Pushing myelination – developmental regulation of myosin expression drives oligodendrocyte morphological differentiation

Myelinating oligodendrocytes (OLs) are highly specialized cells of the vertebrate central nervous system (CNS) (Zalc, 2016). OLs derive from oligodendrocyte precursor cells (OPCs), which during differentiation protrude multiple hybrid (actin and tubulin-enriched) structures that further evolve into lamellipodia-like membranous extensions from which compact myelin is ultimately formed (Song et al., 2001; Wilson and Brophy, 1989; Domingues et al., 2018). Recent experimental evidence suggests that the remarkable morphological and functional transformations underlying OPC differentiation are driven by actin cytoskeleton-based forces (Kippert et al., 2009; Nawaz et al., 2015; Zuchero et al., 2015). In higher eukaryotes, the intrinsic dynamics of actin cytoskeleton is modulated by the presence of myosin motors. Humans display the highest number and classes of myosins, with almost 40 myosins distributed over 18 classes based on their sequence similarity. Such remarkable myosin variability supports its involvement in multiple and essential functions (e.g. cell differentiation, migration and division) (Sellers, 2000; Sebé-Pedrós et al., 2014; Heissler and Sellers, 2016). Structurally, myosins are characterized by three canonical domains: the motor domain with ATPase activity involved in actin binding and movement, the neck domain that binds myosin light chains and calmodulin, and a tail domain of variable length and class-specific composition that serves to anchor and position the motor domain so that it can interact with actin and connect to various cargo-associated proteins (Sellers, 2000).

The conventional non-muscle class II myosins (NM2) are hexameric structures composed of two heavy chains and two pairs of light chains [the regulatory light chain (RLC), and the essential light chain (ELC)]. The most striking characteristic of these myosins is their two-headed structure due to the dimerization of the heavy chain in the tail region to form bipolar filaments. These characteristics favor cell contractility when bound to and sliding on antiparallel actin filaments (Billington et al., 2013). In mammals there are three genes – Myh9, Myh10 and Myh14 – encoding the NM2 heavy chains. The heavy chains, together with the bound RLCs and ELCs, are termed NM2A, NM2B, and NM2C, respectively. In contrast, the unconventional class 1, class 5 and class 6 myosins do not form bipolar filaments and work in monomers or dimers to transport cargo (Li et al., 2016).

Several members of the myosin superfamily are known to be expressed by OLs and contribute to the process of myelination: the conventional NM2A and NM2B, and the unconventional myosin-1D (Myo1d), myosin-5A (Myo5a) and myosin-6 (Myo6). While NM2B function is associated with actomyosin contractility and we previously showed that its depletion in OLs accelerates myelination (Wang et al., 2008, 2012; Kippert et al., 2009; Rusielewicz et al., 2014), inhibition of myosins functionally associated with trafficking mechanisms, such as Myo1d, Myo5a or Myo6, lead to OL differentiation arrest (Sloane and Vartanian, 2007; Yamazaki et al., 2014, 2016, 2017). Despite this experimental evidence, it remains elusive how different myosins dynamically interplay and contribute to large-scale cellular behaviors during OL morphogenesis and myelination.

In this study, we conducted a systems biology-oriented analysis of the subcellular distribution of the OL myosinome, which included: (1) transcriptomic and proteomic databases, (2) myosin domain organization, (3) myosin mechanochemical properties, (4) functional enrichment of myosin interactomes, and (5) quantitative imaging of myelinating co-cultures of OLs with dorsal root ganglion (DRG) neurons depleted of specific isoforms of NM2 myosins. Our analysis reveals that the OL differentiation program is regulated by spatiotemporal synchronization of different myosins able to modulate distinctive mechanical features. We hypothesize that, at early stages of OPC differentiation, cell symmetry breaking and emergence of local protrusions requires non-uniform cortical tension, which are dependent on NM2B topology. The transition to mature OL implies a global decrease of NM2B expression for cortical tension reduction and protrusive membrane growth. At this stage, NM2C could potentially contribute to vesicular transport promoting myelin membrane extension.

We aimed to identify and catalog all myosins expressed in developing OL and establish how their expression temporally correlates with the different stages of differentiation. Combined expression analysis (transcriptomic and proteomic data) (Cahoy et al., 2008; Wang et al., 2008; De Monasterio-Schrader et al., 2012; Zhang et al., 2014; Marques et al., 2016; Thakurela et al., 2016; Yamazaki et al., 2017; Azevedo et al., 2018) identified 16 myosins in the OL lineage belonging to classes 1, 2, 5, 6, 9, 10, 16, 18 and 19: Myo1b, Myo1c, Myo1d, Myo1e, Myo1f, Myh9, Myh10, Myh14, Myo5a, Myo6, Myo9a, Myo9b, Myo10, Myo16, Myo18a and Myo19. Among these, seven were consistently identified in five independent studies: Myo1d, Myh9, Myh10, Myh14, Myo5a, Myo6 and Myo18a (Fig. 1A). Analysis of myosin transcript abundances and expression profiles, as reported in the Brain RNASeq (http://www.brainrnaseq.org; Zhang et al., 2014), clearly shows that seven myosins are developmentally regulated and display two contrasting expression profiles. Specifically, Myh9, Myh10 and Myo5a are markedly downregulated during OL differentiation, while the expression of Myo1d, Myh14, Myo6 and Myo18a transcripts is upregulated (Fig. 1B). Of note, OLs express all the three transcripts for the heavy chains of conventional non-muscle class 2 myosin isoforms – Myh9, Myh10 and Myh14. In agreement with our previous published data (Wang et al., 2008), we observe the expression of NM2A (Myh9) and NM2B (Myh10) transcripts in the OL lineage and their decreased protein expression during differentiation (Fig. 1C). By contrast, we observe that the transcript for NM2C (Myh14) heavy chain as well as its protein levels have an opposite expression profile, that is, increased expression during OL differentiation (Fig. 1A–C). Interestingly, myosin Myo18a, which is phylogenetically related to NM2s (Peckham, 2011), is also upregulated in myelinating OLs (Fig. 1A–C). The same pattern of expression, that is, decreased for NM2B and increased for NM2C as myelination proceeds, is observed in developing rat brain (Fig. S1). These findings are confirmed by the enrichment of NM2C and Myo18a in the transcriptome and proteome of purified CNS myelin fractions of both rodent and human origin (Cahoy et al., 2008; De Monasterio-Schrader et al., 2012), and in agreement with earlier evidence of higher abundance of Myh14 transcripts in human adult tissues, especially in white matter-enriched regions (Golomb et al., 2004). Unlike Myh9 and Myh10, Myh14 is not expressed in mouse embryonic stem cells or fetal tissues (Golomb et al., 2004). At the subcellular level, we have previously observed that Myh14 and Myo18a mRNAs are significantly enriched in OPC protrusions (Fig. 1A) (Azevedo et al., 2018). Although the physiological significance of such enrichment is still unknown, OPC protrusions are important structures for cell branching, local regulation of myelin basic protein (MBP) translation and membrane expansion during myelination (Laursen et al., 2011; Azevedo et al., 2018). Protein expression of the unconventional myosins Myo1d, Myo5a and Myo6 was recently analyzed in developing rat OL cultures (Fig. 1A) (Yamazaki et al., 2017). While increased Myo1d expression during OL differentiation was detected in that study at both the protein and mRNA, the levels of Myo5a and Myo6 proteins levels do not change during OL differentiation. Altogether, the expression analyses of the different myosins in developing OLs indicate that the myosinome is temporally regulated, in particular the three known isoforms of non-muscle class 2 myosins (NM2A, NM2B and NM2C) and their closest relative Myo18a.

In order to understand the functional relevance of contrasting expression profiles of transcripts encoding the four phylogenetically related NM2A, NM2B, NM2C and Myo18a proteins in developing OLs (Peckham, 2011), we conducted a careful re-evaluation of their structural protein domain organization. Combining the information available in UniProt (https://www.uniprot.org) and Pfam databases (https://pfam.xfam.org), we retrieved the data on the conserved domains and functional sites of NM2A, NM2B, NM2C and Myo18a (Fig. 2A), and confirmed that overall there are no significant differences in the domain organization of the three NM2 isoforms. When compared with Myo18a, the tail domain of class 2 and 18 myosins have a similar α-helical coiled-coil region of comparable size, one IQ calmodulin-binding site that binds the ELC and RLC in the neck domain, and one myosin-N-terminal SH3-like region in the motor domain. However, in the motor domain, Myo18a has three other regions that clearly distinguishes it from the three NM2 isoforms. First, Myo18a lacks the canonical 23-amino-acid sequence actin-binding region in the myosin head domain. Instead, Myo18a has a 5-amino-acid sequence actin-binding region outside the myosin head domain that is ATP independent. Also, upstream in its N-terminus, Myo18a has a KE-rich region. Finally, downstream from the KE-rich region, this myosin has a PDZ domain, which is known to be involved in protein–protein interactions (Furusawa et al., 2000; Isogawa et al., 2005).

In addition to mRNA expression profiles and protein structural domains, the different myosin functions can also be explained by mechanochemical properties including: (1) the enzymatic activity of the motor domains that enables its processivity along the actin filaments, (2) the isoform-specific heavy chain phosphorylation by upstream regulators, and (3) interactions with proteins that may regulate myosin filament assembly and its activity upon actin binding (Clark et al., 2007, 2008; Tan et al., 2008; Heissler and Sellers, 2016; Melli et al., 2018). We have summarized what is currently known about the bipolar myosin filaments structure and the mechanochemical properties of these four myosins (Fig. 2B) (Billington et al., 2013, 2015). All form homotypic antiparallel bipolar filaments, but NM2s have longer coiled-coil tails compared to those of Myo18a and form longer myosin filaments. Additionally, there are other relevant differences in filament structure, enzymatic properties and actin binding behavior among the three NM2 isoforms. Specifically, NM2A and NM2B share more common features when compared with NM2C and Myo18a: NM2A and NM2B have higher ATPase motor activity (0.41 s⁻¹, 0.23 s⁻¹ and 0.13 s⁻¹ for NM2A, NM2B and NM2C, respectively), larger filament width (11.2 nm, 11.5 nm and 7.9 nm for NM2A, NM2B and NM2C, respectively) (Billington et al., 2013) and contain higher number of myosin molecules per filament (29, 30 and 14 for NM2A, NM2B and NM2C, respectively) (Billington et al., 2013). By contrast, NM2C has a longer bare zone (M-region) (167 nm, 166 nm and 219 nm for NM2A, NM2B and NM2C, respectively) (Billington et al., 2013). This suggests that NM2C may form bipolar filaments with fewer component molecules. Additionally, NM2C has been previously described to differ from NM2A and NM2B in its binding capacity to actin (Fig. 2C) (Billington et al., 2013). The percentage of NM2C bipolar filaments that are not bound to actin is much higher than those of NM2A and NM2B, and the actin-bound NM2C structures are mainly characterized by a single actin filament attached to the myosin filament via one side only, while a higher proportion of NM2A and NM2B filaments associate more frequently to assembled actin and, most importantly, with more stable engagements (Fig. 2C). Altogether, the structure and mechanochemical properties of NM2A and NM2B are very similar but differ significantly from those of NM2C and Myo18a. The higher ATPase activity and number of myosin heads in NM2A and NM2B filaments, along with more stable engagements with actin determine their processivity along actin filaments (Stam et al., 2015; Melli et al., 2018), suggesting that they may be more capable of generating and maintaining intracellular tension and force output than NM2C and Myo18a.

Previous studies have shown that the myosins NM2A, NM2B, NM2C and Myo18a bind to actin and are able to oligomerize and form antiparallel bipolar myosin filaments. Of note, in vivo Myo18a does not form homotypic filaments, but rather polymerizes with NM2C-based molecules (Billington et al., 2013, 2015). The structural, enzymatic and mechanochemical properties of myosins described above suggest that filaments made out of NM2C and Myh18 are not well suited for processivity along actin filaments and, therefore, may not be able to generate contractile forces as efficiently as NM2A- and NM2B-based filaments. In order to explore this idea, we performed a stringent in silico interactome analysis of NM2A, NM2B, NM2C and Myh18a. To gain confidence that a particular physical interaction may exist in the physiological context, we used the EMBL-EBI IntAct database of molecular interactions (https://www.ebi.ac.uk/intact; Orchard et al., 2014) that collects experimental molecular interaction data curated from peer-reviewed scientific publications or directly submitted by the scientific community. Additionally, we used the web service Interactome3D to explore the structural details of protein-protein interaction networks (https://interactome3d.irbbarcelona.org; Mosca et al., 2013), which uses IntAct as one of the source databases. Of note, in order to decrease the degree of artefactual interactions, our analysis did not consider in silico predictions of physical protein interactions. Our analysis resulted in the identification of 20, 23, 10 and 11 physical interactors for NM2A, NM2B, NM2C and Myo18a, respectively (Table S1). The significant higher number of interactors for NM2A and NM2B in comparison with NM2C and Myo18a is a reflection of their extensive characterization. Next, we analyzed the expression profile of these interactors in the OL lineage using the Brain-RNA transcriptome database (http://www.brainrnaseq.org), focusing only on those expressed during OL differentiation (Fig. 3A). We identified 17, 18, 4 and 7 interactors for NM2A, NM2B, NM2C and Myo18a, respectively, that were expressed in the OL lineage. Interestingly, we found that the expression of the vast majority of these myosin interactors (71% for NM2A, 88% for NM2B, 50% for NM2C and 71% for Myo18a) decreases during OL differentiation. To further understand the global framework of the myosin interactomes, we performed a functional enrichment heatmap analysis of the interactors using the NIH DAVID bioinformatics functional gene ontology (GO) annotation tool (https://david.ncifcrf.gov; Fig. 3B; Tables S2–S6). Since the expression of NM2A and NM2B decreases during OL differentiation, while NM2C and Myo18a expression increases, and it is possible that they may form heterofilaments in vivo as previously described (Billington et al., 2015), we pooled the interactomes of NM2A and NM2B (30 interactors) and compared it with those of NM2C and Myo18a (11 interactors). The GO functional enrichment analysis of these two groups revealed interesting and contrasting findings. Globally, the GO categories for the 30 interactors of NM2A and NM2B suggest that these proteins participate in signaling pathways that link the extracellular environment to the actin cytoskeleton, where NM2A and NM2B may be the downstream effectors of such dynamics. Interesting examples are the small GTPase Rac1 and one of its effectors PAK2, important modulators of cytoskeleton dynamics. In particular, we have previously shown a relevant role of Rac1 in OL myelination (Thurnherr et al., 2006). Previously recognized regulators of NM2 activity, TRPM7 (Clark et al., 2008) and SVIL (Hasegawa et al., 2013), are common interactors of NM2A and NM2B. By contrast, the pooled interactomes of NM2C and Myo18a are enriched in proteins which are expressed mainly in adult brain and converge into functions implicated in cellular trafficking, including transport, secretion and establishment of localization. Some examples include GOLPH3 (Dippold et al., 2009) and Snapin (Tian, 2005). These findings are in agreement with the late upregulation of NM2C and Myo18a in mature OLs, where they may be responsible for regulating myelin production and membrane expansion and deformation.

To further characterize the spatial and temporal localization of NM2B and NM2C during development, we carried out immunocytochemistry analysis of OPC and OL cultures. Fig. 4A shows the localization of NM2B and NM2C in the OPC soma and the cytoplasm of nascent protrusions. Although both motor proteins are found in the same compartments at this stage, the signal for NM2C extends further distally into the growing processes than that of NM2B. Preliminary time-lapse imaging analysis of NM2B–GFP shows that, at initial differentiation stages, a local decrease of NM2B–GFP levels is observed, which leads to a local decrease in cortex tension favoring branching protrusion. As protrusion emerges, localization of filamentous actin becomes confined to the leading tip and the NM2B released from the cortex is redirected to the tip, as shown in Fig. S2. This localization suggests that unidirectional actin polarization/depolarization dynamics and actin crosslinking by NM2B potentially provides guidance for initial protrusion growth and stability. As OPCs further branch and mature into OLs (Fig. 4B), the differential local distribution of these proteins becomes more apparent, with NM2B mostly localized in the cortical cytoplasm of thicker branches, while NM2C further extends into actin protrusions active in membrane extension. Interestingly, this pattern of NM2C staining is reminiscent of the recently described Golgi outpost compartment, which is characterized as being enriched in microtubules and Myo18a, and is critical for myelination (Fu et al., 2019). Indeed, staining with NM2C, Myo18a and tubulin corroborates that these proteins are found in these outposts in mature OLs (Fig. 4C).

Collectively, the pattern of localization and levels of expression of NM2B, NM2C and Myo18a suggest a model whereby at initial differentiation stages, a local decrease of NM2B leads to local decrease in cortex tension favoring branching protrusion (Fig. 4D). While the global reduction of NM2B leads to cortical tension release, which is a requirement for branching initiation, the speed of protrusion growth and myelin sheet formation is yet dependent on the elasticity of the plasma membrane and the robustness of mechanisms of membrane targeting. The mechanochemical properties and interactomes of NM2C and NM18a suggests their putative involvement in processes of myelin membrane extension and MBP production.

The differential expression and distribution of NM2B and NM2C in OPC and OL suggest that these two motors might carry out different functions during development. We hypothesize that protrusion growth and local myelin extension and wrapping requires the coordination between plasma membrane elasticity and sustained protein targeting. We have previously reported that global reduction of NM2A and NM2B leads to cortical tension release, and increased OL branching and myelination, while their overexpression inhibits these processes (Wang et al., 2008, 2012). To test whether NM2B and NM2C are involved in different steps of OPC development (branching initiation versus membrane extension), we performed NM2B and NM2C silencing in myelinating co-cultures of OPCs with DRG neurons and examined the effects on myelin formation with shRNA (denoted shNM2B and shNM2C, respectively) (Fig. 5A). Western blots demonstrated the effective knockdown of these proteins in culture (Fig. S3A). Quantitative imaging analysis confirmed that knockdown of NM2B in OPC resulted in a 2-fold increase (P<0.0001) in the average number of MBP-positive segments formed in cultures compared to control (Fig. 5B). By contrast, knockdown of NM2C caused a 36% reduction in myelin formation (P=0.02) compared to control cultures (Fig. 5B). We found that MBP-positive segments formed in shNM2C-treated cultures appeared abnormal compared to controls and NM2B knockdown cultures (Fig. 5A). In addition, the relative proportion of cultures showing no myelination or less than 40 segments per field (Fig. 5C) was higher in shNM2C (59%) compared to shCtrl (40%) and shNM2B (12%) cultures. By contrast, knockdown of NM2B increases the proportion of fields showing more than 100 myelin segments per field (47%) compared to shCtrl (12%) and shNM2C (6%) (Fig. 5C). Collectively, our results suggest that spatiotemporal fluctuations of NM2B and NM2C control local cortical tension and membrane expansion for initial protrusion growth (NM2B) and subsequent myelin formation (NM2C).

In this study, we have performed an integrative analysis, at the transcriptomic and proteomic level, of all myosin in CNS myelinating glial cells. This work revealed the presence of many myosins in the OL lineage, with some of them not yet functionally characterized. Some of these myosins are expressed in a developmentally coordinated manner, suggesting that they may play distinctive roles during OL differentiation. Here, we highlight seven myosins that have been consistently identified in several independent studies: the three NM2 isoforms and the phylogenetically related Myo18a, and the transport-associated Myo1d, Myo5a and Myo6. While the abundance of Myh9, Myh10 and Myo5a transcripts decreases during differentiation, that of Myo1d, Myh14, Myo6 and Myo18a transcripts increases. Of these, to our knowledge, only a function of NM2C and Myo18a had not previously been described in OL biology.

Our systems biology-oriented analysis revealed members of the NM2 family working as distinctive mechanical regulators of OL cell morphology and differentiation, as proposed in our conceptual model (Fig. 4D). Based on the changes in mRNA and protein expression level of NM2B and NM2C, we propose three stages of the OL development: stage 1 (NM2B is high and NM2C is low), stage 2 (NM2B is reduced and NM2C is elevated), and stage 3 (NM2B is low and NM2C is high). At stage 1, the enrichment of NM2B in the early stages of OL development may contribute to the generation of actomyosin crosslinkage and contractility-based tension. While actin filaments polymerization drives the elongation of OL protrusions (Nawaz et al., 2015; Zuchero et al., 2015) and myosin contractility is released at the protruding edge, NM2B-dependent actomyosin crosslinkage and contractility determines the overall shape of the cell. A non-uniform pattern of NM2B expression at the cortex supports branching initiation, which possibly takes place in regions of decreasing mechanical tension. It can be noted here that the ability of the cortex to elongate (commonly termed as elasticity) is an inverse function of the local mechanical tension. This idea is supported by the observation that NM2B overexpression inhibits OPC branching due to increased cortical tension (Wang et al., 2012). Preliminary data, looking at the local changes in the distribution of cortical NM2B–GFP as OPCs initiate branching (Fig. S2), shows that, as protrusions emerge, localization of filamentous actin becomes confined to the leading tip and the NM2B released from the cortex is redirected to the tip. This redistribution is also in agreement with our previous electron microscopy and immunofluorescence analyses of OPC protrusions (Wang et al., 2012). Subsequently, at stage 2 the global reduction of cortical NM2B expression promotes further branching and protrusion, while relocalization of this motor can be reused for scaffolding of the growing protrusions. Note that at this stage there is no requirement for membrane spreading. The increase in the number of branches is accompanied by the continuous decrease in the NM2B expression, which is not sufficient to promote myelin wrapping as the latter requires increased transport of myelin proteins to the growing myelin sheet. Finally, at stage 3 OL protrusions evolve into myelin-bearing sheets showing low levels of NM2B and higher expression of NM2C. The low expression of NM2B allows further reduction of the cortex tension, which may promote the expansion of the myelin sheet. This condition is necessary but not sufficient, as the cell has to maintain an increased membrane transport to the growing sheet and it is likely secured by the increased expression of NM2C and Myo18a, which are implicated in membrane-bound cargo and trafficking. In agreement with this interpretation, we also found that NM2C and Myo18a colocalized in oligodendrocyte protrusions that are rich in tubulin, and that knockdown of NM2C negatively impacted myelin formation in vitro. Furthermore, Myo18a is also found in Golgi outposts, a recently described domain that is needed for proper myelin formation (Fu et al., 2019), and NM2C protein levels are significantly enriched in white matter regions, such as corpus callosum and pons of the adult CNS (Golomb et al., 2004), while NM2B levels are reduced (Cahoy et al., 2008; Wang et al., 2008; Zhang et al., 2014; Marques et al., 2016).

The integrative analysis of published data and our own results support the emerging idea that the intracellular tension of a developing OL is modulated by varying the levels of the NM2 isoforms in the actomyosin cytoskeleton. Recently, two studies proposed that actin dynamics is the driving force for myelination of axons (Nawaz et al., 2015; Zuchero et al., 2015) and that actin, but not actomyosin modulators, influence membrane tension (Nawaz et al., 2015). However, these results were based on measurements in cells treated with blebbistatin, a drug that does not act exclusively on NM2B but may also inhibit NM2C activity (Rauscher et al., 2018; Zhang et al., 2017; Chinowsky et al., 2020 preprint).

The functional enrichment analysis of the pooled interactors of NM2A and NM2B versus NM2C and Myh18a clearly suggest that NM2A and NM2B may have similar functions during OL differentiation. By virtue of their biophysical properties, NM2A and NM2B effectively generate intracellular cell tension and control cell shape to enable the synchronization of various genetic, metabolic and biochemical processes leading to myelin biogenesis. This goes in agreement with the concept that ablation of both NM2A and B in vitro accelerates OL differentiation, while their overexpression prevents OL branching and differentiation (Wang et al., 2008, 2012). Although NM2B OL-specific ablation in adult brain also accelerates myelin repair (Rusielewicz et al., 2014), early deletion may have adverse consequences for developmental myelination. Accordingly, it would be relevant to address in the future the phenotype of OPC-specific NM2B ablation in early post-natal brain prior to active myelination. In contrast, NM2C and Myo18a are upregulated later in development (Cahoy et al., 2008; Zhang et al., 2014), at a stage of active elaboration of myelin, suggesting a potential role for NM2C and Myo18a in myelin membrane production. In agreement with this hypothesis, some of the identified interactors of NM2C and Myo18a are involved in cell trafficking and membrane homeostasis. Myelin membrane trafficking involves the assembly of particular components in a spatiotemporal-regulated manner and requires specific sorting and transport mechanisms for delivering myelin components to the sites of membrane growth (Baron and Hoekstra, 2010; Simons et al., 2012). Here, we identified Snapin as one of the potential interactors of NM2C. In chromaffin cells, Snapin is important for Ca²⁺-dependent exocytosis of large vesicles (Tian, 2005). Although it will be necessary to confirm in the future that NM2C and snapin interact physically in OLs, this finding further supports the hypothesis that NM2C might be involved in mechanisms of membrane trafficking.

It has also been described that NM2 isoforms co-polymerize, both in vitro and in vivo, to form heterotypic bipolar filaments with hybrid characteristics (Beach et al., 2014). This suggests that the OL during differentiation may fine-tune a broad spectrum of mechanoenzymatic properties with different combinations of myosin heavy chain heterodimers. Additionally, all NM2 isoforms establish heterotypic filaments with Myo18a, which decreases their myosin filament size (Billington et al., 2015). In vitro, addition of Myo18a molecules to NM2 filaments interferes with homotypic filament assembly (Billington et al., 2015). Previous studies also demonstrate that NM2A and NM2B filaments have distinct mechanical properties. NM2A holds a higher ATP hydrolysis rate, propelling actin filaments faster, whereas NM2B can exert tension for extended time periods (Vicente-Manzanares et al., 2009; Melli et al., 2018). OLs may therefore adjust the intracellular force output of NM2 filaments for specific functions by altering their composition and/or stoichiometry of their isoforms, possibly together with Myo18a. Future experiments will address the role of this phenomenon on OL differentiation directly, as we have observed that knockdown of Myo18a in purified OLs also decreased the levels of NM2C (data not shown), further suggesting that formation of these heterodimers among myosin motors might add another level of regulation to the cytoskeletal dynamics of myelination.

Intracellular tension can also be modulated based on the ability of class II myosins to induce actin depolymerization. In budding yeast cytokinesis, for example, the major contractile forces results from actin depolymerization mediated by NM2 motor activity and cofilin (Mendes Pinto et al., 2012). In OLs, actin depolymerization is dependent on cofilin/ADF activity (Nawaz et al., 2015). In this context, it is possible that NM2C and Myo18a, which are present in higher abundances in myelinating OL than NM2A and NM2B, are inducing actin depolymerization together with cofilin proteins. Another study proved that active class II myosin motors can quickly ‘fluidize’ entangled actin networks, decreasing the resistance to cellular deformations (Humphrey et al., 2002). In agreement with this model, we have previously shown that F-actin and phosphorylated MLC2, a read out of active myosin motors, colocalize at the edges of the MBP-positive membranes extended by mature oligodendrocytes in culture (Wang et al., 2008, 2012). This finding indicates that, despite the overall reduction in cortical tension and actomyosin contractily as OLs differentiate, localized activity of myosin motors and actin polymerization/depolymerization act in concert to promote myelin extension.

Overall, this work suggests that intracellular tension generation in a developing OL is crucial for the definition of cell shape and polarization, and this may be acquired by fine-tuning the proportion of NM2 isoforms and the phylogenetically related Myo18a, which associated with actin filaments at each developmental stage. This mechanical output, together with the regulation of actin assembly, will be crucial for the ability of OLs to wrap their myelin sheets around axons. Several examples in other cell and tissue models have nicely illustrated the panoply of possibilities by which different patterns of actomyosin contractility and resulting tension can mold cell and tissue shapes (Heer and Martin, 2017). In the particular case of how developing OL regulate their growth and form, the knowledge of the active molecular mechanisms underlying the resulting forces surely needs to be further characterized in the future.

The mRNA and protein expression profile of the myosin network during oligodendrocyte differentiation was cataloged by integrating data from available transcriptomic and proteomic datasets (Cahoy et al., 2008; De Monasterio-Schrader et al., 2012; Zhang et al., 2014; Thakurela et al., 2016; Azevedo et al., 2018) with previous developmental studies of myosin expression (Wang et al., 2012; Yamazaki et al., 2016) during in vitro, ex vivo and in vivo human, mouse and rat developmental oligodendrocyte differentiation. The results from high throughput data presented in the final table derive from the bioinformatics analysis of the original authors.

The information on domain organization and domain boundaries for the selected myosins was retrieved from the Pfam database (https://pfam.xfam.org) and combined with the functional annotation of specific profiles and patterns as described in UniProtKB (https://www.uniprot.org/uniprot).

The EMBL-EBI IntAct (https://www.ebi.ac.uk/intact; Orchard et al., 2014) and Interactome3D (https://interactome3d.irbbarcelona.org; Mosca et al., 2013) databases were used to identify direct physical interactors and to explore the structural details of protein–protein interaction networks of NM2A, NM2B, NM2C and Myh18a. The identified interactors were inspected for the temporal expression in the OL lineage as reported in the Brain RNASeq (http://www.brainrnaseq.org; Zhang et al., 2014) and only those were considered for subsequent analysis.

Gene-enrichment and functional annotation analysis was performed to understand the global framework of the NM2A and NM2B versus the NM2C and Myo18a interactomes using the NIH DAVID bioinformatics functional gene ontology (GO) annotation tool (https://david.ncifcrf.gov). This enrichment analytic algorithm provided a batch annotation and gene-GO term enrichment analysis to highlight the most relevant gene GO terms, tissue expressions, protein functional domains and bio-pathways associated with a given gene list. The results are displayed in an individual annotation table (Tables S2–S6). We also used the DAVID Functional Annotation Clustering tool to measure the relationships among the annotation terms based on the levels of their co-association genes (Tables S2–S6).

Vertebrate animal tissue was collected in accordance with the guidelines published in the NIH Guide for the Care and Use of Laboratory Animals for the humane treatment of laboratory animals (Publication No. 85-23, revised 1985). All procedures were approved by Hunter College Institutional Animal Care and Use Committee.

Primary OPCs were purified by immunopanning from mixed glial cultures of postnatal day 1 rat (Sprague Dawley) cerebral cortices as previously described (Wang et al., 2008, 2012). Purified OPCs were seeded onto poly-lysine-coated glass coverslips (20,000–30,000 cells) and maintained in Sato medium (DMEM, 100 μg/ml transferrin, 100 μg/ml BSA, 20 nM progesterone, 100 μM putrescine, 30 nM sodium selenite, 2 mM L-glutamine, 5 μg/ml insulin, 60 μg/ml N-acetyl cysteine and 10 μM forskolin) with 10 ng/ml PDGF and 10 ng/ml bFGF for proliferation or induced to differentiate in Sato medium containing T3 (30 ng/ml). DRG explants were isolated from rat embryonic day 16 spinal cords and plated directly onto glass coverslips coated with Matrigel (BD Biosciences). Explants were kept in NB medium (NB: neurobasal medium, containing 2% B27 supplement, 4 g/l D-glucose, 2 mM L-glutamine and 50 ng/ml NGF) and treated with antimitotic agents to eliminate non-neuronal cells. Explants were allowed to extend a dense neurite network for at least 2–3 weeks before their use in co-culture experiments. OPC–DRG co-cultures were established by seeding freshly purified OPCs (50,000–100,000 cells) onto DRG explants and myelination was induced the following day by addition of 1 μg/ml TrkA-Fc (R&D Systems). Cultures were allowed to myelinate for up to 18–21 days, with fresh medium provided every 2–3 days.

Knockdown of NM2B and NM2C in myelinating OPC–DRG co-cultures was performed using specific shRNAs as previously described (Wang et al., 2008, 2012). Briefly, MISSION^® shRNA Lentiviral Transduction Particles were purchased from Sigma. Lentivirus carrying two individual clones (TRCN0000110555 and TRCN0000110558 for NM2B, and TRCN0000110570 and TRCN0000110573 for NM2C) were transduced into OPCs in co-culture with DRG neurons 24 h after seeding. Lentiviral particles were left in culture for 3 days, prior to medium change. In our experience, this approach allows for selective transduction of OPCs over post-mitotic mature neurons and improved OPC survival (Wang et al., 2008, 2012). Co-cultures were maintained in myelinating medium for an additional 18–21 days, prior to fixation and analysis by immunofluorescence. MISSION^® pLKO.1-puro non-mammalian shRNA targeting particles were used as controls.

Cell cultures were fixed in 4% PFA and processed for immunocytochemistry as previously described (Wang et al., 2008, 2012). Antibodies used in these studies included: rabbit anti-NM2A, anti-NM2B and anti-NM2C (Cell Signaling, #3403, 3404 and 3405, at 1:200), mouse anti-NM2B (DS Hybridoma Bank, #CMII-25, at 1:100), mouse anti-NM2C (Proteintech, #66825-I-Ig, at 1:4000), rabbit anti-Myo18a (Proteintech, #14611-1-AP, at 1:100), chicken anti-tubulin (Novus, #NB-100-1612, at 1:100), mouse anti-MBP (Sternberger Monoclonal, #SMI94, at 1:200), chicken anti-neurofilament (NF) (Biolegend, #822701, at 1:2000) and rabbit-anti-Olig2 (Abcam, #33427, at 1:1500). Actin-670 (Cytoskeleton, #PHDN1, at 1:100) and secondary antibodies conjugated to Alexa Fluor 647, Rhodamine, Fluorescein, Coumarin or Cyanin 5 were obtained from Jackson Immunoresearch (West Grove, PA) and used at 1:100 dilution. Myelinating co-cultures were examined by epifluorescence on a Leica DMI4000 microscope with a HCX PL FLUOTAR L 20.0×/0.40 NA DRY objective and LAS 1.7.0 software equipped with a digital camera (DFC350FX; Leica). For OPC and OL cultures, images were acquired with a confocal laser scanning microscope (LSM 510; Carl Zeiss, Inc.) using Plan-Apochromat 20×/0.75 NA or Neofluor 40×/1.3 NA oil objectives and LSM software (Carl Zeiss, Inc.).

Lysates from OPC and OL cultures were prepared as previously described (Wang et al., 2008), subjected to SDS-PAGE and blotted onto nitrocellulose membranes. Appropriate regions of the blots were cut and incubated with specific antibodies and developed using chemiluminescent substrate (Pierce, Rockford, IL). For estimation of changes in protein levels, X-ray films from two or three independent experiments were scanned, and the relative intensity of each protein band was calculated in ImageJ by dividing the absolute intensity of each protein band (the area of the band by the number of pixels contained in that area) by the absolute intensity of the corresponding actin band.

Lysates of rat brains were prepared as previously described (Wang et al., 2008). Briefly, whole-brain homogenates were lysed in a solution containing 1% SDS, 150 mm NaCl, 10 mm EDTA, 1 mm PMSF, 10 μg/ml aprotinin, and 20 μm leupeptin in 50 mm Tris-HCl pH 7.4. Lysates were cleared by centrifugation at 12,000 g for 10 min. Protein concentrations were determined using the BCA method (Pierce, Rockford, IL). The lysates (10–15 µg total protein per sample) were then subjected to SDS-PAGE and blotted onto nitrocellulose membranes. Appropriate regions of the blots were cut and incubated with specific antibodies and developed using chemiluminescent substrate (Pierce, Rockford, IL). For estimation of changes in protein levels, X-ray films from two or three independent experiments were scanned, and the relative intensity of each protein band was calculated in ImageJ by dividing the absolute intensity of each protein band (the area of the band by the number of pixels contained in that area) by the absolute intensity of the corresponding actin band.

Image analysis and processing was performed using the freely available NIH ImageJ software, version 2.0.0, and Photoshop CS6 (Adobe). Adjustment of image brightness or contrast was performed in some cases but without misrepresenting data. OPC–DRG myelinating co-cultures were stained with MBP, Olig2 and NF antibodies, and images acquired at low magnification (25×). MBP-positive segments aligned with the NF-positive axon were identified, and counted in micrographs from 10–12 random fields per coverslip (three or four coverslips per condition, per experiment; total of three experiments). Statistical analyses (one-way ANOVA and multiple comparison post-tests) were performed using GraphPad Prism 6 software. For western blots, analyses of band intensity differences between PDGF and 3dT3 conditions were performed using GraphPad Prism 6 software (Krustal–Wallis non-parametric test, followed by Dunn's multiple comparison).

Mouse OPCs were prepared as previously described (O'Meara et al., 2011). Briefly, mixed glial cultures were prepared from postnatal day 2 mouse cerebral cortices of transgenic mice (MMRRC_034321-UNC) expressing NM2B–GFP. Cultures were used to generate OPC-enriched glial cultures by separating OPCs from the astrocyte monolayer by orbital shaking, followed by purification through differential adhesion to plastic. Purified OPCs (25,000–30,000 cells) were seeded on poly-lysine/laminin-coated coverslips and kept in proliferating medium containing PDGF and bFGF (10 ng/ml, Peprotech) for the duration of the experiment. Imaging was performed at 37°C on a LeicaDMI-4000B inverted microscope (Leica Microsystems, Buffalo Groove, IL) mounted on a TMC isolation platform (Technical Manufacturing Corporation, Peabody, MA), with a Yokogawa CSU 10 spinning disc head and Hamamatsu C9100–13 ImagEM EMCCD camera. The images were acquired as 16-bit data files, with a 63×/1.4 NA oil objective (0.24 μm/pixel), using Volocity acquisition (PerkinElmer, Waltham, MA) and processed with ImageJ.

The authors acknowledge Inês Mendes Pinto lab members for critical reading of the manuscript, Sandra Maya for the artwork and Zoe Pian for assistance with data acquisition.