Building neuromuscular junctions in vitro

At the beginning of the last century the ‘neuron doctrine’ proposed by Ramón y Cajal revolutionized the emerging field of neuroscience. The doctrine firmly established individual neurons as the basic anatomical and physiological unit of the nervous system, and prompted many questions regarding how neurons communicate with one another and how the space between them is crossed by the signals responsible for that communication. We now know that neurons make complex structures called synapses that allow chemical transmission of signals among cells. The neuromuscular junction (NMJ) is a tripartite chemical synapse of the peripheral nervous system. It allows the transmission of motor information that governs muscle contraction from the central nervous system to the muscle fibers. The NMJ comprises a motoneuron on the presynaptic side, a muscle cell on the post-synaptic side and a terminal Schwann cell (tSC) wrapping the NMJ structure. A fourth cell type, a kranocyte, which caps the NMJ and plays a role in synapse regeneration, has recently been identified in rodents (Court et al., 2008) (Fig. 1). Fatt and Katz showed that end-plate potentials (EPPs), which correspond to the chemically induced change in electric potential of the motor end plate, originate from the spontaneous or electrically evoked release of acetylcholine (ACh) from the nerves (Fatt and Katz, 1951). Furthermore, Katz and colleagues proposed the ‘quantal hypothesis’, stating that ACh is released from the nerve terminal in packets or ‘quanta’. The spontaneous release of one quantum results in a miniature EPP (mEPP) in the post-synaptic cell, whereas the synchronized release of multiple ACh quanta leads to EPPs. Synaptic vesicles accumulate at discrete zones of the nerve terminal called ‘active zones’ (Couteaux and Pecot-Dechavassine, 1970). A functional definition of active zones came a few years later, when it was shown that vesicles dock and fuse at the membrane, and their contents are exocytosed in response to nerve depolarization (Heuser et al., 1979). Taken together, these pioneering studies on NMJs provided the basis for our current knowledge of the intricate mechanisms underlying synaptic transmission.

Neuromuscular junctions have since been the most intensively studied synapses, because of their comparatively large size, relative simplicity and accessibility. Indeed, among specific features, neurotransmitter release at this synapse is mediated by multiple docking sites present in long active zones in frogs or several active zones in mammals. In contrast, synaptic transmission in the brain is associated with a single active zone per synapse (Biro et al., 2005; Silver et al., 2003). Another important distinction between NMJs and central synapses is that the post-synaptic element at an NMJ only receives cholinergic inputs, whereas post-synaptic neurons receive a large variety of inputs, which makes the analysis of the response to a single neurotransmitter more difficult to interpret. Therefore, the NMJ remains an excellent model for studying the mechanisms of neurotransmission, and is well suited to study synaptic formation, maintenance and function. Furthermore, the NMJ is at the core of a number of human pathologies that all have the perturbation of neurotransmission in common (Box 1).

Here, we provide an update on the different in vitro models that have been investigated to increase our understanding of the normal and pathological development of NMJ. We present a focus on the recent findings on humanized in vitro models, which have been made possible using human pluripotent stem cells, and discuss the promises and challenges of these new models.

A great deal of information on how synapses form has been obtained through studying muscle cell lines (mostly the murine C2C12 cell line) that form spontaneous acetylcholine receptor (AChR) aggregates, and co-cultures of muscle cells and motoneurons (Sanes and Lichtman, 2001). Early studies were carried out in cultures of neurons and muscles from Xenopus laevis (Anderson et al., 1977) or from chick (Frank and Fischbach, 1979), laying the groundwork for understanding the mechanisms of in vitro NMJ formation and the study of NMJ disorders (discussed below). These datasets have been extended in vivo with the emergence of transgenic mice, along with the use of non-mammalian animal models (Box 2). Here, we briefly describe the steps in NMJ formation as well as the essential molecular and structural markers of NMJ formation and maturation (Figs 1 and 2). For more detailed discussion of the mechanisms of synapse formation, we refer the reader to recent reviews (Burden et al., 2018; Li et al., 2018).

NMJ development requires extensive communication among the three key components of the tripartite synapse: presynaptic motoneurons, post-synaptic muscle fibers and tSCs. NMJ formation is closely intertwined with myogenesis and innervation occurs between embryonic days (E) 12.5 and E14 in mice, immediately after the first wave of myotube formation. This timing varies slightly among specific muscles and their locations (Fig. 1).

The composition of AChRs changes with development as follows: nicotinic AChRs are pentamers composed of five subunits in a stoichiometry of two α-, one β-, one δ- and one γ-subunit in the embryo. These are replaced by a ε-subunit at the postnatal age (Changeux et al., 1992). Within 3 weeks of birth, AChR clusters adopt a complex, pretzel-like configuration in mice (Sanes and Lichtman, 2001). The maturation of the NMJ leads to an increase in number, stability, density and clustering of AChRs in the post-synaptic compartment, as well as the differentiation of the presynaptic membrane into active zones with proper synaptic proteins and associated synaptic vesicles (Burden et al., 2018). The overall process reflects strong reciprocal signaling between pre- and postsynaptic structures.

Another crucial component involved in the formation of the NMJ is the tSC. Although the very first steps in synapse formation do not require the presence of tSCs, these cells are necessary for maintenance and maturation of NMJs (Ko and Robitaille, 2015). Ablation of tSCs after NMJ maturation at P30 in mice, leads to NMJ fragmentation and neuromuscular transmission deficits (Barik et al., 2016).

Over the past decade, extensive progress has been made in understanding the molecular mechanisms involved in NMJ formation. Here, we summarize the current knowledge on some of the key molecules and signaling pathways that are indispensable for NMJ formation and maintenance (Figs 1 and 2).

MuSK (muscle-specific kinase), a transmembrane receptor tyrosine kinase and its co-receptor LRP4 (low-density lipoprotein receptor-related protein 4) are crucial for the aggregation of AChRs before and after innervation. Prior to innervation, the MuSK/LRP4 complex accumulates in the muscle central region where it induces AChR clustering. With innervation, LRP4 binds a neural isoform of agrin (AGRN), a heparan sulfate proteoglycan secreted by the axon terminal (Ferns et al., 1993; Kim et al., 2008; Ruegg and Bixby, 1998; Zhang et al., 2008). Agrin interaction with LRP4 results in the formation of tetrameric agrin-LRP4 complexes that promote activation and transphosphorylation of MuSK, leading to AChR clustering and NMJ formation (Zong et al., 2012). AChR clusters that are apposed to nerve terminals persist and grow, while the non-apposed ones disassemble (Wallace, 1989). Mutations in the human AGRN gene cause congenital myasthenic syndrome (CMS), with distal muscle weakness and atrophy (Nicole et al., 2014).

More recently, two studies have independently revealed that the muscle LRP4 also acts as a retrograde signal for presynaptic differentiation (Wu et al., 2012; Yumoto et al., 2012). MuSK activation requires Dok7 (docking protein 7; an adaptor protein expressed by the muscle), which is indispensable for MuSK activity and therefore AChR clustering. The binding of Dok7 to the cytoplasmic juxtamembrane region of MuSK strongly enhances the tyrosine kinase activity of MuSK, leading to its activation (Hamuro et al., 2008; Inoue et al., 2009). MuSK-deficient mice show neither muscle pre-patterning of AChRs nor NMJ formation, and die perinatally from respiratory failure (Okada et al., 2006). More recently, it has been shown that the postnatal knockdown of Dok7 gene in mice causes structural defects in NMJs and motor dysfunction, supporting a role for Dok7 in NMJ maintenance (Eguchi et al., 2016).

Beginning around the time of birth, fetal AChRs (α2βγδ) are gradually replaced by the adult subtype of AChRs (α2βεδ). Although the physiological significance of this developmental switch remains to be fully understood, different studies have suggested a role of the γ-subunit during the NMJ formation and more specifically in innervation patterning (Koenen et al., 2005; Takahashi et al., 2002). For example, deletion of the mouse AChR γ-subunit leads to perinatal lethality associated with an absence of pre-patterned AChR clusters during initial stages of neuromuscular synaptogenesis (Koenen et al., 2005; Takahashi et al., 2002). A key effector in AChR clustering is the protein rapsyn, a cytoplasmic scaffolding protein expressed in the muscle, which binds directly to AChR and anchors AChR in the post-synaptic membrane (LaRochelle et al., 1990; Sobel et al., 1978). Rapsyn-deficient mice fail to develop post-synaptic specializations (Gautam et al., 1995), indicating that rapsyn is essential for the tethering of AChRs to the post-synaptic apparatus. Although rapsyn assembles with AChRs in the secretory pathway (Marchand et al., 2002; Moransard et al., 2003), the exact mechanisms downstream of MuSK/Dok7 that stimulate rapsyn-induced AChR clustering remain to be determined.

In addition to these core molecules that are essential for NMJ formation, a number of important regulators also participate in this process. Wnts are a group of glycoproteins secreted at the NMJ by the muscle fiber, motoneurons and tSCs. Several Wnt ligands bind to the cysteine-rich domain (CRD) in MuSK (Barik et al., 2016), and play a role in axon extension and branching, as well as in AChR prepatterning (Li et al., 2018). Both canonical and non-canonical planar cell polarity (PCP) Wnt pathways are affected in transgenic mice in which the CRD in MuSK is deleted (Messeant et al., 2017).

Dystroglycan (DG) has a dual function: it organizes extracellular matrix (ECM) scaffolds and stabilizes AChR clusters. DG is formed of two subunits that are generated by proteolytic cleavage of a single precursor protein. The extracellular subunit is necessary to organize a functional scaffold in the basal lamina including perlecan, acetylcholinesterase/ColQ and laminin (Jacobson et al., 2001; Peng et al., 1998). DG is also a receptor for agrin (Gee et al., 1994), although – unlike Lrp4 – it does not activate MuSK.

The synaptic basal lamina contains many components that have important regulatory roles in myogenesis and synaptogenesis. ECM molecules help guide the process of innervation by motor neurons and are crucial to formation of the post-synaptic density as well as organizing and maintaining functional apposition of the pre- and post-synaptic elements (Maselli et al., 2012; Rogers and Nishimune, 2017). Key junctional basal lamina components include collagens [such as collagen IV (α3)(α4)(α5) and (α5)₂(α6) trimers, collagen XIII and ColQ), glycoproteins (such as laminin isoforms α4β2γ1, α5β2γ1 and α2β2γ1, and nidogen 2) and heparan sulfate proteoglycans (HSPGs), such as perlecan and agrin (Fox et al., 2007). The laminin β2 chain is considered as an active zone organizer and binds to the voltage-dependent calcium channels (VDCC) located in the active zones (Nishimune et al., 2004). The interaction between laminin β2 and VDCCs leads to VDCC clustering, which is the first step toward a mature active zone. In laminin β2 knockout mice, the number of active zones and synaptic vesicles are reduced along with a reduction in neurotransmitter release. Mice lacking ColQ, collagen XIII, collagen IV (α3)(α4)(α5), (α5)₂(α6) trimers or collagen VI also show immature nerve terminals and/or NMJs (Cescon et al., 2018; Sigoillot et al., 2016; Zainul et al., 2018). Networks of laminins and collagens are connected to one another by nidogen and are anchored to the sarcolemma and intracellular cytoskeleton by binding to laminin receptors: integrins and, most importantly, DG (Fox et al., 1991; Jacobson et al., 2001).

For over one century, different models and approaches have been developed to decipher the mechanisms involved in the formation, maintenance and function of NMJs. Meanwhile, increased incidence and prevalence of inherited neuromuscular diseases require the development of pertinent pathological models for deciphering the pathophysiology underlying these diseases and developing effective treatments (Bhatt, 2016). Although animal models have been crucial for the NMJ field, they have certain limitations, such as ethical issues involving animal welfare, their capacity to fully reproduce some human pathological phenotypes and in their complexity, which hinders the identification of cell-specific mechanisms, including anterograde and retrograde signals. Researchers have tried to overcome these hurdles by developing in vitro models to recreate the NMJs. In the following section, we focus on the contribution and the evolution of these in vitro models of the NMJ (Fig. 3).

Historically, the first neuromuscular junction obtained in vitro was reported by R. G. Harrison in 1959 (Harrison, 1959). By cultivating medullary cord and axial mesoderm explants from clotted frog lymph between two glass slides, Harrison observed striated and contracting muscle fibers. As the frog muscle cells do not contract spontaneously, he concluded that the contractions were nerve induced. As the embryos are easy to manipulate, collect and dissect, amphibians have been extensively used as animal models for generating in vitro models of neuromuscular junctions (Anderson et al., 1977; Kidokoro et al., 1980; Peng and Nakajima, 1978; Swenarchuk et al., 1990). Among amphibians, two species stand out – Xenopus laevis and Ambystoma mexicanum – which belong to the anuran and the urodele orders, respectively. Xenopus laevis offers a number of advantages, most notably cells that are inexpensive to culture and mature rapidly (Peng et al., 1998; Swenarchuk et al., 1990). Within 3 days of co-culture, myotomal muscle cells are seen to be contracting due to innervation by neurons (Anderson et al., 1977). Cells from Ambystoma mexicanum, like Xenopus cells, can be cultivated at room temperature and in a basic cell culture medium complemented with 1% horse serum or fetal bovine serum (Anderson et al., 1977). Ambystoma cells differ from Xenopus cells in terms of the time of maturation. Ambystoma cells require a longer period to generate mature myofibers and establish functional synapses (Swenarchuk et al., 1990).

In parallel, a number of other species from chick to rodents have been used to build models for NMJs. These models were justified in part by the lack of tools to study cell biology in frog. Burgen and colleagues used rat phrenic nerve-diaphragm explants to investigate the action of the botulinum toxin (Burgen et al., 1949) (Fig. 3A). In these first studies, models were formed by removing a section of embryonic spinal cord with the phrenic nerve and muscle still attached. These explants were maintained as ‘organ cultures’.

The next breakthrough was the development of physically separated explants at difference with the previous system (Fig. 3B). This strategy allowed the observation of newly formed NMJs (James and Tresman, 1968). In these dissociated explants cultures, most of the NMJ models aimed to visualize and understand the process of neuromuscular synapse formation and also to decipher the mode of action of pharmacological drugs (Evans, 1969). At the same time, the first serial electron microscope studies allowed detailed observations of the ultrastructure of mouse neuromuscular synapses in culture (Bornstein et al., 1968). Crain and colleagues showed that explant cultures of rat spinal cord and skeletal muscle tissues from rat and human formed functional NMJs (Crain et al., 1970).

The development of dissociated neuronal cell culture, where isolated neurons were grown on a single layer along with the availability of inverted microscopes, facilitated electrophysiological recordings in vitro (Scott et al., 1969). Using these tools, Fischbach and colleagues demonstrated that co-cultures of cells dissociated from chick skeletal muscle with those from chick spinal cord produced functional synaptic contacts (Fischbach, 1970) (Fig. 3C). These inter-species models, also called chimeric models, confirmed that NMJ formation and function is conserved among a variety of species.

Chicks, frogs and rodents are the most common species used as sources of tissues for studying NMJ formation. Chicks offer easy access and manipulation of the embryo (Frank and Fischbach, 1979; Johnson et al., 1981), and have been used in chimeric models with rat and mouse cells. However, as mammals, the rodents Mus musculus (mouse) and Rattus norvegicus (rat) provide greater insight into human development or pathology than amphibians or birds (Ionescu et al., 2016). Mouse myoblasts can differentiate into fused and striated myofibers 2-4 days after plating (Ionescu et al., 2016). Functional synapses form 4 days later when they are cultured with murine neuronal explants (i.e. 2 days for contact between axons and myotubes followed by 2 days for synapse maturation) (Vilmont et al., 2016) (Fig. 3D). However, mouse co-cultures are rather inefficient systems, with few synapses forming, in contrast to frog co-cultures (Sanes and Lichtman, 1999).

Chimeric cell cultures models have provided answers to the origin – muscle or nerve – of a number of molecules expressed at the NMJ. For example, species-specific antibodies, one targeting specifically the HSPG from anuran species and the second targeting the HSPG from urodela species, have identified that HSPGs originate from the muscle (Swenarchuk et al., 1990). In the same vein, the development of transgenic animals brought new tools in the field of in vitro NMJ modeling. For example, Hb9::GFP transgenic mice express GFP in differentiated motoneurons (Wichterle et al., 2002). The use of Hb9::GFP mouse cells allow for easy visualization and identification of motoneurons in culture (Ionescu et al., 2016; Umbach et al., 2012).

Embryonic spinal cord explants provide motoneurons in situ, within their 3D multicellular environment (Ionescu et al., 2016). Alternatively, dissociated cells from embryonic spinal cord or myotome allow one to enrich specifically for motoneurons or muscle cells (Fischbach, 1972; Nakajima et al., 1980). Clonal and immortalized cell lines are often used to provide a large, convenient and homogenous population of cells ready to culture. However, these cells may harbor abnormal phenotypes that are not completely representative of healthy cells in vivo, either because of the immortalization process they have been through or the fact that they are cancer cells. For example, G-8 clonal myogenic cells have a modal chromosome number of 75 instead of 40 (Christian et al., 1977). Nevertheless, muscle cell lines are used largely to study the mechanism of post-synaptic differentiation and more specifically, the mechanisms of AChRs aggregation. For example, C2C12 cells are a subclone originally isolated by Blau (Blau et al., 1983) from a murine cell line generated in 1977 by Yaffe and Saxel (Yaffe and Saxel, 1977). C2C12 cells are useful to study early muscle differentiation, including myoblast fusion as well as AChR aggregation. However, C2C12 cells do not fully differentiate and do not form aggregates of acetylcholinesterase (AChE). A more recently established murine muscle cell line called MLCL forms AChE clusters in addition to the AChR clusters (Cartaud et al., 2004). Finally, collaborations between biologists and physicists lead to the emergence of new culture devices such as microfluidic devices with the objectives of better modeling NMJ pathologies, as well as providing a more suitable system for some mechanistic studies (Machado et al., 2019; Uzel et al., 2016; Zahavi et al., 2015). These developments are discussed below.

The establishment of reliable protocols for co-cultures led many groups to consider developing humanized in vitro models of NMJs to study human neuromuscular diseases. Whichever in vitro model is used, a number of criteria have to be met to ascertain that an NMJ model is representative of those in vivo in terms of structure and function, and therefore a valid model. The criteria list that has been used is described in the next few paragraphs and is summarized in Tables 1 and 2.

In the first instance, different presynaptic, synaptic and post-synaptic hallmarks can be used as physiological and morphological clues indicative of normal NMJs. Some of them can be detected by immunolabeling of key synaptic proteins (Das et al., 2007; Dorchies et al., 2001; Umbach et al., 2012; Vilmont et al., 2016) (Fig. 2). For skeletal muscle, the first indication of complete differentiation is the morphology of the myofibers. NMJ morphology is dependent on the species as well as the stage of NMJ differentiation and the innervated muscle. Gillingwater and colleagues developed a standardized ‘NMJ morph’ tool to study NMJ morphology from tissue samples, highlighting these specific parameters in function of the species (Jones et al., 2017; Jones et al., 2016). The best approach to ascertain the presence of a synapse is through electron microscopy (EM). EM allows the visualization of a contact between axon terminals and muscle cells, the presence of synaptic vesicles in the presynaptic element and post-synaptic densities apposed to presynaptic elements (reviewed by Slater, 2017). However, it remains to be shown that this synapse is functional; i.e. muscle contractions induced by nerve activity and the presence of spontaneous/evoked transmission.

Striations and spontaneous contractions are also characteristics acquired during differentiation of myoblasts into myotubes in mammals. Mouse, rat and human muscle cells can contract in aneural mono-culture, unlike amphibians (Guo et al., 2014). These contractions are helpful in assessing the differentiation of the cells. Generally, two types of muscle cell contractions can be observed in co-culture models: small twitches independent of innervation (which are short in length and time) and contractions induced by innervation, which are longer and stronger (Ionescu et al., 2016). Innervation-dependent contractions are also characterized by their synchrony from the innervation of several myotubes by axonal branches of the same motoneuron (Vilmont et al., 2016). These two types of twitching movements can easily be distinguished by adding the drugs tetrodotoxin (which inhibits voltage-gated sodium channels) or d-tubocurarine (an antagonist of AChR that prevents ACh binding to its receptor) to the culture medium (Vilmont et al., 2016) (Table S1). The innervation-independent twitches are blocked only by tetrodotoxin, whereas the contractions induced by innervation are abolished by both drugs.

Electrophysiological demonstration of spontaneous and evoked neurotransmission is the most detailed way to characterize the functionality of a NMJ model. mEPPS can be recorded by a microelectrode in the motoneuron and another in the myotube close to the point of contact. This criterion alone is an unequivocal evidence of a synapse, although it may not be fully mature. A functional neuron can be identified upon electrical stimulation of the motoneuron and evocation of EPPs (Nelson et al., 1993). Alternatively, EPPs can be evoked and synapses revealed by applying glutamate from a microelectrode directly onto glutamatergic motoneurons (Puro et al., 1977).

To assess whether the EPPs or contractions are dependent on neurotransmission and AChRs, different toxins or drugs can be used. As mentioned previously, d-tubocurarine is a toxin blocking the AChR functionally, and thus blocking twitching, EPPs and mEPPs mediated by nicotinic AChR (Chipman et al., 2014; Kidokoro et al., 1980). A common test is to record muscle cell twitching, spontaneous EPPs or evoked EPPs, and then add d-tubocurarine in the medium. This addition of d-tubocurarine eliminates twitching and EPPs in the muscle cells, but not the motorneuron action potential (Umbach et al., 2012). This inhibition is reversible and when the d-tubocurarine is removed from the medium, the muscle fibers should re-start twitching or expressing mEPPs and EPPs almost immediately (Anderson et al., 1977; Das et al., 2007; Dorchies et al., 2001; Puro et al., 1977; Umbach et al., 2012; Vilmont et al., 2016).

As the prevalence of neurodegenerative and muscular disorders is rising, and available treatments for these diseases are scarce, it is a matter of urgency to investigate new treatment strategies. Understanding the pathophysiology underlying diseases, such as amyotrophy lateral sclerosis (ALS) or congenital myasthenic syndrome (CMS), is a crucial step for early diagnosis and effective treatment. Therefore, the remaining challenge was to generate a human in vitro NMJ adapted to mimic in vivo biology with the aim to decipher pathological mechanisms and perform functional analysis.

In 1970, Peterson and Crain demonstrated that neuromuscular connections could be formed between explants of embryonic spinal cord from rodents and skeletal muscle tissue from adult humans (Crain et al., 1970). Formed with human myofibers, these cultures were maintained up to 7 weeks and electrophysiological analysis indicated that synaptic connections between rodent neurons and human muscle were functional and could be blocked with d-tubocurarine. They also demonstrated an accumulation of acetylcholinesterase (AChE) at sites of synaptic contact (Crain et al., 1970). Extending this first demonstration, Peterson and Crain observed pathological features in long-term co-cultures with human dystrophic muscle. In co-cultures maintained for more than 4 months, unusual focal myofibrillar lesions in a substantial number of innervated mature Duchenne muscular dystrophy (DMD) muscle fibers were described (Peterson et al., 1986). However, the complexity of preparing these co-cultures from human muscle biopsies has restricted their use.

An important breakthrough in the field of NMJ in vitro models came with the development of co-cultures between dissociated human primary skeletal muscle cells, derived from satellite cells, with rodent embryonic spinal cords (Table 3). Kobayashi and Askanas were the first to demonstrate that monolayer cultures of adult human muscle cells can be innervated by embryonic rat spinal cord neurons and that the differentiation of human skeletal muscle cells, as well as their electrophysiological properties, are greatly enhanced by innervation (Askanas et al., 1987; Kobayashi et al., 1987). Importantly, dorsal-root ganglia were essential for achieving functional innervation, as demonstrated by the presence of regular contractions and well-organized acetylcholinesterase-positive contacts (Askanas et al., 1987; Kobayashi et al., 1987). Further electrophysiological studies also demonstrated mEPPS and EPPs in these co-cultures were abolished by application of AChR antagonists, such as d-tubocurarine or α-bungarotoxin (α-BTX), which also blocks the contractions of human myotubes in these chimeric co-cultures. These studies also highlight that innervation can occur without the original basal lamina (Kobayashi and Askanas, 1985). Finally, the development of these chimeric models of NMJ provided an important basis for pathological studies described for myotonic dystrophy (Kobayashi et al., 1990), DMD (Park-Matsumoto et al., 1991), McArdle's myopathy (Martinuzzi et al., 1993), spinal muscular atrophy (SMA), ALS and facioscapulohumeral muscular dystrophy (FSHD) (Braun et al., 1995). Several of these studies revealed some pathological phenotypes. Thus, human skeletal muscle from children affected by a severe form of SMA began to display degeneration only when co-cultured with embryonic rat spinal cords (Braun et al., 1995). Despite these important advances, the only human tissue that could be used in these co-cultures was skeletal muscle cells. Recent advances in stem cell biology have fueled the prospect of generating the main components of the NMJ, offering new perspectives for pathobiological studies and therapeutic development for neuromuscular diseases. Thus, the availability of human pluripotent stem cells (hPSCs), either of embryonic stem cell (ESC) or induced pluripotent stem cell (iPSC) origin, has made the generation of human muscle and nerve cells possible. The first protocols developed from hPSCs concerned the specification of spinal motoneurons. Based on developmental insights generated from animal studies and those gained on mouse ESCs (Wichterle et al., 2002), much progress has been made on the capacity to convert hPSCs into spinal motoneurons. Li and colleagues were among the first to attempt co-cultures between hPSC-derived spinal motoneurons and murine skeletal muscle cells C2C12 (Li et al., 2005) (Table 4). After 3 weeks of culture, contractions were observed only in the presence of hESC-derived motoneurons and immunostaining analysis revealed the presence of clusters of AChR juxtaposed to neurite contacts (Li et al., 2005). Since then, this approach has been extended to disease-specific hiPSC-derived motoneurons such as SMA (Yoshida et al., 2015). In this context, the use of SMA-hiPSC derived motoneurons led to an impaired AChR clustering on C2C12 that precedes motoneuron loss (Yoshida et al., 2015). However, it is important to emphasize the relatively low rate of differentiation of hPSCs into hPSC-derived motoneurons observed in these studies, with only 25-30% of motoneurons generated on average.

To circumvent this limitation, motoneurons enriched by FACS or directly induced from fibroblasts (hi-motoneurons), without passing through an induced pluripotent stem cell stage, have been used (Du et al., 2015; Liu et al., 2016; Singh Roy et al., 2005). When co-cultured with primary rodent skeletal muscle cells, both sources of motoneurons led to the colocalization of synaptic markers such as synapsin with AChR clusters. Electrophysiological analyses also revealed spontaneous end-plate currents detected in murine myotubes when co-cultured with hi-motoneurons generated from human fibroblasts by overexpression of four transcription factors (SOX11, NGN2, ISL1 and LHX3). This electrophysiological activity can be blocked by using either d-tubocurarine, tetrodotoxin or a specific inhibitor of voltage-dependent sodium channels in muscle: ω-connotoxin. When applied to a pathological condition such as ALS, profound dysfunction in the ability of patient-specific hi-motoneurons to control functional muscle maturation is observed (Liu et al., 2016). Although promising, these first descriptions were all based on chimeric models and to reach the ‘holy grail’ of a ‘real’ human NMJ in vitro model, both co-culture components should be of human origin.

Over the past decade, the literature has documented fully humanized in vitro models of NMJs based on motoneurons mainly derived from hPSCs cultured with human primary skeletal muscle cells (Afshar Bakooshli et al., 2019; Marteyn et al., 2011; Santhanam et al., 2018; Shimojo et al., 2015; Steinbeck et al., 2016; Vila et al., 2019). Although all these studies diverged in their protocols, they all described the presence of AChR clusters in close proximity to neuritic processes, as well as muscle twitching after 2 or 3 weeks of co-culture. This indicates to some extent the functionality of human primary skeletal muscle cells when co-cultured with human motoneurons (Table 5). To go further ahead in the functional characterization of this system, Steinbeck and colleagues were the first to combine hPSC-based in vitro models of NMJs with optogenetics (Steinbeck et al., 2016). The authors used purified hESC-derived motoneurons stably expressing channelrhodopsin 2 (ChR2) coupled to the GFP under the human synapsin promoter. After 6-8 weeks of co-culture with human primary skeletal muscle, muscle twitches were observed and quantified after light stimulation. The addition of vecuronium, an antagonist of the nicotinic AChR, completely blocked the light-induced muscle twitches. Calcium imaging and electrophysiological analyses also confirmed these results. As purification of hPSC-derived motoneurons could lead to the selection of physiologically irrelevant motoneurons, these datasets were reproduced using motoneurons generated with a recent protocol that allows the efficient conversion of hPSC into a homogenous population of spinal motoneurons without using a purification step (Maury et al., 2015). Interestingly, the authors also demonstrated that their ‘all human’ NMJ cultures could be used in the pathological context of myasthenia gravis (Box 1) (Steinbeck et al., 2016).

Despite these developments, several challenges remain. The first limitation is the source of the human skeletal muscle cells. All the models described are based on the use of human primary skeletal muscle cells, which exhibit substantial donor variation in terms of differentiation (Nikolic et al., 2017). The development of an ‘all human’ NMJ model in which both neuron and muscle components are derived from the same individual is of great interest for modeling disease of genetic etiology. With the recent development of protocols allowing the conversion of hPSCs into skeletal muscle cells (reviewed by Pourquie et al., 2018), it is possible to produce co-cultures between motoneurons and skeletal muscle cells derived from hiPSCs from the same donor (Demestre et al., 2015; Maffioletti et al., 2018; Mazaleyrat et al., 2020; Osaki et al., 2018; Picchiarelli et al., 2019; Puttonen et al., 2015). Although the maturation level of these co-cultures generated from ‘embryonic-like’ pluripotent stem cells is still questionable, the detection of AChR clustering, contractions and calcium imaging suggest an acceptable level of functionality of these systems.

Another limitation stems from the fact that most of these studies have used a 2D cell culture system that does not mimic the native tissue structure. Consequently, different attempts have been made recently to evaluate the possibility of developing an ‘all human’ NMJ in vitro model by taking advantage of the advent of 3D cultures and microfluidic lab-on-a-chip technologies (Afshar Bakooshli et al., 2019; Maffioletti et al., 2018; Osaki et al., 2018; Santhanam et al., 2018; Vila et al., 2019). Microfluidics allows precise control over the cell microenvironment, providing an interesting platform for studying the NMJ. Motoneurons and muscle can be in two different chambers connected by axons, which enables spatiotemporal control over the cells. Analysis of the localized actions at the soma, along the axon and at the NMJ can be performed. Santhanam and colleagues were among the first to apply microfluidics to human motoneurons and muscle cells. Two weeks of co-culture were sufficient to observe muscle contractions after electrical stimulation of the motoneurons that can be blocked by treatment with d-tubocurarine (Table S1). However, the percentage of contracting myotubes is lower than the number of NMJs determined immunocytochemically, suggesting that some of the latter may not be mature enough to be functional (Santhanam et al., 2018).

Another approach relies on the use of 3D co-cultures with muscle cells in hydrogels, such as the commercially available basement membrane extract Matrigel followed by differentiating the motoneurons on the surface. 3D cultures tend to be functionally more relevant with respect to contractile muscle. Collagen I is a major component of the native ECM and harbors adequate properties needed for contraction and force generation. However, Collagen I has less stability, degrades faster and appears to be a less effective solution for long-term cultures (Leikina et al., 2002). In contrast to Collagen I, Matrigel seems to be better adapted for muscle differentiation (Grefte et al., 2012). Other alternatives to collagen are fibrin, fibrinogen or thrombin (Wu et al., 2005). Such approaches have recently been applied to the development of fully humanized NMJ in vitro models in which muscle cell laden hydrogels were formed in specific molds with pillar structures at each extremity sufficient. These then created a continuous uniaxial tension and directed orientation of cells along the axis of tension (Afshar Bakooshli et al., 2019; Maffioletti et al., 2018). A comparative analysis of 2D versus 3D co-cultures reveals a faster and more efficient functional innervation in 3D versus 2D co-cultures when reviewed after 2 weeks in culture. The determination was performed using calcium imaging, optogenetics and electrophysiological recordings (Afshar Bakooshli et al., 2019). A more mature expression profile of AChR isoforms was observed in 3D co-cultures with an increased expression of the adult AChR ε isoform in comparison with 2D cultures. Nonetheless, the small size of muscle fibers, as well as the properties of action potentials recorded in these co-cultures when compared with in vivo mammalian models, suggests that the 3D human NMJs co-cultures are not fully mature and that additional cues might be necessary to ensure further maturation of tissues (Afshar Bakooshli et al., 2019). Increasing culture time, adding trophic factors or providing other cell types have been suggested. Recently, complex isogenic multilineage 3D co-cultures have been developed containing skeletal muscle cells, vascular endothelial cells, pericytes and motoneurons, all derived from hiPSCs (Maffioletti et al., 2018). Whereas these complex co-cultures were mainly characterized at the morphological level, these results open new perspectives for the development of humanized organoid-like platforms for studying NMJ development and disease modeling.

Two studies have also described the combination of 3D system and microfluidics (Osaki et al., 2018; Vila et al., 2019). These approaches have the advantage that motoneurons and muscle cells can be compartmentalized while providing spatial guidance cues and signal gradients using 3D hydrogels for better integration. The study uses engineered functional contractile hiPSC-derived skeletal muscle cells co-cultured with hiPSC-derived motoneurons. Interestingly, such approaches are compatible for studying and validating the therapeutic potential of compounds for neuromuscular disorders such as ALS (Osaki et al., 2018).

Finally, 3D bioprinting is an emerging field that could be relevant for the development of in vitro NMJ systems. A recent study described the feasibility of using bio-printed 3D hydrogel skeletons to support culture, differentiation and alignment of human primary skeletal muscle cells compatible with the functional integration of immortalized human neural progenitor cells (Kim et al., 2020). Interestingly, these bio-printed constructs with neural cells facilitate rapid innervation and mature into organized muscle tissue, restoring normal muscle function in a rodent model of muscle defect injury (Kim et al., 2020). This development can be pursued in the future for improving the maturity of NMJ systems and highlights an area of considerable interest in the tissue-engineering field.

For over one century, the field of research around NMJs has grown constantly. Recent advances in stem cell biology have stimulated the development of innovative models of human NMJs. Despite initial successes in developing human motoneuron-muscle cells co-cultures, stem-cell derived NMJs remain immature in comparison with the human and chimeric NMJ models. One challenge is the limited survival of these co-cultures. The first chimeric mouse-human co-cultures could survive for more than 6 months (Peterson et al., 1986), whereas human stem-cell based co-cultures often do not survive beyond 4 weeks. However, a recent protocol of human stem cell co-differentiation enables the co-culture to survive up to 7 months (Mazaleyrat et al., 2020). Another challenge for future studies is to find a different method for generating spinal motoneurons. Among the motoneuron lineages in vivo, seven different lower motoneuron subtypes exist, each of them innervating a different muscle group. Most of the NMJ models from iPSCs in the currently published studies used iPS cell-derived general motoneurons. Efforts are under way to characterize the subtypes obtained and to improve the motoneuron differentiation protocols in order to obtain specific motoneuron subtypes (Hester et al., 2011; Lukovic et al., 2017). Similarly, iPS cell-derived muscle fibers are so immature that they cannot be classified as muscle fiber type I or II. Future work to obtain well-characterized motoneurons and muscle fibers subtypes will be particularly important when it comes to modeling neuromuscular pathologies, as all NMJs are not affected in the same way (Nicole et al., 2014). Additionally, tSCs are still a missing partner in current in vitro models. Protocols enabling the differentiation of iPS cells into myelinated Schwann cells are available (Liu et al., 2012; Ziegler et al., 2011). To date, there are no protocols for generating tSCs. These cells, together with complex and specialized basal laminas, will likely be required to reach a fully mature in vitro human NMJ model. Future research on 3D in vitro human NMJ models will be important in this effort. Recently, Faustino Martins and colleagues developed an elegant strategy to derive neuromuscular organoids from hiPSCs. Over a period of 50 to 100 days, these organoids reached a size of 5-6 mm in diameter and displayed an elongated morphology with neural tissue on one side and muscle cells on the other. In this system, electron microscopy confirmed the presence of presynaptic nerve terminals containing synaptic vesicles and active zones: tSCs and post-synaptic muscle fibers with organized sarcomeres and junctional folds. Importantly, these NMJs were functional, as demonstrated by the detection of muscle contraction after 50 days of culture. The functionality could be inhibited with d-tubocurarine, as well as spontaneous calcium oscillations and electrical activities (Faustino Martins et al., 2020). When employed with patient-specific stem cells together with CRISPR/Cas9 genome editing to restore or create mutations identified in patients, these new models will provide more reliable models of human NMJs in vitro and of neuromuscular diseases. Finally, to address the needs of new therapeutic approaches for neuromuscular diseases, standardized and miniaturized human in vitro NMJ models are a long-awaited drug-screening tool.

I-Stem is part of the Biotherapies Institute for Rare Diseases (BIRD) supported by the Association Française contre les Myopathies-Téléthon. The authors thank Dr Salvatore Carbonetto and Dr Amit Sandra for constructive comments and proofreading of the manuscript.