Neuropathologic basis of quantitative susceptibility mapping in the substantia nigra: contributions of tau, pigmented neurons, and iron

Abstract

Quantitative susceptibility mapping (QSM) on MRI quantifies tissue magnetic susceptibility, which increases with iron accumulation, myelin loss, and neuroinflammation. Elevated QSM in the substantia nigra (SN) has been reported in Lewy body disease and other parkinsonian disorders, but from existing literature it remains unclear whether these findings are driven by neurodegeneration-related iron deposition or other neuropathologic features. We studied 59 autopsied participants who underwent antemortem 3 T MRI with QSM (median age at death, 78.5 years; MRI-to-death interval, 2.0 years), including clinical diagnoses of 18 with Alzheimer’s-type dementia, 15 cognitively unimpaired, 9 with mild cognitive impairment, and 9 with dementia with Lewy bodies. A machine learning-incorporated digital histopathology pipeline quantified tau burden, iron deposition, and neuronal densities. The SN was divided into geometric quadrants, and QSM values were analyzed in relation to corresponding neuropathologic measures within each quadrant. Iron deposition correlated with QSM in all quadrants (ρ = 0.41–0.56, all P < 0.005). Tau burden correlated with QSM in the ventromedial (VM) quadrant (ρ = 0.45, P = 0.002), whereas lower pigmented neuron density was associated with higher QSM in the dorsomedial quadrant (ρ = – 0.35, P = 0.007). Rank regression analysis confirmed iron as the strongest predictor of QSM across all quadrants (β = 0.35–1.06, P ≤ 0.026), with tau independently associated with QSM in the VM (β = 0.45, P = 0.015). Mediation analysis demonstrated that tau exerted direct (0.45, P = 0.018) and indirect effects via iron (0.12, P = 0.046) on QSM in the VM, with 80% of the effect being direct. These findings underscore the contributions of tau pathology, pigmented neuron density, and iron deposition to nigral magnetic susceptibility and highlight the potential for QSM to serve as a sensitive biomarker for diverse neuropathologies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00401-026-02991-x.

Introduction

Quantitative susceptibility mapping (QSM) measures the magnetic susceptibility properties of brain tissue on MRI [23, 58]. QSM values increase with iron and other mineral accumulation in tissue, as well as with myelin loss and neuroinflammation [30, 40, 53]. Lewy body disease (LBD) is a neuropathologic umbrella term that encompasses neurodegenerative disorders with Lewy-related pathology, including Parkinson’s disease (PD) and dementia with Lewy bodies (DLB). Consistent with iron accumulation in the substantia nigra (SN) in LBD, elevated QSM values have been reported in patients with PD and DLB [5, 6, 8, 9, 11, 15, 29, 61]. These studies highlight the potential for QSM to serve as a diagnostic biomarker of LBD, but other contributing factors to iron deposition and elevated tissue susceptibility remain unknown.

In addition to LBD, progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD) are also associated with nigral neurodegeneration and iron accumulation in the SN [13, 59]. Moreover, approximately 10% of patients with Alzheimer’s disease (AD) without concomitant Lewy-related pathology exhibit parkinsonism in association with nigral neuron loss [41]. In accordance with these reports, elevated QSM in the SN has also been documented in the tauopathies, including PSP [4, 21, 27, 31, 47, 50, 51, 57], CBD [15, 31, 57], and AD [12], making interpretation of QSM in the SN complex.

Comprehensive evaluation of neuropathologic measures and their correlation with QSM values is limited [31, 57]. Wang et al. performed a clinicopathologic study with a postmortem cohort including those with parkinsonian tauopathies [57]. They measured tau burden, pigmented neuron and glial cell density, and concluded that tau and glial density contribute to tissue magnetic susceptibility in the SN. However, because tissue iron deposition was not measured, independent contributors to tissue magnetic susceptibility in the SN were not clarified.

To address these challenges, we applied digital pathology with a machine learning-based object detection model to quantify histologic parameters in the SN across neurodegenerative diseases, including participants with autopsy-confirmed LBD and tauopathies. By assessing iron deposition, tau burden, and neuron density in the four quadrants of SN, we sought to elucidate the neuropathologic determinants of antemortem QSM and delineate their independent contributions to magnetic susceptibility.

Materials and methods

Participants

We included 60 participants—33 from the Mayo Clinic Alzheimer’s Disease Research Center (ADRC; Rochester, MN and Jacksonville, FL) and 27 from the Mayo Clinic Study of Aging (MCSA; Rochester, MN) [44]—who underwent antemortem MRI examinations with a multi-echo T2* gradient echo (MEGRE) sequence and autopsy at the Mayo Clinic ADRC Neuropathology Core from 2018 to 2023. One participant was excluded due to substantial number of microbleeds, which affected the iron measurements. Therefore, this study cohort consisted of 59 participants.

Clinical characterization

Participants underwent clinician’s interviews, questionnaire surveys, and neurological examinations [28, 44]. Cognitive function was assessed using the Mini-Mental State Examination (MMSE) [22] and the Clinical Dementia Rating^® Sum-of-Boxes (CDR-SB) [25]. Clinical diagnoses were determined in reference to the previous publication [1, 34, 36, 42, 43]. Parkinsonism was defined as the presence of at least two of the four cardinal features (tremor, rigidity, bradykinesia, and postural instability). To quantify severity, the Unified Parkinson Disease Rating Scale-Part III (UPDRS-III) was used [19]. Since ADRC participants receive the UPDRS-III and the MCSA uses a modified UPDRS form, we combined 11 common components across the forms: speech, tremor, facial expression, rigidity of the neck, rigidity of the right arm, rigidity of the left arm, rigidity of the right leg, rigidity of the left leg, posture, bradykinesia, and gait.

Neuropathologic evaluation

Formalin-fixed brains were sampled and embedded with paraffin. 5-μm-thick sections were mounted on glass slides and stained with hematoxylin and eosin (H&E) staining. For diagnostic purpose, immunohistochemistry (IHC) was performed using α-synuclein antibody (NACP; rabbit polyclonal, 1:3000 Mayo Clinic’s antibody). Neurofibrillary tangles (NFTs) and senile plaques were visualized with a modified Bielschowsky silver stain or thioflavin S staining. Braak NFT stage (0-VI) and Thal amyloid phase (0–5) were determined based on the distribution and counts of NFTs and senile plaques, respectively [10, 54]. Neuropathologic diagnoses were provided by expert neuropathologists (DWD, MEM, AN, and RRR), referencing published criteria [16, 17, 26, 34, 35, 38, 39, 45]. The neuropathologic diagnosis of AD was defined as intermediate or high levels of AD neuropathologic change according to the NIA-AA criteria [38]. Lewy body pathology was classified according to the DLB Consortium criteria [34] (Supplementary Table S1). Amygdala-predominant Lewy-related pathology was excluded from LBD in this study, based on prior literature suggesting that it represents an isolated form of α-synucleinopathy, typically associated with AD dementia and not Lewy body dementia [56]. For the quantitative assessment of iron deposition and tau burden, Prussian blue staining and IHC using a phosphorylated tau antibody (CP13; mouse monoclonal, 1:1000; from the late Dr. Peter Davies, Feinstein Institute for Medical Research, NY) were performed on midbrain sections. The CP13 antibody recognizes tau phosphorylated at Ser202 and has been shown to produce staining patterns comparable to AT8 [37].

Digital pathology

Midbrain sections were stained with H&E, Prussian blue, and tau IHC and digitized at × 20 magnification using the ScanScopeXT (Aperio Technologies, Vista, CA). α-Synuclein IHC was not included in the quantitative analyses, as prior studies have shown that α-synuclein burden assessed by conventional immunostaining may not correlate consistently with disease progression and is not significantly associated with QSM measures [49, 57]. Similarly, amyloid-β assessment in the SN was not included, as no significant correlation was observed between QSM values and Thal amyloid phase in this cohort (Supplementary Table S2). Digital pathologic analyses were performed by an expert neurologist (DO) using QuPath 0.4.3 [7], Python 3.9.15, and related libraries. The border of the SN was defined as a region surrounded by the midbrain tegmentum and cerebral peduncle at the level of the oculomotor nerve roots [41, 46]. To objectively define subregions of the SN, this study adopted a geometric approach based on midlines along the medial–lateral and ventral-dorsal axes, rather than relying on neuronal populations [20]. This method divided the SN into four subregions: ventromedial (VM), ventrolateral (VL), dorsomedial (DM), and dorsolateral (DL) (Fig. 1).

Fig. 1.

Representative images on quadrants in the substantia nigra on QSM and H&E staining. a Quantitative susceptibility map (QSM) of the midbrain at the level of the substantia nigra (SN). b The same slice with color masks delineating the ventromedial (VM; orange), dorsomedial (DM; light green), ventrolateral (VL; brown), and dorsolateral (DL; red) quadrants. c H&E-stained midbrain section with manual segmentation of the SN. d The SN at the level of the oculomotor nerve root is divided into geometric quadrants based on midlines along the medial–lateral and ventral–dorsal axes. e pigmented neurons (light green) and non-pigmented neurons (cyan) are automatically detected with a previously established machine learning pipeline. Scale bars: 2 mm in c and d; 50 µm in e. DL, dorsolateral; DM, ventromedial; QSM, quantitative susceptibility mapping; VL, ventrolateral; VM, ventromedial

As in previous studies, we considered nigral neuropigmentation to primarily reflect neuromelanin, although it may include other pigments [41, 60]. Given the evidence that neuromelanin-containing dopaminergic neurons in the SN are vulnerable to Lewy-related pathology [24], we distinguished pigmented from non-pigmented neurons and calculated their densities in SN quadrants. Pigmented and non-pigmented neurons in the SN on H&E-stained sections were automatically detected using a previously developed pipeline, available at https://github.com/onnonuro/countnigra [41]. Briefly, a machine learning-based object detection model (YOLOv8) was fine-tuned with human annotations and integrated into a Python program that processes whole-slide images. The densities of pigmented and non-pigmented neurons per mm² were then calculated for each quadrant of the SN.

Segmentation of iron-deposited areas was determined by thresholding signal intensity after normalization by estimate stain vectors in QuPath (Supplementary Fig. S1a). Segmentation of tau-immunoreactive areas in the SN were measured using a multiple-thresholding method, as previously described [41]. Briefly, pigments were extracted by subtracting objects filtered by larger area and lower intensity; the remaining signals, corresponding to neuropil threads and NFTs, were segmented (Supplementary Fig. S1b). The segmented areas were quantified as the percentage of the area occupied (%AO) in each region of interest.

MRI acquisition and susceptibility mapping

MRI acquisition and subsequent QSM reconstruction were performed according to the previously published protocols [12, 15]. Briefly, antemortem MRI was acquired for all participants on a 3-Tesla Siemens PRISMA system (Siemens, Erlangen, Germany) with a 64-channel head coil. High-resolution anatomical images were obtained with a three-dimensional T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) sequence for segmentation and labeling. For QSM, phase data were acquired with a bipolar multi-echo three-dimensional gradient-echo (GRE) sequence with five echoes (6.71, 10.62, 14.53, 18.44, 22.35 ms) A TR/TE1/ΔTE = 28/6.71/3.91 ms was used with a field of view = 200 × 200 mm, voxel dimensions = 0.5 × 0.5 × 1.8 mm³, NEX = 1, flip angle = 15°, bandwidth = 280 Hz/pixel.

T1-weighted MRI scans were segmented and corrected for intensity inhomogeneity using SPM12 with tissue priors from the Mayo Clinic Adult Lifespan Template (MCALT; https://www.nitrc.org/projects/mcalt/). The SN region, segmented using the DISTAL atlas [2, 18], was further subdivided manually into four quadrants (VM, VL, DM, and DL) to align with neuropathologic evaluation by SK, as shown in Fig. 1. QSM images were reconstructed using STI Suite software with Laplacian-based phase unwrapping, V-SHARP filtering, and iterative least squares decomposition [32, 33, 48]. Mean susceptibility values were then calculated for each SN quadrant.

Statistical analysis

The clinicopathologic characteristics were reported with medians and interquartile ranges (IQR) for continuous variables and counts and percentages for categorical variables. Due to skewness and influential neuropathology data points, we used a combination of non-parametric and parametric statistical methods. Although transformations were considered, some of the data points continued to be influential, and therefore all continuous predictor and response variables were analyzed with ranks. In addition, because neuropathologic assessment was performed in the hemi-brain, the imaging data were matched to the corresponding side for each participant. Due to limitations in tissue availability and staining quality, histologic assessments were incomplete in 13 participants. Of these 13 participants, tau could not be assessed in 12 and iron in 2. Differences in demographics and clinical characteristics between groups were analyzed with Wilcoxon two-sample rank sum tests for continuous variables and chi-squared tests for categorical variables. Partial Spearman rank correlations, adjusting for age and time from scan to death, were done to look for associations with the QSM values and the neuropathological measures. All predictors and responses were ranked for the multiple linear regression analyses, which adjusted for age and time from scan to death, to assess the contributions of tau burden, neuronal loss, and iron deposition to QSM. Two-way interactions among the neuropathologic measures were evaluated in these regression models. Finally, mediation analyses were conducted, with all continuous variables analyzed on the ranks. This tested whether the direct effects or the indirect effects of the predictors pigmented nigral neuron density or tau burden had on the QSM pathways, while accounting for the mediator variable iron deposition. This was done using the mediation package from R.

Data availability

Data and codes used in the current study are available from the corresponding author upon reasonable request.

Results

Participants’ characteristics

Fifty-nine autopsied participants were included in the analyses, with a median age at death of 78.5 years (range 53.8–94.8) and a median time between MRI and death of 2.0 years (range 0.5–4.1; Table 1). At the time of MRI, the median Mini-Mental State Examination (MMSE) was 23 (range 7–30) and Unified Parkinson's Disease Rating Scale (UPDRS) score was 1 (range 0–24). Clinically, 18 participants were diagnosed as Alzheimer’s type dementia, 15 were cognitively unimpaired, 9 had mild cognitive impairment, and 9 had dementia with Lewy bodies. Postmortem examination established diagnoses of AD in 29 cases, 8 cases with co-pathology of LBD and AD (LBD/AD), LBD in 6 cases, 2 cases with co-pathology of frontotemporal lobar degeneration with TDP-43 pathology (FTLD-TDP) and AD, and 1 case with FTLD-TDP, CBD, PSP, and Pick’s disease. Ten participants showed no neurodegenerative changes on postmortem evaluation (Fig. 2).

Table 1.

Clinicopathologic characteristics of participants

	All participants n = 59
Age at MRI, years	78.5 (69.3, 83.7)
Age at death, years	79.5 (71.9, 86.3)
Time between MRI and death, years	2.0 (1.4, 2.7)
Males, no. (%)	37 (63%)
APOE ε4 carrier, no. (%)	31 (56%)
Education, years	14.0 (12.0, 16.0)
MMSE at MRI	23.0 (20.0, 28.0)
CDR-SB	3.5 (0.0, 8.0)
Amyloid PET positive, no. (%)	37 (73%)
PiB SUVr	2.11 (1.50, 2.72)
Tau PET positive, no. (%)	28 (58%)
Tau SUVr	1.33 (1.25, 1.78)
Braak NFT stage	V (III, VI)
Thal Aβ phase	4 (2, 5)
Hallucinations, no. (%)	8 (18%)
Parkinsonism, no. (%)	23 (45%)
pRBD, no. (%)	12 (23%)
Modified UPDRSs score at MRI	1.0 (0.0, 5.8)

Characteristics table of groups with the median (IQR) listed for the continuous variables and count (%) for the categorical variables. P-values for differences between groups come from a Wilcoxon two-sample rank sum test for continuous variables and a Chi-squared test for categorical variables

Aβ, amyloid β; CDR-SB, Clinical Dementia Rating Sum-of-boxes; LBD, Lewy body disease; MMSE, Mini-Mental State Examination; MRI, magnetic resonance imaging; NFT, neurofibrillary tangles; PET, positron emission tomography; PiB, Pittsburgh Compound-B; pRBD, probable REM sleep behavior disorder; SUVr, standardized uptake value ratio; UPDRS III, Unified Parkinson’s Disease Rating Scale-Part III (3rd edition)

Fig. 2.

Clinical diagnoses and neuropathologic summary of the participants. Clinical diagnoses (left) and their corresponding neuropathologic summary (right) for the participants. ADNC (High/Int), high or intermediate Alzheimer´s disease neuropathologic change; CBD, corticobasal degeneration; FTLD-TDP, frontotemporal lobar degeneration with TDP-43 pathology; LBD, Lewy body disease; No NP, no neurodegenerative pathology; PiD, Pick’s disease; PSP, progressive supranuclear palsy; TDP-43, TAR DNA-binding protein 43

Correlation of QSM with nigral pathology

Pigmented and non-pigmented neuronal densities, tau burden, and iron accumulation were measured in SN geometric quadrants (Fig. 1 and Supplementary Fig. S2), and partial Spearman correlations with QSM values were calculated (Table 2). Iron staining (Prussian blue) %AO was positively associated with QSM in all SN subregions (ρ = 0.41–0.56, all P < 0.005). Tau %AO was positively associated with QSM in the VM subregion of the SN (ρ = 0.45, P = 0.002). Pigmented neuron density was negatively associated with QSM in the DM subregion (ρ = – 0.35, P = 0.007), while non-pigmented neuron density showed no association with QSM in the subregions. Cases with parkinsonism showed lower neuronal densities than cases without parkinsonism, with statistical significance observed for non-pigmented neuron density in the DM quadrant (Supplementary Table S3). Comparison across diagnostic categories with at least three cases (no neurodegenerative change, AD, AD/LBD, and LBD) showed lower pigmented neuron density in the LBD and LBD/AD groups and higher tau burden in the AD group, while no significant differences in QSM values were observed among these major diagnostic categories (Supplementary Table S4).

Table 2.

Spearman’s rank correlation coefficients between QSM and neuropathologic measurements

	QSM
	Ventromedial	Dorsomedial	Ventrolateral	Dorsolateral
Pigmented neuron density	− 0.014 (P = 0.92)	− 0.35 (P = 0.007)	− 0.26 (P = 0.050)	− 0.26 (P = 0.053)
Non-pigmented neuron density	0.085 (P = 0.53)	0.055 (P = 0.68)	0.066 (P = 0.63)	0.026 (P = 0.85)
Tau %AO	0.45 (P = 0.002)	0.26 (P = 0.087)	0.23 (P = 0.13)	0.18 (P = 0.23)
Iron %AO	0.41 (P = 0.002)	0.45 (P =  < 0.001)	0.56 (P =  < 0.001)	0.56 (P =  < 0.001)

QSM values were adjusted for age and time from scan to death

Bold values indicate statistical significance at p < 0.05

%AO, percent area occupied; QSM, quantitative susceptibility mapping

Linear regression analysis done on the ranks

To assess the independent contributions of tau burden, neuronal loss, and iron deposition to susceptibility changes, we performed linear regression analyses using three models (Table 3) where each predictor and response were ranked. Each model considered two-way interactions with the neuropathologic variables, but only one model had a significant interaction. Model 1, which included iron %AO and pigmented neuron density, showed that iron deposition was independently associated with QSM in quadrants (estimate [SE] = 0.425 [0.127] in DM; 0.548 [0.122] in VL; and 0.543 [0.124] in DL; all P ≤ 0.002). Pigmented neuron density was negatively associated with QSM in the DM (– 0.297 [0.122], P = 0.019). In VM quadrants where both iron %AO and pigmented neuron density were associated with QSM, a negative interaction between iron and pigmented neuron density was detected (– 1.018 [0.007], P = 0.017), indicating that the effect of iron accumulation on susceptibility becomes greater as the pigmented neuron density decreases. Model 2, which included non-pigmented neuron density and iron %AO, showed that iron remained strongly associated with QSM in all quadrants (0.433 [0.136] in VM; 0.483 [0.132] in DM; 0.579 [0.119] in VL; and 0.593 [0.122] in DL; all P ≤ 0.002), while non-pigmented neuron density showed no association in any region. Model 3, which included tau %AO and iron %AO, demonstrated that iron %AO was associated with QSM in all quadrants (0.348 [0.151] in VM; 0.502 [0.143] in DM; 0.563 [0.128] in VL; and 0.601 [0.130] in DL; all P ≤ 0.026). Tau %AO was independently associated with QSM in the VM quadrant (β = 0.448 [0.177], P = 0.015).

Table 3.

Rank regression analysis for substantia nigra susceptibility as the outcome

	QSM
	Ventromedial		Dorsomedial		Ventrolateral		Dorsolateral
	Estimate (SE)	P-value	Estimate (SE)	P-value	Estimate (SE)	P-value	Estimate (SE)	P-value
Model 1
Iron %AO	1.057 (0.268)	< 0.001	0.425 (0.127)	0.002	0.548 (0.122)	< 0.001	0.543 (0.124)	< 0.001
Pigmented neuron density	0.598 (0.227)	0.011	− 0.297 (0.122)	0.019	− 0.121 (0.117)	0.3	− 0.126 (0.119)	0.3
Interaction: Iron × pigmented neuron	− 1.018 (0.007)	0.017
Model 2
Iron %AO	0.433 (0.136)	0.002	0.483 (0.132)	< 0.001	0.579 (0.119)	< 0.001	0.593 (0.122)	< 0.001
Non-pigmented neuron density	0.051 (0.128)	0.69	0.096 (0.129)	0.46	0.016 (0.118)	0.89	− 0.074 (0.116)	0.53
Model 3
Tau %AO	0.448 (0.177)	0.015	0.147 (0.168)	0.39	0.254 (0.158)	0.12	0.149 (0.157)	0.35
Iron %AO	0.348 (0.151)	0.026	0.502 (0.143)	0.001	0.563 (0.128)	< 0.001	0.601 (0.130)	< 0.001

These variables are analyzed on the ranks. The results for the final two predictor models with adjustments for rank age and rank time from scan to death. If an interaction was not significant, it was removed, and the results of the two main effects are shown

Bold values indicate statistical significance at p < 0.05

%AO, percent area occupied; QSM, quantitative susceptibility mapping; SE, standard error

Mediation analysis for the pathologic predictors of nigral susceptibility

Mediation analyses were performed to examine the direct effects of pigmented neuron density and tau burden on QSM, as well as the indirect effects mediated by iron accumulation (Table 4). In the first analysis, pigmented neuron density was the primary predictor, iron %AO served as the mediator, and QSM was the outcome variable. Iron accumulation negatively mediated the effect of pigmented nigral neuron density on QSM in the VM (estimate [confidence interval] = − 0.15 [− 0.30, − 0.03], P = 0.012), VL (− 0.13 [− 0.28, − 0.00], P = 0.046), and DL quadrants (− 0.14 [− 0.30, − 0.02], P = 0.026). Next, tau burden was evaluated as the primary predictor, with iron %AO as the mediator and QSM as the outcome. In the VM quadrant, tau burden demonstrated a direct effect on QSM (0.45 [0.10, 0.79], P = 0.018) and an indirect effect mediated by iron accumulation (0.12 [0.001, 0.30], P = 0.046). The total effect of tau on QSM was also statistically significant in the VM quadrant (0.57 [0.23, 0.92], P = 0.002), with 20% of the effect mediated by iron (Fig. 3).

Table 4.

Mediation models evaluating the effects of pigmented neuron density and tau burden on QSM through iron accumulation

	Ventromedial		Dorsomedial		Ventrolateral		Dorsolateral
	Estimate (CI)	P-value	Estimate (CI)	P-value	Estimate (CI)	P-value	Estimate (CI)	P-value
Predictor, pigmented neuron density
ACME (indirect effects via iron)	− 0.15 (− 0.30, − 0.03)	0.012	− 0.07 (− 0.19, 0.03)	0.18	− 0.13 (− 0.28, − 0.00)	0.046	− 0.14 (− 0.30, − 0.02)	0.026
Direct effects	0.14 (− 0.13, 0.41)	0.28	− 0.30 (− 0.53, − 0.06)	0.014	− 0.12 (− 0.35, 0.11)	0.25	− 0.13 (− 0.36, 0.11)	0.25
Total effect	− 0.01 (− 0.28, 0.27)	0.94	− 0.36 (− 0.61, − 0.11)	0.006	− 0.25 (− 0.50, − 0.00)	0.050	− 0.27 (− 0.52, − 0.01)	0.042
Proportion mediated via iron	0.44	0.93	0.17	0.18	0.51	0.080	0.52	0.056
Predictor, tau %AO
ACME (indirect effects via iron)	0.12 (0.001, 0.30)	0.046	0.16 (0.002, 0.37)	0.05	0.07 (− 0.13, 0.27)	0.53	0.07 (− 0.12, 0.28)	0.49
Direct effects	0.45 (0.10, 0.79)	0.018	0.15 (− 0.18, 0.47)	0.36	0.25 (− 0.05, 0.56)	0.1	0.15 (− 0.15, 0.46)	0.32
Total effect	0.57 (0.23, 0.92)	0.002	0.30 (− 0.04, 0.66)	0.098	0.32 (− 0.03, 0.68)	0.076	0.22 (− 0.13, 0.59)	0.23
Proportion mediated via iron	0.20	0.048	0.47	0.14	0.20	0.52	0.32	0.47

All variables were rank-transformed for analysis, with iron %AO as the mediator. Age and time from scan to death were included as covariates

Bold values indicate statistical significance at p < 0.05

ACME, average causal mediation effect; CI, confidence interval; %AO, percent area occupied; QSM, quantitative susceptibility mapping

Fig. 3.

Proposed mechanism of elevated nigral susceptibility. QSM, quantitative susceptibility mapping

Discussion

We leveraged a large postmortem cohort with antemortem QSM measurements and applied an automated digital histopathology pipeline, which enabled direct quantification of tau- and iron-positive areas and separate detection of pigmented and non-pigmented neurons. This study revealed several important findings regarding the neuropathologic determinants of nigral susceptibility: (1) the predominance of tau’s independent influence on QSM (80%) over its indirect effect mediated through iron; (2) the VM quadrant as a sensitive region for QSM biomarker for tau pathology and iron accumulation; (3) QSM in the DM and VM quadrants as independent biomarkers of pigmented neuron loss, interacting with iron accumulation in the VM.

Elevated tissue susceptibility has been reported in tauopathies such as PSP, CBD, and AD, which were attributed to secondary iron accumulation following neurodegeneration [3, 12, 13, 52]. The current study highlights tau’s non-iron-mediated effects on susceptibility in the SN. Mediation analysis demonstrated that, in the VM quadrant of the SN, the direct effect of tau burden on susceptibility was four times greater than the indirect effect mediated through iron accumulation (0.8 vs. 0.2 in Table 4). O'Callaghan et al. performed experiments using tauopathy model mice and observed tau pathology with elevated tissue magnetic susceptibility in the brain, without evidence of iron accumulation [40]. In human tauopathies, tau pathology may similarly increase QSM through mechanisms such as myelin loss and neuroinflammation, which warrants further investigation. This finding is important, as differences in the underlying mechanisms of tauopathy and LBD on SN QSM may distinguish these pathologies by tissue susceptibility in specific SN quadrants.

We previously demonstrated that QSM in the substantia nigra pars compacta (SNc) outperformed QSM of the entire SN in distinguishing LBD [15]. Referring to a postmortem study reporting the lateral portion of the ventral tier of the SNc as the most vulnerable region in LBD [20], the present study further divided the SN into four subregions. Under this geometric definition, the DL and DM quadrants predominantly correspond to the SNc, and the VL and VM quadrants predominantly correspond to the SN reticulata [14, 20]. In Model 3 of the rank regression analysis, iron accumulation was identified as the primary contributor to tissue susceptibility in the VL, DL, and DM quadrants, whereas tau was not (Table 3). In contrast, QSM values in the VM quadrant reflected tau aggregation in addition to iron accumulation. In contrast to vulnerability in lateral part in LBD, tauopathies may preferentially affect medial part of the SN. Previous studies have reported more severe loss of pigmented neurons and higher NFT counts in the medial SN in AD, as well as severe neurodegeneration of the VM region in PSP [20, 55] This regional vulnerability in tauopathies could support the potential of tissue magnetic susceptibility in the VM quadrant as a sensitive biomarker for tauopathies with nigral degeneration.

Pigmented neurons are reportedly more vulnerable than non-pigmented neurons in Parkinson’s disease [24], and in AD with parkinsonism in association with nigral TDP pathology [41]. However, previous literature has not demonstrated a direct association between QSM values and pigmented neuron loss in the SN across parkinsonian disorders [57]. In the present study, rank regression analysis (Model 1 in Table 3), which incorporated iron accumulation and pigmented neuron density, revealed that in the DM quadrant, pigmented neuron density was negatively associated with QSM independent of iron accumulation. This finding suggests that pigmented neuron loss itself—likely driven by pathologies such as α-synuclein, tau, or TDP-43—may contribute to increased QSM values through mechanisms apart from iron deposition measured in the current study.

Model 1 also demonstrated a strong positive effect of iron accumulation (estimate = + 1.057, P < 0.001) and a weaker positive effect of pigmented neuron density (estimate = + 0.598, P = 0.011) on QSM in the VM, along with a significant negative interaction between the two (estimate = − 1.018, P = 0.017). These findings indicate that when pigmented neurons are relatively preserved, the effect of iron accumulation on QSM is modest; however, once pigmented neurons are reduced, the influence of iron accumulation on QSM becomes markedly more pronounced in the VM quadrant. This synergistic effect highlights the potential utility of QSM as a sensitive biomarker for neurodegenerative diseases that predominantly affect pigmented neurons in the SN.

This clinicopathologic study has several limitations. Although our findings suggest direct and indirect contributions of the histopathologic measurements, reproducibility in larger ethnoracially diverse cohorts is needed to establish generalizability. Mediation analyses were performed to evaluate the direct and indirect effects of tau burden and nigral neuron density on QSM via iron accumulation. However, this approach does not account for the mechanism underlying iron-related neurodegeneration in the SN. Incorporation of additional covariates, such as vascular pathology and α-synuclein quantification, may improve our understanding of the contributors to elevated QSM. QSM values did not show a statistical difference across major diagnostic categories (Supplementary Table S4). Given the limited sample sizes within individual categories, analyses in larger cohorts will be required to fully assess potential disease-specific effects. These issues will be addressed in future investigations. Finally, a common limitation of antemortem imaging and pathology correlation studies is the variability in the time interval between imaging and pathology. To address this limitation, we adjusted for the time from MRI to death in our models.

Conclusions

This study demonstrated the contributions of neuropathologic findings to QSM in the SN, highlighting a direct effect of tau in the VM quadrant, independent effects of pigmented neuron loss in the DM quadrant, and a synergistic interaction between iron accumulation and pigmented neuron loss in the VM quadrant. These findings provide a neuropathologic basis for tissue magnetic susceptibility in the SN, supporting the potential applications of QSM as a sensitive biomarker in individuals with diverse underlying neuropathologies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

D.O. and K.K. conceived and designed the study. C.G.S, M.E.M, A.N., R.R.R., M.L.S., J.L.G., C.R.J., T.M., L.K.F., J.A.F., R.S., V.K.R., D.T.J., H.B., E.K.St.L., D.S.K., N.R.G-R., G.S.D., T.J.F., W.K.K., V.J.L., R.C.P., B.F.B., and D.W.D. contributed to data acquisition. D.O., S.K., E.M., S.A.P., and A.F. analyzed data. D.O. wrote the first draft of the manuscript. All authors reviewed and approved the final manuscript.

Funding

This work was supported by the National Institute of Health grants U01 NS10062, R01 AG40042, U01 AG06786, P50 AG16574, P30 AG062677, R37 AG11378, R01 AG41851, and U01 AG082350.

Declarations

Conflict of interest

S.K., E.M., S.A.P., A.F., A.N., R.R.R., J.L.G., T.M., L.K.F., J.A.F., R.S., D.T.J., H.B., N.R.G-R., G.S.D., and W.K.K. report no disclosures relevant to the manuscript. D.O. receives research support from the Japan Science and Technology Agency and the Japan Society for the Promotion of Science. D.W.D receives research support from NIH. C.G.S. receives research support from the NIH. M.L.S. owns or has owned stock in medical-related companies, unrelated to the current work, within the past 36 months: Align Technology, Inc., Inovio Pharmaceuticals, Inc., Mesa Laboratories, Inc., Johnson and Johnson, LHC Group, Inc., Natus Medical Inc., and Varex Imaging Corporation. C.R.J. has consulted for Lily and serves on an independent data monitoring board for Roche and as a speaker for Eisai, but he receives no personal compensation from any commercial entity. He receives research support from NIH and the Alexander Family Alzheimer’s Disease Research Professorship of the Mayo Clinic. T.M. receives research support from National Institute of Health and Zander Family Foundation. V.K.R. receives research funding from the NIH and the Mayo Clinic Dorothy and Harry T. Mangurian Jr. Lewy Body Dementia Program and has provided educational content for Medscape, Expert Perspectives in Alzheimer’s Disease, Clinical Care Options, and Roche/ADLM; has received speaker and conference session honoraria from the American Academy of Neurology Institute; is PI for a clinical trial supported by the Alzheimer’s Association; is site Co-PI for the Alzheimer’s Clinical Trials Consortium; and is a site clinician for clinical trials supported by Eisai, the Alzheimer's Treatment and Research Institute at USC, and Transposon Therapeutics, Inc., unrelated to this work. E.K.St.L. has no disclosures related to this work. He has served as a Co-Investigator on K.K. research grant (DLB U01) and as a NAPS co-investigator. D.S.K. serves on a Data Safety Monitoring Board for the DIAN study, has served on a Data Safety Monitoring Board for a tau therapeutic for Biogen but received no personal compensation, is a site investigator in Biogen aducanumab trials, is an investigator in clinical trials sponsored by Lilly Pharmaceuticals and the University of Southern California, serves as a consultant for Samus Therapeutics, Roche, Magellan Health and Alzeca Biosciences but receives no personal compensation, and receives research support from the NIH. T.J.F. serves as a consultant for Acadia pharmaceuticals and receives funding from NIH and the Mayo Clinic Dorothy and Harry T. Mangurian Jr. Lewy Body Dementia Program. V.J.L. serves as a consultant for Bayer Schering Pharma, Piramal Life Sciences, Life Molecular Imaging, Eisai Inc., AVID Radiopharmaceuticals, and Merck Research and receives research support from GE Healthcare, Siemens Molecular Imaging, AVID Radiopharmaceuticals, and the NIH (NIA, NCI). R.C.P. serves as a consultant for Roche, Inc., Merck, Inc., Biogen, Inc., Eisai, Inc., Genentech, Inc., and Nestle, Inc., served on a DSMB for Genentech, receives royalties from Oxford University Press and UpToDate, and receives NIH funding. B.F.B. has served as an investigator for clinical trials sponsored by Alector, Cognition Therapeutics, EIP Pharma/Cervomed, and Transposon. He serves on the Scientific Advisory Board of the Tau Consortium, funded by the Rainwater Charitable Foundation. He receives research support from NIH, the Mayo Clinic Dorothy and Harry T. Mangurian Jr. Lewy Body Dementia Program, the Little Family Foundation, and the Ted Turner and Family Foundation. K.K. consults for Biogen, Eisai, and BioArctic with no personal compensation, received research support from Avid Radiopharmaceuticals and Eli Lilly, and receives funding from NIH and Alzheimer’s Drug Discovery Foundation.