Sequentially induced motor neurons from human fibroblasts facilitate locomotor recovery in a rodent spinal cord injury model

Generation of autologous human motor neurons holds great promise for cell replacement therapy to treat spinal cord injury (SCI). Direct conversion allows generation of target cells from somatic cells, however, current protocols are not practicable for therapeutic purposes since converted cells are post-mitotic that are not scalable. Therefore, therapeutic effects of directly converted neurons have not been elucidated yet. Here, we show that human fibroblasts can be converted into induced motor neurons (iMNs) by sequentially inducing POU5F1(OCT4) and LHX3. Our strategy enables scalable production of pure iMNs because of the transient acquisition of proliferative iMN-intermediate cell stage which is distinct from neural progenitors. iMNs exhibited hallmarks of spinal motor neurons including transcriptional profiles, electrophysiological property, synaptic activity, and neuromuscular junction formation. Remarkably, transplantation of iMNs showed therapeutic effects, promoting locomotor functional recovery in rodent SCI model. Together, our advanced strategy will provide tools to acquire sufficient human iMNs that may represent a promising cell source for personalized cell therapy.


Introduction
Spinal cord injury (SCI) causes devastating neurological impairments and disabilities (Singh et al., 2014). SCI leads to the loss of sensory/motor functions and malfunctions in other organs such as bladder, kidneys and bowel, provoking an enormous impact on physical, psychological and social behavior of SCI patients (Bradbury and McMahon, 2006;Vismara et al., 2017). Unfortunately, there is no fully restorative treatment for SCI yet (Silva et al., 2014). Medications or surgical decompression are the only options for SCI treatment, however, these approaches are controversial due to the severe side effects and limited clinical efficacy (Cristante et al., 2012;Vismara et al., 2017).
Thus, cell replacement therapy has been proposed as a promising therapeutic intervention to reconstitute the damaged nervous system and improve functional recovery after SCI (Goldman, 2005). A number of studies have shown that transplantation of neural cells derived from pluripotent stem cells (PSCs) is effective in functional and histological restoration in SCI animal model (Lu et al., 2014;Nakamura and Okano, 2013;Tsuji et al., 2010;Vismara et al., 2017). Despite these encouraging advances, ethical issue of embryonic stem cells (ESCs) and tumorigenic potential of induced pluripotent stem cells (iPSCs) have impeded their translations into clinical trials (Deng et al., 2018;Fong et al., 2010;Miura et al., 2009;Ronaghi et al., 2010). To overcome these limitations, direct conversion of somatic cells into other cell types has been achieved while bypassing pluripotent state (Park et al., 2019;Park et al., 2020;Xu et al., 2015). However, application of previous protocols hindered therapeutic translations due to the heterogeneity and low yields, and involvement of multiple transcription factors which may increase the genetic mutagenesis (Kang et al., 2015). Most importantly, directly converted cells are fully differentiated post-mitotic cells which limit the acquisition of cell source in large-scale for therapeutic purpose. Hence, there is a critical need to develop a new method that enables the large-scale production of highly pure and functional target cells with consistent quality. In terms of treating SCI, it must be capable of generating motor neuron (MN) subtypes that are relevant to disease rather than producing general neurons.
In this study, we succeeded in generating induced motor neurons (iMNs) by using a minimal number of transcription factors, POU5F1(OCT4) and LHX3. POU5F1(OCT4) is known to play an important role in regulating pluripotent genes (Shi and Jin, 2010;Wang et al., 2007), and downstream target genes involved in developmental processes (Shi and Jin, 2010). Previously, overexpression of POU5F1(OCT4) allowed the generation of blood progenitor cells from fibroblasts (Szabo et al., 2010) by regulating hematopoietic gene, HOXB4, which is one of the POU5F1(OCT4) targets (Boyer et al., 2005). Furthermore, a number of studies showed that POU5F1(OCT4) can induce various cell fate reprogramming such as neural stem cells into iPSCs (Kim et al., 2009a;Kim et al., 2009b), and fibroblasts into neural progenitor cells (Mitchell et al., 2014b) as well as oligodendrocyte progenitor cells (Kim et al., 2015), defining POU5F1(OCT4) as a versatile reprogramming factor that confers the plasticity in somatic cells (Mitchell et al., 2014a;Mitchell et al., 2014b). Also, it has been reported that POU5F1(OCT4) binds to homeodomain transcription factor ISL1 (Boyer et al., 2005;Jung et al., 2010) which is required for specification of spinal cord MNs (Cho et al., 2014;Liang et al., 2011). So, we hypothesized that activation of POU5F1(OCT4) might have potential to generate MNs from somatic cells through regulating ISL1 expression. Here, we introduced the key cell fate regulator POU5F1(OCT4) and subsequently overexpressed additional MN specification factor LHX3 to induce fibroblasts toward motor neuronal fate. Importantly, we found that iMNs exhibited typical characteristics of MNs on molecular level, electrophysiological activity, synaptic functionality, in vivo engraftment capacity and therapeutic effects. In conclusion, our strategy enables large-scale production of pure iMNs and facilitates the feasibility of iMNs for SCI treatment. Access to high-yield cultures of human MNs will facilitate an in-depth study of MN subtype-specific properties, disease modeling, and development of cell-based drug screening assays for MN disorders.

Generation of induced motor neurons (iMNs) from human fibroblasts by sequential induction of two transcription factors
To generate induced motor neurons (iMNs) from human adult fibroblasts (HF1), we established an advanced direct conversion strategy by inducing POU5F1(OCT4) and LHX3 sequentially at different time points as described in the experimental scheme ( Figure 1A). To rule out the possibility that resulting iMNs are derived from contaminating neural cells in parental fibroblasts, we confirmed that none of the fibroblasts express neural lineage markers by immunostaining with specific antibodies (Figure 1-figure supplement 1A). Firstly, we transduced fibroblasts with POU5F1(OCT4) regarding its critical role in cell fate decision during early development (Yamada et al., 2013). POU5F1(OCT4)induced plasticity has been shown to activate the lineage genes in response to lineage supporting culture condition (Mitchell et al., 2014a;Mitchell et al., 2014b). We cultured POU5F1(OCT4)infected cells in our defined neural induction medium. The morphological changes were observed as   Figure 1 continued on next page early as day 3, and spindle-shaped cells then formed compact colonies by day 10 ( Figure 1C), whereas mock-infected fibroblasts did not change under the same condition ( Figure 1B). From 0.5 Â 10 4 seeded fibroblasts, we could observe 7~12 colonies (Supplementary file 3). We picked colonies individually and plated them in separate wells for clonal culture. The neural progenitor-like cells grew out gradually from attached colonies ( Figure 1D). We found that the number of cells increased over time in neural induction medium. These cells were passaged and maintained as a monolayer cell line which we named iMN-intermediate cells (iMNICs) ( Figure 1E). We could establish six iMNIC clones from HF1; HF1-iMNIC2, HF1-iMNIC5, HF1-iMNIC6, HF1-iMNIC7, HF1-iMNIC11 and HF1-iMNIC12 (Figure 1-figure supplement 2A and Supplementary file 3). In order to determine whether the fibroblasts acquired motor neuronal identity after POU5F1(OCT4)-mediated neural induction, we analyzed the relative mRNA level of MN specification genes ISL1, HB9, NKX6.1 and LHX3 (Davis-Dusenbery et al., 2014) in POU5F1(OCT4)-infected cells at day 7, day 14 and three iMNIC clones (Figure 1-figure supplement 2B). Notably, quantitative reverse transcription polymerase chain reaction (qRT-PCR) analysis revealed that only ISL1 was dramatically upregulated after POU5F1(OCT4) induction ( Figure 1-figure supplement 2B). Consistently, immunocytochemistry also showed that iMNICs robustly express ISL1, whereas other MN markers (HB9, NKX6.1 and LHX3) or neuronal markers (TUJ1 and MAP2) were not detected ( To induce iMNICs into mature iMNs, we plated iMNICs on PDL/Laminin-coated plate in MN induction medium containing spinal ventralizing morphogen, sonic hedgehog (SHH) (Ericson et al., 1996;Jessell, 2000). Most of the cells showed neuronal morphology expressing TUJ1 (>95%), and nearly all TUJI+ cells co-expressed ISL1 at day 28 (Figure 1-figure supplement 3A). These cells also expressed mature neuronal marker MAP2 and presynaptic marker synapsin1 (SYN1) (Figure 1figure supplement 3B). However, terminal MN marker HB9 expressing cells were very rare in population ( Figure 1-figure supplement 3C). These data indicate that POU5F1(OCT4) induction potentially contributes to MN induction, but not sufficient for complete reprogramming toward mature MNs.
Given that iMNICs express endogenous ISL1 gene, we hypothesized that additional transcription factor involved in MN specification might facilitate the conversion of iMNICs toward bona fide MNs. We selected six candidate transcription factors including PAX6, NKX6.1, OLIG2, LIM1 and LHX3 (Davis-Dusenbery et al., 2014;Jessell, 2000) and transduced iMNICs with each gene individually ( Figure 1I). Also, we employed sphere culture to optimize the culture condition for MN induction. When we plated iMNICs on non-coated dishes for sphere culture, the cells formed clusters and became free-floating iMN spheres in MN induction medium ( Figure 1F). The iMN spheres were subsequently transferred onto PDL/Laminin-coated dishes for neuronal maturation, then MN-like cells outgrew from the spheres ( Figure 1G). We could observe robust axon projections and dendritic arborizations, suggesting the neuronal maturation ( Figure 1G'). Remarkably, the converted neurons exclusively expressed TUJ1 and HB9 only after LHX3 induction, but no HB9+ cells were observed in other transcription factor infected cells ( Figure 1H). More than 96% of iMNs co-expressed TUJ1 and Figure 1 continued HB9+TUJ1+ iMNs. The cells were co-stained with HB9 (red) and TUJ1 (green), and the nuclei were counterstained with DAPI. Scale bars, 125 mm. (I-J) Conversion efficiency determined by quantification of TUJ1+ cells (I) and HB9+TUJ1+ cells (J) depending on transduced MN factor and culture condition. Data are presented as mean ± SD (n = 3), and represent triplicate experimental replicates. Related data can be found in Figure 1-figure supplements 1-6. The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. Quantification of TUJ1+/HB9+TUJ1+ iMNs depending on TF, qRT-PCR analysis of MN genes, and qRT-PCT analysis of endogenous expression of POU5F1, NANOG, and SOX2.      HB9 in sphere culture, while the efficiency was considerably lower in adherent culture ( Figure 1H-J and Figure 1-figure supplement 4A). We applied this to other iMNIC clones, and they were converted into HB9+TUJ1+ iMNs with conversion efficiency of 70~90% (Figure 1-figure supplement 4B and Supplementary file 4). Addition of other transcription factors rather impeded neuronal morphology, especially PAX6 and OLIG2 known to contribute to the early stage of MN specification even decreased the efficiency of TUJ1+ cells ( Figure 1I and Figure 1-figure supplement 4A). These results demonstrate that POU5F1(OCT4) and LHX3 with our defined culture condition are sufficient to convert cell fate of human fibroblasts into mature iMNs with high purity.
Direct conversion through self-renewing iMN-intermediate cells (iMNICs) enables large production of iMNs As described above, iMNICs exhibited neural progenitor-like morphology and proliferative capacity. It has been reported that POU5F1(OCT4) can directly convert somatic cells into tripotent neural progenitors (Mitchell et al., 2014b). Considering previous report, we evaluated cellular identity of iMNICs whether the cells are similar to neural progenitors. Immunocytochemistry analysis showed that none of iMNICs expressed early neuroectoderm markers (SOX1 and PAX6), neural progenitor markers (SOX2 and NCAM) or MN progenitor marker (OLIG2), and these markers were never detected during the entire neural induction process (Figure 1-figure supplement 5A). To determine whether this cell conversion occurred through pluripotent state or neural progenitor state, we evaluated the activation of endogenous expression of pluripotent genes (POU5F1(OCT4) and NANOG) and neural progenitor marker (SOX2) after POU5F1(OCT4) induction (Figure 1-figure supplement 5B). We found that none of these markers were detected in POU5F1(OCT)-induced cells, iMNICs and iMNs as confirmed by qRT-PCR. This result implies that fibroblasts transit neither pluripotent state nor neural progenitor state. Although iMNICs were distinct population from neural progenitors, these cells were highly proliferative cells and capable of being maintained for more than 13 passages. We examined the self-renewal capacity of iMNICs. We evaluated iMNICs at early passage (P2) and late passage (P13). Cellular morphology was very similar at both early and late passages ( Figure 1-figure supplement 5C). The cells sustained the proliferation rate presenting the mean doubling times (mDT) of 28.4 hr at P2 and 29.4 hr at P13, respectively (Figure 1-figure supplement 5D). In addition, we confirmed that >97% of iMNICs expressed proliferative cell marker KI67 (Figure 1-figure supplement 5E and F). To verify the continuous capacity of iMNICs to convert into iMNs over multiple passages, we conducted iMN induction using iMNICs at P2 and P13. iMNICs were successfully converted into mature iMNs at both early and late passage (Figure 1-figure supplement 5G). Furthermore, to determine whether generation of iMNICs by initial induction of POU5F1(OCT4) prior to LHX3 induction is necessary, we compared the cellular identity of iMNICs emerged after induction of POU5F1(OCT4) alone and co-induction of POU5F1(OCT4) and LHX3. When we infected fibroblasts with POU5F1(OCT4) and LHX3 simultaneously, the majority of infected cells died and iMNIC colonies were not stably maintained for multiple passages after picking (up to 5 passages) (Figure 1-figure supplement 6A). Proliferation rate and ISL1 gene expression were also decreased in POU5F1(OCT4)/LHX3 induced iMNICs compared to POU5F1(OCT4)-induced iMNICs (Figure 1-figure supplement 6B and C). In addition, induction of POU5F1(OCT4) alone showed higher efficiency in the number of iMNIC colonies (Figure 1-figure supplement 6D). These data indicate that the sequential induction of two transcription factors is essential for generating self-renewing iMNICs more efficiently. Together, these results demonstrate that iMNICs generated by POU5F1(OCT4) are expandable through self-renewing intermediate state which is distinct from neural progenitor state. Therefore, our method can facilitate the generation of iMNs on large-scale.

Characterization of iMNs
To monitor the cell fate conversion process in live culture and further characterization of iMNs, we employed reporter system using a lentivirus expressing the green fluorescent protein gene (GFP) under the control of the HB9 promoter (HB9:GFP) (Marchetto et al., 2008;Toli et al., 2015). After we transfected iMNICs with LHX3 together with HB9:GFP reporter lentivirus, GFP-positive cells gradually emerged from the cell clusters under SHH stimulation ( Figure 2A). HB9:GFP+ iMNs were capable of extending long projections ( Figure 2A). To identify whether these GFP+ iMNs possess the typical characteristics of MNs, we evaluated MN marker expression. Immunocytochemistry , suggesting that our reprogramming method is specific to MN lineage rather than other neuronal subtypes or neural progenitors. Consistent with the immunocytochemistry results, the endogenous mRNA levels of MN marker genes including LIM1, LHX3, NKX6.1, HB9 and ISL1 were upregulated in iMNs relative to fibroblasts ( Figure 2C). Furthermore, we could observe the synaptic activities of HB9:GFP+ neurites expressing presynaptic markers, SYN1 and synaptic vesicle protein 2 (SV2) ( Figure  . Those cells displaying multiple APs were further grouped based on the presence of spontaneous firing (group 1 vs. group 2). We first tested passive properties of neuronal membrane by injecting hyperpolarizing current. Both resting membrane potential and input resistance were comparable between group 1 and group 2 ( Figure 3A-D). Next, we compared the active properties of iMNs between the two groups by injecting depolarizing current ( Figure 3E). Likewise, there were no significant differences in AP threshold, frequency, amplitude, half-width and neuronal excitability between the two groups ( Figure 3F-J), suggesting that these cells are functionally matured regardless of spontaneous firing at resting membrane potential. In voltage clamp mode, depolarizing voltage steps also induced fast inward currents. These currents were completely blocked by the bath-application of tetrodotoxin (TTX). We confirmed that these inward currents were elicited by TTX-sensitive voltage-gated sodium channels ( Figure 3K-M). Together, these data indicate that iMNs are electrophysiologically mature.

iMNs form neuromuscular junctions with myotubes
To evaluate the in vitro functionality of iMNs, we investigated whether iMNs possess the ability to form neuromuscular junctions (NMJs) with muscles, which is the key feature of spinal MNs. We differentiated mouse myoblasts C2C12 into multinucleated myotubes and co-cultured with iMNs. Extensive axons of HB9:GFP+ iMNs projected along the myotubes ( Figure 4A). Noticeably, SV2-positive vesicles were seen in HB9:GFP+ iMNs and enriched at NMJs showing puncta-like structures ( Figure 4A). At the sites of contacts with myotubes, clusters of postsynaptic acetylcholine receptors (AChR) on the surface of myotubes were detected by immunocytochemistry with AChR antibody ( Figure 4B) and a-bungarotoxin conjugated with Alexa 555 (a-BTX) ( Figure 4C). The formation of NMJs was also observed with iMNs generated from iMNICs at late passage ( Figure 1-figure supplement 5I). Importantly, we could observe the rhythmic contraction of myotubes when co-cultured with iMNs. The contractions were blocked by antagonist of nicotinic AChR, curare, indicating that the myotubes contract depending on the activities of iMNs (Video 1). These data demonstrate that  Global transcriptional profiles of iMNICs and iMNs by RNAsequencing To examine the identity of iMNICs and iMNs, we performed RNA-sequencing (RNA-seq) analysis to compare global gene expression profiles of iMNICs and iMNs to that of parental fibroblasts (HF1 and HF2), fetal NPCs (NPC) (Kim et al., 2009a), wild type motor neurons (wtMNs) derived from human ESC (wtMN-1) (Amoroso et al., 2013) and iPSC (wtMN-2) (Ng et al., 2015) and fetal spinal cord (Fetal-SC) (Kumamaru et al., 2018). Heatmap analysis demonstrated that the global gene expression patterns of iMNICs and iMNs were similar to wtMNs, but distinct from fibroblasts and fetal NPCs ( Figure 5A). The hierarchical clustering and principal component analysis (PCA) showed that iMNICs, iMNs and wtMNs are tightly correlated, whereas distinct from NSCs ( Figure 5B and C).  Video 1. Contraction of C2C12 myotubes co-cultured with iMNs. C2C12 myotubes started rhythmic contraction after co-culture with iMN for 3 weeks. To block the acetylcholine receptor on the myotubes specifically, 100 mM curare (final concentration) was added to the culture.   Figure 5D). In contrast, we could observe low similarity between iMNICs vs NPC and parental fibroblasts vs iMNICs/iMNs, especially NPC marker genes (SOX2, OLIG2, PAX6 and SOX1) and fibroblast genes (COL5A1 and COL1A1) ( Figure 5D and Figure 5-figure supplement 1A). Furthermore, we could observe high level of similarity in HOX gene clusters and MN genes in our iMNICs and iMNs compared to wtMNs, which were not expressed in NPCs ( Figure 5E,F and Figure 5-figure supplement 1B). In addition, iMNICs and iMNs expressed similar level of HOXC clusters and ISL1 with wtMN-1, but SOX2 expression was not detected in iMNICs and iMNs ( Figure 5-figure supplement  1D). In contrast, the expression of NPC genes (SOX2, PAX6, OLIG2, ASCL1, FABP7 and EMX), fibroblast-specific genes (COL1A1, COL5A1, SNA1 and SNAI2) and pluripotent genes (NANOG, POU5F1 (OCT4) and TDGF1) were not detectable in iMNICs and iMNs ( Figure 5F and Figure 5-figure supplement 1C). To validate the RNA-sequencing data, we conducted qPCR to analyze mRNA expression of HOX genes in our iMNICs and iMNs. Consistent with RNA-sequencing data, cervical and thoracic spinal cord specific genes (HOX4-9) were upregulated in iMNICs and iMNs as confirmed by qRT-PCR ( Figure 5G). Together, these results indicate that iMNs generated from two fibroblast lines acquired motor neuronal identity and showed a high degree of similarity with wtMNs or fetal spinal cord tissues.

Therapeutic potential of iMNs in spinal cord injury (SCI) model in vivo
In order to examine the in vivo functionality and therapeutic effects of iMNs, we transplanted HB9: GFP+ iMNs into adult rat SCI models ( Figure 6A). We induced compressive damage to thoracic vertebrae 9 (T9) of the spinal cord and injected 1 Â 10 6 iMNs into the upper (T8) and lower (T10) vertebrate after 1 week of injury. We could observe the engraftment of transplanted HB9:GFP+ iMNs in the spinal cord ( Figure 6B). To evaluate the cellular features of iMNs in vivo, the spinal cord tissue was immunostained with neuronal marker (TUJ1) and oligodendrocyte marker (MBP). Most of the GFP+ cells co-expressed TUJ1 ( Figure 6C-E) and were surrounded by host myelinating MBP+ oligodendrocytes ( Figure 6F-H). To confirm the tissue recovery of the injury site, we conducted hematoxylin and eosin (H and E) staining of sagittal sections of the spinal cord after 8 weeks of transplantation to detect the cavity size. Transplanted tissue show less cavity of injured site compared to control ( Figure 6I and J). Next, we evaluated the motor function recovery of the hind limbs by measuring Basso-Beattie-Bresnahan (BBB) scores for 8 weeks in rat SCI models that were treated either with PBS (control) (n = 6) or with iMNs transplantation (n = 10). iMN-transplanted rats were improved in BBB scores from week 4 that persisted steadily through week showing significant improvement in locomotor recovery ( Figure 6K,L and Video 2). To assess the risk of tumor formation, iMNICs and iMNs were subcutaneously transplanted into immune-deficient nude mice and no tumors were observed for 12 months of experiment period (n = 4) ( Figure 6-figure supplement  1A). These results showed that iMN transplantation significantly improved functional recovery after SCI.  (HOXC4, HOXC5, HOXC6, HOXA7 and HOXC8) and thoracic HOX genes (HOXA9 and HOXC9) in HF1-iMNINs, HF1-iMNs, H9-MNs and NPCs. Bars represent fold changes relative to fibroblasts after normalization to GAPDH. Data are presented as means ± SD (n = 3). Related data can be found in Figure 5-figure supplement 1.
The online version of this article includes the following source data and figure supplement(s) for figure 5: Source data 1. qRT-PCR analysis of cervical and thoracic HOX genes.

Discussion
In this study, we established an advanced direct conversion strategy to generate induced motor neurons (iMNs) from human fibroblasts in large-scale with high purity, thereby providing a cell source for the treatment of spinal cord injury (SCI). Previously, we succeeded in generating oligodendrocyte progenitor cells (OPCs) from somatic cells by POU5F1(OCT4) (Kim et al., 2015). In line with this, we hypothesized that combination of POU5F1(OCT4) and defined supplements may convert cell fate toward motor neurons (MNs) which is developmentally derived from the same origin with OPCs, pMN progenitors of the ventral spinal cord (Ravanelli and Appel, 2015). POU5F1(OCT4) could efficiently generate TUJ1-positive neuronal cells under MN induction conditions, however, most of the cells could not reach to HB9 positive mature MNs. Nonetheless, we found that endogenous expression of ISL1, an important transcription factor for MN specification (Liang et al., 2011), was activated in fibroblasts after POU5F1(OCT4) induction under our defined culture condition. Based on this observation, we assumed that an additional transcription factor is required for complete iMN conversion. By screening the MN specification factors, we found that inclusion of LHX3 significantly increased the reprogramming efficiency of HB9 positive iMNs. This result is consistent with the previous report that the LIM complex composed of ISL1 and LHX3 specify spinal MNs by inducing MN gene HB9 in development (Lee et al., 2012). To our knowledge, this is the first time showing that POU5F1(OCT4) can initiate cellular reprogramming toward MN lineage by activating ISL1. This is correlated with previous reports that LIM homeodomain transcription factor ISL1 has been detected as POU5F1(OCT4) targets in human pluripotent cells (Boyer et al., 2005;Jung et al., 2010).
This result has an important implication on how we minimized the transcription factors for generating iMNs. Previously, iMNs were generated by either eight transcription factors (ASCL1, BRN2, MYT1L, LHX3, HB9, ISL1, NGN2, and NEUROD1) (Son et al., 2011) or four transcription factors (NGN2, SOX11, ISL1, and LHX3) (Liu et al., 2016). In ; contrast to previous methods, we identified the minimal two transcription factors, POU5F1(OCT4) and LHX3, for generating iMNs, so that our iMNs would be safer than multiple transcription factors-derived iMNs for therapeutic applications in terms of lower chance of viral integrations (Kim et al., 2009b). Moreover, previous studies have not verified the in vivo functionality or therapeutic potential of human iMNs.
In addition, RNA-sequencing analysis revealed that our iMNs exhibited transcriptional profiles of motor neuronal identity similar with wtMNs. Importantly, iMNICs and iMNs expressed spinal cord specific HOX5-9 gene clusters, whereas NPC genes were not detected in these cells. The ability to generate autologous iMNs with motor neuronal identity can facilitate disease modeling and cellbased therapy for MN diseases or spinal cord disorders. A critical prerequisite for achieving these goals is obtaining the pure population of subtype-specific neurons in high number. However, previous methods have not been feasible since directly converted neurons are terminally differentiated cells that are not scalable. This limitation could be overcome by generating expandable neural progenitors; however, neural progenitors could produce a mixture of heterogeneous neurons (Lujan et al., 2012;Ring et al., 2012;Thier et al., 2012). Herein, sequential introduction of two transcription factors enabled the large production of pure iMNs. We could obtain self-renewing cell line iMNICs after POU5F1(OCT4) induction by clonal reprogramming method as we have demonstrated previously in generating OPCs from fibroblasts (Kim et al., 2015). iMNICs were proliferative, expandable and retained conversion capacity to mature iMNs for long-term culture (>13 passages); however, they were distinct from neural progenitors. Therefore, our method is 'direct conversion' rather than indirect conversion or redifferentiation through pluripotent state or neural progenitor state. Previously, Mitchell et al. generated iNPC from POU5F1(OCT4)-induced fibroblasts by culturing the cells in reprogramming media (RM) for 8 days before inducing in classical NPC media (Mitchell et al., 2014a;Mitchell et al., 2014b). They reported that RM step is essential to induce SOX2+ NPCs after POU5F1(OCT4) induction. In contrast, SOX2 was not activated in POU5F1 (OCT4)-infected cells and failed to generate iNPCs when cultured directly in NPC media (Mitchell et al., 2014a). Consistent with previous report by Mitchell et al., our iMNICs which were generated without RM step did not express neural progenitor marker (Mitchell et al., 2014a;Mitchell et al., 2014b). This finding indicates that extracellular environment and timing for appropriate stimuli are critical for cell identification of POU5F1(OCT4)-induced cells.
Several studies have reported that transplantation of MNs or MN precursors differentiated from pluripotent stem cells (PSCs) showed functional benefits to injured spinal cord (Rossi et al., 2010;Wyatt et al., 2011). However, translation of these cells to the clinic is limited due to the challenges including differentiation efficiency, scalability, purity, and especially tumorigenecity from undifferentiated cells (Nori et al., 2015). Moreover, transplantation of MN precursors remain poorly defined because of stem cell heterogeneity (Trawczynski et al., 2019). In this study, we could obtain pure iMNs in large number and found that hindlimb functional recovery depends on maturity and purity of iMNs. When we transplanted selected HB9+ mature iMNs, BBB score was higher compared to transplanting iMN spheres. We also observed that our iMNs exhibited integration properties when transplanted into injured spinal cord. Transplanted HB9+ iMNs still expressed neuronal markers and interacted with host neural cells such as oligodendrocytes in vivo. We could observe ensheathment by myelinating host oligodendrocytes surrounding our iMNs. Notably, transplantation of iMNs contributed histological and functional recovery in SCI model without tumor formation, and this is the first report demonstrating the therapeutic effects of human iMNs in vivo.
In conclusion, this proof-of-concept study shows that our functional iMNs can be employed to cell-based therapy as an autologous cell source. iMNs possessed spinal cord motor neuronal identity and exhibit hallmarks of spinal MNs such as neuromuscular junction formation capacity and electrophysiological properties in vitro. Importantly, transplantation of iMNs improved locomotor function in rodent SCI model without tumor formation. Although further investigation on mechanism responsible for cell fate conversion may be needed, our strategy is a safer and simpler methodology that may provide new insights to develop personalized stem cell therapy and drug screening for MN diseases or spinal cord disorders.

Study plan and ethics
This study was designed to establish an advanced direct lineage reprogramming strategy to generate autologous iMNs from human fibroblasts by overexpressing POU5F1(OCT4) and LHX3. We further investigated the therapeutic effects of iMNs for treating traumatic spinal cord injury using rodent spinal cord injury model. The experiments were repeated at least three times, and the repli-

Virus construction and production
The cDNAs of candidate MN transcription factors were amplified by PCR, and individually subcloned into the lenitiviral vector backbone (Addgene ID#12252) (Warlich et al., 2011). Plasmids carrying human OCT4 and LHX3 cDNAs were purchased from Addgene (OCT4; ID#130692, LHX3; ID#120456). Also, plasmid carrying the HB9 promoter and GFP protein were purchased from Addgene (ID#37080) (Marchetto et al., 2008). The viruses were produced and harvested as previously described (Zaehres and Daley, 2006). In short, 293 T cells were seeded at 40-50% confluency on 10 cm plates prior to transfection. Individual transfer plasmids, packaging plasmid (PAX2, Addgene ID#12260), and envelope plasmid (VSV-G, Addgene ID#8454) were transfected into 293 T cells using X-treme GENE9 DNA transfection reagent (Roche) according to manufacturer's instructions. After 48 hr of transfection, virus containing supernatants from two 10 cm plates were collected and filtered through 0.45 mm membrane. Virus particles were concentrated by ultracentrifugation (1.5 hr at 80,000 g, 4˚C) and resuspended in 200 ml of fresh DMEM (virus soup). For cell conversion, we applied 15 ml of virus soup per well (six-well plates). Materials and reagents are available upon request.

Generation of iMNs
Human fibroblasts were seeded at 0.3-1 Â 10 4 cells on gelatin-coated 6-well plates. On the next day, the fibroblasts were infected with 2 ml of 15 ml virus soup (lentiviral vector carrying human OCT4) and fibroblast medium mixture containing 6 mg/ml protamine sulfate. The medium was replaced with fresh fibroblast medium after 24 hr of infection. At 3 days post-infection, the medium was switched to neural induction medium (DMEM/F12 (GIBCO) supplemented with N2 (GIBCO), penicillin/streptomycin, 10 ng/ml bFGF (Peprotech), and 10 ng/ml epidermal growth factor (EGF, Peprotech) and 10 ng/ml laminin). Morphologically changed infected cells (compact colonies of neural progenitor-like cells) were mechanically isolated by a glass micropipette and transferred into new wells individually. The cells were expanded in neural induction medium and iMNICs were established after 1 or 2 passages. For further MN induction, we used two methods, sphere culture and adherent culture.
[Adherent culture method] iMNICs were seeded at 5 Â 10 4 cells on PDL/laminin-coated plate in neural induction medium. On the next day, seeded iMNICs were infected with 2 ml of 15 ml virus soup (lentiviral vector carrying human LHX3) and neural induction medium mixture containing 6 mg/ ml protamine sulfate. After 24 hr, the medium was switched into MN induction medium for 7 days. In this process, the MN spheres did not appear. We switched the medium into MN maturation medium and matured for additional 7-14 days.
C2C12 myotube co-culture C2C12 myoblasts (ATCC) were expanded in DMEM with 10% FBS and penicillin/streptomycin. When the culture reached 70% confluency, the medium was switched to 2% horse serum containing medium to induce multinucleated myotubes. iMNs were added to the myotubes in motor neuron medium to induce formation of neuromuscular junctions and spontaneous contractions. After 3-4 weeks, the myotube contractions were observed under the microscope and were inhibited by adding 100 mM curare (Sigma). Neuromuscular junctions were observed by labeling with a-bungarotoxin conjugated with Alexa 555 (Invitrogen, 1:200) and immunostaining with myosin heavy chain (MHC, DSHB), acetylcholine receptor (AChR, DSHB), and synaptic vesicle 2 (SV2, DSHB).

RT-PCR and quantitative RT-PCR
DNA-free total RNA of iMNs was extracted using the RNeasy mini kit (Qiagen). Total RNA (500 ng) was used to synthesize cDNAs using SuperScript III reverse transcriptase (Invitrogen). RT-PCR was performed using recombinant Taq DNA polymerase (Invitrogen). qRT-PCR analysis was conducted on a LightCycler 480 instrument with SYBR Green I Master mix (Roche). The experiments were performed in triplicate, and expression was normalized to the housekeeping gene GAPDH. Gene expression was measured by calculating Ct values. All of the experiments were conducted according to the manufacturer's instructions. The sequences of the primers used are listed in Supplementary file 2.

Electrophysiology
Motor neurons induced from human fibroblasts were placed in a recording chamber and recognized visually by IR-DIC optics. Induced motor neurons were continuously perfused with artificial cerebrospinal fluid (ACSF) containing 125 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 25 mM NaHCO3, 15 mM glucose, 2 mM CaCl2, and 1 mM MgCl2 oxygenated with 95% O2% and 5% CO2 at 30-32˚C. Borosilicate glass pipettes (2.5-3.5 MW) were pulled (P-1000, Sutter Instrument) and filled with potassium-based internal solution containing 133 mM KMeSO3, 3 mM KCl, 10 mM HEPES, 1 mM EGTA, 0.1 mM CaCl2, 8 mM Na2-phosphocreatine, 4 mM Mg-ATP, 0.3 mM Na3-GTP (290-300 mOsm, pH 7.3 with KOH) to make whole-cell configuration. Whole-cell patch clamp recording was performed by using Multiclamp 700B (Molecular Devices) and recording signals were filtered at 2 kHz, digitized at 10 kHz (PCI-6221, National Instruments). Recording data were monitored, acquired by WinWCP (Strathclyde Electrophysiology Software) and further analyzed offline by Clampfit 10.0 (Molecular Devices) and Prism 7.0 (GraphPad). To evaluate passive membrane properties of iMNs, hyperpolarizing step current (50 pA increment, 1000 ms duration) was injected in current clamp mode. To check active membrane properties of iMNs and generate AP, depolarizing step current (50 pA increment, 1000 ms duration) was injected. Electrophysiological properties of AP (threshold, frequency, amplitude, half-width) were analyzed from AP firings induced by 50 pA current injection. To isolate sodium current, whole-cell recording was conducted under voltage clamp mode and voltage was stepped from a holding potential of À70 mV to test potentials from À70 to À25 mV in 5 mV increments (200 ms duration). After recording sodium currents, tetrodotoxin (1 mM, Tocris) was bath-applied and voltage steps were repeated to confirm TTX-sensitivity of sodium currents measured.

Growth curve and mean doubling time
Motor neuron intermediate cells (1 Â 10 4 cells) at P2 and P13 were seeded onto 12-well plates and cultivated for 10 days. The cells were collected from triplicate wells and manually counted every 24 hr using a hemacytometer (Marienfeld). The average cell numbers on each day were plotted, and the mean doubling time (mDT) was calculated based on the growth curve.

Establish of compression spinal cord injury model
All animal experimental procedures were approved by Institutional Animal Care and Use Committee of Yonsei University College of Medicine (Seoul, Korea). Adult male Sprague-Dawley rats (Orient, Seongnam, Korea), weighing 200-220 g were used. Rats were housed in individual cages under standard laboratory conditions of 24-3C and 40-60% humidity, with 12 hr light-12 hr dark cycles, and enrichment conditions. All animals anesthetized with ketamine (100 mg/kg; Yuhan, Korea), xylazine (10 mg/kg; Bayer korea, Korea) and isotropy 100 (Troikaa Pharmaceuticals Ltd, India). A laminectomy was performed to expose the spinal cord, and the spinal cord was compressed at thoracic level 9 using self-closing forceps (Fine Science Tools, Canada) for 40 s. Following the injury, the muscle and skin were sutured with 3-0 Vicryl (Johnson and Johnson, Peterborough, Canada). Body temperature was maintained constant at 37C with a heating pad during surgery and the recovery period. After the procedure, cefazolin (25 mg/kg; Chong Kun Dang, Korea) was injected for 5 days. For immune suppression, Cyclosporine A (10 mg/kg, Chong Kun Dang, Korea) was administered to all animals until sacrifice. Also, animals were taken care of bladder system for urination.

Behavior test
All animals underwent behavioral analysis every week for 8 weeks after SCI to measure locomotor recovery. The Basso, Beattie, and Bresnahan (BBB) motor score was used to evaluate the quality of hind limb movement during open field locomotion. In the first recovery phase, the range of joint movement and the presence of the foot closure on the floor were checked. In the second phase, recovery of weighted stepping was observed. In the third phase, gait coordination and tail movement were observed.

Teratoma formation assay
All mice were purchased from Hyochang Science (Daegu, Korea). Animal handling was in accordance with animal protection guideline of Ulsan National Institute of Science and Technology (Ulsan, Korea). Teratoma formation assay was performed by subcutaneously injecting iMNICs (n = 4) and iMNs (n = 4) respectively on dorsal flank of athymic nude mice (1 Â 10 6 cells/mouse). After 12 months after injection, mice were sacrificed for analysis of teratoma formation.

Isolation of iMNs
To isolate pure iMNs for analysis, we designed the plasmid containing puromycin resistance gene under expression of HB9 promoter. After transfection, cells expressing HB9 were selected by culturing in the presence of 0.5 mg/ml of puromycin for 5 days.

Histology and immunohistochemistry (IHC)
For histological analysis, the rats were deeply anesthetized and perfused with PBS followed by 4% paraformaldehyde as well. The spinal cords were isolated and post-fixed in 4% paraformaldehyde overnight, and then immersed in 30% sucrose for 3 days. The tissues were embedded in frozen section compound (Leica) and sectioned at 16 mm in sagittal plane by a cryostat. IHC was performed as previously described (Kim et al., 2015). The primary antibodies used for IHC are listed in Supplementary file 1.
For hematoxylin and eosin (H and E) staining, the spinal cords were embedded in paraffin blocks. The paraffin blocks were sectioned at 4 mm in sagittal plane. Sections were immersed in Harris hematoxylin solution (Sigma) for 2 min to stain nucleus. Slides were then immersed briefly in 1% acid alcohol (1% HCl in 70% ethanol) and blued in 0.2% ammonium hydroxide, followed by staining with eosin Y solution (Sigma) for 30 s. Each step was followed by several washings with distilled water. The slides were dehydrated with ethanol, cleared with xylene, and mounted with mounting solution (Leica).

Statistical analysis
All data in this article are presented as the means ± SD (standard deviation). Data from at least three independent samples were used for statistical analysis. ANOVA with post hoc testing was performed to compare BBB score. A p-value less than 0.05 was considered statistically significant. Statistical analysis was carried out using EXCEL, and SigmaPlot software.

RNA-seq preparation and data analysis
Total RNAs were extracted from cells using the RNeasy mini kit (Qiagen) according to manufacturer's instructions. The quality of RNA was examined using Agilent 2100 Bioanalyzer. RNA integrity number (RIN) of all samples were higher than 8. Library sequencing was carried out on NovaSeq 6000 instrument. We generated 100 bp paired-end reads, with each library sequenced up to depth of 40 million fragments. We used HISAT2 (Pertea et al., 2016) to align the RNA-seq reads to the human reference genome GRCh38, and Cufflinks (Trapnell et al., 2012) to annotate them. We calculated the counts of aligned reads to each gene with HTSeq (Anders et al., 2015). We equalized the data and stabilized them through the log2 transform of the data plus one. RNA-seq data sequence summary is provided in Supplementary file 5, and related codes are provided in Supplementary files 6 and 7.

RNA-seq data integration analysis
To complement the RNA-seq data generated in this work, we collected RNA-seq data from the Sequence Read Archive (SRA) database for wild-type motor neurons MN, wtMN-2 (SRR2038215) (Ng et al., 2015), wtMN-1 (SRR606336) (Amoroso et al., 2013) and human ESC H9 (SRR3647179) (Kumamaru et al., 2018). As with our own samples, we used HISAT2 to align the RNA-seq reads to the human reference genome GRCh38, Cufflinks to annotate the mapped reads, and HTSeq to calculate the counts. We equalized the data and stabilized them through the log2 transform of the data plus one. To reduce the batch effect of the data integration, we used ComBat (Johnson et al., 2007). We used in-house software to merge the expression results into a single text file used in the downstream analysis in Matlab (MathWorks).

Transcriptomics global analysis
The heatmap of the most highly variable transcripts, the hierarchical clustering dendrograms (calculated using the unweighted pair group method with arithmetic mean and Euclidean distance measure), and the Principal Component Analysis (PCA) were performed using in-house functions developed in Matlab (MathWorks).

Gene-coverage count track plots
We sorted the alignment bam files with samtools (Li et al., 2009) and produced the bed files with bedtools (Quinlan and Hall, 2010). We developed a function in Matlab (MathWorks) that for each gene of interest takes the exon boundary information from the basic annotation file in gtf format from Gencode (https://www.gencodegenes.org/human/) version 33, and plots the gene-coverage count track plots, preserving the same scale for the tracks of the same gene in all the samples.

Data availability
The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE149664 .  . Supplementary file 6. Code for alignment and obtained alignment rates.

Decision letter and Author response
. Supplementary file 7. Code for obtaining genes counts and obtained statistics.
. Transparent reporting form

Data availability
The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE149664 (https://www.ncbi.nlm.nih.gov/ geo/query/acc.cgi?acc=GSE149664). Source data files have been provided for Figure 1, 2, and 5.
The following dataset was generated: The following previously published datasets were used: