A Targeted NKX2.1 Human Embryonic Stem Cell Reporter Line Enables Identification of Human Basal Forebrain Derivatives§


  • Disclosure of potential conflicts of interest is found at the end of this article.

  • Author contributions: A.L.G.: design, collection, assembly, analyses and interpretation of data, manuscript writing; D.A., R.P.D., S.J.M., S.M.L, C-L.S., Q.C.Y., D.A.E., E.S.N., T.H., J.B., and B.W.: collection and analyses of data; R.J.L., J.M.H., and C.W.P.: design, collection, and interpretation of data, manuscript writing; A.G.: conception and design, collection of data; A.O.T.: conception and design; S.A.A.: design, collection and interpretation of data, manuscript writing; E.G.S. and A.G.E.: design, assembly, analyses and interpretation of data, manuscript writing and financial support.

  • §

    First published online in STEM CELLSEXPRESS January 7, 2011; available online without subscription thorugh the open access option.


We have used homologous recombination in human embryonic stem cells (hESCs) to insert sequences encoding green fluorescent protein (GFP) into the NKX2.1 locus, a gene required for normal development of the basal forebrain. Generation of NKX2.1-GFP+ cells was dependent on the concentration, timing, and duration of retinoic acid treatment during differentiation. NKX2.1-GFP+ progenitors expressed genes characteristic of the basal forebrain, including SHH, DLX1, LHX6, and OLIG2. Time course analysis revealed that NKX2.1-GFP+ cells could upregulate FOXG1 expression, implying the existence of a novel pathway for the generation of telencephalic neural derivatives. Further maturation of NKX2.1-GFP+ cells gave rise to γ-aminobutyric acid-, tyrosine hydroxylase-, and somatostatin-expressing neurons as well as to platelet-derived growth factor receptor α-positive oligodendrocyte precursors. These studies highlight the diversity of cell types that can be generated from human NKX2.1+ progenitors and demonstrate the utility of NKX2.1GFP/w hESCs for investigating human forebrain development and neuronal differentiation. STEM CELLS 2011;29:462–473


NKX2.1 is a transcription factor expressed in the ventral forebrain during development. Human expression studies have revealed that NKX2.1 transcripts are localized to the hypothalamic area of the diencephalon and in the telencephalic floor, in a region that will later develop into the globus pallidus [1]. Detailed analyses in the mouse confirmed a similar localization of Nkx2.1 transcripts in the developing forebrain, marking the hypothalamic primordium and prospective ventral telencephalon, including the medial ganglionic eminence (MGE) [2]. Nkx2.1-expressing ventral MGE-derived progenitors migrate tangentially into multiple areas of the forebrain, generating predominantly γ-aminobutyric acid (GABA)-ergic interneurons [3]. The MGE also harbors Nkx2.1+ oligodendrocyte precursors (OLPs), a population marked by the expression of platelet-derived growth factor receptor α (PDGFRα) [4]. In the developing hypothalamus, Nkx2.1 has been implicated in specifying progenitors that generate tyrosine hydroxylase (TH)+ dopaminergic neurons [5].

There is a complex relationship between the expression of Nkx2.1 in the mouse and the ventralizing morphogen sonic hedgehog (Shh). Shh is required for initial patterning of the Nkx2.1 domain within the basal telencephalon and hypothalamus [2, 6] and for the maintenance of Nkx2.1 expression in progenitors during neurogenesis [7]. Conversely, in the absence of Nkx2.1, Shh expression is lost from the basal telencephalon and the hypothalamus [2]. Expression of both Shh and Nkx2.1 within the developing forebrain neuroepithelium appears to be regulated by retinoic acid (RA) [8]. The importance of RA in ventral forebrain specification is underlined by experiments showing that expression of Shh and Nkx2.1 in the basal telencephalon and hypothalamus is reduced in vitamin-A-deficient (VAD) quail embryos [9, 10]. These studies and others reinforce the notion that RA is critical factor for establishing cellular identity within the developing forebrain [8, 10–12].

Studies in the chick indicate that Nkx2.1 first marks the emerging diencephalic-derived Foxg1 ventral hypothalamus [13]. An additional domain of Nkx2.1 expression subsequently develops anteriorly in the Foxg1+ telencephalon. Therefore, within Nkx2.1+ regions of the forebrain, Foxg1 expression demarcates the prospective hypothalamus and the newly formed basal telencephalon. These in vivo studies suggest a defined developmental progression in which Foxg1Nkx2.1+ and Foxg1+Nkx2.1+ populations are formed separately.

Experiments addressing these aspects of human forebrain development have been restricted to in vitro studies using human embryonic stem cell (hESC) differentiation models [14–16]. Similar to studies using mouse ESCs [17, 18], differentiation protocols use signaling pathway inhibitors that bias hESCs toward a neural fate that includes telencephalic (FOXG1+) multipotent progenitors [14, 19–21]. These committed neural precursors are subsequently ventralized to NKX2.1+ telencephalon-like progenitors using Shh [15–18]. However, without the capacity to isolate either NKX2.1+ or FOXG1+ live cells, it is not possible to unequivocally determine the precursor-progeny relationships between populations expressing combinations of these markers.

We inserted sequences encoding green fluorescent protein (GFP) into the NKX2.1 locus of hESCs using homologous recombination, enabling the prospective isolation of viable human NKX2.1+ basal forebrain progenitors by virtue of their expression of GFP (hereafter denoted NKX2.1-GFP+ cells). In vitro and in vivo studies demonstrated that NKX2.1-GFP+ progenitors had the potential to differentiate into neuronal cell types expressing GABA, TH, and somatostatin (SST), as well as PDGFRα+ OLPs. These studies illustrate the utility of the genetically tagged NKX2.1GFP/w hESC line for the ongoing investigation of human forebrain development and lineage commitment.


Generation and Identification of Targeted NKX2.1GFP/w hESCs

Construction of the targeting vector is described in Supporting Information Materials and Methods. The targeting vector was electroporated into HES3 hESCs as described [22] and one correctly targeted clone was identified (1/128). NeoR cassette removal and single cell cloning was performed as described [23]. Karyotype analysis and teratoma assays were carried out as previously described [24].

ESC Culture and Differentiation

HESC lines (HES3, MEL1, and H9) and the induced pluripotent stem cell (iPS) line DF19-9-7T [25] were passaged as described [26] and differentiated as spin embryoid bodies (EBs) in a serum-free medium, bovine serum albumin essential lipids (BEL) [27, 28] supplemented with 10–100 ng/ml fibroblast growth factor 2 (FGF2) (Peprotech Asia, Rehovot, Israel, www.peprotech.com) and 10−8 to 10−4 M all-trans RA (Sigma-Aldrich Australia, www.sigmaaldrich.com) as indicated. For reaggregation/reculture experiments, cells obtained from flow cytometric sorting were aggregated using the spin EB protocol (3–5 × 103 per well) in BEL supplemented with 10 μM ROCK inhibitor Y27632 (Sigma) [16]. Reaggregated cells were cultured further on laminin-coated flasks in BEL supplemented with 20 ng/ml FGF2 (PeproTech) and 20 ng/ml EGF (PeproTech). For coculture experiments, NKX2.1-GFP+ cells (2 × 103 cells per well) were added to primary cortical feeder cultures (1 × 105 cells per 36 mm2 well of a 16-well chamber slide, Lab-Tek, Nunc A/S, Roskilde, Denmark, www.nuncbrand.com) established from the dissociated cortices of neonatal pups as described [7].

Flow Cytometry

EBs were dissociated using TrypLE Select (Invitrogen Australia Pty Limited, Mulgrave, Victoria, www.invitrogen.com), and intracellular flow cytometry was performed as previously described [23]. Mouse anti-human primary antibodies reacting to cell surface antigens are shown in Supporting Information Table S1. Unconjugated antibodies were detected with allophycocyanin (APC)-conjugated goat anti-mouse IgG (BD Biosciences Australia, www.bdbiosciences.com).

Cortical Transplantation of NKX2.1-GFP+ Cells

Details of the transplantation protocol are described in Supporting Information Materials and Methods. In brief, donor cells were injected into neonatal mice into each of four sites placed 1 mm lateral to the midline, and either 1 mm caudal to bregma or 1 mm rostral to the interaural line, targeting layers III, IV, and V. Only GFP+ cells located in cortical layers I–VI were included in the counts.

Immunofluorescence and Immunohistochemistry

To assess maturation of reaggregated sorted cells, EBs were transferred to laminin-coated glass slides and stained as described [29]. Immunostaining of cortical feeder layer experiments was performed as described [7]. Primary antibodies (Supporting Information Table S1) were detected with Alexa Fluorophores (Molecular probes, Invitrogen Australia Pty Limited, Mulgrave, Australia, www.invitrogen.com). Images for extended differentiations were captured using a Leica SP5 confocal microscope running LAS AF software (Leica Microsystems Pty Ltd, North Ryde, Australia, www.leica-microsystems.com). Fluorescent and bright-field images for in vivo and coculture experiments were collected on a Nikon E800 microscope equipped with a Cool SNAP HQ (Photometrics, Tucson, AZ, www.photometrics.com) camera and MetaMorph software (Universal Imaging, Molecular Devices Inc, Sunnyvale, CA, www.moleculardevices.com).

Gene Expression Analysis

RNA was prepared using RNAeasy according to manufacturer's instructions (Qiagen). RNA samples were reverse transcribed using SuperScript III (Invitrogen). RNA samples were labeled, amplified, and hybridized to Human WG-6v2.0 Sentrix BeadChip and HumanHT12v3 and HumanHT12v4 Expression BeadChips (Illumina Australia, Scoresby, Australia, www. illumina.com) according to Illumina protocols. Data was analyzed using Beadstudio Gene Expression Module v3.4 (Illumina) using average normalization across all samples with further analysis performed using GeneSpring GX11 (Agilent Technologies Australia, Forest Hills, Australia, www.chem.agilent.com). Quantitative polymerase chain reaction (Q-PCR) was performed using Taqman gene expression probes and relative gene expression levels calculated as described [30]. Data is available at ArrayExpress (http://www.ebi.ac.uk/microarray-as/ae/) via accession number E-MEXP-2807.

Electrophysiology and Neuropharmacology

Experimental details are described in Supporting Information Materials and Methods. Standard whole cell voltage clamp techniques were as previously described [31]. For neuropharmacology, on the day of experimentation BEL was replaced with media containing a calcium-sensitive fluorophore (Fluo-4 AM 5 μM, Sigma, USA) and analysis performed as previously described [32].


Generation of an NKX2.1GFP/w hESC Line

A linearized NKX2.1-GFP targeting vector (Fig. 1A) was electroporated into the hESC line, HES3 [33], and a colony in which the targeting vector had correctly recombined with the NKX2.1 locus was identified. The G418 resistance cassette was excised and the cell line cloned. Homologous recombination was verified by sequencing PCR products spanning the junctions between the genomic DNA and the integrated targeting vector (Fig. 1B) and Southern blot analysis indicated that the NKX2.1GFP/w hESCs harbored a single copy of the GFP gene (Fig. 1C). Although the targeting of GFP rendered NKX2.1GFP/w hESCs heterozygous for NKX2.1, this strategy was chosen to ensure robust and faithful expression of the reporter. Indeed, benign hereditary chorea, the nonprogressive movement disorder associated with NKX2.1 heterozygosity in humans is only evident postnatally and is usually associated with normal magnetic resonance imaging [34], indicating that early NKX2.1-dependent neurogenesis is not significantly perturbed.

Figure 1.

Generation and characterization of NKX2.1GFP/w human embryonic stem cells (hESCs). (A):NKX2.1 targeting vector. The G418 antibiotic resistance (NeoR) cassette, flanked by loxP sites (black triangles), was removed using Cre recombinase. Polymerase chain reaction (PCR) primers (a, b) and (c, d) were used to identify the targeted clone. (B): PCR analysis using these primers generated specific bands of 4.5 kb (5′ homology arm) and 7.1 kb (3′ homology arm) from targeted NKX2.1GFP/w cells that were sequenced to confirm homologous recombination. (C): Southern blot analysis of EcoRV-digested genomic DNA showed a single GFP insert present in the targeted line prior to (NKX2.1GFP/wPC) and following single cell cloning (NKX2.1GFP/w). (D–F): Images illustrating GFP+ cells in day 24 EBs. (G): Protocol for generating NKX2.1-GFP+ cells. (H–J): Colocalization of GFP and NKX2.1 protein in NKX2.1GFP/w hESC-derived EBs. (K): Summary of data demonstrating the relationship between RA concentration (M), FGF2 concentration (ng/ml), and frequency (%) and number of GFP+ cells (96-well plate) measured at differentiation day 24–26. For RA titration experiments, the FGF2 concentration was 100 ng/ml. For FGF2 titration experiments, the RA concentration was 10−5 M. n = 3, mean ± SEM. Scale bars = 100 μm (D–F); 20 μm (H–J). Abbreviations: BF, bright field, FGF2, fibroblast growth factor 2; GFP, green fluorescent protein; GFs, growth factors; RA, retinoic acid.

NKX2.1GFP/w hESCs retained a normal karyotype, expressed markers of undifferentiated hESCs and formed teratomas containing cell types derived from the three germ layers (Supporting Information Fig. S1). Transcriptional profiles of undifferentiated NKX2.1GFP/w cells, parental HES3 cells, and an independent hESC line, MEL1, were all very similar, with ≥3-fold differences in signal intensity observed in only 0.06% of probes comparing NKX2.1GFP/w with either HES3 or to MEL1 cells and 0.04% of probes comparing HES3 and MEL1 cells (Supporting Information Fig. S2 and Supporting Information Table S2).

To demonstrate the fidelity of the GFP reporter gene, GFP+ cells generated under conditions that promoted the formation of neural progenitors (Fig. 1D–1G) were analyzed for NKX2.1 expression. This analysis showed the emergence of GFP+ structures that coexpressed NKX2.1 protein (Fig. 1H–1J). Intracellular flow cytometry confirmed that GFP+ cells were highly enriched for NKX2.1 protein and that this transcription factor was not detected in GFP fractions (Supporting Information Fig. S1B and Fig. S4D).

FGF2 and RA Promote Formation of NKX2.1+ Neural Progenitors

We employed the NKX2.1GFP/w hESC line to more carefully define conditions for the generation of NKX2.1-GFP+ cells from differentiating hESCs. Recent studies have suggested that FGF2 promoted the allocation of differentiating hESCs to a neurectodermal fate [35] and other works have shown that RA was required for expression of NKX2.1 in the ventral forebrain [8, 10]. As our serum-free BEL medium lacked RA or its precursor (retinol), unlike media formulations containing B27 supplement or serum, we investigated protocols in which EBs were differentiated in combinations of FGF2 and a pulse of RA.

The concentrations of FGF2 and RA influenced both the frequency and number of GFP+ cells observed at day 24. The highest percentage (∼12%–14%) of GFP+ cells was observed in cultures treated with 100 ng/ml FGF2 for 7 days followed by 10−5 M RA from days 7 to 10 (Fig. 1K and Supporting Information Fig. S1K and S1L). Q-PCR analysis revealed that, at lower RA concentrations, differentiation defaulted toward anterior and dorsal neural cell fates marked by expression of genes such as FOXG1 and PAX6 (Supporting Information Fig. S1M–S1P). The concentration of FGF2 had little effect on the frequency of GFP+ cells, but it did influence the overall yield (Fig. 1K). Examination of EBs revealed that GFP+ cells were associated with several morphological structures; tight-bordered epithelial-like masses (Fig. 1D–1F), spreading mesenchymal-like cells, and spheroids within cystic structures (Supporting Information Fig. S1E–S1J).

These studies were complemented by experiments in which NKX2.1GFP/w cells differentiated for 7 days in 100 ng/ml FGF2 and then pulsed with10−5M RA for 24 hours were transplanted under the kidney capsule of immunodeficient mice. The RA-treated EBs generated grafts with multiple GFP+ regions as well as areas of pigmented epithelium and NESTIN+ neural rosettes, suggesting an ectodermal bias to the differentiation (Supporting Information Fig. S3). NKX2.1 immunostaining localized to neuroepithelial structures, suggesting a primitive degree of organization. Very few GFP+ regions were observed when EBs cultured in FGF2 in the absence of a pulse of RA were transplanted.

NKX2.1-GFP+ Cells Emerge from an E-CAD+ Population

We investigated the kinetics of NKX2.1GFP/w hESC differentiation by monitoring the expression of GFP over a 24-day time course (Fig. 2A). Given the appearance of GFP within epithelial structures, we also assessed the expression of E-CAD, a cell surface protein associated with epithelial subdomains within the developing embryonic mouse brain (Fig. 2A) [36]. This analysis indicated that cells within day 7 EBs retained robust levels of E-CAD expression but contained no NKX2.1-GFP+ cells (Fig. 2A). At this stage, immunofluorescence studies identified numerous NESTIN+ cells arranged in neural rosette-like structures, confirming the neural commitment of the cells prior to RA treatment (Fig. 2C, 2D). In spin EBs treated with FGF2 alone, the frequency of E-CAD+ cells decreased to ∼30% by day 14 (Fig. 2A, 2B). However, inclusion of RA from days 7 to 10 maintained E-CAD+ expression on a larger fraction of cells (∼85%–90%) that persisted even after RA withdrawal (Fig. 2A, 2B). NKX2.1-GFP+ cells emerged from the E-CAD+ population in RA-treated cultures at day 14 (Fig. 2A, 2B). We also performed differentiation experiments with two additional hESC lines (MEL1 and H9) and an iPS line (Supporting Information Fig. S4). The hESC lines formed NKX2.1+ cells with similar kinetics using the FGF2/RA protocol, but the iPS line did not differentiate as efficiently. Although this could reflect the need for culture optimization, our findings also support recent data that neural differentiation from a range of iPSs was variable and of lower efficiency compared with neural differentiation from hESCs [37].

Figure 2.

Time course analysis of differentiating NKX2.1GFP/w human embryonic stem cells (hESCs). (A): Flow cytometric analysis demonstrating the requirement for RA for the appearance of GFP+ cells. (B): Summary of flow cytometry experiments showing the proportion of cells (%) expressing E-CAD and GFP in cultures differentiated with (green lines) and without (red lines) RA from day 7 to 10. n = 3, mean ± SEM. (C, D): Immunofluorescence analysis of day 7 EBs demonstrating the presence of NESTIN+ cells and neural rosettes (white arrowhead). (E): Microarray analysis comparing gene expression profiles of day 0 hESCs, day 7 EBs, and day 10 EBs with or without RA treatment. The color of each horizontal stripe represents the signal intensity of genes whose expression varied ≥5.0-fold between the day 0 and day 10 RA-treated samples. (F): Bar graph showing selected genes whose expression was upregulated in RA-treated versus FGF2-treated day 10 EBs. (G): Table showing the fold change in the expression of anterior neural markers from day 7 EBs to day 10 with or without RA. (H): Quantitative polymerase chain reaction (Q-PCR) analysis of the indicated genes over a 24-day time course in which EBs were treated with (green lines) and without (red lines) RA from day 7 to 10. n = 3, mean ± SEM. (I): Q-PCR analysis of NKX2.1 and SHH expression in GFP+ and GFP cells sorted at day 14. Scale bars = 30 μm (C); 15 μm (D). Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; E-CAD, E-cadherin; GFP, green fluorescent protein; NC, no change; RA, retinoic acid.

We compared microarray gene expression data derived from days 7 and 10 EBs differentiated with or without RA treatment (Fig. 2E–2G and Supporting Information Table S3). This analysis demonstrated that days 7 and 10 EBs formed in the presence of FGF2 alone shared similar transcriptional profiles (Fig. 2E and Supporting Information Table S3). Relative to undifferentiated hESCs, day 7 EBs had upregulated genes associated with early pan-neural epithelium, including LMO1, OTX2, and HESX1 (Fig. 2E and Supporting Information Table S3) [21]. By day 10, EBs differentiated in FGF2 alone expressed genes such as FGF8, OTX2, LHX2, and SIX6, consistent with prior studies, suggesting that neural differentiation was biased toward cell types with anterior character (Fig. 2E, 2G and Supporting Information Table S3) [21]. Conversely, the RA-treated EBs had downregulated these broadly expressed anterior markers and upregulated genes associated with neural regionalization and patterning, including PAX6 and DLX5 (Fig. 2F) [21, 38]. As expected, RA also induced expression of HOX cluster genes that are direct targets of RA signaling in neural development (Fig. 2F and Supporting Information Table S3) [39].

These studies were complemented by Q-PCR analysis of differentiated NKX2.1GFP/w, MEL1, H9, and iPSs (Fig. 2H and Supporting Information Fig. S4). Expression of the stem cell marker OCT4 decreased during the first 7 days of differentiation, contrasting with an increase in expression of the anterior marker, HESX1. Consistent with our microarray data, HESX1 expression declined from day 7 onward, accelerated by the addition of RA. Interestingly, the elevated expression of PAX6 within RA-treated day 10 EBs identified by microarray represented a transient peak that abated following RA removal. In contrast, PAX6 expression gradually increased in EBs differentiated in FGF2 alone (Fig. 2H and Supporting Information Fig. S4A).

The results of Q-PCR analysis also substantiated microarray data indicating that the WNT inhibitor, secreted frizzled-related protein 5 (SFRP5), was transiently upregulated in RA-treated cultures at day 10 in hESC lines but not in the iPS line (Fig. 2H and Supporting Information Fig. S4). Like PAX6, expression of this gene in differentiating hESCs decreased after RA removal. The transient expression of PAX6 and SFRP5 in the hESC lines preceded the upregulation of SHH, whose expression is required for the generation and maintenance of Nkx2.1+ progenitors in the basal forebrain [40]. Indeed, cell-sorting experiments at day 14 indicated that SHH expression was restricted to NKX2.1-GFP+ cells (Fig. 2I), echoing the overlapping expression patterns of these genes during early mouse and chick forebrain development [2, 13]. These observations are also consistent with recent data indicating that PAX6 is initially uniformly expressed in early human neurectoderm prior to expression of region-specific genes including NKX2.1, in contrast with the more restricted expression domain of Pax6 in the mouse [41].

The telencephalic marker FOXG1 was not expressed in RA-treated cultures, consistent with the ventralizing effects of RA and the initially reciprocal expression domains of dorsally expressed Foxg1 and the earliest expression of Nkx2.1 in the hypothalamus [13]. In contrast, FOXG1 expression progressively increased in cultures treated with FGF2 alone (Fig. 2H and Supporting Information Fig. S4A).

NKX2.1-GFP+ Cells Express Genes Associated with Basal Forebrain Progenitors

Immunofluorescence analysis of cultures between days 24 and 30 demonstrated that a substantial proportion of GFP+ cells expressed the pan-neuronal markers, β-TUBULIN III and MAP2a,b (Fig. 3A–3D). Consistent with the gene expression data, PAX6 and NKX2.1-GFP were expressed in complementary cell populations (Figs. 2H, 3E, 3J).

Figure 3.

Gene expression analysis of NKX2.1-GFP+ EBs. (A–E): Images of cells derived from day 24 to 30 differentiated NKX2.1GFP/w EBs showing colocalization of the neuronal markers β-TUBULIN III (A, B) and MAP2a,b ([C, D] arrowheads) with GFP and the mutually exclusive expression of PAX6 and NKX2.1-GFP. (F): Flow cytometric profile showing reanalysis of cells sorted on the basis of GFP and E-CAD expression. (G): Microarray analysis comparing gene expression profiles of cell fractions sorted on the basis of GFP and E-CAD expression. The color of each horizontal stripe represents the signal intensity of genes whose expression varied ≥5.0-fold between the day 0 and day 23 E-CADGFP+ fractions. (H): Microarray comparison of gene expression in E-CADGFP and E-CADGFP+ cell fractions. (I): Microarray comparison of gene expression in E-CADGFP+ and E-CAD+GFP+ cell fractions. The fold change in expression between the indicated fractions is listed for key genes. (J): Q-PCR analysis of genes associated with basal forebrain progenitors. n = 3, mean ± SEM. Scale bars = 20 μm (A, B, E); 10 μm (C, D). Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; E-CAD, E-cadherin; GFP, green fluorescent protein; Un, unsorted cells.

Gene expression analysis was performed on days 23 to 25 cells purified on the basis of GFP and E-CAD expression (Fig. 3F, 3G and Supporting Information Table S4). Comparison of the E-CADGFP and the E-CADGFP+ fractions indicated that the former population had upregulated genes (204) associated with dorsal and/or posterior neurogenesis including ZIC2, PAX6, and NKX6.2 (Fig. 3H and Supporting Information Table S4). In contrast, the E-CADGFP+ fraction upregulated genes involved in ventral forebrain neuronal specification such as LHX6 and vesicular inhibitory amino acid transporter (VIAAT; Fig. 3H, Supporting Information Fig. S5 and Table S4). Comparison of the GFP+ fractions identified a group of 253 genes preferentially expressed in the E-CAD+GFP+ population, including CDH3, CDH12, and RARRES3, consistent with our hypothesis that this population contained immature neuroepithelial precursors (Fig. 3I and Supporting Information Table S4). Conversely, transcripts upregulated in the E-CADGFP+ population (98) included genes reflecting ongoing neuronal maturation such as STMN2, DLX1, and glutamic acid decarboxylase (GAD1; Fig. 3I and Supporting Information Table S4).

Similar to Nkx2.1, the basic helix loop transcription factor Ascl1 is expressed in the mouse ventral telencephalon in neuronal and OLPs, although its domain extends more broadly [42, 43]. We observed that ASCL1 expression paralleled NKX2.1 expression during differentiation, and that this was preceded by a rise in SHH expression (Supporting Information Fig. S4E). At day 24, ASCL1 was highly expressed in more mature E-CADGFP+ cells. We also observed elevated levels of ASCL1 in E-CADGFP cells, suggesting that this population included neuronal cells and or OLPs (Fig. 3H, 3J and Supporting Information Table S4).

The results of microarray analysis were supported by experiments in which the expression of key transcripts in sorted cell fractions were examined by Q-PCR (Fig. 3J and Supporting Information Fig. S5A). Expression of PAX6 in E-CADGFP cells suggested that this population contained dorsal neural derivatives (Fig. 3H, 3J). Mirroring the results of gene expression studies on GFP+ cells from day 14 EBs (Fig. 2I), when most GFP+ cells retained E-CAD expression (Fig. 2A), SHH expression remained highest in the less mature E-CAD+GFP+ population at day 24 (Fig. 3J). Loss of E-CAD expression was associated with downregulation of SHH. Expression of PDGFRα was limited to the E-CADGFP population, suggesting that at this stage of differentiation, the GFP+ population did not contain committed OLPs (Fig. 3H, 3J) [4].

NKX2.2 and RAX, both of which are associated with developing Nkx2.1+ hypothalamic cells [5, 44], were expressed in the E-CADGFP+ fraction, consistent with the possibility that this population contained cells analogous to those found in the ventral domains of the diencephalon. Expression of OLIG2, DLX1, and DLX5, which mark the emerging ventral telencephalon [45] and the developing hypothalamus [46], was also highest in the E-CADGFP+ fraction (Fig. 3J). Furthermore, this population also showed the highest expression of LHX6, GAD1, and the VIAAT (Fig. 3J and Supporting Information Fig. S5A). These results support our hypothesis that this population contained the most differentiated NKX2.1+ cells.

In the mouse, Foxg1 expression is often used to distinguish ventral telencephalic Nkx2.1+ cells from their hypothalamic counterparts [40]. The NKX2.1-GFP+ cells generated at this stage of differentiation appeared to be predominantly diencephalic as FOXG1 was not expressed within the GFP+ population based on Q-PCR (Fig. 2H and Supporting Information Fig. S4) and immunofluorescence analyses of GFP+ cells 48 hours postsorting at day 24 (Supporting Information Fig. S5).

Differentiation and Functional Examination of NKX2.1-GFP+ Progenitors

The predilection for neural differentiation of NKX2.1-GFP+ cells was further demonstrated by transplanting flow cytometry sorted, reaggregated cells (Fig. 4A–4C) under the kidney capsule of immunodeficient mice. Examination of grafts after 6 weeks, indicated that the tissue was composed of disorganized β-TUBULIN III+GFP+ neural elements, confirming the neural precursor nature of the NKX2.1+ cells (Supporting Information Fig. S6).

Figure 4.

Differentiation of NKX2.1-GFP+ neural progenitors. (A): Experimental protocol used to assess the differentiation potential of isolated NKX2.1-GFP+ cells. BF (B) and epifluorescence (C) images of reaggregated NKX2.1-GFP+ cells. (D–H): Immunofluorescence of NKX2.1-GFP+ cells cultured on mouse cortical feeder layers. (D, E): NKX2.1-GFP+ cell aggregates and individual cells coexpressing human N-CAM and GABA. (F–H): NKX2.1-GFP+ cells also expressed TH. (I–L): Neural processes emerge from NKX2.1-GFP+ cells transplanted into the cortices of neonatal mice (I). Although most cells remain at the transplantation site, some individual cells migrate further into the brain parenchyma. Single hNCAM+GFP+ cells displayed a neuronal morphology (J) and expressed GABA (K) or TH (L). (M, N): Bright-field (M) and epifluorescence (N) images of an NKX2.1-GFP+ neuron selected for patch clamp recording. (O): Depolarization of NKX2.1-GFP+ neurons evoked both an INa and outward potassium current (i). The potassium current was blocked by caesium (CsSO4) (ii), and the remaining INa blocked by 300 nM TTX (iii). Action potentials were observed in GFP+ cells spontaneously or on depolarizing current injection (iv). Plot of mean current–voltage relationship of INa and IK (v). n = 5–6, mean ± SEM. (P): Elevation in intracellular calcium (Ca2+) for NKX2.1-GFP+ cells in response to both glutamate (black, n = 6) and potassium chloride (KCl; blue, n = 3). Mean ± SEM. (Q, R): Migration of sorted NKX2.1-GFP+ populations, with apparent “tracking” of GFP+ ([R], asterisks) and GFP ([R], arrowheads) cell bodies along fibers. Scale bars = 100 μm (C, D, F, I, K, N, Q); 50 μm (E, L–N, R); 25 μm (G, H, J). Abbreviations: BF, bright field; DAPI, 4′,6-diamidino-2-phenylindole; GABA, γ-aminobutyric acid; GFP, green fluorescent protein; hNCAM, human neural cell adhesion molecule; IK, outward potassium currents; INa, inward sodium currents; TH, tyrosine hydroxylase; TTX, tetrodotoxin.

Next, NKX2.1-GFP+ cells at day 24 were cocultured with cells prepared from dissociated cerebral cortices of neonatal mice [7]. After 11–14 days of coculture, NKX2.1-GFP+ cells had differentiated to GABA+ (31%, 18GABA+/59GFP+) and TH+ (17%, 50TH+/294GFP+) neurons (Fig. 4D–4H). Using an anti-human NCAM antibody, we identified infrequent human NCAM+GFP cells, demonstrating that only a minority of NKX2.1-GFP+ cells differentiating on mouse cortical feeder layers lost GFP fluorescence. We observed TH+ cells (8.5%, 11 TH+/129 NCAM+GFP), suggesting that the cells had differentiated to a diencephalic fate and a small number of GABA+ cells (two GABA+/6 NCAM+GFP) consistent with potential differentiation to telencephalic derivatives.

These data were complemented by neonatal mouse cortical transplantation studies demonstrating that 25% (31GABA+/126GFP+) of GFP+ cells colabeled with GABA, whereas 6% (5TH+/82GFP+) colabeled with TH, 19–27 days post-transplantation (Fig. 4I–4L and Supporting Information Fig. S7A–S7C). Human GFP cells were rare, possibly reflecting the lack of appropriate cues or the requirement for a longer period of differentiation as noted in our serum-free in vitro cultures. Although most GFP+ cells were found close to the transplantation site, cells with both leading and trailing processes could be seen 5 days after transplantation, with more mature cells adopting a multipolar, aspiny phenotype (Fig. 4I–4L).

Patch clamp electrophysiology was performed on NKX2.1-GFP+ progenitors differentiated for a further 6–8 weeks in vitro (Fig. 4M–4O). GFP+ cells displayed both inward sodium (INa) and outward potassium (IK) currents at potentials positive to −60 mV (Fig. 4O, i). In potassium gluconate (K-gluconate)-filled cells, INa increased in amplitude until a peak was reached near −20 mV and reduced during steps to potentials more positive. In contrast, the steady state current of IK progressively increased in amplitude, showing a marked outward rectification at positive potentials (Fig. 4O, i). The true time course of the INa was recorded by replacing intracellular potassium with caesium (CsSO4), therefore blocking all current flow through potassium channels (Fig. 4O, ii). In CsSO4-filled cells, INa was rapidly blocked by the neuronal sodium channel blocker tetrodotoxin (Fig. 4O, iii). Under current clamp, Kgluconate-filled GFP+ neurons that displayed a net inward current (INa) readily discharged action potentials either spontaneously or on electrical current injection (Fig. 4O, iv).

The neuropharmacological potential of differentiated NKX2.1-GFP+ neurons was investigated in cells loaded with 5 μM Fluo-4AM, a calcium (Ca2+) sensitive fluorophore (Fig. 4P). Exposure to the amino acid transmitter, glutamate, elevated intracellular Ca2+ in 87% ± 8% of cells (n = 6 replicates of 9 cells), suggesting the presence of functional glutamate receptors (Fig. 4P). Addition of potassium chloride to the culture also resulted in an elevation of intracellular Ca2+ (87% ± 7% of cells, n = 3 replicates of 9 cells), presumably through a depolarization-induced opening of voltage-operated channels (Fig. 4P). These findings, along with the electrophysiology data, indicate that NKX2.1-GFP+ cells were capable of generating neurons with functional electrochemical and neurochemical characteristics.

A notable characteristic of Nkx2.1+ basal forebrain progenitors is their propensity to migrate tangentially from their site of origin [47]. NKX2.1-GFP+ cells displayed extracellular matrix-dependent migratory behavior (Fig. 4Q, 4R and Supporting Information Fig. S7D–S7L) and migrating cells often formed a mesh-like arrangement, had neural processes and, in some instances, cell bodies appeared to track along fibers (Fig. 4R and Supporting Information Video S1). Migration was often preceded by neurite extension and accompanied by nuclear translocation, with individual cell bodies observed at distances up to 1.0 mm from the central cell aggregate (Supporting Information Fig. S7D and Video S2). The majority of migrating cells retained NKX2.1-GFP expression, although some GFP cell bodies were observed (Fig. 4R).

Maturation of NKX2.1-GFP+ Progenitors

The emergence of both GABA+ and TH+ neurons suggested that the NKX2.1-GFP+ cells were a mixed population that formed a spectrum of basal forebrain derivatives. When sorted NKX2.1-GFP+ fractions were reaggregated and cultured on laminin-coated coverslips (Fig. 5A), GFP+ cells migrated from the central aggregate and displayed neuronal differentiation (Fig. 5B–5D). In some instances, this was accompanied by a loss or reduction in GFP fluorescence. In the case of the E-CAD+GFP+ cell aggregates, migration away from the central core coincided with a loss of E-CAD expression (Fig. 5B). Q-PCR analysis of cultures 14 days after sorting showed a decrease in NKX2.1 expression and increased expression of genes indicative of ongoing differentiation, including DLX1, DLX5, LHX6, and GAD1 (Fig. 5E).

Figure 5.

In vitro differentiation of NKX2.1-GFP+ neural progenitors. (A): Differentiation analysis of isolated NKX2.1-GFP+ cells. NKX2.1-GFP+ cells were flow sorted, reaggregated, and cultured on laminin. (B): In differentiating E-CAD+GFP+ spheroids, E-CAD expression was lost by day 14 migrating cells but retained in the central aggregate (arrowhead). (C, D): Bright-field and epifluorescence images of differentiated GFP+ spheroids after 14 days. (E): Q-PCR analysis of recultured E-CADGFP+ spheroids. n = 3, mean ± SEM. (F, G): Histogram and flow cytometry profile (day 28) showing progressive reduction in GFP expression with culture of day 24 purified NKX2.1-GFP+ cells. (H–J): GFP+ and GFP cells derived from ECADGFP+ fraction expressed β-TUBULIN III and ISL1. (K): Q-PCR analysis of E-CAD+GFP+ cells and their derivatives revealed upregulation of FOXG1. (L–O): Upregulation of FOXG1 in GFP+ (arrowheads in [O]) and GFP cells derived from FOXG1GFP+ precursors. (P–R): Immunofluorescence images showing surface expression of FORSE-1. (S): Flow cytometry revealing FORSE-1+ cells arising from a day 24 FORSE-1GFP+ population. (T–Y): Images showing the generation of TH+(T, U), GAD65+(V, W), and SST+(X, Y) neurons from GFP+ cells. (Z): Flow cytometric analysis demonstrating PDGFRα+ cells in the FORSE-1GFP fraction. Microarray analysis indicating that PDGFRα+ cells expressed genes associated with forebrain oligodendrocyte precursors. (A′, B′): Immunofluorescence images showing expression of PDGFRα on both GFP+ and GFP cells. Scale bars = 100 μm in (D); 50 μm in (L, M); 30 μm in (B, H); 25 μm (I, J, N, T, V, X, A′); 10 μm (O, P, Q, R, U, W, Y); 5 μm in (B′). (R), (V), and (A′) are maximum projections of z-stack images. Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; E-CAD, E-cadherin; FACS, fluorescence-activated cell sorting; GAD, glutamic acid decarboxylase; GFP, green fluorescent protein; IF, immunofluorescence; ISL1, islet 1; PDGFRα; platelet-derived growth factor receptor α; RGE, relative gene expression; TH, tyrosine hydroxylase; Q-PCR, quantitative polymerase chain reaction.

After ∼50 days in culture 30%–40% of E-CADGFP+ cells had lost GFP expression (Fig. 5F, 5G). Most cells expressed the pan-neuronal marker β-TUBULIN III and up to 30% expressed the ventral neuronal marker ISLET1 (Fig. 5H–5J) [15]. Interestingly, Q-PCR and immunofluorescence analysis showed that FOXG1 expression was induced in both GFP+ and GFP progeny, indicating that the initial NKX2.1-GFP+FOXG1 population harbored precursors of telencephalic FOXG1+ cells (Fig. 5K–5O and Supporting Information Fig. S8). Coincident with the upregulation of FOXG1, we also observed expression of the forebrain-associated cell surface marker, FORSE-1 (Fig. 5P–5R and Supporting Information Fig. S8I–S8P). Indeed, cell sorting experiments demonstrated that ∼7% of day 24 FORSE-1GFP+ cells upregulated FORSE-1 expression following 7 days in culture (Fig. 5S).

Similar to the observations with mouse cortical cocultures and in vivo transplantation experiments, TH and GAD65 were expressed in GFP+ and GFP progeny of a proportion of long-term cultured NKX2.1-GFP+ progenitors (Fig. 5T–5W). Furthermore, a small number of GFP+ and GFP cells expressed the neuropeptide, SST (Fig. 5X, 5Y).

In NKX2.1-GFP+ cells cultured for 90–120 days, the proportion of GFP+ cells stabilized at ∼50% (Supporting Information Fig. S8Q). Gene expression studies of subfractions sorted on the basis of FORSE-1 and GFP suggested both ongoing neuronal maturation and lineage diversification. For example, flow cytometric analysis of the FORSE-1GFP population demonstrated that ∼22% of these cells were PDGFRα+ (Fig. 5Z–5B′); a result corroborated by Q-PCR analysis (Supporting Information Fig. S8R). Microarray analysis comparing FORSE-1GFPPDGRFα+ and FORSE-1GFPPDGFRα cells indicated that the former expressed genes associated with OLPs, including MYT1, OLIG2, OLIG1, and SOX10 (Fig. 5Z and Supporting Information Table S5) [4].


We targeted a GFP reporter gene to one allele of NKX2.1 in hESCs, to facilitate investigation of early steps in human forebrain specification. The reporter gene allowed the emergence of NKX2.1+ cells to be monitored in real time and enabled the isolation of viable NKX2.1+ cells at different stages of differentiation, providing a starting point for fate mapping studies and gene profiling experiments. In our differentiation protocol, hESCs differentiated in FGF2 adopted a neural fate, evidenced by the presence of neural rosettes in day 7 EBs (Fig. 6). This outcome was consistent with prior studies showing neural differentiation of ESCs cultured in medium lacking mesendoderm inducing factors [14, 15, 35, 48–50].

Figure 6.

RA-mediated induction of human basal forebrain neurons in differentiating hESCs. (A): Differentiation of hESCs in FGF2 allows cells to adopt a neurectodermal fate by day 7, marked by the expression of NESTIN and presence of neural rosettes. Following RA patterning from day 7 to 10, NKX2.1-GFP+ progenitors emerge at day 14 coexpressing E-CAD and SHH. These progenitors resemble a pool of early neuroepithelial NKX2.1+ cells found in the basal forebrain. As the NKX2.1-GFP+ progenitors mature, E-CAD and SHH expression are reduced, whereas genes representing ongoing forebrain differentiation are upregulated. (B): NKX2.1-GFP+ cells at day 24 were FOXG1 and predominantly FORSE-1. Extended differentiation of NKX2.1-GFP+ progenitors results in upregulation of the telencephalic-specific FOXG1 transcription factor and the emergence of GABA+, tyrosine hydroxylase-positive, and somatostatin-positive neurons as well as PDGFRα+ oligodendrocyte precursors. Abbreviations: FGF, fibroblast growth factor 2; GABA, γ-aminobutyric acid; GFP, green fluorescent protein; hESCs, human embryonic stem cells; PDGFRα, platelet-derived growth factor receptor α; RA, retinoic acid.

We found that a specific concentration of RA administered for a defined period during hESC differentiation was required for the induction of NKX2.1. The role of RA in mouse forebrain development has been obscured by redundancy within the RA signaling network and by the early embryonic lethality of embryos harboring multiple defects in the RA signaling axis [9]. However, analysis of E10.5 embryos harboring an RA responsive LacZ transgene revealed the presence of RA signaling activity throughout the entire telencephalon, except the most dorsomedial area [51]. Experiments in the chick using RA signaling inhibitors [8] and in VAD quail embryos [9, 11], in which RA is absent, clearly demonstrated that RA was required for expression of Nkx2.1 and Shh in specific domains within the basal forebrain.

A role for RA during the development of NKX2.1 progenitors from in vitro-differentiated ESCs has not been previously noted. However, many ESC differentiation protocols incorporate either B27 supplement, which contains retinyl acetate, or serum, a known source of retinoids [15, 18], potentially masking the requirement for this factor. The serum-free medium used in our differentiations contained no source of RA, mirroring conditions observed in VAD quail embryos. Therefore, our results indicate that RA signaling probably plays a role in the emergence of NKX2.1+ progenitors during human ventral forebrain development.

Treatment of day 7 EBs with RA resulted in the upregulation of genes associated with forebrain regionalization, including SFRP5, a WNT signaling inhibitor expressed anterior to the optic placode [52]. Notably, this region encompasses the domain, which gives rise to Nkx2.1+ cells one day later [52]. This transient induction of SFRP5 is particularly interesting because recent work suggests that inhibition of WNT signaling is necessary for the ventralization of hESC-derived naïve neuroepithelium, prior to the emergence of NKX2.1+ neural progenitors [15].

The removal of RA was followed by upregulation of SHH (Fig. 6), a morphogen known to be a key factor in Nkx2.1-dependent basal forebrain regionalization [40, 53]. Emerging evidence suggests a role for RA in patterning the developing forebrain, specifically by modulating the expression of key secreted factors, including Shh, and studies suggest that RA and Shh signaling may synergistically promote Nkx2.1 expression in the basal forebrain [8]. In our differentiation system, SHH transcription correlated with NKX2.1 expression, paralleling the relationship between these genes in mouse and chick forebrain development (Fig. 6) [2, 5, 6, 13].

In prolonged cultures of NKX2.1-GFP+ progenitors, GFP expression waned in ∼50% of cells. During mouse embryogenesis, loss of Nkx2.1 expression is observed during the migration and maturation of basal forebrain precursors [54]. However, loss of Nkx2.1 expression is not an absolute prerequisite for neuronal migration and differentiation, as persistent Nkx2.1 expression is observed in MGE-derived striatal interneurons and hypothalamic neurons [54]. Furthermore, in contrast with the mouse, human NKX2.1 expression may be retained in some differentiated cells [55, 56]. For instance, although pallidal expression of NKX2.1 appears to be similar in human and mouse, NKX2.1 is also strongly expressed in the human fetal cortex, in the proliferative zone, cortical plate, and layer I [55–57].

Our results are consistent with the hypothesis that NKX2.1-GFP marks cell populations representing a range of forebrain regional identities, including TH+, GABA+, and SST+ neurons (Fig. 6). As both the telencephalon and diencephalon produce GABA+ and SST+ neurons, these markers alone cannot distinguish between derivatives of these forebrain regions [58]. Therefore, regional identity is often assigned based on the combinatorial expression of Foxg1 and Nkx2.1, with Foxg1+Nkx2.1+ and Foxg1Nkx2.1+ cells representing telencephalic and diencephalic progenitors, respectively [58]. In this context, our finding that FOXG1+NKX2.1+ cells can arise from a FOXG1NKX2.1+ precursor population is particularly noteworthy, contrasting with previous reports in which FOXG1+ cells gave rise to cells expressing NKX2.1 [15–18]. In those studies, ventralizing (Shh) factors were used to differentiate Nkx2.1+ cells from a Foxg1+ population. Our study shows that, in vitro, there is an alternate pathway by which the telencephalic FOXG1+NKX2.1+ cells can be generated. Whether this pathway also operates in vivo within the developing human basal forebrain remains to be determined.

NKX2.1+ progenitors also generated PDGFRα+ cells that expressed the oligodendrocyte markers OLIG2 and SOX10 [4]. These cells were most frequent in the FORSE-1GFP population, although PDGFRα expression was also observed at lower levels in all of the subfractions derived from NKX2.1-GFP+ precursors. To our knowledge, this is the first evidence suggesting that human OLPs can be generated from hESC-derived NKX2.1+ neural progenitors.


Our work demonstrates a role for RA in the generation of a specific subpopulation of human neural progenitors, marked by expression of NKX2.1. Differentiation studies show that this population has significant potential for ongoing lineage diversification, including the generation of both neuronal and glial cell types. Furthermore, the NKX2.1 reporter cell line generated in this study enabled the identification of a novel precursor-progeny relationship within the development progressions underpinning human forebrain neurogenesis. In this context, NKX2.1GFP/w hESCs represent a unique reagent for future investigations that examine the commitment of human NKX2.1+ progenitors to terminally differentiated neural subtypes.


We thank Robyn Mayberry and Amanda Bruce for the provision of hESCs, FlowCore for flow cytometric sorting, Monash Micro Imaging for confocal microscopy and time-lapse video services. This work was supported by the Australian Stem Cell Centre (ASCC), The National Health and Medical Research Council (NHMRC) of Australia, The Juvenile Diabetes Research Foundation and the Starr Foundation (D.A. and S.A.A.). A.G.E. and E.G.S. are Senior Research Fellows of the NHMRC. R.P.D. is currently affiliated with Department of Anatomy and Embryology, Leiden University Medical Centre, Leiden, The Netherlands.


The authors indicate no potential conflicts of interest.