In vitro myelin formation using embryonic stem cells

Myelination of axons by oligodendrocytes in the central nervous system (CNS) and by Schwann cells (SCs) in the peripheral nervous system is crucial for neuronal function and survival. In the absence of myelin, not only is nerve impulse transduction slower but also axons are vulnerable to atrophy and functional deficits due to a lack of trophic support from oligodendrocytes (Nave, 2010). Dysmyelination or demyelination (improper development or loss of myelin, respectively) occurs in many neurological disorders, such as multiple sclerosis (MS), Pelizaeus–Merzbacher disease (PMD) and other leukodystrophies, and as a consequence of spinal cord injury, leading to disruption of electrical impulse conductivity, atrophy of neurons and permanent functional deficits (Miron et al., 2011; Nave, 2010; Peru et al., 2008; Siegel, 2006). Additionally, these symptoms are aggravated by inadequate myelin repair. Therefore, transplantation of exogenous cells and mobilization and recruitment of endogenous cells have been proposed as therapeutic approaches to restore myelin (Cummings et al., 2005; Gupta et al., 2012; Miron et al., 2011; Peru et al., 2008; Potter et al., 2011; Tsuji et al., 2010). For successful outcomes, it is essential to establish and maintain both stimulating and permissive cues for myelination. Such environments can be uncovered through a better understanding of basic oligodendrocyte and myelin biology and by recognizing the molecules that control the myelination process. Consequently, the development of a robust, reliable, easy to modify and quantitative myelination assay is crucial for a deeper study of myelination and to refine or discover therapies for remyelination.

Currently, the myelination process has mostly been studied in vitro with isolated primary cells and in postmortem tissues or in vivo with animal models (Chen et al., 2010; Dean et al., 2011; Pouya et al., 2011; Watkins et al., 2008). Most of these approaches are restricted by the limited number of cells and animals available and by variability among individual animals, making large-scale experiments difficult. Furthermore, owing to light scattering of the tissues and restrictions on the time of imaging in live animals, observing myelin formation in vivo is not currently practical. Efforts to uncouple axons and oligodendrocytes by co-culturing rat-derived primary oligodendrocytes with nanofibers have been made (Lee et al., 2012), but this system has not yet been used with other sources of oligodendrocytes or to study the dynamic process of myelination. By contrast, stem cell technology offers many benefits to overcome the limitations of existing assays. Embryonic stem cells (ESCs) can be readily expanded to a relatively uniform population comprising a large number of cells. The uniformity of cells minimizes experiment-to-experiment differences and studies can be as large or as small as required because ESCs can be readily stored, expanded and transferred between facilities. However, achieving consistent in vitro myelination with these cells has been challenging (Brüstle et al., 1999; Jiang et al., 2010; Liu et al., 2000). To develop a consistent and robust in vitro myelin formation assay using ESCs, we developed a protocol to differentiate mouse ESCs into myelinating oligodendrocytes. Subsequently, we adapted an existing protocol to generate cortical neurons from mouse ESCs (Gaspard et al., 2009) and established myelin-forming co-cultures. We confirmed functional interaction between neurons and oligodendrocytes and the existence of compact myelin by immunostaining with markers, such as myelin basic protein (MBP) and contactin-associated protein (CASPR; CNTNAP1 – Mouse Genome Informatics), and by transmission electron microscopy (TEM) analysis.

When our co-culture system was grown on conventional cell culture dishes, the effective evaluation of myelin formation was difficult due to the presence of neuronal cell bodies and the random distribution of neurites and myelin within the tissue culture plate. To overcome these limitations, we combined our myelinating co-culture system with microfluidic technology (Taylor et al., 2005). The compartmentalization of oligodendrocytes into a defined region with axons and away from neuronal cell bodies allowed for the direct study of axon-oligodendroglia interactions and myelin formation in the ‘myelination compartment’. With the ability to observe the entire experimental area, we developed a custom semi-automated computer platform, termed computer-assisted evaluation of myelin formation (CEM), to quantify myelin formation. Using CEM, we were able to identify and quantify myelin formation events over a large area with minimal human intervention. In addition to these strengths, our assay is especially suitable for long-term live imaging of the myelin formation process. Longitudinal live imaging allowed us to track oligodendrocytes as they ensheathed and wrapped axons over several days. Our observations revealed that oligodendrocyte processes anchor to axons before wrapping them. By combining live imaging and confocal analysis with previously published data, we propose a new model for myelin formation, in which oligodendrocyte processes survey their environment, anchor to the bare axon and wrap around the axon before forming overlapping sheets of myelin.

To achieve myelination for our study, a consistent source of myelination-competent oligodendrocytes was required. Previously published protocols yielded variable ratios of oligodendrocyte progenitor cells (OPCs) that produced myelin when transplanted into animals; however, in vitro, these protocols either failed to produce myelin, were inefficient, or took several months (Brüstle et al., 1999; Jiang et al., 2010; Liu et al., 2000; Numasawa-Kuroiwa et al., 2014). Therefore, we developed a method to differentiate oligodendrocytes from mouse ESCs. We chose a serum-free approach to minimize the variability that is inherent in different batches of serum and to produce a well-defined differentiation cocktail.

First, we generated neural progenitor cells (NPCs) from mouse ESCs (Fig. 1). NPCs can be expanded as needed or stored for further use. OPCs were induced from NPCs using differentiation medium (Fig. 1; see Materials and Methods for media formulations). At the end of 1 week [day (D) 8], ∼70% of the population expressed markers for OPCs and/or oligodendrocytes (Fig. 2A,B; supplementary material Table S1). In the second week, we switched to maturation medium to promote the maturation of OPCs (Fig. 1). By D15, mature oligodendrocytes were generated expressing O4 and MBP and showing mature oligodendrocyte morphology, i.e. extended membrane sheaths and highly branched processes (Fig. 2A; supplementary material Table S1). Only a small percentage of the population consisted of astrocytes (5.08%) or neurons (8.38%), highlighting the specificity of our protocol (Fig. 2B; supplementary material Table S1).

Moreover, we confirmed the enrichment of oligodendrocytes by comparing the expression levels of several genes from our oligodendrocyte cultures with undifferentiated NPCs by quantitative PCR (qPCR) on D8 and D15 (Fig. 2C). Expression levels of all tested OPC and oligodendrocyte marker genes were increased several fold over those of NPCs by the first week (Fig. 2C). On D15, genes expressed in mature oligodendrocytes, such as those encoding MBP, proteolipid protein 1 (PLP) and myelin oligodendrocyte glycoprotein (MOG), were increased several thousand fold (Fig. 2C), which is indicative of a well-defined and highly efficient oligodendrocyte generation protocol that yielded mature oligodendrocytes in vitro from mouse ESCs.

SCs originate from neural crest cells and myelinate peripheral nerves (Jessen and Mirsky, 2005). To investigate whether SCs were generated alongside oligodendrocytes, we stained cultures for the SC-specific markers myelin protein zero (P0) and peripheral myelin protein 22 (PMP22). No cells positive for either marker were observed (data not shown). In addition, the expression levels of genes encoding P0, PMP22 and periaxin did not show any significant change in D15 oligodendrocyte cultures compared with undifferentiated NPCs by qPCR (supplementary material Fig. S1). We concluded that there was no SC contamination in cultures.

To mimic myelination in the CNS, we established an in vitro myelin formation assay in which both neurons and oligodendrocytes were derived from ESCs. To generate neurons, a previously published protocol that enriches for cortical neurons was used (Gaspard et al., 2009). The cells were plated onto coverslips on D12, as described in the protocol, and neurons were allowed to mature for 10-14 days before adding oligodendrocytes to the neuron culture. Oligodendrocytes that were at the end of the 2-week differentiation protocol (D15) were plated on the neurons, thereby starting the myelin formation co-culture (Fig. 1). The co-cultures were kept in myelination medium to promote myelin formation. Oligodendrocytes formed smooth sheaths around neurites (Fig. 3A) and transverse sections of reconstructed confocal images showed that these tubes fully wrapped the neurites (Fig. 3B), indicating that OPCs generated using our protocol could mature into MBP-expressing oligodendrocytes that ensheathed and wrapped axons in vitro.

Light microscopy can only show if the axons are ensheathed; it cannot confirm whether single or multiple layers of the membrane wrap the axons and if those layers compact to form myelin. To assess the presence of myelin in our co-cultures, we performed TEM analysis and confirmed that the oligodendrocyte membrane wrapped several times around individual axons and formed compact myelin (Fig. 3C). The distance between alternating darker circles (i.e. major dense lines: the electron-dense regions composed of proteins and lipid polar groups) was on average 12.56±0.28 nm (s.d.) (supplementary material Fig. S2), the same as seen in in vivo measurements for mammalian CNS myelin (Baumann and Pham-Dinh, 2001; Quarles, 2002). To confirm that oligodendrocytes, but not SCs, wrapped axons, the co-cultures were stained for P0 and PMP22. No cells positive for either marker were observed, confirming the specificity of co-cultures to mimic CNS myelination (data not shown).

During myelination by oligodendrocytes, CASPR initially clusters at the contact sites between axon and oligodendrocyte processes, then spreads across the entire ensheathed portion of the axon and finally accumulates only at the paranodes (Eisenbach et al., 2009; Pedraza et al., 2009). This premyelination CASPR accumulation does not occur during SC myelination and is characteristic of a reciprocal interaction between oligodendrocytes and neurons. Therefore, we characterized CASPR localization in our co-cultures to assess the functional quality of ESC-derived oligodendrocytes.

CASPR either spread along axons at the regions of contact with oligodendrocytes or densely accumulated at the ends of the region wrapped by oligodendrocytes (Fig. 3D). Thus, similar to primary oligodendrocytes and neurons, ESC-derived oligodendrocytes and neurons interacted reciprocally in myelinating co-cultures. Dual CASPR localization, along both the axons and at the ends of oligodendrocyte wraps, demonstrated that several phases of myelination, including initial contact, ensheathment and compacted myelin, coexisted in our co-cultures. Given that we showed compact myelin formation and that oligodendrocytes interacted with neurons in a functional manner, we concluded that not only did our protocol yield oligodendrocytes that mimicked primary oligodendrocytes but also our co-culture system was representative of biologically relevant myelin formation. Taken together, these results yielded a myelin formation model in an easy-to-manipulate in vitro environment.

On conventional cell culture platforms, neurons grow and extend neurites in random orientations. Consequently, myelin is distributed arbitrarily, and quantifying significant numbers of myelination events requires the monitoring of a large portion of the plate. As a next step to optimizing myelin formation co-culture assays, we sought to adapt the system to a platform in which myelination events could be clustered together. Such clustering into a defined area would improve quantitative measurements and also make it easier to track myelination in real time. Previously, primary neurons were cultured in a compartmentalized manner using microfluidics-based neuron culture devices (Taylor et al., 2005), which allowed for the isolation of axons from cell bodies by using microgrooves that interconnected the two main compartments (supplementary material Fig. S3A).

In the conventional microfluidic device design, the capillary forces inside the enclosed compartment of the neuronal compartment only allow 10-20% of ESC-derived neurons to remain in the compartment, which results in a low density of axons growing into the myelination compartment (supplementary material Fig. S3B). To overcome this limitation of low axon density, we introduced a new design that replaced the enclosed compartment with an open well, which restricted almost all of the neurons to the neuronal compartment due to minimized flow (Fig. 4A,B; supplementary material Fig. S3B).

Similar to the experiments in the standard mixed co-cultures, we plated mouse ESC-derived neurons on D12 into the neuronal compartment. In 10-14 days, neurons extended many axons into the myelination compartment (supplementary material Fig. S3B). At this point, D15 oligodendrocytes were plated into the myelination compartment (Fig. 4B). In 2 weeks, oligodendrocytes ensheathed axons, which appeared as elongated tubes (Fig. 4C). Transverse sections of reconstructed confocal images showed complete wrapping of neurites (Fig. 4D).

Next, we examined whether oligodendrocytes and isolated axons in the microfluidic devices would also interact reciprocally to concentrate CASPR. Consistent with our observation in the standard mixed co-culture, CASPR spread along or concentrated on axons (Fig. 4E), which required interaction between oligodendrocytes and axons, demonstrating the presence of several phases of myelin formation from initial contact to compacted myelin in device co-cultures, similar to standard mixed co-cultures.

In summary, we established an in vitro myelin formation assay through a combinatorial approach to achieve a compartmentalized co-culture of ESC-derived oligodendrocytes and neurons. This platform provides an ideal method for quantifying myelin formation because the experimental compartment is highly enriched in axons and oligodendrocytes, allowing for more accurate quantitation of myelin formation. Moreover, many myelin formation events can be monitored live by focusing on a relatively small area instead of scanning the entire plate.

Although attempts at quantifying myelination events have been made, a number of challenges remain. Counting the number of smooth, tubular-looking segments of oligodendrocyte membrane or manually tracing these segments are two of the main strategies employed (Stancic et al., 2012; Watkins et al., 2008; Yang et al., 2012), but these approaches are often prohibitive when trying to compare several conditions and/or to analyze the entire experimental area due to the large time and manpower requirements. To improve the throughput, automated quantification is the key. Zhang et al. (2011) developed a fluorescence-based method that relies on overlap between mature oligodendrocytes and axons, visualized by MBP and neurofilament, respectively. They showed that a fluorescence-based method could identify factors that affect myelination (Zhang et al., 2011). We took a similar fluorescence-based approach to develop a computer platform (CEM) that could be used broadly. Therefore, CEM was written mainly on ImageJ [see supplementary material Appendix S1 (CEM users guide) and Appendix S2 (CEM package)]. CEM was able to identify and quantify myelin formation within the entire myelination compartment and to detect changes in myelin formation (Fig. 5).

First, the myelination compartment was imaged using confocal microscopy and the entire compartment was reconstructed (∼10×1.5 mm²; Fig. 5A-C; supplementary material Fig. S4A). The algorithm for identification of myelin formation was based on detection of colocalization between neurons and oligodendrocytes after the cell body regions were identified and removed to prevent false positives. To perform calculations, the images were first turned into binary images, i.e. all empty (black) pixels were zero and all pixels with signal were 255 (Fig. 5B,C; supplementary material Fig. S4A). A crucial step in this conversion was to choose a reference slice with a large amount of positive signal and low background noise for optimal thresholding (Padmanabhan et al., 2010). Binary images were further processed to remove cell bodies of oligodendrocytes and neurons, as described in the Materials and Methods (supplementary material Fig. S4B). The overlap between the resulting images was identified as ‘myelin’ that, by the calculation of the algorithm, encompassed ensheathment as well as compact myelin (Fig. 5D; supplementary material Fig. S4D). Finally, the total amount of ‘myelin’ was calculated by counting overlapping pixels and the ratio of myelinated axons was calculated by dividing the former value by the total pixel count of neurons (2.61%, n=11; Fig. 5E).

Some experiments might need to be performed on neurons that are subject to different manipulations or conditions because it is not possible to grow them in separate cultures (for example, shRNA treatment will affect only a subset of neurons). Such a scenario was mimicked by expressing GFP in a subset of neurons in our co-cultures. Using the strategy described above, CEM identified the GFP and nonGFP axons and the myelin formation of each subset (supplementary material Fig. S4C,D). As GFP expression alone does not affect myelination, the ‘myelin’ ratios of each subset were not significantly different (GFP 2.19%, nonGFP 2.78%; P=0.47, n=11; Fig. 5E).

The thyroid hormone triiodothyronine (T3) is known to promote myelination (Barres et al., 1994). ‘Myelin’ with (+) or without (−) T3 supplementation was quantified to evaluate the capability of CEM to assess myelin formation under different manipulations (Fig. 5F, left two columns). To make the comparison clearer, a myelin quotient (MQ) was calculated by normalizing myelin ratios to an average of +T3 ratios, which was assigned a value of 1. −T3 MQ was significantly lower than +T3 MQ, as expected (0.070 and 1, respectively; P=0.007).

To test if the identified ‘myelin’ was a random overlap or had biological significance, the joint probability of a pixel being positive for both MBP and TUJ1 (TUBB3 – Mouse Genome Informatics) due entirely to chance was calculated. When this +T3 chance probability was compared with the experimentally calculated probability of a pixel being identified as myelin, it was significantly different (P=0.0025; supplementary material Fig. S4F). We performed the same analysis for the −T3 condition and the experimental value was again significantly higher than chance (P=0.0079; supplementary material Fig. S4F). Next, to test whether changes in MBP expression alone are sufficient to account for the effects under ±T3, we equalized MBP pixel numbers of +T3 to −T3 and compared overlap probabilities. The two values were significantly different (P=0.0192; supplementary material Fig. S4F), suggesting that a reduction in MBP levels alone did not explain the reduced MQ value.

Another advantage of having a normalized MQ was that we could test CEM against manual quantification. Manually calculated +T3 MQ was again assigned a value of 1. −T3 MQ was 0.059, which is significantly lower than the manually calculated +T3 MQ (P=0.012) but not significantly different from the −T3 MQ calculated by CEM (P=0.833; Fig. 5F).

In conclusion, CEM provided a semi-automated and unbiased quantitative approach for evaluating the biology of myelin formation with statistical authority. It was applied to analyze large data sets and was used to evaluate the effects of genetic factors and/or different treatments within a sample, such as GFP versus nonGFP and ±T3, both necessary precursors to expanding our assay into large screens. Moreover, the resulting output of the algorithm contained values for several parameters relevant to myelin formation, allowing flexibility in the analysis (supplementary material Fig. S4E).

We next tested whether it was possible to improve quantification by adding cell-shape recognition. The images were processed through the filament analysis function of Imaris software (Bitplane). Many weaker-stained processes were omitted by the software and several false-positive filaments were drawn connecting bright points close to each other (data not shown). Moreover, the processing of each image took almost a day and required a computer with a large amount of RAM (>100 gigabytes), making the strategy not easily accessible. Therefore, cell-shape recognition in its current form is not only significantly failure prone but is also time and resource consuming. When better algorithms for cell-shape recognition are developed, CEM can be updated to take advantage of them.

One limitation of all quantification methods that rely on fluorescent colocalization, including CEM, is that it is impossible to distinguish contact on axon/single wrap versus compact myelin. However, given that CEM detected differences in myelination under two conditions comparable to manual quantification and can process several images with minimal human intervention, it is a valuable tool to quantify myelination of large data sets such as drug screens. When necessary, CEM can be followed by a more detailed analysis of selected candidates to quantify compact myelin.

The final structure of myelin – several wraps of oligodendrocyte membrane around an axon – has been well described through electron microscopy (EM) and light microscopy (Knobler et al., 1976; Remahl and Hildebrand, 1990a,b; Siegel, 2006; Snaidero et al., 2014). However, an understanding of the dynamics of wrapping, especially the initial steps, is still incomplete. Serial section EM analysis of the developing CNS showed non-uniformity in the number of layers of the ensheathing oligodendrocyte process along an axon and bidirectional ensheathment within a developing internode, i.e. the inner ensheathing layer became the outer layer (Knobler et al., 1976; Remahl and Hildebrand, 1990a,b). Based on these observations, it has been proposed that the oligodendrocyte sheath around an axon is initially helical and then elongates and finally undergoes extensive remodeling to reach complete wrapping and compaction. Such helical wrapping is also observed by fluorescent labeling of oligodendrocyte membranes and cytoplasm (Ioannidou et al., 2012). Non-uniformity of wrapping along an internode was also reported in a recent study that combined live imaging and serial block-face imaging (Snaidero et al., 2014).

As a consequence, most of the existing models of myelination predict that the front edge of the oligodendrocyte process continuously grows while wrapping the axon: for example, the ‘carpet crawler’ model, in which the oligodendrocyte process flattens out and expands after contact with the axon and then wraps the axon like a rolling carpet (Bauer et al., 2009; Bunge et al., 1989); the ‘serpent’ model, in which the oligodendrocyte process wraps around the axons in a helical fashion first, like a corkscrew, and then expands into overlapping sheets (Asou et al., 1995; Bauer et al., 2009; Ioannidou et al., 2012; Knobler et al., 1976); and the ‘liquid croissant’ model, in which the oligodendrocyte process continuously wraps and spreads sideways along the axon (Snaidero et al., 2014; Sobottka et al., 2011). Alternatively, through a detailed analysis of CASPR and NF155 (NFASC – Mouse Genome Informatics) localization during myelination, Pedraza et al. (2009) proposed that the initial contact between the axon and the oligodendrocyte process persists throughout myelination and the membrane wraps the axon like a yo-yo before expanding laterally along the internode; then, the leading edges of all the overlapping layers expand and draw closer at the paranodes.

Part of the controversy about these current models describing the process of myelination is that they were developed primarily through static images and, when available, with the support of relatively short-term live-imaging data sets, which leaves many questions unanswered. To clarify the dynamics of wrapping, in the current study the initiation and progression of myelination were tracked live for up to almost 8 days by visualizing oligodendrocytes and axons using membrane-localized GFP and RFP, respectively (Fig. 6; supplementary material Fig. S5, Movies 1 and 2). Intriguingly, this was a dynamic process in which oligodendrocyte processes performed surveys around and along the axon by constantly moving, extending and retracting as if they were sensing their environment (Fig. 6A). Next, a process attached itself at one or two points on the axon but was otherwise free and continued to survey, a step we termed anchoring. After a period of time that varied from process to process, the oligodendrocyte membrane wrapped the axon by folding repeatedly around it (Fig. 6B). The processes were relatively stable for ∼20 h, suggesting that the initial wrapping of the axon was complete (Fig. 6C). In a consecutive analysis, we stopped live imaging and imaged the same oligodendrocyte at higher resolution after immunostaining (Fig. 6D). The processes that we observed anchored and wrapped around the axon were reminiscent of previously observed helical spiraling (Ioannidou et al., 2012). Based on live-imaging data, it does not seem plausible that the process crawled around axons like a snake, forming a perfect corkscrew-like structure, as was previously suggested in the variants of the serpent model, because the process was anchored on the axon before wrapping, as was also proposed by Pedraza et al. (2009).

To assess the wrapping process at the ultrastructural level, we performed 3D EM analysis by serial section EM using an ATLAS scanning approach (Hayworth et al., 2006). The inner and outer tongues (the inner and outer layers wrapping the axon) and the mesaxon (tip of the inner tongue) were identified on an axon that was wrapped twice (Fig. 7; supplementary material Movie 3). Strikingly, the mesaxon wound around the axon as would be expected if the oligodendrocyte process wrapped around the axon and started expanding laterally (Fig. 7A,B). Such lateral expansion would push the leading edges of the membrane sheets towards future paranodes, as was proposed by Pedraza et al. (2009) and observed recently by Snaidero et al. (2014) through live imaging of late stage myelination and serial EM. Consistently, we also observed cytoplasmic channels in the form of myelin outfoldings that facilitate membrane delivery to the growing leading edges (Fig. 7C). Hence, after the initial anchorage and wrapping of the process around the axon, myelination should continue by lateral expansion of the ensheathed layers.

Thus, we propose a mechanism to unify our data concerning early wrapping dynamics with our observations and those of others on later stages of myelination, including: (1) anchorage, wherein an oligodendrocyte process first surveys the environment and the axon in a dynamic fashion and then anchors on the axon while continuing to survey; (2) the unanchored part of the process folds over and wraps the axon; and (3) the process expands along the axon as flattened membrane sheaths that slip over each other to form many layers, the leading edges of which eventually meet at the paranodes (Fig. 8; supplementary material Movie 4). We call this mechanism SARAPE (survey, anchor, wrap and expand) – a term derived from the Mexican blanket-like shawl.

The process and mechanism of myelination in the CNS are still poorly understood. We set out to establish an efficient and reliable assay to study this process. First, we developed a protocol to differentiate myelination-capable oligodendrocytes from mouse ESCs and assessed their identity and maturation levels by both immunostaining and qPCR. When co-cultured with ESC-derived neurons, these oligodendrocytes reproducibly formed compact myelin. We adapted this co-culture system as a myelin formation assay to an optimally engineered microfluidic device, thereby establishing an assay that is robust, reliable, easy to modify and quantitative. Introduction of a semi-automated myelin evaluator, CEM, made it feasible to quantify large data sets such as the entire experimental area, i.e. the myelination compartment, circumventing any sampling bias.

With the advancement of induced pluripotent stem cell (iPSC) technology, an increasing number of neurological diseases are being modeled in vitro. Recently, myelin pathology of PMD has been replicated using patient-derived iPSCs (Numasawa-Kuroiwa et al., 2014). iPSCs from MS patients are also available (Douvaras et al., 2014; Song et al., 2012). It is possible to adapt these models to the microfluidic co-culture environment to study myelination. CEM can identify and quantify myelin on images from any myelinating platform, making the results and framework presented here practical to apply to a number of different systems, ranging from iPSC to animal models of neurological diseases.

Finally, we used this myelin formation assay to image the myelination process in real time, following oligodendrocytes over several days as they surveyed around and along axons by constantly extending and retracting their processes. It is possible that this surveying activity allows oligodendrocytes to identify axons that carry the correct signals to be myelinated. Surprisingly, we observed that the process anchored on the axon but continued to survey. This observation is consistent with earlier work by Asou et al. (1995), although that work was conducted at much lower resolution and for a shorter time period. Additionally, it has been suggested that the interaction between CASPR and its binding partner on the glial membrane, NF155, is never broken from the initial contact to completion of myelination (Pedraza et al., 2009). Therefore, it is possible that these proteins are part of the anchor. Further studies are required to explain the nature of the anchor point and the details of the interactions between oligodendrocytes and axons. It will also be intriguing to examine through additional studies whether myelination occurs through SARAPE in vivo in mouse and in other organisms.

We developed an ESC-based in vitro myelin formation assay by combining stem cell and microfluidic technologies that can be quantified by an automated method and is suitable for long-term live imaging. We believe that our assay constitutes a powerful tool to better understand myelination and to identify novel therapeutic avenues for treating dysmyelinating and demyelinating diseases such as MS and leukodystrophies. In its current form, our assay has already led us to investigate myelin wrapping and it has the potential to be modified, depending on the specific question being asked, and to be expanded and optimized to accommodate larger drug screening formats.

A timeline representation of the process is provided in Fig. 1. Mouse D3 ESCs were grown on mouse embryonic fibroblast (MEF) plates by feeding with mouse ESC medium supplemented with 1000 U/ml mouse leukemia inhibitory factor (LIF; Millipore) every day. NPCs were generated and differentiated as modified from a previous study (Marchetto et al., 2008). On D1, OPC differentiation started by switching NPCs into differentiation medium, which comprised N2/B27 medium supplemented with IGF1 and PDGFα (R&D Systems). On D8, OPCs were switched to maturation medium, which comprised N2/B27 medium supplemented with CNTF, NT3 (NTF3) and T3 (Sigma-Aldrich). From D1, cells were fed every other day. For further details of culture media and conditions, see the supplementary Materials and Methods.

Neurons were differentiated following a previously published protocol that enriches for cortical neurons (Gaspard et al., 2009). On D12, neurons were plated either on laminin-coated coverslips or in the neuronal compartment of microfluidic devices. Neurons were maintained in neuron medium for another 1-2 weeks for optimal neurite growth. Either 200,000 (coverslip) or 80,000 (device) oligodendrocytes at D15 of the differentiation protocol were plated on neurons. Every other day, co-cultures were fed with myelination medium, which comprises neuron medium supplemented with 40 ng/ml T3 (Sigma-Aldrich).

Microfluidic devices were fabricated as described (Taylor et al., 2005). A detailed description of the fabrication is provided in the supplementary Materials and Methods.

Cells were fixed in 4% paraformaldehyde in PBS for 15 min, blocked and permeabilized with horse serum (10%) and Triton X-100 (0.1%) in PBS and were incubated overnight with a combination of primary antibodies (see supplementary Materials and Methods). Images were acquired on Zeiss LSM confocal microscopes at 1, 0.5, 0.33 or 0.3 μm intervals and were processed with Zen (Zeiss), Imaris (Bitplane) and ImageJ (NIH).

On D8 and D15, cells were fixed and stained as described above. Images of six to ten randomly selected regions per well were taken on Stereo Investigator (MicroBrightField). Cell counts were performed using Cell Counter on ImageJ. The total number of cells was determined by DAPI staining. Three wells per experiment in two independent experiments were counted.

The expression levels on D8 and D15 were compared with undifferentiated NPCs (D0) by qPCR. The average fold changes of three independent experiments are shown. Three internal replicates were performed and averaged for each experiment. Gapdh expression was used for normalization. The primer pairs used are listed in supplementary material Table S2.

Oligodendrocytes expressing GFP and neurons expressing RFP were grown on gridded glass-bottom dishes and fixed as described above. Regions of interest (ROIs) were located and their locations recorded for future use (supplementary material Fig. S6). The samples were embedded in Spurr's resin (EMS, 14300) and ROIs were marked by hand and later by the UV beam of a Zeiss PALM MicroBeam laser capture microdissection microscope (supplementary material Fig. S6). Transverse 70-nm sections were then taken of the ROI and imaged on a Zeiss Libra 120 PLUS for TEM.

After confirming wrapping events by TEM, 100-nm thick serial sections were taken from the same block using a similar device to that reported elsewhere (Horstmann et al., 2012). Sections were counterstained on the chip (supplementary material Fig. S6) and were then imaged in a Zeiss Sigma VP field emission scanning electron microscope using a four-quadrant solid-state backscatter detector with the quadrants inverted to produce a TEM-like image. Of the 52 serial sections taken, two were omitted due to sectioning and collection artifacts (slices 12 and 32). The resulting dataset was aligned, segmented and 3D rendered on TrackEM2 (FIJI) and Amira (FEI,). For further details of EM and array tomography, see the supplementary Materials and Methods.

Images of the entire myelination compartment of each device were acquired as 15% overlapping tiles on confocal microscopes at 1-μm intervals and were stitched in Zen. A detailed description of the computational and manual methods is provided in the supplementary Materials and Methods, Fig. S4 and Appendix S1. Eleven devices from four independent experiments were quantified for total, GFP and nonGFP myelination. Five devices were quantified without T3 supplementation. Five devices for each condition were quantified manually to evaluate CEM.

In total, we observed 21 complete wrapping and 15 partial wrapping events. SARAPE was observed in all 21 of the complete wrapping events and anchored oligodendrocyte processes were observed in all 15 partial wrapping events. Although in most of these events a single oligodendrocyte process anchored at one location and wrapped the axon, an oligodendrocyte may have sent out more than one process to wrap different axons (as in Fig. 5; supplementary material Movie 1); in two cases (one complete and one partial wrapping), two processes of the same oligodendrocyte were anchored on the axon and wrapped the same segment (as in supplementary material Fig. S4 and Movie 2). Further details are provided in the supplementary Materials and Methods.

Error bars in all figures indicate s.e. For qPCR, graphs are presented in log scale; the error bars are represented as described (Sriramulu et al., 1999). P-values for myelin quantification were calculated using the Mann-Whitney U-test.

Lentiviral reporter plasmids were derived from pCSC-SP-PW (Marr et al., 2004). A mouse Mbp promoter was utilized for oligodendrocyte-specific expression of EGFP (Gow et al., 1992). A human synapsin I (SYN1) promoter was utilized for neuron-specific expression (Kügler et al., 2001) of either tdTomato (Fig. 6; supplementary material Movie 1) or mCherry (supplementary material Fig. S5, Movie 2). EGFP and tdTomato were targeted to the cell membrane through the N-terminal acylation peptide of lymphocyte protein tyrosine kinase (LCK) (Zlatkine et al., 1997) with a five amino acid linker (GPVAT) separating the two protein regions. Lentiviruses were produced as described previously (Tiscornia et al., 2006). See also the supplementary Materials and Methods and Table S3 for oligonucleotides used for reporter plasmid generation.

We thank Ayad Ali and Ozlem H. Caglayan for technical assistance; Yury Sigal, Michael Adams and Jamie Kasuboski for microscopy assistance; Jane Peppard for helpful discussions concerning assay development; and Mary Lynn Gage for editorial comments. CEM is dedicated to the loving memories of Cem and Dilay Kerman, who were taken from us far too early.