Susceptibility MRI correlates with glial density and tau in the substantia nigra pars compacta

Abstract

Background:

Susceptibility MRI is sensitive to iron-related changes in the substantia nigra pars compacta (SNc), the key pathologic locus of parkinsonisms. It is unclear, however, if iron deposition in the SNc is associated with its neurodegeneration.

Objective:

To test whether susceptibility MRI metrics in parkinsonisms are associated with SNc neuropathologic features of dopaminergic neuron loss, gliosis, and α-synuclein and tau burden.

Methods:

This retrospective study included 27 subjects with both in vivo MRI and postmortem data. Multi-gradient echo imaging was used to derive the apparent transverse relaxation rate (R2*) and quantitative susceptibility mapping (QSM) in the SNc. Archived midbrain slides that were stained with hematoxylin and eosin, anti-α-synuclein, and anti-tau were digitized to quantify neuromelanin-positive neuron density, glial density, and the percentages of area occupied by positive α-synuclein and tau staining. MRI-histology associations were examined using Pearson correlations and regression.

Results:

Twenty-four subjects had postmortem parkinsonism diagnoses (Lewy body disorder, progressive supranuclear palsy, multiple system atrophy, and corticobasal degeneration), two only had Alzheimer’s neuropathology, and one only exhibited mild atrophy. Among all subjects, both R2* and QSM were associated with glial density (r-values ≥ 0.67; P-values < 0.001) and log-transformed tau burden (r-values ≥ 0.53; P-values ≤ 0.007). Multiple linear regression identified glial density and log-transformed tau as determinants for both MRI metrics (R²-values ≥ 0.580; P-values < 0.0001). Neither MRI metric associated with neuron density or α-synuclein burden.

Conclusions:

R2* and QSM are associated with both glial density and tau burden, key neuropathologic features in the parkinsonism SNc.

Introduction

Parkinsonisms are characterized by tremor, rigidity, bradykinesia, loss of postural control, and gait dysfunction. Whereas these features are clinical hallmarks of Parkinson’s disease, they also are found in neuropathologically distinct atypical parkinsonisms such as progressive supranuclear palsy, multiple system atrophy, and corticobasal degeneration.¹ Overlapping motor presentations among these disorders are attributed largely to the loss of dopaminergic input to the basal ganglia.² Dopamine deficiency stems from the loss of pigmented neurons in the substantia nigra (SN), specifically the subregion known as the pars compacta (SNc) where the neurons are populated most densely.³

Postmortem examinations have shown that dopaminergic neuron loss in the SNc is a shared neuropathologic feature among parkinsonisms.⁴ Also shared are findings of reactive gliosis and proteinopathy. Proteinopathy is used to differentiate parkinsonisms as they differ in both the type and distribution of protein aggregation. Specifically, Parkinson’s disease is characterized by α-synuclein-positive neuronal inclusions known as Lewy bodies. Multiple system atrophy is also an α-synucleinopathy but is defined by α-synuclein-positive neuronal and glial cytoplasmic inclusions. Progressive supranuclear palsy and corticobasal degeneration both are characterized by tau-positive neurofibrillary tangles and neuropil threads.⁴

Overlapping motor presentations among parkinsonisms continue to challenge early and accurate clinical diagnoses.⁵ There consequently has been continued interest in capturing neuropathologic features using in vivo imaging techniques, both for the earlier detection of neurodegenerative changes and the differential diagnosis of parkinsonisms.⁶ Among the many imaging modalities examined, susceptibility MRI (sMRI) has attracted considerable attention as it is particularly sensitive to magnetic susceptibility differences in the SN and SNc that predominately reflect brain iron accumulation.^{7, 8} Excessive iron deposition is a well-known feature of many neurodegenerative conditions, including parkinsonisms.^{9, 10} Iron accumulation in the parkinsonism SN first was reported by Dexter et al. (1987)¹¹ and Sofic et al. (1988)¹² using postmortem tissue samples. Since then, in vivo sMRI studies have generated findings consistent with higher iron content in the parkinsonism SN and SNc.¹³ More recently, our group has used sMRI to show that SN iron is significantly higher in later-stage Parkinson’s disease patients versus earlier-stage patients.¹⁴ Using longitudinal data, we also have shown that SN iron increases over the disease course.¹⁵

In previous studies, susceptibility differences in the brain often have been captured by the apparent transverse relaxation rate (R2*) and quantitative susceptibility mapping (QSM).^{7, 8} As iron has been implicated in various pathophysiologic processes in the brain – including oxidative stress, DNA damage, abnormal modifications of macromolecules,¹⁶ and interactions with both α-synuclein and tau^17–19 – we hypothesize that R2* and QSM may capture key neuropathologic features in the parkinsonism SN. Our original pilot study of 14 parkinsonism subjects yielded some preliminary data consistent with this hypothesis, but was limited by sample size and semi-quantitative histopathologic grading methods.²⁰ In the present study, we focused specifically on the SNc and re-examined sMRI-histology associations using a quantitative approach to evaluate R2* and QSM associations with neuron density, glial density, and α-synuclein and tau burden in an expanded cohort of 27 subjects.

Methods

Study subjects

This study included 27 subjects from a longitudinal cohort established under the Parkinson’s Disease Biomarker Program.²¹ Subjects were selected based on the availability of in vivo multi-gradient echo MRI and postmortem neuropathology data as of January 2022. Clinical and final neuropathology diagnoses were made, respectively, by a movement disorder specialist and neuropathologist according to published criteria.^{1, 4, 22, 23}

Pathology diagnoses of Lewy-type α-synucleinopathies (e.g. Parkinson’s disease, Parkinson’s disease with dementia, dementia with Lewy bodies) were condensed under a diagnosis of Lewy body disorder,²⁴ and staged by distribution of Lewy body pathology (brainstem predominant, limbic/transitional, neocortical/diffuse).²⁵ Multiple system atrophy diagnoses were further subtyped as striatonigral degeneration versus olivopontocerebellar atrophy based on the predominant neuropathologic phenotype. Alzheimer’s disease neuropathologic change was graded according to the National Institute on Aging-Alzheimer’s Association guidelines for ABC score and severity.²³

For parkinsonism subjects, disease duration was defined as the number of years between date of first diagnosis and date of visit. For all subjects, an MRI-pathology interval was defined as the months spanning a subject’s MRI scan and death. Written informed consent was obtained from all subjects or their legal representatives in accordance with the Declaration of Helsinki, and all data collection was approved by the institutional review board at the Pennsylvania State University.

MRI acquisition and post-processing

Brain MRIs were obtained using a 3-Tesla MRI system (Magnetom Trio; Siemens, Erlangen, Germany). For image registration purposes, T2-weighted (T2w) images were acquired using a fast spin-echo sequence: repetition time (TR)/echo time (TE) = 2500/316 ms, field of view (FoV) = 256×256 mm, matrix = 256×256, slice thickness = 1 mm (no gap), slice number = 176. Multi-gradient-echo images (for R2* and QSM images) were acquired using the following acquisition parameters: 6 TEs ranging from 7 to 47 ms at an equal interval of 8 ms, TR = 54 ms, flip angle = 20°, FoV = 256×256 mm, matrix = 256×256, slice thickness = 1.5 mm (no gap), slice number = 64. Six magnitude images taken from the multi-gradient echo were aligned by affine registration and averaged to generate a mean magnitude image. R2* images were generated from a voxel-wise, nonlinear, Levenberg-Marquardt algorithm to fit the monoexponential function: $s=s_0{e}^{-\mathrm{T}\mathrm{E}\times \mathrm{R}2\mathrm{*}}$. QSM images were generated using morphology-enabled dipole inversion with a nonlinear formulation.^{26, 27} Both R2* and QSM values were calculated in MATLAB (MathWorks, Natick, Massachusetts).

SNc segmentation on MRI

Dopaminergic neurons of the SNc are pigmented by neuromelanin, a byproduct of dopamine synthesis. The paramagnetic properties of neuromelanin allow neuromelanin-sensitive sequences (NMS-MRI) to localize the SNc within the midbrain.²⁸ NMS-MRI scans were not obtained for this cohort. Therefore, left and right regions of interest (ROIs) for the SNc were defined on a Montreal Neurological Institute-normalized T2w template using the NMS-MRI-based probabilistic atlas generated by Safai et al. (2020)²⁸ (https://github.com/apoorvasafai/NMS-SNc-atlas) as a reference. Definition was constrained to six axial slices with the superior limit one slice inferior to the level showing the largest red nucleus radius (Figure 1A).²⁰

Figure 1. MRI segmentations and quantification of histology metrics.

(A) SNc definitions on a T2w template (left: axial; center: coronal; right: sagittal). (B) Normalized R2* and (C) QSM images from a representative subject. (D) Magnified view of the SNc stained with H&E. Scans were color- and intensity-normalized to separate values corresponding to (E) neuromelanin (NM) and (F) hematoxylin to identify respectively NM-positive (NM+) neurons and glial cells (red masks in E and F). (G and I) Magnified views of the SNc stained with anti-α-synuclein. Scans were color- and intensity-normalized to oxidized DAB (H and J) to identify α-synuclein-positive Lewy bodies (large arrows in G; red masks in H) and α-synuclein-positive glial cytoplasmic inclusions (small arrows in I; red masks in J). (K) Magnified view of the SNc stained with anti-PHF-tau. Scans were color- and intensity-normalized to DAB (L) to identify tau-positive neurofibrillary tangles and neuropil threads (red masks in L). (M) Quantification results (grouped by primary pathology diagnosis) for NM+ neuron density, glial cell density, %AO α-synuclein, and %AO tau. Abbreviations: CBD – corticobasal degeneration; LBD – Lewy body disorders; MSA – multiple system atrophy; PSP – progressive supranuclear palsy.

Susceptibility MRI images for each subject were brought into template space by registering the mean magnitude images to the T2w images, and then T2w images to the T2w template. These transformations then were applied to subject R2* and QSM images using B-spline interpolation (Figure 1B,C). All registrations were performed using Advanced Normalization Tools (https://github.com/ANTsX/ANTs)²⁹ and inspected visually at each slice for misalignments. R2* and QSM values were calculated from mean intensities within the left and right ROIs.

Histology image and data acquisition

This study used archival midbrain slides that were acquired postmortem and prepared alongside samples taken from other regions of the brain as part of standard neuropathologic assessments for pathology diagnoses (as mentioned above). Briefly, slides were stained with hematoxylin and eosin (H&E), or anti-α-synuclein (Sigma-Aldrich AB5038, 1:2500) and anti-PHF-tau (Invitrogen MN1020, 1:400) antibodies developed with 3,3’-diaminobenzidine (DAB) and hematoxylin counterstain. Slides were digitized at 40× magnification using an Aperio AT (Leica) whole-slide scanner. SNc ROIs were defined on each scan by circumscribing the extent of neuromelanin-positive neurons. Scans then were color- and intensity-normalized in MATLAB using the Macenko et al. (2009)³⁰ two-stain normalization method (https://github.com/mitkovetta/staining-normalization/) to adjust for variable fading and separate stains by color vectors. H&E scans were separated into images corresponding to neuromelanin and hematoxylin; anti-α-synuclein and anti-tau scans were normalized to separate DAB from hematoxylin.

The resultant images were processed in ImageJ with user-defined intensity thresholding followed by morphology filters to mask and quantify: dopaminergic neurons by their neuromelanin deposits (Figure 1E), glia by their hematoxylin-positive nuclei (Figure 1F), and α-synuclein and tau by their respective morphologic aggregations (Figure 1G–L). This semi-automatic counting was compared against manual counting in sample images, showing strong agreement (Supplemental Figure 1). Quantification of histology metrics on whole-slide scans was conducted by a blinded rater in a single session. Estimated cell densities for each subject were calculated by dividing neuromelanin-positive neuron and glia counts by the ROI area. Estimates for percentages of area occupied by positive staining (%AO) for α-synuclein and tau were calculated by dividing the respective areas of positive staining by the ROI area.

Statistical analyses

Previous postmortem validation studies of R2* and QSM have shown that both susceptibility metrics reflect iron concentrations in strong linear relationships.^{7, 8} To test our hypothesis that R2* and QSM may similarly capture key neuropathologic features, we chose to test MRI-histology associations using two-tailed Pearson correlations and multiple regression models. In these analyses, both %AO α-synuclein and %AO tau were log-transformed to handle skewness. Data were paired according to the midbrain side sampled. Significance was set at P < 0.00625 (0.05/8) by the Bonferroni approach to control for the multiple associations tested under the primary hypothesis (two MRI × four histology metrics).

Given the variable time interval between MRI visit and death, two-tailed Pearson correlations also were used to rule out possible confounding effects of the MRI-pathology interval on histology results (Supplemental Table 1). Nonparametric Spearman correlation tests of the original (non-transformed) data also were used to confirm MRI-histology associations in the event that the parametric testing of log-transformed data would misrepresent the results (Supplemental Table 2).

All statistical analyses were performed using R version 3.6.1 (R Core Team, Vienna, Austria). Where error terms are provided, they are standard error of the mean unless otherwise stated.

Data availability

Raw data were generated at the Penn State Health-Milton S. Hershey Medical Center. Derived data supporting the findings are available from the corresponding author on request.

Results

Demographics, clinical diagnoses, and pathology diagnoses of the study subjects

At their study visit, subjects’ average age was 74.6 years (SE = 1.7) and average disease duration for parkinsonism subjects was 6.7 years (SE = 1.3). Clinical diagnoses were: Parkinson’s disease (n = 11), progressive supranuclear palsy (n = 6), multiple system atrophy (n = 4), atypical parkinsonism of uncertain diagnosis (n = 2), essential tremor (n = 1), and normal controls (n = 3). At death, the average MRI-pathology interval was 23.6 ± 3.4 mo. The pathology diagnoses were: Lewy body disorder (n = 12); progressive supranuclear palsy (n = 6); multiple system atrophy (n = 4); corticobasal degeneration (n = 1); and concurrent Lewy body disorder and progressive supranuclear palsy (n =1). Nine parkinsonism and two control subjects exhibited Alzheimer’s disease neuropathologic changes (ADNC). Subtypes of Lewy body disorder and multiple system atrophy, as well as grading of Alzheimer’s disease neuropathologic change, are shown in Table 1. Anti-α-synuclein slides were not available for one progressive supranuclear palsy subject and anti-tau slides were not available for one control and one Lewy body disorder subject.

Table 1.

Study subjects

		In vivo MRI				Postmortem pathology
ID	Sex	Visit (m/y)	Age (y)	Clinical diagnosis	DD (y)	Death (m/y)	MRI-path. interval (m)	Pathology diagnosis (NIA-AA scoring)
1	F	01/2013	68	PD	22	01/2013	0	LBD-L, ADNC-H (A3/B3/C3)
2	M	01/2013	91	PSP	4	10/2013	9	PSP
3	M	02/2013	63	MSA	6	03/2014	13	MSA-OPCA
4	F	10/2013	78	PD	11	07/2014	9	LBD-N, ADNC-H (A3/B3/C2)
5	M	10/2013	74	PSP	9	10/2014	12	PSP
6	M	04/2013	78	MSA	9	01/2015	21	LBD-B
7	M	06/2014	82	PD	5	02/2015	8	PSP
8	M	06/2014	62	PSP	3	05/2015	11	PSP
9	M	09/2014	71	Control	-	08/2015	11	Mild Atrophy
10	F	06/2015	83	ET	-	07/2016	13	ADNC-L (A3/B1/C1)
11	M	01/2016	71	PD	22	07/2016	6	LBD-B, ADNC-I (A2/B2/C1)
12	M	01/2016	76	PD	6	10/2016	9	CBD, ADNC-I (A2/B3/C2)
13	F	09/2015	70	APUD	9	03/2017	18	MSA-SND
14	F	08/2016	81	PSP	0	04/2017	8	LBD-B, PSP
15	M	09/2015	84	APUD	0	04/2017	19	LBD-L, ADNC-H (A3/B3/C2)
16	F	04/2016	84	MSA	0	11/2017	19	MSA-SND
17	F	06/2015	81	PD	3	12/2017	30	LBD-B
18	M	11/2015	82	PD	16	12/2017	25	LBD-L
19	F	06/2016	54	MSA	0	05/2018	23	MSA-OPCA, ADNC-I (A1/B2/C2)
20	M	07/2016	73	PD	3	02/2019	31	LBD-B, ADNC-I (A2/B3/C2)
21	M	12/2014	58	PD	1	06/2019	54	LBD-N, ADNC-I (A2/B3/C3)
22	F	01/2017	83	PD	4	12/2019	35	LBD-B
23	M	11/2014	63	PSP	9	02/2020	24	PSP
24	M	06/2015	69	PSP	6	04/2020	58	PSP
25a	F	09/2015	77	Control	-	07/2020	58	LBD-N, ADNC-I (A2/B3/C3)
26	M	05/2016	78	Control	-	12/2020	55	ADNC-H (A3/B3/C2)
27	M	04/2016	81	PD	6	01/2021	57	LBD-L

^aSubject was diagnosed with Parkinson’s disease after participating in the study as a control.

Abbreviations: ADNC – Alzheimer’s disease neuropathologic change (L: low; I: intermediate; H: high); APUD – atypical parkinsonism of uncertain diagnosis; CBD – corticobasal degeneration; DD – disease duration; ET – essential tremor; F – female; LBD – Lewy body disorder (B: brainstem predominant; L: limbic/transitional; N: neocortical/diffuse); M – male; MSA – multiple system atrophy (SND: striatonigral degeneration; OPCA: olivopontocerebellar atrophy); PD – Parkinson’s disease; PSP – progressive supranuclear palsy.

Histology results by parkinsonism diagnosis are presented in Figure 1M. Among the histology metrics, neuromelanin-positive neuron density associated negatively with log-transformed %AO α-synuclein (r = −0.58; df = 24; P = 0.002), whereas glial density showed a nonsignificant positive association with log-transformed %AO tau (r = 0.39; df = 23; P = 0.054) (Supplemental Table 1).

Associations between MRI and histology metrics

As shown in Table 2, R2* associated positively with both glial density (r = 0.70; df = 25; P < 0.0001) and log-transformed %AO tau (r = 0.61; df = 23; P = 0.001). QSM also associated with glial density (r = 0.67; df = 25; P < 0.001) and log-transformed %AO tau (r = 0.53; df = 23; P = 0.007), although the latter did not survive Bonferroni correction.

Table 2.

Pearson correlations between susceptibility MRI and histology metrics

	NM+ neuron density			Glial density			log(%AO α-synuclein)			log(%AO tau)
	r	df	P	r	df	P	r	df	P	r	df	P
R2*	−0.09	25	0.656	0.70	25	<0.0001 *	−0.09	24	0.659	0.61	23	0.001 *
QSM	0.05	25	0.812	0.67	25	<0.001 *	−0.01	24	0.945	0.53	23	0.007

Pearson correlation coefficients (r) between susceptibility MRI and histology metrics with degrees of freedom (df) and unadjusted probability values (P).

P < 0.05 are bolded.

^*P < 0.00625 to correct for multiple comparisons.

Abbreviations: NM+ = neuromelanin-positive; %AO = percent of area occupied by positive staining; R2* = apparent transverse relaxation rate; QSM = quantitative susceptibility mapping.

As seen in Figure 2, scatterplots showed that R2* and QSM associations with glial density and log-transformed %AO tau may have been driven by a single subject with the highest recorded values in all four metrics. Without this subject, R2* associations with glial density (r = 0.52; df = 24; P = 0.006) and log-transformed %AO tau (r = 0.53; df = 22; P = 0.007), and QSM associations with glial density (r = 0.40; df = 24; P = 0.041) and log-transformed %AO tau (r = 0.41; df = 22; P = 0.046) remained.

Figure 2. MRI versus histology associations in the substantia nigra pars compacta.

(A) R2* versus glial cell density. (B) R2* versus the log-transformed %AO tau. (C) QSM versus glial cell density. (D) QSM versus log-transformed %AO tau. Lines shown are least-square fitted with Pearson correlations (r), degrees of freedom (df) and unadjusted probability values (P). Abbreviations: ADNC – Alzheimer’s disease neuropathologic change; CBD – corticobasal degeneration; LBD – Lewy body disorders; MSA – multiple system atrophy; PSP – progressive supranuclear palsy.

Multiple regression analyses predicting R2* and QSM by both glial density and tau burden

Multiple linear regression was used to test whether R2* and QSM were explained with both glial density and log-transformed %AO tau as predictors. For R2*, the overall model was significant (R² = 0.651, F(2, 22) = 20.50, P < 0.00001), with prediction by both glial density (β = 0.049, P < 0.001) and log-transformed %AO tau (β = 3.80, P = 0.009). The QSM model also was significant (R² = 0.580, F(2, 22) = 15.16, P < 0.0001), with prediction by both glial density (β = 0.286, P < 0.001) and log-transformed %AO tau (β = 15.59, P = 0.065) (Table 3).

Table 3.

Post hoc regression models

Variable	R2* (s−1) Model 1 (full model)	R2* (s−1) Model 2	R2* (s−1) Model 3	QSM (ppb) Model 1 (full model)	QSM (ppb) Model 2	QSM (ppb) Model 3
Constant	8.85 (8.81)	−5.92 (8.29)	43.89 (3.28)	−27.58 (53.38)	−74.25 (46.46)	176.30 (19.54)
Glial density (cells/mm2)	0.049** (0.012)	0.062*** (0.012)	-	0.286** (0.072)	0.318** (0.071)	-
log (%AO tau)	3.80* (1.32)	-	5.94* (1.59)	15.59 (8.02)	-	28.05* (9.47)
Observations	25	27	25	25	27	25
R 2	0.651	0.487	0.378	0.580	0.445	0.276
Adjusted R2	0.619	0.466	0.350	0.541	0.423	0.245
F-statistic	20.50	23.69	13.95	15.16	20.03	8.77
P-value	<0.00001	<0.0001	0.001	<0.0001	0.0001	0.007

Linear regression models with R2* and QSM as dependent variables, predicted by glial density and log-transformed %AO tau. In Model 1 (full model), MRI metrics were predicted by both glial density and log-transformed %AO tau; in Model 2, MRI metrics were predicted by only glial density; In Model 3, MRI metrics were predicted by only log-transformed %AO tau. All data were obtained from the substantia nigra pars compacta.

^*P < 0.01

^**P < 0.001

^***P < 0.0001

Abbreviations: NM+ = neuromelanin-positive; %AO = percent of area occupied by positive staining; R2* = apparent transverse relaxation rate; QSM = quantitative susceptibility mapping.

Discussion

In this study, we examined in vivo sMRI associations with postmortem histologic features of the SNc in 27 subjects. We found that both R2* and QSM correlated with glial density and tau burden. These results indicate that sMRI is sensitive to neurodegenerative changes in the SNs of patients with parkinsonisms, and support its clinical and translational value in studies of parkinsonism neuropathology.

Nigral iron is localized predominately to glia

R2* and QSM are sensitive to local magnetic field inhomogeneities, of which iron is the predominant contributor.^{7, 8} Iron deposition in the SN has been observed in Parkinson’s disease and other parkinsonisms using both sMRI metrics,⁶ although the localization of iron accumulation in the SN and the neuropathologic features to which it may be associated remain uncertain. It has been hypothesized that neuromelanin may play the leading role in iron sequestration.³¹ This is supported by the substantial iron-chelating ability of neuromelanin as demonstrated ex vivo using various elemental analysis methods,^32–34 and the spatial overlap between sMRI changes and the distribution of neuromelanin in Parkinson’s disease brains.³⁵

In the human SN, glia are known to represent an estimated 95% of brain cells.³⁶ Oligodendrocytes, astrocytes, and microglia have been shown to store the majority of total iron in the brain³⁷ and within the SN.³⁸ Recently, Friedrich et al. (2021)³⁹ presented compelling evidence that iron is localized predominantly to glia in the SNc in Parkinson’s disease by quantifying iron concentrations across immunolabelled cell types using microparticle-induced X-ray emission. Corroborating earlier findings of iron colocalization with neuromelanin, they found iron concentrations were increased within neuromelanin deposits in Parkinson’s disease patients. This increase had a minimal effect on total iron, however, and was outweighed by increased iron concentrations within oligodendrocytes and microglia (but not astrocytes). In total, they estimated 80% of SNc iron was localized to glia in both control and Parkinson’s disease brains.³⁹

Our data support their finding since we observed strong linear associations between sMRI metrics and glial density. The exact glial cell types contributing to this observation, however, are unknown. Because SNc microglial activation has been shown to be particularly robust in Parkinson’s disease⁴⁰ and is associated strongly with postmortem iron staining,⁴¹ we suspect microglia may be the primary storage compartment for excess iron. An oligodendrocyte response is less studied, though the data from Friedrich et al. (2021)³⁹ indicate iron changes may differ dramatically among oligodendrocyte subtypes. They observed large increases in iron concentrations among myelinating subtypes and iron loss among non-myelinating subtypes.³⁹ Further studies are needed to elucidate the glial cell types harboring iron in parkinsonisms.

Tau is independently associated with nigral iron

Tau is a microtubule-associated protein expressed primarily in neurons. In Alzheimer’s disease, tau aberrantly aggregates into neurofibrillary tangles that are believed to cause neuronal dysfunction and death.⁴² Animal studies have shown that endogenous tau expression is much lower in glial cell types than in neurons.^{43, 44} Patients with progressive supranuclear palsy and corticobasal degeneration exhibit tau aggregations in neurons, oligodendrocytes, and astrocytes, however.⁴³

Our observation that R2* and QSM associated with tau burden is interesting as increased iron content, as determined by QSM, has been found to overlap with tau-PET tracer uptake (¹⁸F-AV-1451 and ¹⁸F-RO-948) in studies of Alzheimer’s disease.^{45, 46} There is concern, however, that this primarily reflects off-target binding of ¹⁸F-AV-1451 and other tau-PET tracers to iron, monoamine oxidase, and other molecular structures.^{45, 47, 48} There is other evidence to suggest that iron and tau are directly associated. For example, in a study of progressive supranuclear palsy neuropathology, ferritin was observed to co-localize with tau filaments in tissue samples and co-purify with isolated tau polymers.⁴⁹ Interestingly, ferric iron has been shown in vitro to bind brain-isolated hyperphosphorylated tau and induce its polymerization.¹⁷ This suggests a nonlinear precipitating effect that may explain why we observed R2* and QSM associated linearly with log-transformed tau staining.

In this cohort, glial density and tau associated weakly (Supplemental Table 1). Their association may be related to tau burden distributed to oligodendrocytes and astrocytes, as is found in progressive supranuclear palsy and corticobasal degeneration. Yet multiple regression models for R2* and QSM suggest that glial density and tau burden have distinct contributions to sMRI variance (R² values were improved with both glial density and tau burden in the same model, Table 3), and this independence may reflect those tau neurofibrillary tangles localized within neurons. Understandably, MRI signals are not composed “purely” of a single cellular or biochemical feature. Future studies are needed to better understand the associations among iron, tau burden, and their respective distributions to SNc cell types.

Limitations and future directions

This study had limitations future studies may address. First, although this is the largest parkinsonism study to date to examine in vivo sMRI associations with postmortem histology, our existing cohort still is relatively small. We continue to capture postmortem brain samples from our Parkinson’s Disease Biomarker Program cohort to power additional investigations. Second, our SNc ROIs relied on atlas-based segmentations since we did not obtain NMS-MRI scans. As an improvement on our earlier study,²⁰ we distinguished the SNc from the SN pars reticulata in order to better study the key structure of parkinsonism neuropathology. Third, our use of single archival slides reduced our ability to precisely associate MRI with histology. For each subject, we paired the midbrain side sampled to MRI data from the corresponding SNc ROI. Improving on our previous study, here we quantified histologic features across the entire slide rather than using the averages of cell counts sampled at high-power fields and semi-quantitative ratings for α-synuclein and tau burden.²⁰ Nonetheless, each slide was only a two-dimensional representation of a three-dimensional (3D) structure. Stereotactic association of MRI with 3D tissue blocks is desired in future studies. Last, we used archival slides. Future studies will benefit from immunolabelling specific glial cell types to better distinguish their respective associations with MRI. Co-staining may address additional questions, such as the extent to which tau and iron are distributed to particular cell types, and whether differences in their accumulation drive distinct parkinsonism neuropathologies.

Despite these limitations, the current study demonstrates the feasibility of finding linear associations between quantitative histology and MRI metrics on a subject-to-subject basis. Overall, we believe these results to be more robust than those in our earlier study²⁰ given the larger sample size and improved methods. We previously used nonparametric correlations to show that R2* and α-synuclein were directly associated. We found this observation reasonable as both iron deposition and α-synuclein burden have been shown to be higher in the SN of patients with α-synucleinopathies. ^{9–12, 50, 51} But here we observe that, in the SNc, R2* and α-synuclein are not related in a linear fashion, but that both R2* and QSM are associated linearly with glial cell density and tau burden. In conclusion, we anticipate that ongoing and future studies will address the above limitations and expand into other subcortical, cerebellar, and midbrain structures involved in parkinsonism neuropathology. Comprehensive study of additional neuropathologies will improve our knowledge of their respective contributions to MRI signals. This will allow a better interpretation of MRI biomarkers in the diagnosis of parkinsonisms and monitoring of their progression.