Enhanced Production of Mesencephalic Dopaminergic Neurons from Lineage-Restricted Human Undifferentiated Stem Cells

The differentiation of human pluripotent stem cells (hPSCs) into mesencephalic dopaminergic (mesDA) neurons requires a precise combination of extrinsic factors that recapitulates the in vivo environment and timing. Current methods are capable of generating authentic mesDA neurons after long-term culture in vitro; however, when mesDA progenitors are transplanted in vivo, the resulting mesDA neurons are only minor components of the graft. This low yield hampers the broad use of these cells in the clinic. In this study, we genetically modified pluripotent stem cells to generate a novel type of stem cells called lineage-restricted undifferentiated stem cells (LR-USCs), which robustly generate mesDA neurons. LR-USCs are prevented from differentiating into a broad range of nondopaminergic cell types by knocking out genes that are critical for the specification of cells of alternate lineages. Specifically, we target transcription factors involved in the production of spinal cord and posterior hindbrain cell types. When LR-USCs are differentiated under caudalizing condition, which normally give rise to hindbrain cell types, a large proportion adopt a midbrain identity and develop into authentic mesDA neurons. We show that the mesDA neurons are electrophysiologically active, and due to their higher purity, are capable of restoring motor behavior eight weeks after transplantation into 6-hydroxydopamine (6-OHDA)-lesioned rats. This novel strategy improves the reliability and scalability of mesDA neuron generation for clinical use.


Introduction 49 50
Mesencephalic dopaminergic (mesDA) neurons develop from the ventral midbrain of 51 the neural tube. The morphogen sonic hedgehog (SHH) and members of the WNT family are 52 instrumental in their specification due to their essential role in establishing the dorsal-ventral 53 and anterior-posterior (A-P) axes of the embryo, respectively 1,2 . For the production of 54 midbrain dopaminergic (DA) neurons, high SHH signaling from the notochord is required to 55 specify ventral neural epithelial cells in the neural plate to a floor plate identity 3 , and graded 56 Wnt signaling emanating from the posterior regions of the embryo is required to specify 57 anterior neuroectoderm to a midbrain identity 4 . At later stages in development, neural 58 progenitors in the midbrain region receive Wnt1 and Fgf8 signals from the isthmic organizer, 59 which refines the patterning of the cells to a caudal location within the mesencephalon 5,6 . 60 Recapitulating these developmental steps in vitro with human pluripotent stem cells (hPSCs) 61 is the goal of stem cell transplantation therapies for Parkinson's disease 7 . 62 In stem cell differentiation protocols, early application of high concentrations of SHH 63 is necessary to specify neural progenitor cells to a floor plate identity 8,9 . However, along the 64 A-P axis (also referred to as rostral-caudal axis), specification to a caudal midbrain identity is 65 more complex. A titrated WNT concentration within a precise range can specify anterior 66 neuroectoderm progenitors to a caudal midbrain identity 10,11 . High concentrations result in 67 the production of hindbrain cell types, and lower concentrations result in an anterior midbrain 68 or diencephalic identity. Timed delivery of FGF8 or sequential exposure to high levels of WNT 69 has also been shown to improve specification to a caudal midbrain identity 12,13 . Despite these 70 advances in stem cell differentiation protocols, the overall yield of mesDA neurons after 71 transplantation is extremely low, and increasing the yield would significantly improve cell 72 purity and the reliability of graft function, which is essential for the use of these cells in the 73 clinic 14,15 . In this study, we used a gene knockout approach to restrict cell fate and prevent the 100 differentiation of non-DA cell lineages with the aim of enhancing differentiation to mesDA 101 neurons. Specifically, we focused on the early developmental stages when major lineage 102 choices are made and identified the transcription factor determinates that are critical for 103 those lineages but not required for a mesDA fate. By inducing loss-of-function mutations in 104 lineage determinant genes expressed in non-DA lineages, we were able to bias the 105 differentiation of hPSCs toward a mesDA identity. We generated stem cells that could be 106 expanded in the undifferentiated pluripotent state and were restricted in their potential when 107 differentiated. We named these lineage-restricted undifferentiated stem cells (LR-USCs). Cdx1/2/4 in mice causes posterior truncation, we chose to disrupt the CDX family of genes in 153 addition to knocking out GBX2. We generated homozygous null mutations in all three CDX 154 family members, CDX1/2/4, by targeting their DNA binding domains ( Supplementary Fig. 2). 155 The resulting GBX2 -/-CDX1/2/4 -/-hESC line (hereafter referred to as 4X cells) was differentiated 156 for four DIV using the CNP protocol (Fig. 1a). As expected, we did not detect CDX2 transcripts 7 or CDX2 expression in 4X CNPs (Fig. 1b, d). Strikingly, at 4 DIV, the 4X neural progenitors 158 showed a significant increase in OTX2 transcript levels compared to H9 and GBX2 -/derived 159 CNPs (LogFC to hESC = H9: 0.0032; GBX2 -/-: 0.083; 4X: 0.301; P < 0.0001 for both 4X vs. H9 or 160 GBX2 -/-), and we could readily identify OTX2-positive cells (Fig. 1b, c). 161 To elucidate the effects on gene expression in more detail, we performed RNA 162 sequencing on CNPs of all three cell lines (H9, GBX2 -/and 4X). We assessed HOX gene 163 expression profiles and found that the 4X CNPs showed a significant reduction in the 164 expression of posterior HOX genes, compared to H9, beginning with HOXA3 and moving 165 caudally (HOXA3, P = 1x10 -6 ; HOXA5, P = 1.95x10 -5 ; HOXA7, P = 0.0005; HOXA9, P = 0.0004; 166 HOXA10 P = 0.002; Fig. 1e). These results indicate that 4X cells were unable to generate 167 progenitor cell types caudal to r4 31 . We next questioned whether there were changes in the 168 expression of anterior genes. First, we examined the expression of forebrain genes in 4X CNPs 169 and observed no change in the expression of SIX3, DLX2 and FOXG1 (Fig. 1e). However, the 170 transcript levels of the forebrain/midbrain gene OTX2 were significantly increased in 4X CNPs 171 compared to H9 and GBX2 -/-CNPs (H9, LogFC = 6.83, P = 0.001; GBX2 -/-, LogFC = 3.92, P = 0.003, 172 respectively). The expression of the midbrain genes PAX2, and EN1 was also significantly 173 increased in 4X CNPs compared to H9 (PAX2, LogFC = 2.63, P = 0.03; EN1, LogFC = 4.76, P = 174 0.03) and GBX2 -/-CNPs (PAX2, LogFC = 3.49, P = 0.00005; EN1, LogFC = 7.08, P = 0.0003; Fig.  175 1e). Interestingly, GBX2 -/cells showed a reduction in the expression of anterior hindbrain 176 genes, such as EGR2 (LogFC = -3.86; also known as KROX20) and MAFB (LogFC = -0.67), which 177 was in line with reports showing that disruption of Gbx2 in mice causes loss of r1-3 28 . In 178 contrast, the expression of MAFB significantly increased in 4X cells (MAFB, LogFC = 2.49, P = 179 0.0002), which is in accordance with the loss of posterior HOX expression. 180 Analysis of differentially expressed genes among the three groups showed that the top 181 significantly downregulated genes in 4X cells compared to H9 and GBX2 -/cells included HOX 182 genes (Fig. 1f). Furthermore, the expression of CYP26A1, which is involved in retinoic acid (RA) 183 metabolism and is induced by CDX2, was significantly downregulated in 4X cells compared to 184 GBX2 -/cells (LogFC = -13.43, P = 1.30x10 -12 ). A comparison of GBX2 -/to H9 cells showed that 185 knockout of GBX2 alone resulted in a significant decrease in the transcription levels of the 186 Groucho corepressor proteins TLE1 (LogFC = -2.86, P = 8.05x10 -9 ) and TLE4 (LogFC = -1.54, P = 187 8.32x10 -9 ), which function with GBX2 to repress OTX2 32 . Overall, knockout of GBX2 resulted 188 in disruption of anterior hindbrain patterning, and 4X cells showed that further loss of CDX 189 positive midbrain cells at 3 µM GSK3i is maintained at lower GSK3i concentrations. Using the 226 same 11 DIV caudal protocol, we tested four concentrations of GSK3i from 0.7 µM to 3 µM 227 ( Fig. 2a-b). We examined the percentage of cells expressing OTX2 by flow cytometry. 228 Strikingly, at the lowest concentrations of GSK3i, i.e., 0.7 µM and 1 µM, 71.0% and 34.9%, 229 respectively, of 4X cells were OTX2-positive; there were significantly more OTX2-positive 4X 230 cells than OTX2-positive H9 and GBX2 -/cells at these concentrations (H9: 2.5%; GBX2 -/-: 231 21.8%; 4X: 71.0% P < 0.0001 for both, for 0.7 µM. H9: 4.8%; GBX2 -/-: 5.0%; 4X: 34.9% P < 0.0001 232 for both, for 1 µM). At 2 -3 µM GSK3i, the percentage of OTX2-positive cells decreased 233 dramatically in all groups; however, there were still more OTX2-positive 4X cells than OTX2-234 positive H9 and GBX2 -/cells at 3 µM GSK3i (H9: 0.3%; GBX2 -/-: 0.8%; 4X: 10.1% P < 0.0004 and 235 P < 0.0008, respectively). 236 Our main objective was to generate LR-USCs that more efficiently generate mesDA 237 neurons, even under suboptimal conditions. Thus, we compared H9 and 4X cells and 238 differentiated them using a recently reported mesDA protocol (Fig. 2c) 14 . This protocol is 239 known to require adjustments to the concentration of GSK3i between cell lines; therefore, we 240 started by titrating GSK3i from a concentration of 0.5 µM to 1 µM to identify the optimal 241 concentration for generating posterior midbrain cells from H9 cells. First, we determined that 242 the highest percentage of OTX2-positive H9 cells was obtained with a concentration of GSK3i 243 between 0.5 and 0.7 µM and that there was a significant decrease in the number of these cells 244 when the GSK3i concentration reached 1 µM (Fig. 2d). Second, we examined the expression 245 levels of midbrain genes (Fig. 2e). The transcript level of OTX2 in H9 cells at 16 DIV reached a 246 maximum at a GSK3i concentration of between 0.5 and 0.6 µM, and EN1 expression was the 247 highest at a GSK3i concentration of 0.6 µM. Furthermore, the transcript level of the hindbrain 248 gene HOXA2 reached the lowest point at a GSK3i concentration between 0.5 and 0.6 µM. 249 These results were consistent with previous reports, which indicated that a GSK3i 250 concentrations below 1 µM is necessary for midbrain specification and that concentrations 251 approaching 1 µM or higher result in a dramatic shift to hindbrain identity 11 . Interestingly, 252 HOXA2 was expressed under optimal conditions, demonstrating that the protocol resulted in 253 the production of a wide variety of cell types along the A-P axis, including hindbrain cell types, 254 as reported by others 11 . Overall, we determined that the optimal GSK3i concentration for H9 255 cells at 16 DIV was 0.6 µM (Fig. 2d, e and Supplementary Fig. 3). 256 We next compared H9 cells and 4X cells across multiple GSK3i concentrations and 257 observed a significant improvement in the ability of 4X cells to produce midbrain cells.   Using single-cell sequencing, we determined the cell types that were produced by H9 280 and 4X cells following the mesDA neuron differentiation protocol using 1 µM of GSK3i. 281 Dimension reduction was performed by uniform manifold approximation and projection 282 (UMAP), and there was a noticeable separation between 4X and H9 cells across clusters (Chi-283 square, P < 0.0001) and a difference in the cell types produced by the two cell lines at 16 DIV 284 (Fig. 3a, d). This separation coincided with a marked shift in distribution along the A-P axis. 285 Based on the expression of OTX2 and EN1, we divided the A-P axis into four domains (rostral, 286 OTX2-positive/EN1-negative; caudal midbrain, OTX2-positive/EN1-positive; rhombomere 1, 287 OTX2-negative/EN1-positive; and posterior, OTX2-negative/EN1-negative) (Fig. 3c). 4X cells 288 produced all four populations, with the smallest being the most rostral population (Fig. 3c). 289 The caudal midbrain is the region where DA neurons of the substantia nigra develop, and patterning gene FOXA2 showed that it was highly expressed in both H9 and 4X cells, indicating 296 that both cell lines were efficiently ventralized to a floor plate identity (Fig. 3d). 297 Upon further examination of caudal midbrain cells by graph-based clustering, we 298 identified three clusters (clusters 1, 9, and 8) enriched in caudal midbrain floor plate 299 progenitors expressing FOXA2, OTX2, LMX1A and EN1 (Fig. 3a-d). Two additional midbrain 300 floor plate clusters (clusters 6 and 10) were identified; these populations expressed FOXA2, 301 OTX2 and EN1 but lacked LMX1A, indicating that they were a lateral floor plate population 302 ( Supplementary Fig. 4a). Clusters 8 and 6 were in a proliferative state, as revealed by the 303 expression of MKI67 and TOP2A (Fig. 3b). Cluster 1 was the largest midbrain population and 304 exhibited the highest expression of midbrain markers, with 4X cells making up 98.6% of cells 305 in this cluster. 306 By analyzing the hindbrain cells in more detail, we identified five clusters that we 307 classified as hindbrain floor plate progenitors (clusters 0, 4, 2, 7, and 5), which expressed 308 FOXA2, SHH and CORIN (Fig. 3b). The hindbrain floor plate clusters comprised both H9 and 4X 309 cells (Fig. 3c). Further examination of HOX gene expression within these clusters showed that 310 there was an abundance of cells expressing anterior HOXA/B genes (Fig. 3d). A total of 92.2% 311 of the HOXA/B cells originated from H9 cells, confirming that H9 cells were of a more caudal 312 identity than 4X cells ( Fig. 3d and Supplementary Fig. 4b-c). We also identified a small 313 population of early neural crest progenitors expressing SOX10 and FOXD3 (cluster 13), which 314 were exclusively H9 cells (Fig. 3a, b). 315 By 16 DIV, three neuronal clusters (3, 11, and 12), which were predominately derived 316 from H9 cells, were present (H9: 52%, 95%, 100%; 4X: 48%, 5%, 0%, respectively). Clusters 11 317 and 3 expressed high levels of ONECUT2, PHOX2B and ISL1, which are markers of early-born 318 basal-plate hindbrain motor neurons 35-37 . The remaining cluster, cluster 12, contained only 319 28 cells that expressed high levels of GATA2, GATA3 and MEIS2 but did not express GAD1 or 320 GAD2, indicating that they were immature V2b GABAergic neuroblasts 36 . A population of cells 321 within the neuronal clusters expressed tyrosine hydroxylase (TH); however, after 322 subclustering, we found that these cells did not express NR4A2 (also known as NURR1), LMX1A 323 or EN1, indicating that they were not of a mesDA identity (Fig. 3b, d and Supplementary Fig.  324 4d, e).  In contrast to 4X cells, H9 cells formed one main connected set of clusters (clusters 1, 339 2, 3, and 7) and two small isolated clusters (clusters 9 and 10; Fig 3e). All six clusters were 340

TH-positive neurons (Supplementary
To further support our single-cell sequencing results, histological analysis was 352 performed. We used the same growth factor paradigm but adapted a differentiation protocol 353 to generate organoids to provide an optimal environment for the survival of neurons (Fig. 4a). 354 At 83 DIV, we observed a large population of FOXA2/TH double-positive DA neurons within 355 organoids produced from 4X cells (Fig. 4b). In contrast, TH-positive cells were occasionally 356 scattered throughout organoids produced by H9 cells; however, we rarely detected TH- According to the single-cell sequencing data, the majority of H9 cells were VLMCs 363 (Clusters 1, 2, 3, 7, 9, 10; 93% of H9 cell). To confirm this finding, we examined the expression 364 of VLMC markers in organoids at 83 DIV, and we identified a large population of 365 COL3A1/COL1A1 double-positive cells with a nonneuronal morphology among H9 cells (Fig.  366 4c). No cells positive for COL3A1 or COL1A1 were identified among 4X cells (Fig. 4c). 367

DA neurons derived from LR-USCs exhibit pacemaker activity 369
Having shown that we can generate mesDA neurons from 4X cells under caudalizing 370 conditions, we next wanted to examine the electrophysiological properties of the DA neurons. 371 We performed in vitro electrophysiological recordings in whole-cell patch-clamp configuration 372 between DIV 80 and DIV 84 (Fig. 4d). We observed that the cells developed into 373 electrophysiologically mature neurons, as measured by their ability to generate repetitive 374 action potentials upon somatic current injection (Fig. 4e). Recordings in current-clamp mode 375 revealed spontaneous pacemaker activity characteristic of a DA neuron identity, with a mix of 376 single spikes and phasic bursts (Fig. 4f). Membrane oscillations collapsed at potentials below 377 -50 mV (data not shown). The firing frequency in our sample ranged from 1 to 5 Hz (Fig. 4g). 378 Furthermore, HPLC analysis of cell extracts showed that the DA content in the 4X cells was 379 significantly higher than that in the H9 cells (287.4 nmol/g in 4X cells vs. 65.1 nmol/g in H9 380 cells, P = 0.002; Fig. 4h). 381

Analysis of 4X cells in vivo in a Parkinson's disease rat model 383
When using current DA neuron differentiation protocols, DA neurons account for only 384 a small percentage of cells of the entire graft when mesDA progenitors are grafted in vivo. A 385 protocol using FGF8 to induce midbrain caudalization was shown to result in the production 386 of approximately 3,700 TH-positive DA cells per 100,000 grafted cells 14 , and a more recently 387 reported protocol using a delayed boost of WNT was found to induce the generation of 9,173 388 TH-positive cells per 6.22 mm 3 graft following transplantation of 450,000 cells 13 . We innervation. All three groups showed forelimb asymmetry in the cylinder test, with the rats 400 using mostly the ipsilateral forepaw (6-OHDA 52.4%; H9 70.4% and 4X 67.4% of total) and 401 almost never the contralateral forepaw (6-OHDA 1.3 %; H9 1.3% and 4X 0% of total) to touch 402 the walls or land on the floor after rearing, further supporting the induction of a DA deficit by in the cylinder test confirmed the significant improvement in 4X cell-transplanted rats, as 415 these rats used the contralateral forelimb alone (9.7% of total) or together with the ipsilateral 416 forelimb (both 46.3%) during the test at week 18 (Fig. 5c). However, both H9 cell-transplanted 417 and 6-OHDA lesion rats used mostly the ipsilateral forelimb (76.1% and 79.3% of total 418 respectively), using both forelimbs less than 30% of the time but almost never using the 419 contralateral impaired forelimb when rearing in the cylinder, as observed before 420 transplantation (Fig. 5c). Therefore, 4X cell transplantation significantly improved both drug-421 induced and spontaneous motor behavior after 6-OHDA-induced lesioning of the MFB. 422 Postmortem histological analysis of the brains showed that rats transplanted with 4X 423 cells had graft-derived TH-positive cells in the area of injection, i.e., the striatum, as well as in 424 globus pallidus, the corpus callosum and the area of the cortex above the striatum (Fig. 5d). Interestingly, we also found that TH-positive neurons derived from H9 cells were 447 positive for FOXA2, LMX1A and EN1 ( Supplementary Fig. 6c-e). This was in contrast to our in 448 vitro experiments, in which TH-positive neurons derived from H9 cells rarely expressed FOXA2 449 (Fig. 4b, Supplementary Fig. 5a), suggesting that the in vivo environment is more permissive Thus far, we have described how knockout of four selected genes can dramatically 493 increase the specification of PSCs to mesDA neurons by restricting the cell types along the A-494 P axis that they can differentiate into. It is possible to further restrict cell fate by knocking out 495 additional genes to prevent differentiation into remaining populations of unwanted cells, 496 which would further enhance the ability to generate mesDA neurons. Specifically, our single-497 cell sequencing data showed that 4X cells are capable of producing telencephalic and anterior 498 hindbrain cell types. By targeting transcriptional determinates of these populations, we 499 speculate that we could eliminate these populations. Furthermore, we can also target dorsal 500 populations in addition to populations along the A-P axis. It is conceivable that by restricting 501 the genome even further, we can produce a cell line that is capable of producing a highly pure 502 population of mesDA neurons. suspension. From day 9 to day 11, the supplements in the medium were replaced with FGF8 542 (100 ng/ml; R&D Systems). From day 11, the medium was supplemented with FGF8 (100 543 ng/ml), LM22A4 (2 µM), and ascorbic acid (200 µM; Sigma). On day 16, the cells were 544 dissociated with Accutase and subsequently grown on culture plates coated with 545 polyornithine, fibronectin, and laminin (all from Sigma). Neural differentiation medium 546 consisting of 1% B27 supplement, 25 U/mL pen/strep, 0.5% Glutamax was supplemented with 547 200 µM ascorbic acid, LM22A4 (2 µM), 1 µM DAPT (Tocris Bioscience), GDNF (10 ng/ml), and 548 dcAMP (500 µM). The medium was changed every second day until the end of the experiment. 549 Alternatively, on day 16, the cultured cells were maintained in suspension to generate 550 organoids. 551 552 553

Generation of knockout cell lines 570
Three lentiviral plasmids, pLV-4gRNA-GBX2-RFP, pLV-hUbC-GBX2-CDX124-Cas9-T2A-GFP and 571 lentiCas9-Blast (Addgene # 52962), were used to produce lentiviruses. Lentiviral production 572 was performed as described previously 43 . To generate the GBX2 knockout cell line, H9 cells 573 were transduced with LV-4gRNA-GBX2-RFP and lentiCas9-Blast, and after three days, 574 transduced cells were selected using 10 µg/ml blasticidin for 6 days ( Supplementary Fig. 1). 575 FACS was then used to separate single RFP-positive cells in a 96-well plate using the 561 nm 576 laser on a FACSAriaIII (BD Biosciences, San Jose, CA). Indels at the corresponding target sites 577 in the clones were analyzed by genomic PCR. To generate the 4X knockout cell line, H9 cells 578 were infected with LV-hUbC-GBX2-CDX124-Cas9-T2A-GFP, and after 7 days, single GFP-579 positive cells were sorted by FACS (Supplementary Fig. 2). Allele-specific mutations in both 580 the GBX2 -/and 4X cell lines were confirmed using whole-exome sequencing. Whole-exome 581 sequencing and mapping were performed by BGI (BGI, Copenhagen). Integrated Genome 582 Browser V 2.10.0 was used to identify allele-specific mutations. To identify large deletions that 583 could not be mapped by the alignment tools, individual sequencing reads were extracted from 584 the FastQ files using Grep and manually analyzed. for one hour at room temperature. After the secondary antibodies were removed, the cells 619 were washed three times with PBST for 10 min each in the dark. The nuclei were 620 counterstained with DAPI (1 μg/ml, Sigma) and rinsed with PBS three times for 5 min each. 621 The slides or coverslips were mounted with PVA-DABCO. Images were captured with a 622 confocal microscope (Zeiss LSM 780) and Zen software. that were obtained, 16 were kept for analysis. The rest of the recordings were from neurons 718 that either were nonrespondent to depolarizing steps (putative astrocytes), were unstable, or 719 did not exhibit spontaneous activity; therefore, these recordings were discarded from the 720 analysis. Then, the samples were centrifuged, and the supernatant was collected and spun through a 734 0.2 µm spin filter (Costar Spin-X, Merck) at 14000 × g at 4 °C for 1 min and loaded into an HPLC 735 system (Thermo Scientific Ultimate 3000). The mobile phase was 12.5% acetonitrile buffer (pH 736 3.0, 86 mM sodium dihydrogen phosphate, 0.01% triethylamine, 2.08 mM 1-octanesulfonic 737 acid sodium salt, and 0.02 mM EDTA). The flow rate of the mobile phase was adjusted to 1.5 738 ml/min. The dopamine level was calculated using a standard curve generated using external 739 DA standards (the standard curve coefficient of determination was 0.99946). Dopamine 740 content was then normalized to the protein concentration and is expressed in nmol/g. The rats were anesthetized with isoflurane (5% for induction, 2-3% for maintenance), 1. attached. Following injection, the cannula was left in place for 5 min before being slowly 766 retracted. The incision was sutured, and the animals were injected with buprenorphine (0.36 767 mg/kg) as an analgesic. Once the animals were fully awake, they were placed back into their 768 cages with wet food and 0.009 mg/ml Temgesic in water. 769 Lesioning efficiency was assessed 3 weeks postsurgery using the amphetamine-induced 770 rotation test, and animals that exhibited >5 rotations/min were used for further experiments. 771 The selected rats were divided into 3 groups with a similar average number of amphetamine-772 induced rotations: the 6-OHDA lesion (no transplantation) group (n = 8), the H9 cell-773 transplanted group (n = 9) and the 4X cell-transplanted group (n = 9) (see Supplementary Fig.  774   6a, b). Four weeks after lesioning, the animals in the H9 cell-transplanted and 4X cell-775 transplanted groups were stereotaxically injected into the striatum (AP, +0.5; ML, -3; DV, -776 4.6/4.8) with 250,000 cells of the respective cell type in a volume of 2.5 µl using a protocol 777 similar to the one described above. All three groups were sacrificed 22 weeks postlesioning 778 (i.e., 18 weeks after transplantation). Two transplanted rats did not complete the study and 779 were euthanized due to health issues: one in the 4X cell-transplanted group (week 8 780 posttransplantation) due to a broken tail and one in H9 cell-transplanted group due to 781 hindlimb paralysis (week 17 posttransplantation). 782 783

Amphetamine-induced rotation test 784
An amphetamine-induced rotation test was performed as described previously 50

Cylinder test 796
The cylinder test was used to assess paw use asymmetry three weeks postlesioning (one week 797 prior to transplantation) and 18 weeks posttransplantation. The animals were placed in a 798 transparent Plexiglas cylinder (height of 30 cm, diameter of 20 cm), and two mirrors were 799 placed behind the cylinder so that the cylinder surface could be fully visualized. Spontaneous 800 activity was video recorded for a total of 5 min. Data analysis was performed by a researcher 801 blinded to the groups using VCL Media Player software in slow motion as previously described 802 51 . Because most of the exploratory motor activity of the animals was limited to the first 2 min 803 and there was little movement after this timepoint, activity in the first 2 min were analyzed, 804 and activity after this time point was analyzed only if the animal exhibited fewer than 10 805 movements (wall touches and rears). The following behaviors were scored to determine the 806 extent of forelimb-use asymmetry 51 : a) independent use of the left or right forelimb when 807 touching the wall during a full rear or landing on the floor after a rear and b) simultaneous use 808 of both the left and right forelimb to contact the wall of the cylinder during a full rear, for 809 lateral movements along the wall (wall stepping) and for landing on the floor following a rear. 810 The data are presented as the percentage of time each forelimb (left or right) or both 811 forelimbs were used relative to all movements (wall and floor). 812

Immunohistochemical analysis of brain slices 814
The rats were killed 23 weeks after 6-OHDA-induced lesioning by an overdose of pentobarbital 815 (50 mg/kg i.p.). During respiratory arrest, they were perfused through the ascending aorta 816 with ice-cold saline followed by 4% cold PFA (in 0.1 M NaPB, pH 7.4). The brains were 817