Neuronal signal propagation in vertebrates is sped up by the electrical insulation of axons with an ensheathing, specialized glial plasma membrane: myelin. Myelination of axons reduces their transverse capacitance and increases their transverse resistance [1]. Insulation is achieved by the multilayered arrangement of the myelin membrane (Fig. 1) and its special molecular composition, mainly its very high lipid content. In myelinated axons, action potentials are restricted to periodically spaced small segments spared from coverage with myelin, termed the nodes of Ranvier [2]. In the central nervous system (CNS), any individual oligodendrocyte myelinates up to 50 axon segments, termed internodes [3]. Oligodendrocyte precursor cell division, migration, and regular alignment along the axons have been recently visualized in vivo in zebrafish [4], which today complement rodents as an important model organism for myelin research [58]. Myelin formation proceeds with outgrowth and retraction of glial cell processes, target axon recognition, stabilization of cellular contacts, rapid biosynthesis and trafficking of lipid and protein constituents of the myelin membrane, and its organization as a multilayered structure around the axon [9, 10]. Once myelinated, axons become dependent on glial support [11]. Some of the molecules involved in myelin development and function are known but a detailed molecular picture has not been gained yet.

Fig. 1
figure 1

CNS myelin. a Purified mouse brain myelin was one-dimensionally separated in a 4–12% Bis–Tris gradient gel using a morpholineethanesulfonic acid buffer system. Proteins were visualized by colloidal Coomassie staining. Bands constituted by abundant myelin proteins are annotated. b Schematic depiction of an oligodendrocyte myelinating an axon, cross-sections in the internodal and paranodal segments, and subcellular localization of myelin proteins. Structural proteins of compact myelin (middle), cytoskeletal and vesicular proteins located in uncompacted regions (right), and adhesion proteins mediating association with the axon (bottom) are shown. CNP 2′,3′-cyclic nucleotide phosphodiesterase, Cntn contactin, Caspr contactin-associated protein, Cx29 connexin 29 kDa, DM20 small splice isoform of PLP, ERM ezrin, radixin, moesin, IPL intraperiod line, JAM3 junctional adhesion molecule 3, MAG myelin-associated glycoprotein, MBP myelin basic protein, MDL major dense line, Necl nectin-like protein, NF155 neurofascin 155 kDa, OSP oligodendrocyte-specific protein/claudin-11, PLP proteolipid protein, Rab3 Ras-related protein Rab3, SIRT2 sirtuin 2

That CNS myelin is important for normal sensation, cognition, and motor function is obvious considering that myelin-related disorders often affect humans lethally. Besides the inflammatory demyelinating disease multiple sclerosis [12], there are genetically inherited disorders that affect CNS myelin, collectively termed leukodystrophies [13]. This heterogeneous group of diseases is characterized by the loss of motoric, sensory, and mental capabilities and the susceptibility to seizures. A detailed knowledge of the molecular expression profiles of oligodendrocytes and myelin will be crucial to understand the pathomechanisms of white matter diseases. For example, the mRNAs [1416] and proteins expressed in cultivated oligodendrocytes [17] and oligodendroglial exosomes [18] have been recently examined. This review focuses on systematic analyses of the molecular composition of mammalian CNS myelin, while no such compendium of peripheral nervous system (PNS) myelin proteins has been published yet. Proteomics approaches to myelin provide a valuable resource to understand its biogenesis, function, and pathology. Although only a few comparative studies have been reported to date, novel insights into the molecular basis of myelin-related diseases are beginning to emerge.

A Myelin-Enriched Fraction from the Central Nervous System

A comparatively simple method is available for the isolation of a myelin-enriched fraction from the CNS. Biochemically, myelin is defined as the lightweight membranous material accumulating at the interface between 0.32 and 0.85 M sucrose after sequential ultracentrifugation combined with osmotic shocks [19, 20]. The most commonly used protocol starts from brain homogenate contained in 0.32 M sucrose as the top layer, “spinning-down” myelin to accumulate at the interface with the bottom 0.85 M sucrose layer. One valuable modification is “floating-up” of myelin starting from brain homogenate contained in a more concentrated sucrose solution as the bottom layer (0.85, 1.2, 1.44, or 2 M). During ultracentrifugation, myelin also accumulates at the interface between the upper 0.85 and 0.32 M sucrose layers, while other fractions of interest assemble at higher sucrose concentrations. This method allows the simultaneous isolation of other brain fractions such as rough microsomes [21] or axogliosomes [22, 23]. The lightweight fraction from the interphase between 0.32 and 0.85 M sucrose is the most frequently used one for biochemical and proteomic experiments. This fraction is enriched in the most abundant proteins of compact myelin, proteolipid protein (PLP), and myelin basic protein (MBP), and as revealed by electron microscopy, mainly contains multilamellar membranes with a periodicity comparable to that of myelin in native or perfused brains [24, 25]. However, we suggest to term this fraction “myelin-enriched” rather than “compact myelin”, as it also contains proteins from the noncompacted cytosolic channels in myelin (i.e., adaxonal and paranodal myelin) and proteins associated with the axonal membrane. Myelin purification is very reproducible across different laboratories, even when applied to different species (e.g., mouse–rat) or to mutant mice with altered myelin protein or lipid composition, such as CnpCre/+*Fdftflox/flox [26], Ugt3a1null [27], Arsanull [28], and Plpnull [29] (see below). Thus, the method has proven to be very robust, explaining why the original protocol from the early 1970s is still in common use. It is generally assumed that myelin purification relies on its special lipid content and composition.

Myelin Lipids

The molecular composition of myelin differs from other plasma membranes in that it contains 70–75% of its dry weight as lipid, unusually high compared to other eukaryotic plasma membranes. Also, its molar ratio of lipids with approximately 2:2:1:1 for cholesterol/phospholipid/galactolipid/plasmalogen [30, 31] distinguishes myelin from other cellular membranes. The abundance of cholesterol within a membrane affects its biophysical properties, including fluidity and curving [32]. Cholesterol has earlier been identified as unusually enriched in myelin and constitutes 24–28% of the total myelin lipids [19]. That the cellular cholesterol supply is rate-limiting for myelin membrane biogenesis has been shown in mice lacking squalene synthase (also termed farnesyl diphosphate farnesyl transferase [FDFT]) exclusively in myelinating glia [26]. FDFT mediates a crucial step of cholesterol biosynthesis. CNS myelination is severely delayed in CnpCre/+*Fdft1flox/flox mice, and that any myelin made in these mice is likely due to compensatory cholesterol uptake from other cells [26].

The biophysical properties of myelin are also influenced by its unusually high concentration of the galactolipids galactosylceramide (GalC), its sulfated form 3-O-sulfogalactosylceramide (SGalC), and their hydroxylated forms GalC-OH and SGalC-OH. Together, they add up to 20–26% of total myelin lipids. Myelination is moderately delayed in mice lacking UDP-galactose:ceramide galactosyltransferase (Ugt3a1), an enzyme required for galactolipid synthesis. Additionally, impaired glia–axonal interactions at the paranodes were observed [27, 33, 34]. Paranodal disruption was at least partly due to the lack of SGalC and hydroxylated galactolipids, since the long-term integrity of the sodium channel domain of the nodes of Ranvier was also impaired in mice lacking galactosylceramide-3-O-sulfotransferase (Gal3st1), the enzyme converting GalC into SGalC [3537], and late onset myelin degeneration was also reported for mice lacking fatty acid 2-hydroxylase (Fa2h), the enzyme hydroxylating GalC and SGalC [38]. Absence of functional arylsulfatase A (ARSA), the enzyme degrading SGalC, causes metachromatic leukodystrophy (MLD), illustrating that a regulated galactolipid metabolism is required for long-term integrity of the white matter. SGalC accumulation and many pathological features of MLD are modeled in Arsanull mice and in transgenic mice overexpressing Ugt3a1 or Gal3st1 in neurons or oligodendrocytes [28, 39, 40]. Sulfatide metabolism with respect to myelin and MLD pathology was recently reviewed [41].

Also, the plasmalogen class of phospholipids is associated with white matter disease. Plasmalogens are ether-linked (as opposed to ester-linked) phospholipids, the main species being ethanolamine–plasmalogen. They are ubiquitous structural components of mammalian cell membranes and amount to 12–15% of total myelin lipid [19] and, when processed by plasmalogen-selective phospholipase A2, give rise to the second messengers arachidonic acid and eicosanoids [42]. At low concentrations, these metabolites have trophic effects, but at high levels, they are cytotoxic and may induce inflammation [43]. The reactivity of the alkenyl ether bond makes plasmalogens more susceptible to oxidative reactions than their fatty acid ester analogs. Thus, myelin plasmalogens may act as endogenous antioxidants protecting cells from oxidative stress [44]. Disrupted activity of peroxisomal plasmalogen synthesizing enzymes results in peroxisomal biogenesis disorders such as rhizomelic chondrodysplasia punctata (RCDP) in which hypomyelination of the optic nerve has been observed. Decreased plasmalogen levels [45, 46] and increased levels of reactive oxygen species [47, 48] may also contribute to the demyelination in X-linked adrenoleukodystrophy caused by the mutated peroxisomal transporter ABCD1, suggesting that a normal plasmalogen metabolism may prevent peroxisomal- and myelin-related disease. Mice lacking dihydroxyacetonephosphate acyltransferase (DAPAT) model several aspects of the RCDP pathology, including optic nerve hypoplasia [49]. Interestingly, the association of flotillin-1 and contactin with plasmalogen-deficient brain membrane microdomains was diminished in DAPATnull mice [49], suggesting that the local concentration of membrane lipids dictates the association of particular proteins.

Association of Myelin Lipids and Proteins

Cholesterol assembles with galactolipids and plasmalogens within the plane of the membrane, but how they are enriched to the levels found in myelin is unknown. It has been suggested that lipids are targeted to future myelin membrane by their association with myelin-bound proteins [9]. SGalC appears to be an example to the contrary. SGalC is associated with myelin and lymphocyte protein (MAL) [50]. Lack of SGalC and lack of MAL lead to similar paranodal malformation [35, 51]. The subcellular trafficking of MAL, as well as its abundance in myelin, is determined by SgalC [28], whereas SGalC abundance is not altered in Malnull myelin [51]. It is likely that other myelin proteins are also incorporated into the sheath by attachment with future myelin membrane because of its special lipid composition. Thus, whether myelin proteins dictate the fate of lipids or vice versa may not be generalized. It appears likely that the association of both molecule classes results in each other’s control of abundance and trafficking.

That myelin lipids and proteins are closely associated was suggested earlier after the characterization of two types of protein fractions isolated from the white matter based on their resistance to aqueous or organic solvents or to enzymatic proteolysis. One fraction behaved as a lipid with regard to its solubility and was termed PLP [52, 53]. PLP was later identified to be the most abundant protein of mammalian CNS myelin. It has a high affinity to phospholipids and cholesterol [5456], and impaired interactions of mutant PLP with membrane lipids are a likely key step in the molecular pathogenesis of the leukodystrophy Pelizaeus–Merzbacher disease [57]. The other fraction, termed trypsin-resistant protein residue, was insoluble in organic solvent and attached to the membrane lipid phosphatidylinositol phosphate [58, 59]. The application of extraction methods by Folch became commonly used to categorize myelin proteins according to their biophysical properties.

More recently, the myelin-enriched brain fraction has been chemically subfractionated by differential detergent extraction at low temperatures, resulting in distinct nonidentical but overlapping assemblies of myelin-associated proteins and lipids that were suggested to represent myelin subcompartments [60, 61]. Cholesterol- and galactolipid-rich membrane microdomains (also referred to as “lipid rafts”) have been suggested to deliver myelin proteins to the plasma membrane [6264]. The relevance of applying the analysis of biochemical characteristics established for membrane microdomains to such a large structure as myelin has remained debated. However, it is widely accepted now that lipid-associated cell signaling molecules, such as the protein tyrosine kinase fyn, have central roles in myelination [65, 66].

In oligodendroglial processes, fyn is activated by axonal signals via integrin alpha6beta1 [67]. Among other fyn substrates [68, 69], the protein translation repressor heterogeneous nuclear ribonucleoprotein (hnRNP) A2 upon phosphorylation is released from its binding site in the 3′UTR of mRNA encoding MBP [70], the second-most abundant myelin protein. hnRNP A2 binding represses translation during the translocation of MBP mRNA to distal sites of the cell [71] where newly translated MBP is directly incorporated into the extending oligodendroglial process [21, 72]. It is generally assumed that MBP mediates the adhesion of the cytoplasmic surfaces between the individual layers of compact myelin [73] via binding of its many basic residues with the negatively charged headgroups of membrane lipids. Indeed, membrane association of MBP is controlled by the membrane lipid phosphatidylinositol-(4,5)-bisphosphate [7476]. For over 30 years, it has been known that MBP is highly heterogeneous due to alternative splicing and multiple post-translational modifications (PTMs) [77]. More recently, modern mass spectrometric techniques have been used to compare the PTMs of MBP from normal and multiple sclerosis brains with respect to methylation, phosphorylation, and arginine deimination [78]. PTM alterations affect charge, conformation, and hydrogen bonding of MBP, which may modulate its affinity to the myelin membrane and play a role in myelin compaction and in the pathogenesis of demyelinating diseases. MBP is the only myelin protein that has been shown to be essential for myelin formation, as became obvious with the analysis of the natural mouse mutant shiverer and the rat mutant long evans shaker [79, 80], which are severely hypomyelinated. Interestingly, mice lacking fyn are also hypomyelinated [81, 82], likely due to affected translational regulation of MBP expression [70, 83]. Together, a multitude of factors affects mRNA transcription and transport, translation at axonal contact sites, or membrane binding of MBP, and we speculate that several myelin proteins with yet unidentified roles affect MBP abundance and function.

Systematic Analysis of the CNS Myelin Protein Composition

The relative abundance of myelin proteins has previously been calculated based on their binding to Buffalo black [84], Fast green [85], or Coomassie blue [86] after separation in one-dimensional (1D) sodium dodecyl sulfate (SDS)–polyacrylamide gel electrophoresis (PAGE). In these measurements, a small number of proteins was determined to be extraordinarily abundant in CNS myelin. PLP and its smaller splice isoform DM20 accounted for 30–45% of total myelin protein, two of the four MBP splice isoforms for 22–35%, 2′,3′-cyclic nucleotide 3′-phosphodiesterase (CNP) for 4–15%, and all remaining proteins for 5–25% [19, 85, 87, 88]. Similarly, PNS myelin is also dominated by two proteins, myelin protein zero (MPZ, P0) and MBP, which have been estimated to account for 50–70% and 15%, respectively [89]. In comparison, the most abundant proteins in a brain fraction enriched for synaptic vesicles are synaptobrevin 2 and synaptophysin, which constitute 8% and 10% of the total synaptic vesicle proteins, respectively, as revealed by quantitative immunoblotting [90]. How and why myelin proteins are enriched to their unusual relative abundance is unclear, considering that PLP and CNP are not essential for the formation of normal amounts of CNS myelin [29, 91, 92].

Various proteomic techniques have been applied towards the systematic protein composition analysis of the myelin-enriched fraction. Traditionally, first insights into proteomes of subcellular structures often come from two-dimensional (2D) protein maps generated by utilizing isoelectric focusing (IEF) with immobilized pH gradients in the first and SDS-PAGE in the second dimension (2D-IEF/SDS-PAGE) (Fig. 2a). Proteins of interest are then excised from the gel, proteolytically digested in situ, and finally, identified by mass spectrometry (MS) [93]. Due to its high resolving power, 2D-IEF/SDS-PAGE can be routinely applied for profiling of proteins from complex mixtures and, as protein integrity is retained, also leads to information on protein abundance and processing [94]. However, major shortcomings of 2D-IEF/SDS-PAGE concern a limited dynamic range, the display of basic and hydrophobic proteins, and—most importantly—the under-representation of membrane proteins. As myelin is dominated by MBP (a highly basic protein) and PLP (a hydrophobic tetraspan protein), incremental improvements in 2D-IEF/SDS-PAGE technology were required before the first 2D mapping of myelin was presented [95]. By using the zwitterionic detergent amidosulfobetaine-14 (ASB-14) instead of the most commonly used 3-[(3-cholamidopropyl)-dimethylammonio]-1-propanesulfonate (CHAPS) [96], it was possible to solubilize myelin proteins much more effectively and to identify 98 proteins (91 by MS and seven by immunoblotting) in the myelin-enriched fraction from mouse CNS [95]. This crucial effect of the solubilization conditions is further underscored by two more recent 2D-IEF/SDS-PAGE mapping studies of similar input material. Thirty-eight myelin-associated proteins were identified in one study after CHAPS solubilization [97], but 131 proteins were identified in another study with ASB-14 [25]. Thus, at least in the presence of appropriate detergents, myelin can now be considered as well accessible by 2D-IEF/SDS-PAGE, which not only facilitates protein cataloging but also paves the way for differential myelin proteomics on the basis of the 2D differential fluorescence intensity gel electrophoresis technology (2D-DIGE, see below). It is important to note that all conventional 2D mapping approaches mentioned above failed to appropriately display relatively abundant transmembrane myelin marker proteins such as PLP, myelin-associated glycoprotein (MAG) [98], myelin oligodendrocyte glycoprotein (MOG) [99], tetraspanin 2 [100], M6B [101], or oligodendrocyte-specific protein (OSP/claudin-11) [102104]. A potential remedy is to perform the first dimension separation as nonequilibrium pH gradient electrophoresis for the 2D mapping of myelin proteins [105]. However, although this method appeared promising particularly for displaying highly basic proteins, it did not get as popular as 2D-IEF/SDS-PAGE with immobilized pH gradients, mainly due to limitations in reproducibility and resolution.

Fig. 2
figure 2

Gel-based myelin proteome maps. Purified mouse brain myelin was two-dimensionally separated in different gel systems. Proteins were visualized by colloidal Coomassie staining, and spots constituted by selected myelin proteins are indicated. a 2D-IEF/SDS-PAGE with IEF in a nonlinear pH gradient (pH 3–10) as the first and gradient SDS-PAGE (8–16% acrylamide) as the second dimension. To improve resolution, myelin was delipidated and precipitated by a methanol/chloroform treatment prior to IEF [25]. b 2D-16-BAC/SDS-PAGE with separation in a 16-BAC gel (10% acrylamide) as the first and gradient SDS-PAGE (8–16% acrylamide) as the second dimension. c 2D-CTAB/SDS-PAGE with separation in a CTAB gel (10% acrylamide) as the first and gradient SDS-PAGE (8–16% acrylamide) as the second dimension. To deplete soluble and membrane-associated proteins, myelin was subjected to a multistep wash procedure before separation [25]. 16-BAC and CTAB resulted in similar spot patterns. 2D-IEF/SDS-PAGE provides good resolution but basic, hydrophobic, and transmembrane proteins are under-represented. 2D-16-BAC/SDS-PAGE and 2D-CTAB/SDS-PAGE lead to efficient representation of basic, hydrophobic, and transmembrane proteins but have a lower resolution since separation occurs by protein size in both dimensions

More complete proteome coverage while retaining the benefits of displaying intact proteins can be reached by the additional use of alternative 2D gel systems. Here, the charge-dependent separation in the first dimension (i.e., the IEF) is replaced by a size-dependent separation in the presence of cationic detergents such as 16-benzyldimethyl-n-hexadecylammonium chloride (16-BAC; Fig. 2b) [106] or cetyltrimethylammonium bromide (CTAB; Fig. 2c) [107]. Due to the similar separation principle in both dimensions, proteins are typically dispersed along a diagonal rather than distributed over the entire gel area. Accordingly, these gel systems have a lower resolution compared to 2D-IEF/SDS-PAGE, but can resolve highly basic and even membrane-spanning proteins [108]. Application of 2D-16-BAC/SDS-PAGE to mouse CNS myelin resulted in the identification of 62 proteins and readily enabled displaying of the transmembrane myelin proteins PLP, MAG, MOG, and OSP/claudin-11 [25]. Thus, the combination of 2D-IEF/SDS-PAGE and 2D-16-BAC/SDS-PAGE has, so far, yielded the most comprehensive gel-based proteome compendium of mouse CNS myelin, consisting of 162 nonredundant proteins [25]. Further technical refinements of the method were established in a recent systematic evaluation of five different cationic detergents for the 2D gel electrophoresis of myelin proteins. Here, 16-BAC was the most effective agent for the separation of myelin proteins in the first dimension, while CTAB was most effective for their solubilization [109, 110]. As resolution improves, 2D gel electrophoresis with cationic detergents may be combined with the DIGE technology as a future tool for monitoring abundance changes of highly basic and membrane-spanning myelin proteins [111].

To overcome the limitations of gel-based proteomic methods, in particular those of 2D-IEF/SDS-PAGE, gel-free techniques, commonly referred to as shotgun approaches, have emerged in recent years [93, 112]. Here, separation at the level of intact proteins is omitted and the protein preparation is proteolytically digested at the expense of information related to protein integrity, such as protein size and charge. Separation takes place at the level of proteolytic peptides before interfacing with MS. The tremendous complexity of such peptide mixtures requires a high resolving power and is, therefore, often addressed by the application of 2D liquid chromatography (2D-LC), usually consisting of strong cation exchange in the first and reversed-phase chromatography in the second dimension. In the first application of shotgun proteomics to the myelin-enriched fraction from the mouse CNS [97], 93 proteins were identified resulting—by combination with 2D-IEF/SDS-PAGE (see above)—in a myelin proteome compendium consisting of 103 proteins. The application of a similar shotgun approach to a myelin-enriched fraction from rat CNS led to the identification of 97 myelin proteins [23]. Both shotgun approaches yielded quite a high overlap of approximately 50% with the so far most comprehensive gel-based library [25] and contained transmembrane myelin proteins such as PLP, MAG, and MOG.

Relative Abundance of Myelin Proteins

To understand myelin biogenesis and pathology, a comprehensive knowledge of the proteins associated with myelin is a prerequisite. We have confirmed and expanded the previous myelin protein compendia by applying nanoscale 1D ultra performance liquid chromatography (1D-UP-LC) separation coupled to detection with a quadrupole time-of-flight (QTOF) mass spectrometer (Tenzer et al., unpublished). Data were acquired by LC-MS using an alternating low (MS) and elevated (MSE) collision energy mode of acquisition (LC-MSE), which allows simultaneous identification and label-free relative quantification of the proteins in the sample [113115]. The identified peptides were annotated to a total of 294 myelin-associated proteins (Table 1) based on a minimum of two peptides per protein with an effective false-positive rate of <0.2%. They showed a very good overlap of 141 proteins that were also detected in previous myelin proteome analyses and included several established myelin markers (Table 1 and Fig. 3). We have calculated the relative abundance of the myelin-associated proteins based on the average intensity of the three most abundant peptides per protein. In the few cases where only two peptides were identified, their average intensity was used. Strikingly, PLP, MBP, and CNP constituted only 17%, 8%, and 4% of the total myelin-associated proteins, respectively (Fig. 4). All previously known myelin proteins together constituted 35%, while newly identified myelin-associated proteins accounted for 65%. These quantifications take into question previous estimates based on conventional techniques (Fig. 4b and see above). We suggest that the complexity of myelin protein composition has been overlooked because low abundant proteins did not constitute significant bands on gels when compared to the highly abundant PLP and MBP due to limitations concerning gel separation and/or protein staining.

Table 1 The CNS myelin proteome
Fig. 3
figure 3

Assembling a compendium of myelin proteins. a The number of proteins identified by MS in different approaches to the CNS myelin proteome is plotted. The total number of myelin-associated proteins is unknown. Transmembrane proteins (black) have been categorized based on prior experimental studies or have been predicted using TMHMM and Phobius software. Proteins associated with mitochondria, which copurify with myelin, were omitted. T 2D-IEF/SDS-PAGE [95], V 2D-IEF/SDS-PAGE [97], W 2D-IEF/SDS-PAGE [25], B 2D-16-BAC/SDS-PAGE [25], R shotgun [23], S shotgun [97], E LC-MSE (Tenzer et al., unpublished). b Venn diagram comparing the number of myelin-associated proteins identified by MS after gel separation [25, 95, 97], previous gel-free shotgun approaches by LC/LC-MS/MS [23, 97], with those identified by LC-MSE (Tenzer et al., unpublished). Note the high overlap of proteins identified independent of the technique used. c Venn diagram showing our own experience with the identification of myelin-associated proteins by MS after combined 2D-IEF/SDS-PAGE and 2D-16-BAC/SDS-PAGE separation [25] or by LC-MSE with known myelin proteins according to the literature

Fig. 4
figure 4

Relative abundance of myelin proteins. a The abundance of known myelin proteins was determined by LC-MSE. Note that known myelin proteins constitute less than 50% of the total myelin protein. Mitochondrial proteins were not considered. b Comparison of myelin protein abundance as quantified by LC-MSE with previous estimates based on band intensity after 1D-PAGE and various protein staining techniques [19, 85, 87, 88]. Note that the abundance of PLP and MBP was previously overestimated because low abundant proteins did not constitute significant bands due to limitations in the resolving power of the 1D gels and in the dynamic range of protein staining. c Simulated 2D map of myelin-associated proteins identified by LC-MSE. Proteins are indicated as dots at their molecular weight and isoelectric point as predicted from the amino acid sequence. The size of each dot reflects the relative abundance as determined by LC-MSE. Myelin-associated proteins without transmembrane domains are shown in blue and transmembrane proteins in green, the latter being usually under-represented or absent from conventional 2D gels. Mitochondrial proteins are shown in gray. The red frame indicates the portion of proteins that can be reproducibly displayed by 2D-IEF/SDS-PAGE (see Fig. 2a)

We conclude that modern LC-MS-based approaches—though technically more demanding than gel-based studies—appear to be appropriate for tackling the myelin proteome as they cover several orders of magnitude of protein abundance and detect highly basic, hydrophobic, and membrane-spanning proteins. This tackles the bias towards certain protein classes, which is the major shortcoming particularly of 2D-IEF/SDS-PAGE (Fig. 4c). Moreover, LC-MS-based approaches enable the gel- and label-free quantification of proteins from complex mixtures, which allowed for the systematic reassignment of protein abundance in CNS myelin (see above). Finally, they require only low amounts of sample, which is of special relevance for the proteome analysis of myelin purified from hypomyelinated model animals or human brain autopsy material.

Technical Limitations

How pure is the myelin-enriched fraction? Myelin-associated proteins are defined as proteins in the myelin-enriched fraction since all studies have operationally defined the term “myelin protein” without systematic experimental verification. Although the identification of new myelin proteins by more than one study and the detection of established myelin markers increase confidence, some of these proteins may only have copurified with myelin. The high dynamic range of LC-MSE leads to the new identification of proteins as myelin-associated, but also to the false-positive identification of contaminants. These mainly stem from copurifying mitochondria and synaptic vesicles. In reverse, proteomic compendia of mammalian brain mitochondria [116] or synaptic vesicles [90] include classical myelin proteins such as PLP, MBP, MOBP, and MAG. Notwithstanding that some of these proteins may have a dual localization, cross-contamination occurs likely due to similar floatation properties in sucrose or Percoll gradients and can only be excluded once improved separation protocols become available. Proteins of the axonal plasma membrane, such as potassium channels or Na+/K+-ATPases, have also been detected in the myelin fraction, which can be explained by the tight linkage of the membranes via adhesion proteins, sometimes referred to as the myelin–axolemma complex [24]. Indeed, some adhesion complexes are present in the myelin-enriched fraction, such as the glial neurofascin (NF155) and contactin and their axonal partner contactin associated protein 1 (Caspr) [117120] and the glial nectin-like protein Necl4 and its axonal counterpart Necl1 [121124]. Importantly, myelin proteome analysis also revealed novel candidate proteins to mediate intracellular or intercellular adhesion, such as the immunoglobulin domain superfamily protein Igsf8, also termed EWI-2 [23]. Igsf8 is associated with the myelin tetraspanins CD9 and CD81 and regulates integrin function, at least in vitro [125, 126], but its function in vivo remains to be shown. The experimental validation or falsification of newly identified myelin-associated proteins will be a matter of the systematic application of histological techniques, provided that reliable antibodies are available.

How many proteins can be considered true myelin proteins? Though proteomic compendia aim at completeness, the number can only be guessed at this time. As the dynamic range of current MS-based protein identification schemes is in the range of three to five orders of magnitude, detection of infrequent proteins remains a challenge. Additionally, some technical impediments remain. The myelin proteins CD9 [127, 128], oligodendrocyte myelin glycoprotein [22, 129], and MAL [51] have not yet been detected by proteomic approaches, and the appearance of MAL in one catalog [95] is due to the additional use of immunoblotting. Its nondetectability illustrates the limitations of proteome analysis. MAL is a very hydrophobic protein with four transmembrane domains and very small cytoplasmic and extracellular domains and is, therefore, hardly accessible by MS-based identification. Apart from the membrane-spanning peptides not visible in proteomic approaches, complete tryptic digest of MAL results in only four theoretically detectable peptides: one of 120 amino acids (which is too long for identification by MS), two of two amino acids each (too short to provide useful sequence information), and one of 29 amino acids, which is, in principle, appropriate for identification. However, to obtain a reasonable level of confidence for protein identification, the detection of two peptides per protein is usually set as a prerequisite in the algorithms. This suggests that all proteome approaches requiring protease cleavage have an inherent bias against very small polypeptides or proteins with an unusual cleavage site pattern. In future experiments, the lack of suitable trypsin cleavage sites may be circumvented by the use of endopeptidases with different specificities (e.g., GluC or AspN), although they create proteolytic peptides lacking a basic C-terminal amino acid and are difficult to sequence [130]. This suggests that the detection of more myelin-associated proteins is not just a matter of higher resolving power but also of other technical refinements.

Newly Identified Myelin-Associated Proteins

The compendium of proteins identified in the myelin-enriched brain fraction represents a valuable reference for myelin research. The proteins are candidates for performing important functions in myelin biogenesis and integrity, molecular interactions between myelinating glia and neighboring cells, and white matter homeostasis. By gene ontology terms (, many myelin-associated proteins are implicated in catalytic activities (48%), the cytoskeleton (20%), protein transport (21%), vesicular trafficking (6.8%), cell adhesion (6.3%), phospholipid binding (4.2%), or glycolysis/gluconeogenesis (5.1%). Among the recently identified myelin proteins, some were first and others subsequently detected using proteomic approaches. They include proteins of quite various anticipated functions, such as the NAD+-dependent deacetylase sirtuin 2 (SIRT2, see below), cytoskeletal proteins of the septin family [23, 25, 131], and ermin [132], regulators of intracellular vesicle transport in the secretory pathway, such as cdc42 and Rac1 [133], Rab3A, and other Rab-GTPases [134, 135], the paranodal transmembrane glycoprotein Opalin/TMEM10 with a suggested signaling or adhesive function [136138], the nucleoside diphosphate kinases NM23A and NM23B [95], and a protein particularly abundant in the CNS myelin of teleost fish, the 36K protein, also termed short-chain dehydrogenase/reductase (SDR family) member 12 (DHRS12) [139]. Some of these are quite abundant myelin proteins as judged both by the spots constituted on 2D gels and LC-based quantification, and the challenge to establish their functions in vivo promises a deepened understanding of myelin. Besides, novel myelin proteins are candidates to cause (when mutated), enhance, or ameliorate white matter disease, such as leukodystrophies.

Differential Myelin Proteome Analysis in Myelin-Related Disease

The proteomic comparison of myelin from human patients or animal models with that of respective controls is a powerful approach towards the identification of secondary molecular changes that may contribute to the pathogenesis of myelin-related disease. Such a differential approach has first been applied to myelin purified from PLPnull mice [25], which provide a genuine model for spastic paraplegia (SPG-2) in humans, a comparatively mild variant of the leukodystrophy Pelizaeus–Merzbacher Disease with progressive axonal degeneration in the presence of normal amounts of CNS myelin [29, 140]. In that study, 2D-DIGE [141] was used to screen for candidate proteins that could be involved in the oligodendroglial failure to support the long-term integrity of myelinated axons. Three distinct proteins of the cytoskeletal septin family were found to be reduced, and the deacetylase SIRT2 was virtually absent from PLPnull myelin. SIRT2 is an abundant myelin protein in the CNS and the PNS [23, 25, 142] and regulates microtubule dynamics during oligodendrocyte development [143]. Whether acetylated α-tubulin is a relevant substrate of SIRT2 in vivo remains to be shown. Similar to PLPnull mice, CNPnull mice are also normally myelinated but develop length-dependent axonal loss [92, 144]. It is intriguing that CNP also modulates microtubule dynamics [145, 146]. Taken together, spatiotemporal control of microtubule stability in oligodendrocytes (by SIRT2, CNP, and likely other factors) seems critical for normal axon–glia interaction.

Acetylation is a reversible post-translational modification of numerous mammalian proteins [147, 148], and all acetylated myelin proteins (α-tubulin, MBP, MOG, and several nonidentified proteins of lower abundance) are candidate substrates for SIRT2 [25]. In oligodendrocytes and myelin, SIRT2 activation upon increased axonal NAD+ levels may remove acetyl residues from myelin-associated proteins with consequences for their net charge and function. Interestingly, SIRT2 has been recently shown to interact with 14-3-3 beta and gamma [149], which are myelin-associated as revealed by proteome analysis (Table 1). Their interaction is strengthened by the serine/threonine kinase AKT [149], which is a central signaling molecule for CNS myelination [150]. 14-3-3 proteins have been implicated in membrane protein transport, exocytosis [151], and stress response [152], but their function in myelin has not yet been investigated. 14-3-3 proteins are homologs of the C. elegans partitioning-defective polarity protein Par5 and bind to the tight junction-associated Par3 [153, 154], which is required for establishing polarity prior to myelination, at least by Schwann cells in the PNS [155]. To determine whether SIRT2, 14-3-3 proteins, Par-proteins, protein kinases, and tight junctions indeed interact in myelinating glia will be an important topic of future investigation. We speculate that the competence of oligodendrocytes to dynamically react to NAD+ level changes in white matter tracts is required for their role in maintaining long-term axonal integrity.

With the objective to identify novel therapeutic targets for the treatment of multiple sclerosis, a systematic proteomic profiling of tissue samples from three brain lesions affected to various degrees (acute plaque, chronic active plaque, and chronic plaque) has recently been performed [156]. Material from the respective lesion type was collected by laser-capture microdissection and extracted proteins were separated by 1D gel electrophoresis followed by mass spectrometric protein identification. Unexpectedly, five coagulation proteins, including tissue factor and protein C inhibitor, were only present in chronic active plaque characterized by concomitant inflammation and degeneration, a finding that provided new insights in the relationship between the coagulation cascade and inflammation. Most importantly, administration of inhibitors to tissue factor (i.e., hirudin) and protein C inhibitor (i.e., activated protein C [aPC]) indeed ameliorated the disease phenotype in experimental autoimmune encephalomyelitis, a model of multiple sclerosis. The anti-inflammatory treatment with engineered aPC variants may develop into an alternative route to a therapy of multiple sclerosis. Together, differential proteome analysis has identified secondary molecular changes that contribute to understanding the pathogenesis of myelin-related disease and support the design of rational treatment strategies.