Directing Differentiation of Human Embryonic Stem Cells Toward Anterior Neural Ectoderm Using Small Molecules§



This article is corrected by:

  1. Errata: Erratum to: Directing Differentiation of Human Embryonic Stem Cells Toward Anterior Neural Ectoderm Using Small Molecules Volume 31, Issue 7, 1446, Article first published online: 6 July 2013

  • Author contributions: B.S.: conception and design, collection and assembly of data, data analysis and interpretation, and manuscript writing; H.F.: collection and assembly of data and data analysis and interpretation; A.G.: data analysis and interpretation and manuscript writing; S.L.: conception and design and data analysis and interpretation; P.W.: conception and design, collection and assembly of data, data analysis and interpretation, manuscript writing, and final approval of manuscript.

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

  • §

    First published online in STEM CELLSEXPRESS July 3, 2012.


Based on knowledge of early embryo development, where anterior neural ectoderm (ANE) development is regulated by native inhibitors of bone morphogenic protein (BMP) and Nodal/Activin signaling, most published protocols of human embryonic stem cell differentiation to ANE have demonstrated a crucial role for Smad signaling in neural induction. The drawbacks of such protocols include the use of an embryoid body culture step and use of polypeptide secreted factors that are both expensive and, when considering clinical applications, have significant challenges in terms of good manufacturing practices compliancy. The use of small molecules to direct differentiation of pluripotent stem cells toward a specified lineage represents a powerful approach to generate specific cell types for further understanding of biological function, for understanding disease processes, for use in drug discovery, and finally for use in regenerative medicine. We therefore aimed to find controlled and reproducible animal-component-free differentiation conditions that would use only small molecules. Here, we demonstrate that pluripotent stem cells can be reproducibly and efficiently differentiated to PAX6+ (a marker of neuroectoderm) and OCT4 (a marker of pluripotent stem cells) cells with the use of potent small inhibitors of the BMP and Activin/Nodal pathways, and in animal-component-free conditions, replacing the frequently used Noggin and SB431542. We also show by transcript analysis, both at the population level and for the first time at the single-cell level, that differentiated cells express genes characteristic for the development of ANE, in particular for the development of the future forebrain. Stem Cells2012;30:1875–1884


Human embryonic stem cells (hESCs) have the potential to undergo differentiation into all cell types present in the adult, including various derivatives of the ectodermal lineage such as neurons of central nervous system (CNS) and peripheral nervous system (PNS) origin as well as retinal pigment epithelium and pituitary gland [1–3]. This feature makes hESCs attractive not only for basic research of the fundamental mechanisms involved in lineage commitment but also for therapeutic applications including cell-based therapies, toxicology screens, and drug development.

Published protocols for the differentiation of hESCs, or indeed induced pluripotent stem cells, to neural lineages have tended not to emphasize the need for reproducible culture conditions or approaches that would be animal-component free. Most use serum-containing animal-derived proteins and/or N2B27 media as well as Matrigel (a mouse cell-derived product) that allows cell attachment and growth. In addition, many protocols use an embryoid body (EB) culture step, which introduces a microenvironment that is difficult to control [4–10]. Having animal-component-free, monolayer-based cell differentiation conditions would have significant advantages for the development of cell-based therapies [4]. Using small molecules to modify cell fate brings a powerful and scalable approach for directed differentiation [11–15].

The development of the vertebrate nervous system is regulated temporally and spatially by gradients of signaling molecules that play inhibitory or activating roles. Neuroectoderm specification is known to be coordinated by the underlying mesoderm from the pioneering experiments carried out by Spemann and Mangold [16]. Subsequently, in vivo experiments have demonstrated that this process is under the control of native inhibitors of bone morphogenic protein (BMP), Activin/Nodal signaling [17, 18]. This has been successfully translated to in vitro cell differentiation systems where Noggin (an endogenous polypeptide BMP antagonist), Dorsomorphin (DM) (a small molecule BMP antagonist), SB431542 (a small molecule Activin/Nodal antagonist thorough the inhibition of the ALK4, 5, 7 kinases), and Lefty (an endogenous polypeptide Nodal antagonist) have been shown to induce neuroectodermal fate [9, 10, 19, 20]. A recently published protocol by Chambers et al. brought a significant advance in the induction of neuroectoderm; inhibition of both BMP and Activin/Nodal was shown to be necessary for more efficient induction in vitro [6]. BMP and Activin/Nodal are members of the TGF-β signaling system, which can be divided into two based on the activation of specific Smad transcription factors that are the downstream mediators of TGF-β signaling. Activin/Nodal activates Smad2 and 3, whereas BMP activates Smad1, 5, and 8 phosphorylation [21].

In this study, we describe a protocol for directing hESCs toward anterior neural ectoderm (ANE) in a robust, reproducible manner that uses animal-free components and a defined cell monolayer system. We demonstrate the utility of only using small molecule inhibitors of the TGF-β signaling pathways to direct the differentiation, removing the need to include endogenous signaling polypeptides such as Noggin. Furthermore, we identify novel analogs of the ALK4/5/7 inhibitor SB431542 which are significantly more potent inducers of PAX6 expression. We also perform transcript analysis at both the population and single-cell level to characterize and validate the differentiation process.


Monolayer Differentiation of hESCs

hESCs, Shef1, and induced pluripotent stem cells (iPS) cells: Detroit 551-3 and BJ-1 (Supporting Information Materials and Methods), were routinely cultured on Matrigel (BD) in TeSR1 media (Stem Cell Technologies, Vancouver, Canada, For differentiation, pluripotent cells were dissociated with TrypLE (Invitrogen, Carlsbad, CA,, put through a cell strainer (BD, NJ, USA, to ensure single-cell suspension, and seeded at 30,000 cells per well of 96-well plate (BD-Amine, NJ, USA, on chosen matrix in TeSR2 media (Stem Cell Technologies, Vancouver, Canada, supplemented with Rock inhibitor (Y-27632 10 μM, Sigma, St. Louis, Missouri, Matrices used: Matrigel (BD, NJ, USA,, Vitronectin (Vn), and basement membrane extract (BME) (AMS Bio, Abingdon, UK, After the monolayer of cells reached 90% confluence, media were changed to “induction media”: Dulbecco's modified Eagle's medium (DMEM) knockout (KO) ES optimized (Invitrogen, Carlsbad, CA, supplemented with 20% of Xeno-Free knockout serum replacement (KSRXF, Invitrogen, Carlsbad, CA, or 20% KSR (Invitrogen, Carlsbad, CA, http://www.invitrogen. com/site/us/en/home.html) containing Noggin 50 ng/ml (R&D, Minneapolis, MN,, DM 5 μM (Sigma, St. Louis, Missouri,, LDN193189 1 μM (Stemgent, Cambridge, MA,, SB431542 10 μM (Sigma, St. Louis, Missouri,, or an SB431542 analog where stated.

Western Blotting

hESCs were stimulated ± BMP4 (50 ng/ml), Activin A (50 ng/ml), and LDN193189 (1 μM, 100 nM, 10 nM, 1 nM) for 1 hour in DMEM knockout ES optimized (Invitrogen, Carlsbad, CA, Polyvinylidene fluoride (PVDF) membrane (Invitrogen) was incubated at 4°C overnight in 5% milk in tris-buffered saline with tween 20 with primary antibodies: Smad1 (1:500), phospho-Smad1/5/8 (1:50), Smad2/3 (1:500), and phospho-Smad2 (1:200) (all from Cell Signaling, Danvers, MA, following 1 hour room temperature incubation with IR-dye-conjugated secondary antibodies (LI-COR Biosciences, Lincoln, Nebraska,


Cells were fixed in 4% paraformaldehyde for 30 minutes then incubated with 10% donkey serum, 0.3% Triton X-100 in phosphate buffered saline (PBS) containing primary antibody for 1.5 hour. Subsequently, cells were incubated for 1 hour with fluorescence-labeled secondary antibodies (1:1,000; Invitrogen, Carlsbad, CA, and 5 minutes with Hoechst 33342 500 ng/ml (Promega, Fitchburg, Wisconsin, Primary antibodies used: PAX6 (1:300, rabbit, Covance, Princeton, New Jersey,, OCT4 (1:300, goat, Santa Cruz, California,, β-III TUBULIN (1:1,000, mouse; Sigma), TH (1:500, rabbit, Sigma, St. Louis, Missouri,, GABA (1:500, rabbit, Sigma, St. Louis, Missouri,, and PERIPHERIN (1:200, goat, Santa Cruz, California,

Imaging and Quantifications of Cells

Twelve fields, covering the edges and the center of each well in the 96-well plate, were imaged (at ×10 magnification) using ImageXpress high throughput cellular imaging system (Molecular devices, Sunnyvale, CA, At least six wells per tested condition in each experiment were imaged and the cell count was averaged. Total cell number and percentage of cells positive for certain marker (OCT4 and PAX6) were assessed using MetaXpress3.1 software (Molecular Devices, Sunnyvale, CA, Identification of objects was by set size of nucleus with fixed intensity threshold, for each of the channels, across experiments and time points.

DiscoveRx Kinase Assay—KINOMEscan Technology (KdELECT)

Kd against selected kinases was determined using 11 point concentration-response curves (top concentration 30 μM, threefold dilutions) in duplicate. (

Quantitative Real-Time Polymerase Chain Reaction

Total RNA was extracted using an RNeasy micro kit (Qiagen, Hilden, Germany, DNase treatment was included into RNA preparation. One microgram of total RNA was transcribed using the high capacity RNA-to-cDNA kit (Applied Biosciences, Foster City, CA, Amplified material was detected using TaqMan gene expression assays (Supporting Information Table S1). Results were normalized to Gapdh and 18s and are from three technical replicates of two independent biological samples.

Microarray Studies

mRNA was hybridized and scanned on Illumina HT-12v4 BeadChips by Aros Inc. according to the manufacturer's instructions. The significance of expression changes throughout the whole time course was measured using ANOVA (analysis of variance). Gene set enrichment analysis was performed on gene sets from the gene ontology (GO) [22] using a hypergeometric test with DAVID [23]. All quoted p values are corrected for multiple testing (Benjamini and Hochberg [24]). All analysis were performed using R/Bioconductor [25] (Supporting Information Materials and Methods)

Single-Cell Quantitative Real-Time Polymerase Chain Reaction

Cell monolayer was treated with TrypLE (Invitrogen, Carlsbad, CA, for 15 minutes to achieve single-cell suspension, triturated, passed through a cell strainer (BD, NJ, USA,, and then single-cell sorted (Beckman Coulter MoFlo, High Wycombe, UK, into wells of a 96-well polymerase chain reaction (PCR) plate containing preamplification mastermix (5 μl Cellsdirect 2× reaction buffer (Invitrogen, Carlsbad, CA,, 2.5 μl 0.2× primer/probe mix, 1.2 μl Tris and EDTA buffer (TE) low EDTA (Applichem, Darmstadt, Germany,, 0.2 μl Superscript III RT/Platinum Taq mix (Invitrogen, Carlsbad, CA, http://www.invitrogen. com/site/us/en/home.html), and 0.1 μl SUPERase-in (Ambion, Carlsbad, CA, html) per well. The cDNA was diluted 1:5 in TE low ethylenediaminetetraacetic acid (EDTA) then run on a Fluidigm BioMark 96.96 Dynamic Array against 22 Taqman assays (Supporting Information Table S1) in triplicate. Data from each Fluidigm chip were normalized to the geometric mean of Actb and Gapdh expression detected in the interchip calibrator sample using the efficiency corrected calculation model [26]. (Supporting Information Materials and Methods).


Differentiation of hESCs to PAX6+ Cells in Animal-Component-Free Condition and Monolayer-Based System

Current protocols for directed differentiation have a number of drawbacks, including lack of reproducibility, use of animal-derived components both in the media and in the matrix [6, 8, 10], and the use of an EB stage during the induction period that introduces a microenvironment that is not easy to control [6, 8, 9].

We therefore first investigated conditions that would provide us with a robust protocol for induction of neuroectodermal PAX6+ cells from hESCs in xeno-free conditions. As markers of the differentiation protocol, we used two transcription factors, Oct4 and Pax6. Oct4 is a key pluripotency regulator that is downregulated when cells differentiate, whereas Pax6 plays a crucial role in the development of the central nervous system and the eye and is upregulated after induction of the neuroectodermal fate [27, 28].

To determine whether we can remove the necessity for Matrigel (an extracellular matrix material made from mouse sarcoma cells), we evaluated other types of substrates that support growth of hESCs. Human BME and human Vn provide similar support for cell growth as compared to Matrigel (Fig. 1A). To increase the reproducibility, homogeneity, and efficiency of PAX6 induction, we adapted a protocol for differentiation of cells in an adherent monolayer-based system. In this system, hESCs were seeded as single dissociated cells in defined, animal protein-free TeSR2 medium in the presence of Rock inhibitor, 10 μM (Y-27632) [29]. After 2 days (D2-0), when the cells reach 80%–90% confluence, the culture medium was changed to induction medium supplemented with Noggin and SB431642 (SB) or Noggin, DM, and SB, as reported previously [6, 10]. SB was kept in the media for 6 days (D2-6) (Fig. 1C). We investigated different types of induction media that would replace the mix of KSR and N2B27. Supplementing knockout DMEM with KSRXF (Xeno-free KSR) leads to an increase in the percentage of PAX6+ cells as compared to when KSR was used: 59.8% ± 2.8% PAX6+ cells were detected in KSRXF conditions in comparison to 42.1% ± 8.8% in KSR. No significant change in the percentage of OCT4+ cells was detected between these two conditions (Fig. 1B). Together these data demonstrate monolayer-based differentiation system of hESCs to PAX6+ cells in xeno-free culture conditions.

Figure 1.

Induction of PAX6+ cells from human embryonic stem cells (hESCs) using different matrices and media. (A): Percentage of OCT4+ cells 2 days postseeding is similar for Matrigel, BME, and Vitronectin. (B): Percentage of PAX6+ cells is greater in KSRXF media (mean ± SEM; *, p < .05, Student's t test). (C): Schematic of induction of hESCs to PAX6+ cells. D2-0 is equivalent to 2 days postseeding in TeSR2 media. Abbreviations: BME, basement membrane extract; KSR, knockout serum replacement; LDN, LDN193189; MTG, Matrigel.

Small Molecules can Replace Growth Factors in the Efficient and Reproducible Differentiation of hESCs to ANE

Use of Noggin for the differentiation of hESCs is not ideal: it is expensive to use for any potential clinical application of differentiated cells and is not routinely produced according to good manufacturing practice standards. DM, a small molecule BMP inhibitor, was tested as a substitute for Noggin [30]. However, in the differentiation system described here, DM alone did not induce PAX6 expression to the same extent as a Noggin/DM combination (data not shown). To find a small molecule that would replace Noggin and/or DM, we identified compounds that share structural homology with DM and tested their ability to induce PAX6 expression using our established monolayer differentiation protocol. We found that treatment with 1 μM LDN193189 (LDN) in the presence of SB (LDN/SB) leads to an increase in the number of PAX6+ cells induced as compared to treatment with Noggin/DM plus SB (N/DM/SB). PAX6+ cells (79% ± 3.1%) were detected 8 days after induction with LDN/SB, in comparison to 59.9% ± 8.3% with N/DM. The percentage of OCT4+ cells was lower after treatment with LDN/SB suggesting quicker and/or more efficient differentiation (Fig. 2A). Furthermore, a uniform induction of PAX6 protein and a uniform decrease of OCT4 over the time course of differentiation were seen (Fig. 2B). This was observed not only on the whole surface of one well of a 96-well plate (Fig. 2C) but similarly in all the wells within the plate. This suggested that intraplate variability in the induction of PAX6+ cells could be minimized using LDN/SB, and reproducibility would be high. Comparable results were observed when other pluripotent cell lines were treated with LDN/SB (Supporting Information Fig. S1A, S1B), demonstrating broad applicability. We evaluated the coefficient of variation (CV) and the Z-factor (Z′), for PAX6 and OCT4, from eight biological replicates at D2-0, D2-6, and D2-8. CV determines whether an assay can identify significantly different data points with confidence and Z′ provides a useful tool for the evaluation of the quality of data between assays.31 The OCT4+ signal has the lowest CV at D2-0 while the reverse is true for PAX6, with the CV being close to acceptable at D2-6 and best at D2-8 (23.33% and 8.37%, respectively). This is reflected in the Z′, with D2-8 being the best time point to reproducibly detect a difference between the two signals (Fig. 2D). Induction of PAX6+ cells was not through a process of cell selection. A significant increase in total cell number together with minimal cell death was observed during the time course of induction (Supporting Information Fig. S1C, S1D).

Figure 2.

Efficient and consistent induction of PAX6+ cells with LDN/SB. (A): Induction of human embryonic stem cells with LDN/SB results in higher percentage of PAX6+ cells, *, p < .05 (Student's t test). (B): Increase of PAX6+ cells with LDN/SB is associated with decrease of OCT4+ cells. (C): Induction of PAX6 occurs evenly across the well of 96-well plate: composite of seven images. (D): Summary of coefficient of variation and Z′ (n = 8 experiments). Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; DM, dorsomorphin; LDN, LDN193189; SB, SB431542.

We also determined whether concentrations of LDN lower than 1 μM are similarly effective in PAX6 induction. This would minimize the possibility of activity at targets other than those through which the desired biological effect is achieved. A concentration-response curve demonstrated that LDN is potent at inducing PAX6 expression, with an EC50 of 4.84 nM (Fig. 3A). To further understand the mechanism through which LDN acts to induce PAX6 expression, we investigated the phosphorylation of Smad 1/5/8. LDN abolishes the effect of BMP4 on phosphorylation of Smad1/5/8 in a concentration-dependent manner (Fig. 3B), consistent with its activity being mediated through inhibition of this pathway. As expected, BMP4 has no effect on phosphorylation of Smad2, mediated by Activin A signaling.

Figure 3.

Potency and mechanism of action of LDN in the induction of PAX6+ cells. (A): Concentration response curve for LDN normalized to the max response of positive control for PAX6 (D2-6). (B): LDN attenuates BMP4-mediated Smad 1/5/8 phosphorylation. The graph shows quantification of the p-Smad signal normalized to Smad1 and P84. Abbreviation: LDN, LDN193189.

SB is generally used in cell differentiation protocols at a concentration ≥10 μM [8, 9]. Indeed, we observed that concentrations below 10 μM are suboptimal in the induction of PAX6+ cells. We therefore determined whether a number of structural analogs of SB were more potent: Cpd2 [32], Cpd3 [33], Cpd4 [34], Cpd5 [35], Cpd6 [36], Cpd7 [37], Cpd8 [38], and Cpd9 [39] (Supporting Information Fig. S2). The potency of these compounds was determined using the differentiation protocol described above; immunocytochemistry for PAX6 and OCT4 was performed 6 days postinduction (D2-6) and the SB analogs were kept in the media in the presence of LDN (1 μM) throughout the induction. The results were normalized to the maximum response for PAX6 and OCT4 generated by a positive control (LDN/SB; 100% dotted line) on each experimental plate (70% ± 5%) and are shown as a percentage of this response. Three out of eight tested compounds demonstrated an increase in the percentage of PAX6+: Cpd2, Cpd3, and Cpd4 (135% ± 1.9%, 116% ± 2.96%, 135.5% ± 5.8%, respectively). The increase in the percentage of PAX6+ cells correlated with a decrease in the number of OCT4+ cells (Fig. 4A). Comparison of the EC50 for the induction of PAX6+ cells by Cpd2, Cpd3, and Cpd4 with the EC50 for SB demonstrates that the analogs are up to 30× more potent (Fig. 4B, 4C). To confirm that the presence of both LDN and SB is important for expression of PAX6, we tested LDN, SB, or DM alone, and observed that individually each compound was unable to induce PAX6 expression (Supporting Information Fig. S3).

Figure 4.

Analogs of SB are more potent and have improved efficacy in the induction of PAX6+ cells. Concentration response curves normalized to the response of positive control (LDN/SB) for both PAX6 and OCT4. (A): SB analogs. (B): SB. (C): Summary table showing comparison of potency (EC50) between tested SB analogs and SB431542.

Specificity of the Small Molecule Inhibitors used for Induction of PAX6 Expression

We used KINOMEscan-scanELECT technology [40] to determine which type-1 and type-2 TGF-β receptors were inhibited by the SB analogs identified previously as able to induce PAX6 expression. The three most effective compounds (Cpd2, Cpd3, and Cpd4), as well one compound that did not induce PAX6 expression (Cpd7), were profiled. A summary table with Kd values for ALK1, ALK2, ALK3, ALK4, ALK5, ALK6 and ACVR2A, ACVR2B, BMPR2, TGFBR2 is available in Supporting Information Table S2. By comparing the fold differences in Kd values between SB and the selected compounds, we observed that Cpd3 and Cpd4 have up to 150-fold higher affinity for inhibition of type-1 receptors. Specifically, Cpd3 and Cpd4 showed 75- and 120-fold higher affinity for ALK4 inhibition and 94- and 68-fold higher affinity for ALK5 inhibition, respectively. In addition, Cpd3 and Cpd4 had 12- and 13-fold higher inhibition of ALK6. Cpd2 had a higher affinity at all tested type-1 receptors (25-, 24-, 68-, 34-, 22-, and 178-fold higher potency for ALK1, ALK2, ALK3, ALK4, ALK5, and ALK6, respectively) (Fig. 5A). When we similarly compared inhibition of type-2 receptors, we observed that Cpd2, Cpd3, and Cpd4 are higher affinity inhibitors (54-, 22-, and 83-fold, respectively) at ACVR2B and at TGFBR2 (270-, 81-, and 144-fold, respectively). In addition, Cpd2 had 210-fold higher affinity for ACVR2A compared to SB (Fig. 5A). Cpd7 did not inhibit type-1 and type-2 receptors to a greater extent than SB, in line with the data shown in Figure 5A. Together these data indicate that while none of the newly identified analogs are more selective for inhibition of specific ALK subtypes, they are significantly more potent than SB at inhibiting type-1 and type-2 TGF-β receptors. We further examined the activity of Cpd2 at a panel of 138 human kinases (MRC Centre for Protein Kinase Profiling). At concentrations of up to 1 μM, Cpd2 showed significant inhibition only at RIPK2, PKD1 (not expressed in hESCs), MINK1, YES1, p38aMAPK, and VEGF-R (Supporting Information Fig. S4A, S4B). The affinity of Cpd2 at these kinases is however at least 10-fold lower when compared with ALK4/5, consistent with the proposed mechanism of action.

Figure 5.

SB analogs are more potent inhibitors of type-1 and type-2 TGF-β receptors than SB. (A): Fold difference in affinity between SB and SB analogs. Data generated comparing Kd values of SB and analogs (mean ± SEM) (Supporting Information Table S2). (B): Gene expression profile of ALK2-7 during the time course of induction (mean ± SEM). Ct values for Gapdh and 18s were 23 and 13, respectively. ALK1 did not reach the detection threshold. Ct values for ALK1/2/3/4/5/6/7 were 44/28/27/29/27/33/33, respectively.

We also profiled the activity of LDN and DM at eight type-1 and type-2 TGF-β receptors using the KINOMEscan-scanELECT assay. LDN had a higher affinity for most of the type-1 and type-2 receptors than DM. The biggest difference in affinities was at ALK1, ALK2, ALK3, ALK4, and ALK5 where LDN showed 58-, 38-, 50-, 38-, and 37-fold, respectively, greater inhibition (Supporting Information Table S2).

We determined the expression of the ALK subtypes present in cells during the differentiation. Using quantitative PCR (qPCR), we found that ALK2/3/4/5/6/7 are expressed in the hESCs seeded as single cells, at the very beginning of induction (D2-0) as well as during the induction until D2-8 (Fig. 5B). ALK1 was not detected.

PAX6-Positive Cells Induced with LDN/SB, LDN/Cpd2, and LDN/Cpd4 Express Genes Characteristic for ANE and can Differentiate into Neurons

To fully characterize the induction, we performed microarray profiling of samples taken throughout the time course. We also profiled noninduced samples at two time points (D2-3 and D2-8) and samples induced with SB analogs: Cpd2 and Cpd4 at D2-8. Three biological replicates were used at each time point in order to assess reproducibility.

The time course reveals the details of the dynamics of the induction process. Figure 6A shows the expression levels of three upregulated (Pax6, Otx2, and Lhx2) and three downregulated (Pou5f1, Nanog, and Dnmt3b) marker genes. A clear sigmoidal curve is observed, with the majority of the expression changes occurring between D2-3 and D2-8. A principal component analysis (PCA) shows that this pattern of expression is also reflected genome-wide (Supporting Information Fig. S5A), suggesting that large-scale expression changes have effectively stopped by D2-8 and are slowing as early as D2-6. In contrast, the noninduced samples at D2-8 and D2-3 lag several days behind the induced samples. The noninduced D2-8 samples are most similar to the induced samples at D2-4, implying that, even by D2-8, they are still changing their gene expression toward the more differentiated cell state.

Figure 6.

Gene expression changes upon induction with LDN and SB. (A): Expression profile over time of three of the most upregulated and downregulated genes. (B): Induction leads to the expression of genes specifically implicated in forebrain as opposed to hindbrain development.

Clustering the samples based on the correlations between their expression profiles and a PCA (Supporting Information Fig. S5A, S5C) shows that contemporaneous samples cluster together, suggesting a high degree of temporal reproducibility. The clustering analysis also reveals that this reproducibility is not observed in the noninduced samples. The variance between replicates for many genes, including important markers such as POU5F1 and PAX6 (Supporting Information Fig. S5D, S35E), is significantly higher in the noninduced D2-8 sample when compared with the other samples. This suggests that directed differentiation using small molecule inhibitors increases the reproducibility of the differentiation as well as the speed.

We observed almost no difference between the cells that were treated continuously with LDN until D2-11 versus cells exposed to LDN only until D2-8 (Supporting Information Fig. S5B), and when we tested for differential expression (p < .01, more than twofold change) between these two groups, we saw no differentially expressed (DE) genes, suggesting that the effect of LDN at these later time points is negligible. The clustering also demonstrates the functional equivalence of the two SB analogs. For both analogs tested, at D2-8, they cluster closely with the SB treated D2-8 sample. We tested for differential expression (p < .01, more than twofold change) between the three groups (SB, Cpd2, and Cpd4) and found no DE genes between Cpd2 and Cpd4 and only 39 DE genes between the analog-treated samples and the SB treated. Although this number is small, a gene set enrichment analysis (Supporting Information Table S3) showed that these genes are enriched for those involved in nervous system development (hypergeometric test; p < .01), suggesting there may be subtle differences between the effects of the analogs and SB, perhaps due to activities at molecular targets beyond those currently known. Expression profiles of the 14 DE genes involved in nervous system development are given in Supporting Information Figure S6.

To formally demonstrate that the induction is specific for the generation of anterior (forebrain) neuroectoderm, as opposed to hindbrain lineages, we made an unbiased selection of the top 20 forebrain and hindbrain development-specific genes as measured by the significance of their expression changes throughout the time course. Forebrain and hindbrain development-specific genes were defined using GO terms. A heatmap of the expression changes of these genes is shown in Figure 6B. The majority of forebrain-specific genes are significantly induced, while in the hindbrain set only Gas1 is. This outlier appears to represent a misannotation in GO as Gas1 has been clearly linked to forebrain development [41] as well as hindbrain [42]. This effect is not cut-off dependent. When tested with a range of significance thresholds, hindbrain genes were not significantly enriched among DE genes (hypergeometric p < .01) at any point, whereas forebrain genes were at all the thresholds tested (hypergeometric p < 1e−24 − 1e−3; Supporting Information Fig. S5F). We next looked at the potential of the PAX6+ cells generated using LDN, SB, and LDN/Cpd2 to differentiate into neuronal subtypes. We were able to show that PAX6+ cells generated according to our differentiation protocol can differentiate to neuronal subtypes such as dopaminergic, GABAergic neurons as well as neuronal cells expressing PERIPHERIN (Supporting Information Fig. S7)

Single-Cell Analysis During the Time Course of Induction Reveals Transient Subpopulations of Cells

Measuring expression levels in single cells allows us to assess population heterogeneity, delineate subpopulations, and observe their dynamics. Figure 7A shows the expression levels of 22 genes across 470 individual cells taken from six time points. The 22 genes were chosen to cover genes of marking pluripotent, epiblast, neural, and eye field cells as well as house-keeping genes. The microarray data were used to select for genes likely to show a significant change in expression over the time course. Two genes commonly associated with muscle development (Acta2 and Tnnt2) that were expressed at later time points were also selected in order to determine whether the cells expressing them formed a distinct subpopulation.

Figure 7.

Expression profiles of 22 selected genes at single-cell level during the time course of induction. (A): Expression heatmap of all cells grouped according to time of collection. (B): Proportion of cells at each time point belonging to each of the clusters. (C): The numbers and colors in the grid show the proportion of cells within each cluster.

Figure 7A shows that between D2-0 and D2-2 there is a reduction of Nanog and Pou5f1 levels consistent with loss of pluripotency. Subsequently, between D2-2 and D2-4, we see expression of Pax6 increasing followed by increasing expression of other brain markers such as Emx2 and Lhx2 and at D2-6, eye field markers such as Rax.

A PCA of the data (Supporting Information Fig. S8A) allows two further observations: first that no significant differentiation occurs between D0-0 and D2-0 and between D2-6 and D2-8. Indeed, close examination of Figure 7A and Supporting Information Figure S6A shows that there is higher expression of pluripotency markers and reduced expression of Otx2 at D2-0 when compared with D0-0. These observations are consistent with the first 2 days of the time course acting as a selection step, removing spontaneously differentiating cells.

The second observation is that the overall heterogeneity of the population increases over time. Supporting Information Figure S8B shows how the mean cell-cell distance (defined as 1 − R; where R is the correlation in expression levels between two cells) at first decreases (consistent with the initial stage of the time course acting as a selection step) and then steadily increases with time.

The increase in heterogeneity at later time points implies that distinct subpopulations may be present. To better visualize these populations, we used k-means clustering to extract seven clusters of cells within the complete dataset (Supporting Information Fig. S8C).

Although the clustering is based on all 22 genes, the clusters detected can be summarized according to the presence or absence of eight principal marker genes: Nanog, Tdgf1, Otx2, Lhx2, Pax6, Six3, Rax, and Acta2. Figure 7B shows the proportion of cells within each cluster expressing each one of these marker genes. Clusters 1 and 2 clearly represent the pluripotent cells where 100% express Nanog and Tdgf1. These two clusters are differentiated by the expression of higher levels of Nanog and Pou5f1 and lower levels of Otx2 in cluster 2. Clusters 3 and 4 represent cells that have lost Nanog expression but have retained Tdgf1 expression. In this case, they are differentiated by expression of Pax6, which is only seen in cluster 4. Clusters 5–7 are all Pax6 positive and Nanog and Tdgf1 negative but are differentiated by Acta2 expression (seen only in cluster 6) and Lhx2/Six3/Rax expression (seen only in cluster 7).

Figure 7B shows how the proportion of the total population represented by each cluster varies with time. At D0-0 and D2-0 clusters, 1 and 2 dominate, although, consistent with previous observations, D2-0 is more homogeneous than D0-0, being comprised almost entirely of cells from cluster 2. At D2-2, the proportions change dramatically: cluster 2 disappears and cluster one is now in a minority; cluster 3, representing Pax6-negative cells that have nonetheless lost Nanog expression (but retained Tdgf1) dominates. Between D2-2 and D2-4, the three Pax6-positive clusters emerge. The Rax/Six3/Lxh2-positive cluster 7, which represents prospective eye field cells, is most common at D2-6 (comprising ∼ 20% of the population). Although it appears to drop as a proportion of the total population at D2-8, a Fisher's exact test suggests that the change in proportions between D2-6 and D2-8 is not significant, although all the earlier changes are (Fisher p < .05).


Development of efficient and reproducible conditions for directed differentiation of hESCs into specific cell types is an important goal not only to study early human development but also to enable more practical applications, such as in vitro models, drug discovery, and cell therapies. To achieve the more applied goals, robust, scalable, and cost-effective methodologies are required. For cell therapy, xeno-free conditions are highly desirable. Previously described neural differentiation protocols [7–10], however, incorporate an EB step and use proteins and media containing animal-derived components. While a recent study describes a monolayer differentiation condition, it does not show the reproducibility of the protocol [6]. Several approaches have been described for xeno-free hESC culture, but none have been extended for use in differentiation protocols [4]. Here, we describe a method for the robust and reproducible differentiation of a hESC monolayer using media that contains Xeno-Free KSR and a MTG replacement of human origin. In addition, our data suggest that this protocol is easily scalable to T25 flasks (data not shown).

We focused upon the replacement of Noggin and/or DM with small molecules that would potently inhibit BMP signaling. DM has been reported previously to convert hESCs to neural fate [10]. We were unable to reproduce these observations under our differentiation conditions and found that DM is toxic to cells at concentrations above 5 μM. This toxicity might be due to the activity of DM at other molecular targets within the cell. Indeed, a recent publication [43] shows that at 1 μM, DM inhibits many more kinases than just ALK2/3/6. We found that the small molecule inhibitor LDN (in the presence of SB) is a very potent and efficient inducer of PAX6+ cells and that, in hESCs, LDN inhibits BMP4-induced Smad1/5/8 phosphorylation, consistent with a mechanism of BMP inhibition.

SB has been reported previously to differentiate hESCs to neuroectoderm with caudal positional identity. We observed however that under differentiation conditions described here, SB alone does not increase the number of PAX6+ cells. The discrepancy could be due different culture conditions used by Patani et al. [8]. SB (in the presence of LDN) was found to exert its effect on PAX6 induction only at concentrations above 4 μM, consistent with its low affinity at the various ALK subtypes (Supporting Information Table S2) [43]. Our search for more potent structural analogs of SB identified a group of small molecules that were up to 30-fold more potent than SB in PAX6 induction, and had up to 150-fold higher affinity at ALK subtypes, particularly ALK4 and ALK5. This study therefore describes novel, potent Activin/Nodal pathway inhibitors that can be used at nanomolar concentrations to differentiate hESCs.

While methods for derivation of neuroectodermal cells from hESCs have been reported in the literature, to our knowledge, whole genome transcript profiling of cells across the time course of induction from has not been described. We use transcript profiling to show that the cells generated using the small molecule inhibitor differentiation protocol described here reproducibly express genes characteristic of ANE and to define the dynamics of the differentiation process.

Understanding the dynamics is crucial, as the time point of addition of differentiation inducers (polypeptides or small molecules) during induction/differentiation is vital to steer the fate of cells to a desired cells type [44]. This whole genome transcript profiling approach may therefore help to elucidate why some differentiation inducers have a precise temporal requirement. In this context, we observed that dual inhibition with LDN/SB for 1 day only + additional 1 day with LDN was sufficient to drive the efficient generation of PAX6+ cells. This suggests that the signaling changes evoked upon addition of the two inhibitors are rapid and do not require extended inhibition of the Activin/BMP signaling pathways. Therefore, Noggin/DM/SB application throughout the first 5–11 days neural induction, as described in the literature [6, 8, 9, 19], appears unnecessary.

To explore the heterogeneity within the cell population, we performed single-cell qPCR. This shows the existence of transitory populations of cells coexisting during the time course. Consistent with the results of the whole genome transcript analysis, the majority of the changes in these populations occur between D2-0 and D2-6. While, the ontogenic relationship between the subpopulations of cells is unknown, as is whether there is a difference in the plasticity/fate choice between subpopulations of cells, nevertheless single-cell analyses open the opportunity of refining our understanding of cell differentiation and fate choice.


We have developed a protocol for efficient and reproducible induction of hESCs to ANE cells using potent small molecules to inhibit Activin/BMP signaling. Transcript analysis at the single-cell level reveals the coexistence of transient and dynamic subpopulations of cells during the time course of differentiation.


This work was supported by Worldwide Research&Development Postdoctoral Training program at Pfizer. Julie Kerby, Juliette Steer, and Maria Isabel Rosa are thanked for their help with hESC culture, Magda Bictash and Emma Jarvie for help with compounds screening, and Rachel Walker from Flow Cytometry Core Facility at Welcome Trust Centre for Stem Cells Research, University of Cambridge, for cell sorting. We would also like to thank Jamie Bilsland and Caroline Benn for help in reviewing the manuscript.


A.G. has employment/leadership position and Ownership interest with Pfizer Ltd., B.S. has employment/leadership position with Pfizer Neusentis Ltd., and P.W. is a Pfizer Neusentis employee and owns Pfizer and Merck stock.