Human neuroepithelial stem cell regional specificity enables spinal cord repair through a relay circuit

Traumatic spinal cord injury results in persistent disability due to disconnection of surviving neural elements. Neural stem cell transplantation has been proposed as a therapeutic option, but optimal cell type and mechanistic aspects remain poorly defined. Here, we describe robust engraftment into lesioned immunodeficient mice of human neuroepithelial stem cells derived from the developing spinal cord and maintained in self-renewing adherent conditions for long periods. Extensive elongation of both graft and host axons occurs. Improved functional recovery after transplantation depends on neural relay function through the grafted neurons, requires the matching of neural identity to the anatomical site of injury, and is accompanied by expression of specific marker proteins. Thus, human neuroepithelial stem cells may provide an anatomically specific relay function for spinal cord injury recovery.

T raumatic spinal cord (SC) damage results in cell loss at the injury level, as well as disconnection of surviving neurons, with an irreversible interruption of the information flow to and from the brain. The implantation of neural stem cells (NSCs) at the lesion site has been considered an appealing potential treatment for decades, and several approaches have been proposed. Mechanistically, the hypothesized benefits of transplantation are diverse, including replacement of lost neurons, creation of a conducive axon growth environment for host axons, production of growth factors, and provision of glial cells to assist function of surviving neurons. In order for these mechanisms to occur, graft integration into the host is critical and defining the parameters that regulate its success is fundamental to facilitate translation of cell-based therapies to the clinic. Unfortunately, at present, neither the identity nor the selection path for the most appropriate cell population for optimal graft integration are known.
Human NSC transplants for spinal cord injury (SCI) have been limited to partially characterized human cell lines [1][2][3] or to fetal NSCs collected after 8 post-conceptional weeks (PCW) [4][5][6] . Although fetal NSCs can be propagated in vitro, neither their long-term stability nor the preservation of their regional identity in vivo have been demonstrated 7 . Fetal NSCs exhibit molecular markers suggestive of radial glia and appear to differentiate more easily toward the glial fate, whereas their neurogenic potential is largely restricted to GABAergic neurons both in vitro and in vivo 7,8 . In most previous reports, NSCs were cultured in suspension as neurospheres, a method that often leads to a significant reduction in self-renewal competency and in the neurogenic capacity of the cells 9,10 .
As an alternative, human embryonic stem (ES) or induced pluripotent stem (iPS) cells are an in vitro source of neural progenitors and their application to SCI treatment is currently being investigated [11][12][13][14] . During human pluripotent stem cell differentiation, neural progenitors exhibit spontaneous selforganization into transient structures termed "rosettes". Cells within rosettes exhibit morphological and gene expression markers of neuroepithelial progenitors and are molecularly distinct from radial glia-like NSCs 15 . However, the identity and the physiological relevance of cells derived in vitro from pluripotent sources are unclear because cells could acquire transcriptional and epigenetic programs in vitro that diverge from cell states in vivo 16 .
To understand how regional cell identity affects graft integration, we analyzed the engraftment of a novel human NSC population that retains over time the transcriptional profile acquired in vivo. In contrast to other NSC sources, human neuroepithelial stem (NES) cells derived from tissues collected at an embryonic stage of the neural tube development, typically from 5 to 8 PCW, possess unique advantages. NES cells can be propagated as monolayers for a virtually unlimited number of passages, retain a high and unaltered neurogenic potential over time and preserve the molecular and transcriptional signature of their tissue of origin 17,18 .
We derived SC-NES cells from human post-mortem specimens and propagated them without genetic manipulation. Human SC-NES cells exhibited excellent integration properties in a rodent SCI model and established functional connections with local neurons. Through the application of chemogenetics to diverse behavioral paradigms, we show that SC-NES cells form a relay system through the lesioned area reconnecting spared host neural elements. In contrast, NES cells derived from neocortex (NCX-NES cells) fail to acquire a mature neuronal phenotype when transplanted into SC, fail to integrate and fail to extend neurites. Importantly, NCX-NES cell integration is dramatically enhanced in the cerebral cortex, demonstrating that anatomical matching of graft with recipient tissue is critical for functional neuronal networks. These findings provide key mechanistic, molecular and practical information to develop human cell transplantation therapy for SCI.

Results
Human SC-NES cells are tripotent and highly neurogenic. Here we derived human SC-NES cells from six embryonic postmortem specimens in a range of 5-8 PCW (Fig. 1a) 18 . The SC samples ( Supplementary Fig. 1a) were dissected free of meninges and dorsal root ganglia and dissociated to a singlecell suspension. After 24 h of plating, SC-NES cells formed neural rosettes with typical radial organization and apico-basal orientation of the developing neural tube (Fig. 1a and Supplementary Fig. 1b). SC-NES cells exhibit progenitor cell characteristics and express the canonical NSC markers nestin, SOX2, vimentin, phospho-vimentin and PAX6 ( Fig. 1b-d, f). KI67 staining confirmed the state of active proliferation of SC-NES cells in vitro ( Supplementary Fig. 1c). Next, we investigated whether SC-NES cells retain their positional identity after long-term expansion. Therefore, we verified the expression of the caudal transcription factor HOXB9 (Fig. 1e) and confirmed immunolabelling specificity by lack of HOXB9 expression in NCX-NES cells ( Supplementary Fig. 1d). Self-renewing SC-NES cells were maintained for >35 passages with no signs of senescence or chromosomal instability (Fig. 1g).
A 30-day differentiation protocol involving mitogen withdrawal coupled with neurotrophin addition induced the acquisition of a neuronal phenotype. SC-NES cells gave rise to RBFOX3-(also known as NeuN) positive neurons (Supplementary Fig. 1e) accounting for 79.2 ± 0.1% of cells (Supplementary Fig. 1i). Neurons also expressed the pan-neuronal markers MAP2, neurofilament (NEFL) and TUBB3 (Fig. 1h-i) whereas 11.2 ± 0.2% of cells were positive for the early postmitotic neuronal marker doublecortin (DCX) ( Supplementary  Fig. 1f, i). Among mature neurons, 1% of the cells expressed choline acetyltransferase (CHAT), a feature of spinal motor neurons, in agreement with SC-NES cell anatomical origin ( Fig. 1i Supplementary Fig. 1i). A subpopulation of cells expressed astrocyte and oligodendrocyte markers including GFAP and O4 (Supplementary Fig. 1g-i), confirming SC-NES cell tri-potent differentiation properties. In addition, we tested SC-NES-derived neurons for the expression of pre-and postsynaptic markers, PSD-95 and synaptophysin (SYP) (Fig. 1j, k), and for electrophysiological properties. Upon 90 days of in vitro differentiation SC-NES-derived neurons displayed ionic currents and elicit spontaneous repetitive action potentials (Fig. 1l, m), consistent with a fully functional neuronal phenotype (Supplementary Table 1).
Human SC-NES cells survive and differentiate in a SCI model. Human SC-NES cells were transduced with GFP-coding lentiviral vector and grafted into the lesion site of genetically immunodeficient mice 10 days after thoracic SC dorsal hemisection (Fig. 2a). Anatomical analysis revealed that SC-NES cells survived in all grafted animals and consistently filled the lesion cavity at 8 weeks post-grafting (Fig. 2b, c). Grafts in 3 out of 9 animals exhibited a rift near the graft center ( Supplementary Fig. 2a) which contained collagen and SOX10-positive host Schwann cells ( Supplementary  Fig. 2b-c). The rift separated the graft into two segments and contained very few grafted-cell-derived neurites ( Supplementary  Fig. 2d).
In accordance with literature for other transplants 6,11,19,20 , we observed that some grafted cells migrated into host SC and formed satellite clusters. We analyzed the SC from the brainstem to conus medullaris, and found that 5 out of 7 animals exhibited ectopic nodules. The vast majority of clusters were found just below the meninges ( Supplementary Fig. 2e) within 11 mm from the graft center. Dispersion of cells into parenchyma occurred in one single recipient animal at a 100-200 µm distance (Supplementary Fig. 2f). In addition, some grafted cells were located in the central canal ( Supplementary Fig. 3c) at 7-12 mm from the injection site.
A subset of grafted cells (6.6 ± 1.7%) were positive for KI67 or expressed nestin and DCX, thus indicating that a fraction of cells remained in active phases of the cell cycle after 8 weeks ( Supplementary Fig. 2g-i). However, the percentage of proliferating graft cells was not higher than the fraction of mitotic cells present in uninjured SC, which accounts for 10% of cells including central canal, gray and white matter 21 . Of note, SC-NES cells retained their regional identity in vivo and expressed the SCspecific transcription factor HOXB9 ( Supplementary Fig. 2j-l).
Human SC-NES cells elongate long distance axons in the host. Graft-derived neurons extended many projections into host SC ( Supplementary Fig. 3a). Emerging GFP-labeled processes were observed mostly in white matter (69.8 ± 3.5%) rather than in gray matter (30.1 ± 3.5%) (Fig. 3a, b), proving that SC-NES cells are not affected by myelin-associated inhibitors present in adult recipient tissues. We verified that SC-NES cell-derived neurons   do not express Nogo receptor 1 (NgR1), which is known to mediate axonal growth repression exerted by ligands Nogo A, Myelin Associated Glycoprotein (MAG) and Oligodendrocyte Myelin Glycoprotein (OMgp) [22][23][24][25] , thus providing a possible mechanism for the robust axonal growth observed in white matter ( Supplementary Fig. 3b). Notably, GFP-positive projections grew in parallel linear trajectories in white matter and were ramified in gray matter within close proximity to host neurons in a pattern that is very similar to endogenous axonal fibers ( Fig. 3c-e). The axonal nature of projections was confirmed by GFP colocalization with the axon-specific marker, human neurofilament (NEFL) (Fig. 3f, g). SC-NES cell-derived axons did not appear to be myelinated by host oligodendrocytes (Fig. 3h). The absence of myelination of human axonal fibers in rodent hosts has also been reported by others 11 and may be attributable to a lack of inter-species recognition.
Human SC-NES cell-derived axons extended over the entire length of the SC from the injection site in the thoracic segment. GFP-labeled terminals were observed cervically up to the medulla oblongata and caudally in the conus medullaris ( Fig. 3i-  Thus, human SC-NES cells exhibit excellent survival properties and appear impervious to myelin-associated inhibitors with consequent growth of long-distance projections to re-establish a connection between the SC segments above and below the trauma. Human SC-NES cells establish a functional neuronal network. We sought to determine whether SC-NES cells form neuronal networks within the host. To this purpose we labeled grafts with anti-human synaptophysin (HSYP) and examined RBFOX3positive neurons in the gray matter of recipient SC (Fig. 4a). We found HSYP-positive puncta surrounding host neurons, consistent with synapses between grafted cells and the host (Fig. 4b). Similarly, we investigated the formation of synaptic connections between host fibers and the graft. In particular, we analyzed two descending pathways: the corticospinal tract (CST) and the raphespinal tract. Biotin dextran amine (BDA)-labeled CST fibers regenerated into the graft (Fig. 4c, d) and colocalized with the pre-synaptic marker SYP, as well as the GFP-positive human projections, thus establishing a mechanism for host-graft connectivity (Fig. 4e). No CST axons extended beyond the graft into the caudal SC. Raphespinal 5-HT-positive fibers were also found to grow into the graft (Fig. 4f, g) and formed synapses with human cells (Fig. 4h). Notably, 5-HT fibers were observed throughout the entire graft ( Fig. 4j-l).
Grafted animals were also subject to functional analysis. Behavioral outcomes of control group and SC-NES cell recipients were measured by Basso Mouse Scale (BMS) starting 7 days after injury, but prior to engraftment 26 . At the pre-graft evaluation, both groups had substantial loss of locomotor function with an average BMS score <1 (slight ankle movement). The SC-NES cell recipient animals showed a mild improvement by weeks 7 and 8 with a final score of 4 (occasional stepping), whereas control animals scored 3 (paw placement with or without weight support) (Fig. 5a). We determined the extent of tissue spared from injury, and confirmed that the lesion was comparable between controls and SC-NES grafted mice (Fig. 5b). Next, we performed a correlation analysis between spared tissue and BMS score at 8 weeks. We observed a statistically significant difference in the  linear regression between the two groups indicating that cell recipient mice had better locomotor performance than control animals with a comparable injury (Fig. 5c). The correlation analysis revealed that the divergence in the behavioral outcome was most prominent for animals with mild to moderate injury, defined as spared tissue >25% (Fig. 5c). Therefore, we stratified the behavioral data, including mice with a lesion not greater than the 75%. For mice with such mild to moderate injuries, there was significantly greater performance by cell recipient animals with a BMS score of 5 (frequent or consistent plantar stepping) by week 8 post-implant, as compared to similarly injured control animals scoring <3 (Fig. 5d). The beneficial effect of neural grafts within the injured SC has been associated with diverse mechanisms. However, at present, no direct evidence has been provided to demonstrate that neural grafts can generate novel circuits within recipient SC. Here, to assess the necessity for SC-NES cell neural relay function, we transiently silenced their electrical activity using the hM4Di-DREADD receptor. hM4Di is an engineered human muscarinic receptor designed to be responsive to the otherwise inert ligand Clozapine-N-Oxide (CNO) and mediates neuronal membrane hyperpolarization with the consequent reduction in action potential firing 27,28 . First, we assessed the efficacy of hM4Di-CNO for SC-NES cells in vitro. We differentiated SC-NES cells to achieve firing of spontaneous action potentials, then, we transduced SC-NES cell-derived neurons with an hM4Dimcherry lentiviral construct ( Supplementary Fig. 4a). Application of CNO caused significant hyperpolarization of membrane potential and decrease in action potential frequency (Supplementary Fig. 4b-d). SC-NES cells expressing hM4Di-mcherry were then transplanted into hemisected SCs of immunodeficient mice. Animals were assessed by BMS scale at 8 weeks postimplantation before CNO administration (Pre-CNO), 30 min after injection (CNO) and 24 h later (wash-out). Upon CNO administration, 6 out of 8 animals showed a significant decline in locomotor skills, with restoration of function after CNO wash-out (Fig. 5e, f). Control lesioned animals underwent identical functional testing but did not exhibit any significant variation in their performance with CNO treatment (Fig. 5f).
The same cohort of grafted and control mice was also evaluated by grid-walking test where animals are assessed for their ability to walk on a suspended grid by quantifying the percentage of incorrectly placed steps 29 . SC-NES cell grafted animals showed a significantly greater percentage of missed steps during CNO treatment as compared to pre-CNO and post-CNO wash-out performances (Fig. 5g). For control animals there was no detectable effect of drug on the grid-walking competency ( Supplementary Fig. 4e).
Finally, gait was assessed by the foot-print test 29 . We measured the distance between the hindlimbs (base of support) and the stride length (Fig. 5h). Paired testing revealed that, compared to the pre-CNO assessment, drug treatment resulted in a significant reduction of the stride length for SC-NES cell recipient animals with a recovery after wash-out ( Fig. 5i-j). CNO did not alter control animals' performance ( Supplementary Fig. 4f). Base of support was not changed by CNO administration in either group (Fig. 5k).
These data show that the transient silencing of SC-NES cells abrogates the beneficial effect of transplantation in three different functional paradigms. The temporary interruption of SC-NES cell electrical activity through a pharmaco-genetic tool has a direct impact on the functional outcome of grafted mice thus providing evidence that while SC-NES cells can act as a scaffold for severed axons to regenerate, they can also create a relay system essential for functionally reconnecting supraspinal axons with denervated target neurons below the level of the trauma.
Human NES cell regional origin effects graft integration. Next, we explored whether the beneficial effects of SC-NES cells were related to the regional identity of the cells. To this purpose, we compared human SC-NES cells with human NES cells derived from the NCX of an age-matched embryo. Upon in vitro differentiation, SC-and NCX-NES cells appeared indistinguishable: both displayed a mature neuronal phenotype and gave rise to a similar number of RBFOX3-positive neurons ( Supplementary  Fig. 5a-c) 18 . However, when transplanted in the injured SC, NCX-NES cells did not integrate in the parenchyma and formed dense clusters at the injection site with drastically reduced axonal elongation compared to their SC-derived counterparts (Fig. 6a-c). Notably, SC-NES cells distributed more broadly into SC tissue over a much larger area (Fig. 6d-f). The significant reduction in the volume of NCX-NES cell grafts did not appear to be associated with a higher number of apoptotic events as no statistical difference was observed in the quantification of active caspase 3 positive-cells in both SC-and NCX-NES cell implants (Supplementary Fig. 5d). In addition, KI67 staining performed at an early stage of in vivo cell differentiation (i.e., 2 weeks post-implantation) did not reveal a significant difference between the two cell types ( Supplementary Fig. 5e), thus excluding enhanced mitosis as a possible mechanism behind the larger volume of SC-NES cell transplants. NCX-NES cells in vivo also appeared to be substantially undifferentiated. Immunostaining for pan-neuronal markers RBFOX3 and TUBB3 revealed no colocalization with NCX-NES cells (Fig. 6g-h), whereas the vast majority of the cells remained immuno-positive for the neural progenitor marker nestin (Supplementary Fig. 5f-g).
We analyzed interactions between NCX grafts and the host, but found no BDA-labeled host CST fibers establishing connections with grafted NCX-NES cells ( Supplementary Fig. 5h). Unlike SCderived NES cells, NCX-NES cells failed to produce any functional improvement in recipient animals over 8 weeks (Fig. 6i and Supplementary Fig. 5i). This cannot not be attributed to greater injury severity, since spared tissue was similar in graft recipients and controls (Fig. 6j).
Of note, when placed into the motor cortex of immunodepressed mice, NCX-NES cells displayed a mature morphology at 8 weeks after engraftment with highly branched neurites and a wide distribution in the tissue (Fig. 6k, m). In the same location, SC-NES cells formed small tight clusters with short linear projections and a drastically reduced spreading in the parenchyma (Fig. 6l-m).
Our data show that the integration of human NES cells into a SCI model is strictly related to cellular regional identity and that the anatomical match of neural implants with the recipient tissue is a fundamental requirement for connectivity.

Global transcriptome analysis of grafted human SC-NES cells.
To analyze molecular differences between SC-and NCX-NES cells associated with their radically different adaptation in the lesioned SC, we performed a global RNA-sequencing (RNA-seq) analysis of tissue from both SC-and NCX-NES grafts. In order to segregate intrinsic from extrinsic factors contributing to graft integration, differentially expressed (DEX) genes were also compared to the transcriptional profile of SC-and NCX-NES cells in vitro before and after neural differentiation (Fig. 7a). Principal component analysis (PCA) of SC-and NCX-NES cells confirmed a clear separation of groups (Fig. 7b). In particular, we identified a first principal component (PC1) segregating in vitro from in vivo samples, whereas a third principal component (PC3) separates samples based on anatomical derivation ( Supplementary Fig. 6ab). Gene-ontology analysis showed that DEX genes upregulated in SC-NES cell grafts belonged to the "neurogenesis" ontology  cluster (GO: 0007399; P value 1.3 × 10 −2 ), supporting our histological data reporting the in vivo neural differentiation of SC-NES cells compared to undifferentiated NCX-NES cells. Among the top 50 up-regulated genes in the SC-NES cell grafts were genes related to neuronal maturation, such as maturin (MTURN) 30 and caytaxin (ATCAY) 31 , as well as genes involved in axonal elongation (pleiotrophin, PTN) 32 and cell adhesion (anosmin, KAL1) 33 . The maturation of SC-NES cells was also confirmed by the up-regulation of synaptic genes SYT4 and SYT13, reinforcing our observation of graft-host interaction (Fig. 7c). Differential expression of SC-and NCX-positional genes, such as HOXB4 and HOXB9 or FOXG1 and OTX2 respectively, confirmed the longterm preservation of positional identity after transplantation (Fig. 7c). Interestingly, while MTURN and ATCAY were upregulated by in vitro NCX-NES cell-derived neurons, their expression was strongly reduced in the grafts. On the other hand, the expression of these genes was maintained by SC-NES cells during engraftment, with no significant difference between SC-NES cell-derived neurons and SC-NES cell grafts (Fig. 7d). PTN expression was drastically reduced in NCX-NES cell grafts compared to NCX-NES cell-derived neurons, with less reduction observed in SC-NES cell grafts compared to the same cells after in vitro neural differentiation (Fig. 7d). Finally, we validated our transcriptome analysis with immunostainings for MTURN and PTN. The cytosolic protein MTURN was visualized in the soma and neurites of grafted SC-NES cells (Fig. 7e-f), whereas PTN was secreted into the extracellular environment ( Fig. 7i-j). Neither protein was detected in the NCX-NES cell grafts (Fig. 7g-l).
To extract additional biologically relevant information, we applied weighted gene co-expression correlation network analysis (WGCNA) to identify gene modules with similar variation across in vitro and in vivo samples and identified overall 23 modules ( Supplementary Fig. 7, Supplementary Data 1). We observed that module 1 (M1) identifies genes upregulated in grafted cells from both NCX and SC, suggesting that implantation itself impacts NES cell transcriptional profile. Genes in M9 were upregulated in grafted SC-NES cells and mostly related to synaptic signaling (GO: 0099536; P value 4.48 × 10 −14 ), ion transmembrane transport (GO: 0034220; P value 3.48 × 10 −10 ), neurogenesis (GO; 0048666 P value 1.3 × 10 −6 ) and neuron projection development (GO: 0031175; P value 4.43 × 10 −6 ). Module M8, on the other hand, clustered genes related to forebrain neuronal differentiation, revealing that while NCX-NES cells have the potential to differentiate, they fail to do so upon implantation into SC.
Thus, global RNA profiling provides strong evidence for selective integration into SC of human SC-NES cells as compared to NCX-NES cells. The findings indicate that upregulation of genes involved in neural precursor maturation, neurite extension, and synapse formation plays a crucial role in graft integration, and that host environment determines the transcriptional profile of implanted cells. Therefore, SC-NES cell integration results from both intrinsic and extrinsic factors, with an anatomical match of grafted cells to recipient tissue being required for successful SCI treatment.
Transplantation in a model of spinal cord contusion. To explore the translational impact of our findings, we challenged SC-derived NSCs with an allograft procedure in a clinically relevant model of SCI. For this purpose, we derived NSCs from the embryonic SC primordia of Sprague Dawley rats (r-NSCs) ubiquitously expressing the GFP reporter gene and transplanted into rats with a contusion of thoracic SC. We used rats as recipients since contusion is more controllable and reproducible in larger species, and we administered cyclosporine A as immunosuppressant in place of genetic immunodeficiency. Anatomical analysis 2 months after engraftment confirmed that r-NSCs successfully survived in host SC, filled the contusion cavity and elongated bilaterally many GFP-positive axons, most of which were myelinated by host oligodendrocytes ( Supplementary  Fig. 8a-d). No graft rifts were observed. Functional analysis by Basso Beattie and Bresnahan (BBB) test 34 revealed improved motor performance in the transplanted group as compared to control rats, which reached significance at 7 and 8 weeks postgraft ( Supplementary Fig. 8e). These results confirmed that SCderived NSCs yield benefit in a clinically relevant SCI model.

Discussion
In this study, we report the derivation and characterization of human SC-NES cells as candidates for cell-based therapy in SCI treatment. Previous investigations on SC-derived NSCs were limited to the application of one single fetal cell line (566RSC) 1,2,7 or to fetal NSCs cultured as neurospheres with the consequent reduction of the neurogenic potential 4,5 . Here we show that SC-NES cells can be successfully derived from human SC primordia and can be cultured long-term in adhesion without genetic immortalization. Further, we demonstrate that human SC-NES cells are highly expandable and retain over time intact karyotype and neurogenic capacity. In vitro experiments confirm that SC-NES cells are immunopositive for canonical NSC markers and can be efficiently differentiated toward the neuronal and glial fate. Importantly, SC-NES cells retain their regional commitment after propagation as demonstrated by the expression of specific transcription factors and the neurochemical identity of neural progeny 35,36 .
When transplanted into injured SC, SC-NES cells exhibit excellent integration properties: they differentiate into neurons and extend long-distance axons reaching the rostral and caudal extremities of the recipient SC. Previous work has emphasized the importance of mechanical and trophic support to guarantee cell survival and neurite extension from grafted cells 1,11 . Human SC-NES cells, on the other hand, are intrinsically prone to elongate long distance axonal connections with no need of growth factorenriched matrices. We also observed that SC-NES cells produce an amelioration of locomotor deficits and that the extent of the functional improvement depends on lesion severity. The infiltration of collagen and the subsequent formation of an impervious fissure within some grafts may have attenuated the behavioral outcome. However, the formation of a rift could be due to the hemisection surgical procedure which inevitably compromises meningeal integrity. In fact, no rifts were observed in the SC contusion experiment where dura mater was left intact. However, since contusion is the most common type of clinical SCI lesion, the formation of the rift may not be a limiting factor for potential clinical applications.
The relevance of graft rostro-caudal origin has not been systematically explored with human cell sources, though brain precursor transplantation has been described 37,38 . For rodent neural precursor transplantation, there has been a direct comparison of cells established from SC versus hindbrain versus telencephalon 6 and substantially greater host CST growth was observed with SCmatched transplants. However, the relative degree of engraftment for different source cells, and the converse engraftment of different source cells into NCX have not been reported previously. Here, in order to assess whether the engraftment potential of SC-NES cells is correlated to their anatomical derivation, we challenged NCX-NES cells with an identical experimental paradigm. Interestingly, we found that NCX-NES cells aggregate, forming dense clusters in the SC with a remarkably poor graft neurite extension. On the other hand, when placed into NCX, they distribute widely and elongate branched projections suggestive of efficient integration. Thus, matching NES cell sources to engraftment site is essential not only for host CST axon growth into the graft, but also for human neuron engraftment per se, and most critically for graft axon extension.
To investigate the molecular mechanism responsible for the striking difference in the engraftment properties of SC-and NCX-NES cells, we performed global transcriptional investigation of implanted cells. Grafted SC-NES cells upregulate genes related to neuronal maturation (MTURN, ATCAY), cell adhesion (KAL1), axonal elongation (PTN), and synapse establishment (SYT13, SYT4). These six genes could be considered as positive markers of engraftment in order to delineate the molecular signature of cell populations suitable to achieve a successful integration. Moreover, release of secreted PTN into cerebrospinal fluid may provide an engraftment biomarker monitored by lumbar puncture during future transplantation studies. Two negative markers of engraftment have also been proposed by Ladewig and colleagues 39 who found that the secretion of FGF2 and VEGF by implanted neural progenitors impairs the integration in the tissue through chemoattraction. Additional studies may determine whether these markers could be used to screen cell populations before their implantation, thereby optimizing the experimental procedure and facilitating clinical transition. Such biomarkers may be especially relevant in the development of novel differentiation protocols from patientderived pluripotent stem cells in order to enhance their integration potential and maximize their potential benefit.
The beneficial effect of NSCs in rewiring the lesioned SC occurs through a range of mechanisms including growth factor secretion [40][41][42] , severed axon myelination 43 , and enhancement of severed axon regeneration 1,6,11 . Grafted cells can provide a scaffold for severed axons to regenerate, and descending fibers from the CST, rubrospinal or raphespinal tracts have been observed to enter the graft 1,6,11 . However, the extent of axonal regeneration beyond the lesion is much reduced if not completely absent, and it is unlikely to support a functional benefit. It has also been proposed that NSCs can form relay circuits in the injured SC which are comparable to those formed by propriospinal circuits after incomplete SCI [44][45][46][47] . One key aspect of a relay system is that injured axons do not need to regenerate long distances and the onus for axon growth is placed on the transplant-derived neurons, which can be selected for their ability to extend long axons in vivo. Nevertheless, no proof has been provided previously to demonstrate a graft-mediated neural relay function. Using a pharmaco-genetic approach to obtain a transient silencing of implanted cells, we provide the first evidence that human SC-NES cells establish a functional circuit in recipient SC. Interruption of the circuit by the temporary silencing of SC-NES cell-derived neurons resulted in loss of the functional benefit in three different behavioral paradigms. Thus, SC-NES cells restore the disrupted connectivity by forming an intermediate station between the rostral and caudal segments of damaged SC.
The transition of human cell therapy to the clinic will likely require more systematic investigations. While recent studies show that human cell transplantation is possible in non-human primates 48 , practical issues remain to be addressed. For instance, further studies are needed to understand guidance of extending axons to appropriate targets and to determine what effect newly generated circuits may have for autonomic and sensory function.
Our investigation of human NES cells provides tools to understand the molecular features necessary to achieve successful engraftment in terms of both efficient cell integration and host axonal growth. Having the means in hand to delineate the transcriptional profile of human cells with a high engraftment potential provides an important instrument for future development in the field.

Methods
Human tissue procurement. All work was performed according to the NIH guidelines for the acquisition and distribution of human tissue for bio-medical research purposes and with the approval by the Human Investigation Committees and Institutional Ethics Committees of each institute from which samples were obtained. De-identified postmortem human brain and SC specimens were provided by the Joint MRC/Wellcome Trust (grant #099175/Z/12/Z) Human Developmental Biology Resource (http://hdbr.org). Appropriate informed consent was obtained and all available non-identifying information was recorded for each specimen. Tissue was handled in accordance with ethical guidelines and regulations for the research use of human brain tissue set forth by the NIH (http://bioethics.od.nih. gov/humantissue.html) and the WMA Declaration of Helsinki (http://www.wma. net/en/30pubblications/10policies/b3/index.html).
Maintenance and differentiation of NES cell lines. NES cells were kept in proliferation in T25 flasks coated with poly-L-ornithine (0.01%), laminin (5 µg/ml) and fibronectin (1 µg/ml). Cells were expanded in NES medium and reached confluency (0.5-1 × 10 5 cells/cm 2 ) in about 7 days. Cells were split 1:2 every 5-7 days with 0.25% trypsin. The expansion rate of the cells was constant over time. Half volume of the medium was changed every 2-3 days. For cryopreservation, cells were trypsinized and~2 × 10 6 cells were pelleted (200 × g for 3 min) and resuspended in cryopreservation medium containing 10% DMSO and 90% NES medium. For resuscitation, cells were rapidly thawed at 37°C, resuspended in pre-warmed culture media, centrifuged for 3 min at 200 × g, resuspended in NES medium, and replated on coated plates. Neuronal differentiation of NES cells was performed in two steps. For the pre-differentiation step NES cells were seeded at a density of 0.5 × 10 5 cells/cm 2 in a T25 coated flask in NES medium without EGF and FGF2. After 7 days, cells were dissociated and replated at a density of 0.8-1 × 10 5 cells/cm 2 in BrainPhys Medium (Stem Cell Technologies, Cambridge, MA, #05790) supplemented with B27 (2%), N2 (1%) and BDNF (20 ng/ml). Half volume of the medium was changed every 2-3 days and neurons were differentiated up to 3 months.
Animals. Mice: 8-10 weeks old female NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ mice (Jackson Laboratories) were used in the present study as NES cell recipients. Rats: SD-Tg(CAG-EGFP)CZ-004 Osb rats were a generous gift from Dr. Jeffery Kocsis (Yale University, Department of Neurology) and were used to derive GFP-positive NSCs. At 10-11 weeks old female wild type Sprague Dawley rats were used as recipients for rat NSCs grafts. All animals were housed on a 12 h light/12 h dark cycle and had free access to food and water throughout the study. All experimental procedures were performed in compliance with animal protocols approved by the Institutional Animal Care and Use Committee at Yale University.
Karyotype analysis. Karyotype analysis was performed by core facility at the Department of Genetics at the Yale School of Medicine. Briefly, the cells were cultured in the presence of 0.1 µg/ml colcemid for up to 4 h and then fixed. Metaphase spreads were analyzed.
Electrophysiology. Electrophysiological experiments were performed on differentiated SC-NES cells infected with the hM4Di-mcherry lentiviral construct. Recordings were performed at 21-23°C using the whole-cell patch-clamp technique in voltage-and current-clamp configurations. Cells were continuously perfused with an extracellular solution composed by 125 mM NaCl, 2.5 mM KCl, 26 mM NaHCO 3 , 15 mM glucose, 1.3 mM MgCl 2 , 2.3 mM CaCl 2 , 1.25 mM NaH 2 PO 4 (bubbled with 95% O 2 , 5% CO 2 ) and were visualized using an Axioskop FS2 microscope (Zeiss) equipped with epifluorescence and infrared differential interface contrast (DIC) optic. An X-Cite 120LED lamp (Excelitas) and an appropriate filter set were used to identify mcherry-positive cells. Pipettes were produced from borosilicate glass capillary tubes (Sutter Instruments) by mean of a horizontal puller (P-2000, Sutter instruments) and filled with the following intracellular solution: 130 mM K-gluconate, 4 mM NaCl, 2 mM MgCl2, 1 mM EGTA, 5 mM creatine phosphate, 2 mM Na2ATP, 0.3 mM Na3GTP, 10 mM HEPES (pH 7.3 with KOH). Membrane voltage was corrected off-line for a calculated liquid junction potential of −10 mV. Series resistance was always compensated by 70-80% and monitored throughout the experiment. Clozapine-N-oxide [ Western blot. NES or HEK 293 cells were washed with PBS and harvested in radioimmunoprecipitation assay (RIPA) buffer. WT or NgR1 KO mouse forebrain were dissected and homogenized in RIPA buffer. The samples were then centrifuged at 20,000 × g for 20 min. The lysate was resolved by SDS-PAGE, transferred to nitrocellulose membranes, immunoblotted with anti-β-actin (1:3000 Sigma-Aldrich, #A1978), and anti-NgR1 (1:1000 R&D Systems, AF1440) primary antibodies o/n at 4°C. After primary antibody incubation, secondary antibodies (Odyssey IRDye 680 or 800) were applied for 1 h at room temperature. Membranes were then washed and visualized using a Licor Odyssey Infrared imaging system.
Thoracic dorsal hemisection surgery. Mice were first anesthetized with 4% isofluorane and maintained with 2% isofluorane throughout the whole procedure. A laminectomy was performed to expose the dorsal portion of the SC corresponding to the T8 and T9 levels. The dorsal hemisection was performed at the T8 level with a pair of microscissors to a depth of 1.1 mm to completely sever the dorsal and dorsolateral CST. The lateral portions of the SC were scraped with a 30 gauge needle to ensure the completeness of the lesion. Muscle and skin overlying the lesion were sutured with 4.0 vicryl. All animals received subcutaneous injection of 100 mg kg −1 ampicillin and 0.1 mg kg −1 buprenorphine twice a day for the first two days after surgery. Bladders of injured animals were expressed manually on a daily basis throughout the whole experiment.
SC contusion surgery. Female Sprague-Dawley rats (10-11 weeks, 220-240 g) were used in this study. Animals were anesthetized with intraperitoneal injection of ketamine (60 mg kg −1 ) and xylazine (10 mg kg −1 ) mixture. A laminectomy was conducted at the caudal portion of T6 and all of T7 spinal levels. A T7 severe contusion injury (weight of 10 g, height of 50 mm) was produced with the MASCIS impactor. After the spinal contusion, muscle and skin layers were sutured with 4.0 polyglactin. Bladders of injured animals were expressed manually twice a day throughout the whole experiment.
Cell engraftement. Ten days after mouse dorsal hemisection surgery, lesioned animals underwent a second procedure for NES cell implantation. Mice were anesthetized with 4% isofluorane and maintained anesthetized with 2% isofluorane until the procedure was completed. The original incision was reopened and the SC re-exposed. NES cells were resuspended in BrainPhys medium supplemented with 10 µM Rock Inhibitor and 20 ng/ml BDNF at a density of 250,000 cells/µl. NES cells were kept on ice throughout the procedure. Two injections were performed on the injury site, each delivering 1 µl of cell suspension at a distance of~1 mm one from the other. The injection was performed using a glass capillary attached to a syringe (Hamilton, Reno, NV) and a micropump (Ultramicropump III, World Precision Instruments, Sarasota, FL). The rate on the injection was 250 nl/min and the capillary was left in place for two additional minutes before withdrawal. Control animals underwent the same surgery and received a vehicle injection. At the end of the procedure, the muscle and the skin were sutured with vicryl 4.0 and animals received ampicillin 100 mg kg −1 subcutaneous injection for 3 days. Animals survived 8 weeks before being sacrificed for the histological analysis.
Nine days after rat SC contusion, animals underwent surgery for NSCs engraftment. Embryonic day 13.5 (E13.5) SC from transgenic SD-EGFP rats provided donor tissue for grafting. Rat NSCs were resuspended in a fibrin matrix enriched with growth factors, as previously described 1 . Starting 24 hours before cell transplantation and throughout the whole experiment rats were treated with cyclosporine A (15 mg kg −1 ) as immunosuppressant drug administered i.p. on a daily basis. The graft mixture (250,000 cells/µl) was microinjected into three sites on the lesion delivering 3 µl per injection. Animal underwent functional testing for up to 8 weeks and were sacrificed for the anatomical analysis through transcardial perfusion with 4% formaldehyde.
For NCX transplantation, mice were anesthetized with 4% isofluorane as previously described. A craniotomy was performed using a micromotor system (Foredom, Bethel, CT) exposing the motor cortex. The cortex was damaged by aspiration of a portion 1 × 1 mm 2 . A volume of 1 µl of NES cell suspension (250,000 cells/µl) was injected onto the damaged area using a glass capillary attached to syringe (Hamilton, Reno, NV) and a micropump (World Precision Instruments, Sarasota, FL). Animals were kept for 8 weeks before being sacrificed for histological analysis.
Anterograde labeling of the CST. Six weeks after injury, descending CST fibers of grafted mice were labeled with BDA (0.1 g/ml in sterile normal saline, Thermo-Fisher Scientific # D1956) by injection into five spots of the right motor cortex. The skulls of anesthetized mice were tightly fixed to a stereotaxic apparatus (Kopf, Tujunga, CA). Craniotomy over the motor cortex area was carried out using a micromotor system (Foredom, Bethel, CT). The injection area on the right hemisphere was defined in a rectangle with one side measuring 2 mm (from 1.0 mm anterior to −1.0 mm posterior to the bregma) and one side measuring 1.5 mm (lateral to the bregma). Injections were performed using a glass capillary attached to a microsyringe (Hamilton, Reno, NV) at a 0.7 mm depth. Each injection delivered 75 nl of BDA solution into the motor cortex at a rate of 75 nl/min. The tip of the glass capillary was left in place for two additional minutes before withdrawal. Two weeks later, the animals were sacrificed by transcardial perfusion with PBS followed by 4% formaldehyde. To visualize the BDA, a tyramide signal amplification fluorescence system (Perkin Elmer, Waltham, MA) was used.
Immunohistochemistry. After perfusion tissues were removed, post-fixed overnight in 4% formaldehyde at 4°C, and cryoprotected in 30% sucrose (Sigma-Aldrich, St. Louis, MO) until they sank before being sectioned on a Leica cryostat. SC sections were cut horizontally at a thickness of 30 µm. Brain sections were cut sagitally at a thickness of 40 μm. Sections were washed three times in PBS between each step in the immunodetection protocol. For detection of the antigens with fluorescent antibodies, the protocol was performed as follows: sections were left in blocking solution (PBS supplemented with 1% horse serum and 0.1% Triton) for 1 h at RT and then incubated overnight at 4°C with primary antibodies diluted in blocking solution. Primary antibodies were diluted as follows: Millipore #MAB5504); 5HT (1;10000 Immunostar #20080); Alexa 568-conjugated streptavidin (1:500 Thermo Fisher Scientific #S11226) to label BDA traced CST axons. The following day, sections were incubated for 1 h at RT with appropriate Alexa 488, Alexa 568 or Alexa 647 (1:500 Invitrogen) secondary antibodies. The sections were finally covered with mounting medium (Life Technologies #P36935) containing DAPI before being coverslipped. All images were acquired using a Carl Zeiss LSM 710 confocal microscope. Image processing was performed using Zeiss Zen, Adobe Photoshop and Illustrator.
Quantification of cell types in the graft and axons. Cellular differentiation in the grafts was determined by counting individual cells labeled for RBFOX3, GFAP, KI67, CHAT or caspase 3 in 8 randomly selected fields (4 sections per animal, 2 fields per section) and expressing them as a percentage of the total number of human nuclei positive cells in the field. Cells were visualized using a Carl Zeiss LSM 710 confocal microscope at a magnification of ×63. The counting was performed using the built-in plugin of the Image J program. The number of GFP labeled human axons emerging from the graft was quantified using a Carl Zeiss LSM 710 confocal microscope. For every 6th consecutive horizontal section, a mediolateral line was drawn 250 and 500 µm caudal to the graft/host interface under a ×40 magnification. The tissue was then examined under a ×600 magnification and GFP labeled axons that intersected this line were marked and counted. To estimate the total number of axons/subjects, the number of axons counted on sections was multiplied by 6. To visualize axons in the gray matter sections were co-stained with RBFOX3 antibody.
Behavioral analysis. A total of 64 mice underwent SC mid-thoracic dorsal hemisection studies in two batches. Forty-nine (49) mice underwent hemisection injury and were tested for their locomotor ability using the Basso Mouse Scale (BMS) 26 7 days after lesion. Seven out of 49 mice were eliminated from further study because of unsuccessful hemisection surgery, defined by BMS score higher than 2 on day 7. The remaining 42 mice were randomized to receive vehicle injection (14 mice), SC-NES cell injection (19 mice) or NCX-NES cell injection (9 mice) at 10 days after hemisection. These 42 animals were followed by BMS behavioral test performed by two investigators blinded to the treatment for a total of 8 weeks after cell implantation, then received BDA injection and were processed for histological analysis. A post-hoc subgroup analysis of behavior in 25 (9 controls, 10 SC-NES and 6 NCX-NES cell recipients) out of 42 animals were used for spared tissue assessment and correlation studies.
In a separate experiment, 15 mice underwent hemisection injury and received SC injection of either hM4Di-expressing SC-NES cells (8 mice) or vehicle injections (7 mice) 10 days after hemisection. Their behavior was assessed by BMS, grid walking and footprint test with and without CNO administration 8 weeks after transplantation. Grid walking and footprint analysis were performed for animals able to walk, i.e. scoring at least 4 at the BMS test. For the footprint analysis animals had water-based nontoxic paint placed on the plantar surfaces of all four limbs (red on forelimbs and blue on hindlimbs) and were then run three times along a 1 m long narrow corridor lined with absorbent paper. Stereotyped gait and motor coordination parameters, including hindlimb stride length and hindlimb base of support, were measured from three complete step cycles from the middle of the runway. Animals were trained for one week in order to collect the baseline scores. To validate the behavioral outcome upon silencing NES cells, animals were given an i.p. injection of CNO, (1 mg kg −150 ) and were tested 30 min later. The wash-out scores were collected 24 h later. For the grid-walking test mice were placed on an elevated 45 × 45 cm metal grid with 2.5 × 2.5 cm square spacing covered by a dark box to make the environment more comfortable for the animals. Mice were videotaped via reflection from a mirror placed under the grid and allowed to explore the grid for 3 min. Videos were scored for the percentage of impaired steps out of the first 50 steps for each hindlimb individually. Impaired steps included a foot slip where the limb fell between the rungs or an incorrectly placed step where either the ankle or the tips of the digits were placed on the rung instead of proper grasping of the rung. Similar to the footprint test, baselines scores were collected after one week of training. The CNO effect was recorded 30 min after i.p. administration, whereas the drug wash-out assessment was performed on the following day.
Rats were evaluated using the BBB locomotor scale 34 two days before cell implantation and every week until killing.
Transcriptome analysis. For the transcriptome analysis each sample was handled in triplicate. Total RNA was extracted using the RNeasy plus minikit (Quiagen, #74134) according to manufacturer's instruction. mRNA libraries were prepared according to the TruSeq RNA Illumina kit protocol and sequenced using a HighSeq 2500 sequencing system (Illumina). The sequenced reads were clipped one nucleotide in both ends, leaving 74 nucleotide long reads for sequence alignment to human reference genome (GRCh38/hg38) by using the STAR software 51 . Beyond default parameters in the alignment procedure, human gene annotation retrieved from the GENCODE project (version 21, http://www.gencodegenes.org/releases/21. html) was additionally provided to improve exon-exon junction mapping. To quantify accurately the origin of the genomic locus of short sequenced reads, only the uniquely mapped reads were used for downstream analyses. FeatureCounts was used to quantify expression profiles of each type of annotation entry retrieved from GENCODE v21 52 . Additional quality control measures were introduced to assess potential issues including reads classification, mitochondrial contamination, and gene coverage uniformity. R package DESeq2 was used to identify DEX genes 49 . We generated two types of gene expression value, read counts and RPKM (reads per kilobase of exon model per million mapped reads). The reads count per gene served as the input for DESeq2. When performing the comparisons, DESeq2 first gets the mean expression level as a joint estimate for both groups, and then calculates the difference as well as the P value for the statistical significance of this change. The adjusted P value was calculated based on multiple testing with the Benjamini-Hochberg procedure, estimating the false discovery rate (FDR). To detect statistically significant DEX genes, FDR was set to be minor than 0.01. The gene ontology (GO) analysis was performed using The Database for Annotation, Visualization and Integrated Discovery (DAVID) version 6.8.
Weighted gene correlation network analysis (WGCNA). The R WGCNA package was used to find modules of highly correlated genes and characterize their expression patterns 53 . Genes expressed in at least 25% of samples were included for the analysis. Prior to module detection, scale-free topology was calculated to determine soft power, which was set at 6 and used to generate dissimilarity matrix. Then, module clustering was process based on dissimilarity matrix with minimum module size setting at 100. Detected modules were further merged together if they have very low dissimilarity (threshold = 0.2) and 23 modules were finally determined. To characterize the module expression pattern, we calculated the eigengene of each module via moduleEigengenes function. Eigengene expression pattern was shown using box plot.
Statistical analysis. For comparison between two groups two-tailed Student's ttest was used at a designated significance level of P < 0.05. Measurements taken at different time points were compared using the repeated measures one-way ANOVA. Analyses were conducted using Excel and SPSS. For the electrophysiology experiments, statistical significance was determined by one-sample Student's t-test. All data are presented as mean ± s.e.m. The statistical details of the experiments can also be found in the figure legends. No statistical methods were used to calculate sample size estimates. No animal exclusions were made except for the subgroup specified in Fig. 5d. Animals were allocated to injection groups in alternating order by an observer unaware of treatment. All behavioral and histological analyses were conducted without knowledge of treatment group.
Data availability. All relevant data that support our experimental findings are available from the authors. Supporting RNA-sequencing data are available on NCBI GEO (Gene Expression Omnibus), accession number GSE107514.