Subjects and tissue collection

In this study, we included 13 cases with confirmed progressive MS and 13 pathologically confirmed non-neurological controls. Shortly after death, MRI in MS cases was performed and tissue was collected through the standardized MS center Amsterdam rapid autopsy protocol in collaboration with the Netherlands Brain Bank (NBB) [5]. Healthy volunteer data collection followed similar protocols according to protocols of the Normal Aging Brain Collection Amsterdam (NABCA) [22]. Prior to death, all subjects had registered with the NBB or full body donation program at the department of Anatomy and Neurosciences at Amsterdam UMC, providing informed consent for autopsy and use of imaging, tissue and medical records for research purposes. Permission for the protocols was provided by the institutional ethics review board. Clinical and demographic information is summarized in Table 1. In more detail, 92.3% of the MS cases were female, with a mean age and standard deviation at time of death of 63.76 ± 12.10 and a disease duration of 26.54 years ± 0.04; 61.5% of the pathologically confirmed non-neurological controls were female, with a mean age and standard deviation at time of death of 70.38 ± 6.64. For both groups, the most common cause of death was euthanasia and lung complications (e.g., infections, failure). After MRI scanning, formalin-fixed paraffin-embedded (FFPE) blocks of the thalamus were cut into 5 and 10μm sections and stored at room temperature until further use.

Table 1 Clinical data of the cohortMagnetic resonance imaging (MRI)

MRI was performed on a 3T whole-body scanner (GE MR750 Discovery) with an eight channel phased-array head coil, as previously described [5]. Imaging was performed both in-situ (with brain still in cranium) and of subsequently derived 1cm-thick coronal brain slices, from which thalamic sections were cut. By matching the slices with the in-situ 3D imaging, the precise anatomical locations of the thalami were determined. The imaging protocol consisted of a sagittal three dimensional (3D) T1-weighted fast spoiled gradient-echo sequence (TR = 6.7 ms; TE = 3 ms, TI = 450 ms, slice thickness = 1.0 mm, in-plane resolution = 0.98 × 0.98 mm2), a sagittal 3D T2-fluid-attenuated inversion recovery (FLAIR) scan (TR = 8000 ms; TE = 125 ms; TI = 2000ms; slice thickness = 1.2 mm, in-plane resolution = 0.98 × 0.98 mm2) and an axial two dimensional (2D) echo-planar imaging (EPI) pulse sequence with diffusion gradients applied in 30 non-collinear with b = 1000s/mm2, and five volumes acquired without diffusion weighting (TR = 8000 ms, TE = 85 ms, slice thickness = 2.0 mm, in-plane resolution = 2.0 × 2.0 mm2). 3D-FLAIR was used for white matter lesions segmentation while 3D-T1 was used for volumetry. In control donors, the few segmented lesions likely reflect age-related white matter changes. Smaller vascular-type lesions were not included, as they did not meet the minimum size threshold for MS lesion segmentation. These lesions were segmented primarily to improve the accuracy of T1-weighted image processing and avoid segmentation errors during volumetric analysis. Imaging parameters have been published previously [5, 22].

MRI processing: volumes

Lesions were segmented on 3D-FLAIR using a semi-automated technique and manual corrections. Lesion maps were then co-registered to 3D-T1 and used for lesion filling. Subsequently, AssemblyNet software was used to determine global brain volumes and deep grey matter volumes [9].

Thalamic nuclei segmentation was performed with a modified version of Thalamus Optimized Multi Atlas Segmentation (THOMAS) method [43], which uses T1-weighted (T1w) imaging as an input image to segment the thalamus [47]. This method specifically uses two separately trained convolutional neural networks (CNN) to synthesize white matter-nulled (WMn)-MPRAGE images from T1w data and then segment the synthetic data based on the histological definitions of the Morel atlas [33]. The analysis results in the segmentation of 10 thalamic nuclei, which were then grouped into four main regions based on their anatomical and functional location: anterior group (anteroventral: AV; ventral anterior: VA), lateral group (ventral posterolateral: VPL; ventral lateral anterior: VLa; ventral lateral posterior: VLp), medial group (mediodorsal: MDn; centromedian: CM) and posterior group (pulvinar: Pul; medial geniculate nucleus: MGN; lateral geniculate: LGN). Each group was blindly checked and manually corrected if needed. For the volumetric analysis of all participants, the volumes of the right and left sides were summed and assessed as a fraction of the volume of the intracranial cavity (IC) extracted through AssemblyNet to account for variations in head size [9].

Tractography and diffusion tensor imaging

Probabilistic tractography was performed using MRtrix3 software (http://www.mrtrix.org) [45]. A whole-brain mask was created for each subject after the diffusion data had been corrected for eddy current distortions and motion artefacts. Diffusion tensors were also generated from the whole brain mask, from which fractional anisotropy (FA) and mean diffusivity (MD) maps were obtained with tensor2metric. Using the ss3t_csd_beta1 function of MRtrix3Tissue (https://3Tissue.github.io), a fork of MRtrix3, a single-shell response function was calculated to estimate the Fibre Orientation Distributions (FOD) based on constrained spherical-deconvolution [46]. Ten million whole-brain tracts were produced using five tissue-type segmented T1 images and anatomically constrained tractography [45]. To reduce reconstruction bias and increase biological plausibility, these streamlines were ultimately filtered to one million using the spherical-deconvolution-informed filtering of tractograms [42]. We also assessed the microstructural integrity of each thalamic nuclei group using both FA and MD.

Construction of connectivity atlas and structural connectomes

FreeSurfer (v5.3) (http://surfer.nmr.mgh.harvard.edu/) was used for whole-brain segmentation using T1-weighted images [11]. To construct a connectivity matrix for our analysis, the Destrieux cortical atlas was used as the basis for our custom-made atlas [13]. The cortical atlas was registered to diffusion space, followed by the registration of the T1-image of segmented thalamic groups to diffusion B0 space using the flirt function of FSL [3]. All registrations were checked visually. The registered Destrieux cortical atlas was then modified to maintain only labels of interest, and thalamic groups were added to the image to create a custom-made atlas. Finally, the personalized atlas featured 172 regions.

The 1 million streamlines previously obtained were mapped into the custom atlas, and a 172 × 172 structural connectivity matrix was produced. Each matrix element reflected the number of streamlines for each subject. The structural connectivity matrix was then normalized by the maximum number of tracts for each participant, to bring the connectivity values within a consistent range. Connectivity matrices were also weighted by the mean values of FA per streamline.

Immunolabelling

Thalamic sections were deparaffinized in xylene and rehydrated in a series of graded ethanol (100%, 90%, 80% and 70% for 3 min each). For staining for Nissl bodies, sections were stained with Thionin (Thermo Fisher Scientific) directly afterwards and mounted with Entellan. For all the other sections, after a rinse in milliQ, epitope retrieval was performed in citrate buffer pH 6 for 30 min in a water bath at 95ºC. Sections were cooled on ice and washed with PBS, followed by a blocking step with a PBS solution containing 10% normal serum (from the host of the secondary antibody) and 0.05% Tween-20 (Sigma) for 30 min at room temperature. Sections were incubated overnight at 4ºC with primary antibodies (Table 2) diluted in 3% normal serum and 0.05% Tween-20.

Table 2 Primary antibody details

For immunohistochemistry, sections were subsequently incubated with EnVision + Dual Link System-HRP (Agilent Dako) with 3,3’-diaminobenzidine (DAB) as the chromogen. Sections were counterstained with hematoxylin, dehydrated in alcohol and xylene series, and embedded in Entellan medium (Merck). Sections were stored at room temperature until image acquisition. For immunofluorescence, primary antibody incubation was followed by a 2h incubation of Alexa fluorophore-labeled secondary antibodies at room temperature (1:400, Thermo Fisher Scientific). After washing with PBS, sections were incubated with DAPI (1:10,000, Molecular Probes) for nuclear visualization. Brain auto-fluorescence was quenched with 0.03% Sudan Black (Sigma) in 70% ethanol for 5 min. After washing, sections were mounted with Mowiol medium and stored at 4ºC in the dark until image acquisition.

Microscopy image acquisition and image analysis

Bright-field images were taken using a 20 × objective on a Vectra Polaris whole-slide scanner (Akoya Biosciences). Fluorescent-labelled sections of microglia were imaged on a Vectra Polaris whole-slide scanner, a Nikon A1R laser-scanning confocal microscope, a Nikon AXR confocal microscope and a Leica TCS SP8 confocal microscope. Sections for synaptic markers were acquired on the Olympus VS200 slide scanner. Specific details for each staining are described in the sections below.

Mediodorsal nuclei location

The thalamus is divided into different nuclei and each nucleus has unique inputs and projections towards different cerebral cortical regions. Hence, the thalamus contains histologically a high heterogeneity depending on the specific nucleus [20]. To avoid misleading results, we decided to focus specifically on the mediodorsal nucleus (MDn), which is involved in cognition, emotion and attention, and it was present in all the blocks of our tissue cohort [15]. To pinpoint the different thalamic nuclei, we combined the autopsy images of the thalamus removal with the MRI post-processed images to define the anterior–posterior organization. We used bright field images for Nissl, PLP and synaptophysin to identify structures such as the choroid plexus or ependymal layers and to define the medial and lateral organization of the thalamic block. With the help of the neuroanatomy book [30], and by looking at the neuronal and synapse stainings, we identified the corresponding nuclei of each section.

Demyelination and neuronal loss

Whole slide scans of PLP stained sections were imported in ImageJ and MDn was outlined to calculate the surface area. All MDn lesions were manually annotated to calculate the respective area covered by lesions. These two areas were used to calculate the percentage of demyelinated MDn.

Per section, 3 confocal images for HuC/HuD / DAPI within the thalamic MDn were taken using a 20 × objective and a z-stepside of 0.5 µm on a Nikon AXR confocal microscope. To quantify neuronal density, these images were imported in QuPath (version 0.2.1). Cell detection function was used to count all DAPI+ nuclei. The numbers of HuC/HuD+DAPI+ neuronal cells were manually counted in QuPath. We determined the percentage of positive cells divided by the total number of DAPI+ nuclei and multiplied by 100 and the absolute count of neuronal cells divided by the total area of the images.

Mediodorsal adaptive immune cells

Fluorescent whole-slide images stained for CD3/CD19/DAPI were acquired with a 20 × objective on a Vectra Polaris scanner. Of the whole slides, we blindly selected 10–15 random areas of the MDn. These regions were saved as TIFF files using QuPath (version 0.2.1). NIS Elements software (version 5.42.01) was used to count all DAPI+ nuclei. The numbers of CD3+DAPI+ and CD19+ DAPI+ immune cells were manually counted in FIJI. We determined the percentage of positive immune cells divided by the total number of DAPI+ nuclei and multiplied by 100 and the absolute count of immune cells divided by the total area of the images.

Microglia density, protein expression and morphology

Per section, 5 confocal images for IBA1/P2Y12/DAPI within the thalamic MDn and 5 confocal images for IBA1/HLA-DR/DAPI within the thalamic MDn were taken using a 40 × objective and a z-stepside of 0.5 µm on a Nikon A1R confocal microscope and a Nikon AXR confocal microscope, respectively. In addition, 10 confocal images for IBA1/P2Y12/DAPI randomly distributed, regardless of MDn position, were taken in the whole thalamus using a 40 × objective and a z-stepside of 0.5 µm on a Nikon A1R confocal microscope.

For microglial density, NIS Elements software (version 5.42.01) was used to count all DAPI+ nuclei and IBA1+DAPI+ cells in the confocal images for IBA1/P2Y12/DAPI within the thalamic MDn and the images randomly distributed in the whole thalamus. The average count of the images was used for the following calculations: we determined the percentage of microglia as IBA1+DAPI+ cells divided by the total number of DAPI+ nuclei and multiplied by 100; we calculated the microglia density as IBA1+DAPI+ cells divided by the volume of the images.

For P2Y12 and HLA-DR protein expression, Imaris software (version 9.9.1, Bitplane AG) was used to segment all IBA1+ microglia using the surface function. From these surfaces, we measured the total fluorescence intensity sum of P2Y12 or HLA-DR divided by the total microglial volume. The final mean fluorescence intensity corresponded to the average of the 5 images.

For microglial morphology, we generated 2D maximum intensity projections of the aforementioned confocal images and manually traced IBA1+ P2Y12+ cells in a blind setting using FIJI as previously described [48]. We randomly selected microglia following this inclusion criteria: microglia should be included within the borders of the image, microglia should not overlap with any other cell and they should not be interacting with vessels. All traced cells were analysed using the Sholl Analysis Plugin [17] with a 0.3 µm step size from the cell soma. Per image, we analysed around 10 cells.

Pre-synaptic density and pre-synaptic microglial phagocytosis

Sections stained for IBA1/LAMP1/Synaptophysin/DAPI were acquired on a Leica TSC SP8 confocal microscope with a 100 × objective and a z-stepside of 0.1 µm. Per donor, 20 confocal images were taken in the MDn including one or two microglia in each image. We analysed pre-synaptic phagocytosis by microglial cells as previously described [48]. Briefly, we generated surfaces for DAPI+, IBA1+, LAMP1+ and IBA1+LAMP1+ signal with the Imaris software (version 9.9.1, Bitplane AG). Using the spot function with a spot diameter of 0.5µm and a quality filter, Synaptophysin+ pre-synapses were detected. All the steps were blindly individually set per donor. Spots in IBA1+ and IBA1+LAMP1+ surfaces were identified as spots that had a maximum distance of 0µm to these respective surfaces. Microglia phagocytosis was calculated by dividing the number of spots in IBA1+ and IBA1+LAMP1+ surfaces by the total number of spots or the total volume of the images. The volume of lysosomes in microglia was calculated as the average of the 20 images of the volume of LAMP1+ in IBA1+ surface.

To quantify pre-synaptic density, sections stained for Synaptophysin were scanned on an Olympus VS200 slide scanner. Specifically, 20 images per donor with a 60 × objective and a z-stepside of 0.25µm were acquired. Images were blindly taken in different areas of the MDn. The images were uploaded in Huygens software (version 23.04) to remove autofluorescence and deconvolve. Deconvolved images were uploaded in Imaris software (version 9.9.1) and using the spot function with a spot diameter of 0.5µm and a quality filter, Synaptophysin+ pre-synapses were quantified. Results are shown as the total number of pre-synapse spots divided by the total volume of the images and the percentage of the total volume occupied by pre-synapse spots divided by the total volume of the images.

Statistics

For the MRI analysis, SPSS (version 28) was used for all statistical tests, while for the microscopy analysis, GraphPad Prism 9 was used. To assess the differences between control and MS donors across thalamic nuclei volumes, multivariate analysis of variance (MANCOVA) was conducted. Sex and post-mortem delay (time between death and in-situ MRI) were included in the model as covariates. To investigate whether cortical regions structurally connected to the MD nuclei of the thalamus also exhibit atrophy in MS, we identified corresponding frontal regions using FreeSurfer-based anatomical labels and tractography data from each subject. These included the anterior cingulate gyrus, middle anterior and middle posterior cingulate gyri, and the middle frontal gyrus, known to receive projections from the MD nuclei. We extracted the cortical volumes of these regions and performed a MANCOVA comparing MS and control donors, including sex and post-mortem delay to MRI as covariates. Effect sizes were calculated using Partial eta squared (η2). For microscopy analysis, we first tested for normality using Shapiro–Wilk to decide the appropriate statistical tests. Unpaired two-tailed Student’s t-test with or without Welch’s correction for unequal variances, or the non-parametric Mann–Whitney test were used accordingly. For correlations between MRI and histopathological variables, MRI data was selected from the same side as the thalamus block collected during autopsy for immunolabeling and partial Spearman’s correlations with post-mortem delay as a covariate were performed in SPSS. Data were considered significant when P < 0.05 and reported in the figures using the corresponding significance levels and statistical tests indicated in the figure legends. To plot the heatmaps of the r value from the correlations, we used the package pheatmap in R studio (version 4.2.1). Exclusion criteria from certain analysis was the absence of antibody reactivity or poor-quality diffusion measurements, as indicated in Table 1.