Introduction

Deep gray matter (DGM) iron dyshomeostasis has been linked to neuroinflammation in people with multiple sclerosis (pwMS). It remains unclear if short-term (<1 year) changes can indicate MS-related iron dyshomeostasis. In the present work, we hypothesized that a novel correlation analysis of over-time magnetic susceptibility changes between DGM regions exposes short-term MS-related iron dyshomeostasis that is undetectable with conventional ROI-based methods. We then used in silico simulations to test whether small aging-related susceptibility changes could explain the observed network behavior. This novel method may inform understanding of neuroinflammatory-related pathology and has potential clinical utility due to its high sensitivity.

Methods

MRI Acquisition, Reconstruction, and ROI analysis
29 pwMS and 29 healthy controls (HCs) underwent at least three MRI scans within one year. Imaging protocols included 3D gradient-echo (GRE) and high-resolution T1w imaging at 3T. GRE imaging parameters were: matrix size of 512x192x64, voxel size of 0.5x1x2mm³, flip angle of 12°, TE/TR of 22ms/40ms, and bandwidth of 13.89kHz. Susceptibility maps were generated as previously described.¹ We segmented 14 DGM regions in both hemispheres using FSL FIRST after registering the T1w images to the susceptibility maps: accumbens, amygdala, caudate, hippocampus, pallidum, putamen, and thalamus.

Assessment of Inter-Regional Correlations in vivo
The proposed network approach assesses the relationship of susceptibility fluctuations between regions. Susceptibility values were Pearson-correlated across timepoints, resulting in subject-specific Pearson's R values for each of the 196 region-pairs. Subject-specific, region-average correlation values for each region were obtained by averaging the correlations between the region and all other regions.

Simulation of Aging-related Inter-Regional Correlations (in silico)
We evaluated aging-related susceptibility changes as a potential mechanism underlying the short-term correlations by comparing the real data to simulated data. Susceptibility changes were simulated by inputting baseline ages and follow-up times from the HC group into aging trajectories of magnetic susceptibility previously published for 6 bilateral DGM structures (amygdala, caudate, hippocampus, pallidum, putamen, and thalamus).² These changes were added to each subject's baseline susceptibility along with zero-mean Gaussian noise (~N(0, 0.0001), selected heuristically) to simulate short-term susceptibility values for each DGM region. Pearson's R values were calculated for each region-pair, averaged across subjects, and compared to the in vivo data (excluding accumbens) to assess the similarity of correlation patterns between the simulated aging and observed in vivo data.

Statistics
Subject age was compared between HC and MS groups using t-tests, and sex was compared using a chi-squared test. Changes in ROI-specific mean susceptibilities and whole-DGM correlations were compared between groups using independent samples t-tests. The number of negative R-values in the MS and HC correlation matrices were compared using chi-squared tests. Alpha < 0.05 was considered statistically significant.

Results

There were no significant differences between pwMS and HCs in baseline age (42.1 years for HC vs 42.6 years for MS, p = 0.884) or sex (20/29 female for HC vs 20/29 female for MS, p = 1.0).

Average correlation matrices for pwMS and HCs are shown in Fig. 1. The MS-minus-HC subtraction matrix (Fig. 2) shows most correlation values decrease in the MS group compared to the HC group. The MS correlation matrix had about 50% more negative R-values than the HC group (82/182 for MS vs 52/182 for HC, p = 0.001).

Simulated R-values (Fig. 3, bottom) demonstrated a pattern that closely resembled the pattern of the real R-values (Fig. 3 left; Pearson's R = 0.50, p < 0.001).When comparing susceptibility changes between pwMS and HCs using the standard ROI-based method, only the right amygdala exhibited statistical significance (0.71 ppb for MS group vs -6.7 ppb for HC group, p = 0.043). In contrast, the average short-term correlations of 7/14 regions showed a significant decrease in pwMS compared to HCs (Fig. 4): left amygdala (-0.006 vs 0.163, p = 0.038), left hippocampus (0.198 vs 0.319, p = 0.0406), left pallidum (0.094 vs 0.242, p = 0.040), left putamen (0.193 vs 0.361, p = 0.016), left thalamus (0.007 vs 0.248, p < 0.001), right putamen (-0.005 vs 0.190, p = 0.008), and right thalamus (0.075 vs 0.233, p = 0.029).

Discussion

Network analysis revealed neuroinflammatory-related brain iron abnormalities that were not detected using the standard ROI-based approach. The average correlation values were more sensitive to pwMS vs HC differences than the standard ROI-based approach, possibly because correlations are independent of variations in susceptibility reference values, which is a considerable source of within-subject variability in QSM. The observation of decreased correlations in the MS-minus-HC correlation difference matrix (Fig. 2) and a greater number of negative R-values in pwMS compared to HCs suggests a negative feedback mechanism to iron change. In the MS group, negative correlations were prominent in the left thalamus. This region regulates iron influx in other DGM regions in mice.³

Our simulations showed that minuscule age-related changes in susceptibility largely explained the correlation structure seen in healthy controls, indicating that the MS-related susceptibility variations are related to a disruption of normal aging-related iron changes.

Conclusion

Our novel methodology was able to detect neuroinflammatory-related susceptibility abnormalities in MS over a significantly shorter timescale than previously possible. This work may have implications in understanding and early diagnosis of neuroinflammatory diseases such as MS.

Acknowledgements

No acknowledgement found.