Rapid generation of regionally specified CNS neurons by sequential patterning and conversion of human induced pluripotent stem cells

The di ﬀ erentiation of patient-speci ﬁ c induced pluripotent stem cells (iPSCs) into speci ﬁ c neuronal subtypes has been exploited as an approach for modeling a variety of neurological disorders. However, achieving a highly pure population of neurons is challenging when using directed di ﬀ erentiation methods, especially for neuronal subtypes generated by complex and protracted protocols. In this study, we e ﬃ ciently produced highly pure populations of regionally speci ﬁ ed CNS neurons by using a modi ﬁ ed NGN2-Puromycin direct conversion pro- tocol. The protocol is amenable across a range of iPSC lines, with more than 95% of cells at day 21 positive for the neuronal marker MAP2. We found that conversion from pluripotent stem cells resulted in neurons from the central and peripheral nervous system; however, by incorporating a short CNS patterning step, we eliminated these peripheral neurons. Furthermore, we used the patterning step to control the rostral-caudal identity. This approach of sequential patterning and conversion produced pure populations of forebrain neurons, when patterned with SMAD inhibitors. Additionally, when SMAD inhibitors and WNT agonists were applied, the approach produced anterior hindbrain excitatory neurons and resulted in a neuronal population containing VSX2/SHOX2 V2a interneurons. Overall, this sequential patterning and conversion protocol can be used for the production of a variety of CNS excitatory neurons from patient-derived iPSCs, and is a highly versatile system for investigating early disease events for a range of neurological disorders including Alzheimer ’ s disease, motor neurons disease and spinal cord injury.


Introduction
Since the derivation of induced pluripotent stem cells (iPSCs), patient-specific iPSCs have been produced from a broad range of patients with sporadic or familial forms of neurological diseases. When combined with gene editing techniques to correct or introduce mutations, the diseased and isogenic iPSCs can be used together to investigating disease mechanisms in vitro. A constant focus in the field has, therefore, been towards the development of differentiation protocols that can more closely recapitulate sequential in vivo patterning events and direct cell fate towards specific brain regions, in order to generate appropriate neuronal subtypes for modeling a neurological disease (Arenas et al., 2015;Davis-Dusenbery et al., 2014). One of the current challenges in the field is that directed differentiation protocols take several weeks or even months to produce high percentages of post-mitotic neurons, the cells become more asynchronous as the protocol becomes more protracted, and cultures come to contain varying amounts of progenitor cells at differing maturity. These restraints hamper the use of highthroughput analyses or sequencing methods. Several approaches have been adopted to remove progenitors or accelerate their differentiation to a mature neuronal state. However, accelerated differentiation using various combinations of inhibitors has only been successful at producing a limited range of neuronal subtypes of either forebrain or sensory neurons, depending on the inhibitors used (Chambers et al., 2012;Qi et al., 2017;Telezhkin et al., 2016).
An alternative to directed differentiation is direct reprogramming, which can be performed from fibroblasts or other somatic cells and is capable of directly generating post-mitotic neurons (Ambasudhan et al., 2011;Chanda et al., 2014;Vierbuchen et al., 2010). Applying this method to pluripotent stem cells overcomes limitations in the T availability of precious patient cells and, when combined with a resistance gene to select for transgene expression, the method can result in a highly pure population of excitatory neurons (Chanda et al., 2014;Pang et al., 2011;Thoma et al., 2012;Zhang et al., 2013). Conversion to alternate neuronal lineages can also be achieved by supplementing the neuronal conversion genes with lineage-specific transcription factors. However, the inclusion of multiple conversion factors to generate alternate neuronal subtypes introduces additional inefficiencies and requires an understanding of which transcription factor determinants are specifically required for each lineage (Blanchard et al., 2014;Pfisterer et al., 2011;Son et al., 2011;Theka et al., 2013). Having the ability to directly convert cells to a broader range of neurons without the need to identify a specific transcription factor determinate would considerably increase the applicability of direct conversion as a method to generate neurons for disease modeling.
In this study, we sought to develop a platform for converting iPSCs into region-specific neurons. By combining the early patterning events of a differentiation protocol with a direct conversion approach, we produced highly pure, regionally specified induced neurons (iNs). To achieve this, we began by optimising an NGN2-Puromycin direct conversion protocol to produce pure populations of excitatory neurons. We demonstrated that optimisation of the antibiotic selection period and concentration enables highly pure neuronal cells to be obtained, and the addition of mitotic inhibitors can be avoided, resulting in a protocol that is amenable across a broad range of iPSC lines. Interestingly, the NGN2-conversion from the pluripotent state yielded a mixed neuronal culture containing a significant number of peripheral neurons. To overcome this and have the ability to control the differentiation towards specific central nervous system (CNS) fates, we incorporated a short differentiation-patterning step that successfully prevented the generation of peripheral neurons. By using factors known to alter the rostral-caudal identity, we successfully specified the neurons to specific brain regions and produce pure populations of either forebrain or anterior hindbrain excitatory neurons. Thus, this approach can produce regionally specified post-mitotic CNS excitatory neurons from patientderived iPSCs, which stands as a highly relevant system for investigating a range of neurological disorders.

Electrophysiology
For the electrophysiological recordings, iNs were replated on individual 13 mm coverslips (VWR, Ø 13 mm) on day 5 following an additional 8 weeks of maturation time. The iNs on the coverslips were transferred into a recording chamber and continuously perfused with Artificial Cerebrospinal Fluid containing (in mM): NaCL 119, KCl 2.5, NaHCO3 26, 1 NaH2PO4, D-glucose 11, CaCl2 2, MgCl2 2, adjusted to pH 7.4 and bubbled with a mixture of CO2 (5%) and O2 (95%), at room temperature. The chamber was mounted on an upright microscope (Scientifica) linked to a digital camera (QImaging Exi Aqua) and the cells were visualized using 63X water-immersion objective (Olympus). Acquisitions were performed in a whole-cell configuration using Clampex 10.6 software connected to a Multiclamp 700B amplifier via a Digidata 1550A digitizer (all from Molecular Devices). Voltage-clamp data were low-pass filtered at 200 Hz and digitized at 10 kHz and the whole-cell capacitance was compensated. Patch pipettes (3-7 MΩ of resistance) were filled with internal solution containing (in mM): Kgluconate 170, Hepes 10, NaCl 5, KCl 10, EGTA 0.6, Na2GTP 2, and NaGTP 0.3. The pH and osmolarity of the internal solution were close to physiological conditions (pH 7.25, osmolarity 285 mOsm). The access resistance of the cells in our sample was~19 MOhm. Spontaneous excitatory postsynaptic currents (sEPSC) were recorded in voltageclamp gap-free mode (clamped at −65 mV). Current-clamp mode was used to analyse the intrinsic firing properties of the cell. Briefly, repetitive depolarizing current pulses (800 ms) of incremental amplitude (10 pA) were used to determine the rheobase corresponding to the minimum depolarizing current intensity needed to generate an action potential (AP). Amplitude and spike-width of APs were measured on the first AP at the rheobase level. Repetitively firing neurons were defined as those capable of firing ≥ 3 APs in response to a depolarizing current step. The amplitude of APs was measured from the point preceding the fast rising phase of the spike to the peak of the spike, whereas APs duration was defined as the width at half-maximal spike amplitude. Data analysis was performed by Clampfit 10.6 (Molecular device). Data are shown as the mean ± S.E.M.

RNA sequencing and analysis
For RNA sequencing, we used the hESC line and differentiated them into the three iN groups NGN2, SB-CHIR-NGN2 and SB-LDN-NGN2 and extracted RNA at day 21. RNA from undifferentiated hESCs were collected and used as a control. Three biological replicates from each group were collected, resulting in a total of 12 samples sequenced. Specifically, RNA was extracted using RNeasy Mini Kit (Qiagen) following the manufacture's instruction, and sent to BGI-China for mRNA library construction, which consisted of mRNA enrichment by oligo dT selection using magnetic beads. RNA was fragmented and reverse transcribed to cDNA by N6 random primers. cDNA was amplified by PCR and single-end sequencing performed with a minimum of 20 M clean reads acquired per sample on a BGISEQ500 platform. Bioinformatic analysis followed standard procedures of removing adaptors, mapping to the reference human genome (hg19) using HISAT and Bowtie2 tools, and differential gene expression detected using DEseq2. For principal component analysis, the full transcriptomic profile (21,774 genes) was included for analysis. Principal components were calculated by singular value decomposition of the centred and scaled data matrix using the prcomp function in R stats package. Heatmaps were generated by log2 transforming and mean scaling FPKM values of selected genes using heatmap.2 function in R stats package. A Venn diagram was constructed using differentially expressed genes (adjusted p-value lower than 0.05) identified by comparison of NGN2, SB-LDN-NGN2 or SB-CHIR-NGN2 group to hESCs. Bar graphs of the top 10 significantly up/down-regulated genes present Log2 (Fold Change) values of differentially expressed genes, identified by comparison of SB-LDN-NGN2 or SB-CHIR-NGN2 group to NGN2. Threshold bar (white line) indicates a fold change of ± 2. RNAseq data reported in this paper is deposited in the Gene Expression Omnibus (GEO) with accession number GSE149864.

Multi-electrode array (MEA) recording and analysis
Three groups of iNs were plated on day 5 with a density of 100,000 cells/well on poly-L-ornithine (0.1 mg/mL), laminin (10 μg/mL) and fibronectin (10 μg/mL) (all from Sigma-Aldrich) triple-coated 24-well MEA plate (Multi Channel Systems MCS GmbH) in Neurobasal medium supplemented with B27 2%, penicillin/streptomycin 0.5%, ascorbic acid (200 μM; Sigma-Aldrich), DAPT (2.5 μM; Tocris), doxycycline (1 μg/mL), and BDNF (20 ng/ml; R&D systems). Since there was no obvious spontaneous activities detection on the day 21 culture for the NGN2 group, human astrocytes (Sciencell) were added with a density of 25,000 cells/well to support the culture system for electrophysiological recordings for the all the three groups. Spontaneous firing was recorded for two minutes each time, twice a week for up to 10 weeks with Multiwell-Screen software (Multi Channels Systems MCS GmbH). Signals were sampled at 20 kHz and low-pass Butterworth filterer cutoff was 3500 Hz and high-pass Butterworth filter cutoff was 100 Hz. Data analysis were performed with Multiwell-Analyzer software (Multi Channels Systems MCS GmbH). A spike detection was recorded when signal exceeded a threshold of ± 5% of standard deviation of the baseline noise during quiescent periods. A burst detection was recorded following these settings: maximum interval to start and end burst was 50 ms; the minimum interval between bursts was 100 ms; the minimum duration of burst was 50 ms, and there were more than 4 spikes counts in a burst. A network burst detection was recorded when minimum 6 channels participating and minimum simultaneous channels are 3.

Direct conversion to iNs from human ESCs or iPSCs
Previous studies have demonstrated that direct conversion of hESCs M. Chen, et al. Stem Cell Research 48 (2020) 101945 or iPSCs into neurons can be achieved via a single proneural transcription factor (Chanda et al., 2014;Goparaju et al., 2017). NeuroD1 or NGN2, are among the most potent; ASCL1, whilst also effective, has been reported to have a slower conversion rate (Goparaju et al., 2017;Zhang et al., 2013). We, therefore, selected NGN2 as the conversion factor in our study and first sought to determine its robustness at converting human iPSCs into neurons. By lentiviral delivery, we overexpressed either NGN2 and Puromycin N-acetyl transferase or NGN2, EGFP and Puromycin N-acetyl transferase; both constructs contain a doxycycline inducible tetO promoter (Zhang et al., 2013). Two days after lentiviral transduction, a high percentage of cells were EGFP positive, with only a few non-transduced EGFP negative cells present ( Supplementary Fig. 1A). Despite the high transduction rate, cells without puromycin selection showed a vast difference in cellular morphology by day 7 and only a few cells presented with a neuronal morphology ( Supplementary Fig. 1B). Conversely, after five days of puromycin selection at 2 μg/ml, we observed an abundance of neurons; intriguingly, however, RBFOX3 positive cells with a non-neuronal morphology were present within these cultures (Fig. 1B). To prevent the appearance of these non-neuronal cells, we first tested a range of puromycin concentrations, up to 10 μg/ml. We determined that after five days of puromycin selection, a concentration of 3 μg/ml was sufficient to dramatically reduce the presence of non-transduced cells (EGFP negative) and non-neuronal RBFOX3 positive cells in the culture by day 7. Secondly, we included a replating step to further enhance the purity of converted cells ( Supplementary Fig. 1C-D). The percentage of EGFP positive cells at a replating density of 250,000 cells/cm 2 on day 7 was 89.84% ± 4.13 S.D. Decreasing the plating density to 120,000 cells/ cm 2 resulted in no detectable EGFP negative cells. By combining the optimised puromycin concentration and replating density, we efficiently and reproducibly converted all iPSC lines to neurons (Fig. 1C-F). Across all the different lines, the percentage of transduced cells on day 14 and day 21 was above 97.93% ± 1.31 S.D (Fig. 1D). The neuronal marker MAP2 increased slightly from day 14 to day 21, all lines were above 89.51% ± 3.84 S.D at day 14, and above 95.06% ± 6.38 S.D at day 21 (Fig. 1E).

Electrical activity of the converted neurons
Whole-cell electrophysiological recordings confirmed that the neurons exhibited neuronal activity, indicative of mature, functional neurons. At days 84-86, the recorded neurons showed sEPSC, with an average amplitude of 46,0 ± 11,4 pA and a frequency of 20,4 ± 4,9 events per minute, confirming the formation of excitatory synapses ( Fig. 2A). The cells' intrinsic membrane properties were also measured as an indication of a functional maturation: the membrane resistance was 500,8 ± 138,2 MOhm, and mean capacitance was 39.90 ± 5.28 pF (Fig. 2B). We then asked if the cells were able to fire action potentials (APs) in response to the injection of current, and observed that repetitive firing can be induced in sufficiently depolarized cells. In our experiment, the injected current necessary to induce a first AP (i.e, rheobase, red trace) was 27.5 ± 6.3 pA (Fig. 2C). The rheobase was directly linked to the resting membrane potential of the cell (Vm). At days 84-86, the Vm of our recorded cells was −50,5 ± 4,1 mV (Fig. 2D). Detailed analysis of AP properties revealed a mature shape of the APs, with a discharge threshold (APthresh) of −31.6 ± 1.9 mV, AP amplitude averaged 38.5 ± 8.0 mV, and a half-width of 6.5 ± 1.7 ms (Fig. 2D). Finally, we saw that the cells responded by an increase in the number of APs to an increment of the amplitude of the current injected (Fig. 2D).

Combining differentiation and direct conversion to generate pure CNS neurons
Having optimised an efficient neuronal conversion method, we then further characterised the types of neurons generated from the Ngn2conversion protocol. In agreement with our electrophysiological recordings, the iNs displayed punctate staining for both synaptophysin (SYP) and postsynaptic protein PSD-95, which indicates synaptic maturity (Fig. 3A). Furthermore, we observed the expression of neuronal marker RBFOX3 (Fig. 3B) and glutamate transporter SLC17A6 (VGLUT2) in the neurons (Fig. 3C). We could not detect GABA or THpositive neurons, or astrocytes (GFAP-positive; data not shown), although interestingly, a population of peripheral neurons was observed in the cultures (Fig. 3D). The percentage of PRPH + /DAPI neurons was 9.40% ( ± 3.29 S.D), 10.24% ( ± 3.05 S.D), and 9.20% ( ± 3.55 S.D) across three different cell lines.
To prevent the presence of peripheral neurons, we incorporated a short differentiation step before the direct conversion protocol (Fig. 4A). Two differentiation methods were developed that patterned the hESCs/iPSCs into neural progenitors of either the forebrain or hindbrain. For forebrain specification we patterned the cells with SB431542 and LDN193189 referred to as SB-LDN-NGN2 and for hindbrain patterning we used SB431542 and CHIR99021, referred to as SB-CHIR-NGN2. Previously, we reported both of these patterning methods to successfully differentiate iPSCs to NSCs of these regions (Chen et al., 2018;Denham et al., 2015Denham et al., , 2012. Regardless of the patterning, upon neuronal conversion and selection, the two new protocols produced neurons with high efficiency similar to the NGN2 group. We next used RNAseq to determine how the patterning had altered their neuronal identity. Firstly, gene expression from all the samples was normalized and compared using principal component analysis (Fig. 4B). We observed that PC1 had the highest variance (50.7%) and revealed a clear separation between the undifferentiated stem cell state and the neuronal groups, whereas PC2 could separate the neuronal groups based on anterior-posterior identity: the SB-CHIR-NGN2 group separated from the NGN2 and SB-LDN-NGN2 groups with a variance of 18.1%. Within the four groups, we compared the expression of pluripotent genes (Fig. 4C). We observed that the pluripotent genes NANOG and POU5F1, as well as SOX2, were highly expressed in the hESCs group, but not in the three iN groups. Secondly, we examined the gene expression of neuronal lineage markers (Fig. 4D). As expected, all the three direct conversion groups, when compared to hESCs, expressed high levels of the mature neuronal markers SYP, DLG4, and MAP2, as well as higher levels of neuronal lineage markers (Fig. 4D). Uniquely to the NGN2 group, a high level of the peripheral neuronal marker PRPH was observed ( Fig. 4D), which is in accordance with our immunostaining results (Fig. 3D). The inclusion of a patterning step, for both SB-LDN and SB-CHIR, significantly reduced the PRPH transcript levels (−4.14 log2 fold change of SB-LDN vs. NGN2 at a padj = 2.11 × 10 −185 and −5.09 log2 fold change of SB-CHIR vs. NGN2 at a padj = 2.96 × 10 −233 ) and no PRPH positive cells could be identified by immunostaining in the SB-LDN or SB-CHIR groups ( Supplementary Fig. 2). Compared to hESCs, all iN groups expressed significantly higher vesicular glutamate transporter SLC17A7 (SB-LDN padj = 3.50 × 10 −38 and SB-CHIR padj = 1.77 × 10 −39 ) and SLC17A6 (SB-LDN padj = 1.02x10 -282 and SB-CHIR padj = 4.12x10 -143 ) levels, while very few transcripts were detected for GABA transporter SLC6A1 in SB-CHIR-NGN2 and SB-LDN-NGN2 groups, and no detectable transcripts were identified in any of the groups for GAD1 and GAD2 (< 1 FPKM).
We next examined the expression of anterior-posterior neuronal markers and forebrain markers. The SB-LDN-NGN2 group expressed the highest level of SIX3 and FOXG1 (Fig. 4C). The forebrain/midbrain marker OTX2 was also higher in SB-LDN-NGN2 group, compared to NGN2 group and SB-CHIR-NGN2 group. At the midbrain level, the markers EN1 and PAX5 were not detected in any of the groups, and only very few transcripts for PAX2 were detected in the NGN2 group (1.24 FPKM). At the hindbrain level, the SB-CHIR-NGN2 group expressed the highest level of HOXA1 and HOXA2 (Fig. 4C). For cervical, thoracic, lumbar and sacral markers; no transcripts were detected in the four groups. Gene expression comparisons showed that all three direct-converted groups shared 7650 genes (out of 12,264 genes) and, in accordance with PCA analysis, the SB-CHIR-NGN2 group contained the most unique genes (1793) from the other two neuronal populations, which accords with its more distinct hindbrain identity (Fig. 4E). The most differentially expressed genes were identified by direct comparison between the SB-CHIR-NGN2 group and the NGN2 group, as well as between the SB-LDN-NGN2 group and the NGN2 group (Fig. 4F). Of significance, in the SB-CHIR-NGN2 group, VSX2 (previously known as CHX10), NEUROG3, POU3F1, and SHOX2 were highly upregulated, while in the NGN2 group, ISL1, NEUROG2 and PRPH were highly upregulated. In the SB-LDN-NGN2 group, SIX3 and FOXG1 were among the most differentially expressed genes (Fig. 4F). Overall, the results of the RNAseq analysis revealed that the SB-LDN-NGN2 group resulted in a forebrain cortical neuron identity, and that the SB-CHIR-NGN2 group resulted in a hindbrain identity.
VSX2 and SHOX2 are both expressed in V2a interneurons, and both genes were highly expressed in the SB-CHIR-NGN2 group. V2b interneurons that express GATA3 are closely related to V2a interneurons, but GATA3 was not detected in any of the groups. We were therefore interested in confirming the presence of V2a interneurons within the SB-CHIR-NGN2 group and, by immunostaining, we readily observed VSX2 and SHOX2 double-positive cells (Fig. 4G), with an efficiency of 43.17% ± 16.81 S.D, 45.83% ± 2.93 S.D, and 43.07% ± 8.97 S.D ((VSX2 + SHOX2 + )/DAPI)% for V2a interneurons converted from hESCs, GBA-006-C6, and GBA-002-C3 respectively. Immunostaining also confirmed that NEUROG3-positive cells featured prominently in SB-CHIR-NGN2, but not in NGN2 and SB-LDN-NGN2 (Fig. 4H). In the SB-LDN-NGN2 group, only a few SHOX2-positive neurons were observed and no VSX2-positive neurons were found. In the NGN2 group, neither VSX2 nor SHOX2 neurons were observed (Fig. 4G). Additionally, RNA in situ images (from the Allen Developing Mouse Brain Atlas in E11.5) of mouse embryos were used to confirm the spatial location of Vsx2, Neurog3, and Shox2, which identified these markers to be expressed in the ventral hindbrain and spinal cord, where V2a interneurons are located (Supplementary Fig. 3).
In the SB-LDN-NGN2 group, TBR1 expression levels were significantly higher than in any of the other groups. To confirm its expression, we examined all neuronal conversion groups by immunostaining. We identified TBR1 in both the NGN2 and SB-LDN-NGN2 groups and not in the SB-CHIR-NGN2 group (Fig. 5). For the SB-LDN-NGN2 group, the efficiency of (TBR1 + /DAPI)% was 65.55% ± 4.16 S.D for hESCs and 62.84% ± 6.68 S.D for GBA-002-C3.

Electrophysiological activity and functionality of the three iN groups by MEA
We used MEA to examine the electrophysiological activity of the neurons in all the three groups. We detected neural activity, recorded as spikes, first in the SB-CHIR-NGN2 group on day 16, and bursts and network bursts were seen on day 23. Following this, in the SB-LDN-NGN2 group, bursts were detected on day 30 and network bursts on day 54. The NGN2 group was the last group to show bursts (on day 33) with no network bust detection throughout the 65 days culture. On day 49, the SB-CHIR-NGN2 group showed a synchronized network burst with voltage around 400 µV. The SB-LDN-NGN2 group had network bursts with voltage around 50 µV. The NGN2 group had bursts but not network burst with the voltage around 30 µV (Fig. 6).
All the three iN groups responded to the treatment of AMPA/kainite glutamate receptor antagonist NBQX. The spontaneous firing decreased dramatically for both SB-CHIR-NGN2 and SB-LDN-NGN2 groups, while there were still some firing signals for the NGN2 group. With TTX treatment, the firing was eliminated from all the three groups ( Supplementary Fig. 4).

Discussion
In this study, we established a robust and rapid protocol for generating regionally specified iNs from human iPSCs by combining a short patterning step with a direct conversion approach. We started out by examining the previously described NGN2-puromycin method, which has been shown capable of rapidly converting human iPSCs and hESCs into excitatory neurons with a cortical identity (Zhang et al., 2013). We determined that, in order to achieve a highly pure population of neurons, and to overcome variability in efficiencies between experiments and individual hiPSC lines, we needed to make several important modifications to the existing protocol.
We first optimised the concentration and duration of puromycin, as well as including a cell-replating step that further removed any remaining non-selected cells and dead cells and allowed for neurons to be replated at a consistent density, which was essential for the downstream transcriptomics and also provided a scalable and standard platform for converting numerous iPSCs lines in parallel.
After implementing these modifications, the percentage of transduced cells was above 97.93% ( ± 1.31) on day 14 and day 21, and the percentage of cells expressing the mature neuronal marker MAP2 was above 95.06% ( ± 6.38) (Fig. 1). Furthermore, we demonstrated that the neurons were synaptically mature: the 21-day culture contained SYP-positive synaptic punctate and PSD95 (Fig. 3A), and patch-clamp electrophysiology showed that neurons exhibited mature functional activity at 12 weeks of culture (Fig. 2). Overall, our replating step and puromycin optimisation successfully overcame the problem of cell line variability, ensured a consistent neuronal density at the end-point for all iPSC lines used, and resulted in functionally active neurons without the need to supplement the cultures with astrocytes, which thus ensured that the transcriptome analysis was purely from converted neurons.
Close examination of our NGN2 converted neurons revealed that peripheral neurons were present within the culture (Fig. 3D). To address this heterogeneity, we made a second modification to the protocol: we combined a differentiation/patterning step with the conversion to generate pure CNS neurons. Previous research has shown that hESCs can be patterned towards a forebrain identity with SMAD inhibition or towards a hindbrain fate by SMAD inhibition and activation of WNT signalling . We examined the effect of combining either of these two differentiation protocols with the NGN2conversion approach. Our RNA sequencing analysis showed that both patterning steps, when followed by direct conversion, completely eliminated the presence of peripheral neurons (Fig. 4B). Additionally, the identity of the converted neurons was differentially altered according to the extrinsic patterning. We observed that SB-LDN patterning, when combined with NGN2 conversion, was an effective strategy for generating an enriched culture of forebrain cortical neurons. Both SB-LDN-NGN2 and NGN2 conditions converted hESCs towards a cortical identity (Fig. 4B, D), however, SB-LDN patterning produced a more pure population of forebrain neurons, whereas the NGN2 alone could also produce peripheral neurons (Fig. 3D); RNAseq analysis revealed expression of genes representative of broader brain regions (Fig. 4C). In accordance with this, the gene expression of forebrain markers SIX3 and FOXG1 in the SB-LDN-NGN2 group was significantly higher than in the NGN2 group, and the SB-LDN-NGN2 group had no detectable levels of the midbrain markers PAX2, PAX5 or EN1 (Fig. 4C).
Patterning of hESCs/iPSCs into hindbrain progenitors by treatment with SB431542 and CHIR for 4 days, followed by NGN2 conversion, generated anterior hindbrain excitatory neurons (Fig. 4). VSX2 was seen to be significantly upregulated in the SB-CHIR-NGN2 group and we could detect an abundance of these neurons within the culture (Fig. 4G). VSX2 is a transcription factor and, within the hindbrain and spinal cord, it is specific to V2a interneuron (Al-Mosawie et al., 2007;Clovis et al., 2016;Li et al., 2005;Lundfald et al., 2007). To further confirm the identity of the interneuron, we examined the expression of SHOX2 by immunostaining, which co-localized with VSX2. Significantly more transcripts for SHOX2 were also detected in the SB-CHIR-NGN2 group. Interestingly, we observed that not all VSX2 interneurons expressed SHOX2. Neurons double-positive for VSX2 and SHOX2 are known to be excitatory interneurons that control motor neurons, while the VSX2 + SHOX2activate the commissural pathways involved in left-right coordination. (Dougherty et al., 2013;Ha and Dougherty, 2018).
A recent study demonstrates that V2a interneurons can be differentiated from human PSCs by dual SMAD inhibition, followed by retinoic acid and low concentrations of a Smoothened agonist, and Notch inhibition (Butts et al., 2017). By contrast, we obtained V2a interneurons without the inclusion of extrinsic factors to specifically activate the SHH pathway, which suggests that some basal plate SHH targets may be activated directly by the NGN2 conversion. In accordance with this, Ngn2 is regionally expressed in the basal plate of the developing mouse spinal cord and has been shown to be involved in regulating the expression of ventral transcription factors (Scardigli et al., 2001). The neuralizing transcription factor in direct conversion protocols may therefore also influence the neuronal subtypes generated. Furthermore, in direct conversion experiments NGN2 has been shown to preferentially give rise to excitatory neurons. As expected, by inhibiting the AMPA receptor, we found that indeed excitatory neurons were the main neuronal cell type present in all of our iN groups ( Fig. 6 and Supplementary Fig. 4).
The Notch pathway has previously been shown to control the specification between V2a and V2b interneurons, with Notch inhibition specifying V2a over V2b interneurons (Del Barrio et al., 2007;Peng Fig. 4. Combining CNS patterning and direct conversion to generate pure CNS neurons. (A) Flow diagram of combining CNS patterning before the direct conversion protocol. (B) Principal component analysis shows a separation of the iNs groups from the undifferentiated stem cell state on PC1 and a separation of the iNs groups based on anterior-posterior identity on PC2. (C) Gene expression of known markers indicative of the pluripotent, anterior, and posterior patterning. (D) Gene expression of known markers indicative of neuronal lineages. (E) Schematic Venn diagram showing the overlap of genes within the three iNs groups. (F) The top 10 and selected additional (bold) downregulated (blue) and upregulated (red) genes in the CNS pre-patenting groups, versus the NGN2 group. Threshold bar (white line) indicates a fold change of ± 2. (G) Representative images of DAPI, VSX2, SHOX2 and TUBB3 staining. Arrowheads point to VSX2 and SHOX2 double positive cells. Bar = 10 µm. (H) Representative images of DAPI, TUBB3 and NEUROG3 staining. Bar = 20 µm. RNAseq data were from hESCs and hESC-derived iNs, immunostaining results were represented images from iNs converted from hESCs and two iPSCs lines.  et al., 2007). In all of our iN groups, we could not detect GATA3 expression, a marker of V2b interneurons. DAPT was also included during the NGN2 conversion, to enhance neuronal conversion, which in combination with NGN2 likely helped promote the specification of V2a over V2b cell types. This suggests that the influence of patterning factors, specifically DAPT, on the neuronal subtypes would also be effective during the later NGN2 conversion phase. Indeed, simultaneous patterning and neuronal conversion from either fibroblasts or iPSCs can result in a more robust conversion and with the inclusion of dual SMAD factors and WNT antagonists enhance the specification of forebrain neurons (Ladewig et al., 2012;Nehme et al., 2018).
In conclusion, we demonstrated that sequential patterning and neuronal conversion of human PSCs is a highly successful method for manipulating the identity of the neurons with the ability to regionally specify them along the rostral caudal axis, which results in a more defined population of neurons. This approach can be used for Fig. 6. Electrophysiological activities of the three iN groups by multi-electrode array (MEA). Raster plots of spontaneous activities for 2 mins (left, x axis is time and y axis is channel ID) and 30 s (right, numbers in the corners of boxed plots are the channel IDs) of raw data from the representative electrodes on week 7. Each black line represents a spike detection. Each red line represents a single channel burst (a collection of at least 4 spikes). Each blue line represents a network burst detection (with minimum of 50% of participating electrodes across entire well and minimum 25% of simultaneous channels). The images on the right side were 30 s recording from the left (marked with black rectangles).
investigating a range of neurological disorders such as AD, MND or spinal cord injury.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.