Mapping of Thalamic Magnetic Susceptibility in Multiple Sclerosis Indicates Decreasing Iron with Disease Duration: A Proposed Mechanistic Relationship between Inflammation and Oligodendrocyte Vitality

Abstract

Recent advances in susceptibility MRI have dramatically improved the visualization of deep gray matter brain regions and the quantification of their magnetic properties in vivo, providing a novel tool to study the poorly understood iron homeostasis in the human brain. In this study, we used an advanced combination of the recent quantitative susceptibility mapping technique with dedicated analysis methods to study intra-thalamic tissue alterations in patients with clinically isolated syndrome (CIS) and multiple sclerosis (MS). Thalamic pathology is one of the earliest hallmarks of MS and has been shown to correlate with cognitive dysfunction and fatigue, but the mechanisms underlying the thalamic pathology are poorly understood.

We enrolled a total of 120 patients, 40 with CIS, 40 with Relapsing Remitting MS (RRMS), and 40 with Secondary Progressive MS (SPMS). For each of the three patient groups, we recruited 40 controls, group matched for age- and sex (120 total). We acquired quantitative susceptibility maps using a single-echo gradient echo MRI pulse sequence at 3 Tesla. Group differences were studied by voxel-based analysis as well as with a custom thalamus atlas. We used threshold-free cluster enhancement (TFCE) and multiple regression analyses, respectively. We found significantly reduced magnetic susceptibility compared to controls in focal thalamic subregions of patients with RRMS (whole thalamus excluding the pulvinar nucleus) and SPMS (primarily pulvinar nucleus), but not in patients with CIS. Susceptibility reduction was significantly associated with disease duration in the pulvinar, the left lateral nuclear region, and the global thalamus. Susceptibility reduction indicates a decrease in tissue iron concentration suggesting an involvement of chronic microglia activation in the depletion of iron from oligodendrocytes in this central and integrative brain region. Not necessarily specific to MS, inflammation-mediated iron release may lead to a vicious circle that reduces the protection of axons and neuronal repair.

Graphical abstract

1. Introduction

Atrophy of the thalamus is one of the earliest hallmarks of brain pathology in Multiple Sclerosis (MS) (Audoin et al., 2009; Bergsland et al., 2012; Calabrese et al., 2011; Henry et al., 2009; Henry et al., 2008; Ramasamy et al., 2009; Zivadinov et al., 2013), which correlates with physical disability (Rocca et al., 2010) and fatigue (Calabrese et al., 2011), and continues with the progression of the disease (Bergsland et al., 2012; Henry et al., 2008; Preziosa et al., 2017; Ramasamy et al., 2009). Thalamic atrophy has also been shown to correlate with cognitive dysfunction (Batista et al., 2012; Bergsland et al., 2016; Bisecco et al., 2017; Houtchens et al., 2007) in MS, which is in line with other studies showing an association between thalamocortical connectivity and diverse functions of higher level cognitive processes (Fama and Sullivan, 2014), including attention, speed of information processing, working memory, and episodic memory processes (Hughes et al., 2012; Philp et al., 2014; Ystad et al., 2010; Ystad et al., 2011).

However, despite its potential direct involvement in cognitive dysfunction, disability, and disease progression, comparatively little is known about the mechanisms of thalamic involvement in MS. One reason for the relative scarcity of mechanistic studies is that the thalamus exhibits similar MR-relevant properties as circumjacent white matter (WM) tissues (Kanowski et al., 2014; Tourdias et al., 2014), like T₁ and T₂ relaxation times and proton density, which hampers in vivo investigations with clinical MRI. In particular, the anatomical complexity of the human thalamus, which consists of approximately 100 distinct cell groups or nuclei (Axer and Niemann, 1994) with distinct connectivity profiles (Fama and Sullivan, 2014; Postuma and Dagher, 2006; Sherman and Guillery, 2013b), has been inaccessible in vivo until recently. The increasing availability of higher magnetic field strengths and advanced MRI acquisition and analysis techniques has enabled imaging studies with improved resolution and sensitivity toward new biophysical tissue properties, recently enabling a more detailed in vivo assessment of the thalamus (Abosch et al., 2010; Behrens et al., 2003; Bisecco et al., 2015; Unrath et al., 2008; Wiegell et al., 2003). These developments have increased the interest in exploring the involvement of the thalamus in neurological diseases (Minagar et al., 2013).

Quantitative Susceptibility Mapping (QSM) is such a novel advanced MR-based technique (Duyn, 2013; Haacke et al., 2015; Liu et al., 2015; Reichenbach et al., 2015; Schweser et al., 2011; Schweser et al., 2016; Wang and Liu, 2015) that allows for the precise anatomical depiction of intra-thalamic nuclei and even allows quantifying tissue property alterations with high spatial resolution (Deistung et al., 2013) and reproducibility (Deh et al., 2015; Feng et al., 2017; Lin et al., 2015; Santin et al., 2017) at clinically feasible scan times. Various histological validation studies have demonstrated that magnetic susceptibility, the quantity provided by QSM, reflects the tissue concentrations of paramagnetic iron complexes (Langkammer et al., 2012; Schenck, 1992; Stüber et al., 2014; Stüber et al., 2016; Zheng et al., 2013) as well as, in an opposite way, myelin (Groeschel et al., 2016; Schweser et al., 2011; Stüber et al., 2014) and calcium (Chen et al., 2014b; Schweser et al., 2010; Straub et al., 2016a; Stüber et al., 2014). Within the MS research, QSM is increasingly being used for the characterization of iron load in the deep gray matter (DGM) (Al-Radaideh et al., 2013; Blazejewska et al., 2015; Hagemeier et al., 2017; Langkammer et al., 2013; Ropele et al., 2017; Rudko et al., 2014; Schmalbrock et al., 2016) and lesions (Bian et al., 2016; Chen et al., 2014a; Cronin et al., 2016; Eskreis-Winkler et al., 2014; Harrison et al., 2016; Kakeda et al., 2015; Li et al., 2016; Wisnieff et al., 2015; Zhang et al., 2016).

In the present study, our central hypothesis was that MS is associated with increased magnetic susceptibility in the thalamus. We further hypothesized that the most substantial differences between patients and controls would be observed in the pulvinar (PUL) nucleus and the lateral nuclear region (LNR) because of potential trans-synaptic degeneration emerging from the motor and visual cortices, regions that are highly affected by MS (Calabrese et al., 2007) and maintain relatively rich structural connectivity with the PUL and LNR. We based our hypothesis on a significant body of literature on DGM iron accumulation in MS (Ropele et al., 2017; Stephenson et al., 2014; Stüber et al., 2016) as well as the well-documented fact that MS leads to demyelination and atrophy. Both the accumulation of iron and loss of myelin increase the tissue’s magnetic susceptibility, which leads to a hyper-intense appearance on the susceptibility maps. Furthermore, thalamic atrophy, which occurs early in the course of MS due to demyelination and neurodegeneration, would increase the observed susceptibility simply because of a condensation of the iron present in the tissue.

Several previous MRI studies indicated increased thalamic iron in MS (Drayer et al., 1987; Rudko et al., 2014; Zivadinov et al., 2012), such as our previous work with susceptibility-weighted imaging (SWI) MRI, in which PUL was atrophied and signal changes inidcated an increased susceptibility early in the disease course of pediatric and clinically isolated syndrome (CIS) patients (Hagemeier et al., 2012; Hagemeier et al., 2013b). Further support for our hypothesis was provided by previous studies using positron emission tomography (PET) (Banati et al., 2000; Herranz et al., 2016; Kauzner et al., 2016; Rissanen et al., 2014) and histopathology (Haider et al., 2014; Vercellino et al., 2009) indicating increased microglia activation and influx of highly iron-laden macrophages in the thalami of MS patients, respectively. Also, histopathologic evidence exists for substantial focal demyelination and neuronal loss in the thalamus (Cifelli et al., 2002; Haider et al., 2014; Vercellino et al., 2009). Overall, previous literature strongly argued for a susceptibility increase in the thalamus of MS patients.

To test our hypothesis, we used QSM to examine intra-thalamic alterations of magnetic susceptibility in patients with CIS and MS as compared to matched healthy controls. QSM combines two crucial properties for the study of such a complex structure as the thalamus: first, it delivers high resolution images and enables the depiction of anatomical details that have been inaccessible with other in vivo imaging techniques, forming the basis of a volumetry of the thalamic substructures. Second, QSM represents a unique tool to assess the tissue composition via quantitative measurements of the tissue’s magnetic susceptibility. We combined volumetry and susceptibility quantification to dissect thalamic iron and myelin mass changes from apparent susceptibility increases that may result from atrophy without actual iron transported into the brain across the blood-brain barrier. A voxel-based analysis (VBA) of the susceptibility maps was applied to reveal disease-related thalamic susceptibility changes without prior assumptions about the location of the changes within the thalamus.

We compared each MS phenotype separately against a dedicated control group that was group matched for age and sex to the different MS/CIS cohort because MS-specific thalamic pathology develops on top of a well-established background of normal aging-related linear atrophy rates (Cherubini et al., 2009; Sullivan et al., 2004), and non-linear tissue iron concentrations. In particular, the latter peaks in the fourth decade of life and decreases thereafter (Bartzokis et al., 2007; Hagemeier et al., 2013a; Hallgren and Sourander, 1958; Mitsumori et al., 2009; Persson et al., 2015), which is difficult to model statistically, complicates the comparison of groups that are not perfectly age-matched, and could affect the comparison of the different clinical phenotypes of MS, including CIS, Relapsing-Remitting (RR) and Secondary Progressive (SP) patients. These disease-subtypes are difficult to match properly on age and typically center on average ages right before, at, and after the characteristic peak of iron concentration in the thalamus, respectively.

2. Subjects and Methods

2.1 Subjects

The study was approved by the local Ethical Standards Committee at the University at Buffalo, and a written informed consent form was obtained from all participants. We enrolled 40 patients in each of the three patient groups (CIS, RRMS, SPMS; 120 patients total) and 40 normal controls (NC) in each of the three control groups (CIS-NC, RRMS-NC, and SPMS-NC; 120 NCs total). The female to male sex ratio was approximately 3, reflecting the average sex incidence rate ratio of MS in the general population of North America (Trojano et al., 2012). Exclusion criteria were pregnancy and pre-existing medical conditions known to be associated with brain pathology (e.g., cerebrovascular disease or a positive history of alcohol dependence). MS/CIS patients were diagnosed using the revised McDonald criteria (Polman et al., 2011), and clinical disease severity was measured using the Expanded Disability Status Scale (EDSS) (Kurtzke, 1983). Table 1 summarizes the demographics and clinical details of the study groups. Patients with CIS had the lowest EDSS and disease duration, followed by patients with RRMS, and patients with SPMS. The average ages of the three patient groups were significantly different (CIS/RRMS: p=0.004; RRMS/SPMS: p<0.001; CIS/SPMS: p<0.001).

Table 1.

Demographics and clinical details of the study groups in this work.

	Patients			Normal controls (NC)
	CIS	RRMS	SPMS	CIS-NC	RRMS-NC	SPMS-NC	All patients	All NCs
N	40	40	40	40	40	40	120	120
Age (years)	36.9±9.8 [20–58]	43.6±10. [27–65]	52.0±7.0 [33–64]	37.0±11.6 [20–59]	44.0±10.0 [19–60]	52.6±11.9 [23–76]	44.1±10.9 [20–65]	44.5±12.8 [19–76]
Sex ratio (F/M)	29:11	27:13	29:11	28:12	30:10	29:11	85:35	87:33
EDSS	1.5, 0.5– 2.5	2.0, 1–3	6.5, 4–9	–	–	–	2.5, 0–7	–
dd (years)	2.2±2.6	9.8±6.0	24.0±10.2	–	–	–	11.9±11.4	–

Values stated as n, m-p represent median, IQR; n±m indicates mean and standard deviation; [m-p] indicates The range. The abbreviation dd stands for disease duration and N is the number of subject.

The three groups of each 40 age- and sex-matched normal controls (NC; CIS-NC, RRMS-NC, and SPMS-NC) had a normal neurological examination and no history of neurologic disorders or chronic psychiatric disorders. Neither ages nor sex-ratio of the NC groups were significantly different from the respective patient groups (CIS-NC: age p=0.96, sex p=0.81; RRMS-NC: age p=0.84, sex p=0.46; SPMS-NC: age p=0.79, sex p=1.00). The average ages of the three NC groups were significantly different (CIS-NC/RRMS-NC: p=0.005; RRMS-HC/SPMS-HC: p=0.001; CIS-HC/SPMS-HC: p<0.001).

2.2 MRI

Participants were imaged with a clinical 3T GE Signa Excite HD 12.0 scanner (General Electric, Milwaukee, WI, USA) using an eight-channel head-and-neck coil. Data for QSM were acquired using an unaccelerated 3D single-echo spoiled gradient recalled echo (GRE) sequence with first-order flow compensation in read and slice directions, a matrix of 512×192×64 and a nominal resolution of 0.5×1×2 mm³ (FOV=256×192×128 mm³), flip angle = 12°, TE/TR=22ms/40ms, bandwidth=13.89 kHz, and a total measurement time of 8 minutes and 46 seconds (Zivadinov et al., 2012). The anisotropic voxel size resulted from a reduction of the number of phase and slice encoding steps to minimize the total measurement time and, hence, motion artifacts. Because the scanner software did not allow an online reconstruction of phase images for QSM, we saved the raw k-space data for each coil channel. To allow a determination of the total brain volume, we applied an axial high-resolution 3D magnetization prepared T₁-weighted (T1w) fast spoiled gradient-echo pulse sequence with inversion recovery (IR-FSPGR) using the following parameters: TE/TI/TR=2.8ms/900ms/5.9ms, matrix=256×192×128 matrix, nominal resolution of 1×1×1.5mm³ (FOV=256×192×192mm³), and flip angle = 10°. No hard- or software upgrades of the MRI system occurred during the duration of the study.

2.3 Data Processing and Analysis

The data processing is schematically illustrated in Figure S.1. The analysts performing the data processing were blinded to the study groups and the scope of the study.

QSM

QSM-related processing was performed by a fully automated pipeline with in-house developed MATLAB programs (2013b, The MathWorks, Natick, MA) on a Linux workstation (Ubuntu 12.04) with 48 cores (Intel Xenon E5-2697v2 at 2.7Ghz) and 396 GB RAM. Susceptibility maps were reconstructed as described previously (Hagemeier et al., 2017); details are given in the Supplementary Material 1.

Voxel-based analysis (VBA)

For the VBA, susceptibility maps were normalized with the diffeomorphic Greedy-SyN transformation model (Advanced Normalization Tools; version 2.1; http://stnava.github.io/ANTs) to an in-house generated susceptibility brain template with 1mm³ isotropic voxel size. The template had been created independently from the present study dataset based on sixty randomly selected susceptibility maps of patients with different diseases and NCs over a wide range of ages using the diffeomorphic Greedy-SyN transformation model with an intensity rescaling strategy (DIR-R) described previously (Hanspach et al., 2017). The template is illustrated in Figure 1.

Figure 1.

Brain template and custom thalamus atlas. Pulvinar (PUL) indicated in blue, lateral nuclear region (LNR) in yellow, and medial nuclear region (MNR) in green. The global thalamus (GT) is indicated in red (other areas overlaid). The white circle indicates the location of the slices used in the enlarged views in Fig. 2. The cutout view on the left visualizes the three-dimensional spatial location in the brain.

After smoothing with a 1mm Gaussian kernel, we compared the control groups to one another to determine if previously reported aging-related differences in iron concentrations between the groups could be replicated. Furthermore, we performed a voxel-wise statistical analysis via non-parametric permutation tests (FSL randomise (Winkler et al., 2014); 5000 permutations) using age and sex as covariates to identify susceptibility differences between patient groups and their respective control groups. Threshold-free cluster enhancement (TFCE) while controlling for family-wise error (FWE) rate revealed significant differences between groups at the level of p<0.05. Given the study objectives, we restricted the statistical analysis to voxels within the thalamus, which also increased the statistical power. The procedure used to define this region is described below.

Atlas-based analysis

To gain insight into volume changes of thalamic subnuclei and to quantify susceptibility differences between groups, we applied an atlas approach. Toward this end, we created a custom thalamic nuclear atlas by identifying and outlining clearly identifiable thalamic substructures on the susceptibility brain template with Freeview (FreeSurfer 5.3.0, Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA). The atlas was created slice-by-slice by a trained image analyst (A.L.R.D.M.; 2 years of experience in neuroimaging) in consultation and based on consensus with an expert in brain susceptibility contrast (F.S.; 9 years of experience in neuroimaging). Identification of substructures was facilitated by a direct comparison of the template with cytoarchitectonic plates of the thalamus in the Schaltenbrand and Wahren atlas (Schaltenbrand et al., 1977), which has been shown to be the most suitable atlas for a segmentation of in vivo MR images (Deistung et al., 2013; Kanowski et al., 2014). Anatomical regions that appeared as a contiguous structure on the susceptibility template and could not be sub-segmented were treated as a single structure in the atlas, even if cytoarchitectonic plates and ultra-high field QSM (Deistung et al., 2013) suggested subdivisions. In addition to thalamic subnuclei, we outlined the thalamus as a whole (global thalamus; GT) to facilitate a comparison with previous studies.

We transformed the atlas to the original susceptibility maps with nearest neighbor interpolation by applying to the atlas the inverse of the non-linear transformations to the template space. We calculated the volume of each region of the warped susceptibility atlas and normalized it to the head volume, as determined by FMRIB’s SIENAX cross-sectional software tool (version 2.6) (Smith et al., 2002).

Statistical analysis of regional average values of volume and susceptibility

The statistical analysis of regional average values was performed using the Statistical Package for the Social Sciences (SPSS; version 24; IBM, Armonk, NY) and Excel (version 1701; Microsoft, Redmond, WA). All sample distributions were tested for normality using the Shapiro-Wilk test and visual examination of Q-Q-plots. If distributions were normal, paired t-tests were used to determine whether mean values differed between left and right hemispheres for each anatomical region. If distributions were non-normal, we used the Wilcoxon signed rank test. If a significant inter-hemispheric difference was found in either the patient group or the respective control group, further analyses of the structure were carried out for both hemispheres separately; otherwise, further analyses used the mean of left and right hemisphere values. Univariate ANCOVA with sex as a covariate was applied to determine if mean values of controls and patients differed significantly from one another.

We have previously estimated that increased caudatal susceptibility observed in MS may be driven by a loss of tissue compartments with little iron (Hagemeier et al., 2017). To understand if this effect could also be the driving force leading to increased PUL susceptibility, we calculated the Pearson correlation coefficients for volume and susceptibility. For simplicity, this correlation analysis was performed for bi-hemispheric average values. To investigate the presumably nonlinear effect of normal aging on thalamic susceptibility and volume, we calculated the Spearman rank correlation coefficients for susceptibility and volume, respectively, using all 120 subjects of the three NC groups.

Effect sizes were estimated using Cohen’s d. Statistical significance levels were corrected for FWE rate using the Bonferroni procedure. Findings with p≤0.05 that did not reach statistical significance after the correction are reported as “trends” to counter-balance the conservative nature of the multiple comparison corrections.

To characterize the relative effects of age, disease duration (dd), and sex on the observed variables, we performed a multiple regression analysis. Variable transformations were performed if histograms and scatterplots indicated they were necessary. Since results from multivariate analyses with transformed variables are difficult to interpret, we repeated all analyses with untransformed variables. Reliability of fitting coefficients was determined by performing a collinearity analysis with a condition index threshold of 15 and variance proportions exceeding 90 in two or more variables.

3. Results

We report anatomical locations within the thalamus according to the recommendations of the Federative Committee on Anatomical Terminology (Federal Committee on Anatomical Terminology, 1998) and use neurological display convention in all figures (subject’s left is shown on the left).

Voxel-based analysis

Susceptibility differences between the three NC groups, CIS-NC, RRMS-NC, and SPMS-NC, did not reach statistical significance (F-test). For demonstrative purposes, we calculated the group differences and Z-score maps of the control groups, which may be found in Supplementary Figure S.2. While some nuclear groups could be discerned on these maps, group-average difference maps had a noisy appearance, and Z-scores did not exceed 0.5.

Figures 2a and S.3 show the regions of significant susceptibility differences between patients and controls after Threshold-Free Cluster Enhancement (TFCE). Figures 2b and S.4 show the corresponding group-average difference maps and Z-score maps. While the comparison of CIS patients with CIS-NCs did not reach statistical significance, it indicated a relatively localized disease-related susceptibility increase in the PUL and susceptibility decrease in the remaining part of the thalamus, Z-scores were low overall.

Figure 2.

Results of the voxel-based analysis. (a) Regions in which differences between patients and controls reached statistical significance. Shown are voxels with a p-value below 0.05 (color-coding) after TFCE in selected slices (white circles in Fig. 1). The corresponding overview of the results in the whole thalamus is shown in Fig. S.3 in the Supplementary Material. (b) Voxel-based group-average differences of magnetic susceptibility between the patient and their corresponding NC groups (top to bottom) in selected slices, illustrating the magnitude and direction of the group differences in every voxel. Note the differences in contrast between the groups (color bars). The corresponding overview of the results in the whole thalamus is shown in Fig. S.4 along with the Z-scores

RRMS patients showed a statistically significant susceptibility reduction compared to RRMS-NCs, which reached statistical significance primarily in the left thalamus, whereas the right thalamus was largely unaffected apart from a small region with susceptibility reduction located in the central medial intralaminar nucleus of the thalamus (arrow I in Fig. 2a and S.3; between the medial dorsal nucleus and the anterior nuclei of the thalamus). The reduced susceptibility in the left hemisphere was localized in the medial dorsal nucleus (arrow II) and the ventral posterolateral nucleus (also nuclei ventrocaudales; arrow III). Reduced susceptibility was also observed bilaterally in a small region localized at the medial pulvinar (PUL; arrow IV; see (Stepniewska, 2004) for PUL subdivisions).

In SPMS, we found susceptibility reductions compared to SPMS-NCs that were more symmetric than in the RRMS group. In particular, significantly reduced susceptibility was found bilaterally in the medial dorsal nucleus (arrow II) and the PUL (arrow V). While the whole PUL was affected in the left hemisphere, changes in the lateral division of the medial PUL (arrow VI) in the right hemisphere did not reach statistical significance. A comparison with the group difference maps and Z-scores indicated that both PULs were homogeneously affected and the lack of statistical significance was due to a higher variability between subjects in the center of the PUL (Fig. S.4). Contrary to the RRMS group, we did not find significant differences in the ventral posterolateral nucleus.

Altas-based analysis

Comparison of the susceptibility brain template with the Schaltenbrand and Wahren atlas (Schaltenbrand et al., 1977) led to the unambiguous identification of the following three major nuclear regions of the thalamus: PUL, medial nuclei region (MNR), and lateral nuclei region (LNR). The anterior nucleus appeared hyper-intense but blended over into the cerebrospinal fluid, rendering a reliable segmentation difficult. A further parcellation of the nuclear groups, as reported in previous ultra-high field work (Deistung et al., 2013), was not possible. Figure 1 illustrates the anatomical locations of the identified regions.

The average susceptibilities and volumes in the patient and control groups are listed in Tabs. 2 and 3, respectively. The PUL showed the highest average magnetic susceptibility among all regions, in line with earlier QSM-based work (Deistung et al., 2013) and iron stains showing the strongest reactivity in this region (Morris et al., 1992; Spatz, 1922). Although both LNR and anterior nucleus had been described among the regions with the strongest iron reactivity (Morris et al., 1992), the LNR appeared less paramagnetic than the anterior nucleus (data not shown) and the MNR. This observation may be explained by the relatively high myelin content of the LNR compared to other thalamic regions (Schaltenbrand et al., 1977), which counteracts the effect of iron on the voxel susceptibility.

Group differences

The distributions of susceptibilities and volumes were normal in all groups and regions except for the PUL volumes in SPMS. Consistent with the VBA-results, differences between CIS and CIS-NCs did not reach statistical significance (p≥0.11; |d|≤0.36). In RRMS patients, we found statistically significant reductions of susceptibility compared to RRMS-NCs in the GT (−5.9 ppb; d=−0.86; p<0.001) and, consistent with the VBA, in the left MNR (−9.7 ppb; d=−0.84; p<0.001) and left LNR (−5.1 ppb; d=−0.75; p=0.001). Trends toward lower susceptibility in patients were observed in the PUL (−5.3 ppb; d=−0.50; p=0.028) and in the right MNR (−7.8 ppb; d=−0.57; p=0.019). In SPMS patients, magnetic susceptibility was significantly reduced compared to SPMS-NCs in all regions (≤−7.0 ppb; d≤−1.18) except in the LNR (d=−0.44; p=0.051), in line with the VBA findings.

Volume differences between CIS and CIS-NCs did not reach statistical significance, but disease-related atrophy was indicated by trends in the right GT (−0.28 ml; d=−0.47; p=0.043) and the right PUL (−0.15 ml; d=−0.49; p=0.034). In both RRMS and SPMS patients, volume reductions relative to the individual NC groups reached statistical significance in all regions (except for a trend in the right MNR of RRMS). Effect sizes were |d|≥1.0 in all areas except the MNR (|d|≥0.6).

Associations between susceptibility and atrophy

Table 4 lists the correlations of susceptibility and volume for all groups. Correlations were positive in all regions. Statistical significance was reached in the PUL of all patient groups (r>0.399, p<0.011), the PUL of SPMS-NCs (r=0.414, p=0.008), and the MNR of RRMS-NCs (r=0.508, p<0.001) and SPMS-NCs (r=0.525, p=0.001). Trends were observed in PUL (CIS-NC and RRMS-NC), MNR (all patient groups), and GT (all patient groups and SPMS-NC).

Table 4.

Spearman rank correlation coefficients for volumes and susceptibility.

	CIS-NC	CIS	RRMS-NC	RRMS	SPMS-NC	SPMS
GT	0.249 (0.12)	0.328 (0.039*)	0.237 (0.14)	0.333 (0.036*)	0.390 (0.013*)	0.347 (0.028*)
PUL	0.379 (0.016*)	0.419 (0.007**)	0.315 (0.048*)	0.539 (<0.001**)	0.414 (0.008**)	0.399 (0.011**)
MNR	0.237 (0.14)	0.332 (0.037*)	0.508 (0.001**)	0.384 (0.014*)	0.525 (0.001**)	0.390 (0.013*)
LNR	0.163 (0.315)	0.110 (0.50)	0.222 (0.17)	0.111 (0.49)	0.292 (0.068)	0.234 (0.15)

P-values are shown in parentheses. Bonferroni corrected threshold for statistical significance was 0.05/4=0.0125 for each group. * indicates trends, statistical significance after Bonferroni correction is indicated by ** and boldface.

Multivariate linear regression

A histogram analysis revealed a positively skewed distribution of the dd, which could be mitigated by a log-transform of the variable. Scatterplots correlating susceptibility with age and dd, respectively, indicated non-linear associations of susceptibility with age in the PUL and with the log-transformed dd in PUL and MNR, which could be described by a quadratic relationship. A similar analysis for the volumes revealed non-linear associations of left/right PUL with transformed dd and LNR with age and transformed dd. Hence, we performed a multivariate regression analysis with susceptibility and volume as dependent outcomes, respectively, and the following transformed independent variables in all 120 patients: age² [susceptibility: PUL; volume: LNR], ln(dd) [susceptibility: GT, LNR; volume: all], and ln(dd)² [susceptibility: PUL, MNR; volume: PUL, LNR], respectively.

We found significant negative associations of right GT susceptibility with ln(dd) [p<0.001, partial correlation R_p=−0.34], average PUL susceptibility with ln(dd)² [p<0.001, R_p=−0.42], and left MNR susceptibility with age [p=0.004, R_p=−0.28]. Trends toward negative associations were found for the left GT susceptibility with ln(dd) [p=0.029, R_p=−0.21] and age [p=0.035, R_p=−0.21], for the right MNR susceptibility with ln(dd)² [p=0.017, R_p=−0.23], and for the left LNR susceptibility with age [p=0.028, R_p=−0.21]. For volumes, we found significant negative associations between right GT and ln(dd) [p<0.001, R_p=−0.33], PUL and ln(dd)² [p<0.001, R_p=−0.32/−0.38], left MNR and age [p<0.001, R_p=−0.34], and LNR and ln(dd)² [p<0.001, R_p=−0.35]. Trends were found for negative associations of left GT with age [p=0.012, R_p=−0.24] and ln(dd) [p=0.045, R_p=−0.20] and right MNR with ln(dd) [p=0.019, R_p=−0.23]. Associations with sex did not reach p≤0.05 in any of the regions.

Age (and age²) correlated significantly with ln(dd) [and ln(dd)²] (Pearson 0.63≥R≥0.61, p<0.001), but condition indices of the collinearity analyses between the variables did not exceed 14 in any of the analyses. Supplementary Material 2 summarizes the results of the multivariate analyses with untransformed variables for susceptibility and volumes, respectively. The directions of the dependencies were largely in line with the analyses using transformed variables. Statistically significant univariate correlations are plotted in Figure 3.

Figure 3.

Top and middle row: Scatter plots illustrating univariate associations of susceptibility for regions that were significant in the multivariate analysis (top row and middle row for patients) and, for illustrative purposes, associations with age for the same regions (middle row). The shapes of the markers indicate the clinical phenotypes: circle – CIS; square – RRMS; cross - SPMS. In the top row, linear regression curves are used, whereas in the middle row cubic curves were used, except for the MNR, where we used a quadratic fit. Bottom row: Scatter plots of the susceptibilities in corresponding regions of NCs over age. The regression line for the left GT was determined by fitting a + b · (0.3 mg/100g · age −5.82·10⁻³ mg/100g · age² + 3.15·10⁻⁵ mg/100g · age³) to the susceptibility values (Levenberg-Marquardt), which relies on the known iron aging-trajectory in the GT (see Figure S.6 in the Supplementary Material). The fitted coefficients were a = (−15.3±6.9) ppb and b = (5.22±1.58) ppb·100g/mg. For the other NC plots, we used a cubic regression (PUL: R²=0.13, ANOVA p=0.0011, 22.0ppb – 7.18·10⁻³ ppb/a · age + 1.62·10⁻² ppb/a² · age² − 2.41·10⁻⁴ ppb/a³ · age³; MNR: R²=0.05, p=0.11; LNR: R²=0.22, p<0.001, 20.3ppb – 8.36·10⁻¹ ppb/a · age + 1.77·10⁻² ppb/a² · age² − 1.48·10⁻⁴ ppb/a³ · age³). For improved visualization, all ordinates, except that of the MNR, cover a susceptibility interval of 65ppb (all data points are visible).

4. Discussion

13 This work is the first systematic assessment of intra-thalamic susceptibility variations across the clinical spectrum of MS. We found significantly reduced magnetic susceptibility in specific thalamic subregions of patients with RRMS and SPMS, in particular in the PUL, but not in patients with CIS. Susceptibility reduction was significantly associated with increase in dd. In the following, we discuss our findings in the light of previous works and provide a mechanistic interpretation. In particular, we believe that the integrative role of the thalamus in brain function and the common notion of iron increase as a driver in neurodegeneration justify a detailed discussion of the potential biophysical origin and mechanistic processes underlying our findings.

Previous susceptibility-based MRI studies in MS that are consistent with our findings

In CIS patients, we found increased susceptibility in the PUL (Figure 2b), but group differences did not reach statistical significance (d=0.14; Tab. 2). This finding is (partially) in line with previous studies by Al-Radaideh et al. (Al-Radaideh et al., 2013), Langkammer et al. (Langkammer et al., 2013) and Elkady et al. (Elkady et al., 2017), which did not find statistically significant changes in the thalamus of CIS patients with QSM, and previous studies by our group reporting significantly increased iron load in the PUL of CIS patients (Hagemeier et al., 2012; Zivadinov et al., 2012) using a predecessor techniques of QSM. Although statistically insignificant, the group differences between CIS patients and controls found by Elkady et al. (Elkady et al., 2017) (Fig. 3a in that publication) closely resembled our findings of increased susceptibility in the PUL and decreased susceptibility in the rest of the thalamus (Fig. 2b).

In RRMS patients, we found significantly decreased GT susceptibility. This finding is consistent with recent QSM-based studies by Burgetova et al. (Burgetova et al., 2017) (−3.3ppb, p=0.004) and by our group (Hagemeier et al., 2017) (−5.7ppb, d=0.71–0.94, p<0.001). Also similar to the present work, Burgetova et al. observed reduced thalamic susceptibility in RRMS only when excluding the PUL and not in the PUL itself. This observation is consistent with group differences shown by Elkady et al. (Elkady et al., 2017) (Fig. 3b in that publication).

Previous susceptibility-based MRI studies in MS that are inconsistent with our findings

Several authors did not find significant GT changes in patients with RRMS (or mixed phenotype groups) using either QSM (Elkady et al., 2017; Fujiwara et al., 2017), the effective transverse relaxation rate (R₂*) (Fujiwara et al., 2017), or different iron-sensitive non-QSM techniques (Du et al., 2015; Hagemeier et al., 2013b; Raz et al., 2014). Other studies suggested increased GT iron in RRMS with QSM (Cobzas et al., 2015) and other techniques (Cobzas et al., 2015; Ge et al., 2007; Modica et al., 2014). Rudko et al. (Rudko et al., 2014) reported increased susceptibility relative to controls in a mixed group of CIS and RRMS patients, but differences in R₂* between CIS and RRMS patients of similar age did not reach significance (Khalil et al., 2009).

In the present study, the PUL appeared unaltered in RRMS patients and showed significantly decreased values in SPMS when compared to NCs. In line with previous work (Henry et al., 2008), we found both atrophy and reduced susceptibility in the left MNR of our RRMS group and bilaterally in SPMS patients. However, we could not confirm the previous findings of significant atrophy of the bilateral MNR and bilateral PUL in CIS (Henry et al., 2008). Inconsistent with the present study is also the increased PUL iron load that has been suggested in RRMS using QSM (Al-Radaideh et al., 2013; Elkady et al., 2017; Rudko et al., 2014), R₂* (Elkady et al., 2017; Lebel et al., 2012; Quinn et al., 2014; Walsh et al., 2014), and other techniques (Haacke et al., 2010; Habib et al., 2012; Hagemeier et al., 2013b; Modica et al., 2014; Zivadinov et al., 2012; Zivadinov et al., 2010).

Heterogeneity of the literature on brain-iron in MS

Overall, the literature is highly heterogeneous with respect to the outcomes of brain iron studies in MS. This observation may be explained by the wide variety of techniques employed for imaging, reconstruction, and analysis. Even a direct quantitative comparison between studies that employed QSM is difficult because the studies employed different reference regions (or did not specify the region used), phase processing techniques (Özbay et al., 2017; Schweser et al., 2017b), and QSM algorithms (Wang and Liu, 2015). A comparison with non-QSM techniques, such as R₂* mapping, is even more problematic because of confounding sensitivities on the microdistribution and chemical form of iron (Dietrich et al., 2017) as well as the microstructure. Furthermore, the non-linear aging trajectory of the thalamic iron concentration (Hallgren and Sourander, 1958) and the heterogeneity of the disease render a comparison of groups with different clinical and demographic characteristics challenging. An overview of the substantial differences between the cohorts studied in the literature may be found in the Inline Supplementary Tab. S.3. Interestingly, Burgetova et al.’s study, which is largely consistent with the present work, relied on a cohort relatively similar to our RRMS group.

The comparison of the group characteristics and outcomes of published studies led us to the proposal of a early-rise late-decline hypothesis: Increased PUL susceptibility may be found predominantly in younger patients with an average age below 40 years (all studies published before 2017; cf. Tab. S.3), whereas in older patients, such as those studied in the present and in our previous work (Hagemeier et al., 2017), PUL susceptibility in patients was, on average, lower than in controls. The observed significant reduction of PUL susceptibility with dd (p<0.001) further supports this hypothesis. Moreover, in a recent R₂*-based follow-up study, Khalil et al. (Khalil et al., 2015) suggested that not only in the thalamus but also in other brain regions iron accumulation is more pronounced in the early stages than in the later stages of the disease. Showing that iron decreases over time already in CIS, this study also suggests that peak iron concentration may be observed (best) before the first clinical symptoms. However, since the authors did not enroll a control group, it remained unclear if the observed temporal iron decrease exceeded the normal aging-related iron decrease. Hence, the authors’ conclusion, although consistent with our findings, warrants further investigation.

Biophysical explanation of the observed thalamic susceptibility alterations

A decrease of magnetic susceptibility toward more diamagnetic values can be explained by both a reduction of the contributions from paramagnetic compounds (such as ferritin-bound iron) and an increase in the contributions from diamagnetic compounds (such as myelin or calcium); a change toward higher susceptibility values vice versa. While calcium may play a role in MS (see Supplementary Material 3 for a detailed discussion), no strong evidence exists for a substantial calcium accumulation in the thalamus. Hints on the underlying biophysical mechanisms of the observed susceptibility decrease may be obtained from correlations of susceptibility and R₂*. The two measures have a similar dependence on paramagnetic but an opposite dependence on diamagnetic tissue compartments; iron increases both of them, whereas calcium decreases/increases susceptibility/R₂*. Hence, thalamic susceptibility and R₂* values would be expected to correlate negatively if calcium had a significant contribution to the variations between observed measures. Fujiwara et al. (Fujiwara et al., 2017) recently performed such a partial correlation analysis correcting for age and sex in the DGM of MS patients. However, they reported a significant positive correlation in the thalamus (r=0.6; p<0.001) as well as all other DGM. This observation indicates that variations in the tissue concentration of iron, but not calcium, dominate the inter-subject variations of susceptibility and R₂* in MS. Interestingly, the authors found significant positive correlations also in the DGM of NCs, but not in their thalami (r=0.03; p=0.89). This discrepancy between MS patients and NCs may be explained by the relatively high amount of diamagnetic myelin in the thalamus of NCs, which counteracts the positive correlation, and the reduced amount of myelin in MS.

The most plausible explanation of increased PUL magnetic susceptibility early in the disease (Fig. 2b and previous work) is the iron accumulation of microglia (Banati et al., 2000; Haider et al., 2014; Herranz et al., 2016; Kauzner et al., 2016; Rissanen et al., 2014). Recent evidence for a chronic intravascular haemolysis (Lewin et al., 2016) and a correlation of serum iron with DGM susceptibility changes (Bergsland et al., 2017) in MS point toward a translocation of blood-iron into the brain, where it accumulates in microglia, a process that has been hypothesized already by Metz and Spatz in 1924 (Metz and Spatz, 1924). DGM may be particularly prone to the inlux of heme-iron due to their high perfusion compared to WM. In principle, also the reduction of the relative voxel volume fraction of myelin associated with the focal thalamic demyelination in MS (Haider et al., 2014; Vercellino et al., 2009) could increase the susceptibility. However, while the thalamus is higher myelinated than other DGM regions, myelin stains show that the PUL is one of the regions with the least amount of myelin in the thalamus (Schaltenbrand et al., 1977), rendering it unlikely that demyelination in this region explains a substantial increase of the tissue susceptibility.

The decrease of magnetic susceptibility in later stages of the disease most likely involves iron-containing thalamic cells. In both WM and DGM of normal brain, most of the iron is found in oligodendrocytes and their processes (Bagnato et al., 2011; Francois et al., 1981; Haider et al., 2014; Hill and Switzer III, 1984; Meguro et al., 2008). Iron-staining in rats indicated a higher number of iron-laden oligodendrocytes as well as a higher iron load of individual oligodendrocytes in the DGM than in the WM (Meguro et al., 2008). In human DGM, approximately 30% of the cell bodies have been reported to be oligodendrocytes (in the caudate) (Myers et al., 1991b), most of them are probably perineuronal satellite oligodendrocytes with unknown functions (Verkhratsky, 2013). Hence, the substantial reduction of the magnetic susceptibility below that in NCs, as observed in definite MS (Tab. 2 and Fig. 2b), points toward a decrease of the oligodendrocyte density or a depletion of iron from oligodendrocytes. In fact, using diaminobenzidine (DAB)-enhanced Turnbull blue staining for ferrous and ferric iron, Hametner et al. (Hametner et al., 2013) reported a decrease in the iron concentrations in both the WM close to the neocortex (including the iron-rich U-fibers, personal communication) and in the deep normal appearing WM (NAWM; sampled from the whole telencephalon, personal communication) of MS tissue specimens relative to that in specimens of NCs. Similar to our findings in the thalamus, the reduction of WM iron correlated significantly with the dd (R²=0.31, p<0.001). Furthermore, the authors reported a loss of oligodendrocytes and a reduction of non-heme iron within oligodendrocytes and myelin in NAWM. Using ionised calcium-binding adapter molecule 1 (IBA-1) immunoreactivity, the same group later demonstrated a shift of iron from oligodendrocytes to microglia in the DGM (Haider et al., 2014). While the authors assessed iron by semi-quantitative densitometry, which may not represent iron concentration linearly, Popescu et al. (Popescu et al., 2017) recently confirmed the decrease of iron with age and disease duration in NAWM of MS patients using X-ray fluorescence imaging, which is an element-specific and quantitative technique. Reduced iron has also been reported in inactive cortical MS lesions (Yao et al., 2014), i.e. lesions with sparse inflammatory cells.

Mechanistic interpretation of the observed thalamic susceptibility alterations

Due to their high oligodendrocyte density, the DGM may be subject to similar pathological processes as the WM. In particular, both NAWM (Frischer et al., 2009) and DGM (Herranz et al., 2016) show chronic microglial activation in MS, with the highest activity in the thalamus (Banati et al., 2000; Herranz et al., 2016; Kauzner et al., 2016; Rissanen et al., 2014). The pro-inflammatory cytokines TNF-α and interferon-γ, expressed by microglia upon activation, have an iron-mediated toxic effect on oligodendrocytes (Zhang et al., 2005) and trigger a release of iron from the cells (Zhang et al., 2006). While protective against the direct toxicity of cytokines, this oligodendroglial iron release may itself have several detrimental consequences. First, before the liberated iron is detoxified and cleared from the region, it might directly contribute to mitochondrial dysfunction and neurotoxicity by fueling the creation of reactive oxygen and nitrogen species via Fenton chemistry. Second, iron deficiency negatively affects several pathways of normal oligodendrocyte function (Connor and Menzies, 1996), the proliferation of oligodendrocyte progenitor cells (Schonberg et al., 2012), their differentiation, and remyelination (Stephenson et al., 2014). Third, microglia that pick up the iron increase the release of proinflammatory cytokines and switch from a quiescent to a pro-inflammatory phenotype (Zhang et al., 2006), further fueling the release of iron from oligodendrocytes in a vicious circle.

The observed positive correlation between structural volumes and magnetic susceptibility observed in the present study (Tab. 4) points toward an association between iron depletion and the relatively high number of focal demyelinating lesions previously observed in the thalamus (Haider et al., 2014; Vercellino et al., 2009). This association might result from a loss of iron-containing oligodendrocytes secondary to the lesion formation or be related to a retraction of oligodendrocyte processes under an iron-depletion related stress condition (Rone et al., 2016). Hence, it may be speculated that the bystander damage of chronic inflammation and, in particular, its effect on the local availability of storage iron, represents the missing link between inflammatory and neurodegenerative disease components in MS, partially explaining the slow transition from RRMS to SPMS. In this context, it is an interesting observation that thalamic T₂-hypointensity, indicative of high iron load, predicts the progression of brain atrophy over 1 year in untreated MS patients, whereas this relationship diminishes upon immune-modulating treatment with interferon β-1 (Bermel et al., 2005) despite no measurable effect of the drug on the atrophy rate during this time.

An association between chronic microglial activation and iron depletion would provide the missing link also for a mechanistic explanation for the peculiar normal aging trajectory of thalamic iron (Fig. S.6). Microglial activation is a poorly understood phenomenon of normal aging, which shows the strongest correlation with age in the thalamus (Schuitemaker et al., 2012).

What distinguishes the thalamus from other brain regions that show predominantly increased magnetic susceptibility in MS?

Strong evidence exists for the common notion that MS is associated with an increase in the DGM iron concentration. This evidence may give rise to the question why the WM and the thalamus show the opposite behavior. However, increased iron concentration, such as that observed in the DGM, does not necessarily imply iron deposition. We have estimated in a recent longitudinal QSM-based MS study that the expected increase of caudatal susceptibility due to demyelination and atrophy of cells with little or no iron, exceeds the observed susceptibility increase (Hagemeier et al., 2017). In other words, the removal of iron-containing cells or depletion of iron from cells on top of the major demyelination is required to explain the observed longitudinal increase of magnetic susceptibility in the caudate of MS patients. This estimation was recently supported by independent preliminary data on total iron estimated from R₂* and volumetric measurements, which indicated decreased total iron in the caudate, pallidum, and thalamus of MS patients (Hernández-Torres et al., 2017). Reduced DGM iron has also previously been observed in other neurological diseases (Doring et al., 2016; Kanaan et al., 2017). In summary, the depletion of iron may not be restricted to the thalamus but may instead be a general hallmark of MS.

However, its rich connectivity profile and central role in several brain networks might render the thalamus particularly susceptible to secondary effects from remote injury in other parts of the brain via Wallerian degeneration or hypometabolism. This hypothesis is supported by studies showing that the reduction of thalamic NAA in MS (Cifelli et al., 2002; Wylezinska et al., 2003), indicative of neurodegeneration, is correlated with NAA in the frontal NAWM (Wylezinska et al., 2003), and thalamic atrophy and hypometabolism are correlated with WM lesion volume (Blinkenberg et al., 2000; Houtchens et al., 2007). Furthermore, a study by Henry et al. (Henry et al., 2009) suggested a mechanistic relationship between thalamic atrophy and WM lesions in thalamocortical tracts. A recent study in the experimental autoimmune encephalomyelitis (EAE) mouse model for MS reported chronic inflammation and demyelination of the spinothalamic tract at the level of the spinal cord paired with neuronal loss in the thalamic target region of the tract (Wagenknecht et al., 2016). The absence of indicators for autoimmune attack of the thalamus during the acute stage of the disease suggested that the observed thalamic neurodegeneration was secondary to the autoimmune attack in the spinal cord.

It is also known that neuroinflammatory responses can be ‘projected’ bidirectionally along cortico-thalamic tracts (Banati, 2002), where cortical injury induces a remote microglial response in the ipsilateral thalamus (Banati et al., 2001; Kuchcinski et al., 2017; Myers et al., 1991a; Pappata et al., 2000; Sørensen et al., 1996). Consequently, the widespread cortical pathology in MS (Calabrese et al., 2007; Kutzelnigg and Lassmann, 2005), including chronic active lesions with microglial activation (Pitt et al., 2010) and meningeal inflammation (Howell et al., 2011; Kutzelnigg et al., 2005), may spread along cortico-thalamic tracks and, as such, “focus” in the thalamus. In fact, thalamic [11C]-PBR28 binding correlates with cortical thinning and reduced cognitive performance (Herranz et al., 2016).

Remote cortical injury as a driving factor for the observed susceptibility reduction is also supported by the fact that the PUL was the most affected region in the present work. Being the largest of the “association” nuclei (Shipp, 2003), the PUL receives the majority of its input directly from the cerebral cortex and participates primarily in reciprocal cortico-corical interactions. In addition, also the MNR, including the mediodorsal nucleus, is primarily associative, whereas the LNR, which showed only little alteration in the present study, consists of “relay” nuclei, which receive input from the periphery and only forward it to the cortex (Jones, 1991).

It is likely that the strong susceptibility decrease in the PUL observed in the present study is secondary to injury elsewhere in the optical pathways. Cortico-pulvinar-cortical circuitry has an active participation in the processing of visual information and selective attention through the promotion of synchronized activity in different cortical areas (Benarroch, 2015; Fama and Sullivan, 2014). Early studies in monkeys have demonstrated that lesions in the visual cortex (Mathers, 1972; Ogren and Hendrickson, 1979) cause degenerative changes in the PUL. Visual impairment is a frequent symptom of MS and both retinal nerve fiber layer thickness, and retinal periphlebitis have been associated with disease activity in MS (Sepulcre et al., 2007). More recently, Mühlau et al. (Muhlau et al., 2013) found that WM lesions and PUL atrophy are spatially related in MS, with axonal transection within WM lesions and subsequent degeneration along the axonal projections being the most plausible explanation. Gabilondo et al. (Gabilondo et al., 2014) showed retro- and anterograde trans-synaptic neuronal degeneration in the visual pathway and Zivadinov et al. (Zivadinov et al., 2014) reported a trend toward associations between retinal nerve fiber layer thickness in RRMS patients and microstructural damage in the thalamus, in particular, volume changes in the GT and PUL.

Interestingly, thalamic lesions in MS have previously been reported primarily in the MNR and the anterior nuclear region, but not in the LNR (Vercellino et al., 2009). These observations are in line with the locations of significant susceptibility reduction in our VBA (Figs. 2a and S.3). Several studies have shown that the LNR connects to premotor areas (anterior portion) (Behrens et al., 2003; Bisecco et al., 2015; Johansen-Berg et al., 2005), M1, and somatosensory cortices S1/S2 (posterior portion) (Behrens et al., 2003; Johansen-Berg et al., 2005). The intra-thalamic spatial patterns of regions with thalamo-occipital, -temporal, and -somatosensory projections shown in these publications resemble surprisingly well our QSM-based findings in RRMS, and areas with thalmo-occipital, -parietal, and -frontal projections those in SPMS (Figs. 2a and S.3). Also a recent R₂*-based study in stroke patients (Kuchcinski et al., 2017) reported increased iron in the MNR and PUL following infarcts in more anterior brain regions and more posterior regions, respectively. This selective involvement of areas with specific connectivity profiles further supports a direct relationship between thalamus pathology and pathology in the cortex.

Laterality of susceptibility changes

We observed strong laterality in some regions. Although the reason for the laterality remains unclear, it is not an unexpected finding. Cobzas et al. (Cobzas et al., 2015) reported left-right asymmetry of the p-values in their VBA of thalamus susceptibility. Asymmetry of deep brain regions has also recently been shown in a large multi-center study. In particular, thalamus was among the structures with the highest inter-hemispheric asymmetry (Guadalupe et al., 2016). A meta-analysis by Lansley et al. (Lansley et al., 2013) showed increased atrophy of the left thalamus in patients with RRMS and Preziosa et al. (Preziosa et al., 2017) recently reported a stronger association of atrophy in the left than in the right thalamus with worsening disability and cognitive deterioration over 5 years.

Limitations of the study

Our work has conceptual and technical limitations. Conceptual limitations of the study include the lack of a detailed assessment of the disability profile in patients and its correlation with the intra-thalamic susceptibility changes. It is likely that a link exists between the alterations in sub-nuclei and functional, emotional, and cognitive dysfunction primarily associated with these regions and basal ganglia-thalamocortical pathways. In particular, the globus pallidus is the major input region to the thalamus (Alexander et al., 1986) and, at the same time, one of the DGM regions that show the greatest susceptibility increase in MS (Hagemeier et al., 2017).

To investigate if our findings are strongly associated with clinical disability, we performed an exploratory post hoc analysis in which we added EDSS to the multivariate regression analysis (log-transformed due to platykurtic distribution). We observed a statistically significant positive association between the transformed disability and susceptibility in the left (p=0.034) and right (p=0.025) LNR. In all other regions, associations between disability and susceptibility did not reach significance (p>0.05). In particular, we failed to confirm the negative association of the global thalamus susceptibility with EDSS scores previously reported by Burgetova et al. (Burgetova et al., 2017). Considering that the LNR is a region that contains primarily relay nuclei, which forward information from the periphery to the cortex, a more direct association of the LNR with EDSS scores (compared to the association nuclei) seems plausible. A more detailed investigation of the associations of susceptibility alterations in thalamic sub-nuclei with clinical disability will be the subject of future research.

Another limitation of our study is that we could not demonstrate that thalamic susceptibility explains more variance than other, more conventional MRI measures and did not take into account the thalamic lesion burden. In addition, although the design of our cross-sectional study controlled for confounding effects to the extent possible, longitudinal studies are needed to control for cohort effects that affect the whole cohort equally.

We discuss the technical limitations of the study in the Supplementary Material 4.

Conclusion

In conclusion, the results of the present study falsified our hypothesis of increased thalamic susceptibility. Our findings suggest that thalamic susceptibility decreases with dd and that thalamic sub-regions follow distinct temporal trajectories. The inconsistency of this result with some previous studies may be attributed to a substantial heterogeneity of clinical characteristics, imaging methods, and group demographics, particularly a lower group-average age in previous studies. Although the biophysical origin of the decreased thalamic susceptibility remains unclear, a plausible explanation that is supported by independent previous work is the depletion of iron from oligodendrocytes, which may be a side effect related to chronic microglia activation and ultimately lead to neurodegeneration. Because the different thalamic nuclei maintain distinct connectivity profiles that form a closely coupled system with virtually all cortical and subcortical areas (Fama and Sullivan, 2014; Postuma and Dagher, 2006; Sherman and Guillery, 2013b), the observed pathologic alterations in thalamic tissue properties may have wide-spread interaction effects with other disease mechanisms and symptoms in MS.

Table 2.

Average magnetic susceptibility (parts-per-billion; ppb) in thalamic subnuclei.

	CIS-NC	CIS	RRMS-NC	RRMS	SPMS-NC	SPMS
GT
Avg (CI)			9.2 ± 6.0 (1.85)	3.3 ± 3.3 (1.85)	7.08±5.7 (1.76)	0.079±6.1 (1.90)
Left (CI)	7.4 ± 7.2 (2.23)	6.1 ± 5.2 (1.6)
Right (CI)	10.0 ± 5.3 (1.63)	9.0 ± 4.7 (4.7)
Difference (d)	−1.3 (−0.21) / −1 (−0.20)		−5.9 (−0.86)		−7.0 (−1.18)
p	0.34 / 0.37		<0.001**		<0.001**
PUL
Avg (CI)	29.2 ± 9.2 (2.85)	30.4 ± 8.2 (2.54)	30.9 ± 9.3 (2.90)	25.6±11.4 (3.55)	29.9±10.4 (3.21)	16.6±13.6 (4.21)
Difference (d)	+1.2 (0.14)		−5.3 (−0.50)		−13.3 (−1.10)
p	0.56		0.028*		<0.001**
MNR
Avg (CI)					3.9 ± 11.2 (3.48)	−7.8±12.1 (3.74)
Left (CI)	6.7 ± 13.2 (4.09)	4.3 ± 7.7 (2.39)	7.3 ± 10.8 (3.91)	−2.4±12.4 (3.80)
Right (CI)	6.1 ± 9.5 (2.95)	6.9 ± 8.5 (2.63)	4.7 ± 12.6 (3.91)	−3.1±14.9 (4.60)
Difference (d)	−2.4 (−0.23) / 0.8 (0.08)		−9.7 (−0.84) / −7.8 (−0.57)		−11.·7 (−1.00)
p	0.30 / 0.74		<0.001*/0.019*		<0.001**
LNR
Avg (CI)					2.6 ± 5.0 (1.56)	0.19 ± 5.8 (1.81)
Left (CI)	5.2 ± 7.1 (2.20)	3.6 ± 6.0 (1.85)	4.9 ± 6.5 (2.01)	−0.2 ± 7.1 (2.21)
Right (CI)	7.9 ± 5.4 (1.68)	5.9 ± 5.4 (1.68)	5.6 ± 7.2 (2.22)	2.5 ± 6.9 (2.15)
Difference (d)	−1.6 (−0.24) / −2.0 (−0.36)		−5.1 (−0.75) / −3.1 (−0.44)		−2.41 (−0.44)
p	0.28 / 0.11		0.001* / 0.058		0.051

Values are reported as mean ± standard deviation. CI denotes the 95% confidence interval, d is Cohen’s effect size. For each region, values are reported either as bi-hemispheric average (Avg) or for each hemisphere individually, depending on the statistical significance of differences between hemispheres. If sample distributions were different in either the patient or the control groups, hemispheres were analyzed separately for both groups. All sample distributions were normal. For the comparisons of hemispheres, differences are reported as left / right. The p-values were determined by univariate ANCOVA adjusted for sex (we found no significant interactions with sex except for a trend in the right MNR with p=0.042). Bonferroni corrected thresholds for statistical significance were 0.05/7=0.0071 for CIS, 0.05/6=0.0083 for RRMS, and 0.05/4=0.0125 for SPMS. Trends are indicated by *, statistical significance after Bonferroni correction is indicated by ** and boldface.

Table 3.

Average volumes (in milliliters; ml) in thalamic subnuclei.

	CIS-NC	CIS	RRMS-NC	RRMS	SPMS-NC	SPMS
GT
Sum (CI)			16.81 ± 0.13 (0.416)	14.90 ± 1.72 (0.531)	16.07 ± 1.31 (0.407)	14.00 ± 1.85 (0.572)
Left (CI)	8.22 ± 0.64 (0.200)	8.05 ± 0.71 (0.221)
Right (CI)	8.45 ± 0.53 (0.165)	8.17 ± 0.68 (0.211)
Difference (d)	−0.17 (−0.25) / −0.28 (−0.47)		−1.91 (−1.2)		−2.07 (−1.1)
p	0.28 / 0.043*		<0.001*		<0.001*
PUL
Left (CI)	2.14 ± 0.32 (0.0982)	2.05 ± 0.30 (0.0918)	2.22 ± 0.30 (0.0925)	1.84 ± 0.35 (0.109)	2.05 ± 0.27 (0.0833)	1.71 ± 0.38 (0.117)
Right (CI)	2.35 ± 0.29 (0.0904)	2.20 ± 0.33 (0.101)	2.39 ± 0.29 (0.0885)	2.01 ± 0.35 (0.109)	2.25 ± 0.29 (0.0899)	1.88 ± 0.41 (0.128)
Difference (d)	−0.09 (−0.30) / −0.15 (−0.49)		−0.38 (−1.1) / −0.38 (−1.2)		−0.34 (−1.0) / −0.37 (−1.0)
p	0.19 / 0.034*		<0.001* / <0.001*		<0.001* / <0.001*
MNR
Left (CI)	0.79 ± 0.12 (0.0380)	0.77 ± 0.13 (0.0405)	0.78 ± 0.13 (0.0395)	0.69 ± 0.11 (0.0345)	0.76 ± 0.15 (0.0474)	0.64 ± 0.14 (0.0427)
Right (CI)	0.87 ± 0.10 (0.0326)	0.85 ± 0.12 (0.0385)	0.85 ± 0.16 (0.0500)	0.75 ± 0.15 (0.0461)	0.81 ± 0.14 (0.0424)	0.67 ± 0.12 (0.0365)
Difference (d)	−0.02 (−0.21) / −0.02 (−0.15)		−0.09 (−0.75) / −0.10 (−0.60)		−0.12 (−0.82) / −0.14 (−0.82)
p	0.36 / 0.50		0.0013* / 0.012*		<0.001* / <0.001*
LNR
Sum (CI)	2.063 ± 0.187 (0.0580)	2.035 ± 0.22 (0.0692)	2.10 ± 0.23 (0.0717)	1.83 ± 0.26 (0.0817)	2.00 ± 0.21 (0.0660)	1.69 ± 0.29 (0.0901)
Difference (d)	−0.024 (−0.086)		−01.26 (−1.0)		−0.36 (−1.1)
p	0.56		<0.001*		<0.001*

Values are reported as mean ± standard deviation. CI denotes the 95% confidence interval, d is Cohen’s effect size. For each region, values are reported either as bi-hemispheric sum or for each hemisphere individually, depending on the statistical significance of differences between hemispheres. If sample distributions were different in either the patient or the control groups, hemispheres were analyzed separately for both groups. All sample distributions were normal, except the left and right PUL in SPMS (compared using Wilcoxon test). For the comparison of hemispheres, differences are reported as left / right. The p-values were determined by univariate ANCOVA adjusted for sex (we found no significant interactions with sex except for a trend in the right MNR with p=0.027). Bonferroni corrected thresholds for statistical significance were 0.05/7=0.0071 for CIS, 0.05/6=0.0083 for RRMS and SPMS. * indicates trends, statistical significance after Bonferroni correction is indicated by ** and boldface.

Abbreviations

Avg

Average

CI

Confidence Interval

CIS

Clinically Isolated Syndrome

DAB

Diaminobenzidine

dd

Disease Duration

DGM

Deep Gray Matter

EDSS

Expanded Disability Status Scale

FOV

Field of view

FWE

Family-Wise Error

GRE

Gradient Recalled Echo

GT

Global Thalamus

IBA-1

Ionised Calcium-Binding Adapter Molecule 1

IQR

Interquartile Range

IR-FSPGR

Fast Spoiled Gradient-Echo Pulse Sequence With Inversion Recovery

LNR

Lateral Nuclear Region

MNR

Medial Nuclear Region

MS

Multiple Sclerosis

NAWM

Normal Appearing White Matter

NC

Normal Controls

PET

Positron Emission Tomography

ppb

Parts Per Billion

PUL

Pulvinar

QSM

Quantitative Susceptibility Mapping

R₂*

Effective Transverse Relaxation Rate

RRMS

Relapsing-Remitting MS

SPMS

Secondary Progressive MS

SWI

Susceptibility-Weighted Imaging

T₁w

T₁-Weighted

TFCE

Threshold-Free Cluster Enhancement

WM

White Matter