Fusion protein Isl1–Lhx3 specifies motor neuron fate by inducing motor neuron genes and concomitantly suppressing the interneuron programs

Abstract

Combinatorial transcription codes generate the myriad of cell types during development and thus likely provide crucial insights into directed differentiation of stem cells to a specific cell type. The LIM complex composed of Isl1 and Lhx3 directs the specification of spinal motor neurons (MNs) in embryos. Here, we report that Isl1–Lhx3, a LIM-complex mimicking fusion, induces a signature of MN transcriptome and concomitantly suppresses interneuron differentiation programs, thereby serving as a potent and specific inducer of MNs in stem cells. We show that an equimolar ratio of Isl1 and Lhx3 and the LIM domain of Lhx3 are crucial for generating MNs without up-regulating interneuron genes. These led us to design Isl1–Lhx3, which maintains the desirable 1:1 ratio of Isl1 and Lhx3 and the LIM domain of Lhx3. Isl1–Lhx3 drives MN differentiation with high specificity and efficiency in the spinal cord and embryonic stem cells, bypassing the need for sonic hedgehog (Shh). RNA-seq analysis revealed that Isl1–Lhx3 induces the expression of a battery of MN genes that control various functional aspects of MNs, while suppressing key interneuron genes. Our studies uncover a highly efficient method for directed MN generation and MN gene networks. Our results also demonstrate a general strategy of using embryonic transcription complexes for producing specific cell types from stem cells.

Developing central nervous system (CNS) produces a vast number of neuronal types, but adult CNS has only limited capacity to regenerate neurons. This has prompted great interest in identifying methods to produce specific neuronal types from stem cells. Production of differentiated cell types from pluripotent stem cells, such as embryonic stem cells (ESCs), should enable a continuous supply of diseased cell types for drug screening and cell replacement therapy and provide valuable insights into the pathophysiology of human diseases. One important challenge in this effort is to steer stem cells into specific cell types. Recapitulation of normal developmental processes using embryonic inductive signals has been used to drive differentiation of pluripotent stem cells into specific cell types (1). However, this strategy tends to trigger formation of mixed cell types rather than a targeted cell type, because each inductive signal is used in multiple developmental pathways. This shortcoming might be circumvented by using more specific, downstream transcription factors of inductive signals. In this regard, it should be noted that many transcription factors function in combination to determine cell fates during development, suggesting that coexpression of multiple transcription factors could be a more effective method to generate a particular cell type from pluripotent stem cells.

Motor neurons (MNs) in the spinal cord project axons to muscles and control their contraction. The developmental pathways to generate MNs have been relatively well studied. In the developing spinal cord, sonic hedgehog (Shh) signal triggers the expression of two LIM homeodomain (HD) transcription factors Isl1 and Lhx3 in differentiating MN cells (2, 3). Then, Isl1 and Lhx3 form a transcriptional activating MN-hexamer complex, in which two Isl1:Lhx3 dimers are assembled into a complex via a self-dimerizing cofactor nuclear LIM interactor (NLI, also called LDB for LIM domain binding) (Fig. 1A) (4, 5). This complex is sufficient to induce ectopic MN formation in the dorsal spinal cord, which is not exposed to a high concentration of Shh (4, 5). Lhx3 alone directs the specification of V2 interneurons (V2-INs) by forming the V2-tetramer complex, consisting of two Lhx3 and two NLI molecules (Fig. 1A) (4, 6). Thus, the combinatorial action of Isl1 and Lhx3, exerted via the formation of the MN hexamer, is critical to induce MN differentiation without triggering V2-IN differentiation. Identification of the downstream genes that are controlled by the MN hexamer would provide important insights into the developmental processes to generate functionally mature MNs.

Fig. 1.

Isl1–Lhx3 is a specific and potent MN inducer. (A) MN-hexamer and V2-tetramer complexes direct the differentiation of MNs and V2-INs, respectively, in the developing spinal cord. HxRE, MN-hexamer response element; TeRE, V2-tetramer response elements. (B) Hb9⁺ MN and Chx10⁺ V2-IN specification analyses in chicks electroporated with Lhx3 and Isl1 in indicated ratios. The efficiency of MN and V2-IN induction was quantified by the number of ectopic Hb9⁺ MNs or Chx10⁺ V2-INs among all Isl1⁺ electroporated cells. *P < 0.001 in the two-tailed t test. (C) Schematic representation of MN-hexamer mimetic fusions. (D and E) Luciferase reporter assays in P19 cells using HxRE:LUC (D) or MN-enhancer:LUC (E) reporters. (F and G) MN and V2-IN specification analyses in chicks electroporated with indicated constructs. The efficiency of MN and V2-IN induction was quantified by the number of ectopic MNs or V2-INs among all Lhx3⁺ electroporated cells. Error bars represent the SD (B and D–G).

MNs differentiated from stem cells have proven to be useful for developing potential therapies for human MN diseases. Shh, when combined with retinoic acid (RA), converted ESCs and induced pluripotent stem cells (iPSCs) to MNs (7–10). However, under this condition, Shh also differentiates ESCs into spinal interneurons (7). To develop new methods that resolve this specificity issue and generate spinal MNs from stem cells with higher fidelity and efficiency, we explored the possibility of using the embryonic transcription program for MN generation, the MN hexamer, instead of Shh signal. We first investigated the mechanisms underlying the MN-hexamer function in MN differentiation and then applied this information to design a strategy for stem cell differentiation to MNs. Here we report that an equimolar ratio of Isl1 and Lhx3 is critical for specifically generating MNs without activating V2-IN pathway and that the LIM domain of Lhx3 is required for the effective recognition of MN-hexamer response elements (HxREs) by the MN hexamer. These findings led us to develop a MN-hexamer mimetic fusion, Isl1–Lhx3, which directs highly specific and efficient differentiation of ESCs into MNs. The RNA-seq analyses of Isl1–Lhx3-induced MNs revealed that Isl1–Lhx3 up-regulates a battery of genes that control a wide range of MN functions and concomitantly suppresses the interneuron differentiation programs. To our knowledge, this is a unique demonstration that a fusion mimicking an embryonic transcription complex can serve as an ideal tool to generate a targeted cell type from stem cells.

Results

Ratio Between Isl1 and Lhx3 Is Critical for the Specific Generation of MNs.

Once Lhx3 is incorporated into the MN hexamer, it cannot form the V2 tetramer (Fig. 1A). Thus, an equimolar expression of Isl1 and Lhx3 should be critical for MN-specific differentiation of stem cells without inducing V2-IN differentiation. To test this idea, we expressed Isl1 with an increasing amount of Lhx3 in the chick neural tube and monitored the ectopic formation of Hb9⁺ MNs and Chx10⁺ V2-INs in the dorsal spinal cord (Fig. 1B and Fig. S1). When the ratio of Lhx3 to Isl1 was 0.5, only Hb9⁺ MNs, but no ectopic Chx10⁺ cells, were formed. However, increasing the amount of Lhx3 led to the generation of ectopic Chx10⁺ cells even in the presence of Isl1 (Fig. 1B and Fig. S1). When the ratio of Lhx3 to Isl1 was 8, several cells acquired MN–V2-IN hybrid characteristics expressing both Hb9 and Chx10 (Fig. S2). The ectopic generation of Chx10⁺ cells following coelectroporation of Isl1 and Lhx3 likely results from an excess of Lhx3 molecules, which form the V2-tetramer. Thus, expression levels of Isl1 and Lhx3 should be tightly controlled at or close to an equimolar ratio to differentiate neural stem cells specifically to MNs.

Isl1–Lhx3 Fusion Is a Specific and Efficient Inducer of the MN Fate.

In keeping the optimal equimolar ratio of Lhx3 to Isl1, we generated three fusions of Isl1 and Lhx3, which are predicted to mimic the MN hexamer structurally (Fig. 1C). DD–Isl1^HD–Lhx3^HD consists of the dimerization domain (DD) of NLI fused to the DNA-binding HDs of Isl1 and Lhx3 (4). Isl1–Lhx3^HD consists of full-length Isl1 fused to the HD of Lhx3. Isl1–Lhx3 is a fusion protein of full-length Isl1 and Lhx3. Isl1–Lhx3^HD and Isl1–Lhx3 can form MN-hexamer mimetic complexes with widely expressed endogenous NLI (4, 11). To test whether the fusions activate the MN-hexamer target enhancers, we performed luciferase assays in P19 mouse embryonic cells using the luciferase reporters linked to HxREs and to the MN-specific enhancer in the Hb9 gene, in which the MN-hexamer transcriptionally synergizes with the proneural basic helix–loop–helix (bHLH) factor NeuroM (NeuroD4) or Ngn2 (Neurog2) (5, 11–13). Isl1–Lhx3 was efficient in activating HxRE:LUC, whereas DD–Isl1^HD–Lhx3^HD and Isl1–Lhx3^HD were much less effective than Isl1 plus Lhx3 (Fig. 1D). Similarly, Isl1–Lhx3 collaborated with NeuroM to trigger a potent activation of the MN enhancer, whereas DD–Isl1^HD–Lhx3^HD and Isl1–Lhx3^HD showed only a marginal level of activation even in the presence of NeuroM (Fig. 1E). These results indicate that Isl1–Lhx3 is a powerful activator of the HxREs and cooperates with NeuroM for the transcriptional activation of MN genes.

To test whether the fusions induce MN generation in vivo, we expressed each fusion in chicken embryonic spinal cord using in ovo electroporation and monitored the ectopic formation of Hb9⁺ MNs in the dorsal spinal cord. Consistent with the reporter assays, Isl1–Lhx3 triggered MN generation significantly more efficiently than coexpression of Isl1 and Lhx3, and DD–Isl1^HD–Lhx3^HD and Isl1–Lhx3^HD were much less effective (Fig. 1F and Fig. S3). All three fusions did not induce ectopic Chx10⁺ cells, unlike coexpression of Isl1 and Lhx3, which produced several Chx10⁺ cells in the dorsal spinal cord (Fig. 1G and Fig. S3). These results indicate that the three MN-hexamer mimetic fusions do not form a V2-tetramer–like complex, because the LIM domain of Lhx3 is either deleted or unavailable to bind to NLI (Fig. 1C). Together, these results identify Isl1–Lhx3 as a potent and specific MN inducer that overcomes the specificity issue associated with coexpression of Isl1 and Lhx3.

LIM Domain of Lhx3 Is Critical to Induce MN Differentiation.

In the assembly of the MN hexamer, the Isl1–LIM domain functions to bind to NLI, whereas the Lhx3–LIM domain is important to bind the C-terminal domain of Isl1 (Fig. 1A) (4). Our results suggest an unexpected role of the Lhx3–LIM domain in MN specification, apart from its known function to bind Isl1. To test whether the Lhx3–LIM domain within Isl1–Lhx3 is needed to provide optimal distance between two DNA-binding HDs of Isl1 and Lhx3, we made an Isl1–L1–Lhx3 fusion in which the Lhx3–LIM domain is replaced by the LIM domain of Lhx1 (Lim1) (Fig. 2A). As the LIM domains of Lhx3 and Lhx1 are highly homologous with each other (∼70% homology, Fig. S4), the HDs of Isl1 and Lhx3 within Isl1–Lhx3 and Isl1–L1–Lhx3 are similarly spaced in primary sequences. We compared the activation of the MN-hexamer target genes by Isl1–Lhx3 and Isl1–L1–Lhx3 using the HxRE:LUC and MN-enhancer:LUC reporters. Isl1–Lhx3 potently activated both reporters, whereas Isl1–L1–Lhx3 was ineffective, despite their comparable expression levels (Fig. 2B). Similarly, Isl1–Lhx3, but not Isl1–L1–Lhx3, synergized strongly with Ngn2 in stimulating the MN enhancer (Fig. 2C).

Fig. 2.

The Lhx3–LIM domain is needed for a potent MN-inducing activity of Isl1–Lhx3. (A) Schematic representation of various fusions. (B and C) Luciferase reporter assays in P19 cells using HxRE:LUC (B) or MN-enhancer:LUC (C) reporters. (D) GFP expression in chicks electroporated with HxRE:GFP and Isl1–Lhx3 or Isl1–L1–Lhx3. Isl1–Lhx3 triggered the ectopic GFP expression in the dorsal spinal cord, but Isl1–L1–Lhx3 did not. (E) Hb9⁺ MN specification analyses in chicks electroporated with indicated constructs above. Isl1–Lhx3 induced ectopic Hb9⁺ MNs above the horizontal lines, whereas Isl1–L1–Lhx3 did not. (F) Efficiency of MN induction was quantified by the number of ectopic MNs among all Lhx3⁺ electroporated cells. Error bars represent the SD (B, C, and F). (G) In vivo GST pull-down assays in HEK293 cells expressing GST–NLI and HA-tagged fusions.

Next, we tested the ability of Isl1-Lhx3 and Isl1-L1-Lhx3 to activate MN genes in the developing spinal cord. The electroporation of HxRE:GFP results in MN-specific GFP expression in chick neural tube as the endogenous MN hexamer in MNs activate HxREs (5). Isl1–Lhx3, but not Isl1–L1–Lhx3, triggered ectopic expression of HxRE:GFP in the dorsal neural tube (Fig. 2D). Likewise, Isl1–L1–Lhx3 was inert in ectopic MN generation in chick neural tube despite its high level of expression, whereas Isl1–Lhx3 induced ectopic MN formation (Fig. 2 E and F). Thus, the Lhx3–LIM domain within Isl1–Lhx3 plays an active role for efficient MN generation, rather than playing a passive role as a spacer.

The LIM domain of Lhx3, but not that of Lhx1, binds to the C-terminal region of Isl1 (14). Thus, it is possible that, within the Isl1–Lhx3 fusion, the Lhx3–LIM domain interacts with Isl1, allowing Isl1–Lhx3 to assume the native conformation of the MN-hexamer. To test this idea, we examined whether the LIM domain of Lhx3, which is fused with the C-terminal region of Isl1, is available to interact with NLI. If the LIM domain of Lhx3 is preoccupied due to the intra- or intermolecular interaction with the Isl1–C-terminal region of the fusion, it would not be available for NLI interactions. We expressed Isl1^HD–Lhx3 or Isl1^HD–L1–Lhx3 (Fig. 2A) with GST–NLI in HEK293 cells and purified NLI-associated proteins using glutathione beads. Whereas Isl1^HD–L1–Lhx3 efficiently associated with NLI in cells, Isl1^HD–Lhx3 did not (Fig. 2G), indicating that the Lhx3–LIM domain within Isl1^HD–Lhx3 is not available for NLI interaction. These results support our idea that, in Isl1–Lhx3, the C-terminal region of Isl1 interacts with the Lhx3–LIM domain intra- or intermolecularly.

NLI-Mediated Dimerization of Isl1–Lhx3 Is Important for MN Differentiation.

We considered the possibility that the Lhx3–LIM domain within Isl1–Lhx3 binds to the Isl1–C-terminal domain in another Isl1–Lhx3 molecule in trans, leading to the formation of Isl1–Lhx3 homodimer without NLI, and that this Isl1–Lhx3 homodimer is sufficient to activate the MN genes (Fig. 3A). To test this possibility, we examined whether Isl1^HD–Lhx3, which contains both the Isl1–C-terminal domain and the Lhx3–LIM domain, self-dimerizes (Fig. 3A). The CoIP assays revealed that Flag-tagged Isl1^HD–Lhx3 associates with HA-tagged Isl1^HD–Lhx3, and that this interaction was not disrupted by the dimerization domain of NLI (NLI-DD), an inhibitor of NLI self-dimerization (4) (Fig. 3B). Combined with the finding that Isl1^HD–Lhx3 does not bind NLI (Fig. 2G), these data suggest that Isl1^HD–Lhx3 fusion forms a homodimer without NLI (Fig. 3A). Isl1^HD–Lhx3 neither activated the HxRE:LUC reporter in P19 cells nor induced ectopic MNs in chicken embryos (Fig. 3 C and D). These data indicate that the Isl1^HD–Lhx3 homodimer is unable to activate the MN differentiation program, unlike Isl1–Lhx3.

Fig. 3.

NLI-mediated dimerization of Isl1–Lhx3 is important for MN differentiation. (A) Schematic representation of possible complexes. (B) CoIP assays using HEK293 cells expressing HA–Isl1^HD–Lhx3 and Flag–Isl1^HD–Lhx3 along with or without HA-tagged NLI-DD. Flag and HA antibodies were used for IP and Western blotting, respectively. *, heavy chain bands. (C and E) Luciferase reporter assays in P19 cells using HxRE:LUC reporter. Error bars represent the SD. (D and F) MN specification analyses in chicks electroporated with indicated constructs. (D) Isl1^HD–Lhx3 failed to trigger ectopic MN formation in the dorsal neural tube. (F) NLI-DD inhibited Isl1–Lhx3-triggered ectopic MN induction (MNs above the red horizontal line). (G) Gel-shift analyses with proteins indicated above using the MN enhancer (MNe, Upper) and HxRE (Lower) as ³²P-labeled probes. (H) ChIP assays in P19 cells transfected with fusions indicated above using IgG or α-Lhx3 antibody. Isl1–Lhx3, but not Isl1–L1–Lhx3, is recruited to the MNe of the Hb9 gene. (Lower) Expression of Isl1–Lhx3 or Isl1–L1–Lhx3 in P19 cells using Western blotting assays with α-Lhx3 antibody.

Isl1–Lhx3 could form both the homodimer via intermolecular interactions without NLI and the MN hexamer via NLI self-dimerization (Fig. 3A). To test which of the two complexes functions to induce MN differentiation, we used NLI-DD, which disrupts assembly of the MN-hexamer, but not formation of Isl1–Lhx3 homodimer (4). NLI-DD strongly inhibited the HxRE activation by Isl1–Lhx3 in P19 cells as well as ectopic MN formation by Isl1–Lhx3 in the developing spinal cord (Fig. 3 E and F). These data establish that the MN-hexamer, not the Isl1–Lhx3 homodimer, is the functional complex that directs MN specification.

Isl1:Lhx3 Interaction Is Needed for the MN Hexamer to Bind the HxREs.

To test whether the interaction between the Lhx3–LIM domain and the Isl1–C-terminal domain aligns the HDs of Isl1 and Lhx3 in Isl1–Lhx3 for efficient binding to the HxREs, we monitored the HxRE-binding ability of Isl1–Lhx3 and Isl1–L1–Lhx3. In gel-shift analyses, Isl1–Lhx3 bound to both HxRE and the MN enhancer of the Hb9 gene as efficiently as the mixture of Isl1 and Lhx3, but the binding of Isl1–L1–Lhx3 was barely detectable (Fig. 3G). Similarly, chromatin immunoprecipitation (ChIP) assays revealed that Isl1–Lhx3 was recruited to the MN enhancer of the Hb9 gene in P19 cells, whereas Isl1–L1–Lhx3 was not (Fig. 3H). Together, these results uncover that the interaction between the Isl1–C-terminal domain and the Lhx3–LIM domain is important to ensure efficient binding of the MN-hexamer to the HxREs.

Isl1–Lhx3 Efficiently Directs Differentiation of Mouse ESCs to MNs.

Our results suggest that Isl1–Lhx3 is an ideal tool to specifically induce MN differentiation in stem cells. To test this idea, we established inducible (i)MN-ESCs, in which Isl1–Lhx3 coding sequence was inserted downstream of the tetracycline response element (TRE) and the reverse tetracycline transactivator (rtTA) was integrated into the constitutively active ROSA26 locus (Fig. 4A and Fig. S5) (15). The expression of Isl1–Lhx3 was robustly induced by doxycycline (Dox) treatment (Fig. S6).

Fig. 4.

Isl1–Lhx3 expression triggers efficient and specific differentiation of ESCs to MNs. (A) Schematic representation of iMN-ESCs. Dox treatment up-regulates Isl1–Lhx3, which induces MN genes in iMN-ESCs. TRE, tetracycline response element; rtTA, reverse tetracycline transactivator. (B) Experimental design to differentiate ESCs to MNs. (C and D) Cell differentiation analyses in iMN-ESCs treated with RA alone, RA plus Dox, or RA plus Shh. (D) Efficiency of MN induction was quantified by the number of Hb9⁺TuJ⁺ MNs among all TuJ⁺ neurons. Error bars represent the SD; *P < 0.001 in two-tailed t test. (E) Isl1–Lhx3 expression induces VAChT⁺NF⁺ cholinergic neurons. (Scale bar, 25 μm.) (F) RT-PCR analyses to test expression of markers for neurotransmitter phenotypes. (G) Isl1–Lhx3-induced MNs form neuromuscular junctions (arrows) with myotubes. Clustering of acetylcholine receptors (AChR) was determined by patched α-bungarotoxin staining. (Scale bar, 20 μm.)

Given that Isl1–Lhx3 directs ectopic MN formation in the dorsal spinal cord without additional Shh signaling activation (Figs. 1F and 2E), we hypothesized that expression of Isl1–Lhx3 is sufficient to induce MN differentiation in ESCs bypassing the need of Shh signaling in the conventional method (7). To test this hypothesis, we subjected iMN-ESCs into three differentiation conditions: culturing embryoid bodies (EBs) with RA alone, RA plus Dox, and RA and a Shh agonist purmorphamine (Fig. 4B). iMN-ESCs cultured with RA alone differentiated to TuJ⁺/neurofilament (NF)⁺ neurons, but failed to form Hb9⁺ MNs (Fig. 4 C and D). Cotreatment of RA and a Shh agonist induced MN differentiation in ∼40% of TuJ⁺ neurons in an optimized condition (Fig. 4 C and D). Isl1–Lhx3 expression, without exogenous Shh activation, resulted in differentiation of 77% of TuJ⁺ neurons to Hb9⁺ MNs, which is substantially more efficient than Shh signaling activation (Fig. 4 C and D). One of essential characteristics of MNs is that they use acetylcholine as a neurotransmitter, unlike spinal interneurons that are glutamatergic or GABAergic (16). Immunostaining assays revealed that vesicular acetylcholine transporter (VAChT), a well-established marker for cholinergic neurons, is expressed only in Dox plus RA-treated ESC-derived neurons, but not in RA alone-treated ESC-derived neurons (Fig. 4E), indicating that Isl1–Lhx3-induced MNs are cholinergic. Together, our data demonstrate that Isl1–Lhx3 is capable of promoting robust differentiation of ESCs to MNs independently of exogenous activation of Shh signaling.

Isl1–Lhx3 Is More Specific than Shh Signaling in Driving Motor Neuron Differentiation.

Isl1–Lhx3 activates the MN-specific gene program without up-regulating V2-IN genes (Fig. 1 F and G), whereas Shh signaling also triggers specification of ventral interneurons (2). Thus, Isl1–Lhx3 expression likely drives MN differentiation more specifically than Shh signal in ESCs. To test this idea, we monitored induction of various neurotransmitter phenotypes in iMN-ESC–derived neurons using RA alone, RA plus Dox, or RA plus a Shh agonist. We analyzed the expression profile of cholinergic markers VAChT and choline acetyltransferase (ChAT), a glutamatergic neuronal marker vesicular glutamate transporter 2 (VGluT2), and a GABAergic neuronal marker GAD1, which encodes the γ-aminobutyric acid (GABA) synthesis enzyme GAD67 (Fig. 4F). RA treatment induced expression of VGluT2 and GAD1, but not VAChT and ChAT genes, consistent with its inability to induce MNs. Interestingly, Isl1–Lhx3 expression not only increased cholinergic gene expression but also suppressed VGluT2 and GAD1, compared with RA alone-treated sample. In contrast, RA/Shh-treated cells displayed high levels of GAD1 and VGluT2 as well as cholinergic markers, in agreement with the ability of Shh to specify multiple types of neurons. These results demonstrate that Isl1–Lhx3 promotes cholinergic MN differentiation in ESCs at the expense of glutamatergic and GABAergic neuronal cell types. Furthermore, our data indicate that Isl1–Lhx3 drives stem cells to differentiate into cholinergic MNs more specifically than Shh signal.

Isl1–Lhx3-Induced MNs Form Neuromuscular Junctions.

To test whether Isl1–Lhx3-induced MNs can form neuromuscular junctions with myotubes, we performed MN–myotube coculture assays. Either Shh-induced MNs or Isl1–Lhx3-induced MNs were dissociated and plated onto myotubes differentiated from C2C12 cells and cultured for 4 d. TuJ⁺ motor axons innervated the myotubes and triggered clustering of acetylcholine receptors on myotubes, as detected by patched α-bungarotoxin staining (Fig. 4G), indicating that Isl1–Lhx3-induced MNs establish neuromuscular junctions with muscle cells.

Isl1–Lhx3 Induces the MN Transcriptome and Suppresses the Developmental Programs for Spinal Interneurons.

To investigate how Isl1–Lhx3 affects the transcriptome during neuronal differentiation of ESCs in an unbiased genome-wide manner, we performed RNA-seq assays with RA alone-treated iMN-ESCs and RA/Dox-treated iMN-ESCs (Fig. 5A). To capture relatively early changes of the transcriptome by Isl1–Lhx3 expression, Dox was treated for 2 d. This high-throughput comprehensive analysis revealed that the levels of 444 genes were significantly changed by Isl1–Lhx3 expression, whereas many genes that are commonly expressed in neurons, such as β2-tubulin, are highly expressed in both samples (Dataset S1 and Fig. S7A). A total of 79% of the significantly changed genes exhibited induction and 21% displayed reduction, consistent with the notion that the MN hexamer functions as a transcription activator (5, 11). Isl1–Lhx3 highly induced the expression of prototypical markers of MNs (Fig. 5 B and C). Hb9, a direct target of the MN hexamer (11), was induced by 28-fold. ChAT, VAChT, and a high-affinity choline transporter CHT, were induced by 218-, 46-, and 203-fold, respectively, indicating that Isl1–Lhx3 directs the cholinergic differentiation. Lhx4, a LIM-HD factor functioning redundantly with Lhx3 for MN differentiation (17), two LIM-only proteins (LMOs) that are expressed in MNs, LMO1 and LMO4 (5, 18), Nkx6.2, a marker of the lateral motor column (LMC) type of MNs (16), and Isl2 are also significantly up-regulated. The induction of Hb9 and Isl1/2 proteins was confirmed using immunoblotting analyses (Fig. S7B). Isl1–Lhx3 expression also triggered the expression of a battery of genes involved in axon guidance and cell adhesion, many of which have been shown to control motor axon guidance, cell body positions, and MN survival (Fig. 5C). These genes include Met, Neuropilin1 (Nrp1), FGF9, Neurotrophin-3 (NTF3), PlexinA4 (PlxnA4), and Semaphorins (19–24). This induced gene profile suggests that Isl1–Lhx3 directs the expression of MN genes that are important for functional maturation and survival of MNs.

Fig. 5.

Isl1–Lhx3 induces a gene expression profile of MNs, while suppressing the developmental programs for spinal interneurons. (A) Experimental design to prepare samples for RNA-seq analyses. (B) Scatter plot to show RNA-seq results. x axis indicates the mean value of the normalized number of reads for each gene transcript in logarithm scale. y axis shows the log fold change between Dox-treated and control samples. Red spots represent the significantly induced genes by Isl1–Lhx3 expression, and blue spots indicate the significantly down-regulated genes by Isl1–Lhx3 expression. Cutoff is false discovery rate <10%. Arrows mark corresponding spots for several MN genes (red spots) and interneuron genes (blue spots). (C) List of significantly altered genes by Isl1–Lhx3 expression. y axis shows the log fold change between Dox-treated and control samples. Induced genes and repressed genes are marked in red and blue, respectively.

The expression of Sox2 and Pou3f3 (Brn1), transcription factors that establish neural progenitor identity (25), was suppressed by Dox treatment (Fig. 5C and Fig. S7B), suggesting that Isl1–Lhx3 facilitates differentiation of neural progenitors to neurons. Interestingly, Isl1–Lhx3 expression repressed class II progenitor factors, whose expression is suppressed by Shh signal, such as Irx3/5, Dbx1/2, and Pax3/6 (2, 3) (Fig. 5C). These progenitor factors mark the progenitor domains giving rise to spinal interneurons. Likewise, Isl1–Lhx3 expression also inhibited many transcription factors that determine the identity of spinal interneurons, such as Ptf1a, Olig3, Gsx1/2 (Gsh1/2), Msx3, FoxD3, Pou4f1 (Brn3A), Pax2/8, and bHLHe22 (bHLH5) (Fig. 5C) (16, 26). In addition, the expression of bone morphogenetic protein (BMP) and Wnt signaling molecules, which are important for the development of dorsal spinal cord (26), was repressed by Dox treatment (Fig. 5C). These results indicate that Isl1–Lhx3 expression inhibits the gene programs directing spinal interneuron differentiation.

Together, our RNA-seq data indicate that Isl1–Lhx3 expression is sufficient to induce a signature of MN transcriptome by triggering MN gene expression while suppressing interneuron differentiation.

Discussion

During CNS development, many transcription factors function in combination to specify greatly divergent cell types that constitute the functional CNS. Formation of cell type-specific transcription complexes serves as an important mechanism underlying the combinatorial actions of transcription factors. Lhx3 drives either MN fate or V2-IN fate, depending on the transcription complex that it forms in a given cellular context. When coexpressed with Isl1, Lhx3 forms the MN-hexamer that activates MN genes, whereas Lhx3 without Isl1 turns on V2-IN genes (Fig. 1A) (4, 5). This combinatorial action of Lhx3 and Isl1 in MN generation provides a useful model to test the strategy to differentiate stem cells to a specific neuronal type by expressing a defined set of transcription factors. Indeed, the coexpression of Isl1 and Lhx3, along with other transcription factors that induce neurogenesis, is capable of directing differentiation of ESCs, iPSCs, and fibroblasts to MNs (27, 28). However, a major challenge remains that each individual transcription factor has its own activity in cell lineage determination, distinct from a combinatorial function, and thus the expression ratio of the transcription factors should be optimized. Here we demonstrated that Isl1–Lhx3, a fusion molecule mimicking the MN hexamer, directs highly efficient and reproducible generation of MNs from ESCs. Our approach offers two critical advantages over the conventional method that uses combined exposure of ESCs to RA and Shh signals (7, 8). First, by using the MN-hexamer, a MN-specific transcription complex downstream of the multifunctional Shh signal, we were able to minimize differentiation of ESCs to mixed repertoire of neuronal cell types, which arise due to the broad spectrum of biological activities of Shh. Second, our strategy maximizes the specificity and efficiency of MN generation from stem cells by intrinsically maintaining the required 1:1 ratio of Lhx3 to Isl1. This suppresses erroneous formation of the V2-tetramer by excess Lhx3 protein, which can drive stem cells to an unwanted V2-IN pathway.

How are the proper stoichiometry of Isl1 and Lhx3 and selective formation of the MN-hexamer over the V2-tetramer ensured during MN differentiation in vivo? Selective degradation of Isl1, Lhx3, or NLI, depending on their status in complex formation, might contribute to achieving the proper stoichiometry. For instance, single-stranded DNA-binding proteins (SSBPs) and an E3 ubiquitin ligase RLIM/Rnf12 regulate the abundance of LIM-HD factors (29, 30). In addition, LMOs play important roles in regulating the assembly of LIM-HD complexes (5, 6, 31, 32). Specifically, LMO4 inhibits the assembly of V2-tetramer in MNs, thereby promoting MN-hexamer formation (5). These multiple layers of in vivo stoichiometric regulation might not fully operate in stem cells, and thus it is important to keep the correct ratio of transcription factors in generating a specific type of neurons from stem cells by expressing combinations of transcription factors.

Our RNA-seq data from Isl1–Lhx3-induced MNs provide important insights into the MN gene networks regulated by the MN-hexamer. Isl1–Lhx3 up-regulates the expression of Isl2 and Lhx4, which function redundantly with Isl1 and Lhx3, respectively, in forming the hexamer complex and inducing MN differentiation (32–34). The level of NLI was also increased by Isl1–Lhx3 expression. Thus, Isl1–Lhx3 expression leads to the higher levels of the MN-hexamer complexes, indicating a positive regulatory feedback. Isl1–Lhx3 also induces the expression of Hb9 and LMO4, which inhibit the transcription of V2-IN genes and the V2-tetramer assembly, respectively (5, 35, 36), reinforcing the previous model that the MN-hexamer actively blocks the V2-IN differentiation pathway (5). Once MNs are specified, Lhx3 expression is down-regulated in all MNs except MMCm-type MNs, which innervate axial musculature (17). Although it remains to be determined whether Isl1–Lhx3 is capable of directing generation of all subtypes of MNs, it is interesting to note that Isl1–Lhx3 induces not only the MMCm genes, such as Lhx3, Lhx4, and LMO4, but also the genes enriched in LMC-type MNs innervating limb muscles, such as Nkx6.2 (16). It is noteworthy that Isl1–Lhx3 induces a panel of genes that control MN cell body position, motor axonal trajectory, cholinergic neurotransmission, and MN survival, suggesting that the MN-hexamer controls a wide range of MN functions beyond initial cell-type specification. This result suggests a role of the MN-hexamer as a master regulator of the MN fate. The transcriptome analyses also uncovered that the MN-hexamer blocks the spinal interneuron differentiation programs by suppressing the expression of many key determinants of multiple interneuron fates. Whether the MN-hexamer actively suppresses the interneuron genes, for instance via induction of microRNAs that block the interneuron gene expression, or passively represses the interneuron programs by driving MN generation at the expense of interneurons, remains to be studied. In the future, it will also be interesting to investigate which of the induced and repressed genes are direct targets of the MN-hexamer complex.

In summary, our study demonstrates that the activation of an embryonic differentiation program using developmental transcription complex mimetic fusions can be explored as a strategy to direct stem cell differentiation to a specific cell type. This method is relatively free from the specificity issue associated with the application of widely acting inductive signals or with coexpression of individual transcription factors in cells. In addition, our approach of using inducible ESCs provides powerful model system to define target genes and downstream events of developmental transcription factors.

Materials and Methods

Details are provided in SI Materials and Methods. For differentiation assays, the EBs of iMN-ESCs were treated with RA (0.5 μM) alone for 2 d and then cultured without or with Dox (2 μg/mL) in the presence of RA for 2–3 d. In ovo electroporation, immunohistochemistry, gel-shift and ChIP assays were performed as described (4, 5, 11). RNA-seq libraries were prepared according to the Illumina TruSeq protocol and sequenced on an Illumina HiSeq 2000.