choices that participants were presented with on a trial-by-trial basis.Participants were encouraged to make use of all four buttons rather than selectively responding with only a subset of buttons, to vary the finger that was used to respond during trials where multiple response options were presented, and to respond as quickly and as accurately as possible.Cues remained on the screen for a maximum of 2 seconds or until a response was recorded and were immediately followed by the presentation of a new fixation cross.The task consisted of 132 trials and lasted for approximately 10 minutes, depending on performance.There were 60 one-choice trials (15 per finger), 30 two-choice trials, and 30 three-choice trials.The additional 12 trials consisted of 6 one-choice trials, 3 two-choice trials, and 3 three-choice trials where circles were outlined in red.Participants were instructed to withhold a response during these "catch" trials.Trials were presented in three blocks, each consisting of 44 trials.Out of these 44 trials, 20 were one-choice, 10 were two-choice, 10 were three-choice, and 4 were catch.The ordering of trial conditions within each block was pseudo-randomized.Blocks were separated by 20 seconds of rest.Prior to entering the scanner, participants practiced by performing one continuous block of 68 trials (30 one-choice trials, 15 two-choice trials, 15-three choice trials, 8 catch trials).Trials of one-, two-, and three-choice conditions were classified as correct, incorrect, or miss.Correct responses were defined as button presses executed within two seconds following cue onset that corresponded to a single, highlighted circle (in the case of catch trials, correct responses were defined as no response).Incorrect responses were defined as button presses executed within two seconds following cue onset that did not correspond to a highlighted circle.
Misses were defined as cues where no button press was executed within two seconds following cue onset.Patients were asked to perform the task with their most-affected hand.The responding hands of healthy controls were matched to the 56 patients who were assessed in an off-medicated state.

Preprocessing details
Registrations of functional images to anatomical space were estimated using linear transformations with boundary-based registration and six degrees-of-freedom. 11Non-linear transformations were estimated from anatomical to MNI152Lin6Asym-space. 12,13 A single interpolation step was used to carry out all transformations in combination with motion correction 14 and slice time correction. 15nfound time series were generated for framewise displacement and DVARS, 16 along with 24 motion derivatives.An anatomical principal component analysis was performed to derive time series for cerebrospinal fluid and white matter signal. 17ICA-AROMA was used to derive time series of motion-related noise 18,19 whose classification was further refined using custom methodology (see section below).A series of discrete cosine-basis functions were derived for highpass filtering (>0.008Hz).Tremor regressors were generated from the accelerometry data of 69 patients.

Refinement of ICA-AROMA component selection
Additional steps were taken to ensure that the removal of confounding motion-related variability through ICA-AROMA during the first-level analysis did not adversely affect the estimation of task-related regressors.For each participant, the classification of ICA-AROMA components was refined in a multiple regression analysis where each noise time series was modelled as a function of task regressors for choice and catch conditions.Time series that shared more than 5% explained variance, as assessed by the r 2 of the regression model, were reclassified as non-noise and subsequently left out of the first-level design.

Quantification of tremor
For all patients, tremor severity was quantified using a three-axial accelerometer placed on the dorsum of the most-affected hand.Accelerometry preprocessing involved detrending, demeaning, transformation to scan-to-scan tremor power at peak frequency, and log-transformation, after which the tremor signal was convolved with a canonical hemodynamic response function. 20The resulting tremor regressors were added to the first-level models of 69 patients whose tremor was confirmed through visual inspection.

Exclusions and final sample sizes for analyses of behavioral performance and task-related activity
All healthy controls had full data.For patients, final sample sizes were determined separately for analyses of task performance and brain activity.Exclusions were carried out sequentially in the order that they are reported.Note that most analyses depended on data availability along multiple data streams and were therefore subjected to further constraints with respect to final sample size.

Behavioral performance
Out of 367 patients assessed in the on-medicated state, 14 were excluded due to misdiagnosis, 5 had missing behavioral data, and 11 were excluded due to poor performance.This led to a total sample size of 337 patients.50 of these patients had data from assessments in the off-medicated state, 306 had sufficient data to carry out subtype classification, 335 had usable bradykinesia scores, and 321 had usable cognitive composite scores.

Task-related activity
Out of 367 patients assessed in the on-medicated state, 14 were excluded due to misdiagnosis, 7 had missing imaging data, 9 had insufficient behavioral data to carry out 1 st -level analyses, 3 were lost due to technical issues, 8 were excluded based on excessive movement, and 3 were excluded due to poor accuracy.This led to a total sample size of 323.48 of these patients had data from assessments in the off-medicated state, 296 had sufficient data to carry out subtype classification,

Supplementary material 2 Task-related activity across groups
Task-related activity was investigated across groups to generate activation maps of motor-activity, catch-related activity, intermediate selection-related activity, and high action selection-related activity.

Motor-related activity
A conjunction analysis of one-choice, two-choice, and three-choice activity yielded a network of cerebellar, visual, parietal, insular, and sensorimotor activity (Supp.Fig. 1A).Sensorimotor activity was more extensive in the left hemisphere, as would be expected given that this was the responding side of all participants, either naturally or as a result of horizontal flipping of contrast images.

Action selection-related activity
Moderate (Supp.Fig. 1C) and high (Supp.Fig. 1D) demand on action selection elicited increased activity in a frontoparietal network and decreased activity in regions involved in the default mode network.This is consistent with the idea that action selection led to the recruitment of additional cognitive processing beyond simple motor responses.

Bradykinesia severity
Bradykinesia severity was associated with selection-related activity in a network of parietopremotor regions, primarily in the left hemisphere (Supp.Fig. 7).It may be argued that increased parieto-premotor activation could reflect alterations in saccadic eye-movement control rather than the activation of compensatory processes. 21,22Arguing against this interpretation for correlations related to bradykinesia, we found no clusters that centred on typical saccade control regions, such as the frontal and posterior eye fields.

Cognitive performance
Like bradykinesia severity, correlations with cognitive performance are located primarily in a network of parieto-premotor regions (Supp.Fig. 7).However, these regions tended to be adjacent to the regions implicated in bradykinesia rather than overlapping with them, and were preferentially located in the right hemisphere.Here, we observed a single cluster in the frontal eye fields of the right hemisphere.While we cannot exclude the possibility this effect may relate to saccadic eye-movement control, we note that the frontal eye fields also play an important role in goal-directed guidance of attention, an important aspect of general cognitive function.Anatomical labels were derived from the Anatomy Toolbox v3.0.Area labels were derived from the Glasser atlas.

Overlap between domains
Interestingly, there was little overlap between cluster related to bradykinesia severity and cognitive performance.

Supplementary material 4 Do individual differences in cortical compensatory capacity depend on underlying differences in reserve?
Cerebral compensation refers to functional adaptations that enable patients to meet behavioral demands despite the presence of some form of underlying dysfunction in the mechanisms that ordinarily support behavioral performance, particularly when the task at hand is relatively demanding.6][27][28][29] Patients with higher reserve may experience less severe symptoms compared to patients with low reserve despite having similar levels of pathology because their brains have more resources to draw from.Reserve typically manifests as a trait-like property in the sense that its effect on behavior tends to be present across multiple clinical domains.In contrast to compensation, which can be observed as a recruitment of additional resources in response to increasing task demands (through upregulation, selection or recruitment of additional mechanisms), reserve can only be estimated indirectly through proxy measures that are argued to have conferred beneficial effects on neural resources prior to disease onset.For example, higher levels of educational attainment may strengthen brain structure and function, thereby increasing resistance to pathology.We should note that some researchers define reserve in terms of both behavior (cognitive 28 or motor 26 reserve) and underlying brain function (neural reserve). 28For example, it has been suggested that cognitive reserve denotes factors that reduce susceptibility to cognitive decline, whereas neural reserve denotes factors that reduce susceptibility to decline in brain function.According to this conceptualization, cognitive reserve can be supported by both neural reserve and cerebral compensation.However, the distinction between behavioral reserve and neural reserve can appear somewhat artificial, given that behavior depends on the brain.For the purposes of this study, we focus solely on compensation and reserve in the context of brain activity, where the two concepts can be more clearly distinguished from each other.
Cerebral compensation can be disentangled from reserve by assessing the relationship between brain activity and clinical severity while simultaneously controlling for reserve proxies.
Individual differences in cerebral compensation should be more strongly related to behaviour than with reserve proxies and can be expected to be sensitive to manipulations of task difficulty.Neural reserve should be more related to proxy measures of reserve, and, given its trait-like nature, may not be as sensitive to manipulations of task difficulty.However, compensation may depend on reserve.It is conceivable that higher levels of reserve may enable patients to more effectively recruit compensatory resources.When investigating cerebral compensation, it is therefore useful to explore whether individual differences in compensatory activity depend on underlying reserve.
In our between-subtype comparisons and brain-clinical association analyses we already controlled for several proxies of reserve, such as age, 30 sex, 31,32 and years of education, 33 which are considered prominent contributors to reserve in relation to both motor and cognitive domains. 26,34However, there are additional proxy measures of reserve that may probe different aspects of reserve in relation to Parkinson's disease.For example, there is evidence suggesting lower body-mass index, more engagement in physical activities, lower non-motor burden, and smoking history may decrease susceptibility to the behavioral effects of pathological decline. 26,35,36 therefore performed an exploratory re-analysis of our between-subtype comparisons and brain-clinical association analyses to test whether the relationship between increased taskrelated activity and clinical severity (as quantified through subtyping and clinical scores of symptom severity) remained after controlling for additional reserve proxies: body-mass index (continuous measure), smoking history (no/yes), engagement in physical activity (Physical Activity Scale for the Elderly [PASE]; 37 continuous measure), and non-motor burden (MDS-UPDRS part I; 2 continuous measure).Body-mass index and physical activity should be viewed as approximations of pre-morbid lifestyle choices given that they were measured at baseline.
In comparisons between subtypes, we observed that regions showing stronger selectionrelated activity in both mild-motor predominant and intermediate subtypes compared to the diffuse-malignant subtype remained significant (Supp.Table 1).Furthermore, we observed additional parietal clusters of stronger motor-related activity in the mild-motor predominant subtype compared to the diffuse-malignant subtype.This supports our hypothesis that upregulation of parieto-premotor activity is more strongly related to compensation than reserve.
The inverse relationship between selection-related activity and bradykinesia severity remained significant in parieto-premotor cortex, but only at moderate action selection demand (Supp.Table 1).Correlations between selection-related activity and cognitive performance did not remain significant, with the exception of a cluster in the secondary visual cortex.However, better cognitive performance was now associated with greater motor-related activity in primary somatosensory cortex and supplementary motor area.In combination, these findings suggest that reserve may influence the degree to which compensatory cortical adaptations contribute to counteracting specific clinical domains.Anatomical labels were derived from the Anatomy Toolbox v3.0.Area labels were derived from the Glasser atlas.
Voxel-wise whole-brain comparisons revealed decreased gray matter volume in patients compared to controls in a network consisting of large of visual, temporal, orbitofrontal, posterior parietal cortex, and the anterior striatum (Supp.Fig. 7A).Parietal cortex and striatal atrophy was primarily lateralized to the left hemisphere.

Supplementary figure 1
Task-related activity across patients and healthy controls.(A) Sensorimotor network activation common to all levels of action selection demand.(B) Response withholding preferentially activates visual and prefrontal cortex.(C) Moderate and (D) high action selection demand leads to activation of the frontoparietal network and deactivation of the default mode network.

Supplementary table 1 Voxel-wise correlations of bradykinesia severity
Anatomical labels were derived from the Anatomy Toolbox v3.0.Area labels were derived from the Glasser atlas.