Although iron-mediated oxidative stress has been proposed as a potential pathomechanism in Parkinson’s disease, the global distribution of iron accumulation in Parkinson’s disease has not yet been elucidated. This study used a new magnetic resonance imaging contrast, quantitative susceptibility mapping, and state-of-the-art methods to map for the first time the whole-brain landscape of magnetostatic alterations as a surrogate for iron level changes in n = 25 patients with idiopathic Parkinson’s disease versus n = 50 matched controls. In addition to whole-brain analysis, a regional study including sub-segmentation of the substantia nigra into dorsal and ventral regions and qualitative assessment of susceptibility maps in single subjects were also performed. The most remarkable basal ganglia effect was an apparent magnetic susceptibility increase—consistent with iron deposition—in the dorsal substantia nigra, though an effect was also observed in ventral regions. Increased bulk susceptibility, additionally, was detected in rostral pontine areas and in a cortical pattern tightly concordant with known Parkinson’s disease distributions of α-synuclein pathology. In contrast, the normally iron-rich cerebellar dentate nucleus returned a susceptibility reduction suggesting decreased iron content. These results are in agreement with previous post-mortem studies in which iron content was evaluated in specific regions of interest; however, extensive neocortical and cerebellar changes constitute a far more complex pattern of iron dysregulation than was anticipated. Such findings also stand in stark contrast to the lack of statistically significant group change using conventional magnetic resonance imaging methods namely voxel-based morphometry, cortical thickness analysis, subcortical volumetry and tract-based diffusion tensor analysis; confirming the potential of whole-brain quantitative susceptibility mapping as an in vivo biomarker in Parkinson’s disease.
Idiopathic Parkinson’s disease is a common progressive degenerative disorder characterized pathologically by proteinaceous aggregates of α-synuclein in the form of Lewy bodies and Lewy neurites. These are thought to occur early in the brainstem, gradually spreading across vulnerable sites in the allocortex and temporal paralimbic cortex before reaching prefrontal and sensory-association isocortex (Braak et al., 2004). It is postulated that projection neurons with disproportionally long, thin and poorly myelinated axons are particularly susceptible to degeneration in Parkinson’s disease (Braak et al., 2004), of which the most studied to date have been the neuromelanin-pigmented dopaminergic neurons of the basal ganglia. It should be noted, however, that the mechanisms that render certain neuronal populations vulnerable to degeneration are unclear; furthermore, the motor dysfunction in Parkinson’s disease related to nigrostriatal degeneration is only one aspect of this multi-faceted disease that includes numerous non-motor manifestations. As such, new in vivo approaches to identify possible pathomechanisms across the entire Parkinson’s disease brain are highly desirable.
One such potential mechanism is oxidative stress due to excessive brain iron accumulation. In the ageing brain, this has been implicated in neurotoxicity, mitochondrial dysfunction and chronic inflammation, as well as promotion of α-synuclein aggregation and dopaminergic alterations (Sian-Huelsmann et al., 2011; Ayton and Lei, 2014). The relevance of iron in the pathogenesis of Parkinson’s disease lies in its capacity to generate free-radical species that may act in concert with α-synuclein to induce Lewy pathology (Ostrerova-Golts et al., 2000; Li et al., 2011), and may catalyse dopamine oxidation reactions that exacerbate the formation of other neurotoxic byproducts (Hare and Double, 2016). The study of brain iron, however, is particularly challenging due to its ubiquitous involvement in numerous biological processes (Ward et al., 2014). Under normal conditions, it is well established that a high proportion of brain iron is bound to ferritin in the redox-inactive ferric state (Hallgren and Sourander, 1958), whereas only small quantities of redox-active ferrous iron are required to ensure a readily available supply of labile iron for cellular metabolism. It is thus largely assumed that all mechanisms involving iron management, i.e. transport, uptake, storage, efflux and redistribution, must be stringently regulated to prevent free ferrous iron from indiscriminately catalysing additional unwanted toxic reactions. To date, however, we lack conclusive evidence incriminating free-iron mediated mechanisms as a primary cause rather than a secondary consequence of neurodegeneration in Parkinson’s disease. Nonetheless, numerous studies have confirmed iron elevation in the substantia nigra pars compacta of post-mortem brains using histochemical (Dexter et al., 1991; Sofic et al., 1991) and X-ray methods (Popescu et al., 2009), and in vivo, using semi-quantitative iron-sensitive techniques such as transcranial sonography and MRI (Groeger and Berg, 2012); an increase that might lead to iron-mediated toxic interactions in Parkinson’s disease. Thus far, however, iron analyses have been limited to regions of interest, and though several brain areas beyond the substantia nigra have been identified as having increased (Griffiths et al., 1999) or decreased (Dexter et al., 1991; Popescu et al., 2009) iron content, the global distribution of iron dysregulation in Parkinson’s disease has not yet been elucidated. The present study used quantitative susceptibility mapping (QSM)—a newly validated, iron-sensitive MRI measure (Langkammer et al., 2012; Zheng et al., 2013; Sun et al., 2015)—to describe the whole-brain distribution of magnetostatic alterations in Parkinson’s disease, and contrasted these results with those from ‘standard’ structural and microstructural MRI markers, none of which, however, probes the integrity of the substantia nigra.
Materials and methods
Twenty-five patients with clinically definite idiopathic Parkinson’s disease according to UK Parkinson’s Disease Brain Bank criteria (Hughes et al., 1992), were recruited from the specialist movement disorder clinic of the Otto von Guericke University Neurology Department. All patients had their clinical assessments in the ‘ON’ state with respect to anti-parkinsonian medication with the exception of one mild, drug-naïve, tremor-dominant individual who wished to delay commencing therapy until symptoms worsened. For imaging comparisons, 50 healthy control subjects—a subset of a larger, previously studied, ageing cohort (Acosta-Cabronero et al., 2016)—were twice-paired for age while keeping a non-significant sex ratio. All controls performed normally on cognitive screening, Mini-Mental State Examination (MMSE; Folstein et al., 1975), and had no history of neurological disease. Group demographic details are summarized in Table 1.
|Control (n = 50)||Parkinson’s disease (n = 25)|
|Sex (male:female)||28:22||20:5 (χ2 = 9.7, NS)|
|Age (years)||63.6 (8.5)||63.6 (8.6)|
|Disease duration (years)||NA||6 (4)|
|MMSE (≤30)||29.0 (0.9)||26.7 (3.0)|
|UPDRS-III (≥0)||NA||16.3 (8.0)|
|Tremor dominant/akinetic-rigid dominant/mixed (n, %)||NA||12/52/36|
|Modified Hoehn and Yahr stage (≥1)||NA||2.2 (0.3)|
|Levodopa equivalent dose (mg)||NA||748 (434)|
|Control (n = 50)||Parkinson’s disease (n = 25)|
|Sex (male:female)||28:22||20:5 (χ2 = 9.7, NS)|
|Age (years)||63.6 (8.5)||63.6 (8.6)|
|Disease duration (years)||NA||6 (4)|
|MMSE (≤30)||29.0 (0.9)||26.7 (3.0)|
|UPDRS-III (≥0)||NA||16.3 (8.0)|
|Tremor dominant/akinetic-rigid dominant/mixed (n, %)||NA||12/52/36|
|Modified Hoehn and Yahr stage (≥1)||NA||2.2 (0.3)|
|Levodopa equivalent dose (mg)||NA||748 (434)|
Where appropriate values are given as mean (SD); not significant (NS, P > 0.01); NA = not applicable; MMSE; ON-medication, UPDRS-III – motor evaluation.
Prior to inclusion in the study, which was approved by the ethics committee at Otto von Guericke University, subjects gave their written informed consent according to the Declaration of Helsinki.
The imaging protocol and processing methods used in this study are identical, except for the addition of diffusion tensor imaging (DTI), to those developed and optimized for a previous ageing study (Acosta-Cabronero et al., 2016), which are summarized below:
MRI measurements were performed on a Siemens Verio 3 T with a standard Siemens 32-channel receive array coil.
Susceptibility weighting was sensitized with a 3D, flow-compensated, spoiled gradient-echo pulse sequence: flip angle was 17°; echo time, 20 ms; receiver bandwidth, 100 Hz per pixel; and repetition time, 28 ms. Matrix size was 256 × 224 × 80, with slices in straight-axial orientation (0.4-mm interspacing and 20% oversampling to reduce crosstalk effects and aliasing artefacts, respectively), and voxel resolution of 1 × 1 × 2 mm3. GRAPPA (generalized autocalibrating partially parallel acquisitions; Griswold et al., 2002), was used with acceleration factor of 2 and 24 reference lines, giving a scan time of 5 min 32 s.
MPRAGE (magnetization-prepared, 3D, rapid gradient-echo) anatomical images were also acquired with the following acquisition parameters: inversion time = 1100 ms; flip angle, 7°; echo time = 4.37 ms; receiver bandwidth, 140 Hz/pixel; echo spacing = 11.1 ms; and repetition time = 2500 ms. Matrix dimensions were 256 × 256 × 192 (0.5-mm interslice gap), 7/8 partial Fourier, and 1 × 1 × 1 mm3 voxel size. GRAPPA was also enabled with acceleration factor of 2 and 24 reference lines. To aid co-registration, the field of view was prescribed consistent with the susceptibility-weighted scan, i.e. with slices in straight-sagittal orientation.
T2-weighted turbo spin echo images were also acquired in the same scanning session, and were visually inspected to ensure vascular pathology—as for standard clinical practice—was not significant in any subject. To facilitate visual inspection, the field of view was aligned at acquisition to the anterior commissure–posterior commissure line. Scan parameters were: flip angle, 150°; echo time = 96 ms; receiver bandwidth, 220 Hz/pixel; turbo factor, 18; 13 echo trains with echo spacing, 9.64 ms; and repetition time = 8160 ms. In-plane matrix was 320 × 320 (resolution: 0.7 × 0.7 mm2) for 45 axial slices (thickness: 3 mm; gap: 0.9 mm); GRAPPA factor of 2 with 51 reference lines.
Diffusion MRI data were acquired using a twice-refocused, single-shot echo-planar imaging pulse sequence: repetition/echo time = 9000/94 ms; matrix, 120 × 120; 63 contiguous slices aligned parallel to the anterior commissure–posterior commissure line; voxel size: 2 × 2 × 2 mm3; 7/8-phase partial Fourier; bandwidth of 1667 Hz/pixel and echo spacing of 0.68 ms. Diffusion gradients were applied along 30 non-collinear directions with two non-zero b-values (b = 700 and 1000 s/mm2), and 12 interleaved reference scans. Parallel imaging was enabled (GRAPPA, acceleration factor of 2 and 38 reference lines) for a total scan time of 11 min 15 s.
A thin pillow was placed on the base of the coil surrounding the sides and the back of the head to minimize motion and increase intersubject reproducibility in positioning.
Structural MRI statistical analyses
VBM was performed in SPM12b v6080 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12) with default settings. Following ‘unified segmentation’ and ‘modulation’, grey matter segments were smoothed with an 8 mm full-width at half-maximum Gaussian kernel prior to statistical analysis—a two-sample (Parkinson’s disease versus control) t-test controlling for total intracranial volume; these were determined using a previously validated method (Pengas et al., 2009).
Cortical thickness analysis
A standard whole-brain routine for vertex-wise surface reconstruction, inflation, smoothing and statistical group analysis of cortical thickness was carried out in the FreeSurfer v5.3.0 framework (http://surfer.nmr.mgh.harvard.edu).
Left and right thalamus, caudate nucleus, putamen, globus pallidus, hippocampus and amygdala were segmented automatically from radio-frequency bias-corrected MPRAGE images using the FIRST algorithm (Patenaude et al., 2011)—an automated surface-aware method that incorporates prior anatomical knowledge (included in FMRIB’s software library, FSL v5.0.9). FIRST-derived region of interest volumes were normalised by total intracranial volume using a well-established covariance method (Jack et al., 1989). Finally, 12 two-tailed Wilcoxon rank-sum tests of equal medians were performed to assess for group differences in each region of interest.
Diffusion tensor imaging analysis
Diffusion MRI datasets were processed with a standard procedure in FSL v5.0.9 (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FDT). DTI metrics [fractional anisotropy (FA) and mean diffusivity (MD) calculated from weighted least-squares tensor fits] were introduced into the tract-based spatial statistics (Smith et al., 2006) v1.2 routine with default settings, which generated a mean FA skeleton (default FA threshold set to 0.2) to assess white matter tracts, i.e. it did not include the substantia nigra or other grey matter regions. Finally, whole-brain group contrasts (Parkinson’s disease versus controls) for FA and MD were inferred using Randomise (Winkler et al., 2014) v2.9 in FSL with 5000 permutations of the data and default threshold-free cluster enhancement (Smith and Nichols, 2009) settings, i.e. 2D optimization for tract-based DTI analysis.
Multi-channel complex data were combined using a modified adaptive algorithm (Walsh et al., 2000). Combined phase images were then unwrapped with a continuous Laplacian approach (Schofield and Zhu, 2003), and the local field was revealed through global extraction of the background field using the spherical mean value filtering method (Schweser et al., 2011) with sphere radius of 5 mm. Finally, susceptibility maps were estimated with the morphology-enabled, non-linear dipole inversion formulation (Liu et al., 2013) solved by the conjugate gradient method on iterative blocks of 500 operations nested in a Gauss-Newton loop with stopping tolerance ratio for the outer loop set to 0.01. For consistency with our previous ageing study (Acosta-Cabronero et al., 2016) using the exact same MRI scan and QSM routine, the Lagrangian multiplier was set to λ = 1000. Our previous study also suggested reference normalization is only a small adjustment relative to ageing effects. In this study, thus, to avoid making assumptions about areas being spared in Parkinson’s disease, QSM values were not referenced.
QSM spatial standardization
For template creation, radio-frequency bias corrected MPRAGE images were spatially normalized using the ‘Greedy-SyN’ approach in ANTs v2.1 (http://stnava.github.io/ANTs) with a maximum of 90 × 30 × 30 multi-resolution iterations and template-update step size set to 0.1 mm. Rigid-plus-affine initial alignment was followed by six full runs of the above routine. Subsequently, bias-corrected magnitude images were affine co-registered to their corresponding MPRAGE volume. QSM spatial standardization was achieved through the warp composition of the above transformations and third-order b-spline interpolation. An average template was then calculated from all warped QSMs.
Whole-brain QSM statistical analyses
Analyses were performed both for signed and absolute QSM. The absolute value was taken to QSM to improve statistical conditioning for whole-brain analysis (Betts et al., 2016) and to ameliorate the spurious impact of blooming effects, i.e. residuals in the vicinity of steep magnetic susceptibility gradients. Smoothing is a requirement in this context to correct for co-registration errors and other imperfections, thus a 3D Gaussian kernel [standard deviation (SD): 3 mm] was applied, followed by a previously proposed smoothing-compensation strategy (Betts et al., 2016). Whole-brain (Parkinson’s disease versus control) permutation analysis was performed with Randomise v2.9 and threshold-free cluster enhancement (with ‘-T’ settings, http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise) in FSL. Significant clusters were inferred from 20 000 data permutations and were reported at a family-wise error (FWE) corrected threshold level of 0.05. The group comparison was performed in two stages: first, we tested the null-hypotheses that age and sex did not co-vary with group behaviours, which returned a widespread distribution of strong positive age effects, whereas the same null hypothesis for sex could not be rejected. The second (reporting) stage, therefore, tested whether age-adjusted QSM group medians were equal. Finally, to assess whether QSM might predict disease severity, permutation-based whole-brain QSM regression analyses versus severity scores, i.e. Unified Parkinson’s Disease Rating Scale (UPDRS-III) and MMSE, were also performed.
Subsequently after analysis, the QSM template and statistical maps were warped into MNI152 space (Montreal Neurological Institute, McGill University, Canada) using ANTs tools. Affected brainstem structures were discerned using Duvernoy’s atlas—the upper pons was identified axially on the standardized QSM template across z = −28 mm (in MNI coordinates); high pons at the level of the nucleus coeruleus, across z = −25 mm; high pons at the level of the lateral lemniscus, across z = −22 mm; mesencephalon at the level of the inferior colliculus, across z = −19 mm; mesencephalon at the level of the inferior pole of the red nucleus, across z = −13 mm; and mesencephalon at the level of the red nucleus, across z = −10 mm.
Regional QSM statistical analysis
The primary objective of the present study was to map the whole-brain landscape of QSM alterations in Parkinson’s disease; post hoc regional analyses in this context, therefore, aimed at providing additional insight into the nature of such alterations. Median QSM values were extracted bilaterally from the following structures: caudate nucleus, putamen, globus pallidus, hippocampus, amygdala and thalamus—inferred automatically from the study-wise MPRAGE template using FSL-FIRST. In addition, QSM data were also extracted from previously validated (Acosta-Cabronero et al., 2016) manual segmentations of the red nucleus, substantia nigra and dentate nucleus. Further segmentation of the substantia nigra is particularly complex because its two main substructures—dorsally, pars compacta and ventrally, pars reticulata—overlap (Lehericy et al., 2014). Neuromelanin-rich dopaminergic neurons of the dorsal substantia nigra, however, possess differential spin-lattice relaxation properties resulting in T1 shortening effects. Such effects are typically too subtle for reliable identification in single subjects but they can be resolved on state-of-the-art T1-weighted templates. Figure 1 illustrates the nigral subdivision into dorsal (hyperintense) and ventral (hypointense) tiers. Both subregions were traced three-dimensionally using FSL’s image viewer (FSLView). Finally, all subcortical regions of interest were eroded by convolution with a 1-mm radius spherical kernel and were visually inspected on each template-warped QSM to ensure partial-volume contamination due to warping errors was not significant.
In light of the whole-brain results, a set of cortical regions of interest were also examined; these were defined from the gyral-based Desikan-Killiany-Tourville digital atlas (Klein and Tourville, 2012) overlaid onto the statistical map of QSM group differences. The OASIS-30 template and OASIS-TRT-20 joint fusion atlas (in OASIS-30 space) were obtained from Mindboggle’s repository (http://www.mindboggle.info/data). The study-wise to OASIS-30 space non-linear warp field was calculated with a deformable b-spline co-registration routine in ANTs (‘antsRegistrationSyN’). Desikan-Killiany-Tourville labels were brought into the study space using the inverse of such transformation and nearest-neighbour interpolation. Finally, to minimize partial-volume contamination, each cortical region of interest was intersected with a study-wise grey matter mask—the control-average grey matter segment binarized at a grey matter density cut-off of 0.5. Grey matter probability segments were inferred from anatomical MPRAGE images using SPM12 with default ‘unified segmentation’ settings, and were then spatially normalized using ANTs template transforms. The pattern of QSM involvement in the Parkinson’s disease cortex led to the selection of six regions of interest: four putative and two control regions. The former were lateral occipital, middle temporal, posterior parietal (including inferior and superior parietal lobules plus precuneus regions), and rostral middle prefrontal cortex. Pre- and postcentral gyri were selected as control regions.
Interhemispheric differences were probed using Wilcoxon rank-sum tests, which revealed that left/right measurements were comparable across patients and controls both in the cortex and in deep grey matter structures (all P > 0.15). Thus, to improve measurement stability, median region of interest values were averaged across hemispheres and Pearson age-correlation tests were performed; these, together with the age effects observed in stage 1 of the whole-brain analysis, confirmed that the present age range is sufficiently large to modulate QSM behaviours. Consequently, prior to computing Parkinson’s disease versus control Wilcoxon rank-sum statistics, QSM values were age-corrected using the covariance method (Jack et al., 1989). Finally, Pearson correlation tests were performed to test the null hypotheses: QSM in patients does not covary with UPDRS-III or MMSE. All regional tests were performed two-tailed, and their resulting P-values were Bonferroni-adjusted to minimize multiple testing effects.
Except where stated otherwise, processing tasks (including QSM reconstruction and regional data analysis) were prototyped and executed in the MATLAB (R2012a) environment (The Mathworks Inc., Natick, MA, USA).
The structural measures—subcortical volumetry, voxel-based morphometry, cortical thickness analysis and tract-based DTI statistics—were all negative in the Parkinson’s disease versus age-matched control group comparison, i.e. they failed to reject the null-hypothesis of equal means or medians at an uncorrected statistical threshold of P < 0.01.
Whole-brain QSM study
In contrast, the whole-brain QSM group results (Figs 2 and 3) revealed widespread absolute susceptibility increases (FWE corrected P < 0.05) in Parkinson’s disease that involved brainstem, cerebral and cerebellar structures. Largely bilateral abnormalities were identified in the rostral pons (z = −34 to −25, y = −25 and x = −8 to 15 mm, including the site of the basal pontis and superior areas of the pontine tegmentum); in the superior cerebellar peduncle; caudal mesencephalon (z = −22 to −13, y = −16, x = −8 to 8 mm, including dorsal substantia nigra and inferior midbrain tegmental areas); temporal lobe structures (z = −43 to 11, y = −58 to −16, x = −44 and 44 mm, including portions of superior, middle and inferior temporal, fusiform, parahippocampal, entorhinal regions and the hippocampal head, with a marked right bias and largest cluster centred around the intermediate part of the right temporal lobe, z = −19 to−13 mm, y = −25 to −16, x = 44 mm); occipital regions (z = −13 to −53, y = −87 to −58, x = −44 to 44 mm, most intense in left lateral areas but also involving lingual, pericalcarine and cuneus cortices); posterior parietal regions (z = 17 to 68, y = −70 to −58, x = −15 to 15 mm, also left lateralized including large sections of the inferior and superior parietal lobules, precuneus and isthmus of the cingulate cortex); and prefrontal areas (z = −25 to 50, y = 24 to 50, x = −44 to 44 mm, most markedly across the rostral middle prefrontal cortex but also touching on orbitofrontal areas and the pars triangularis of the inferior frontal lobe). Although more patchy, some regions in the insula and cerebellar cortex were also involved. In contrast, QSM in primary sensory-motor fields of the precentral and postcentral gyri, most of the remaining insula, extra-nigral nodes of the basal ganglia and diencephalic structures did not return statistical differences between groups. The opposite behaviour (i.e. reduced absolute susceptibility in Parkinson’s disease) was not statistically significant at a whole brain-corrected level. Significant clusters for increased signed QSM—albeit less extensive—were highly co-localized with those for absolute QSM (not shown).
QSM regression analyses versus symptom severity measures, UPDRS-III or MMSE, did not return any significant correlation cluster at the whole-brain level (PFWE < 0.05).
Single-subject QSM examination
Closer inspection of midbrain and cerebellar nuclei demonstrated QSM abnormalities that were apparent on visual inspection of individuals (Fig. 4). The most striking effect—consistent with the whole-brain results—was observed in the dorsal substantia nigra region. Particularly remarkable was the disappearance in most Parkinson’s disease cases of two lateral hyperintense pockets that can be readily identified in the caudal region of most healthy subjects (Fig. 4). Increased susceptibility, however, was also noticeable in the ventral substantia nigra—chiefly in rostral areas. The cerebellar dentate nucleus (Fig. 4) was also abnormal in Parkinson’s disease but in the opposite direction with several subjects presenting very little or no contrast on QSM.
Regional QSM study
In the control group, strong QSM-age dependencies (i.e. statistically significant Pearson correlations) were identified in several subcortical and cortical regions of interest including putamen, middle temporal and prefrontal structures (Fig. 5). Group statistics were thus calculated with age-corrected absolute QSM data (Fig. 6), which returned strong alterations in the substantia nigra—mostly driven by increased susceptibility in the dorsal tier though also apparent in ventral regions—lateral occipital, posterior parietal and rostral middle prefrontal cortical regions with additional, though less pronounced, increases in the middle temporal gyrus and hippocampus. The cerebellar dentate nucleus region of interest again showed the opposite pattern with significantly decreased susceptibility. Although effect sizes were overall smaller, regional behaviours for signed QSM (not shown) were highly concordant with those reported in Fig. 5 and 6 for absolute QSM.
In general, consistent with the whole-brain analyses, regional QSM values did not correlate with UPDRS-III or MMSE (
Using a whole-brain approach for the first time to map the landscape of magnetostatic alterations in Parkinson’s disease, widespread QSM changes across the brainstem and cortex were revealed (Fig. 2). In the brainstem, abnormalities were identified in the rostral pons, including pyramidal tracts and pontine tegmental areas co-localized with the site of the locus coeruleus; the superior cerebellar peduncle; and caudal mesencephalon—seemingly spreading across pars compacta/ventral tegmental substantia nigra subregions and midbrain tegmental areas, possibly also including dorsal raphe and oculomotor nuclei (Fig. 3). Parts of the temporal paralimbic, prefrontal and occipito-parietal cortex and, less markedly, insular and cerebellar areas were also involved (Fig. 2). The striatum as well as primary motor and somatosensory fields, in contrast, were relatively spared.
Superior pontine and caudal mesencephalic Parkinson’s disease involvement (Fig. 3) was an expected result in light of the many past studies that reported extensive Lewy body pathology and loss of pigmented neurons in brainstem nuclei such as locus coeruleus, and substantia nigra pars compacta (Hirsch et al., 1988; Braak et al., 2004). Past reports have shown that iron overload in Parkinson’s disease substantia nigra can be readily identified by the absence of a bilateral pocket of low paramagnetism located in the dorsolateral region—a caudal substantia nigra subregion commonly designated as substantia nigra nigrosome-1— detectable both with 3 T (Schwarz et al., 2014) and 7 T MRI (Kwon et al., 2012; Blazejewska et al., 2013; Lehericy et al., 2014). The nigrosome-1 is a large cluster of melanin-rich dopaminergic neurons that do not require as much iron for their normal activity as their adjacent neighbours (Lehericy et al., 2014). In Parkinson’s disease, however, nigrosome-1 dopaminergic neurons progressively scavenge excess iron (Good et al., 1992; Oakley et al., 2007) and degenerate (Fearnley and Lees, 1991; Damier et al., 1999); observations consistent with the present study in which nigrosome-1 of Parkinson’s disease patients appeared overloaded with iron, which might be more readily detectable on coronal views (Fig. 4). To our knowledge—although this was not our primary objective, hence image resolution was suboptimal for this purpose—this is the first report of nigrosome-1 visualization with QSM. In future, quantitative susceptibility evaluations of nigrosome-1 load in parkinsonism with higher resolution MRI prescriptions optimized for midbrain imaging are warranted. On a different note, it is also worth highlighting that substantia nigra free water levels estimated with a diffusion MRI bi-tensor model have recently been proposed as a novel biomarker in Parkinson’s disease (Ofori et al., 2015)—a selective pattern of progressive free water elevation that stands in stark contrast to the more widespread involvement found for other extra-pyramidal disorders (Planetta et al., 2016). Combining QSM and free water levels—as surrogates for iron content and cellularity, respectively—in future studies may, therefore, offer added value in exploring nigral integrity.
In contrast to the substantia nigra, only a non-significant QSM increase was observed in the globus pallidus of Parkinson’s disease patients. Notably, post-mortem reports of pallidal iron content are inconclusive; while some studies have shown an increase in Parkinson’s disease (Chen et al., 1993), others showed normal levels (Riederer et al., 1989; Sofic et al., 1991) or even reductions (Dexter et al., 1991). Differences across experimental procedures and quantification techniques could explain the discrepancy (Hare et al., 2012), though a key possibility might be inferred from one study (Griffiths et al., 1999) that revealed iron content decreased in pars interna and increased in pars externa in Parkinson’s disease. It is conceivable, therefore, that these effects could cancel each other when studying the whole globus pallidus with regions of interest or by the effect of spatial smoothing (a necessary step in co-registration-based analyses of the whole brain). There may also be a temporal dimension that could explain the discordant pallidal iron results in Parkinson’s disease: QSM elevation has been reported in moderate—mean UPDRS-III = 44.5 ± 13.1—but not early disease stages—mean UPDRS-III = 22.8 ± 11.7 (Guan et al., 2016)—consistent with the present negative pallidal results in a Parkinson’s disease cohort with a mean UPDRS-III of 16.3 ± 8.0, therefore suggesting that pallidal iron loading may be a relative late disease feature.
Another relevant negative in the present study was the lack of significant alterations in the normally iron-rich striatum. This negative is compatible with many post-mortem (Riederer et al., 1989; Dexter et al., 1991; Sofic et al., 1991) and in vivo region of interest studies using the apparent proton transverse relaxation rate i.e. R2* (Martin et al., 2008; Du et al., 2012; Barbosa et al., 2015; He et al., 2015; Murakami et al., 2015), homodyne-filtered phase mapping (Han et al., 2013; Kim and Lee, 2014) or QSM (Barbosa et al., 2015; He et al., 2015; Murakami et al., 2015). This observation finds resonance with the view that dopamine responsiveness in Parkinson’s disease relates to the preserved striatal targets of the degenerating nigrostriatal dopaminergic neurons, which could also lead to the prediction that Parkinson-plus diseases such as progressive supranuclear palsy and multiple system atrophy might show additional QSM changes in the corpus striatum (Dexter et al., 1991; Han et al., 2013).
An interesting observation in the age-corrected region of interest analysis was a significant QSM reduction in the cerebellar dentate nucleus (Fig. 6). This structure has typically not been considered in post-mortem Parkinson’s disease analyses, but there is at least one precedent where a marked iron reduction was found in the dentate nucleus (Riederer et al., 1989), though the observation was not discussed further. In health, the cerebellar dentate is one of the most iron-rich structures in the brain (Hallgren and Sourander, 1958) and therefore returns high QSM values (Acosta-Cabronero et al., 2016). In the elderly, however, dentate susceptibility estimates become dispersed—in comparison to young adults—including a small minority of cases with very low values (Acosta-Cabronero et al., 2016). It appears that low values become very prevalent in Parkinson’s disease. The reduction in some individuals was extreme in the present study (Figs 4 and 5); whether this might have phenotypic significance should be a topic for future research. The patient cohort in this study was predominantly akinetic-rigid Parkinson’s disease (only three tremor-dominant cases), thus decreased QSM appears to be concordant with a recent study that has hinted at dentate QSM being associated with phenotypic differences in that tremor-predominant cases returned increased values, whereas there was a trend to reduced susceptibility in akinetic-rigid Parkinson’s disease (He et al., 2016). One other study found no changes in the dentate in Parkinson’s disease, although the MRI acquisition was very brief and relatively low resolution, therefore risking loss of sensitivity (Guan et al., 2016). There is growing interest in the role of the cerebellum in Parkinson’s disease with the discovery of disynaptic pathways linking cerebellum to basal ganglia (Bostan et al., 2013) and with numerous studies highlighting cerebellar hyperactivation during motor tasks or increased resting functional connectivity—reviewed elsewhere (Wu and Hallett, 2013). Physiological changes such as these may be secondary or compensatory but the present finding of apparent loss of iron content in the dentate points to local cerebellar alterations—a finding that should be investigated further histologically.
An advantage of the present MRI approach over other established means of probing brain iron such as histochemical analysis, staining methods or transcranial sonography is that MRI can image the whole brain in a few minutes, thus making it an ideal tool to characterize the cortical pattern of iron deposition in vivo. The present whole-brain results (Fig. 2)—the first of their kind in Parkinson’s disease—revealed a distribution of QSM abnormalities in the cortex that was in close agreement with the suspected landscape of Lewy inclusion pathology (Hughes et al., 1992; Braak et al., 2004) and glucose hypometabolism in autopsy-confirmed Lewy body disease (Minoshima et al., 2001). In clinical stages, these predict concurrent involvement of limbic and paralimbic structures of the temporal lobe, prefrontal cortex and isocortical sensory-association areas with relative sparing of primary sensory areas—a distribution highly consistent with the QSM results in this study. Such concordance between markers of pathology, iron accumulation and neuronal loss does not yet clarify whether iron is the cause or consequence of Parkinson’s disease degeneration, it cannot resolve either whether iron dysregulation occurs through uptake, storage or transport mechanisms, or whether excess iron is in free form or bound to macromolecules, but reinforces the notion that in Parkinson’s disease there is a deleterious cocktail of Lewy inclusions and iron elevation across the isocortex. The causative theory that iron accumulation is an upstream event to Lewy pathology, however, could be entertained. It could be argued that although the distribution of cortical increase in QSM in the present study shows homology with the known distribution of glucose hypometabolism and Lewy pathology once this pathology has reached the neocortex, it does not follow that pathological burden causing neuronal dysfunction and iron accumulation are contemporaneous events. In fact, given that the current cohort generally had preserved cognition, it is possible that the cohort were not yet significantly affected by neocortical Lewy pathology—symptomatic Parkinson’s disease is thought to begin with Braak stage 4 α-synuclein pathology (Braak et al., 2003) in which the neocortex is yet to be affected. There is no way at present to directly investigate this in vivo because imaging techniques that visualize α-synuclein are not yet available. Though it would be intriguing to investigate, post-mortem, whether increases in neocortical iron concentration are detectable in Parkinson’s disease brains without neocortical Lewy pathology (i.e. Braak stage <5). Nonetheless, even if iron imbalance was an indirect consequence of other primary causes, a mirroring distribution would still be highly relevant for the purpose of disease monitoring as it would mean iron mapping could act as proxy signature of cortical pathology.
The QSM findings stand in stark contrast to the lack of significant findings using conventional MRI methods namely voxel-based morphometry, cortical thickness analysis, subcortical volumetry and tract-based DTI analysis. The negative findings with these methods were not unexpected; for instance, even in dementia with Lewy bodies, structural imaging to detect atrophy is largely unremarkable (Whitwell et al., 2007) while previous tract-based DTI studies have also been negative (Worker et al., 2014). Though notably, diffusion MRI using a bi-tensor model, as previously discussed, could offer a sensitive marker to study the substantia nigra (Ofori et al., 2015), which is usually excluded from DTI analyses due to low diffusion anisotropy and other reliability concerns. The absence, however, of consistent findings with standard structural MRI/DTI methods highlights the potential of probing tissue iron across the whole brain with QSM as a useful biomarker in Parkinson’s disease.
Note on QSM interpretations
A broad assumption in this study was that QSM alterations in Parkinson’s disease reflect changes in iron content. Such interpretation, however, merits further discussion on the basis that the intrinsic magnetostatic properties of human brain tissue are not be driven by iron alone. Copper(II) and manganese compounds are magnetically reactive and could perturb the susceptibility measurement; although with concentrations 15 and 50 times lower, respectively, than iron in the substantia nigra (Krebs et al., 2014), it is highly improbable they could have played a significant role in the present results. Furthermore, post-mortem studies have suggested that copper levels are slightly reduced in the substantia nigra in Parkinson’s disease (Dexter et al., 1991; Davies et al., 2014), while manganese levels are largely unaltered (Dexter et al., 1991). Abnormally high levels of zinc, in contrast, have been detected in several post-mortem Parkinson’s disease brain regions. Zinc too, however, is scarce relative to iron content and does not have an unpaired electron configuration; consequently, the impact of zinc on QSM can also be discarded. Diamagnetic alkaline earth metals, calcium and magnesium, are more abundant in human brain tissue than the aforementioned transition metals, with regional concentrations in post-mortem tissue similar—often greater—than those for iron (Krebs et al., 2014). Conceivably, the effect of diamagnetic mineralization could reduce QSM sensitivity to iron-related effects. Signed QSM, however, although less sensitive overall than absolute QSM presumably due to its greater vulnerability to neighbouring contamination (Betts et al., 2016), returned increased susceptibility—consistent with increased iron content—across all affected areas. The exception was the dentate nucleus meaning that, in addition to a loss of iron, micro-calcifications could explain our QSM observations, and should thus be considered as a plausible theory in future post-mortem studies.
Three post-mortem validation studies that, to date, have been carried out to investigate the relationship between QSM and tissue iron concentrations (Langkammer et al., 2012; Zheng et al., 2013; Sun et al., 2015) should also be considered in interpreting the present study. The conclusion from these studies was that ferric iron, presumably stored in glial ferritin, is the dominant source of magnetic susceptibility in metal-laden deep brain nuclei. In the cerebral cortex, however, it has been shown that the myeloarchitecture can modulate local tissue susceptibility (Fukunaga et al., 2010; Stüber et al., 2014). It is therefore plausible that demyelination processes—yielding a net increase in paramagnetism—could drive QSM effects in areas of relatively low iron concentration, e.g. prefrontal cortex. Both studies, however, demonstrated that ferritin iron co-localized with intracortical myelin is the dominant source of local susceptibility—approximately two-thirds of the overall contrast proportion in grey matter (Stüber et al., 2014)—suggesting, therefore, only a massive myelin-driven susceptibility effect could perturb the measurement. In future, a recently proposed multivariate model could be applied to disentangle iron/myelin contributions using multi-contrast information (Stüber et al., 2014). All of the above, nonetheless, suggests the present in vivo results are most likely driven by changes in iron content, although QSM’s inability to provide information about the exact cellular distribution or the valency of iron ions means the present results cannot discern whether QSM alterations in Parkinson’s disease are primarily driven by excess free iron, greater ferritin numbers, greater iron loading within ferritin or neuromelanin, haemosiderin burden, or whether they represent neuronal or glial iron accumulation.
Finally, other technical limitations must be highlighted. The medulla oblongata has been postulated as one of the earliest sites of Lewy pathology in the Parkinson’s disease cascade (Braak et al., 2004), though here it did not return a significant QSM alteration. A plausible explanation for such absence could be that baseline iron levels are very low in medullar nuclei, thus small disease-related alterations—potentially large in relative terms—may lie undetected with current technology. In contrast, one should be careful of not over-interpreting whole-brain results in small anatomical structures. Prior to performing statistics, QSM data were spatially smoothed to account for possible errors in spatial normalization, and although a smoothing-compensation strategy was applied, one should assume a degree of spatial uncertainty for the resulting clusters. This is particularly relevant to Braak stage 1–2 (Braak et al., 2004) brainstem nuclei such as the motor nucleus of the vagus nerve, locus coeruleus or raphe nucleus that are probably too small to be reliably assessed using the present whole-brain methodology. On a final note, we must highlight that in the present study we were unable to establish QSM correlates of symptom severity measures, i.e. UPDRS-III and MMSE. Given the narrow spread of severity scores for the patient cohort under investigation (particularly for MMSE, see
This study revealed a spatial distribution of QSM alterations in Parkinson’s disease highly consistent with the landscapes of glucose hypometabolism and Lewy pathology that are known to emerge as the disease evolves. In contrast to QSM, no significant effects were observed with standard structural MRI and microstructural DTI measurements. These results, therefore, demonstrate the relevance of mapping the biochemical environment of the whole Parkinson’s disease brain in vivo, and provide new insights into the behaviour of QSM as a disease biomarker. MRI’s safe and non-invasive nature means, in addition, that QSM might be suitable for longitudinal monitoring in clinical trials.
We are very grateful to all the participants who made this research possible. We also thank Mrs Urte Schneider, Ms Sabrina Rädler, Ms Deike Hartmann and our local radiographers—Mrs Kerstin Möhring and Mrs Ilona Wiedenhöft—for their invaluable help with subject recruitment, data collection and their dedicated efforts in scanning standardisation and quality assurance.
German Center for Neurodegenerative Diseases (DZNE).
diffusion tensor imaging
Mini-Mental State Examination
quantitative susceptibility mapping
Unified Parkinson’s Disease Rating Scale