Neurogenin 2 regulates progenitor cell-cycle progression and Purkinje cell dendritogenesis in cerebellar development

Cerebellar neurogenesis is compartmentalized, with progenitors located in the cerebellar ventricular zone (cbVZ) giving rise to all GABAergic neurons, including Purkinje cells (PCs), and rhombic lip (RL) precursors producing glutamatergic neurons, including granule cells (GCs) (Carletti and Rossi, 2008). The mechanisms by which diverse GABAergic populations diverge from their common progenitors are still unclear. All GABAergic cell types in the cbVZ share lineage relationships (Mathis and Nicolas, 2003) and require pancreatic transcription factor 1a (Ptf1a) (Sellick et al., 2004; Hoshino et al., 2005; Pascual et al., 2007). In the mouse, a population of Pax2⁺ progenitors appears in the cbVZ around embryonic day (E) 12 and spreads into the prospective white matter (Maricich and Herrup, 1999). All cerebellar inhibitory interneurons derive between E12 and postnatal day (P) 15 from those Pax2⁺ precursors, and their subtype specification is entirely dependent on environmental cues (Leto et al., 2006; Glassmann et al., 2009; Leto et al., 2009).

Unlike GABAergic interneurons, PCs are born between E11 and E13 as their progenitors exit the cell cycle and start to leave the cbVZ, migrating to the cortex, forming the PC plate, and projecting an axon and an intricate dendritic arbor (reviewed by Sotelo and Rossi, 2011). The regulatory cascades that preside over PC fate specification and maturation are only partially dissected. Further work is required to identify the factors that regulate PC fate choice, cell-cycle progression and exit, differentiation (Zhao et al., 2007), migration (Goldowitz et al., 1997; Rice et al., 1998; Trommsdorff et al., 1999; Park and Curran, 2008), survival (e.g. Croci et al., 2006; Croci et al., 2011), axonogenesis (Sillitoe et al., 2009; Miyata et al., 2010) and dendritogenesis (Boukhtouche et al., 2006; Poulain et al., 2008).

Proneural genes play a pivotal role in neuronal development, triggering a generic neuronal differentiation program and contributing to the specification of distinct neuronal phenotypes throughout the central nervous system (CNS) (reviewed by Bertrand et al., 2002; Ross et al., 2003). Three proneural genes are expressed in the cbVZ (Zordan et al., 2008; reviewed by Consalez et al., 2011), namely Ascl1 (Mash1), Neurog1 (Ngn1) and Neurog2 (Ngn2), and represent suitable candidate regulators of PC development. They are expressed in partially overlapping domains in the Ptf1a⁺ cbVZ, possibly contributing to its regionalization. Ascl1⁺ progenitors contribute to PCs and GABAergic interneurons. Ascl1 null mice lose all GABAergic cerebellar interneurons, whereas PCs are unaffected (Kim et al., 2008; Grimaldi et al., 2009; Sudarov et al., 2011). Neurog1⁺ progenitors give rise to inhibitory cortical interneurons and to some PCs (Kim et al., 2008; Lundell et al., 2009). However, nothing is known regarding the fate of Neurog2⁺ progenitors or the role(s) played by Neurog2 in cerebellar neurogenesis. Neurog2 is a direct downstream target of the Ptf1a-Rbpj transcription complex (Henke et al., 2009).

In the present work, through a combination of genetic inducible fate mapping (Joyner and Zervas, 2006), targeted deletion and overexpression, we: (1) illustrate the roles played by Neurog2 in regulating cell-cycle progression and neuronal output in the cerebellar VZ; (2) identify the cerebellar lineages, primarily PCs, that originate from Neurog2⁺ progenitors; and (3) describe the non-redundant function played by Neurog2 in the context of PC dendritogenesis.

The experimental plan was designed in agreement with the stipulations of the San Raffaele Institutional Animal Care and Use Committee. Surgical procedures were performed on mice anesthetized with Avertin (Sigma, St Louis, MO, USA).

A BAC clone spanning the mouse Neurog2 locus was modified by homologous recombination in E. coli (Copeland et al., 2001) by inserting the ERT2iCreERT2 construct (Casanova et al., 2002) into exon 2, replacing the entire Neurog2 coding sequence. A plasmid was generated from the recombinant BAC clone by gap repair (Liu et al., 2003) and used to electroporate mouse embryonic stem (ES) cells (strain V6.4). Homologous recombinant ES clones were identified by Southern blotting and injected into C57BL/6 blastocysts that were transferred to CD1 pseudopregnant females.

All experiments were conducted on congenic embryos (generation N₈) obtained by backcrossing heterozygotes into the C57BL/6 strain. Neurog2^CreER/+ knock-in mice were crossed with Rosa26^lacZ/lacZ (Soriano, 1999) or Rosa26^YFP/YFP (Srinivas et al., 2001) mice. Because null mutants on the C57BL/6 background die around birth (Fode et al., 1998), postnatal stages were analyzed in F2 mutants obtained by outcrossing C57BL/6 heterozygotes with outbred CD1 mice, then intercrossing the F1 progeny. Four percent of null mutants survived past P15.

Digoxygenin-labeled riboprobes were transcribed from linearized Neurog2 and CreER plasmids. Hybridizations were performed as described (Croci et al., 2006).

Pregnant dams were injected intraperitoneally (i.p.) (50 μg/g body weight) with the following synthetic thymidine analogs: 5-bromo-2-deoxyuridine (BrdU; Sigma), 5-chloro-2-deoxyuridine (CldU; Sigma), 5-iodo-2-deoxyuridine (IdU; Sigma) and 5-ethynyl-2′-deoxyuridine (Edu; Invitrogen, Carlsbad, CA, USA). For cumulative EdU labeling, timed females were injected i.p. every 3 hours for up to 18 hours. BrdU/IdU and BrdU/CldU were detected with a mouse anti-BrdU (1:100; BD Biosciences, Franklin Lakes, NJ, USA) and a rat anti-BrdU (1:100; ABD Serotec, Kidlington, UK) antibody, respectively. Immunohistochemistry was performed as described (Garel et al., 1997). EdU incorporation was revealed using the Click-iT EdU Imaging Kit (Invitrogen).

Cell-cycle progression in the cerebellum was analyzed as described (Takahashi et al., 1993). Cell-cycle phases in Neurog2^CreER/+ and Neurog2^CreER/CreER were compared using a t-test for paired samples.

Tissues were prepared as described (Croci et al., 2006). Mouse primary antibodies: calbindin, parvalbumin (1:1500; Swant, Bellinzona, Switzerland); SMI32 (1:250; Sternberger Monoclonals, Lutherville, MD, USA); NeuN (1:500; Millipore, Billerica, CA, USA); S100β (1:1000; Sigma); ER [1:100; Santa Cruz Biotechnology (SCB), Santa Cruz, CA, USA]; TuJ1 (1:300; Covance, Princeton, NJ, USA); MAP2 (1:300; BD Biosciences); and RC2 (1:20; DSHB, Iowa City, IO, USA). Rabbit primary antibodies: β-galactosidase (β-gal; 1:20,000; Cappel, Westchester, PA, USA); GFP (1:700; Invitrogen); calretinin (1:2500; Swant); neurogranin (1:250; Millipore); Olig2, PH3 (1:500; Millipore); ER (1:100; Thermo Fisher Scientific, Waltham, MA, USA); Dcx (1:100; SCB); NICD (1:100; Cell Signaling, Beverly, MA, USA); Pax2 (1:300; Invitrogen); GABA (1:2500; Sigma); and Vglut2 (1:100; Synaptic Systems, Goettingen, Germany). Chicken primary antibody: GFP (1:500; Abcam, Cambridge, UK). Goat primary antibody: Neurog2 (1:200; SCB).

For epitope unmasking, cryosections were boiled in 10 mM citrate buffer (pH 6) for 5 minutes. The Tyramide Signal Amplification Kit (Perkin Elmer, Waltham, MA, USA) was used to detect β-gal, Neurog2 and NICD.

Neurog2^CreER/+ and Rosa26^lacZ/lacZ mice were crossed. Pregnant females were injected i.p. with tamoxifen (TM; Sigma; 3-5 mg/25 g body weight). Progeny were harvested at E18, given to foster mothers and sacrificed at P30. Sections from mice receiving TM at E11.25 (TM^E11.25) (N=15,089 cells/3 cases), TM^E11.75 (N=5507 cells/2 cases) and TM^E12.25 (N=6919 cells/2 cases) were immunostained for β-gal and specific cerebellar markers, and lacZ-expressing cells from three series were quantitated.

Donor cells were isolated and dissociated at E12.5 from Neurog2^CreER/+; Rosa26^lacZ/+ embryos (TM^E11.75) and grafted to wild-type P2 or P5 cerebella (Jankovski et al., 1996; Carletti et al., 2002). A single-cell suspension (1 μl at final concentration 5×10⁴ cells/μl), obtained by mechanical dissociation of cerebellar primordia, was injected using a glass capillary. Grafted cells were recognized by β-gal expression. Phenotypes were scored as described (Carletti et al., 2002; Leto et al., 2006). Results from the same experimental set were pooled.

Wholemount lacZ staining was performed as previously described (Corradi et al., 2003).

Total RNA was extracted from embryos at different stages using the RNeasy Micro Kit (Qiagen). Total RNA (1 μg) was retro-transcribed using first-strand cDNA M-MLV reverse transcriptase (Invitrogen) and random primers (Invitrogen) and 2 μl of 1:10 cDNA was used for each reaction. RT-qPCR was performed using the LightCycler 480 SYBR Green I Master Mix (Roche, Mannheim, Germany). Each gene was analyzed in triplicate. Data analysis was performed with the ΔΔCt method. All RNA levels were normalized to Gapdh. The following primers (5′-3′) were used: Gapdh-F, TGAAGCAGGCATCTGAGGG; Gapdh-R, CGAAGGTGGAAGAGTGGGAG; Neurog1-F, ATCACCACTCTCTGACCC; Neurog1-R, GAGGAAGAAAGTATTGATGTTGCCTTA; Ascl1-F, AGGAACAAGAG CTGCTGGAC; Ascl1-R, TCGTTGGCGAGAAACACTAA; Map2-F, AGCACTTGATTGGGAAGCACT; Map2-R, CGCCAATTCAAGGAAGTTGT; Sclip-F, AGACTGCAGCTCCCAGACAC; Sclip-R, AGCGACAGCACAGATAACTCC; Rora-F, CGAGCTCCAGCCGAGGTATCTCA; Rora-R, GCCCTTGCAGCCTTCACACGTAA.

The flagNeurog2 cDNA (kind gift of M. Goetz, Helmholtz Center, Munich, Germany) was subcloned into pCCL.sin.cPPT.SV40polyA.eGFP.min-CMV.hPGK.ΔLNGFR.WPRE (kind gift of L. Naldini, TIGET-San Raffaele, Milan Italy) replacing the ΔLNGFR sequence (Amendola et al., 2005).

VSV pseudotyped third-generation LVs were produced as described (Follenzi et al., 2000; Muzio et al., 2010). Expression titers, as determined on HEK293 cells by FACS analysis (FACSCalibur, BD Biosciences), were 3.6×10⁹ TU/ml (control) and 2.4×10⁹ TU/ml (flagNeurog2). Western blots of infected HEK293 lysates were hybridized for flagNeurog2 (flag M2, Sigma) and GFP (Molecular Probes, Eugene, OR, USA).

Cntrl-GFP and Neurog2-GFP LVs were mixed with 0.01% Fast Green (Sigma) and injected in utero into the E12.5 fourth ventricle. Pregnant dams received a single CldU pulse before LV injection, and single EdU and IdU pulses 24 and 48 hours, respectively, thereafter. Brains were collected at E17.5.

For loss-of-function (LOF) experiments, E17.5 cerebellar sections from Neurog2^CreERT2/+; Rosa26^YFP/YFP and Neurog2^{CreERT2/CreERT2}; Rosa26^YFP/YFP embryos were analyzed. YFP⁺ neurons were scored from random fields (5±1 neurons per field). In gain-of-function (GOF) experiments, E12.5 embryos were infected with vectors encoding either Neurog2 and GFP, or GFP alone. GFP⁺ neurons were scored from random fields (5±2 neurons per field). In both GOF and LOF experiments, only neurons exhibiting detectable neurites were included in the analysis.

Sholl analysis was performed with ImageJ (NIH), using the Concentric Circles plug-in (rsbweb.nih.gov/ij/plugins/concentric-circles.html) and Sholl Analysis plug-ins (Ghosh laboratory, biology.ucsd.edu). Measurements were performed within seven concentric circles centered on the cell soma and spaced by 10 μm. Intersections within each concentric circle were counted, and average branch numbers were compared by means of a t-test. Results are presented as mean ± s.e.m.

E13.5 cerebellar primordia were dissected from wild-type and null embryos (n=3) to obtain total mRNA preparations (RNeasy Micro Kit, Qiagen). RNA (500 ng) was used for cRNA generation (Illumina Total Prep Kit, Ambion). Hybridization of MouseWG-6_V2 arrays (Illumina, Eindhoven, The Netherlands) was performed as recommended by the manufacturer. Image files were acquired on a BeadXpress scanner (Illumina). After data normalization and filtering, differentially expressed genes (significance threshold P<0.01; fold changes <0.7 or >1.7) were included in the analysis. Unsupervised hierarchical clustering showed that all transcriptional signatures correctly separated controls from nulls. The array contained 45,281 probes representing transcripts listed in the NCBI-Refseq, Riken and Meebo databases. Raw data were background-subtracted and cubic-spine-normalized using GenomeStudio GX software (Illumina) and further processed using MultiExperiment Viewer 4.7.1. (Saeed et al., 2006).

All experiments (including controls) were conducted in triplicate. Except where noted otherwise, statistical analysis was conducted using the unpaired Student’s t-test and P<0.05 was used as the threshold for statistical significance.

The Neurog2::ER^T2iCreER^T2 mouse line, which expresses tamoxifen (TM)-inducible Cre recombinase under the control of Neurog2 cis-acting sequences (see Materials and methods; hereafter referred to as Neurog2^CreER), was produced by homologous recombination. As illustrated in Fig. 1A, exon 2, which is the only Neurog2 coding exon, was replaced by the improved Cre recombinase (iCre) (Shimshek et al., 2002) fused in frame N- and C-terminally with the modified human estrogen receptor (ER) termed T2 (Indra et al., 1999). This fusion protein (Casanova et al., 2002) displays high recombination efficiency in the presence of TM and very low background activity in its absence (Jullien et al., 2008).

Neurog2^CreER heterozygotes were bred into Rosa26 Cre reporter lines, harboring either lacZ (Soriano, 1999) or YFP (Srinivas et al., 2001). Upon administration of TM, the Cre recombinase activates reporter gene expression (Fig. 1B). Reporter activity becomes detectable 12 hours after TM injection and increases thereafter, reaching a plateau between 18 and 24 hours (Fig. 1E,F and supplementary material Fig. S1A-C), in agreement with previous reports (Hayashi and McMahon, 2002).

The Neurog2 transcript is detectable in the cerebellar primordium from E10.75 to E13.5, peaking between E11.5 and E12.5 (Zordan et al., 2008). We compared the distribution of Neurog2 and CreER transcripts in the cerebellar primordium of Neurog2^CreER/+ embryos by in situ hybridization and showed that the CreER and Neurog2 territories overlap (Fig. 1C,D). To demonstrate colocalization of Neurog2 and CreER proteins, we immunostained E12.5 Neurog2^CreER/+cerebellar sections. Our results indicate that Neurog2 and ER colocalize in an overwhelming majority of cells (Fig. 1H). Thus, in this study, ER staining will be considered synonymous with Neurog2 distribution.

In the embryonic telencephalon, Neurog2 is expressed in a salt-and-pepper pattern due to Notch-mediated lateral inhibition. The expression of Neurog2 and of Notch effectors (such as Hes genes) in neural progenitors are inversely correlated (Shimojo et al., 2008). However, no information is available as regards the embryonic cerebellum. We compared the distributions of ER and Notch intracellular domain (NICD) in the cbVZ of E11.5 and E12.5 Neurog2^CreER/+ embryos. Our results indicate that, at both stages, ER expression is virtually restricted to NICD-negative progenitors (Fig. 1I,J).

Basic helix-loop-helix (bHLH) proteins are known to function as master regulators of mammalian neurogenesis, linking neuronal differentiation to cell-cycle regulation (Farah et al., 2000). To determine whether Neurog2 in the cbVZ is expressed by cycling progenitors or by postmitotic neurons, we immunostained for ER and the early postmitotic marker Tuj1 (Tubb3 – Mouse Genome Informatics) (supplementary material Fig. S2A-D). Only a small minority of Neurog2⁺ cells stained positive for TuJ1 [8.6±4.2% (mean ± s.d.); plotted in supplementary material Fig. S2E], indicating that Neurog2 is almost exclusively expressed by dividing cells.

As they progress through the cell cycle, apical progenitor nuclei undergo interkinetic migration along the apical-basal axis, with mitoses occurring at the apical surface and S phase taking place at a basal location, with apical-to-basal nuclear migration in G1 phase and the reverse in G2 phase (Sauer and Walker, 1959). Restricted Neurog2 distribution in the cbVZ (Zordan et al., 2008) suggests that, in cerebellar neurogenesis, Neurog2 transcription might be cell cycle dependent, and that lack of Neurog2 might hamper cell-cycle progression. However, no information is available to date as regards cell-cycle kinetics in the cbVZ. We analyzed cell-cycle parameters in cbVZ progenitors, then applied our results to the evaluation of cell-cycle kinetics in Neurog2 nulls.

Cumulative S-phase labeling was achieved by pulsing E12.5 embryos every 3 hours with EdU. The labeling index (ratio of labeled nuclei to total nuclei, LI) was plotted at each time point. The growth fraction (ratio of proliferating to total cells, GF), overall cell-cycle length (T_C) and S-phase length (T_S) were calculated as described (Nowakowski et al., 1989). T_C–T_S corresponds to the interpolated value at which LI reaches GF. GF, T_C and T_S were calculated based on the value of the y intercept [(T_S/T_C)×GF] and on the time at which the asymptote was reached (T_C–T_S, i.e. the time required to label the complete GF) (Takahashi et al., 1995). At E12.5, T_C equaled 13.9 hours, with an 85% GF and 9 hours to reach GF (T_C–T_S). T_G2+M length was estimated as 2 hours (Takahashi et al., 1993); the length of G1 was calculated by subtracting T_S+T_G2+M from T_C. The resulting cell-cycle durations were: T_G1, 7.0 hours; T_S, 4.9 hours; T_G2+M, 2 hours (Table 1).

These results were used to map the cell-cycle stage(s) at which Neurog2 is expressed in E12.5 Neurog2^CreER/+ embryos (Britz et al., 2006; Shimojo et al., 2008; Sunabori et al., 2008). Sections were immunostained for ER and EdU. Neurog2⁺ S-phase progenitors (ER⁺ EdU⁺/total ER⁺) were labeled through a single 30-minute pulse (EdU^30′) prior to sacrifice; 4.9±3.1% of Neurog2⁺ progenitors were EdU positive (Fig. 2A,A′). For S+G2+M phase, embryos were labeled by EdU^3hours; 8.4±6% of ER⁺ progenitors were EdU⁺ (Fig. 2B,B′), with hardly any double-positive cells located at the apical surface (M phase). Accordingly, only 0.8±0.5% stained double positive for ER and phospho-histone H3 (PH3) (Fig. 2C,C′). Progenitors located in early and late G1 were labeled by EdU^8hours and EdU^14hours; 24.4±9.8% of ER⁺ progenitors stained positive for EdU using EdU^8hours (Fig. 2D,D′), whereas 63.4±10.6% of ER⁺ progenitors colocalized with EdU using EdU^14hours (Fig. 2E,E′), indicating a specific enrichment of Neurog2 expression in late G1-phase progenitors. Although similar studies have been conducted in other brain regions (Miyata et al., 2004; Britz et al., 2006; Yi et al., 2008), this is an entirely novel finding in the context of cerebellar neurogenesis.

The restriction point (RP) is a cell-cycle checkpoint that occurs after early G1. Cells passing through RP re-enter S phase, while the others enter G0 (the G1/G0 transition) and start differentiating (Takahashi et al., 1999). To determine whether Neurog2⁺ progenitors in late G1 are poised to leave the cell cycle, E11.75 Neurog2^CreER/+; Rosa26^YFP/+ embryos were TM treated and cerebella collected at E12.5, after an EdU^30′ pulse. All Neurog2⁺ YFP⁺ progenitors examined stained positive for the postmitotic markers TuJ1 and Dcx (supplementary material Fig. S3A,B). Hardly any cells (<0.2%) were double positive for YFP and dividing radial glia markers, such as RC2 (Ifaprc2 – Mouse Genome Informatics) (supplementary material Fig. S3C), Blbp (Fabp7 – Mouse Genome Informatics) or nestin (not shown), and virtually no tagged progenitors (<0.1%) colocalized with either PH3 or EdU (supplementary material Fig. S3D-F). In the Neurog2^CreER/+; Rosa26^lacZ/+ line, reporter activity is detectable 12 hours after TM administration (supplementary material Fig. S1A). In the E12.5 cbVZ, the cell cycle lasts ∼14 hours. If any cells were to re-enter S phase due to high CreER protein levels, then reporter activity would be detected in the cbVZ. Our results indicate that progenitors that become robustly positive for Neurog2 leave the cell cycle almost immediately and undertake differentiation and radial migration. In other words, in the cbVZ, Neurog2 is overwhelmingly expressed by G1 progenitors that are bound to withdraw from the cycle.

Cumulative S-phase labeling revealed a significant decrease in the rate of cell-cycle progression in the Neurog2 null cbVZ. At E12.5, T_C was increased by 2.9 hours (Fig. 3A,G), and T_G1 and T_S by 2 and 0.9 hours, respectively (P=0.02, Table 1). CreER distribution was unaffected in S, G2 and M, with only sparse ER⁺ progenitors detected at these stages (plotted in Fig. 3F). However, striking differences became apparent in G1. Specifically, Neurog2^CreER/CreER exhibited an increase in ER⁺ progenitors in early G1 versus controls (EdU^8hours, 60.4±9.1% versus 24.4±9.8%; P<0.01; Fig. 3B′,C′,F), with a corresponding decrease in the late G1 fraction (EdU^14hours, 31.3±15.5% versus 63.4±9.8%; P<0.01; Fig. 3D′,E′,F). Accordingly, ER⁺Neurog2 null nuclei were detected in a more apical position relative to their wild-type counterparts (supplementary material Fig. S4), suggesting that Neurog2 null progenitors stall in their progression through G1. An alternative scenario is that Neurog2 knockout elicits a feedback upregulation of the Neurog2 promoter in early G1 progenitors. At odds with this possibility, the total number of ER⁺ progenitors is not significantly increased in Neurog2 null cerebella. Thus, Neurog2 appears to affect cell-cycle kinetics by modifying the pace of early to late G1 transition (Fig. 3G). As in controls, YFP⁺Neurog2 null E12.5 progenitors (TM^E11.75) do not re-enter the cell cycle and instead become postmitotic, indicating that the delay in G1 progression does not ultimately prevent cell-cycle exit (not shown).

Next we asked whether other proneural genes undergo a compensatory upregulation in the Neurog2 null cbVZ. RT-qPCR revealed a significant upregulation of both Ascl1 and Neurog1 (Fig. 3H). These results are in keeping with recent findings showing that Ascl1 rescues the stall of neurogenesis observed in the developing retina of Neurog2 null mutants (Hufnagel et al., 2010).

To confirm the evidence obtained in Neurog2 nulls, we took a GOF approach. The fourth ventricle of E12.5 wild-type embryos was injected with a lentivirus (Amendola et al., 2005) encoding Neurog2 and GFP (hereafter referred to as Neurog2-GFP LV) or with a control GFP vector. Embryos were pulsed with CldU, EdU and IdU (1 hour before, 24 and 48 hours after injection, respectively) and E17.5 cerebella were immunostained for GFP and EdU. The LI profile of Neurog2-GFP LV-infected cells is represented for each time point in Fig. 4D, revealing that the GFP⁺ EdU⁺ population was strikingly reduced in Neurog2-GFP LV-infected cells as compared with controls (Fig. 4A,B). Thus, Neurog2 overexpression accelerates cell-cycle withdrawal, exhausting the proliferating pool.

Although Neurog2 is expressed in a portion of the Ptf1a expression domain, the contribution given by Neurog2⁺ progenitors to the mature cerebellum is unknown. We used the Neurog2^CreER/+; Rosa26^lacZ/+ line to tag Neurog2⁺ progenitors and follow their eventual fate in the adult cerebellum. Embryos were TM treated at three different stages spanning the Neurog2 expression window (TM^E11.25, TM^E11.75, TM^E12.25). β-gal-tagged cells were visualized on P30 and identified via cerebellar neuron markers (Fig. 5A-P).

At all induction stages, Neurog2⁺ precursors yielded three GABAergic neuron classes: (1) a vast majority of PCs (Fig. 5A-I); (2) SMI32⁺ (SMI32 detects non-phosphorylated neurofilament H) presumptive nucleo-olivary neurons (Leto et al., 2006) (Fig. 5L,M); and (3) calretinin⁺ (calbindin 2 – Mouse Genome Informatics) dorsal cochlear nucleus (DCN) interneurons (Fig. 5N). At E12.5, a small subset of Neurog2⁺ β-gal⁺ progenitors (TM^E11.75) also expresses Pax2 (Fig. 5Q,S), a marker of all cerebellar GABAergic interneuron precursors (Maricich and Herrup, 1999; Leto et al., 2006; Carletti et al., 2008). Cortical GABAergic interneuron types (Golgi, Lugaro, basket and stellate) were virtually absent from the tagged neuronal pool (<1%). Finally, Neurog2⁺ progenitors gave rise to a small percentage of S100β⁺ Bergmann glial cells (Fig. 5O), possibly as products of symmetric, self-consuming radial glial division. Distributions of cerebellar cell types for each stage are plotted in Fig. 5P.

Fate mapping suggests that Neurog2⁺ progenitors are lineage restricted, producing only a defined subset of cerebellar neurons, primarily PCs. When heterochronically transplanted to the postnatal cerebellum, E12.5 cbVZ progenitors generate the entire repertoire of cerebellar neurons (Carletti et al., 2002; Carletti and Rossi, 2008; Williams et al., 2008). To determine whether E12.5 Neurog2/β-gal-tagged progenitors (TM^E11.75) are cell-autonomously restricted to their endogenous fates or acquire late-generated interneuron identities, we grafted them to wild-type cerebella at P2 and P5 and examined the resulting phenotypic repertoires at P30. Embryonic donor cells either reaggregated into mini-cerebellar structures (Fig. 6A-D) or engrafted as isolated scattered elements (Fig. 6E-O) (Jankovski et al., 1996). At both stages, grafted (Fig. 6P,Q) donor β-gal⁺ cells differentiated exclusively into PCs (83.4±7.6%; Fig. 6A-J) or DCN GABAergic neurons (14.0±3.9%; Fig. 6N,O). β-gal⁺ cells did not incorporate BrdU pulsed immediately after transplantation (Fig. 6K-M), confirming that they are postmitotic. This indicates that Neurog2⁺ progenitors are, to a large extent, cell-autonomously committed to adopt PC and DCN neuron fates.

To determine whether Neurog2 is required for cbVZ progenitors to adopt a PC fate, we compared the identity of β-gal-tagged neurons in Neurog2^CreER/+; Rosa26^lacZ/+ versus Neurog2^CreER/CreER; Rosa26^lacZ/+ embryos (TM^E11.75) at E17.5, using FoxP2 for PCs and DCN neurons, Pax2 for all types of GABAergic interneurons, Pax6 for GC precursors and S100β for astroglia. No relevant trans-specification was observed in null mutant cerebella: Neurog2 null neurons correctly differentiated into FoxP2⁺ PCs. Likewise, LV overexpression of Neurog2 did not affect the number of PCs in the E17.5 cerebellum (data not shown). Thus, Neurog2 may either play a minor role in PC fate specification or act redundantly with other proneural genes.

Although pure-bred Neurog2 nulls die around birth (Fode et al., 1998; Fode et al., 2000; Hand et al., 2005), some outbred mutants survive until P15 (Andersson et al., 2006). In surviving mutants, null cerebella are decreased in volume (by 20%; Fig. 7) and exhibit a lobulation defect (Fig. 7A,C-C″) both in the vermis (lobules VI, VII and IX) and hemispheres (fusion of crus 1 and simplex lobule). The thickness of all cortical layers is also reduced (Fig. 7D; see Discussion) and PC number is decreased by 20%, whereas PC density is unaffected (Fig. 7E). TUNEL staining conducted at E14.5 and E17.5 was unchanged (data not shown). As regards PC subtypes, Neurog2 expression does not match the distribution of zebrin II (Aldoc – Mouse Genome Informatics), which is the main marker of parasagittal domains in the cerebellum (supplementary material Fig. S5). Further studies are required to determine whether Neurog2 deletion affects any specific PC subpopulations.

Next, we assessed whether Neurog2 regulates PC differentiation. Embryos (TM^E11.75) were analyzed at E17.5, at the end of PC migration. At this stage, PCs adopt a bipolar elongated shape, develop a branched apical dendrite and start generating new and thinner processes radiating from the cell body (Sotelo and Rossi, 2011). Neurog2 null PCs migrate normally underneath the external granular layer. However, they exhibit a striking reduction in neurite branching. Compared with control PCs, which are fusiform in shape and have extensively branched processes, Neurog2 null PCs are round, with simple and stunted neurites, if any, as shown by YFP and MAP2 (Mtap2 – Mouse Genome Informatics) co-immunostaining (Fig. 8A-B′).

As illustrated in Fig. 8C and plotted in Fig. 8D, less than 1% of mutant PCs exhibited a branched apical dendrite, while 47% carried a single stunted process stemming from the cell soma, and 52% lacked any surface processes. Conversely, all control PCs had an elongated dendrite featuring 1, 2 or 3 or more main branches (20, 54 or 25%, respectively). Accordingly, Neurog2-overexpressing PCs exhibited a dramatic increase in process branching as compared with controls (Fig. 8H-J). Changes in dendritic branching were quantitated by Sholl analysis (Ristanovic et al., 2006), revealing a significant reduction in both the total length and average number of dendrites in Neurog2 null PCs (Fig. 8G), whereas Neurog2-overexpressing PCs exhibited a sharp increase in both the length and number of dendritic processes (Fig. 8K). In the cerebellar primordium, Neurog2 expression is detectable by RT-qPCR until E15 (Ct=32.5±0.07).

The expression of several established genetic regulators of early PC dendritic growth was examined in wild-type and null embryos by RT-qPCR of E14.5 cerebellar RNAs. Three genes were found to be downregulated in the Neurog2 null cerebellum: Rora and Sclip (Stmn3 – Mouse Genome Informatics), which are known to control the early stages of PC dendritogenesis (Boukhtouche et al., 2006; Poulain and Sobel, 2007), and Map2, which encodes a cytoskeletal protein involved in dendritic outgrowth and stabilization (Fig. 8L). All three genes are expressed in early postmitotic neurons, suggesting that Neurog2 might act upstream of them as a master regulator of the intrinsic phase of PC dendritogenesis. While the above genes were downregulated in Neurog2 null embryos at E14.5, the expression of many others was found to vary in a microarray analysis of wild-type versus null cerebella at E13.5, a stage spanning the end of PC neurogenesis and the onset of PC differentiation (supplementary material Fig. S6). An in-depth analysis of these results is in progress and is beyond the scope of the present paper.

At the end of development, PCs feature monoplanar, highly branched dendrites and are characterized by their apical dendritic tree, which stems from a single vertically oriented primary dendrite (Fig. 9A,C). We analyzed the long-term effects of Neurog2 knockout on PC dendritogenesis in outbred Neurog2 mutants. At P10, the dendrites of null mutant PCs, as stained with SMI32 antibody, are still dramatically stunted. Many Neurog2 null PCs have thinned primary dendrites and a poorly branched arbor (Fig. 9B,D). Climbing fiber terminals, as stained with Vglut2 (Slc17a6 – Mouse Genome Informatics) antibody, impinge upon the main dendrites of both wild-type and mutant neurons (Fig. 9E,F), indicating that the defective dendritic development does not abolish PC innervation.

Here, we describe a new knock-in line and its analysis aimed at characterizing the roles of Neurog2 in cerebellar development. This model, which permits the long-term stage-specific tagging of Neurog2⁺ progenitors and of their adult derivatives, will facilitate developmental and transplantation studies, making it possible to assess the ability of Neurog2-tagged progenitors to integrate into the postnatal and adult CNS.

For the first time, we show that Neurog2 is expressed by dividing NICD-negative progenitors in the cbVZ. Thymidine analog incorporation indicates that the early to late G1 transition is slower in Neurog2 nulls and, accordingly, that Neurog2 overexpression accelerates cell-cycle withdrawal, depleting the progenitor pool. Thus, Neurog2 acts astride the G1 restriction checkpoint to affect cell-cycle progression in the cbVZ.

GIFM shows that all progenitors robustly expressing Neurog2 exit the cell cycle and upregulate early differentiation markers within 12-24 hours. Thus, although Neurog2 expression may oscillate repeatedly at low levels in the cbVZ [as shown in telencephalic neurogenesis (Shimojo et al., 2008)], when it reaches robust expression levels it dictates cell-cycle withdrawal. Interestingly, although Neurog2 is expressed exclusively in daughter cells fated to exit the cell cycle, cellular progeny bound for cell-cycle re-entry are also delayed, suggesting non-cell-autonomous regulation.

Although Neurog2 null progenitors are delayed in cell-cycle progression, with a decrease in neuronal output, Neurog2 knockout does not completely abolish PC generation, possibly owing to compensatory Neurog1 and Ascl1 upregulation. As a result of the decreased PC progenitor output, the entire cerebellar cortex is thinner in null mutants. This is not surprising because, starting shortly before birth, PCs promote clonal expansion of GC progenitors (e.g. Wechsler-Reya and Scott, 1999). Thus, a reduction in PC numbers or a decrease in mitogen secretion by PCs is expected to translate into fewer GCs and a thinner molecular layer occupied by GC axons. Taken together, our findings establish that Neurog2 plays an important role in PC development by regulating cell-cycle progression and neuronal output, a novel finding in cerebellar development.

Long-term GIFM indicates that Neurog2⁺ progenitors give rise mainly to PCs and also to GABAergic DCN projection neurons, i.e. to the entirety of cerebellar GABAergic projection neurons. Moreover, the combination of Neurog2 and Pax2 expression at E12.5 represents the earliest known genetic signature of cerebellar DCN interneurons, an important foundation for future studies of cerebellar neurogenesis. Likewise, grafting experiments indicate that E12.5 Neurog2⁺ progenitors are committed to a restricted set of neuronal identities. When grafted to the postnatal cerebellum, E12.5 Neurog2⁺ precursors are cell-autonomously fated to PC or DCN neuron identities and they fail to acquire other interneuron fates. This indicates that Neurog2⁺ cbVZ precursors are irreversibly committed to a positional identity code and mature by unfolding a cell-intrinsic ontogenetic program.

Is Neurog2 part of the combinatorial code that bestows a PC identity on cbVZ progenitors? The lack of transdifferentiation observed in genetically tagged Neurog2 null progenitors suggests otherwise. Yet, Neurog2 might act redundantly with other proneural genes in this context, warranting the generation and analysis of double (Neurog1/Neurog2) and triple (Neurog1/Neurog2/Ascl1) mutants to establish the roles, if any, of proneural genes in PC fate specification. Alternatively, PC identity might be conferred on multipotent Ptf1a⁺ progenitors by other, as yet unidentified, signals, which most likely intervene before M phase of the last round of cell division. A major novel finding of this study is that Ptf1a⁺ progenitors are split into Neurog2^– and Neurog2⁺ precursors, the latter giving rise to mature PCs and DCN neurons. Our findings establish Neurog2 as the earliest conclusive marker of this restricted subset of GABAergic lineages. Further analysis is required to determine whether Neurog2, by dictating cell-cycle exit, controls the binary choice between early and late GABAergic interneuron fates.

Once they arrive at the PC plate, immature PCs lose the bipolar morphology of migratory neurons and undergo complex modifications of their neurites (Miyata et al., 2010) before extending apical dendrites, from which the mature dendritic tree will arise. Thus, dendritic differentiation involves both regressive and growth changes, which follow a characteristic differentiative sequence (Ramón y Cajal, 1911; Armengol and Sotelo, 1991). Whereas the processes and mechanisms underlying the late phases of PC dendritogenesis have been extensively investigated (reviewed by Sotelo and Rossi, 2011), much less is known about the initial stages of dendrite formation and the relevant regulatory cascades that control the early steps of dendritic differentiation.

Phosphorylated Neurog2 controls apical dendrite morphology in neocortical pyramidal neurons by promoting the outgrowth of a polarized leading process during the initiation of radial migration (Hand et al., 2005). Our results indicate that Neurog2 is required for the initial dendritic rearrangement at the end of PC migration (Armengol and Sotelo, 1991; Miyata et al., 2010). Again, both LOF and GOF results converge to implicate Neurog2 as a key early regulator of PC dendritogenesis. Neurog2 null PC precursors have stunted dendritic arbors. Likewise, during postnatal development, Neurog2 null PCs exhibit hypomorphic dendrites and often fail to achieve their typical bipolar arrangement, in which the axon and a single primary dendrite emanate from opposite poles of the cell body (Sotelo, 2004; Sotelo and Rossi, 2011). Thus, Neurog2 is important during the GC-independent phase of PC dendrite formation. The course of such early stages, however, influences the outcome of postnatal PC dendritic arbor expansion, which is GC dependent (Sotelo, 2004). These effects of Neurog2 become evident long after its downregulation, highlighting how instructive mechanisms that take place astride cell-cycle exit strongly affect the entire course and outcome of PC development.

Neurog2 null cerebella express reduced levels of two genes previously implicated in PC dendrite formation: Rora, which encodes the retinoic acid-like orphan receptor alpha, a transcription factor that plays crucial roles in PC development (Sotelo and Changeux, 1974; Herrup, 1983; Zanjani et al., 1990; Hamilton et al., 1996; Steinmayr et al., 1998); and Sclip, which encodes the homonymous stathmin-like phosphoprotein, a regulator of cytoskeletal dynamics (Poulain et al., 2008). Rora expression in fusiform PCs is crucial for dendrite regression and subsequent onset of dendritic process extension (Boukhtouche et al., 2006). Likewise, Sclip is implicated in the retraction of primitive processes and the formation and branching of new dendrites at early postnatal stages (Poulain et al., 2008).

Although Rora expression has its onset (E12.5) at the peak of Neurog2 transcription, Rora clearly outlives Neurog2, which declines by E14.5. Neurog2, possibly in its phosphorylated form, might initiate a long-lasting and self-sustaining regulatory cascade that supports PC differentiation long after Neurog2 downregulation. Further studies are required to assess whether Neurog2 phosphorylation is required for PC dendritogenesis.

We are grateful to Richard Hawkes for sharing unpublished results, to Gianvito Martino for his support of this project and to Cinzia Marinaro for help with viral vectors. We thank the Alembic and CFCM facilities (San Raffaele Scientific Institute) for image analysis and ES cell manipulation/blastocyst injection, respectively.