Myeloid cell iron uptake pathways and paramagnetic rim formation in multiple sclerosis

Cohort A was used to explore iron and important iron-related proteins in MCs across different activity stages of MS lesions (Supplementary Table 1). It consists of formalin-fixed, paraffin-embedded (FFPE) autopsy brain tissue samples from 18 controls and 24 pwMS, approved by the ethics committee of the Medical University Vienna (535/2004/2016). Confirmatory cohort B consists of 23 snap-frozen tissue samples obtained from 6 controls (6 samples) and 14 pwMS (17 samples), provided by the UK Multiple Sclerosis Tissue Bank at Imperial college in London, following ethical approval by the National Research Ethics Committee in the UK (18/WA/0238). Controls displayed absence of neurological disease in the medical history and no brain lesions at neuropathological examination. For cohort B, only MS cases with chronic active lesions were included. Further epidemiological, clinical and basic pathological information for both samples is provided in Supplementary Table 1. Primary antibodies and antigen retrieval methods are listed in Supplementary Table 2.

For cohort A, 3 µm thick FFPE tissue sections were stained with hematoxylin & eosin (H&E) and Luxol fast blue–periodic acid Schiff (LFB-PAS) myelin staining. MS lesion staging was performed based on LFB-PAS stainings and immunohistochemistry (IHC) for the myelin proteolipid protein (PLP; n = 28; see Supplementary Table 2). For detection of iron, diaminobenzidine (DAB)-enhanced Turnbull blue (TBB) staining was applied as described [38]. IHC using DAB as chromogen was performed as described [6]. Iron, transferrin receptor (TfR, encoded by TFRC), scavenger receptor class A member 5 (Scara5, encoded by SCARA5), divalent metal transporter (DMT1, encoded by SLC11A2), natural resistance-associated macrophage protein 1 (NRAMP1, encoded by SLC11A1), CD163 (encoded by CD163), ferroportin (encoded by SLC40A1), hephaestin (encoded by HEPH) and hepcidin (encoded by HAMP) immunoreactivities were assessed on consecutive tissue sections. For cohort B, 16 µm thick snap-frozen tissue sections were stained for LFB-PAS and DAB-enhanced TBB staining for ferric and ferrous non-heme iron. For IHC, antibodies against MOG, CD68 and CD163 were applied (see Supplementary Table 2). MOG and CD68 IHCs were performed as described previously [53]. CD163 IHC on frozen sections was done on a DAKO Link 48 autostainer with one hour incubation at room temperature (RT).

For duplex and multiplex single molecule RNA in situ hybridization (ISH) performed on cohort B, the ACD bio-techne protocols for the RNAscope 2.5 HD Duplex [53] and Fluorescent Multiplex V2 assays were applied as previously described [53, 59]. The following human ACD bio-techne RNAscope assay probes (cat. no.) were used: C1QA (485451), CD163 (417061), HAMP (411911), HMOX1 (319851), IL10 (602051), P2RY12 (450391). In addition, human ACD bio-techne 3-plex negative (320871) and positive ISH probes (320861) were run in parallel as a quality control.

In cohort A, lesions were staged according to myelin degradation products in lysosomes of macrophages based on LFB-PAS histology and PLP IHC and the presence of activated CD68⁺ MCs. The following regions of interest were investigated: normal white matter (WM) of controls (CTRL), MS normal-appearing WM (NAWM) in at least 10,000 µm distance to any discernible lesion rim (LR) [21], WM surrounding lesions (peri-plaque WM, PPWM), early active areas (LC-EA) and late active areas (LC-LA) of active lesions (A) (n = 12, see Supplementary Table 1). For chronic active (CA) lesions (n = 11), we investigated PPWM, LRs, and inactive lesion cores (LC-I). Active lesions were identified by a hypercellular lesion core with a high density of MCs containing myelin degradation products. In active lesions, LR areas—although not as distinct as in chronic active lesions—were characterized by a border zone with a relative loss of myelin extending into the LC associated with the presence of phagocytosing MCs. Chronic active lesions showed a hypocellular, demyelinated lesion center but distinct rim formation by CD68⁺ cells. The inactive lesion (consisting of one NAWM, one PPWM and one LC-I) was characterized by a fully demyelinated, hypocellular lesion center and low presence of CD68⁺ cells throughout lesion areas. The identification of MCs was based on morphology. Cells were quantified with an Olympus BX50 microscope using a 40 × objective. 9 fields per ROI were counted using a grid covering 0.24 × 0.24 mm resulting in a total area of 0.52 mm² per ROI. One lesion per case and one ROI per lesion and ROI type were evaluated for the data presented (Fig. 2). For cohort B, chronic activity of lesions was confirmed based on LFB-PAS and MOG IHC for few rim-associated myelin degradation products and CD68 IHC for the presence of MCs.

For cohort B, images for quantification were taken using a Leica DMi 8 microscope equipped with a Leica DFC 7000 GT camera. Focus points were set at 20 × magnification at the area of interest, pictures were then imaged and exported as LIF files. Confocal images were taken using Leica TCS SP8, Nikon AX R or Nikon A1 microscopes. All confocal pictures were taken as z-stack images consisting of 10 to 20 layers with a 0.5 to 0.7-μm step size. Heights for z-stacks were identified manually by imaging DAPI.

To detect gene expression changes in chronic active lesion areas of cohort B, we defined 4 ROIs that were examined in every lesion. As described before [24], the following ROIs were examined: control WM (CTRL), MS NAWM in at least 10,000 µm distance from the lesion, MS PPWM in 100–400 µm distance to the LR, MS LC in 500–1000 µm distance to the LR. For the quantification of each sample, we took 4 images within 4 different ROIs (LC, LR, PPWM, NAWM) using a 20 × objective. Each image covers 0.05–0.1 mm² depending on the lesion size and resulting in a total of 16 images per lesion. For controls, 4 CTRL ROIs per sample, each covering an area of 0.05–0.1 mm², were quantified accordingly. Quantification analysis was performed manually using FIJI ImageJ (version 2.0.0). For RNA in situ hybridization assays, cells presenting two or more signals were considered as positive cells.

In the cross-sectional, double center retrospective MRI study part, 86 pwMS from the Department of Neurology of the Medical University of Vienna and 12 pwMS from the University Medical Centre Mannheim were included, yielding a pooled cohort of 98 pwMS. Clinical and demographic characteristics of the cohort are given in Table 1. Diagnosis of clinically definite MS was established according to the McDonald criteria at the time of diagnosis between Jan 1st, 1991 and Dec 31st, 2020. All pwMS met the following inclusion criteria: age ≥ 18 years, availability of a T1, FLAIR and SWI-based brain MRI scan at 3 T and a blood sample drawn at any time independent of the MRI. The study was approved by the ethics committees of the Medical University of Vienna (EK 1599/2021) and Mannheim (2017-830R-MA). MRI analysis and processing of blood samples for Hp genotyping were performed at the Medical University of Vienna.

In Vienna, all brain scans were performed on a Siemens Magnetom 3 T Trio (2015–2017) and Vida (2018–2020) MRI system, using a 32-channel radio frequency (RF) coil. Images were acquired between Jan 1st, 2015, and Dec 31st, 2020: isovoxel 1 mm³ 3D fluid-attenuated inversion recovery (FLAIR) (echo time TE = 398 ms, repetition time TR = 5000 ms, inversion time TI = 1800 ms, field of view (FOV) = 240 mm², resolution = 0.9 × 0.9 × 0.9 mm³), 3D T1-weighted images (TE = 2.92 ms, TR = 1800 ms, TI = 900 ms, FOV = 256 mm², resolution = 1 × 1 × 1 mm³) and multi-echo susceptibility-weighted image sequence (SWI) (TE 1 = 7.67 ms, TE 2 = 24.60 ms, TR = 35 ms, FOV = 220 mm², resolution = 0.3 × 0.3 × 1.5 mm³) were acquired consecutively. In Mannheim, all MRI scans were performed on a Siemens Magnetom 3 T Skyra System, using a 20-channel head coil. The following sequences were used: 3D FLAIR (TE = 398 ms, TR = 5000 ms, TI = 1800 ms, FOV = 240 mm², resolution = 0.5 × 0.5 × 0.9 mm³), 3D MPRAGE (TE = 2.49 ms, TR = 1900 ms, TI = 900 ms, FOV = 240 mm², resolution = 0.9 × 0.9 × 0.9 mm³),, and SWI (TE = 20 ms, TR = 27 ms, FOV = 220 mm², resolution 0.9 × 0.9 × 1.5 mm³).

All supratentorial lesions of the periventricular, juxtacortical and deep WM in the frontal, parietal, temporal and occipital lobes [19] as well as infratentorial lesions of the cerebellum were analyzed by two independent raters (ADB, NK) with a long-standing expertise in MS imaging analysis. PRLs were defined as FLAIR-hyperintense lesions that were partially or completely surrounded by a pronounced and distinct SWI-hypointense rim. The presence of a central vein was not considered for rim evaluation. After both raters had made their decision, unclear lesions were discussed and an agreement was reached. The inter-rater agreement before matching was 98.7%. Volume of FLAIR and T1 lesion, and total brain volume were automatically assessed using the MorphoBox prototype imaging software normalized for age from Siemens Healthineers [54].

Blood was drawn from the patients into EDTA tubes and immediately processed by centrifugation for 15 min at 15,000 g at RT. Next, plasma was removed, and remaining blood cells were stored at − 80 °C. For DNA isolation, the Qiagen DNeasy blood and tissue kit together with the corresponding protocol for non-nucleated blood cells was used. All centrifugation steps were carried out at RT. Blood samples were removed from the freezer and instantly incubated in the water bath (37 °C) until the blood was liquified. For sample lysation and RNA subtraction, 100 µl of blood was mixed with 100 µl 1 × PBS, 20 µl Proteinase K (Qiagen) and 4 µl RNAse A (Qiagen) and incubated for 2 min at RT. After adding 200 µl Buffer AL (Qiagen) and incubation for 10 min at 56 °C on a rocking platform, 200 µl ethanol 100% was added and the mixture was transferred into the provided DNeasy mini spin column, placed in a 2 ml tube, and centrifuged at 8000 rpm for 1 min. The DNeasy mini spin column was transferred into a new 2 ml tube container, and enzyme inhibitors were removed by the following two washing steps: 500 µl Buffer AW1 (Qiagen) centrifuged at 8,000 rpm for 1 min followed by 500 µl Buffer AW2 and again centrifuged at 14,000 rpm for 3 min. DNA elution of the membrane was performed by placing the spin column into a new 2 ml tube, adding 100 µl RNAse free water, incubated for 1 min at RT and then centrifuged for 1 min at 8000 rpm. To increase DNA yield, this procedure was repeated. DNA yields (range of 3–90 ng/µl) were measured using the Qubit RNA/DNA work station (Thermo Fisher) to ensure a sufficient concentration of extracted DNA.

Hp haplotypes were determined by allelic amplification in two distinct PCR reactions and separated by agarose gel electrophoresis using primers A (5 -GAGGG GAGCTTGCCTTTCCATTG-3) and B (5–GAGATTTTTGAGCCCTGGCTGGT-3) as well as primers C (5 -CCTGCCTCGTATTAACTGCACCAT-3) and D (5 -CCGAG TGCTCCACATAGCCATGT-3) according to Koch et al. [26]. All experiments were performed in duplicates.

CSF samples were obtained at the time of diagnosis with a median time of ± 4 months from the MRI, according to Consensus Guidelines for CSF and Blood Biobanking [58]. After centrifugation, CSF supernatant and cell pellet were stored separately at − 80 °C. CSF analysis was optimized and performed by two independent investigators, blinded for the patients’ clinical and MRI features. The concentration of sCD163 (ng/mL/mgProt) was assessed by using the ultrasensitive Ella-Simple Plex technology (ELLA microfluidic analyzer, Protein Simple, Bio-techne), an automated ELISA, according to the manufacturer’s protocol. To verify the reproducibility and consistency of the results, all samples were run in triplicates.

In FFPE samples of the exploratory cohort A, we found accumulation of iron-laden MCs in late active lesion zones (LC-LA) and chronic active lesion rims (LR) (Figs. 1 and 2), as reported [24]. In LC-LA regions, this was paralleled by elevated MC numbers IHC-positive for the iron uptake markers TfR (TfR, TFRC), DMT1 (SLC11A2), NRAMP1 (SLC11A1), Scara5 (SCARA5) and – by far the highest—CD163, the scavenger receptor for hemoglobin-haptoglobin complexes. In LR regions of chronic active lesions, however, only DMT1 and CD163 were increased, when compared to either CTRL (DMT1 and CD163) or NAWM (only CD163) (Fig. 2). Of note, the transferrin receptor, while readily detectable on membranes of MCs in early active lesion zones (LC-EA) and LC-LA regions (Fig. 1I), was not enriched on MCs at LR regions. In addition, increased numbers of MCs immunoreactive for the iron exporter proteins ferroportin (SLC40A1) and hephaestin (HEPH) as well as the functional ferroportin antagonist hepcidin (HAMP) were found in LC-LA and LR regions, suggesting iron-related increase of ferroportin translation and its functional downregulation via hepcidin expression. Correlation analysis between MC counts of the mentioned iron importers and iron across samples with active and chronic active lesion ROIs revealed by far the strongest correlations between iron and CD163 (r² = 0.4017, p < 0.001) as well as DMT1 (r² = 0.4491, p < 0.001) in chronic active lesions (Supplementary Fig. 1). Due to the explorative approach in cohort A, not all stainings were available for the entire cohort. In some cases, active lesions only had early active and no late active zones, hence, resulting in different numbers of data points. In the confirmatory cohort B of frozen samples containing chronic active lesions only, we found increased numbers of CD68⁺ MCs at the LR vs. LC (p = 0.016), NAWM (p = 0.002) and CTRL (p = 0.001) (Fig. 3). CD163⁺ cells were more numerous at the LR, compared to NAWM (p = 0.001) and CTRL (p = 0.008). Iron-positive MCs were likewise more numerous at LR areas when compared to LC (p = 0.004) and CTRL (p = 0.028). Quantification levels correlated for CD163 and CD68 (r² = 0.1821, p < 0.001), CD163 and iron (r² = 0.0496, p = 0.033) and CD68 and iron (r² = 0.3527, p < 0.001). These data suggest that MC iron uptake at MS lesion rims is less likely mediated by transferrin but rather driven through CD163 and haptoglobin-hemoglobin complexes.

By RNA ISH on cohort B consisting of frozen samples, upregulation of CD163 was detected at the LR compared to LC (p < 0.001), NAWM (p = 0.037) and CTRL (p = 0.001) areas (Fig. 4). Moreover, we detected increased numbers of CD163-expressing cells at PPWM areas compared to LC (p = 0.012) and CTRL (p = 0.050). Both HAMP-only (encoding hepcidin) (p = 0.010) and CD163/HAMP- (p = 0.007) co-expressing MCs were found to be increased at LR areas compared to LC. Conversely, P2RY12-expressing MCs, indicating a homeostatic MC phenotype [65], were found to be decreased at LC areas vs. CTRL (p = 0.005), NAWM (p < 0.001), PPWM (p < 0.001) and LR (p = 0.001), confirming previous data [65]. HMOX1, encoding heme oxygenase 1 that catalyzes heme to iron, carbon monoxide and biliverdin [11], was upregulated at the LR, compared to NAWM (p = 0.005) and CTRL (p = 0.033) (Fig. 5). Cells co-expressing CD163 and HMOX1 accumulated at the LR, compared to LC (p < 0.001) and CTRL (p < 0.001). Of note, numbers of CD163-expressing MCs correlated with HMOX1 (r² = 0.1495, p = 0.003) and HAMP expression (r² = 0.0632, p = 0.010) as well as presence of iron-laden MCs (r² = 0.2155, p < 0.001). Furthermore, HAMP expression and presence of iron-laden MCs correlated with each other (r² = 0.1814, p < 0.001). Taken together, we could demonstrate that the CD163-HMOX1-HAMP axis was upregulated in MCs at chronic active lesion rims in a spatially restricted pattern.

To gain further insight into the functional properties of CD163-expressing MCs at MS lesion rims, we focused on IL10 and C1QA expression as two signature markers for regulatory (IL10) [41] and pro-inflammatory (C1QA) functions [1, 16, 33, 52]. C1QA expression was elevated at the LR vs. LC (p = 0.0087) and NAWM (p = 0.042), as described [1] (Fig. 6). Next, we assessed co-expression with CD163 and found that C1QA co-expressing cells were increased at the LR vs. LC (p = 0.042) (Fig. 6), while IL10 co-expressing cells were upregulated at PPWM vs. LC (p = 0.030). These findings suggest that both pro-inflammatory and, to a certain extent, regulatory functions are associated with CD163 expression at spatially restricted lesion border areas in MS (Supplementary Fig. 2).

A total of 341 PRLs in 61 pwMS (62.2%) were analyzed; 37 pwMS (37.8%) had no PRLs. Median number of PRLs per individual was 1 (IQR 0–4). PRLs were more commonly seen in men (75.6% vs. 52.6%, p = 0.034) with longer disease duration (8 [2.5–13] vs. 4 [2.5–7.5], p = 0.034) and higher disability scores measured by EDSS (2.5 [1.4–4.0] vs. 1.0 [0–2.8], p = 0.002) and MSSS (3.65 [1.74–5.87] vs. 2.34 [0.45–5.35], p = 0.020). Male sex (OR 3.83; 95% CI 1.44, 10.19; p = 0.007) and longer disease duration (OR 1.10; 95% CI 1.00, 1.22; p = 0.044) were both independently associated with higher risk for the PRL presence, whereas age, EDSS and Hp genotype were not. Individuals with early (0 [0–2]) and intermediate RMS phases (1 [0–2]) had significantly lower number of PRLs compared to pwMS in the late RMS phase (2.5 [1–8.5]) (p = 0.013 and p = 0.027, respectively) (Fig. 7). No differences in PRL counts were observed between pwMS with RMS and SPMS (4 [1.5–13], p = 0.606). Among pwMS without PRLs, only 4 (10.8%) had PMS (p = 0.045) (Table 2). The number of PRLs correlated with the age of pwMS (r_s = 0.2264, p = 0.025), disease duration (r_s = 0.2774, p = 0.006), EDSS (r_s = 0.4036, p < 0.001), and MSSS (r_s = 0.2620, p = 0.010). In a multivariate linear regression model, the number of PRLs was associated with EDSS (β = 0.22; 95% CI 0.04, 1.34; p = 0.039) and disease duration (β = 0.31; 95% CI 0.10, 0.51; p = 0.003) but not with age, sex or Hp haplotype of pwMSs (Fig. 7). Sensitivity analyses removing individuals from the Mannheim MRI cohort did not significantly change the overall results or impact of single variables (data not shown).

To gain insight into the relationship between Hp haplotype and clinical as well as MRI parameters, we analyzed DNA samples from pwMS: Hp1-1 (n = 8, 8.2%), Hp2-2 (n = 39, 39.8%) and Hp2-1 (n = 51, 52.0%). Of note, this distribution was in Hardy–Weinberg equilibrium [45]. pwMS with Hp2-2 and Hp2-1 pooled genotypes showed higher EDSS (2.0 [1.0–3.5] vs. 0.5 [0–1.8]; p = 0.041) and MSSS levels (3.17 [1.27–3.87] vs. 0.56 [0.34–3.84]; p = 0.038) compared to pwMS with an Hp1-1 haplotype. Differences between the Hp haplotype in age (38.5 [10.8] vs. 31.9 [3.9], p = 0.088), sex (52 [57.8] vs. 5 [62.5], p > 0.999), disease duration (6 [2–10.5] vs. 7 [3–11.8], p = 0.800), RMS phase (69 [76.7] vs. 7 [87.5], p = 0.679), and the number of PRLs were not significant (1 [0–4] vs. 1.5 [0–4.5], p = 0.883).

In 38 pwMS with available CSF sample, sCD163 measurements were performed. The median CSF sCD163 concentration was 28.3 ng/mL (22.9–39.2). pwMS with \(\ge\) 4 PRLs had higher CSF sCD163 levels than pwMS with 1–3 PRLs (39.5 [29.9–56.5] vs. 26.7 [17.6–37.9], p = 0.047) (Fig. 7). As no differences in the sCD163 concentration between pwMS without PRLs (25.8 [23.0–29.7]) and pwMS with 1–3 PRLs were seen (p > 0.999), those were pooled. pwMS with \(\ge\) 4 PRLs had significantly higher sCD163 levels than individuals with \(\le\) 3 PRLs (p = 0.009). CSF sCD163 concentration correlated with the number of PRLs (r² = 0.1351, p = 0.023) but not with age, disease duration or EDSS (Fig. 7). After excluding two outliers from the analysis, statistical significance was lost, yet a trend towards higher sCD163 in pwMS with \(\ge\) 4 PRLs was still present (p = 0.059) Fig. 8.