Distinct transcriptional changes distinguish efficient and poor remyelination in multiple sclerosis

Abstract

Multiple sclerosis (MS) is a highly heterogeneous disease, with varying remyelination potential across individuals and between lesions. However, the molecular mechanisms underlying the potential to remyelinate remain poorly understood. In this study, we aimed to take advantage of the intrinsic heterogeneity in remyelinating capacity between MS donors and lesions to uncover known and novel pro-remyelinating molecules for MS therapies.

To elucidate distinct molecular signatures underlying the potential to remyelinate, we stratified MS donors from the Netherlands Brain Bank cohort (n = 239), based on proportions of remyelinated lesions (RLs), into efficiently remyelinating donors (ERDs; n = 21) and poorly remyelinating donors (PRDs; n = 19). We performed bulk RNA sequencing of RLs, active lesions with ramified and amoeboid microglia/macrophage morphology (ALs non-foamy), active lesions with foamy microglia/macrophage morphology (ALs foamy) and normal-appearing white matter (NAWM) from ERDs and PRDs.

We found that ALs non-foamy were positively correlated with remyelination, whereas ALs foamy were not, indicating a role for microglia/macrophage state in influencing remyelination potential. Bioinformatics analyses were performed to identify key pathways and molecules implicated in the remyelination process. We found distinct differences between the donors with differing remyelination potential in comparable MS lesion types. The RLs and ALs non-foamy of ERDs versus PRDs showed upregulation of the epithelial–mesenchymal transition pathway, whereas in ALs foamy of PRDs, inflammation and damage-associated pathways (i.e. MTORC1 signalling, TNF signalling and oxidative phosphorylation) were upregulated in comparison to ALs foamy of ERDs, suggesting that these latter pathways might counteract remyelination. We found genes significantly upregulated in RLs and/or ALs non-foamy of ERDs that have previously been associated with remyelination, including CXCL12, EGF, HGF, IGF2, IL10, PDGFB, PPARG and TREM2, illustrating the strength of our donor and lesion stratification. TGFB1, TGFB2, EGF and IGF1 were determined to be key upstream regulators of genes upregulated in RLs and ALs non-foamy of ERDs. We also identified potential novel pro-remyelinating molecules, such as BTC, GDF10, GDF15, CCN1, CCN4, FGF5, FGF10 and INHBB.

Our study identified both known and novel genes associated with efficient remyelination that might facilitate the development of therapeutic strategies to promote tissue repair and clinical recovery in MS.

active lesions, microglia, growth factors, repair

Introduction

Multiple sclerosis (MS) is a chronic inflammatory disease of the CNS, characterized by intermittent attacks (relapses) and a subsequent progressive increase in neurological disability.^1-3 Although current MS treatments can prevent relapses and delay disability in some patients, regenerative therapies to limit progression remain a major unmet medical need in MS. Pathologically, people with MS show focal areas of demyelination, inflammation and axonal degeneration. Endogenous myelin repair, known as remyelination, occurs in the CNS and is an important component for the restoration of damaged tissue, because it mediates protection from further axonal degeneration and promotes functional recovery.^4-6 Previous studies have shown that in response to demyelinating injury, new myelin is synthesized either by newly formed oligodendrocytes generated from oligodendrocyte precursor cells (OPCs)^7,8 or by pre-existing mature oligodendrocytes.^9,10 A recent study showed that myelin regeneration by newly formed oligodendrocytes is most efficient, with the remyelination process being both more accurate and more extensive.¹¹

The current perspective on myelin regeneration is that it involves OPC proliferation and migration to the demyelinated lesion site and their subsequent differentiation into mature oligodendrocytes.^5,6,12 These processes are controlled/affected by growth factors (e.g. EGF,^13,14 TGFβ,^15,16 INHBA,¹⁷ IGF,¹⁸ PDGF,^18-20 FGF^18,21,22 and HGF¹⁶), neurotrophic factors (e.g. CNTF²³ and BDNF²⁴), chemokines (e.g. CXCL12²⁵) and guidance molecules (e.g. semaphorins 3A and 3F²⁶). Additionally, factors that promote clearance of myelin debris following demyelination (e.g. TREM2^27,28) are required for myelin repair to proceed, primarily expressed by microglia/macrophages.

Microglia and macrophages play a role in both demyelination and remyelination through secretion of pro-inflammatory (e.g. TNF²⁹), anti-inflammatory (e.g. IL-10²⁹) and regenerative factors (e.g. IGF-1, HGF and VEGF³⁰) and by the phagocytosis of myelin debris.³¹ The heterogeneous role of microglia and macrophages in de- and remyelination is attributed to diverse activation states of these cells.^17,31 In MS, microglia/macrophages present with different morphologies in active lesions (ALs): ramified, amoeboid or foamy.² Although previous studies have documented inconsistent findings with regard to the pro- or anti-inflammatory role of specific microglia/macrophage phenotypes,^32-34 we recently demonstrated high acute axonal damage and a correlation with increased CSF neurofilament levels in MS white matter (WM) ALs and mixed active/inactive lesions (MLs) containing foamy, but not ramified, microglia/macrophages.³⁵

In general, remyelination capacity declines with age in people with MS owing to impaired OPC recruitment and differentiation.³⁶ Remyelination generally occurs in relapsing MS and in a subset of people with progressive disease,³⁷ and it can still be extensive even after a long disease course.³⁸ However, for reasons that remain unknown, remyelination capacity varies significantly between individuals with MS. We previously observed that Netherlands Brain Bank (NBB) donors die with a large variability in the proportion of remyelinated lesions.² Moreover, there is increasing evidence for an effect of lesion stage on remyelination, with higher remyelinating potential in ALs in comparison to MLs.³⁹

In this study, we take advantage of the intrinsic heterogeneity in remyelination between MS donors and lesions to uncover pro-remyelinating molecules for new MS therapies. We stratified MS brain donors based on their proportion of remyelinated lesions (RLs) in 239 donors of the NBB MS autopsy cohort, developing a cohort of efficiently and poorly remyelinating donors (ERDs and PRDs, respectively). From these donors, we selected lesions with different potential for remyelination, including ALs with different microglia/macrophage morphologies, to capture potential ongoing remyelination, in addition to already remyelinated lesions, and performed RNA sequencing. By comparing pathological and molecular characteristics between brain tissue from ERDs and PRDs, we reveal both known and novel molecules relating to efficient remyelination. These promising molecules might facilitate the development of therapeutic strategies to promote tissue repair and clinical recovery in MS.

Materials and methods

Human post-mortem tissue

All human post-mortem brain tissues were provided by the NBB. Informed consent was obtained from donors during life for the use of tissue and clinical data for research purposes. The NBB autopsy procedures were approved by the Ethics Committee of the VU University Medical Center (Amsterdam, The Netherlands). Neurological diagnosis was confirmed post-mortem by a certified neuropathologist. Clinical and pathological information was analysed from 239 MS donors from the NBB autopsy cohort (collected from 1990 to 2020). For RNA sequencing, a total of 46 snap-frozen tissue blocks from 40 unique MS donors were collected, containing 21 RL, 14 AL non-foamy, 14 AL foamy and 47 normal-appearing WM (NAWM) samples (Supplementary Table 1). For immunohistochemical analyses of TGFβ1, TGFβ2, EGF, BTC and BCAS1, a total of 33 formalin-fixed paraffin-embedded tissue blocks from 30 MS donors were collected, including 25 RLs, 15 ALs non-foamy and 16 ALs foamy.

Donor selection

MS donors were stratified as efficiently remyelinating donors (ERDs) and poorly remyelinating donors (PRDs), based on the proportions of RLs at autopsy. MS donors with a proportion of remyelinated lesions (RLs) above the median (≥0.27) of the NBB cohort (1990–2020) were considered ERDs, and donors with an RL proportion below the median (<0.27) were considered PRDs. In our study, we selected 21 ERDs, with 55% of all lesions being RLs on average. A total of 19 PRDs were selected, with only 12% of all lesions being RLs on average. From each MS donor, one to three WM lesions and one to three NAWM samples were collected. For histological validation, a total of 13 formalin-fixed paraffin-embedded tissue blocks from 13 MS donors were collected (Supplementary Table 2).

Calculations of lesion proportions and lesion load

The proportions of lesion subtypes were calculated from all MRI- and macroscopically identified lesions and from all lesions found in standardly dissected tissue blocks (brainstem and spinal cord), with an average of 47.0 ± 43.7 lesions per donor at autopsy. The proportion of RLs was calculated by dividing the amount of RLs by all RLs plus inactive lesions. Proportions of all ALs and MLs, including AL non-foamy, AL foamy, ML non-foamy and ML foamy, were calculated by dividing by all ALs plus MLs. Lesion load [log(x + 1)] from standardly dissected brainstem tissue was calculated and transformed as done previously.²

Lesion classification and tissue dissection

RLs, ALs non-foamy, ALs foamy and NAWM from the same tissue block were included in this study. MS lesion types were classified based on double immunostaining of human leucocyte antigen (HLA-DR-DP-DQ, referred to as HLA) [M0775; DAKO, with 3,3′-diaminobenzidine (DAB)-nickel] and proteolipid protein (PLP) (MCA839G; Serotec, with DAB) as previously described.² Frozen sections were stained with Luxol Fast Blue for additional visualization of myelin lipids.²⁶ The characterization of MS lesions was done as previously described.^2,40,41 Briefly, RLs displayed partial myelination and sparse HLA⁺ microglia/macrophages, whereas ALs displayed loss of myelination and accumulation of HLA⁺ microglia/macrophages throughout the lesion. Dissection of MS tissue samples from frozen sections for molecular analyses was guided by staining tissue sections (8 µm thick) with Sudan Black B (Sigma-Aldrich) and thionin (Merck). Adjacent sections (50 µm thick) were cut from the frozen tissue blocks for RNA sequencing analysis. Individual lesion and NAWM samples were dissected in a cryostat with a prechilled scalpel and collected separately. All samples were stored in 500 µl TRIsure™ (Bioline) at −80°C until further use.

Literature search for remyelination-associated genes

The PubMed database was searched for studies published before 8 December 2022, using the combination of search terms ‘growth factor’ and ‘remyelination’, present in titles and/or abstracts. Only peer-reviewed articles published in English were considered. Cross-references were additionally assessed for eligibility.

Immunohistochemistry

Formalin-fixed paraffin-embedded human brain sections (8 µm thick) were deparaffinized and rehydrated in a series of xylene and ethanol. Next, antigen retrieval was performed by microwaving in citrate buffer pH 6.0 or citraconic anhydride pH 7.4 for 10 min at 700 W. All sections were blocked with blocking buffer (1% bovine serum albumin + 0.5% Triton X-100 + 10% normal horse serum in Tris-buffered saline, pH 7.6). Sections were incubated with primary antibodies overnight at 4°C. Antibodies included TGFβ1 (Ab215715, 1:100; Abcam), TGFβ2 (116020, 1:100; NovoPro, Shanghai, China), EGF (Ab9695, 1:100; Abcam), BTC (AF-261, 1:100; R&D Systems) and BCAS1/NaBC1 (Sc-136342, 1:15 000; Santa Cruz). Subsequently, sections were incubated with biotinylated secondary antibodies (1:400), followed by incubation with an avidin–biotin complex kit (PK-6100, 1:800; Vector Laboratories). Additional signal enhancement was achieved via incubation with biotinylated tyramide (1:10 000 in PBS with 0.001% H₂O₂) for 10 min. Sections were visualized with the DAB⁺ DAKO REAL Envision detection kit (K500711-2, 1:100; DAKO), counterstained with Haematoxylin and dehydrated.

For fluorescent double-labelling, sections were blocked with avidin/biotin blocking solution for 15 min (R37627; Thermo Fisher Scientific). Primary antibodies HLA (M0775, 1:1000; DAKO), SOX10 (AF2864, 1:20; R&D Systems), Nogo-A (MAB3098, 1:50; R&D Systems), GFAP (Z0334, 1:800; DAKO) and Cy3-conjugated GFAP (C9205, 1:500; Sigma-Aldrich) were applied overnight at 4°C. Fluorescent-labelled secondary antibodies or biotinylated secondary antibodies were incubated with sections for 1 h at room temperature, followed by incubation with an avidin–biotin complex kit, and streptavidin-conjugated fluorophores for 1 h. All sections were incubated with Hoechst 33342 (H3570, 1:1000; Thermo Fisher Scientific) for 10 min and 0.1% Sudan Black in 70% ethanol for 5 min.

Images were acquired at ×20 magnification using a Zeiss Axio Scan.Z1 slide scanner (Carl Zeiss) and at 40× or 63× magnification using a stimulated emission depletion (STED) microscope (STEDYCON; Abberior Instruments). Data were processed using the Fiji plugin for ImageJ software and analysed using QuPath (v.0.4.0).

Statistical analysis

All correlations were tested using Pearson’s correlation coefficient. The Benjamini–Hochberg false discovery rate (FDR) test was used for multiple testing correction. The Kruskal–Wallis rank sum test was performed to test differences in clinical information across multiple groups. Differences between MS lesion subtype proportions were tested with quasi-binomial generalized linear models. The χ² test was also used to compare categorical variables across groups. A negative binomial generalized linear model with Tukey’s post hoc test was performed to compare multiple MS tissue groups for quantified protein levels of TGFβ1, TGFβ2 and EGF (glmmTMB package, v.1.1.7). For BTC, the percentage of DAB⁺ area was measured, and values were transformed using the qlogis() function. MS tissue groups were compared using the restricted maximum likelihood linear mixed model (nlme package, v.3.1-152). All analyses were performed using R (v.4.2.2).

Results

Active non-foamy lesions are correlated with high proportions of remyelinated lesions

RLs were observed in 74% of NBB MS donors (n = 239). The mean proportion of RLs (±SD) was 0.30 (±0.27) (median: 0.27, interquartile range: 0.48). To identify MS lesion types that might be associated with remyelination, we correlated the RL proportion of all MS donors with the AL proportion, AL subtypes and clinical information in the NBB autopsy cohort (Fig. 1A). We found a positive correlation between the RL proportion and the AL proportion (FDR = 4.8 × 10⁻⁵, R = 0.29), which was driven by ALs non-foamy, containing microglia with a ramified and amoeboid morphology (FDR = 4.2 × 10⁻⁴; Fig. 1A). A negative correlation was found between the RL proportion and the ML proportion (FDR = 4.4 × 10⁻⁵, R = −0.29; data not shown). This suggests that ALs non-foamy, but not ALs foamy or MLs, give rise to RLs. Accordingly, we found a significant positive correlation between the AL non-foamy proportion and clinical and pathological features indicative of a less severe disease course, such as decreased lesion load (FDR = 0.035, R = −0.17), longer disease duration (FDR = 1.9 × 10⁻³, R = 0.24) and older age at death (FDR = 0.013, R = 0.19). There was a negative correlation between the AL foamy proportion and disease duration (FDR = 9.7 × 10⁻³, R = −0.20; Fig. 1A). Hence, both RLs and ALs non-foamy were considered to be associated with remyelination, and ALs foamy were considered to be negatively associated with myelin repair. Consequently, these three lesion types (RLs, ALs non-foamy and ALs foamy), together with NAWM, were selected for further study based on Luxol Fast Blue and HLA-PLP double staining (Fig. 1B).

Figure 1

Identification and selection of multiple sclerosis (MS) donors and MS lesions with high remyelination potential. (A) Comparisons between MS lesion subtypes and clinical data of 239 MS donors (donated between 1990 and 2020) from Netherlands Brain Bank (NBB) show a positive correlation between remyelinated lesions (RLs) and active lesions (ALs). Donors with higher proportions of ALs non-foamy, but not ALs foamy, show lower lesion load, higher disease duration and older age at death. Statistics were performed using Pearson’s correlation with Benjamini–Hochberg multiple testing correction. *False discovery rate < 0.05, **false discovery rate < 0.01 and ***false discovery rate < 0.001. Grey bands in scatter plots represent the 95% confidence interval. (B) MS brain tissue [normal appearing white matter (NAWM), RLs, ALs non-foamy and ALs foamy] was assessed by Luxol Fast Blue and HLA-PLP staining. Dashed lines indicate dissection outlines. (C) The MS cohort selected for this study [(n = 21 efficiently remyelinating donors (ERDs) and n = 19 poorly remyelinating donors (PRDs)] reflects pathological properties of the large MS database, showing higher RL and AL non-foamy proportions, but not a higher AL foamy proportion in donors with a high remyelinated lesion proportion.

Select MS donors and lesions reflect pathological and clinical properties of a larger MS cohort

To explore the molecular pathways contributing to high potential for remyelination, MS donors with proportions of RLs above the median (≥0.27) of the whole NBB cohort (n = 239) were stratified as efficiently remyelinating donors (ERDs), and MS donors below the median (<0.27) were stratified as poorly remyelinating donors (PRDs). In the entire NBB MS cohort, the average proportion of RLs [±standard deviation (SD)] in ERD and PRD groups was 0.52 ± 0.19 and 0.08 ± 0.09, respectively (P < 0.0001). Both the AL proportions (ERDs: 0.40 ± 0.34; PRDs: 0.32 ± 0.28, P = 0.03) and AL non-foamy proportions (ERDs: 0.29 ± 0.32; PRDs: 0.21 ± 0.23, P = 0.03) were higher in ERDs. From the ERD and PRD subgroups within the whole NBB cohort, we selected a total of 40 MS donors (n = 21 ERDs; n = 19 PRDs), with multiple samples per donor, to analyse the gene expression profile using next generation RNA sequencing, and its association with remyelination capacity. In this selected cohort (n = 40), the average proportions of RLs in ERD and PRD groups were 0.55 ± 0.20 and 0.12 ± 0.08, respectively (P < 0.0001; Table 1), representative of the entire NBB cohort. Both the AL proportion and AL non-foamy proportion were significantly higher in ERDs compared with PRDs (ERDs: 0.52; PRDs: 0.39; P = 0.001; and ERDs: 0.41, PRDs: 0.23; P = 0.002, respectively; Fig. 1C), which is in line with the finding for the larger MS cohort. A lower total lesion load (P = 0.002) was also observed in ERDs in comparison to PRDs (Table 1), in agreement with a higher remyelination capacity in this group. No significant differences were found between either donor or lesion groups for age, post-mortem delay, sex, pH of CSF, brain weight, age at onset, disease duration, and years to expanded disability status scale 6 (EDSS6) (Table 1). In summary, the selected MS tissue and donor cohort for this study (n = 40) is pathologically and clinically representative of a large MS autopsy cohort (n = 239). The selected cohort therefore represents a powerful resource for the study of the molecular underpinnings of remyelination heterogeneity in MS.

Table 1

Demographic characteristics and lesion proportions for selected donor and lesion groups

Parameter	All donors (N = 40)	ERDs (N = 21)	PRDs (N = 19)	P-value	ERDs			PRDs			P-value
					RL (n = 13)	AL non-foamy (n = 6)	AL foamy (n = 7)	RL (n = 7)	AL non-foamy (n = 7)	AL foamy (n = 6)
Disease subtype	6 R-MS 12 PPMS 18 SPMS 4 unknown	6 R-MS 5 PPMS 9 SPMS 1 unknown	0 R-MS 7 PPMS 9 SPMS 3 unknown		6 R-MS 3 PPMS 4 SPMS	1 R-MS 1 PPMS 3 SPMS 1 unknown	2 R-MS 1 PPMS 4 SPMS	0 R-MS 2 PPMS 3 SPMS 2 unknown	0 R-MS 4 PPMS 2 SPMS 1 unknown	0 R-MS 2 PPMS 4 SPMS
Number of tissue blocks	45	25	20		13	6	7	7	7	6
Number of lesions	46	26	20		13	6	7	7	7	6
Number of NAWM	45	25	20		13	5	7	7	7	6
RL proportion	0.35 (0.26)	0.55 (0.20)	0.12(0.08)	P<0.0001a	0.59 (0.18)	0.54 (0.29)	0.54 (0.16)	0.17 (0.06)	0.12 (0.07)	0.10 (0.09)
RIN values
All	6.2 (1.5)	6.4 (1.4)	6.0 (1.7)		6.2 (1.4)	6.3 (1.3)	7.3 (1.0)	5.3 (1.4)	7.1 (1.3)	5.5 (1.8)	0.001c
NAWM	6.6 (1.2)	6.7 (1.2)	6.4 (1.3)	0.31c	6.4 (1.2)	6.5 (1.1)	7.7 (0.5)	5.8 (1.2)	7.0 (0.7)	6.1 (1.8)
Lesion	6.0 (1.7)	6.1 (1.6)	5.7 (2.0)	0.65c	5.9 (1.6)	6.2 (1.5)	6.9 (1.2)	4.7 (1.4)	7.2 (1.7)	4.8 (1.8)
Age (years)	62.4 (12.1)	65.7 (13.0)	58.8 (10.2)	0.07c	66.2 (13.9)	69.2 (3.1)	64.4 (15.3)	63.1 (10.2)	56.0 (8.9)	57.5 (10.8)	0.22c
Post-mortem delay (min)	517.5 (113.0)	527.9 (108.6)	506.1 (119.6)	0.72c	489.6 (96.3)	600.0 (114.8)	514.3 (126.7)	553.6 (109.3)	478.6 (125.8)	507.5 (135.7)	0.45c
Sex, female (%)	57.5	52.4	63.2	0.71b	61.5	50	28.6	57.1	57.1	83.3	0.52b
pH of CSF	6.5 (0.3)	6.4 (0.2)	6.5 (0.4)	0.31c	6.4 (0.3)	6.3 (0.2)	6.4 (0.2)	6.7 (0.5)	6.5 (0.2)	6.2 (0.3)	0.14c
Brain weight (g)	1192.2 (119.9)	1212.3 (128.4)	1168.8 (107.9)	0.25c	1167.2 (116.3)	1213.8 (156.0)	1226.3 (131.3)	1169.1 (118.6)	1203.0 (115.5)	1107.2 (49.3)	0.47c
Age at onset (years)	32.8 (10.0)	33.1 (10.2)	32.3 (10.0)	0.84c	33.1 (11.7)	34.7 (7.7)	33.9 (9.0)	37.4 (13.6)	31.1 (12.9)	35.7 (9.9)	0.88c
Disease duration (years)	29.6 (12.7)	32.3 (14.1)	26.5 (10.5)	0.27c	33.2 (12.9)	33.3 (6.7)	31.0 (17.2)	30.0 (13.4)	25.0 (8.9)	21.7 (9.4)	0.27c
Years to EDSS6 (years)	14.3 (10.5)	15.4 (10.0)	13.1 (11.1)	0.33c	16.8 (11.6)	16.6 (6.2)	12.7 (5.2)	18.5 (15.1)	11.3 (9.3)	9.3 (5.9)	0.43c
Lesion load (standardized)	14.5 (12.8)	9.1 (9.3)	20.7 (13.7)	0.002c	6.5 (8.4)	13.5 (10.9)	8.1 (6.5)	16.8 (7.9)	17.3 (11.0)	27.7 (18.5)	0.009c

Data are presented as the mean (standard deviation), unless otherwise indicated. Bold values indicate significant P-values.

AL (foamy) = active foamy lesion; AL (non-foamy) = active non-foamy lesion; EDSS6 = expanded disability status scale 6; ERDs = efficiently remyelinating donors; MS = multiple sclerosis; PPMS = primary progressive MS; PRDs = poorly remyelinating donors; RIN = RNA integrity number; RL = remyelinated lesion; R-MS = relapsing MS; SPMS = secondary progressive MS.

^aQuasi-binomial generalized linear model.

^bχ² test.

^cKruskal–Wallis test.

Comparable cell-type composition between efficiently and poorly remyelinating donors in the same lesion types

Bulk sequencing of mRNA isolated from ERD and PRD samples was performed. Principal component analysis revealed clear segregation of samples dependent on sex after omission of Y, XIST and TSIX sex chromosome genes (Supplementary Fig. 1). That transcriptomic differences were mainly driven by sex chromosome genes was also shown in a previous study in human brain samples⁴² (Fig. 2A). Therefore, comparison of gene expression profiles between male and female MS donors was not explored further.

Figure 2

Transcriptional differences between multiple sclerosis donors with efficient or poor remyelination potential. (A) Principal component analysis on the distribution of all included samples (n = 91). (B) Cell composition analysis (estimated from the single-nucleus RNA sequencing study by Schirmer et al.⁴³) shows that most cell composition differences can be detected between multiple sclerosis (MS) lesions and normal-appearing white matter (NAWM). (C) Venn diagram shows no or few differentially expressed genes (DEGs) between efficiently remyelinating donors (ERDs) and poorly remyelinating donors (PRDs) in NAWM, remyelinated lesion (RL) and active non-foamy lesion (AL non-foamy) samples but many DEGs in active foamy lesion samples (AL foamy; false discovery rate < 0.05). (D) Gene set enrichment analysis of hallmark gene sets shows regulation of specific pathways in RLs, ALs non-foamy and ALs foamy between ERDs and PRDs. Epithelial–mesenchymal transition (EMT) is enhanced in RLs and ALs non-foamy of ERDs. On the contrary, inflammation- and damage-associated pathways are upregulated in ALs foamy of PRDs. Symbols indicate upregulation in ERDs (triangle) or PRDs (circle). (E) Identification of novel soluble potential pro-remyelinating factors with comparison of fold change expression between donor groups in remyelinated and active non-foamy lesions. Green area indicates the area of interest for molecule selection. Genes below the lower line indicate genes with 2-fold upregulation in ERDs compared with PRDs.

Cell-type composition was estimated by using specific cell-type markers for astrocytes, endothelial cells, microglia, oligodendrocytes, OPCs and stromal cells, using the single nucleus RNA (snRNA) sequencing dataset of Schirmer et al.⁴³ (Fig. 2B). For both donor groups, differences in cell-type composition were found mainly between MS lesions and NAWM, with significantly more astrocytes, endothelial cells and stromal cells in MS lesions compared with NAWM (Supplementary Table 3). However, no differences in cell-type composition were found in the same MS lesion types between ERDs and PRDs. Given that no significant differences in cell composition were found between the MS donor groups, deconvolution correction was not applied to our gene expression analysis.

Major transcriptional changes in active foamy lesions of efficiently versus poorly remyelinating donors

When directly comparing gene expression profiles of ERDs and PRDs, we found that the NAWM did not show differences between the donor groups (Fig. 2C). Only few differentially expressed genes (DEGs) were observed in RLs (17 DEGs) and ALs non-foamy (8 DEGs) between ERDs and PRDs (FDR < 0.05) (Fig. 2 and Supplementary Fig. 2). The DEGs detected in ALs non-foamy, such as SNORD3A, SEPTIN7P6, ROGDI, MT-ND4L, MTATP6P1, ATP5ME and SCARNA7, were all upregulated in PRDs. Interestingly, we found a more distinct difference between MS donors with different remyelinating capability in ALs foamy (502 DEGs). Many DEGs that were detected in ALs foamy between ERDs and PRDs were predominantly upregulated in ERDs, including SNORA37, ARF1P2, MKRN5P, CNPY1, PAOX and STRADA. The most significant DEGs upregulated in ALs foamy of PRDs included SNX3, IFIT1, IFIT3, MOCS2, EIF2AK2 and IFIH1.

Efficiently and poorly remyelinating donors show differences in gene sets within the same MS lesion types

Gene set enrichment analysis was performed to assess potential differences in biological processes within RLs, ALs non-foamy, ALs foamy and NAWM of the two donor groups (Fig. 2D). Gene set enrichment analysis revealed one hallmark gene set upregulated in RLs and ALs non-foamy of ERDs related to ‘epithelial–mesenchymal transition (EMT)’ (Fig. 2D). Of the 200 genes included in this hallmark gene set, 190 genes were detected in our RNA sequencing dataset. We identified 59 genes of interest (significantly expressed with FDR < 0.05, fold change >2 in RLs and/or ALs non-foamy versus NAWM of ERDs), including CAP2, CAPG, CCN1, CRLF1, COL1A1, DAB2, VCAM, PTHLH, PCOLCE, CXCL12, TGFB1 and TGFBR3 (Supplementary Table 4). This suggests a potential role for EMT in promoting efficient remyelination in the regenerative lesions of donors with high remyelinating capability. In ALs foamy, many hallmark gene sets were upregulated in PRDs compared with ERDs, in pathways associated with inflammation (‘TNFa signalling via NF-kB’, ‘interferon gamma response’, ‘inflammatory response’, ‘complement’ and ‘interferon alpha response’), cell death (‘apoptosis’) and tissue damage (‘DNA repair’). This suggests that these inflammation- and tissue damage-associated pathways might hinder remyelination in less regenerative lesions of donors with low remyelinating capability. Notably, we did not detect differences in hallmark gene sets in NAWM of ERDs and PRDs. Together, we demonstrate prominent differences in gene set enrichment between ERDs and PRDs within the same MS lesion types.

Known remyelination-associated genes are enriched in donors and lesions with efficient remyelination

To validate our novel approach of donor and lesion selection based on remyelination capacity, we investigated whether genes that have previously been associated with remyelination in the literature were indeed enriched in lesions of ERDs in comparison to PRDs. Of the 109 remyelination-associated genes identified in our literature search, 99 genes were present in our RNA sequencing dataset. A total of 49 unique genes were significantly expressed in any donor or lesion group. The 45 significantly expressed genes detected in RLs and the 21 significantly expressed genes detected in ALs non-foamy were all expressed in ERDs only. Genes of particular interest (significantly expressed with a fold change >2 in RLs and/or ALs non-foamy of ERDs) were CNR1, CNTFR, CXCL12, CXCR4, EGF, ERBB2, FABP7, FGF2, FGF10, HGF, IGF2, IGFBP6, IL10, IL10RA, NRP1, PDGFB, PPARG, SEMA3A, TGFB1, TGFB2, TGFBR3 and TREM2 (Table 2 and Supplementary Table 5). For a number of these genes of interest, both ligands and interacting receptors were upregulated, additionally pointing towards a role in the promotion of remyelination: CXCL12–CXCR4, IL10–IL10RA, SEMA3A–NRP1, TGFB1–TGFBR3 and TGFB2–TGFBR3. Thus, as expected, we identified genes previously associated with remyelination in RLs and ALs non-foamy of ERDs, confirming the value of our dataset for the identification of novel pro-remyelinating molecules.

Table 2

Fold changes of genes previously associated with remyelination enriched in remyelinated and active non-foamy lesions of efficiently remyelinating multiple sclerosis donors

Gene	Protein class	Fold change		Fold change ratio	Fold change
		ERDs, RL versus NAWM	PRDs, RL versus NAWM	ERDs/PRDs, RL versus NAWM	ERDs, AL (non-foamy) versus NAWM	PRDs, AL (non-foamy) versus NAWM	ERDs, AL (foamy) versus NAWM	PRDs, AL (foamy) versus NAWM
Known remyelination-associated factors
CNR1	G-protein-coupled receptor	2.14a	2.40	0.89	2.35	1.29	2.06	1.54
CNTFR	Transmembrane signal receptor	1.97	1.41	1.40a	2.37a	1.02	1.27	1.36
CXCL12	Chemokine	3.18a	2.67	1.19a	2.15a	2.21	2.38	2.06
CXCR4	G-protein-coupled receptor	3.36a	2.31	1.45a	2.69a	2.51	4.29	2.95
EGF	Growth factor	3.82a	1.76	2.17a	3.72a	2.08	1.98	1.37
ERBB2	Transmembrane signal receptor	2.69a	2.14	1.26a	2.38a	1.56	2.19	2.05
FABP7	Transfer/carrier protein	2.47a	1.55	1.60a	1.33	1.72	1.69	1.68
FGF2	Growth factor	2.32a	1.28	1.81a	1.32	1.70	1.78	1.83
FGF10	Growth factor	6.08a	1.90	3.21a	4.12a	3.55	2.78	2.93
HGF	Growth factor	4.33a	2.03	2.13a	3.14a	3.63	4.44	3.01
IGF2	Growth factor	4.69a	1.80	2.60a	0.40	2.28	4.71	2.76
IGFBP6	Protease inhibitor	1.90	0.89	2.14a	3.81a	0.88	1.22	0.86
IL10	Interleukin superfamily	3.54a	3.55	1.00	2.06a	2.02	4.18	3.79
IL10RA	Transmembrane signal receptor	2.80a	2.28	1.23a	2.11a	2.11	2.51	2.36
NRP1	Interleukin superfamily	3.54a	3.55	1.00	2.06a	2.02	4.18	3.79
PDGFB	Growth factor	1.86	2.70	0.69	2.34a	1.72	1.53	1.52
PPARG	C4 zinc finger nuclear receptor	2.63a	1.19	2.20a	2.28a	1.86	2.85	2.66
SEMA3A	Membrane-bound signalling molecule	2.17a	2.01	1.08a	5.10a	1.65	1.30	1.72
TGFB1	Growth factor	2.31a	2.17	1.07a	1.47	1.85	2.22	3.03
TGFB2	Growth factor	2.55a	1.48	1.72a	2.23a	2.12	2.48	1.91
TGFBR3	Serine/threonine protein kinase receptor	2.13a	1.88	1.13a	1.91	1.57	1.86	1.90
TREM2	Immunoglobulin receptor superfamily	3.35a	2.46	1.36a	1.92	2.34	2.23	2.36

Bold values indicate adjusted P-values <0.05.

AL (foamy) = active foamy lesion; AL (non-foamy) = active non-foamy lesion; ERDs = efficiently remyelinating donors; NAWM = normal-appearing white matter; PRDs = poorly remyelinating donors; RL = remyelinated lesion.

^aA >2.0 upregulation in RL versus NAWM or AL non-foamy versus NAWM of ERDs, or ERD/PRD fold change ratio >1.0.

Identification of novel molecules highly enriched in donors and lesions with efficient remyelination

To identify novel potential remyelination-promoting molecules in the CNS, we undertook a different approach by comparing the fold change expression between donor groups, as visualized by quadrant plots (Fig. 2E). We focused on soluble factors, because these are more easily targetable for therapeutic approaches, and thus analysed growth factors, peptide hormones, chemokines and cytokines that were twice as highly upregulated in ERDs versus PRDs or were upregulated in ERDs but downregulated in PRDs, in RLs and ALs non-foamy (threshold of log₂ fold change > 1.5). Based on these criteria, we identified the following genes: ADM2, BMP3, CCN4, CXCL9, BTC and CALCB (in both RLs and ALs non-foamy), CCL20, PTHLH, FGF10, CXCL10, IGF2 and CCL19 (in RLs only), and CXCL3, AREG, INHBC, LIF, FGF5, GDF15, INHA, GDF10 and GHRL (in ALs non-foamy only) (Table 3).

Table 3

Fold changes of genes of interest in search for novel soluble factors potentially associated with remyelination

Gene	Protein class	Fold change		Fold change ratio	Fold change
		ERDs, RL versus NAWM	PRDs, RL versus NAWM	ERDs/PRDs, RL versus NAWM	ERDs, AL (non- foamy) versus NAWM	PRDs, AL (non- foamy) versus NAWM	ERDs, AL (foamy) versus NAWM	PRDs, AL (foamy) versus NAWM
Growth factors from WGCNA (MEturqoise)
BMP3	Growth factor	6.12a	2.16	2.83a	4.59a	1.10	1.71	1.20
BMP4	Growth factor	1.86	1.76	1.06a	2.28a	0.61	1.27	1.18
BTC	Growth factor	3.20a	0.87	3.69a	5.57a	2.04	2.78	10.18
CCN1	Growth factor	2.84a	1.77	1.61a	2.32a	2.53	2.53	1.21
CCN4	Growth factor	3.53a	1.13	3.12a	3.14a	1.12	2.28	1.20
EPGN	Growth factor	0.35	1.27	0.28	1.21	0.87	1.69	0.58
FGF10	Growth factor	6.08a	1.90	3.21a	4.12a	3.55	2.78	2.93
GDF10	Growth factor	2.26a	2.31	0.98	2.91a	0.70	0.83	0.74
IGF2	Growth factor	4.69a	1.80	2.60a	0.40	2.28	4.71	2.76
INHA	Growth factor	3.51a	2.17	1.62a	8.41a	0.76	2.27	4.14
INHBB	Growth factor	2.37a	1.49	1.60a	2.91a	1.79	2.33	8.13
INHBC	Growth factor	1.80	0.92	1.95a	3.53a	1.07	3.26	4.39
Soluble factors from quadrant plots
ADM2	Peptide hormone	3.74a	1.78	2.11a	3.54a	0.83	1.73	6.44
AREG	Growth factor	1.36	1.03	1.32a	6.09a	1.96	1.84	2.22
BMP3	Growth factor	6.12a	2.16	2.83a	4.59a	1.10	1.71	1.20
BTC	Growth factor	3.20a	0.87	3.69a	5.57a	2.04	2.78	10.18
CALCB	Peptide hormone	4.08a	1.04	3.91a	7.03a	1.35	2.24	1.47
CCL19	Cytokine	14.28a	1.16	12.29a	2.48a	2.68	5.70	9.09
CCL20	Cytokine	3.71a	0.98	3.79a	2.35a	2.56	3.65	1.60
CCN4	Growth factor	3.53a	1.13	3.12a	3.14a	1.12	2.28	1.20
CXCL10	Chemokine	3.76a	0.89	4.24a	1.02	1.10	2.73	9.21
CXCL3	Chemokine	2.02a	1.00	2.02a	4.80a	1.45	2.96	2.10
CXCL9	Chemokine	3.75a	1.22	3.08a	8.99a	1.21	6.05	3.47
FGF5	Growth factor	0.67	0.79	0.84	3.55a	1.47	1.54	1.89
FGF10	Growth factor	6.08a	1.90	3.21a	4.12a	3.55	2.78	2.93
GDF10	Growth factor	2.26a	2.31	0.98	2.91a	0.70	0.83	0.74
GDF15	Growth factor	0.91	1.62	0.57	3.16a	0.93	2.08	3.53
GHRL	Peptide hormone	2.43a	1.75	1.39a	3.23a	0.91	1.42	0.98
IGF2	Growth factor	4.69a	1.80	2.60a	0.40	2.28	4.71	2.76
INHA	Growth factor	3.51a	2.17	1.62a	8.41a	0.76	2.27	4.14
INHBC	Growth factor	1.80	0.92	1.95a	3.53a	1.07	3.26	4.39
LIF	Cytokine	0.95	2.09	0.46	3.81a	1.28	2.76	4.99
PTHLH	Peptide hormone	2.98a	1.42	2.10a	3.68a	2.10	2.45	2.12

Gene names shown in bold indicate genes identified in both the weighted gene co-expression network analysis (WGCNA) and quadrant plot analysis. Bold values indicate adjusted P-values <0.05.

AL (foamy) = active foamy lesion; AL (non-foamy) = active non-foamy lesion; ERDs = efficiently remyelinating donors; NAWM = normal-appearing white matter; PRDs = poorly remyelinating donors; RL = remyelinated lesion.

^aA >2.0 upregulation in RL versus NAWM or AL non-foamy versus NAWM of ERDs, or ERD/PRD fold change ratio >1.0.

Unique co-expression networks associated with efficient or poor remyelination potential

Next, we investigated molecular pathways associated with efficient remyelination by exploring the association of gene sets (modules) with MS tissue/lesion type and with the MS donor RL proportion, using weighted gene co-expression network analysis (WGCNA) (Fig. 3). We identified a total of 12 gene modules. We found that modules were more correlated with the MS tissue/lesion types, rather than with the RL proportion of MS donors. One module, MEturquoise, was associated with regenerative MS lesion types: RLs (FDR = 8 × 10⁻⁵, R = 0.44) and ALs non-foamy (FDR = 0.01, R = 0.32). This module had GO terms relating to ‘microtubule-based movement’, ‘axonal assembly’ and ‘extracellular transport’. From a total of 1757 genes in MEturquoise, the hub genes (i.e. genes with the highest correlation with the module eigengene) included DRC7, ODAD2, CFAP52, LRRC43 and DNAH5 (Supplementary Table 6). Modules enriched in the less regenerative MS lesion type, ALs foamy, were MEgreenyellow (FDR = 0.004, R = 0.36), MEpink (FDR = 0.001, R = 0.39) and MEblue (FDR = 1 × 10⁻⁷, R = 0.55). These had GO terms associated with ‘ATP metabolic process’, ‘inflammatory response’, ‘foam cell differentiation’, ‘leucocyte mediated immunity’ and ‘IFNγ production’, respectively, reminiscent of the gene set enrichment analysis pathways.

Figure 3

WGCNA for multiple sclerosis (MS) normal-appearing white matter, MS lesion types and donors with different remyelinating capability. WGCNA identified 12 modules. The numbers of genes in the modules are displayed in parentheses. Correlation and significance are shown for gene traits with false discovery rate < 0.05. Gene Ontology (GO) processes of modules of interest are highlighted. The most significant upstream regulating molecules are indicated in blue, and genes regulating most other genes in the module are indicated in red. AL (foamy) = active foamy lesion; AL (non-foamy) = active non-foamy lesion; NAWM = normal-appearing white matter; RL = remyelinated lesion; WGCNA = weighted gene co-expression network analysis.

Given that MEturquoise hub genes showed more significant differential gene expression in RLs and ALs non-foamy in ERDs compared with PRDs (data not shown), we focused on this module. Assessment of secreted molecules within MEturquoise revealed 12 growth factors within the module: INHBB, BTC, BMP3, INHA, FGF10, INHBC, GDF10, IGF2, CCN1, CCN4, BMP4 and EPGN (Table 2). Ingenuity pathway analyses additionally revealed ATP8BI, RFX3, ISL1, TGFB2 and FOXA2 as the most significant predicted upstream regulating genes of MEturquoise and TGFB1, AGT, IGF1 and EGF as predicted upstream genes regulating most other genes within the turquoise module (Fig. 3). This implicates ATP8BI, RFX3, ISL1, TGFB2, FOXA2, TGFB1, AGT, IGF1 and EGF as key potential molecules involved in remyelination.

Cellular protein expressions of TGFβ1, TGFβ2 and EGF are in line with gene expression in MS pathology

Given that TGFβ1, TGFβ2 and EGF were identified as key predicted upstream regulators of remyelinating lesions, and a significant differential expression was found for these genes in RLs and/or ALs non-foamy of ERDs (TGFB1, RLs: fold change = 2.31; TGFB2, RLs: fold change = 2.55; TGFB2, ALs non-foamy: fold change = 2.23; EGF, RLs: fold change = 3.82; EGF, ALs non-foamy: fold change = 3.72), we performed immunofluorescence to investigate their cellular expression (Fig. 4A–C). The cellular expression of BTC, which we identified as a novel soluble factor to potentially promote remyelination (BTC, RLs: fold change = 3.20; ALs non-foamy: fold change = 5.57), was also investigated (Fig. 4D). To assess potential ongoing remyelination, we examined these factors in ALs non-foamy of ERDs. Additionally, we assessed the cellular gene expression of the target molecules and of their interacting receptors, using an snRNA sequencing dataset of MS lesions⁴⁴ (Fig. 4E).

Figure 4

Cellular expression of genes of interest in human multiple sclerosis brain tissue. (A–D) Immunofluorescent double-stained images of TGFβ1 (A), TGFβ2 (B), EGF (C) and BTC (D) with GFAP (astrocytes), HLA (microglia/macrophages) and SOX10 or Nogo-A (oligodendrocytes) in an active non-foamy lesion. Scale bars: 15 μm. Arrows indicate target⁺/cell marker⁺ double-staining, arrowhead indicates target⁺/cell marker⁻ double-staining, and asterisks indicate target⁻/cell marker⁺ double-staining. (E) Percentage of cells per cell type expressing selected ligand and receptor pairs and average expression of each gene in each cell type (adapted from Absinta et al.⁴⁴). OPC = oligodendrocyte precursor cell.

We found that TGFβ1 mainly co-localized with HLA⁺ microglia/macrophages, whereas most TGFβ1⁺ cells were negative for GFAP⁺ astrocytes and SOX10⁺ oligodendrocytes (Fig. 4A). This finding is consistent with MS snRNA sequencing data⁴⁴ showing that TGFB1 is predominantly expressed by immune cells, including microglia/macrophages (Fig. 4E), and with gene expression data of isolated human⁴⁵ and mouse⁴⁶ microglia in control conditions. For TGFβ2, we identified co-localization with all three cell-type markers (Fig. 4B). In line with the snRNA sequencing data,⁴⁴ the majority of TGFβ2⁺ cells expressed the astrocytic marker GFAP (Fig. 4E). Notably, in NAWM, TGFβ2⁺ cells were detected only in astrocytes, and not in microglia/macrophages or oligodendrocytes. Most EGF⁺ cells were also GFAP⁺ and mainly negative for HLA and SOX10 (Fig. 4C), which is also in agreement with the snRNA sequencing dataset⁴⁴ (Fig. 4E). Lastly, we assessed the cellular localization of BTC and found a high number of BTC⁺ HLA⁺ cells. Furthermore, we found a number of BTC⁺/HLA⁻ cells around blood vessels, which we identified as endothelial cells using Ulex europaeus I agglutinin (data not shown). BTC⁺ cells were predominantly negative for GFAP and Nogo-A⁺ (mature) oligodendrocytes (Fig. 4D), which was unexpected according to the snRNA sequencing dataset.⁴⁴ Receptors for EGF family members EGFR, ERBB3 and ERBB4 are highly expressed by OPCs and/or oligodendrocytes, whereas receptors for TGFβ family members TGFBR1 and TGFBR2 are highly expressed by microglia/macrophages, astrocytes and/or endothelial cells (Fig. 4E). Together with previous studies, these data show that the cellular protein expression pattern of TGFβ1, TGFβ2 and EGF proteins (Fig. 4A–C) is in line with their gene expression in MS pathology (Fig. 4E). Additionally, we found a high cellular expression of BTC protein in microglia/macrophages (Fig. 4D).

TGFβ1 expression is positively correlated with presence of the remyelination marker BCAS1 in MS

Next, we quantified the presence of TGFβ1⁻, TGFβ2⁻ and EGF⁺ cells per millimetre squared, and the percentage of BTC⁺ area in relationship to remyelination, as identified by BCAS1-immunoreactive oligodendrocytes^47,48 in RLs, ALs non-foamy, ALs foamy and NAWM (Fig. 5). TGFβ2, EGF and BTC were elevated in RLs and/or ALs non-foamy compared with NAWM (TGFβ2, RL: P = 0.023; BTC, RL: P < 0.001, ALs non-foamy: P < 0.001; EGF, ALs non-foamy: P = 0.004; Fig. 5A and B). Although not significant, we found a trend towards more TGFβ1 in ALs non-foamy compared with NAWM (P = 0.065) and ALs foamy (P = 0.093) (Fig. 5A and B). Interestingly, TGFβ2, EGF and BTC were increased further in ALs foamy compared with NAWM, and for EGF, also compared with RLs and ALs non-foamy.

Figure 5

Association between factors of interest with remyelinating multiple sclerosis tissue. (A and B) Histological quantification of targets of interest, TGFβ1, TGFβ2, EGF and BTC, in remyelinated lesions (RLs), active non-foamy lesions (ALs non-foamy), active foamy lesions (ALs foamy) and normal-appearing white matter (NAWM). Scale bars: 100 μm. Arrows indicate TGFβ1⁺, TGFβ2⁺, EGF⁺ and BTC⁺ cells, respectively. (C) Immunohistochemical staining of BCAS1. (D–I) Heat map of quantified proteins of interest and BCAS1 in RLs, ALs non-foamy, ALs foamy and NAWM shows a positive correlation between TGFβ1 and BCAS1 and between EGF and BTC. Statistics were performed using negative binomial generalized linear model or restricted maximum likelihood with Tukey’s post hoc test to compare values between groups. *P < 0.05, **P < 0.01 and ***P < 0.001.

To determine the relationship of the target proteins with remyelination, we correlated the proteins of interest, TGFβ1, TGFβ2, EGF and BTC, with BCAS1⁺ oligodendrocytes (Fig. 5C–I). Numbers of BCAS1⁺ cells were not different between RLs, ALs non-foamy, ALs foamy and NAWM (data not shown). Nonetheless, we found a positive correlation between TGFβ1 and BCAS1⁺ cells (R = 0.31, P = 0.042; Fig. 5D and E), while a positive trend was observed between TGFβ2 and BCAS1⁺ cells (R = 0.27, P = 0.090). Expression of the two EGF family members, EGF and BTC, was not correlated with BCAS1⁺ cells (Fig. 5G and H), but was correlated with one another (R = 0.40, P = 0.015; Fig. 5I). In sum, we demonstrated higher levels of potential remyelination-promoting factors in RLs and/or ALs non-foamy, and in ALs foamy. Furthermore, we found that TGFβ1⁺ cells are positively correlated with numbers of BCAS1⁺ cells, suggesting a role in ongoing remyelination.

Discussion

The molecular and cellular factors contributing to heterogeneity in remyelination potential among individuals with MS and among different lesion types remain poorly understood. Here, we took advantage of a large well-characterized MS autopsy cohort (n = 239) by selecting donors with either high or low proportions of remyelinated lesions. From these donors, we selected both lesions which had already remyelinated (RLs) and lesions which were potentially likely to remyelinate (ALs),³⁹ allowing robust identification of pro-remyelinating molecules. We found many genes specifically upregulated in RLs and/or ALs non-foamy of ERDs that were previously shown to be associated with remyelination, such as CXCL12, EGF, HGF, IGF2, IL10, PDGFB, PPARG and TREM2, confirming the validity of our donor and lesion stratification. Importantly, we also identified novel genes, including BTC, AREG, GDF10, GDF15, CCN1, CCN4, FGF5, FGF10 and INHBB in RLs and/or ALs non-foamy of ERDs, that potentially promote remyelination.

Microglial activation states have been implicated in determining the fate of remyelination.^17,31 Indeed, we found a positive correlation between proportions of ALs non-foamy and proportions of RLs, and with a more favourable MS clinical outcome and a lower total lesion load. On the contrary, a negative correlation was found between proportions of ALs foamy and disease duration, and there was no relationship between proportions of ALs foamy and proportions of RLs. Therefore, ALs non-foamy, but not ALs foamy, were hypothesized to have higher remyelination potential. This is in line with previous findings that ALs are correlated more with RLs,³⁹ compared with MLs, and that foamy microglia/macrophages are correlated with acute axonal damage.³⁵ Accordingly, we show that active lesions containing foamy microglia/macrophages are less regenerative than those containing ramified/amoeboid cells. Our finding builds on previous work showing that microglia and their states are important in determining the success or failure of remyelination,^17,31 by now showing in human tissue, for the first time, that microglial morphology in MS lesions is an important criterion for remyelination potential.

Interestingly, ALs foamy exhibited the most DEGs between ERDs and PRDs in comparison to RLs, ALs non-foamy and NAWM. Exploration of biological processes underlying differences between ERDs and PRDs in these lesions showed enriched hallmark gene sets associated with inflammation and tissue damage in PRDs compared with ERDs. This could suggest that the predominance of inflammation- and damage-associated pathways might hinder efficient remyelination in ALs foamy of PRDs. Indeed, these ALs foamy were also enriched with remyelination-associated pathways, such as mTORC1 signalling,⁴⁹ further implicating a remyelination block at play.

In both ALs non-foamy and RLs, gene set enrichment analysis showed enriched ‘epithelial–mesenchymal transition (EMT)’ in ERDs compared with PRDs. EMT is a process best characterized in the context of cancer, where epithelial cells lose their polarity and adhesive contacts and dedifferentiate into mesenchymal stem cells (MSCs).⁵⁰ EMT has also been associated with wound healing and tissue regeneration.⁵¹ In the CNS, epithelial cells can be found in the choroid plexus,⁵² which might be a source for EMT detected in regenerative lesions of ERDs in our dataset. However, the cellular origin of EMT-related signals in the CNS remains unclear, owing to the rarity of epithelial cells in the brain. Other sources or processes might explain a gain of EMT marker expression. For instance, the upregulation of EMT markers might be a response to inflammatory signals, rather than mediating a direct role in the remyelination process. Hence, EMT-related changes attributable to inflammation and those genuinely associated with regenerative processes need to be distinguished. Nevertheless, MSC transplantation has been found to promote remyelination in animal models of MS,^53-56 and MSCs have been proposed as attractive therapeutic targets for demyelinating diseases owing to their immunomodulatory and self-renewal capacities.⁵⁷ A randomized controlled trial of MSC transplantation showed beneficial clinical effects in MS patients with a progressive disease course.⁵⁸ Despite this suggested therapeutic potential, translation to clinical practice is hampered by lack of knowledge regarding the exact mechanisms by which MSCs might promote remyelination, in addition to lack of data regarding its long-term effects. Targeting of MSCs to MS lesions appearing at random CNS locations is challenging. Additionally, safety concerns include putative local inflammatory responses to MSC-based therapies, in addition to tumour formation. Therefore, further investigation is required to understand the potential role of EMT in MS and in remyelination.

WGCNA identified one module (MEturquoise) that was specifically correlated with regenerative MS tissue, the RLs and ALs non-foamy. This module, characterized by gene ontology processes such as ‘microtubule-based movement’, points to a pivotal role of cytoskeletal reorganization in remyelination. Importantly, the top 25 molecules in this module were highly enriched in both RLs and ALs non-foamy of ERDs (e.g. DRC7, ODAD2, CFAP52, LRRC43 and DNAH5). We also identified TGFβ1, TGFβ2, EGF and IGF1 as predicted master upstream regulators of this remyelination-associated module. Studies have shown that TGFβ can induce EMT via cytoskeleton rearrangements (e.g. reorganization of the intermediate filaments, microtubules and actin microfilaments), making the cells more motile.⁵⁹ Actin filaments play a key role in process extension and axon ensheathment by oligodendrocytes.^60,61 Microtubules, another component of the oligodendrocyte cytoskeleton, play a role in survival, differentiation and myelination of oligodendrocytes.^62,63

Previous studies have already identified the involvement of TGFβ1, TGFβ2, EGF and IGF1 in remyelination. An RNA sequencing study of WM MS lesions also identified expression of TGFB1, TGFB2 and the receptor TGFBR3 in remyelinated and active lesions.⁶⁴ Here, we extend this finding by showing that enrichment of these molecules is found particularly in MS donors with efficient remyelination. We found that EGF was upregulated in RLs and ALs non-foamy of ERDs, but not PRDs. Functional studies of TGFβ1 have demonstrated its role in regulation of oligodendrocyte differentiation⁶⁵ and myelin integrity⁶⁶ and in driving remyelination in vivo.¹⁵ Additionally, TGFβ1 promotes homeostatic function of microglia.⁴⁶ Studies of TGFβ2 have implicated a role in immune regulation^67,68 and neuronal cell survival.⁶⁹ No studies have reported a role in remyelination for TGFβ2 before. Our findings suggest, for the first time, that TGFβ2 might have a role in remyelination.

Former studies of EGF have predominantly demonstrated beneficial effects of targeting the ligand or its receptor to promote remyelination through migration and proliferation of OPCs, oligodendrogenesis and oligodendrocyte maturation.^13,70,71 EGF treatment was also shown to promote functional recovery⁷² and prevent disease onset⁷³ in MS mouse models. EGF might also have an effect on the extracellular matrix and cell interactions through integrins, which play a role in myelin wrapping in the CNS.⁷¹

We also focused as a therapeutic approach on soluble factors that, by nature, are easy to deliver to the CNS. In the WGCNA and quadrant analyses, we found soluble factors related to the TGFB family (INHA, INHBB, INHBC, BMP3, BMP4, GDF10, GDF15, TGFB1 and TGFB2),⁷⁴  EGF family (BTC, AREG, EPGN and EGF),⁷⁵  CCN family (CCN1 and CCN4),⁷⁶  FGF family (FGF5 and FGF10)⁷⁷ and IGF family (IGF1 and IGF2).⁷⁸ BTC, a member of the EGF family, has been implicated in enhancing Schwann cell proliferation/migration, myelin formation by Schwann cells⁷⁹ and/or peripheral nerve regeneration.⁸⁰ However, less is known about its role in CNS regeneration. Thus, together with TGFβ1, TGFβ2 and EGF, we identified BTC as an interesting novel candidate to validate for a potential role in the promotion of remyelination.

Via immunohistochemistry, we found increased expression of the TGFβ2, EGF and BTC proteins in RLs and/or ALs non-foamy compared with NAWM, in line with the concept that remyelination is ongoing in these MS lesion types. However, quantification of the active remyelination marker BCAS1⁴⁷ did not show differences between RLs, ALs non-foamy, ALs foamy and NAWM. Recent literature has shown that BCAS1, besides being a marker for remyelinating oligodendrocytes, also marks degenerating glial cells, characterized by absent or very short and dystrophic-appearing processes, in diffusely abnormal WM,⁸¹ which could explain the discrepancy. TGFβ1⁺ cells did, however, show a positive correlation with the number of BCAS1⁺ cells, indicating a role in remyelination. This finding aligns with an animal study by Hamaguchi et al.,¹⁵ which demonstrated that TGFβ1 promotes remyelination in vivo. Notably, TGFβ2, EGF and BTC proteins were also enriched in ALs foamy. We hypothesize that the enrichment of remyelination-associated pathways and molecules might be the result of a compensatory mechanism to overcome pathways (i.e. inflammation- and damage-associated pathways) that inhibit efficient remyelination. Nevertheless, based on their enrichment in RLs and ALs non-foamy at both the gene and protein levels, we can conclude that the TGFβ and EGF family are promising molecules for the promotion of remyelination.

Previous studies have shown the beneficial effect of activin-A (INHBA), a close relative of activin-B (INHBB), on remyelination.^17,82-84 Although INHBA was not significantly expressed in our gene expression dataset, the gene encoding the receptor ACVR1 was significantly increased in RLs and ALs non-foamy of ERDs (1.25- and 1.40-fold change, respectively). Less is known about the regenerative potential of activin-B, although it has been shown to promote oligodendrocyte development and myelination in combination with TGFβ.⁸⁵ It would therefore be interesting to explore the remyelinating potential of activin-B further. Additionally, we detected molecules within the TGFβ family, such as INHA and BMP3, which inhibit remyelination by antagonizing activin-A.^86,87 In addition to promoting molecules that facilitate the activin pathway, targeting inhibitors within this pathway presents another potential avenue for therapeutic intervention. GDFs are other members of the TGFβ family that have a protective effect on axons by stimulating axonal sprouting/regeneration.^88-91 We identified GDF10 and GDF15 owing to >2-fold upregulation in ALs non-foamy of ERDs versus PRDs using quadrant plots. Studies of GDF10 and GDF15 have been related to repair in the peripheral nervous system or in stroke, not in MS. In MS, it has been shown that the serum concentration of GDF15 is increased in the context of a stable disease course.⁹² GDF10 and GDF15 should be studied further to determine the therapeutic potential of these molecules for promoting remyelination.

From the EGF family, we detected the novel targets BTC, AREG and EPGN. Studies on AREG have shown an immunomodulatory function for this molecule.^93,94 More recently, AREG was shown to improve the disease course in an animal model of MS.⁹⁵ In our study, we show, for the first time, the presence of AREG in regenerative lesions in MS, with enhanced gene expression in ERDs compared with PRDs. The EGFR, ERBB3 and ERBB4 receptors that bind to the EGF family members EGF, BTC and AREG are expressed by OPCs and/or oligodendrocytes. This might implicate a direct role in OPC differentiation or remyelination by oligodendrocytes.

In our data analyses, we detected FGF5 and FGF10 members of the FGF family. Previous studies have already shown the involvement of FGF/FGFR signalling in the remyelination process, being either beneficial (e.g. FGF1) or detrimental (e.g. FGF2), depending on the FGF subtype.^96-98 Previous research on FGF5 has shown a regulatory role in Schwann cell migration and adhesion,⁹⁹ and FGF10 has been linked to peripheral nerve regeneration and to inhibition of neuroinflammation and fibrosis.^100-102 Considering their established reparative functions in the peripheral nervous system, it would be interesting to investigate the potential roles of FGF5 and FGF10 in CNS remyelination.

CCN1 and CCN4 from the CCN family were highly expressed in remyelinated lesions of ERDs in our dataset. CCNs have been implicated in wound healing, inflammation and regulation of extracellular matrix differentiation.¹⁰³ Additionally, CCN molecules have a function in regulation of fibrosis.¹⁰⁴ Future research on CCN1 and CCN4 might elucidate their potential to promote remyelination in MS.

Overall, we have identified, for the first time, BTC, AREG, GDF10, GDF15, CCN1, CCN4, FGF5, FGF10 and INHBB as potential pro-remyelinating targets by stratifying MS donors and lesions based on remyelinating capacity. Further investigation of these promising molecules might yield future therapeutic approaches for the promotion of remyelination to limit disease progression in MS. Our dataset serves as a rich and valuable resource for the international MS community, available open access as a basis for future studies of the molecular mechanisms underpinning successful remyelination.

Limitations and strengths

A limitation of our study is that the use of post-mortem brain samples provides data at only one cross-sectional time point. Given that remyelination is a highly dynamic process, involving OPC recruitment, proliferation and differentiation, its temporality is not captured in our study. However, a strength of our study is the validation of target proteins in different MS lesion/tissue types, in a separate cohort from the RNA sequencing cohort.

Conclusion

Here, we studied the molecular underpinnings of remyelination in MS using stratification of MS brain donors and lesions with high or low remyelinating capabilities. This novel approach serves as an important starting point to elucidate cellular and molecular mechanisms underlying efficient remyelination in MS and holds the potential to uncover novel targets for therapeutic intervention in progressive MS in the future.

Data availability

The RNA sequencing dataset is available in the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov) under accession number GSE283092.

Acknowledgements

We are grateful to the MS patients who donated their brains to the Netherlands Brain Bank for research.

Funding

This study was supported by the Start2Cure Foundation (project 0-TI-01).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.