Locus Coeruleus atrophy doesn’t relate to fatigue in Parkinson’s disease

Fatigue is a frequent complaint among healthy population and one of the earliest and most debilitating symptoms in Parkinson’s disease (PD). Earlier studies have examined the role of dopamine and serotonin in pathogenesis of fatigue, but the plausible role of noradrenalin (NA) remains underexplored. We investigated the relationship between fatigue in Parkinsonian patients and the extent of degeneration of Locus Coeruleus (LC), the main source of NA in the brain. We quantified LC and Substantia Nigra (SN) atrophy using neuromelanin-sensitive imaging, analyzed with a novel, fully automated algorithm. We also assessed patients’ fatigue, depression, sleep disturbance and vigilance. We found that LC degeneration correlated with the levels of depression and vigilance but not with fatigue, while fatigue correlated weakly with atrophy of SN. These results indicate that LC degeneration in Parkinson’s disease is unlikely to cause fatigue, but may be involved in mood and vigilance alterations.

Additionally, subjects performed an auditory oddball detection task to assess their levels of vigilance, or ability to maintain attention over time. Vigilance is commonly measured as performance decrement over time during monotonous tasks 15 . Here we assessed it by looking at the increase in RT with time-on-task. As expected, we found that reaction time (RT) increased overtime (t-test on RT slopes: p = 0.01). Neither mean RT nor change of the RT overtime correlated with any of the questionnaires completed by the patients (all p > 0.05, all BF < 0.79).
We performed two additional exploratory analyses. First, we quantified the degeneration of the Substantia Nigra (SN, see Methods, Supplementary Fig. 2), and found that, in contrast to LC, SN degeneration correlated weakly with fatigue scores (SN-PFS, r K = −0.281, p = 0.018, BF: 3.350, weak evidence for H1). Second, we attempted to determine whether pupil response to oddball target detection ( Supplementary Fig. 3)

Discussion
In this study, we explored the plausible and under-investigated link between degeneration of LC, the main source of cerebral NA, and fatigue in PD. We found that LC degeneration was associated with depression and vigilance loss but, contrary to our predictions, not with fatigue ( Fig. 4). In contrast, in a post-hoc analysis, we found weak evidence for a link between fatigue score and SN degeneration. The relationship between neuromodulatory systems, fatigue, depression and sleep disturbance is considerably intricate. There is inconsistent evidence for the role of dopaminergic system in PD fatigue pathogenesis 1,[16][17][18] , and some studies suggest a possible involvement of cholinergic 19 and serotoninergic 18 pathways. Depression is a frequent symptom among PD patients 20 , and is linked with severity of LC degeneration, as found in the present, and earlier studies 12,13 . However, fatigue often remains unrelieved following either dopaminergic replacement Locus Coeruleus atlas mask Figure 2. Schematic illustration of Locus Coeruleus isolation. We normalized the whole brain anatomical image (not shown) to the atlas template, and then projected the location of LC defined in atlas space to the individual neuromelanin-sensitive image (left), followed by an additional precision-fitting step. Afterwards, the LC index was computed from the voxels with high signal intensity (extracted with the unsupervised k-means algorithm, see Methods). The masks for all subjects can be found in the Supplementary Fig. 1.  22 , suggesting different pathophysiological mechanisms for motor, mood and fatigue symptoms. Like depression, disturbance of sleep is highly debilitating for PD patients 23 , and in our study we also found a link between severity of fatigue and sleep disorders. However, sleep does not always have restorative effect on fatigue and patients complain of fatigue even in the absence of sleep disorder 1,24 .
It is also worth to mention several limitations of our study. First, fatigue assessment was based on psychometric methods, and was therefore submitted to the usual limitations of this approach. Notably, fatigue assessment through self-questionnaires can be confounded with similar concepts such as sleepiness or depression 25 . This cross-influence was partly accounted for in our study, in which we included separate questionnaires for these three concepts. Second, our estimates of LC and SN degeneration were based only on voxel intensities, neglecting the spatial extent of the structures. It remains currently unclear which approach provides the best estimate of degeneration. Third, the relationship between neuromelanin content and NA denervation requires careful considerations. It has been shown that the number of cells that contain neuromelanin (a byproduct of catecholamine synthesis) is roughly the same as the number of LC cells that contain tyrosine hydroxylase (a critical enzyme used for catecholamine production), at least in older adults 26,27 . These findings suggest, albeit quite indirectly, that presence of neuromelanin is specific to cells that produce NA and that reduction in neuromelanin signal reflects degeneration of NA-producing cells. Finally, we only tested the patients while they were on the dopamine replacement therapy. Despite the fact that non-motor symptoms of Parkinson's disease such as fatigue and depression are notoriously resistant to medication 1,28 , it would be interesting to also test the patients off-treatment in a future study.
In our additional exploratory analyses, surprisingly, we found a relationship between SN degeneration and pupil responses, while the correlation with LC was absent. This finding appears to contradict earlier claims that have emphasized the link between LC activity and pupil responses 29 . However, the causality of the LC-pupil relationship has been put into question recently 30 and the present results are the first to address the effect of LC degeneration on pupil size. Additionally, compensatory mechanisms could very well intervene to countervail the loss of noradrenergic neurons. For instance, the remaining population (or other NA producing nuclei 31 ) could increase its average discharge rate to maintain overall cortical concentration of NA unchanged. The correlation of pupil responses with SN degeneration is also surprising given that activity of neurons in SN pars compacta was shown to be unrelated to pupil responses in primates 32 . However, the complex interrelation between dopamine and NA systems 33 suggests that SN degeneration may impact on pupillary responses through its effect on general arousal mechanisms. These puzzling findings will require further investigation. Similarly, we found that SN degeneration correlated with fatigue, a link that may be worth exploring in a future study, despite the lack of cohesion in the literature [16][17][18] . At the same time, SN signal intensity failed to correlate with disease severity. This might seem surprising, however, previous findings are highly controversial, with some reporting significant correlations 34 while others failing to find significant effect 35,36 . The latest results highlight the importance of spatial specificity of SN measurement, as disease severity was correlated only with subparts of SN 37 , or only with SN contralateral to the clinically impaired body side 38 .
All in all, we found that the decrease in neuromelanin signal in LC is not related to the severity of fatigue but is associated with increased level of depression and decreased vigilance. Elucidating the underlying mechanisms of fatigue will have a major impact on development of pharmacotherapy to improve patients' quality of life and we believe that our study will advise further research on the neural correlates of fatigue.

Methods
The experiment comprised one unique session, divided in two parts, whose order was counter-balanced across subjects. One part consisted of the MRI acquisition and the other part corresponded to the auditory oddball detection task. For each patient, we fine-aligned the LC mask derived from atlas (see Methods). Afterwards, in order to isolate high intensity region of LC, we applied a k-means clustering algorithm on the intensity of voxels within the aligned atlas region (Fig. 2, Supplementary Fig. 1). The cluster with the highest intensities was assigned to LC and the values of the cluster were normalized with respect to mean brainstem intensity 39  Due to the lack of data on safinamide LED, the safinamide doses were not integrated in the calculation of the total LED for 2 patients. Additionally, MDS-UPDRS, age, duration of the disease and LED were not assessed for 5 patients; PFS, BDI, PDSS for 3 patients, and the scanning was not performed for 2 patients due to contraindications revealed on the day of the experiment. The datasets are available from the corresponding author on reasonable request. All statistical analyses were performed using JASP 0.8.2 (JASP Team, 2017) and MATLAB (The MathWorks, Natick, MA, USA). Correlations between variables were performed with Kendall correlation in order to mitigate the influence of outliers and non-normality of LC index distribution (Fig. 3B,C), and are reported along with the Bayes factors (BF).
Participants were seated in front of a 19″ CRT screen cadenced at 100 Hz, with their head resting on a chinrest. They were instructed to respond as quickly as possible to target tones (1000 Hz) by a mouse click, while ignoring the standard tones (500 Hz). Targets were randomly interspersed throughout the task such that 4 to 10 standard tones (drawn from a uniform distribution) were included between 2 successive targets. The time interval between successive tones, irrespective of whether they were targets or standard tones, ranged between 2 and 3 seconds, distributed uniformly. Consequently, the minimum inter-target interval was 8 s. The number of trials was adjusted so that the total duration of the task was 30 min. The patients were installed in front of a computer screen with a grey background and were instructed to fixate on a white cross in the center of the screen. Task performance was evaluated by measuring the time between each target onset and the following response (reaction time, RT). The data was separated in 10 equally-sized epochs, and the mean of the RT was computed within each epoch, as well as for the entire experiment (RT mean). We then evaluated the progression of the RT over time by computing the Spearman correlation between the epoch index and the corresponding average RT (RT change).
Pupillometry and pupil data analysis. We recorded pupil size continuously during the oddball detection task. Given the assumed link between LC and pupil response 29 , this measure was acquired in order to obtain, in addition to the structural measure provided by the neuromelanin sequence, an estimate of the functional integrity of the noradrenergic system. We used the Eyelink 1000 eyetracker (SR Research, Ottawa, Canada), situated just below the screen and centered on the dominant eye of the participant. The 500 Hz pupil size signal was then normalized, downsampled to 10 Hz and analyzed by means of an autoregressive model with exogenous inputs (ARX 45 ). This method allowed us to isolate the pupillary response to distractors, targets and blinks. Given the non-stationarity of the pupil signal during prolonged recordings, which violated the assumption of the ARX method, we sliced the 30-minute data into six 5-minute chunks. Because of this additional block variable, we analyzed pupil responses with linear mixed models (fitlme in Matlab; model: Pupil~SN + LC + Block + (Block| Subject)).

MRI data acquisition.
All subjects were scanned using 3 T scanner (Achieva, Philips Healthcare, Eindhoven, the Netherlands) with a SENSE 32-channel phased array head coil. Neuromelanin-sensitive images were obtained using T1 turbo spin-echo sequence: FOV 220 × 220 mm 2 , 16 slices aligned perpendicularly to the commissural-obex plane, 0.43 × 0.43 × 2.5 mm resolution, TR/TE/FA = 600 ms/14 ms/90°. We also acquired whole brain Turbo Field Echo anatomical images with the following parameters: FOV 220 × 197 mm 2 , 150 slices, 0.81 × 0.95 × 1 mm resolution (reconstruction 0.75 × 0.75 × 1), TR/TE/FA = 9.1 ms/4.6 ms/8°. LC mask alignment. The MRI data was analyzed using SPM8 (Wellcome Department of Imaging Neuroscience, London, UK; www.fil.ion.ucl.ac.uk/spm) and in-house developed MATLAB code. The normalization consisted first in aligning the individual neuromelanin-sensitive and whole-brain anatomical images and then normalizing the whole-brain anatomical image to the atlas template. Having established the mapping between neuromelanin-sensitive, whole-brain and atlas images, we used the corresponding transformation parameters to project the location of LC defined in atlas space 46 to the individual neuromelanin-sensitive image. This approach resulted in only approximate positions of LC due to the idiosyncrasies of individual brain anatomy and inevitable imperfections of the normalization procedure. Therefore, we performed an intensity-driven alignment by shifting the atlas masks derived from the normalization step to the position with the highest average signal intensity. The space of possible shifts was defined in all three axes. Inward-outward shifts between left and right LCs was also permitted, maximal x -y -z -in/out: 4-4-2-2 voxels (Supplementary Fig. 1). The shifts were smaller in z axis because slice thickness was considerably bigger than in-plane voxel size (2.5 vs 0.4 mm). Following the alignment, we used K-means algorithm to split all voxels within the mask into K clusters, based on signal intensity. K-means iteratively assigns each data point to one of the K clusters, minimizing the sum of squared distances from the data point to cluster mean. The value of K has to be chosen a priori, and we performed the correlations with a range of K values between 2 and 4. Since results did not depend on K, we finally used the LC metrics based on 2 clusters for the sake of brevity. SN degeneration was quantified with an identical procedure, following initial isolation based on probabilistic SN atlas 47 (Supplementary Fig. 2).