Corpus callosum and optic chiasm were the most significantly affected CNS regions in a study, and myelinating oligodendrocytes were most enriched with differentially expressed genes (51). Heat shock proteins were upregulated in all examined brain regions (HSPA1A, HSPA1B, HSPA7, HSPA6, HSPH1, HSPA4L) (52). Genes important in antigen-presentation, inflammation and hypoxia-induced responses were altered in the corpus callosum and optic chiasm (TAPBP, IRF4, CTSB, CD79A), while STAT6 and HLA-DRB5 were only increased in the optic chiasm. However, these regional differences may also reflect the presence of different cell types expressing different types of regional specific “housekeeping genes” with distinct physiological functions and purpose.

DNA methylation was altered, and RNA levels of DNA mehyltransferase were increased in MS hippocampus following demyelination (40). This study identified hypomethylation upstream of six genes including ANKA, a major regulator of CD40-CD40L, and hypermethylation upstream of ten genes e.g. WDR81, NHLH2, PLCH1 involved in neuronal survival, synaptic density and memory.

In the choroid plexus (CP), 17 genes were significantly upregulated in progressive MS patients (54). These genes were related to hypoxia, neuroprotection and secretion (e.g. CXCL2, LYVE1, SNHG15, MT1X, non-coding HIFA1-AS3), while strong inflammatory reactions were absent.

Comparing NAWM to control WM, 465 genes were differentially expressed (48). Among the top ten upregulated genes were immune-related (IGHG1, HLA-DRB5, GPNMB, CD163) and mitochondria-related (MTRNR2L12, MTRNR2L8). NAWM was also different from control WM by a global defense against oxidative stress based on upregulation of STAT6, HIFα and its target genes (21, 26, 61). Genes in the STAT-6 signaling were upregulated in oligodendrocytes (61) ( Figure 3 ). These alterations were accompanied by upregulation of nNOS, HO-1 and HLA-DR, suggesting an inflammatory and oxidative-stress related reaction in oligodendrocytes outside of lesions. A combined methylome and transcriptome study found downregulation and hypermethylation of oligodendrocyte survival genes in NAWM (BCL2L2, NBRG1) (46). Besides oligodendrocytes, several dysregulated genes in MS suggested alterations in subcortical WM neurons (21).

NAWM microglia upregulated STAT4 and HLA-DRα (26), and had a lipid metabolic gene expression profile (e.g. EEPD1, PPARG, LPL) with unchanged expression of the homeostatic signature (P2RY12 and TMEM119) (50). Additionally, a subtype of microglia (48) had increased expression of CD26/DPP4 in the NAWM (46). Astrocytic markers (GFAP, AQP4) were also altered in the NAWM (61) ( Figure 3 ).

Genes of several chemokines and cytokines (21, 26) were upregulated in NAWM reflecting the low level inflammation even without lesion formation. A mild disease course was also associated with a different molecular profile with altered expression of genes related to immune-regulation, myelination, anti-oxidative mechanism and neuroprotection together with a high hypothalamus-pituitary-adrenal (HPA) axis activity (35).

The difference in WM vs. GM microglia gene expression was significantly lower in MS compared to non-neurological disease brains. This suggests that microglia cells are losing region-specific profile in MS (50). However, while NAWM microglia have a lipid signature, NAGM microglia have increased expression of genes related to glycolysis and iron homeostasis (SCL25A37, ABCB6) and a neurodegenerative profile (CXCR4, GPNMB, OPN/SPP1) ( Figure 3 ). Furthermore, in HLADRB1*15:01 positive patients, HLA-DRB1 and B5 were the highest expressed genes in NAGM (37).

A continuum of dysfunctional homeostasis (e.g. VIM, HBB, MAF) and inflammatory changes (e.g. CASP1, IRF5, MMP2) between active lesions and NAWM supports the concept of MS involving the whole CNS (24). However, the lesions differed from NAWM by high expression of genes related to immunoglobulin synthesis (IGKC, IGL, IGGL1, ILR6) and neuroglial differentiation (SNAP25, CAP2, NFL/M) (24). Upregulated genes in active lesions compared to NAWM also included chemokine genes and receptors (MIP-a, RANTES, CCR1, CCR4, CCR5, VLA-4, CCR8) genes, interferon- and tumor-necrosis factor receptors (17), and cytokines (TGFB, IL-3, OPN, IL-5, IL6) (18, 44, 53) ( Figure 4 ). Two highly expressed genes encoded the Th cell marker (CD4) and the antigen-presenting gene (HLA-DRa) (18). Additionally, CD8+ T cells containing cytotoxic granules were suggested to communicate with mononuclear phagocyte cell expressing CD163 and CD11b in the lesions (49). Genes encoding multiple autoantigens were also found in MS lesions indicating a secondary autoimmune stimulation that could exacerbate the ongoing inflammation (43).

Mitochondrial injury in initial WM lesions was indicated by increase of ND1-6, CYTB, COX1, CYBA, MPO, PTGS1, PXDN, GPX4, PRDX1, SGK2, ALOX12, EPHX2 expression, which were related to degeneration of oligodendrocytes and neurons and contributed to reactive oxygen species production by activated microglia and macrophages (28) ( Figure 4 ).

Active and chronic active lesions shared upregulation of a number of genes coding for e.g. iron-binding protein (TF), chemokine and its receptor important for T cell accumulation in CNS (CXCR3, CXCL10), the myelin-binding protein (MBP), the first subcomponent of the complement system (C1QB), oxidative protection (GPX1, SOD1) and cytokines (IL-6, IL-17, INFg) (20, 23). However, 70 uniquely differentially expressed genes were also found: e.g. coding for the receptor related to differentiation (EPHB6), the granulocyte-macrophage colony-stimulating factor (GM-CSF), and a MHC class I molecule (HLA-A) in active lesions or e.g. genes coding for the chaperone protein (HSPA1A), component of MHC class I (B2M) or complement factor 4B (C4B) in chronic active lesions.

Differences have also been found on an epigenetic level, as the microRNA profile was different between active and inactive lesions (39). In the active lesions, microRNA-34a, -155 and -326 were all upregulated and targeted the CD47 in brain resident cells to release inhibitor control and promote phagocytosis ( Figure 4 ). Moreover, upregulated miR-22, miR-320 in active lesion and upregulated miR-30d in inactive lesions (39) were related to pathogenic changes (41), while downregulation of miR-18a, miR23b in inactive lesions (39) were related to protective changes correlating with MRI abnormalities (41).

An in-depth investigation of different lesion types (active, early remyelinating, chronic active, inactive) in the WM showed extreme diverse events at transcriptome level. More differential expressed genes were unique than shared. Among the 282 altered genes common to all lesion types were genes related to inflammation (STAT6, CXCL12, TNFs, DPP4/CD26, ITGA4, GPNMB, IL16, HLA-DRB5, MAFB, IGHG1, IGF2, MMP2), phagocytosis (SCD, CD163, MERTK) complement pathway (CFH, C7, CFI), apoptosis/necroptosis (FADS1, CASP1,-4, MLKL) (48). Immunoglobulin genes were among the top 10 in all WM MS tissues, but the most heterogeneous expression pattern was detected in early remyelinating lesions. TGFBR2 was the major molecular hub of the largest shared lesion network and was highly expressed in remyelinating lesions by astrocytes (48) ( Figure 4 ). The most different signatures were found between remyelinating and chronic active lesions. Chronic active lesions had the highest number of unique genes reflecting intrinsic neuronal alterations, and de novo networks suggested an end-stage exhaustion (48). Most of the uniquely expressed genes in the early remyelinating lesions were non-coding RNAs, while others were related to lymphocytes and NKT cells (e.g. CD8a, TIAM1, CTSW, CCL5/RANTES), growth and development (e.g. PEG10, BMP4, GDF10), vascular changes and remodeling (e.g. PLAU, VEGFA, CTGF), mitochondria and protective stress responses (e.g. NDUFA4, NOSTRIN), lipid metabolism (e.g. ACACA, ACOX2, ADH6, CA3), and neurons (e.g. NEUROD1, NLGN1, GRIA3) ( Figure 5 ). Another study found CXCL12, SCD, STAT6 increased in all lesion types, and transcriptional differences between lesion types reflected a heterogeneous oligodendrogliogenesis (34). Quantitative changes of oligodendrocyte regulators were also found in remyelinating lesions (30). Compared to demyelinating lesions, remyelination was accompanied by significant changes in the expression of myelin proteins (CNP, MAG, MBP, MOBP, MOG, OMG, PLP1), anti-inflammatory IL10, and semaphorins (SEMA3C, SEMA4D, SEMA6A, SEMA6D, SEMA7A) ( Figure 5 ). The growth factor gene FGF1 was significantly increased in remyelinating lesions compared to both control WM and demyelinating lesions. In functional experiments, FGF1 promoted both developmental myelination and remyelination by inducing LIF and CXCL8 in astrocytes to recruit oligodendrocytes. GFAP was also significantly increased in active and remyelinating lesions (55) ( Figures 4 , 5 ). The glia receptor protein tyrosine phosphatase gene PTPRG was increased in all MS WM tissues, and was also significantly increased in the CSF of MS patients compared to healthy and other neurological disease controls (55). CHI3L1 was increased in astrocytes in the chronic active lesion rim (55), and by microglia in active lesions compared to NAWM (50).

In a single-nucleus study of WM lesions, the majority of cells were oligodendrocytes, and oligodendrocytes represented the most heterogenous cell population (56). One of the seven oligodendrocyte populations was termed immune oligodendroglia (imOLG) due to expression of APOE and CD74. OPCs were reduced in lesions and NAWM compared to control WM. One oligodendrocyte population was depleted, whereas three others and imOLG were enriched in MS. Several myelin protein genes were upregulated in mature oligodendrocytes in MS, however some of those (e.g. CNP, MAG) were downregulated in remyelinating lesions.

Excessive expression of the antioxidant transcription factor NRF2 in oligodendrocytes indicated oxidative stress and degeneration at sites of initial demyelination in active lesions (31). NRF2 in astrocytes and macrophages were mainly seen in the later stages of active lesions with profound loss of oligodendrocytes. NRF2 in neurons was low or absent despite NRF2-positve oligodendrocytes in close proximity indicating cellular differences in reaction to oxidative stress and inflammation ( Figure 4 ).

In chronic active lesions,14 genes were significantly upregulated in the rim vs the center (e.g. IFNG, NGF2, CD4, CASP9, MAPKK1) (22, 25, 29) ( Figure 6 ). Inflammatory genes were upregulated in chronic active lesion center (CCL4, IL6, CD27, TNFA) ( Figure 6 ), while upregulation of NCAM, CSF1, HSP60, HSP90A, BCL2L1 in inactive lesion center and rim highlighted different inflammatory responses, beside apoptosis and stress ( Figure 5 ). Heat shock protein genes in inactive lesions (48) and in the rim of chronic active lesions were upregulated, especially the heat shock factor 4 (HSF4) (29).

Upregulation of 165 genes and downregulation of 35 genes were identified in the chronic active lesion/slowly expanding lesions compared to inactive as well as NAWM (38). The upregulated genes suggested accumulation of microglia with proinflammatory differentiation at the lesion edge (e.g. CD163, CD68, CSF3R, IGFBP5, ALOX15B, MME, TNFRSF25) ( Figure 6 ). A study that investigated the rim and peri-lesional regions of both chronic active and inactive lesions, found upregulation of previously not reported genes in the rim of chronic active lesions (NPY, KANK4, NCAN, TKTL1, ANO4) (33) ( Figure 6 ). They also found that foamy macrophages in the rim upregulated genes involved in lipid binding and uptake indicating the expansion of demyelination (e.g. MSR1, CD68, CXCL16, OLR1, CHIT1, GPNMB all ( Figure 6 ). Stressed oligodendrocytes with iron overload, reactive astrocytes and activated phagocytosing cells were also detected in the rim of chronic active lesions (58). These findings were confirmed and elaborated in a recent snRNA-study, where they found immunological-active OPCs, inflamed astrocytes (AIMS) and microglia (MIMS) in the chronic active rim (60). These were strongly connected to a high number of T cells and plasma cells suggesting an active role of the adaptive immune system in lesion expansion in collaboration with the glia cells in the smoldering inflammatory lesions (60). Microglia consisted of two distinct functional subtypes: the MIMS-foamy characterized by myelin phagocytosis and clearance properties, and the MIMS-iron, characterized by expression of complement C1q-complex, antigen-presentation and direct propagation of inflammatory damage at the lesion edge. The inflamed astrocytes were enriched for response to lipid, corticosteroids, wounding and expression of C3 similar to the A1 phenotype identified in the GM (62).

A combined microRNA and mRNA profiling in GM lesions vs NAGM found significantly regulated microRNAs in GM lesions, which target genes of axonal guidance, TGFβ-signaling and FOXO signaling (63). Out of 27 significantly altered microRNAs, four microRNAs (mir149, mir20a, mir29c, mir25) and their targets (e.g. HIF1A, VEGFA, TGFBR1, TGFBR2, NFKBIB, FGFR1, TNFSF10, BCL2, MAP2K4, STAT3, MMP2, PTEN, CD44) were associated with GM atrophy ( Figure 7 ) (63). Three of the 27 significantly altered microRNAs were also detected in another GM lesional microarray study (mir181b, mir129-5p, mir1180) (64). Additionally, miR-330-3p, miR-4286, miR-4488, let-7e-5p and miR-432-5p shared the same mRNA target, the Syt7 gene coding for the neuroaxonal protein normally transported to synapses. These 5 microRNAs may be protective against Syt7 accumulation in the soma resulting in disturbed axonal transport.

TNF signaling was also significantly increased in GM lesions. Increased meningeal inflammation was associated with a shift from TNFR1/TNFR2 and NFkB-mediated anti-apoptotic pathways towards TNFR1- and RIPK3-mediated pro-apoptotic/pro-necroptotic signaling (36) ( Figure 7 ). TNFR1 was expressed by neurons and oligodendrocytes, while TNFR2 was predominantly expressed by astrocytes and microglia. The authors suggest that immune cells in meninges generate a milieu of increased demyelination and neurodegeneration by changing the balance of TNF signaling.

Another study found a selective loss of neurons expressing the transcription factor CUX2 in upper cortical layer lesions associated with pronounced meningeal B cell infiltration (58). These neurons expressed markers of cellular stress (PPIA, NORAD, PUMILIO, RBMX), and their loss may be a key event in MS progression and cortical atrophy ( Figure 7 ).

A study focused on endothelial cells in vessels found 52 genes significantly altered in chronic active or inactive lesions compared to control WM or NAWM (27). The majority of these genes belonged to endothelial cell activation, while VEGFA was the only one belonging to angiogenesis. Most of the genes were highly expressed in chronic active lesions compared to control WM (ANXA5, CSF3, FGF1,-2, FLT1,-4, ICAM1, MMP1, -2) ( Figure 6 ) and compared to NAWM (FGF2, FLT1,-3, MMP14, PLAU, RIPK1). Several endothelin genes (1,2,3,A) involved in constriction of blood vessels and supply were increased in inactive lesions compared to NAWM ( Figure 5 ).

Transcriptional profiling of isolated astrocytes in NAWM also revealed increased gene expression related to iron metabolism, oxidative stress, and inflammatory response (32) ( Figure 3 ). An astrocyte single-cell study identified an expanded astrocyte population in active lesions characterized by decreased NRF2 and increased MAFG, GM-CSF signaling, pro-inflammatory pathway activity and DNA methylation (DNMT1) (59) ( Figure 4 ). This astrocyte population is characterized by a MAFG/MAT2α-driven pro-inflammatory genomic program contributing to the pathology and may be induced by GM-CSF produced by infiltrating T cells ( Figure 4 ). This corresponds to the high GM-CSF in active lesions (23), and low NRF2 in astrocytes in initial demyelinating lesions (31).

Seven microglia cell populations expressing the core microglial genes (TMEM119, P2RY12) in the WM were discovered in a single-cell study (57). Two of these clusters were enriched in brains of MS patients and one was associated with MS. These three populations had increased levels of APOE and MAFB ( Figures 4 , 8 ). The MS-associated microglia subset highly expressed CTSD, APOC1, GPNMB, ANAX2, LGALS1, while the two MS-enriched clusters showed high expression of either CD74, HLA-DRA, HLA-DRB1 or OPN/SPP1, PADI2, LPL ( Figure 8 ). These findings suggest distinct disease-related subtypes of microglia in the MS brain, which were similar to microglia subtypes in a demyelination model. However, subsets of microglia varied substantially between individual patients indicating high inter-individual heterogeneity. Additionally, the different microglia populations appeared as a transcriptional continuum of the local populations, which could reflect the ability of microglia to easily adapt to changes in the surroundings.

Methylome changes within neuronal nuclei in WM suggested alterations in axonal guidance, synaptic plasticity and CREB signaling in MS (42). The CREB activity was reduced in NAWM compared to WM neurons suggesting alteration of CREB signaling prior focal tissue damage ( Figure 3 ). Neurons from MS patients displayed epigenetic alterations affecting several genes of the glutamate/GABA signaling along with interconnected cellular networks (semaphoring/plexin, Slit/ROBO, Shh/Wnt signaling). Lesion-associated changes in genes implicated neuronal projections and synaptic processes (e.g. GABRA5, PRKG1, DLGAP3/SAPAP3) (42) ( Figure 4 ).

Amplicon next-generation sequencing of the human endogenous retrovirus (HERV)-W group found very similar transcript level between of WM lesions and control WM but evidence for interindividual differences in HERV-W transcript levels (45). In another study, genome-wide HERVs expression level between MS WM and control WM was not different (65). However, transcripts of HERV-W were reduced in chronic active and repairing lesions. Additionally, four different transcripts of HERV-W on chromosome 7 were only present in the MS brain (65). Another study found HERVs significantly overexpressed in demyelinating brain tissue including several retroviral domains (core, envelope, integrase, reverse transcriptase) (47). However, the overexpression was small. Due to multiple similar HERV transcripts incorporated and spread out throughout the human genome, examination of them is difficult.

Presence of microbial RNA sequences and bacterial antigens were associated with demyelinating brain lesions (66). In the study, they found 29 MS microbial candidate genera from 11 different phyla, most of which were anaerobic.

Based on these transcriptomics data, novel interactive online databases were generated. The MS Atlas (www.msatlas.dk), comprises comprehensive high-quality transcriptomic profiles of 98 different WM lesion types (53). The user-friendly MS Atlas was designed to provide information about significant expression of candidate genes and their participation in de novo protein-protein interaction networks in different MS lesions (53, 67). The OligoInternode database (https://ki.se/en/mbb/oligointernode), and the single cell expression atlas (https://www.ebi.ac.uk/gxa/sc/experiments/E-HCAD-35/results/tsne) give information about gene expression from single cells in MS lesions.

A proteome study found that 109 proteins could separate WM lesions from adjacent NAWM and control WM (70). Overlap was only observed between NAWM and WM lesions, but not between NAWM and control WM.

To characterize the MHC-bound peptide repertoire in MS brains, proteomics was performed on captured HLA-A, B, C, and DRs. 118 amino acid sequences from MHC I and 191 from MHC II were eluted corresponding to 174 identified proteins including both known and novel autoantigens (72). Some were involved in apoptosis (annexin A1, BCL2-associated TF1), enzymes (GDH, GS, G3PD, NADH dehydrogenase), cytoskeleton (actin, α-ubulin), immune responses (CXCR1, IL12R), CNS structure (NFL, GFAP, MBP, α-synuclein), and serum proteins/iron-related/coagulation (APOD, APOE, ferritin, transferrin, von Willebrand factor). These proteins within the MHC ligandome mirror the proteins involved in different features of the MS pathology.

Combined proteomics and genomics on two acute MS autopsied brain samples detected seven unique mutations of PLP1 (68). This was confirmed with in-depth genomic analysis on mRNA, but not in the genomic DNA, highlighting how results from integrative approaches can strengthen the discovery of specific and precise pathogenic mechanisms in MS.

Myeloid cells from active lesions, NAWM and WM in progressive MS (PMS) were analyzed using single-cell mass cytometry and found lower abundance of microglial homeostatic proteins in active lesions (P2Y12, TMEM119, CXC3R1, GPR56) (79). The myeloid cells in the active lesions were highly phagocytotic and activated indicated by upregulation of CD45, HLADR, CD44, CD114, CD11c, CD68, MS4A4A, CCR2, CD64, CD32, AXL, NFAT1, CD95, Clec7a, CD47, MIP-1β (CCL4) OPN (SPP1) ( Figure 4 ). However, infiltrating myeloid cells were scarce in active lesions in PMS. Additionally, the TNF^hi microglia population was reduced in active lesions compared to NAWM.

Unsupervised clustering of proteomics data led to discovery of cortical lesions, which were not detectable by routine histology (77). They identified tymosin beta-4 mainly expressed in macrophages and activated microglia at the rim of chronic active WM lesions and in the GM ( Figure 6 ). Tymosin beta-4 is involved in neurite extension and plays a role in restoring and remodeling neurons and in remyelination.

Another study found upregulation of the receptor tyrosine kinase Ephrin3 in the MS lesions. Tissue extracts from MS lesions inhibited OPC, while antibody-mediated masking of EphrinB3 epitopes promoted it (76) ( Figure 5 ). These proteomics studies suggest that EphrinB3 and tymosin beta-4 may be potential targets to promote remyelination.

Proteomics of microdissected active, chronic active and inactive lesions showed that chronic active lesions displayed the highest number of uniquely dysregulated proteins, and proteins of unknown function made up more than half of the unique proteins (71). This was supported by an independent study in 2011 (73). Five proteins involved in coagulation were unique to chronic active lesions (tissue factor, PCI, thromobospondin, fibronectin, vitronectin) (71) ( Figure 6 ). Coagulation factors in the CNS interfere with synaptic homeostasis and neuronal networks, and act pleiotropic on different receptors of both resident and circulating cells as well as the extracellular matrix (80).

Another study found dysregulated proteins associated with extracellular matrix, oxidative stress and myelin sheath (73). There was decreased abundance of MAG (oligodendrocytes) and contactin-1 (neurons), while increase in GFAP (astrocytes) in the chronic active lesions in a milieu with abundant anti-oxidant PRX6 and metabolic processes (alfa-enolase).

Proteome studies with co-immunoprecipitation have discovered that hemoglobin β may play a role in neuronal energetics by interacting with histones in the nucleus and by binding to proteins in mitochondria (74, 75) ( Figure 7 ).

Studies on post-translational modifications will be the next layer of valuable information. Recently, a comprehensive analysis of citrullinated peptides in WM and GM of MS patients identified novel citrullinated sites of MBP, GFAP and vimentin, but their functional role remains unknown (78).

Sample size is often low due to high experimental cost, the need of specific laboratory equipment, and limited access to human MS brain tissue. Most studies conducted on brain tissue include a restricted number of patients, and overlapping these studies is also complicated due to inter-individual and inter-study variations. Additionally, availability of tissue from the early timeframe of the disease course or from the transition to progression is largely missing. Autopsy brain tissues often represent advanced stages of disease from older patients, while biopsy brain tissue is very limited, taken from specific sites and most often from patients with atypical MS. The postmortem delay of tissue varies considerably even within the same study ( Tables 2 , 3 ). In transcriptomic studies, the RNA integrity number (RIN) value is often not mentioned, but the threshold for integrity also depends on which downstream approach is used ( Tables 2 , 3 ). Qualities and quantities also differ, where most identified proteins are the highly abundant (86, 87), and low abundant proteins, likely to be involved in the distinct specific processes, remain to be discovered. Consequently, to find the true pathological signatures, reproducible and robust results are needed generated by well-designed studies including sample size power calculation, standardization of experimental as well as computational pipelines and independent validation. Furthermore, the high experimental costs and limited material demand consortiums and larger studies in collaborations across disciplines and nations using experimental and computational consensus pipelines. This kind of international network of MS experts have already begun as with the “Mystery Solved Project” (88).

Omics provides only static snapshots of cells at different states in a limited area: only a moment is captured of the highly dynamic variations derived from the cell state kinetics, daily biological rhythms and even stratification of patient populations over time. Longitudinal studies or individual cell trajectory tools might be helpful, but the same cell can only be measured once. To overcome this, increasing the data size by learning a latent factor model would be necessary, which encodes some unknown cell state coupled with the cell type for deconvolution. This leads to another problem, where the rapidly produced comprehensive omics data challenge the current computational methods and tools for integrative analyses.

There is a danger that too much trust is given to the output data without comprehending, how those data were obtained. Especially, there is no criteria for the sample size, the quality and standardized computational pipelines. Difficulties in combining different datasets have also been emphasized by a comparison of proteome, mRNA and protein abundance profiles of oligodendrocytes and myelin (89). The challenges to develop true robust integrative methods include different modalities, batch effects between experiments, low sequencing depth and high-modality interactions.

Furthermore, directly translate changes in the transcriptome to the dysregulated proteome is improper due to posttranscriptional regulations and spatial and temporal differences in the production of RNA and proteins. On top of that, protein function and turnover are intensely regulated by posttranslational modifications. Phosphorylation and cysteine modifications regulate protein activity; glycosylation affects protein-protein interactions; and ubiquitination affects protein localization and turnover. Activity of a protein, and its abundance in a cell cannot be deduced with certainty from the level of the corresponding mRNA.

Another challenge is to clarify, how single features are associated through multiple interactions across distinct systems and networks, and how to validate them in simplified “artificial” functional assays and models. Functional follow-up studies of the discovered networks and molecules in the right context are required to obtain specific functional annotations as discussed in 4.6. A potential approach to gain full mechanistic insight will require coordinated sets of molecular and cellular multilayer omics data obtained at multiple time points and collected from disease-relevant tissues representing different stages of damage or repair. Additionally, combination of different omics in different human compartments, and combination of omics in the human disease with animal models may help to assess the biological significance (55, 81). However, such combination of omics techniques needs high-level integration. Combination of data-driven and knowledge-driven models into integrative models may define, whether the altered pathways are related to cause or effect. Here, in situ RNAseq will also help in elucidating these aspects of cell-cell interaction without the need of artificial in silico and in vitro modeling.

With the rapid acquirement of data, the concept to understand the heterogeneity of MS may change, starting from the causative molecular signature rather than the clinical phenotype (90). The classic approach (analytical forward approach ) ( Figure 9 ) applies omics of a patient group with a particular phenotype and determines, which variants these people have in common. In contrast, analyses may also start from large omics datasets by examination what human variants have in common in a clinical setting and connect it to endophenotypes (biological reverse approach) ( Figure 9 ). Applying this latter strategy for understanding the mechanism behind MS phenotypes, the interaction of functional subsets of single cells and their unique intracellular systems should be analyzed, where macromolecules and key hubs interact with each other in networks. The observed heterogeneity of cell subtypes (endophenotypes) in individual MS lesions may be responsible for the evolution of different lesion types, and the heterogenous composition of these lesion types may contribute to changes in specific brain networks that are ultimately responsible for the clinical heterogeneity ( Figure 9 ). However, here the snapshot problem will also still be an obstacle.

Finally, MS disease classification is only based on clinical phenotypes and not endophenotypes. Differential signatures in the CSF may reflect the presence of particular lesion types in the brain but also highlight the heterogeneity of lesion/pathogenesis subtypes (endophenotypes) in phenotypically similar patient groups. However, such heterogeneity may also arise from the timepoint of sampling. Avoiding this, repeated analyses of samples with large sample size are needed. While solving the “snap-shot problem”, and also adding the endophenotypic signatures for the patients may specify the pathological events, and thereby use more targeted therapies. A recent study found strong association between severe cortical pathology and a distinctive CSF inflammatory profile (91). Additionally, using positron-emission tomography (PET), potential future targets for biomarkers could be identified in different MS lesion types in vivo. By combining different sources of information, such as omics and structural/functional neuroimaging, it may be possible to obtain a new integrated picture of the pathophysiological process in MS that could span from molecular alterations to cognitive manifestations.