Reorganisation of diffusion microstructure in the precuneus is associated with preserved cognitive function in Parkinson’s disease


 Functional neuroimaging studies of patients with Parkinson’s disease (PD) have repeatedly identified over-activations in midline structures (medial prefrontal cortex, anterior cingulate cortex, posterior cingulate cortex, and precuneus), especially in those without comorbid dementia. Here, we investigated whether the different cognitive profiles in PD were linked to measures of diffusion microstructure in medial regions of the brain. Using magnetic resonance based diffusion weighted imaging (DWI) in healthy volunteers (HV) and PD patients with and without mild cognitive impairment (PD-nonMCI and PD-MCI), applying diffusion tensor imaging (DTI) and high angular resolution diffusion imaging (HARDI) techniques, we observed: 1) increased fractional anisotropy (FA) in the precuneus and the anterior cingulate in the PD-nonMCI participants compared with the HV; 2) an association between precuneus FA and executive and memory function, respectively, in PD and HV; 3) a negative correlation between age and midline structure FA in PD but not HV; and 4) a differential association between cognitive scores and apparent fiber density (AFD) of the posterior cingulate-precuneus bundle in HV vs. PD. Together, these findings suggest that white matter reorganization of the posterior medial microstructures might serve a compensatory role for damaged basal ganglia function in PD-nonMCI.


Introduction
Parkinson's disease is a progressive neurodegenerative disorder caused by the loss of dopamine neurons 1 . Motor symptoms emerge after more than 50 to 80% of nigrostriatal dopaminergic neurons have lost their function 2-4 with a prodromal phase occurring at least a few years beforehand 3,5 . The main symptoms are considered to be the motor dysfunction, but, in some patients, cognitive de cits are present even at early stages of the disease 6 . The motor and cognitive impairments are associated with dopamine de ciency in the striatum within parallel cortico-basal ganglia-thalamocortical circuits 7-10 , and positron emission tomography (PET) studies of PD patients have identi ed impaired striatal dopaminergic function associated with both motor [11][12][13][14] and cognitive problems 11,13,15,16 . These observations are paralleled by functional magnetic resonance imaging (MRI) studies, which have shown reduced motor circuit activity associated with motor dysfunction [17][18][19] and reduced cognitive circuit activity associated with cognitive impairments [20][21][22][23] .
In contrast to these neurological de cits, PD patients have also been repeatedly reported to exhibit overactivations in the medial prefrontal cortex 18,19 and lateral prefrontal cortex 17,24 during the performance of motor tasks, and over-activation in the precuneus during cognitive tasks 21,25 . These increases appear to become greater as the disease progresses, and larger task-induced activations in the medial prefrontal cortex and precuneus have been observed at 2-year follow-up compared to the rst scan in non-demented PD patients both with and without mild cognitive impairment 26 . Moreover, resting state fMRI measures have identi ed increased hub function in the anterior part of the precuneus and the posterior cingulate cortex in PD patients without cognitive impairment (PD-nonMCI), but not in PD patients with mild cognitive impairment (PD-MCI), and hub function in these posterior medial structures was positively correlated with cognitive function in all PD patients 27 . In part, these alterations might be related to dopamine perturbations, and a positron emission tomography study identi ed evidence of upregulated anterior cingulate dopaminergic projections associated with motor dysfunction 14 . Together, these observations may indicate that an upregulation of midline structures play an important role in supporting motor and cognitive functions in PD patients, likely by compensating for the dysfunction of cortico-basal ganglia thalamocortical circuits, possibly beginning during the prodromal phase.
The hypothesized upregulation of midline neurocircuitry might be associated with microstructure changes in structural connectivity. This can be tested with local microstructural measurement of water diffusion through diffusion-weighted MR imaging (DWI), with diffusion tensor imaging (DTI) 28 [38][39][40] . Cognitive function of PD patients was associated with widespread degeneration of white matter, especially, in the frontal and temporal, and cingulate areas [41][42][43][44] . DTI requires an assumption of Gaussian diffusion of water, which may not occur in biological tissues.
To overcome this limitation, newer techniques have been developed to analyse DWI 35,36,45 . Using one of them, diffusion kurtosis imaging, it was reported that FA in the anterior cingulate, but not in the posterior cingulate, was signi cantly lower in PD patients, compared with HV 46 . This may indicate that the microstructure of the posterior cingulate is relatively preserved in PD patients, and that it could compensate PD's dysfunction of the cortico-basal ganglia-thalamocortical circuits, especially for cognitive function in early stage. High angular resolution diffusion imaging (HARDI) technique is another one. It further enables us to create ber orientation distribution function (fODF) models [32][33][34][35][36] , which offer accurate tractography and apparent ber density (AFD) for each streamline 47,48 . Here, we aimed to see if changes in microstructure measures of the medial regions are associated with cognitive function in PD patients, using DTI and HARDI techniques. With this method, we predicted associations between a reorganization of the posterior part of the midline structure and cognitive function in PD.

As shown in Supplementary
Group comparison of fractional anisotropy (FA) and mean diffusivity (MD) After pre-processing the diffusion weighted imaging (DWI) data, diffusion tensor image (DTI) analyses were applied to create fractional anisotropy (FA), and mean diffusivity (MD) matrices for each participant, and a region of interested analysis was done, using four sets of regions of interest (ROIs) ( Fig. 1-a). Correlation analysis of FA with age, Brixton, and RAVLT Zscores Results are summarized in Tables 1, for HV and PD, separately. RAVLT Z-scores and age tended to positively correlate with the precuneus FA in HV (r = 0.691, p = 0.003, and r = 0.440, p = 0.088, respectively), whereas in PD participants, precuneus FA was positively and negatively correlated, respectively, with Brixton Z-scores and age (r = 0.350, p = 0.031, and r = -0.390, p = 0.015, respectively) (Tables 1, Fig. 3).
When age was covaried out, the signi cant correlation survived between the Precuneus FA and RAVLT Zscore (r = 0.62, p = 0.015), in HV, and marginal correlation was observed between Precuneus FA and Brixton Z-scores in PD (r = 0.277, p = 0.097). The FA in all the midline structures were strongly negatively correlated with age in PD, but not in HV. Correlation ratios were signi cantly different between the HV and PD, for Precuneus-FA and age, Precuneus-FA and Brixton Z-scores, and Precuneus-FA and RAVLT Z-scores, respectively (Z = 2.724, p = 0.0032; Z = 2.0719, p = 0.0191, and Z = 2.9033, p = 0.0018, respectively) ( Fig. 3).

Tractography
High angular resolution diffusion imaging (HARDI) was applied on the DWI data to create ber orientation distribution function (fODF) models. From the models, we extracted ve bundles, for each hemisphere; 1) including the anterior part of the precuneus, but not the posterior part of the precuneus, and the posterior cingulate (PCG-Precuneus), 2) including the posterior cingulate and the parahippocampal area (PCG-PHG), 3) including the precuneus and the parahippocamal area, but not the posterior cingulate (Precuneus-PHG), 4) including the precuneus and the paracentral area, the primary motor area, but not the posterior cingulate (Precuneus-Cortex), and 5) including the precuneus but not extending to the posterior cingulate, parahippocampal area, or paracentral area (Precuneus-Local). Additionally, 6) one crosshemisphere bundle was extracted including the both sides of the precuneus via the corpus callosum (Precuneus-Cross). The averaged track density images in MNI standard-space, of the six maps are shown in blue-green for HV ( Fig. 1-c, the location of the corresponding maps not being different for the other two groups, PD-nonMCI and PD-MCI). The PCG-PHG bundle ( Fig. 1-c, top middle) was located more laterally than the PCG-Precuneus bundle ( Fig. 1

Group comparison of apparent ber density (AFD) index
For each streamline, with fODF models, apparent ber density (AFD) index was estimated using the MRtrix3 package 49 . Signi cant differences in AFD index were observed in the Precuneus-Cortex between HV and PD-nonMCI (p = 0. 0036), and in PCG-PHG between HV and PD-nonMCI (p = 0.047) and between PD-nonMCI and PD-MCI (p = 0.048) (Fig. 4). The latter effect was weakened but otherwise similar when age was added as a covariate (p = 0.098).

Discussion
We interrogated a DWI dataset with DTI and HARDI techniques to test whether brain midline microstructure is associated with cognitive function in patients with PD. Our primary ndings were, 1) increased FA in the precuneus and the anterior cingulate in the PD-nonMCI participants compared with HV, which was not observed in PD-MCI participants; 2) precuneus FA was associated with executive and memory function, respectively, in PD and HV; 3) midline structure FA was negatively correlated with age in PD, but not in HV; and 4) the AFD index of the posterior cingulate-precuneus bundles were differentially associated with cognitive scores in HV and PD.
To our knowledge, this is the rst study showing increased FA in the midline structure (the cingulate gyrus) in PD-nonMCI patients, compared with HV. FA in DTI is the most robust metric for quantifying diffusion anisotropy 50,51 . Increased FA can re ect diminished axonal branching, increased axonal myelination 52 , or axonal sprouting 53 . Functional neuroimaging studies have repeatedly shown increased activity in the midline structure in PD patients vs. controls. This over-activation has been reported in the medial prefrontal cortex 18,19 and precuneus 21,25 while performing motor and cognitive tasks. During cognitive tasks, reduced deactivation in the default mode network, including the anterior and posterior midline structure, has also been reported in PD patients [54][55][56] . This might include a dopamine component, and upregulation of dopaminergic projections in the anterior cingulate cortex are associated with PD motor dysfunction 14 . These features appear to change as the disease progresses, and a longitudinal study showed greater activation in the midline structure at the second time compared to the rst time in PD with and without mild cognitive impairment (Nagano-Saito et al, 2016). Increased hub function in PD-nonMCI has also been observed in the posterior part of the midline structure 27 . Our present ndings regarding the FA change, together with these previous studies, possibly re ects increased axonal packing or sprouting, rather than diminished axonal branching, in the midline structure of the PD-nonMCI patients.
Mean diffusivity (MD) represents the average diffusivity of the component and provides a generalized measure of diffusivity in the area, and is nonspeci c with respect to the directionality of the diffusion process 57 . Increased MD has been reported in disease-related neurodegeneration 58,59 , and is thought to re ect decreases in membrane density due to cell degeneration 60 . We observed increased precuneus MD but not decreased precuneus FA, in the PD-MCI, compared with PD-nonMCI (Fig. 4). In addition to dysfunction of the cognitive cortico-basal ganglia-thalamocortical circuit in PD patients 7-10 , the cognitive de cits in PD patients are also originating from cortical Lewy Body/alpha-synuclein deposits that can occur in posterior regions of the brain 61 . We speculate that the possible increased axonal packing or sprouting speculated in PD-nonMCI is not preserved in PD-MCI because of the increased neurodegeneration occurring in the latter group.
Correlation analyses with the cognitive Z-scores yielded statistically signi cant associations only in the precuneus FA. In the HV group, RAVLT Z-scores was positively correlated with the FA, whereas in the PD group, Brixton Z-scores was positively correlated with FA (Tables 1, Fig. 3). This would indicate that the precuneus cognitive contribution has a bigger effect on memory in HV, and a bigger effect on executive function in PD. Supporting this interpretation, there is evidence that the precuneus is involved in memory retrieval in HV 62 . This observation may indicate that the possible increased axonal packing or sprouting in midline regions in PD-nonMCI patients compensates for executive dysfunction resulting from the impairment of the cortico-basal ganglia-thalamocortical circuits in PD, as we have previously proposed based on resting state fMRI data 27 .
Although FA and MD are robust metrics for quantifying diffusion anisotropy, they do not provide information about streamlines passing the ROIs. They are voxel averages. This is why the AFD 47,48 were calculated. AFD indicates relative white matter bre density per unit volume of tissue and we computed total AFD of associated streamlines belonging to speci c bundles, normalized by the mean streamline length 47,63,64 . Based on the results of group comparisons of FA and MD, we were most interested in the bundle of the PCG-Precuneus and Precuneus-PHG, which included the precuneus ROI in Fig. 1-a. While signi cant differences in AFD index were not observed in PD-nonMCI compared with HV in those bundles (Fig. 4), signi cant correlation of the Precuneus-FA was only observed with the PCG-Precuneus AFD index (r = 0.64, p = 0.0070, in HV; r = 0.39, p = 0.015 in PD, respectively), and not with other bundles AFD index (Supplementary Table 3).
There were signi cant differences of correlation patterns between the HV and PD, for PCG-Precuneus AFD index and age, and AFD index and RAVLT Z-scores (Fig. 5). In the HV, signi cant positive correlation with age, marginally negative correlation with the Brixton Z-scores, and marginal positive correlation with the RAVLT Z-scores with AFD index were observed, whereas in PD patients, negative correlation with the RAVLT Z-scores occurred. We propose that the PCG-Precuneus bundle, which are important for HV to support memory, is used for supporting executive function in PD, but that reorganization of other bundles also could support cognitive function, resulting in the positive correlation between Precuneus FA and Brixton scores in PD.
We have previously shown that the dopamine in the postmedial cortex including the precuneus has an important role for network regulation 65 . The location of the PCG-Precuneus bundle seems to follow the D2/3 dopamine receptors distribution in the midline of our previous positron emission tomography study 65 . Therefore, we compared the location of the six bundles of our present study to the D2/3 receptor distribution from our previous study, and the PCG-Precuneus bundle overlapped signi cantly with the D2/D3 distribution (Supplementary Fig. 1). Although all of the neurotransmitters associated with this possible compensation are unknown, we propose that dopamine in the postmedial cortex might at least partially contribute to this.
In HV, marginally positive correlations between the Precuneus FA and age, and signi cantly positive correlation between PCG-Precuneus AFD index and age, were observed (Table 1). To our knowledge, signi cant positive correlation between the posterior part of the midline microstructure and age in HV has not been reported. FA decreases in most part of the brain with age; however, in some speci c regions, FA increases with age 66 . The microstructure in the cingulum shows relative preservation with age 67 . Thus, microstructure measurements for speci c regions and their corresponding bundles may show positive correlations with age. More studies are required to investigate this nding further. Contrary to the HV group, we observed a strong negative correlation between the FA in the midline structure and age in PD patients (Table 1). With AFD, the Precuneus-Cross bundle, connecting the left and right hemisphere via corpus callosum, showed a signi cantly negative correlation with age in PD patients (Table 1, Fig. 5). Moreover, the positive correlation observed between the PCG-Precuneus bundle and age in HV disappeared in PD patients, and when the correlation rate in HV and in PD was compared, a signi cant difference was observed (p = 0.015). We hypothesize that the midline structure upregulates its hub function for cognition in PD patients 27 , as long as the compensation potential is preserved.
This compensation process would start during the prodromal phase, which occurs at least a few years beforehand 3,5 , accompanied by axonal packing or sprouting. A previous study indicated that early-stage PD patients (Hoehn and Yahr 1) showed increased white matter density of the corpus callosum, compared with HV and compared with more advanced stage PD patients (Hoehn and Yahr 2) 38 , in accordance with our present results. At the time of disease onset, the compensation could occur in a limited fashion, resulting in the negative correlation between precuneus FA and age.
The proposed compensation mechanism in the prodromal phase, possibly associated with motor dysfunction, would be supported by reorganization in the posterior part of the brain prevalently, through the midline structure. This may explain why FA in the precuneus, but not the AFD index of PCG-Precuneus bundle, reached statistical signi cance when comparing HV and PD-nonMCI. Finally, although the present observations support and extend previous ndings, DWI acquisition b-values were relatively low and the sample sizes were modest, underscoring the need for replication.
Using DWI data with DTI and HARDI techniques in HV and PD patients, we observed increased FA in the precuneus in PD-nonMCI but not in PD-MCI compared with HV, and change of association between AFD of the posterior midline structure and cognition. In accordance with our previous studies (Nagano-Saito et al., 2016, 2019), we propose that the posterior medial structure is overrecruited, plausibly reorganizing parts of the memory-related diffusion microstructures into executive function-related ones, in PD-nonMCI patients, to possibly compensate for damaged basal ganglia function.

Participants
The participants were 38 PD patients at stages I and II of Hoehn and Yahr (mean age +/-SD, 62,0 +/-5.0 years; range, 53 to 69; 24 male and 14 female) diagnosed by a movement disorder neurologist and who met the UK brain bank criteria for idiopathic PD 68 , and 16 healthy volunteers (HV; mean age +/-SD, 61.8 +/-5.5 years; range, 53 to 72; 7 male and 9 female) without cognitive impairment. All participants underwent a cognitive assessment and MRI session. The majority of participants (PD 35; HV 16) were reported in a previous study 69 , and 31 PD patients were involved in a functional MRI study 23 . All provided informed consent, and the protocol was approved by the Research Ethics Committee of the Regroupement Neuroimagerie Québec. Based on a comprehensive neuropsychological assessment, the PD patients were divided into 2 groups: those with MCI (n = 18) and those who were cognitively intact (non-MCI, n = 20). MCI inclusion criteria were as follows: (1) objective evidence of cognitive decline: performance > 1.5 SD below standardized mean on 2 or more subtests within a cognitive domain ; (2) subjective complaint of cognitive decline from the patient or accompanying person; (3) absence of signi cant decline in daily living activities; and (4) absence of dementia as diagnosed by the evaluating neuropsychologist; consistent with newly proposed guidelines (Level II, comprehensive assessment) for diagnosis of MCI in PD patients by the Movement Disorder Society task force 70 . The HV group also underwent the same neuropsychological assessment and none of them met the criteria for MCI.

Neuropsychological assessment
A screening test, the Montreal Cognitive Assessment 71 , and a comprehensive neuropsychological evaluation was administered before the scanning session. The comprehensive neuropsychological evaluation targeted 5 cognitive domains: attention and working memory, executive functions, language, memory, and visuo-spatial abilities (Supplementary Table 1).

MRI scanning
Participants were scanned at the Institut Universitaire de Gériatrie de Montréal's 3T Siemens TIM MRI scanner. Sessions began with a high-resolution, T1-weighted, 3D volume acquisition for anatomic localization (1 mm 3 , voxel size), followed by a diffusion-weighted image (DWI) acquisition using a 2-D spin-echo EPI sequence, consisting of 64 diffusion-encoding gradients with a b of 700 s/mm 2 , 75 slices (matrix size, 128 x 128 pixels, voxel size, 2 x 2 x 2 mm 3 ).

MRI data preprocessing
The DWI data were processed following the same methods as our previous study (Hanganu et al, 2018).
Brie y, after denoising, up-sampling, and brain-masking, diffusion tensor image (DTI) analyses were applied to create fractional anisotropy (FA), and mean diffusivity (MD) matrices for each participant, and high angular resolution diffusion imaging (HARDI) was applied to create ber orientation distribution function (fODF) models. From fODF models of each participant, tractography of ve million streamlines of the whole brain was generated using a step size of 0.5mm.

DTI matrices
The T1-weignted image (T1WI) was non-linearly co-registered into the DWI b0 image, and the coregistered T1WI was non-linearly transformed into MNI-152 space. Using the transformation parameters, the FA and MD images were resampled into MNI-152 space. A region of interest (ROI) analysis was done. According to our hypothesis, based on our previous studies (Nagano-Saito et al., 2004, 2016, and 2019), four sets of ROIs, one for each hemisphere were located in the precuneus, isthmus, posterior cingulate, and anterior cingulate white matters. The diameter of the ROI was 7mm. These ROIs are shown in Fig. 1

HARDI tractography and apparent ber density
The T1WI image co-registered into the DWI b0 image was segmented into subcortical, cortical, and white matter parcellations, using the Freesurfer package 72 , and thirty-four gyral based regions were parcellated per hemisphere using the Desikan-Killiany atlas 73 . The parcellations of the precuneus were further separated into anterior and posterior subregions. To get the subregions, rst, the MNI-152 template was segmented, using the Freesurfer package. Then, the segmented precuneus on the MNI-152 template was manually divided into anterior and posterior subregions using the precuneal sulcus as the boundary. These subregions were used to extract speci c streamlines, which included the precuneus ROI above. These subregions were then transformed into the individual space, using FSL package, applying the parameter of transformation from the MNI-152 space into the individual space. Using white matter query language (WMQL; Wassermann et al., 2016), we extracted ve bundles, for each hemisphere; 1) including the anterior part of the precuneus, but not the posterior part of the precuneus, and the posterior cingulate (PCG-Precuneus), 2) including the posterior cingulate and the parahippocampal area (PCG-PHG), 3) including the precuneus and the parahippocamal area, but not the posterior cingulate (Precuneus-PHG), 4) including the precuneus and the paracentral area, the primary motor area, but not the posterior cingulate (Precuneus-Cortex), and 5) including the precuneus but not extending to the posterior cingulate, parahippocampal area, or paracentral area (Precuneus-Local). Additionally, 6) one cross-hemisphere bundle was extracted including the both sides of the precuneus via the corpus callosum (Precuneus-Cross). The extracted methods for 1) ~ 4) are summarized in Fig. 1-b. The PCG-Precuneus bundle was set based on our previous study indicating increased hub function in the anterior part of the precuneus and the posterior part of the cingulate gyrus 27 . The PCG-PHG and Precuneus-PHG bundles were set based of our previous studies indicating that hippocampal compensation is used to maintain cognitive abilities in PD patients 26,75 . We also considered the cingulum bundles in the human beings, reported previously 76 .
For each streamline, with fODF models, apparent ber density (AFD) index was estimated using the MRtrix3 package 49 . All tract-speci c bre orientation distribution integrals within the bre voxels belonging to a speci c bundle were added and divided by the mean pathway length resulting in the AFD index 47,63,64 . The AFD index were averaged for corresponding left and right bundles for each of the ve bundles. We did not include Precuneus-striatum bundle, as the number of them were very limited.
Group comparisons, and correlation analysis with diffusion microstructures vs age and cognitive scores With the FA and MD, as well as with the AFD index of each bundle, group comparisons were performed, between the HV and non-MCI PD patients, and between the non-MCI and MCI PD patients. Based on our previous study (Nagano-Saito et al, 2014, 2016 and 2019), correlations were calculated between the diffusion microstructure markers, FA and AFD, and age, Z-scores of cognitive tasks of Brixton Switching test (executive function) 77 and RAVLT (delay-recall list: memory function) 78 in HV and in PD patients collapsing both MCI and nonMCI groups, separately. These three factors (age, Brixton Z-score, and RAVLT Z-score) did not show any signi cant correlations each other (p > 0.1) in HV and in PD, respectively. For the group comparisons and correlation analyses, the signi cance threshold was set at p = 0.05. Because of our strong hypothesis, based on our previous studies (Nagano-Saito et al., 2014, 2016, and 2019), correction for multiple comparison was not applied. Declarations PCG; Posterior cingulate gyrus, PCN; Precuneus, PHV; Parahippocampal gyrus. Bold letters indicate signi cant correlation (P < 0.05), and bold italic letters indicate marginally signi cant correlation (p < 0.1). Figure 5