Author contributions: R.S.: generation, characterization, differentiation of mouse stem cells, real-time PCR, immunocytochemistry, fluorescent microscopy, imaging, cell counting, collection, analysis and interpretation of data, and writing of material and methods section; M.K.: stem cells culture, differentiation, immunocytochemistry, immunohistochemistry, confocal microscopy, and imaging; S.P.: electrophysiological experiments and analyses; Z.C.: in vivo implantations and surgery; S.M.: conception of the study, data interpretation, and manuscript writing; P.G.: data interpretation, financial support, and final approval of manuscript; V.E.G.: conception and design of the study, Western blots, data analysis and interpretation, financial support, manuscript writing, and final approval of manuscript. S.M. and P.G. contributed equally to this article.
Disclosure of potential conflicts of interest is found at the end of this article.
First published online in STEM CELLSEXPRESS December 7, 2012.
Directing differentiation of embryonic stem cells (ESCs) to specific neuronal subtype is critical for modeling disease pathology in vitro. An attractive means of action would be to combine regulatory differentiation factors and extrinsic inductive signals added to the culture medium. In this study, we have generated mature cerebellar granule neurons by combining a temporally controlled transient expression of Math1, a master gene in granule neuron differentiation, with inductive extrinsic factors involved in cerebellar development. Using a Tetracyclin-On transactivation system, we overexpressed Math1 at various stages of ESCs differentiation and found that the yield of progenitors was considerably increased when Math1 was induced during embryonic body stage. Math1 triggered expression of Mbh1 and Mbh2, two target genes directly involved in granule neuron precursor formation and strong expression of early cerebellar territory markers En1 and NeuroD1. Three weeks after induction, we observed a decrease in the number of glial cells and an increase in that of neurons albeit still immature. Combining Math1 induction with extrinsic factors specifically increased the number of neurons that expressed Pde1c, Zic1, and GABAα6R characteristic of mature granule neurons, formed “T-shaped” axons typical of granule neurons, and generated synaptic contacts and action potentials in vitro. Finally, in vivo implantation of Math1-induced progenitors into young adult mice resulted in cell migration and settling of newly generated neurons in the cerebellum. These results show that conditional induction of Math1 drives ESCs toward the cerebellar fate and indicate that acting on both intrinsic and extrinsic factors is a powerful means to modulate ESCs differentiation and maturation into a specific neuronal lineage. STEM CELLS2013;31:652–665
The directed differentiation of embryonic stem cells (ESCs) to specific neuronal subtype is critical for modeling disease pathology in vitro especially because of the strong link between neurodevelopment and neurodegeneration . A highly controllable in vitro system that recapitulates neurodevelopment would play an important role in distinguishing between cause and effect in disease progression and phenotype. In addition, directed differentiation of specific neurons in culture is very important for drug screens and therefore efficient ways of achieving this can substantially lower the cost of such screens . Specification of neuronal subtype involves interpreting extrinsic signals in terms of cell intrinsic mechanisms that initiate subtype-specific neuronal differentiation program . Toward this end, specific extrinsic signals operate at different points during neuronal differentiation to specify neuronal subtype identity in a sequential manner . The initial events of neuronal subtype specification involve specification of regional identity such as those that specify anterior-posterior and dorsal-ventral identity . Within this domain other molecules specify the different types of neurons that must be generated from this region. These further regional subdivisions involve gradients of extrinsic signals that are read out as expression of transcription factors by the cell. This has been especially well elucidated in the spinal cord with respect to the specification of the various classes of cells that are present from dorsal to ventral in the spinal cord . In addition, the specification of neuronal subtype involves combining subtype-specific information with those that are part of a pan-neuronal differentiation program such as cell cycle exit and acquisition of mature neuronal morphology . To understand subtype specification and mature neuronal differentiation, it is important to understand what part of the specification program becomes cell intrinsic and what part remains under the control of extrinsic cues.
The specification of granule neurons of the cerebellum is an interesting example of how neuronal subtype specification takes place. Unlike other glutamatergic neurons, the specification of granule neurons has two distinct steps. First, granule neuron progenitors (GNPs) are specified in the rhombic lip (RL) by factors such as BMP4 and FGF8 that induce the expression of the basic helix-loop-helix transcription factor Math1 in a domain that is positive for En1 and Pax6. Following this, a subset of Math1-positive precursors migrate out of the RL taking a dorsal route and cover the cerebellar anlage to populate the outer external granule layer (EGL) where they subsequently proliferate, differentiate, and tangentially migrate within the EGL. Finally, they migrate inward to form the mature granule neurons of the inner granule layer (IGL) . In the absence of Math1, granule neurons fail to form . Furthermore, extrinsic signals that control the proliferation and maturation of GNPs have been well characterized . Given this unique spatial temporal separation of GNPs development, our aim was to see how much of granule neuron identity in ESCs is specified by the cell intrinsic transcription factor Math1 and how much is dependent on external cues. Previous studies have established that markers of granule neurons can be induced in mouse and human ESCs by external signals [10–12]. However, the contribution of Math1 to the differentiation process is not known. In this study, we hypothesized that Math1 could be sufficient to specify ESCs to GNPs, thus promoting their differentiation and maturation upon subsequent addition of specific growth factors. To address this hypothesis, we created an inducible system that allowed us to induce Math1 transiently. We then allowed these cells to undergo terminal neuronal differentiation in the absence or presence of extrinsic signals and compared granule neuron specific as well as pan-neuronal gene expression to delineate the contribution of Math1 and that of extrinsic factors to the efficiency of deriving mature granule neuron from ESCs.
MATERIALS AND METHODS
Construction of Lentivector and Lentivirus Production
The insulated lentivector pLTET-Math1 was engineered by modification of pLTET-Luc (gift from Dr. Goodell) by site-directed mutagenesis (Stratagene, (Agilent technologies), Santa Clara, CA, www.agilent.com, 200522) to get two unique restriction sites (EcoR1/Cla1) flanking Luciferase gene. Math1 full-length cDNA was amplified from the cDNA IMAGE clone 4218223 (OriGene, Rockville, MD, www.origene.com, MC208159, NM_007500) with the 5′-EcoR1 and 3′-Cla1 overhangs. The amplicon and modified pLTET-Luc were digested with EcoR1 and Cla1. Following ligation of pLTET with Math1, the final vector was used to transform Stbl2 competent cells (Invitrogen, Life technologies: Carlsbad, CA (www.lifetechnologies.com) 10268-019). Positive clones were checked by digestion with EcoR1/Cla1 and direct sequencing. Production and titration of lentiviruses were performed by Sigma-Aldrich (St. Louis, MO, www.sigmaaldrich.com) and transduction-ready viral particles (1 × 108 TU/ml) were stored at −80°C.
Generation of the Math1-Inducible ESC Line
Mouse ES cells (ESD3) were modified in a two-step process to generate cell lines with Tet-ON system of gene expression, which conditionally expresses Math1 upon doxycycline exposure (Fig. 1A). In the first step, the undifferentiated (UD) cells were transduced with lentivirus containing pLTET-Math1. In this lentivector, the TRET promoter allows doxycycline-dependant control of the Math1 gene and the strong mammalian promoter EF1α constitutively drives the expression of DsRedEX, a variant of red fluorescent protein with increased solubility and faster maturation . Passage 11 cells were plated at a density of 8 × 104 per well in a gelatin-coated 96-well plate and allowed to grow for 20 hours. Following a 15-minute treatment with 8 μg/ml Polybrene (Sigma, (St. Louis, MO, www.sigmaaldrich.com) H9268) at 37°C, cells were transduced with the lentivirus at a multiplicity of infection of 10 for 24 hours. The DsRedEx-positive cells were first manually enriched and then clonally selected by serial dilution method to get pure TRET-Math1 clones.
In the second step, the TRET-Math1 clones were stably transfected with the modified form of the pTet-On Advanced regulator plasmid (Clontech, Mountain View, CA, www.clontech.com, 630930) encoding the transactivator protein rtTA2s-M2 under EF1α promoter. In the presence of doxycycline, rtTA2s-M2 binds the TRET promoter and activates transcription of the downstream gene. Around subconfluent passage 5, TRET-Math1 cells were transfected in 12-well culture plates using 4 μl Lipofectamine 2000 (Invitrogen, Carlsbad, CA, www.lifetechnologies.com, 11668-019) with 2 μg of linearized plasmid. Selection of successfully transfected cell was done with 400 μg/ml Geneticin (Invitrogen, Carlsbad, CA, www.lifetechnologies.com, 11811-031), followed by serial dilution for clonal selection. Many clones of Math1-inducible ESCs were obtained and assessed by real-time PCR for their ability to overexpress Math1 in UD stage upon Doxycycline treatment. Three different lines, namely clone 1.2, clone 1.3, and clone 2.8 were selected for experiments as they gave the maximum levels of expression.
Primary Culture of Neonatal Cerebellar Granule Neurons
Primary cultures of neonatal cerebellar granule neurons were obtained by following a procedure adapted from  Briefly, cerebella were isolated from P5 OF1 mouse brains and dissected under microscope to remove meninges. Tissue was treated in Phosphate buffer saline (PBS)/Glucose containing 0.25% trypsine EDTA (Sigma, St. Louis, MO, www.sigmaaldrich.com, T-4049) and 1 mg/ml DNase1 (Sigma, St. Louis, MO, www.sigmaaldrich.com, DN25) and dissociated manually by repeated pipetting with a fire polished glass pipette, centrifuged, then again dissociated in the presence of DNase1 and recentrifuged. Cell pellet was resuspended in β Mercapto Ethanol (BME) media containing 1% FBS, 45% Glucose, 1× Glutamax, 1× Pen-strep, and 1× B27 supplement (Gibco, Carlsbad, CA, www.lifetechnologies.com, 17054-044) and plated on wells coated with 30 μg/ml polyornithine (Sigma, St. Louis, MO, www.sigmaaldrich.com, P3655) and 5 μg/ml laminin (Sigma, St. Louis, MO, www.sigmaaldrich.com, L2020). Media were changed every alternate day and cells were fixed and used after 7–9 days.
Differentiation of the Math1-Inducible ESCs
For basic neuronal differentiation (referred as basal condition in Fig. 1B) , UDs were differentiated over a period of 31 days following the successive steps of early embryonic bodies (eEBs), late embryonic bodies (lEBs), neurospheres (NS), and ultimately final differentiation (FD) stage. To form EBs, passage 15–18 UDs were seeded at the density of 28 × 105 on 60 mm low adherent culture plates in ESIM media containing 10% FBS, 1× Glutamax, 1× Pen-strep, and 1× MEM-NEAA in DMEM either in the absence (eEBs) or presence (lEBs) of 1 mM retinoic acid (RA, Sigma, R-2625). Glass coverslips (14 mm) were coated with 30 μg/ml p-ornithine (Sigma, P-3655) for 4 hours, washed, UV treated, and incubated overnight at 37°C with 5 μg/ml laminin (Sigma, L-2020) made in PBS. The EBs were resuspended in NS media, plated on coated coverslips, and grown for 10 days. NS media consisted of 1× Glutamax, 1× Pen-strep, 1× ITS supplement (Gibco, Carlsbad, CA, www.lifetechnologies.com, 51500-056), and 20 ng/ml basic fibroblast growth factor (R&D Systems, Minneapolis, MN, www.RnDSystems.com, 234-FSE-025/CF) in DMEM-F12 (Gibco, Carlsbad, CA, www.lifetechnologies.com, 11330-032). For final stage, the NS were grown in media consisting of 1× Glutamax, 1× Pen-strep, 1× N2 supplement (Gibco, Carlsbad, CA, www.lifetechnologies.com, 17502-048), 1× B27 supplement (Gibco, Carlsbad, CA, www.lifetechnologies.com, 17054-044), and 1 μg/ml Laminin in Neurobasal-A media (Gibco, Carlsbad, CA, www.lifetechnologies.com, 10888-022) for 14 days.
To obtain a specific cerebellar granule cell differentiation, the cells were cultured as described above, along with the sequential addition of specific inducible factors, adapted with modifications from Salero and Hatten  (referred to as basal condition + growth factors cocktail in Fig. 5A). All factors except Wnt1 (Abcam, Cambridge, United Kingdom, www.abcam.com) were purchased from R&D systems (Minneapolis, MN, www.RnDSystems.com). Factor concentrations used were according to . Fresh media were replenished every alternating day for the respective stage.
Doxycycline Induction of Math1 Overexpression
Math1 induction was performed at various stages in the basal conditions either with or without the growth factors cocktail. Doxycycline (Sigma, St. Louis, MO, www.sigmaaldrich.com, D-9891) was reconstituted in sterile water, filter sterilized, and added in the stage-specific media to get the final concentration of 2 μg/ml. For 7 days experiments, doxycycline was replenished every alternating day together with fresh ESIM media alone or with supplements according to the experiment. To stop the induction, old media were removed, cells given two 1× PBS washes, and finally replenished with the next stage media.
RNA Extraction and Relative Quantification of Gene Expression by Real-Time PCR
Total RNA was extracted from the cells collected at the end of UD, eEB, lEB, NS, and FD stages using TRI reagent (Sigma, St. Louis, MO, www.sigmaaldrich.com, T9424). Diethylpyrocarbonate DEPC (0.1%) (Sigma, St. Louis, MO, www.sigmaaldrich.com, D5758) treated autoclaved water was used for all the molecular work. RNA was quantitated using Nanodrop1000 spectrophotometer, treated with DNase (Roche, Basel, Switzerland, www.roche.com, 04716728001), and again quantitated. Two hundred and fifty nanograms of total treated RNA was used to synthesize cDNA (Bio-Rad, Hercules, CA, www.bio-rad.com, 170-8891) in 20 μl reaction mix according to the manufacturer's protocol. Quality of cDNA was validated by semiquantitative PCR (Taq DNA polymerase: Roche, Basel, Switzerland, www.roche.com, 11647679001; dNTP set: Amersham Biosciences, GE Healthcare Life Sciences, Little Chalfont, United Kingdom, www.gehealthcare.com, 27-2035-01) for the amplification of housekeeping gene Hypoxanthine-guanine phosphoribosyl transferase (HPRT), which was chosen as the internal control for real-time PCR on the basis of its stable expression during all the stages of stem cell differentiation. PCR products were checked on a 2% agarose gel.
Primers (Supporting Information Table S1) were either designed using program PRIMER3 (http://frodo.wi.mit.edu/primer3/) or chosen from Primer Bank site (http://pga.mgh.harvard.edu/primerbank/), and conditions standardized by semiquantitative PCR. iQ-SYBRGreen Supermix (2×) (Bio-Rad, Hercules, CA, www.bio-rad.com, 170-8885) was used for real-time PCR with the following conditions: 95°C 3 minutes/45 cycles (95°C 30 seconds, annealing temperature 30 seconds, 72°C 35 seconds). Dissociation curve was generated for checking the amplification specificity. The standard curves for both internal control and gene of interest were generated to determine the PCR efficiency. All the samples were run in triplicate and negative controls without cDNA were run each time together with the samples for both internal control (HPRT) and gene of interest. The data were analyzed by comparative CT method  to determine fold differences in expression of target genes with respect to the internal control.
Immunocytochemistry and Immunofluorescence Microscopy
Cultures were fixed for 20 minutes with 4% paraformaldehyde. For immunocytochemistry, the cells were permeabilized with 0.3% Triton X-100 for 15 minutes, washed, blocked for 1 hour with 4% Bovine serum albumin (BSA)/10% serum in PBS, and incubated with primary antibody overnight at 4°C. Following washing, cells were treated with secondary antibody (1:1,000 dilution) for 1 hour in dark at room temperature. For immunocytochemistry on EBs, EBs were collected in a 1.5 ml microcentrifuge tube and allowed to settle properly by gravitational pull. The media were discarded and EBs were given one 1× PBS wash. The PBS was removed and EBs were incubated in 4% paraformaldehyde for 20 minutes at room temperature, after which four 1× PBS washes were given. Whole EBs were stained by immunofluorescence similarly in the microcentrifuge tube. For costainings, the second antibody was added and processed as for the first. Cells were finally mounted with mounting media containing 4′, 6-diamidino-2-phenylindole (DAPI) (Vector labs, Burlingame, CA, www.vectorlabs.com, H-1200). Specificity of the stains was checked using nonrelevant primary antibodies that do not react with any antigen (mouse IgG1 isotype control [Abcam, Cambridge, United Kingdom, www.abcam.com, ab126026] and rabbit IgG isotype control [Abcam, Cambridge, United Kingdom, www.abcam.com, ab27478]) as shown in Supporting Information Figure 7. Antibodies and dilutions used are listed in Supporting Information Table S2. All images were captured on Zeiss microscope IMAGER.Z1, using the Apotome imaging system coupled to AxioCam MRm and the Axiovision Rel.4.8 software.
Cell Implantation and Immunohistochemistry
Clone 1.2 cells were induced with doxycycline during EB stage in the presence of growth factors, as described in Figure 5A and collected at NS stage. Cells were trypsinized and made single-cell suspension in PBS. P60 adult OF1 mice (n = 5) under isoflurane anesthesia were mounted on a stereotaxic frame. Injections were made with glass micropipettes implanted into the cerebellum (lambda, −2 mm; lateral, 0 mm; depth, −2.2 mm). A volume of 2 μl containing 4 × 105 cells was injected at a rate of 0.134 μl per minute. The micropipette was left in place for an additional 3 minutes to reduce backflow. One week after implantation, mice were deeply anesthetized and transcardially perfused with 0.9% NaCl for 1 minute followed by cold 4% paraformaldehyde. Brains were then dissected out, postfixed with same fixative overnight at 4°C, cryoprotected with 30% sucrose in 0.12 M phosphate buffer, and frozen at −45°C in isopentane for 3 minutes before sectioning. Sagittal floating sections (30 μm) were prepared using a freezing microtome and collected in PBS. Implanted cells contain the DsRed gene under the control of the constitutive EF1α-promoter (Fig. 1A) and although natural fluorescence disappeared upon neuronal differentiation, DsRed-expressing cells could still be revealed using an anti-DsRed antibody. For immunostaining, sections were first treated with citrate buffer at 80°C for antigen retrieval. After several washes in PBS, they were incubated for 1 hour at room temperature in blocking buffer (2% BSA, 5% serum, and 0.3% Triton in PBS), followed by primary antibody incubation in similar buffer with 0.1% Triton at 4°C overnight. Then sections were rinsed in PBS and incubated for 1 hour in dark with the secondary antibody. After several washes in PBS, sections were counterstained with DAPI for 1 minute and mounted with antifading medium (DAKO, Glostrup, Denmark, www.dako.com).
Cell countings were performed using ImageJ and Adobe Photoshop. For counting cells positive for a marker with respect to total number of cells (DAPI positive), 25–35 random fields (depending on the cell density, covering the maximum coverslip) were captured under ×40 and counted; for finding cells positive for a marker with respect to Tuj1, MAP2, or Nestin, seven random fields were captured on the coverslip and counted. For EBs imaging and cell counting, z-stack images at an interval of 1 μm for each EB were taken at ×20 magnification. For cell counting one image out of the stack of around 14 images of each EB containing defined and countable DAPI-positive nucleus and properly focused Ki-67 or Clv caspase 2-positive cells was selected and the cells were manually counted with the image-J software. All the counts were repeated for six independent experiments.
Western Blot Analyses
For Math1 detection, control tissue samples were dissected out from P4 mice cerebella, sonicated in ice-cold lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% NP40, 0.25% sodium deoxycholate, and protease inhibitors), and centrifuged. Clones 1.2, 1.3, and 2.8 were used for producing EBs for 7 days in control and Dox-induced conditions. Proteins were extracted from EBs in the same buffer. Protein concentrations were determined according to Bradford . Lysates were denatured, separated by SDS-PAGE, and transferred to Polyvinylidene fluoride (PVDF) membranes (Millipore, Temecula, CA, www.millipore.com, RPN303F). Membranes were probed with either anti-Math1 (Acris, Herford, Germany, www.acris-antibodies.com, AP00308PU-N dilution 1:200) or anti-Actin antibodies (Millipore, Temecula, CA, www.millipore.com, MAB1501, dilution 1:10,000) and signal was detected by enhanced chemiluminescence.
In Vitro Electrophysiology
Experiments were performed on stem cell cultures plated on 40 mm culture dishes. Cells were continuously superfused in recording solution which comprised (mM): NaCl, 124; KCl, 3; NaHCO3, 26; NaH2PO4, 1.25; CaCl2, 2; MgSO4, 1; glucose, 10. Visually guided, whole-cell recordings were obtained at room temperature from the soma of neuronal-like cells using patch electrodes (4–6 MOhms) that contained (mM): KCl, 140; HEPES, 10; NaCl, 8; EGTA, 0.5; Mg-ATP, 4; Na-GTP, 0.3. Voltage was recorded on-line using current-clamp techniques. Depolarizations by current injection were performed to induce action potential. Data were stored and analyzed using the LTP Program [17, 18].
Quantitative data are expressed as mean ± SEM for each treatment group. Results were compared using the Mann-Whitney comparison test Mann-Whitney U test (GraphPad Prism Software). A two-tailed p < .05 was considered as significant.
Math1 Expression Can Be Induced in a Temporally Controlled Manner During ESCs Differentiation
Math1 is developmentally regulated in GNPs and therefore we developed an inducible gene expression system whereby the expression of this transcription factor could be temporarily regulated. To this end, we generated clone 1.2, a stable mouse ESC line capable of inducing Math1 expression in a reversible manner through the control of a stably expressed doxycycline (Dox)-regulated reverse transcriptional transactivator (rtTA2s-M2) (Fig. 1A). Such improved TET (Tetracycline) regulators have been used for conditional gene expression in several ESCs models and have proved to be highly efficient without altering the pluripotency of the UDs . Consistent with this, UD ESCs from our clone 1.2 cultures were found to express extensively the pluripotency-related markers SSEA1 and Oct4, suggesting that, like native ESD3 cells, they have kept their pluripotency (Supporting Information Fig. S1). Next, we quantified the ability of our transgene expression system to induce Math1 expression at different stages of ESC differentiation. The differentiation protocol giving the five different stages at which Dox was added to induce transgene expression is schematized (basal condition, Fig. 1B) and the detailed methods for achieving this are as described in Materials and Methods. Following addition of Dox for 48 hours at the beginning of each of these five stages, we either collected the cells pellets directly after induction or cells were washed to remove Dox and let the differentiation reach the final stage. Dox addition was able to induce transgene expression at all stages of differentiation and removal of Dox led to a return of Math1 expression that were comparable to that of preinduction levels (Fig. 1C) thus showing that we had a temporally regulated highly inducible transgene system. Quantitative PCR showed that induction of Math1 using our system was most efficient at the EBs stage and less efficient at UD and FD stage (Fig. 1D). Immunocytochemistry and Western blot analysis on late EBs stage cells confirmed that the induction led to increased Math1 protein (Fig. 1E, 1F). Together, these data indicate that we derived an ESC line able to drive Math1 overexpression upon Dox induction and that conditional Math1 induction is most efficient during EBs stage.
Mbh1 and Mbh2 are Induced by Math1 in a Dose-Dependent Manner and are Maintained During ESCs Differentiation
To test whether the induction of Math1 resulted in the activation of genes that are important for the specification of granule neuron identity, we looked at the Bar-class homeobox genes Mbh1 and Mbh2 that are the downstream targets of Math1 [19, 20]. We found a significant stimulation of Mbh1 and Mbh2 when Math1 was induced at early EBs (eEBs) and late EBs (lEBs). Taking advantage of the fact that induction of the transgene was most efficient at EBs, less efficient at neurosphere (NS), and not very efficient at other stages, we looked at the dose-dependent effect of Math1 expression on Mbh1 and Mbh2. Results showed that the induction of Mbh1 and Mbh2 was dependent on the levels of Math1 (Figs. 1D, Fig. 2A).
We next tested whether the expression of Mbh1 and Mbh2 persists once induced by Math1. Dox was added during the whole EB stage (eEBs+lEBs) and expression of Mbh1 and Mbh2 was analyzed. From seventh day to the end of the differentiation process, both Mbh1 and Mbh2 expressions were increased by two- to threefolds (Fig. 2B). Thus, although the induction of Math1 was transient, the downstream targets Mbh1 and Mbh2 stayed on throughout the differentiation suggesting that once the downstream program is activated it is independent of Math1 expression. Immunocytochemistry experiments using antibodies against Mbh1 and Mbh2, 17 days after initiation of induction confirmed that induced neural precursors were much more immunoreactive than noninduced controls (Fig. 2C). Together, these data indicate that Math1 induction at EBs stage is sufficient to trigger a persistent activation of target genes required for specification of granule neurons identity, during ESC differentiation.
Transient Expression of Math1 During the EBs Stage Leads to Increased Neuronal and Decreased Glial Differentiation
To determine the effects of Math1 transient expression on neuronal differentiation, we induced Math1 during EBs stage and analyzed the cells at the end of FD stage. Quantitative PCR showed an increase in the neuronal markers Tuj1 and MAP2, and a decrease in glial fibrillary acidic protein (GFAP) and Olig2 upon Math1 induction (Dox) (Fig. 3A). Immunocytochemistry confirmed an increase in the number of Tuj1- and MAP2-positive cells and a decrease in GFAP-positive cells (Fig. 3B, 3C). We did not detect a decrease in the number of Olig2-positive cells possibly because very few cells express Olig2 under our differentiation conditions. These data were confirmed in two additional independent clones, clone 1.3 and clone 2.8 (Supporting Information Fig. S5). Interestingly, the total number of neural cells remained unchanged in induced and control conditions (73.60% and 74.51% of total cells, respectively, Fig. 3B). To check that these results were not due to some “pro-neural” characteristics of our clones, we compared their neural properties to those of native ESD3 and ESB6 cells, two unmodified mouse ESC lines. In all cell lines, expression levels of the neural markers Tuj1 (32.2%–37.9%), GFAP (33.6%–38.6%), and Olig2 (5.7%–9.7%) were measured in comparable ranges upon differentiation in basal conditions (Supporting Information Fig. S4A). The proportion of neural cells versus non-neural cells was almost identical between clone 1.2 and its parental ESD3 cell line while ESB6 cells showed a higher trend in favor of neural differentiation (Supporting Information Fig. S4A, S4B). Furthermore, although Ki67 was slightly increased at the end of EBs stage, Math1 expression did not result in an increased cell proliferation at NS stage (Supporting Information Fig. S2C) or later. Furthermore, there was no increase in apoptotic cell death at any stage tested, as assessed by cleaved caspase-3 immunostaining (Supporting Information Fig. S2C). Taken together, these observations suggest that transient activation of Math1 during an early-stage of ESC differentiation results in Tuj1-positive cells being produced at the expense of GFAP-positive cells.
Transient Expression of Math1 Results in the Differentiation of Neurons that Express Granule Cell Markers in the Correct Temporal Sequence
Since transient Math1 expression resulted in an increase in neuronal differentiation, we asked whether it was sufficient to activate the expression of markers of granule neuron subtype. Quantitative PCR showed an increase in the mRNA levels of Zic1 and Pde1c that are early markers of granule neuron subtype which remain expressed throughout differentiation, of TAG1, expressed in migrating postmitotic granule neurons and finally of GABAα6r expressed specifically in mature granule neurons (Fig. 4A). To check whether indeed GABAα6r was expressed in mature neurons, we compared the expression of GABAα6r in Tuj1-expressing and in MAP2-expressing neurons after terminal differentiation. Results showed that there was an increase in the number of MAP2-positive cells expressing GABAα6r, many with higher intensity and that the number of Tuj1-positive cells expressing GABAα6r remained the same as in noninduced condition (Fig. 4B, 4C). However, Zic1 and Pde1c had increased expression in Tuj1-positive cells that is an earlier marker for neuronal differentiation (Fig. 4B, 4C). Although not apparent at RNA level, Zic2 also showed an increase in induced condition in Tuj1-positive cells (Fig. 4A–4C). Thus, transient Math1 expression resulted in an increase in the expression of GABAα6r only in mature neurons, whereas earlier markers were found increased in immature neurons suggesting that the timing of expression of the various subtype-specific markers was recapitulated during ESC differentiation. Consistent with this idea, the expression of engrailed (En1) and NeuroD1, specific of early progenitors, was found increased at the late EB stage but not at later stages (Supporting Information Figs. S2B, S3B). Finally, induction of Math1 had no effect on the number of cells expressing the mature pan-neuronal marker MAP2 (Fig. 4B), indicating that the increased number of MAP2 neurons shown in Figure 3 directly resulted from the increase of Tuj1-expressing cells upon Math1 activation. Similar data were obtained upon induction of the two other clones 1.3 and 2.8 (Supporting Information Fig. S6). Taken together, these results suggest that transient induction of Math1 not only causes more cells to become neuronal but also is sufficient to direct cells into neurons that express granule cell markers in the correct temporal sequence.
Maturation of Granule Neurons Requires Extrinsic Signals
Although induction of Math1 resulted in a significant increase in the number of mature neurons expressing MAP2 and GABAα6r, 40% of the Tuj1-expressing cells was still negative for MAP2 (Fig. 4B) and 65% was found positive for Pax6, a marker of proliferative granule cells (data not shown and Fig. 4A). To determine the contribution of extrinsic cues to the differentiation process, we induced transgene expression in the presence or absence of stage-specific extrinsic cues that had previously been shown to result in granule neuron differentiation from ESCs  (growth factors cocktail, Fig. 5A). Very interestingly, expression of GABAα6r and Zic1 was significantly enhanced by the transient induction of Math1 even in the presence of factors (Fig. 5B). However, colabeling of cells with Tuj1 and MAP2 revealed that the vast majority of the Tuj1-expressing neurons were also positive for the mature neuronal marker MAP2 when factors were added (92%) as compared to basal conditions (58%) irrespective of Math1 induction (Fig. 5C). Therefore, while Math1 activates subtype-specific markers, it is not sufficient for pan-neuronal differentiation and maturation. Moreover, cell counting showed that the number of neurons that expressed GABAα6r was maximal when Math1 was induced in the presence of factors (66% vs. Tuj1 and 73% vs. MAP2, Fig. 5C). The number of Tuj1-neurons that expressed Zic1 was also significantly increased when Math1 induction and extrinsic factors were combined (86%) as compared to all other conditions (Fig. 5C). A similar trend was observed with Pde1c (Supporting Information Fig. S3B). These increases were not due to a greater number of neurons since Math1 activation in the presence of factors had no effect on Tuj1 expression and did not change the number of Tuj1-positive cells that expressed MAP2 (Fig. 5C and Supporting Information Fig. S3A). By contrast, the number of cells that expressed markers of nongranule fate like serotonin (5-HT) and tyrosine hydroxylase was significantly decreased by the combined action of Math1 and factors. Similarly, no effect was seen on MyosinVIIa (Fig. 6B), a marker of mature sensory hair cells, which require Math1 expression outside the nervous system . Similar data were obtained with the two other clones (Supporting Information Fig. S6 and not shown). Together, these data indicate that maturation of granule neurons requires extrinsic signals and that the combined action of Math1 transient expression and extrinsic factors has a stronger effect on the specific stimulation of granule neurons markers than that triggered by either Math1 expression or factors addition.
Expression of Granule Cell Markers Is Coincident with Mature Granule Neuron Morphology and Functional Neurons
We examined the morphology of the differentiated cultures and performed electrophysiological measurements in vitro. In order to demonstrate functionality, we recorded in the condition that gave the maximum percentage of neurons positive for GABAα6r and Zic1 (Fig. 6A). In these clone 1.2 cultures, the majority of the neurons harbored a small ovoid cell body with a T-shaped axon, a signature of differentiated granule cells, established synapses, and expressed synaptophysin (Fig. 6A). These cells were recorded in current-clamp mode to assess whether they were able to generate action potential. We first investigated if at resting potential these cells were able to generate spontaneous action potential and then if they were able to generate firing of action potentials under sustained controlled depolarization. Two categories of cells were visually selected. The first type presented long T-shaped processes and ovoid cell body, typical of granule cells. The second type presented only short processes and large-cell body. Action potentials were never recorded in patched cells of this phenotype. By contrast, spontaneous action potentials as well as sustained firing under controlled depolarization were systematically recorded in the first type of cells (n = 8) (Fig. 6C). Similar results were obtained with clone 1.3 and clone 2.8 (Supporting Information Fig. S6). This suggests that the differentiated granule neurons were indeed functional neurons capable of generating action potentials.
To determine whether the newly generated neurons could also integrate in the cerebellum in vivo, we implanted Dox-induced clone 1.2-derived progenitors at NS stage into the cerebellum of young adult mice (P60) and tracked down the cells in sagittal cerebellar sections 1 week after implantation. In all mice processed (n = 5), DsRed-positive cells could be detected in cerebellar lobules close to the injection site, mainly in the molecular layer (Fig. 7A). Colabeling with an anti-Tuj1 antibody showed that many of the DsRed-positive cells strongly expressed Tuj1 (Fig. 7B), displayed a neuronal shape, and started to colonize the granule cell layer (Fig. 7C, 7D). A few implanted neurons located in the granule cell layer were found positive for the GABAα6receptor (Fig. 7E, 7F). Taken together, these data suggest that Dox-induced progenitors can give rise to functional neurons in vitro and in vivo.
The data presented in this study indicate that conditional induction of Math1 is sufficient to drive ESCs into granule neuron lineage, that extrinsic signals are required for granule neuron maturation, and that the combination of Math1 temporally controlled expression with these extrinsic cues efficiently increases the proportion of mature cerebellar granule neurons. Given the importance of developing in vitro models of development of specific neuronal types, it is critical to understand how cell intrinsic and extrinsic cues interact with each other to specify neuronal subtype in vitro. We have delineated the contribution of Math1 as a cell intrinsic determinant of subtype identity in directing ES differentiation into neurons versus the requirement of other extrinsic cues and parallel signaling pathways in the specification of GNPs. The question is important because specification of granule neurons and their proliferation and differentiation are spatially and temporally separated.
Overexpression and deletion studies have shown that correct levels of Math1 are required for granule neuron differentiation making it obligatory to use an inducible system [8, 22]. Using an inducible system, we were able to determine that the most appropriate window for Math1 transient expression corresponded to the EBs stage that mimics signaling centers . Our first major finding is that transient expression of Math1 in the absence of more specific cues for cerebellar patterning is sufficient to induce markers of granule neurons during RA-induced differentiation of ESCs into neurons. Our second major result is the increased efficiency of reprogramming when we use transient gene expression system along with extrinsic cues. We have shown here that combining a temporally controlled expression of Math1 along with extrinsic signals leads to 73% of neurons acquiring a granule fate. This approach has recently brought encouraging results when trying to derive difficult to get neuronal subtypes such as was shown with serotonin [23–25].
In terms of how closely ESC differentiation follows what happens in vivo, we first show that transient induction of Math1 is enough for the persistent activation of the downstream target genes Mbh1 and Mbh2 that are essential for granule neuron specification . Increase in Mbh1 and Mbh2 in turn led to an increase in its downstream target TAG1 showing the sustained downstream activation of GNPs specification pathway by transient Math1 induction. Thus, we saw increased expression of TAG1 as well as Pde1c, a specific marker for the granule neuron lineage while sensory hair cells markers like Myosin VIIa were not favored in our system.
NeuroD and Pax6 are two other downstream targets of Math1 . Previous studies have suggested that NeuroD is an early marker for neuronal differentiation [27, 28]. Our result showing increased NeuroD at early stages but not at later stages of granule cell differentiation is consistent with the fact that downstream targets are activated in the correct temporal sequence by transient Math1 induction. Pax6-like En1 is expressed in the RL from which Math1-positive cells are generated [20, 29], and in addition, Pax6 is expressed at low levels in proliferating granule neurons and higher levels in differentiating cells . Consistent with this, our results also show an expression of Pax6 maintained throughout differentiation.
Proneural genes activate neural versus glial differentiation program and in this context Math1 has been shown to not only induce the expression of downstream genes that are important in granule neuron specification but also is thought to increase the expression of Tuj1 through a parallel pathway [20, 31]. Math1 expression in ESC differentiation recapitulated both aspects of proneural gene function by increasing the expression of pan-neuronal and neuronal subtype-specific marker. The increase in Tuj1 expression resulted in a decrease in GFAP labeled cells but not in a decrease of non-neuronal cells. This is consistent with studies that have shown that proneural genes including Math1 bias neuronal differentiation at the expense of glial differentiation [32–34].
Regulation of neuronal progenitor proliferation and cell cycle exit is an important component of the pan-neuronal differentiation program. In our study, we observed a slight but significant increase in cell proliferation at late EBs stage due to Math1 induction and subsequent upregulation of Zic1. This is consistent with what has been seen in vivo where Math1 regulates the Notch pathway and in its absence there is decreased cell proliferation [35, 36] and Zic1/2 also plays a role in cell proliferation .
En1 marks the entire mesencephalon/rhombomere one territory [38, 39] and its expression is induced by extrinsic cues . During development, En1 is also coexpressed by Math1-positive granule neurons early in development but is not expressed by mature granule neurons of the IGL. Interestingly, En1 was upregulated even in the absence of extrinsic cues by Math1 induction. Furthermore, this increased En1 expression was seen only at early and not at later stages as predicted from developmental data.
Zic1 and Zic2 are two transcription factors that are expressed in the RL and have overlapping patterns of development and cooperate in cerebellar development . Like En1, we could induce Zic1 and Zic2 by transient expression of Math1 without the addition of extrinsic factors. This shows that even in the absence of extrinsic cues, Math1 activation is sufficient for induction of En1, Zic1, and Zic2 in the correct temporal sequence. However, Zic1 can also be induced by BMP6 and BMP7  and it binds to the upstream region of Math1 to repress its transcription . We found that even in the absence of Math1 induction, Zic1 levels increased significantly upon factors addition. This goes along with the observation that Zic1 may be a parallel pathway involved in granule neuron proliferation which could be under the control of extrinsic factors and crosstalk with Math1 .
Math1 induction did not impact neuronal maturation as the number of neurons expressing MAP2 remained stable but resulted in a significant increase in the number of MAP2-positive neurons expressing GABAα6r, a specific marker for mature granule neurons (+16%). The temporal regulation of GABAα6r is particularly interesting as its expression depends on very precise temporal dynamics of two transcription factors NF1 and REST . The fact that we see an upregulation of GABAα6r with Math1 induction in the absence of specific extrinsic cues suggests that this full program of neuronal maturation can occur in the absence of any specific extrinsic cues.
A recent study showed that several of Math1 target genes are those that are involved in extrinsic signals . Extrinsic factors increased neuronal maturation and thus led to a 33% increase in MAP2 positive and a significant increase in GABAα6r in Tuj1 (+35%) and MAP2-positive cells (+15%). The net result of these two effects was that transient expression of Math1 along with extrinsic factors resulted in 73% of the MAP2-positive cells expressing the mature granule neuron marker GABAα6r that is more than can be achieved by either condition alone (+31.7%). Similarly, our data indicate that Pde1c is increased by either Math1 transient expression (+19%) or extrinsic factors addition (+15%), but that the combination of both effects results in even more Pde1c-positive neurons (+31.7%, Supporting Information Figs. S3, S6). Two studies have reported successful differentiation of mouse ESCs into cerebellar granule neurons [10, 12]. In the study by Su and colleagues, Math1-positive cerebellar progenitors were produced by the combination of rostral CNS induction and subsequent BMP4/Wnt3a treatment. With regard to the efficiency, yields of obtained granule progenitors have been provided based mainly on Zic1 expression (around 60% of total cells), and therefore it is difficult to estimate the percentage yield of mature granule neurons. In this study, combining Math1 transient expression with extrinsic factors generated 86% of Zic1-expressing neurons. Moreover, differentiation into granule cell lineage was achieved using a coculture system in the presence of neonatal cerebellar granule neurons from mouse, and therefore the study was not intended to directly address the contribution of Math1 induction versus extrinsic cues. Similarly, in the study by Salero and Hatten, progenitors and mature cerebellar neurons were obtained in vitro using successive defined media that mimic the sequential differentiation steps of cerebellar differentiation in vivo. However, yields reported for final stage used differentiated neurons obtained with media conditioned by purified cerebellar granule neurons or glia.
In conclusion, we were able to show that transient Math1 expression at EBs stage was enough to drive these cells into granule neuron lineage in the absence of other specific inducing factors that have been used previously. However, the addition of extrinsic cues activated other granule neuron-specific genes with greater efficiency and increased neuronal maturation and thus overall led to an increase in granule neuron differentiation. This result thus identifies factors that will now help us to achieve even greater homogeneity of granule neuron subtype. In future, more use of such a strategy should enable us to identify critical pathways that extrinsic cues induce to generate any kind of neuronal subtype in vitro and pave the way to advance our understanding of basic mechanisms of neuronal subtype specification and pathophysiological mechanisms underlying several neural diseases.
We are grateful to Prof. Margaret Goodell, Baylor College of Medicine, Texas, for providing us the lentivirus pLTET-Luc, which was modified and used in the construction of cell line, to Sophie Lebon (INSERM U676, Paris, France) for her help and advice in molecular biology steps, and to Cécile Martel for invaluable support. This study was supported by the Institut National pour la Santé et la Recherche Médicale (INSERM, France), the Centre National de la Recherche Scientifique (CNRS, France), Denis Diderot University (Paris7, France) and National Brain Research Centre (Haryana, India), and by grants from IFCPAR/CEFIPRA (project N°3803-3), Department of Biotechnology (project N°BT/PR11653/MED/31/602008), French National Research Agency (project ANR-09-GENO-007), the Princesse Grâce de Monaco Foundation, and the Roger de Spoelberch Foundation. S.P. was supported by grant from the Medical Research Council.
DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST
The authors declare that no conflict of interest exists.