Imaging Biomarkers for Early Multiple System Atrophy

Abstract

Objective:

To systematically evaluate structural MRI and diffusion MRI features for cross-sectional discrimination and tracking of longitudinal disease progression in early multiple system atrophy (MSA).

Methods:

In a prospective, longitudinal study of synucleinopathies with imaging on 14 controls and 29 MSA patients recruited at an early disease stage (15 predominant cerebellar ataxia subtype or MSA-C and 14 predominant parkinsonism subtype or MSA-P), we computed regional morphometric and diffusion MRI features. We identified morphometric features by ranking them based on their ability to distinguish MSA-C from controls and MSA-P from controls and evaluated diffusion changes in these regions. For the top performing regions, we evaluated their utility for tracking longitudinal disease progression using imaging from 12-month follow-up and computed sample size estimates for a hypothetical clinical trial in MSA. We also computed these selected morphometric features in an independent validation dataset.

Results:

We found that morphometric changes in the cerebellar white matter, brainstem, and pons can separate early MSA-C patients from controls both cross-sectionally and longitudinally (p<0.01). The putamen and striatum, though useful for separating early MSA-P patients from control subjects at baseline, were not useful for tracking MSA disease progression. Cerebellum white matter diffusion changes could aid in capturing early disease related degeneration in MSA.

Interpretation:

Regardless of clinically predominant features at the time of MSA assessment, brainstem and cerebellar pathways progressively deteriorate with disease progression. Quantitative measurements of these regions are promising biomarkers for MSA diagnosis in early disease stage and potential surrogate markers for future MSA clinical trials.

INTRODUCTION

Multiple System Atrophy (MSA) is a rare neurodegenerative disorder characterized by the abnormal aggregation of alpha-synuclein (αSyn) as neuronal or glial inclusions. Unlike Parkinson’s disease (PD), MSA is a rapidly progressive and universally fatal multisystem degenerative disease. It is characterized by autonomic failure, including orthostatic hypotension and neurogenic bladder, plus poorly levodopa responsive parkinsonism or cerebellar ataxia.¹ Misdiagnosis is common at earlier disease stages and by the time patients fulfill clinical consensus criteria and are considered eligible for clinical trials, the disease is often quite advanced with survival less than 3 years.² The failure of essentially all therapeutic trials to date has been at least in part attributed to enrollment of patients at a disease stage that is too advanced for disease-modifying therapies to be effective.^3–5 As there is a number of new therapeutic strategies in the pipeline to slow disease progression in MSA, there is a critical need to develop robust imaging tools for diagnosis and tracking progression of early MSA to facilitate the use of MRI as a diagnostic biomarker on one hand and as a surrogate marker of disease progression and secondary trial outcome measure on the other hand.

MSA results in structural and signal abnormalities on MRI imaging affecting cerebellum, brainstem, and basal ganglia. Signal abnormalities include posterior putaminal hypodensity, a hyperintense lateral putaminal rim, hyperintensity of the middle cerebellar peduncle, and the characteristic “hot cross bun” sign on T2 weighted image. Structural abnormalities relate to atrophy of the putamen, cerebellar peduncles, pons, and cerebellum, and are listed as supportive features in the consensus criteria for MSA.^{6, 7} Although the previous studies have utilized structural and diffusion MRI markers which showed promise for the differential diagnosis as well as tracking the disease progression^8–11 the findings were inconsistent across studies as they used manual segmentation, different variants, and also the data acquired from multiple scanners. Hence a systematic evaluation of information available from structural and diffusion MRI is important.

Our primary goal with this study was to systematically evaluate key morphometric features in structural MRI (sMRI) and diffusion features in diffusion MRI (dMRI) that can aid in both the cross-sectional discrimination as well as tracking of longitudinal disease progression in early MSA patients. In this work, we included early MSA and control participants from the MONITOR study, an observational study at Mayo Clinic as part of our primary cohort (discovery dataset). We followed a three-step process. First, cross-sectional imaging data from MONITOR was used to identify key morphometric features that differentiated MSA-C (predominant cerebellar ataxia subtype) and MSA-P (predominant parkinsonism subtype) participants from controls (CON). Second, we evaluated these key regions for tracking longitudinal progression in MONITOR. Finally, we validated the identified sMRI features in an independent early MSA-C and MSA-P dataset.

MATERIALS AND METHODS

Participants

The primary participants in this study were part of a prospective, longitudinal study of synucleinopathies; Mayo Longitudinal Synucleinopathy Biomarker Study (MONITOR), NIH R01 NS092625. Patients with MSA received a diagnosis of MSA-C or MSA-P by a Mayo Clinic movement disorder specialist. All patients had autonomic function testing to support the diagnosis. In order to be enrolled, patients were required to fulfill consensus criteria for possible or probable MSA and to have a score ≤17 (omitting the erectile dysfunction score) on part one of the Unified MSA Rating Scale (UMSARS) to ensure enrollment at an early disease stage¹. This cutoff has been used in several major trials aiming at disease modification in MSA to define early disease^{4, 12, 13}. Healthy controls were without evidence of neurologic disease or autonomic dysfunction.

Participants were excluded if they were pregnant or breastfeeding, scored 24 points or less on the Mini-Mental Status Examination, had a clinically significant or unstable medical or surgical condition that might preclude safe completion of the study or might affect the results of the study, or had taken any investigational products within 60 days prior to baseline. The discovery dataset from MONITOR consisted of cross-sectional data from 14 controls (CON), 15 MSA-C and 14 MSA-P participants. Of these, 13 CON, 13 MSA-C, and 10 MSA-P participants had follow-up MRI scans available. We also considered 9 Parkinson’s disease (PD) participants with longitudinal MRI in MONITOR for sensitivity analyses. The validation data set included separate prospective cohort of early MSA patients (23 MSA-C and 6 MSA-P) that underwent identical study procedures and had a single MRI available.

Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the Mayo Clinic Institutional Review Board, and written informed consent was obtained from all participants (and their proxies).

Clinical and Laboratory Study Assessments

A medical and neurologic history was obtained from all participants, and all underwent a comprehensive general and neurologic examination. Medications that could potentially bias evaluations were held for 5 half-lives prior to neurologic assessments, autonomic testing, and MRI studies. Neurologic impairment and deficits were quantified in MSA patients using the Unified MSA Rating Scale (UMSARS, consisting of part I which quantifies patients’ symptoms and function, and part II which quantifies findings on neurologic examination).¹⁴ All participants underwent standardized autonomic function testing including autonomic reflex screen and thermoregulatory sweat test. In order to quantify autonomic deficits, the Composite Autonomic Severity Score (CASS), a validated instrument to quantify the overall severity and distribution of autonomic failure based on standardized autonomic testing, was derived.¹⁵ Autonomic symptoms were assessed using the Composite Autonomic Symptom Score, overall COMPASS.¹⁶ Disease duration in years was measured from date of MSA symptom onset to date of first MRI scan.

MRI acquisition and Processing

Discovery dataset: The discovery dataset was acquired on a 3-Tesla Siemens Prisma whole body scanner (Siemens Medical System, Erlangen, Germany), using a 32-channel head coil. The high resolution T1 weighted (T1w) 3D structural scans were obtained using an MPRAGE sequence with 3D distortion correction, repetition time (TR) = 2300 milliseconds (ms), echo time (TE) = 2.95 ms, flip angle = 9°, voxel size = 1.05 × 1.05 × 1.20 millimeter (mm), acquisition matrix 256 × 240, and total scan time = 312 s with 176 sagittal slices. The diffusion MRI acquisition was performed by a multiband (3 × slice acceleration) single shot spin echo axial echo planar imaging (EPI) sequence with the parameters: TR = 3400 ms, TE = 71 ms, flip angle = 90°, acquisition matrix = 116 × 116, voxel size 2.0 mm isotropic, and NEX=1. Data were acquired at 3 different diffusion weightings (b-values): 16 at b = 0, 48 at b = 1000, and 64 at b = 2000 s/mm². The directions were evenly spread over the entire sphere in both shells.¹⁷ Validation dataset: To evaluate the consistency of the computed morphometric imaging measures, we used an independent validation dataset with 23 MSA-C and 6 MSA-P patients scanned on a different scanner – Siemens Skyra (Siemens Medical System, Erlangen, Germany) with similar MPRAGE parameters.

Image Processing: Trained image analysts inspected all the imaging data and rejected scans with serious quality problems such as motion artifacts. Shading artifacts in the sMRI images were corrected using a combination of SPM12 segmentation and N3. We then computed regional MRI morphometric features using Freesurfer v6.0 for all sMRI images for both discovery and validation datasets using the Desikan-Killiany atlas.¹⁸ Middle cerebellar peduncle atrophy, often found in MSA patients, is included as part of the cerebellum in FreeSurfer. The regional volumes calculated as a fraction of the total intracranial volume (TIV) that was estimated in house¹⁹ were used as a morphometric feature for the statistical analysis. dMRI processing: An intracranial mask was made for the diffusion MRI scan,²⁰ the noise in the raw diffusion images was estimated and removed using mrtrix3’s dwidenoise,²¹ and then FSL’s eddy_cuda was used to correct for head motion and eddy current distortion. Gibbs ringing was corrected using unring,²² and Rician bias was removed using dwidenoise’s noise estimate and the algorithm in Koay et al.²³Then diffusion tensors were fit for the multishell data using a nonlinear least squares fitting algorithm implemented in dipy.²⁴ Diffusion tensors were then fit using nonlinear minimization after which Fractional Anisotropy (FA) and Mean Diffusivity (MD) were computed. ANTS²⁵ was used to nonlinearly register an in-house modified version of the JHU “Eve” WM atlas²⁶ to each subject’s FA image to compute regional median FA and MD. Voxels with MD > 2 × 10⁻³ or < 7 × 10⁻⁵ mm²/s were excluded as mostly CSF or air, respectively. ROIs with < 7 diffusion voxels in subject space were excluded as being too small to be reliably registered. MD values were scaled by 10^6 for ease of interpretation

Statistical Analyses: Characteristics of participants in each group were summarized using means and standard deviations for continuous variables and counts and percentages for discrete variables. Sex was compared across the three discovery groups using Fisher’s exact test, and age was compared using a one-way ANOVA. Clinical scores were compared between MSA-C and MSA-P using Welch’s t-test. All comparisons in the validation cohort were done between MSA-C and MSA-P using Welch’s t-test as well. We examined area under the receiver operator characteristics (AUROC) curves for the regional morphometric features and ranked all regions by their ability to distinguish between CON and MSA-C and CON and MSA-P. Each AUROC was tested for significance (vs. the null of 0.5) using a Wilcoxon rank sum test. For the top performing regions based on AUROC, we evaluated the utility for tracking longitudinal disease progression by calculating change from baseline to 12-month follow-up for each participant in the discovery cohort. We identified WM tracts that were top performing for sMRI and plotted diffusion measures in these regions. We used boxplots to visualize the differences in changes among CON, MSA-C and MSA-P groups and consistency of measures in both discovery and validation cohorts. Welch’s t-tests were used to test for differences in changes between CON and MSA-C and CON and MSA-P. We computed sample size estimates via bootstrapping based on the estimated reduction (35%) in MRI rate of change over time. The boot package was used with 5000 repetitions, and the 95% bias accelerated and corrected (BCa) confidence intervals were reported. We used SAS Studio version 9.4 for data management and RStudio version 4.0.3 for analyses.

RESULTS

The participants characteristics are shown in Table 1. There were no differences in sex and age across the groups within the discovery and validation datasets. We only compared the clinical scores between MSA-P and MSA-C because controls perform well on autonomic testing. We found that MSA-C and MSA-P performed similarly on UMSARS, but MSA-P had worse performance on COMPASS compared to MSA-C in both the cohorts. Patients with MSA-P had longer disease duration at baseline than patients with MSA-C. Most participants in the discovery dataset had MRI scans and had useable volume and dMRI data available from the first follow-up visit.

Table 1.

Characteristics table of participants in groups of interest in the discovery and validation datasets. Mean (SD) is listed for the continuous variables and count (%) is listed for the categorical variables.

	Discovery				Validation
	CON (N = 14)	MSA-C (N = 15)	MSA-P (N = 14)	p-value1	MSA-C (N = 23)	MSA-P (N=6)	p-value 1
Males	7 (50%)	10 (67%)	10 (71%)	0.53	18 (78%)	3 (50%)	0.30
Age, yrs.	62.4 (6.2)	59.7 (6.7)	63.1 (5.9)	0.33	57.2 (6.1)	57.6 (6.6)	0.91
COMPASS	4.1 (3.4)	26.4 (13.3)	43.7 (18.6)	0.009*	38.6 (18.5)	59.1 (19.2)	0.075
UMSARS I	0.5 (0.7)	12.2 (3.8)	13.6 (2.5)	0.25*	15.7 (2.1)	17.6 (3.9)	0.36
UMSARS II	0.5 (0.6)	15.5 (4.7)	15.7 (3.7)	0.89*	20.6 (4.2)	20.7 (6.3)	0.97
UMSARS Total	1.0 (1.1)	27.7 (8.1)	29.2 (5.2)	0.53*	36.3 (5.5)	38.3 (9.7)	0.68
Disease duration, yrs.**	n/a	3.2 (2.2)	4.8 (2.2)	0.0499	3.5 (1.7)	3.6 (2.1)	0.93
Follow-up volume available	11 (79%)	13 (87%)	10 (71%)	0.59	n/a	n/a	n/a
Follow-up dMRI available	13 (93%)	11 (73%)	10 (71%)	0.41	n/a	n/a	n/a

¹P-values for differences between diagnostic groups are unadjusted. The statistical test for categorical variables is Fisher’s exact test and the test for continuous variables is an ANOVA for 3-group comparisons and Welch’s t-test for 2-group comparisons.

^*The p-values for comparisons for the clinical scores were only made between MSA-C and MSA-P.

^**In the validation cohort, disease duration data were available for patients with MSA-C and 5 patients with MSA-P.

Structural MRI measures

The area under the ROC (AUROC) in the discovery dataset for distinguishing MSA-C and MSA-P groups from controls is shown in Figure 1. The regions that are significant at p < 0.001 are shown in red. We used a stringent p-value to account for multiple comparisons. The four regions that significantly separated MSA-C from CON at p<0.001 were cerebellar white matter (WM) (AUROC=1.00), pons (AUROC=0.98), superior cerebellar peduncle (AUROC=0.93), and brainstem (AUROC=0.93). There was a much greater involvement of cerebellum WM in comparison to cerebellum gray matter (GM) which also survived p-value threshold of p<0.05. The two regions that separated MSA-P from CON were putamen (AUROC=0.93) and striatum (AUROC=0.90) with significant differences also seen in the cerebellum WM. We had sufficient samples to detect these differences as we had 80% power to detect an AUROC as small as 0.783 in the control vs. MSA-C comparison and as small as 0.788 in the control vs. MSA-P comparison.

Figure 1.

AUROC forest plots with 95% confidence intervals for distinguishing between control and MSA patients for MRI volumes. Red indicates p < 0.001, green indicates 0.001 < p < 0.05, and black indicates p > 0.05. AUROC values are written for the most significant regions. P-values are from an unadjusted Wilcoxon rank-sum test. WM=white matter; GM=gray matter; SCP=superior cerebellar peduncle.

Figure 2 shows the box plots of longitudinal change in volumes over one year using the most significant regions from the AUROC plots in Figure 1. Change in volumes of brain stem and pons were all significantly different between CON and MSA-C and CON and MSA-P (p<0.01). These results provide evidence that regions that showed lower volume at baseline in MSA-C and MSA-P also had significant decline over time.

Figure 2.

Boxplots of change in adjusted volume between the second and first MRI scan for patients in the control, MSA-C and MSA-P groups, using the most significant regions from the AUROC plots in Figure 1. An unadjusted Welch t-test was used for between-group comparisons. Regions separating the two groups at p<0.01 are marked. WM=white matter; SCP=superior cerebellar peduncle.

Diffusion measures

For dMRI, we focused on the two white matter tracts in the JHU atlas that were impacted on structural MRI scans – brainstem and cerebellar WM. At p<0.001 threshold, cerebellum WM FA (AUROC=1.0) and cerebellum WM MD (AUROC=1.0) separated MSA-C from controls. Figure 3 shows the box plots of longitudinal change in MRI measures over one year for cerebellum WM tract and brainstem, showing that cerebellum WM MD significantly increased in both MSA-C and MSA-P.

Figure 3.

Boxplots of change in FA and MD between the second and first MRI scan for patients in the control, MSA-C and MSA-P groups of the MONITOR study, using the most significant regions from the AUROC plots. MD values are scaled by 10^6 for ease of interpretability. P-values for comparisons of longitudinal change in mean FA and MD using unadjusted Welch’s t-test. Regions separating the two groups at p<0.01 are marked. WM=white matter; FA=fractional anisotropy; MD=mean diffusivity.

Utility of MRI markers for Clinical Trials

For the usefulness of imaging measures in clinical trials, two important features are reproducibility of the measures in independent cohorts as well as sufficient power to detect changes over time. In Figure 4, we plotted the gray matter volumes subset from Figure 1 in both the discovery and validation datasets. Even though these are data from independent groups of MSA patients acquired on a different MRI scanner with similar acquisition parameters, the range of data across the two datasets are highly similar providing indirect evidence that imaging data could be similar in a harmonized clinical trial where data was acquired on two different scanners. In Supplemental Table 1, we provide sample size estimates for detecting 35% reduction in rate of change of longitudinal measurements in Figure 2 and 3. Sample sizes for a composite dataset of MSA-C and MSA-P were calculated via bootstrapping. In a hypothetical clinical trial in MSA, brainstem (sample size estimates [95% confidence interval] = 41 [24, 88]) and pons (49 [29, 85]) measurements from sMRI and cerebellar WM MD (73 [43,150]) from dMRI would be most useful as outcomes to measure efficacy of the clinical trial.

Figure 4.

Boxplots of TIV-adjusted baseline volume by diagnostic group and study. The gray background on the right corresponds to boxplots for patients in a validation dataset. WM=white matter; SCP=superior cerebellar peduncle.

DISCUSSION

In this study, we performed a systematic cross-sectional and longitudinal comparison of sMRI and dMRI brain changes observed in MSA-C and MSA-P patients compared to controls. The main conclusions were that: 1) morphometric changes in the cerebellar WM, brainstem, and pons were able to separate early MSA-C patients from control participants both cross-sectionally and longitudinally (p<0.01); 2) though morphometric changes in the putamen and striatum were able to separate early MSA-P patients from control participants cross-sectionally, they did not provide the best measurements for tracking progression of disease in MSA-C and MSA-P. This provides support for MSA as a condition affecting greater atrophy of brainstem and cerebellar pathways with disease progression regardless of clinically predominant feature at the time of assessment; and 3) cerebellum WM dMRI changes were significantly different between MSA patients and controls cross-sectionally and cerebellum WM MD changes were significant longitudinally (p<0.01). These findings suggest that quantitative MRI analysis of selected regions represents a promising biomarker for the diagnosis of MSA and a potential surrogate marker of disease progression for future clinical trials in MSA.

Cerebellum white matter and brainstem as biomarkers in MSA

MSA is a rare neurodegenerative disorder. This characteristic has made it difficult to acquire large imaging datasets to compute sample size estimates for MRI as surrogate marker of disease and disease progression for future clinical trials that are on the horizon. However, a systematic evaluation of both cross-sectional discrimination and longitudinal change as presented here based on a single center experience can aid in identifying key imaging features to aid diagnosis and assessment of disease progression for clinical as well as research implications. In this work, we found that there was greater involvement of cerebellum WM and brainstem in both MSA-C and MSA-P. The cerebellar findings are in line with a recent meta-analysis of voxel level analyses across all studies between 1974–2020 confirming cerebellum atrophy in MSA patients in general.²⁷ By comparing the GM and WM changes, we were able to discern that WM changes were more extensive in MSA. The pathological observations of greater accumulation of misfolded αSyn in oligodendrocytes²⁸ though supportive of these findings, the numerical confirmation here provides strong evidence for the use of cerebellum WM as a biomarker in MSA.

The clinico-pathological correlations with brainstem involvement in MSA have been widely discussed.²⁹ Since both MSA-C and MSA-P had significant longitudinal changes in the brainstem (Figure 2), we found that brainstem volume provided the lowest sample size estimate needed to document a disease modifying effect for MSA as a single entity among all morphometric measures. Greater putamen and striatum involvement were seen in MSA-P at baseline as expected because basal ganglia degeneration is the underlying driver for parkinsonian symptoms. However, changes in these regions were not extensive in MSA-P longitudinally in comparison to brainstem and pons suggesting that basal ganglia involvement is likely early in the disease process, which is followed by more discernable degeneration in other areas known to be affected by MSA. Degeneration of brainstem and pons over time was greater in both early MSA-C and MSA-P supporting the utility of these morphometric measurements as outcome measures in clinical trials. Though not the focus of the paper, we also computed the volume loss in these regions to the volume loss seen in PD (the comparison group available in MONITOR study) as part of a sensitivity analysis. Supplemental Figure 1 shows that the volume change in PD is minimal and was similar to controls confirming that the morphometric features we identified here are specific to MSA. Superior cerebellar peduncle, which is related to midbrain atrophy and a key area of involvement in progressive supranuclear palsy, was significantly different between MSA-C and controls cross-sectionally but the change in volume over time did not differ between groups as seen in Figure 2.

Though dMRI analyses were exploratory, we found intriguing change over time and low sample size estimates with cerebellum WM MD. dMRI measures in cerebellum WM performed better than volume measures in the same region suggesting that it may be an earlier and more sensitive indicator of neurodegeneration than volume. Poewe et. al. investigated several regional dMRI as outcome measures to test the efficacy of treatment in MSA-P patients. The work here allows the testing of a limited number of regions as markers of disease progression³⁰. Further work is warranted to establish the utility of dMRI as an imaging biomarker in MSA. Future work leveraging methods for multi-shell acquisitions such as NODDI can further shed light on the disease mechanisms in MSA providing more biologically meaningful measurements.

Utility of MRI for clinical diagnosis and tracking of disease progression

Misdiagnosis is common in MSA, particularly among community neurologists.³¹ Even though autonomic biomarkers can enhance the certitude of the clinical diagnosis and provides excellent agreement between the clinical and pathological diagnosis, autonomic testing is only available at a few specialized centers, and it may be less informative at early disease stages. Thus, MRI can play an important role for clinical diagnosis and prognosis in early disease stages of MSA.

MRI has been studied extensively in MSA with reported findings including prominent cerebellum, brain stem, basal ganglia changes and of course the classic, well recognized sign of “hot-crossed-bun”. Even though the current MSA consensus¹ supports the use of MRI changes as additional features to support the diagnosis, the literature has suggested that these visual changes are often not detected until a more advanced disease stage.^{32, 33} Majority of the studies so far have focused on manual measurements and voxel based analyses which limit the wide usage of these methods in clinical studies. ^34–39 This work presents a systematic, prospective exploration of the whole brain morphometric and dMRI along with an independent validation dataset. These results highlight the usefulness of MRI for diagnosis and tracking of disease progression in MSA and support its utility as a surrogate for disease in MSA.

There are some limitations to this study. We limited our focus to early MSA and did not have imaging data to consider other late-onset cerebellar ataxias and other parkinsonisms as comparison groups.