NeuroD2 controls inhibitory circuit formation in the molecular layer of the cerebellum

The cerebellar cortex is involved in the control of diverse motor and non-motor functions. Its principal circuit elements are the Purkinje cells that integrate incoming excitatory and local inhibitory inputs and provide the sole output of the cerebellar cortex. However, the transcriptional control of circuit assembly in the cerebellar cortex is not well understood. Here, we show that NeuroD2, a neuronal basic helix-loop-helix (bHLH) transcription factor, promotes the postnatal survival of both granule cells and molecular layer interneurons (basket and stellate cells). However, while NeuroD2 is not essential for the integration of surviving granule cells into the excitatory circuit, it is required for the terminal differentiation of basket cells. Axons of surviving NeuroD2-deficient basket cells follow irregular trajectories and their inhibitory terminals are virtually absent from Purkinje cells in Neurod2 mutants. As a result inhibitory, but not excitatory, input to Purkinje cells is strongly reduced in the absence of NeuroD2. Together, we conclude that NeuroD2 is necessary to instruct a terminal differentiation program in basket cells that regulates targeted axon growth and inhibitory synapse formation. An imbalance of excitation and inhibition in the cerebellar cortex affecting Purkinje cell output may underlay impaired adaptive motor learning observed in Neurod2 mutants.

ML preferentially innervate Purkinje cell dendrites while basket cells in the deeper ML contact the perisomatic compartment of Purkinje cells 12,13 . However, it is not finally settled whether these interneurons represent two distinct cell types or one functionally continuous population 14 . MLIs control Purkinje cell output by providing feed-forward inhibition in response to PF and CF activation [15][16][17][18][19] . MLIs are therefore essential to cerebellar processing, but the transcriptional mechanisms that regulate the diversification of MLIs and their differentiation remain incompletely understood 11 . Transcription factors from the bHLH family frequently act in cascades during development. The proneural bHLH factors Atoh1 and Ptf1a are necessary and sufficient to specify granule cell and MLI lineages, respectively 8,20,21 , consistent with a function of additional bHLH proteins in these cell lineages during later developmental stages. Indeed, NeuroD1 and NeuroD2, members of the NeuroD subfamily of neuronal bHLH proteins, have been implicated in the timing of 'transit amplification' of granule cells in the EGL 22 and the support of granule cell survival in the IGL 23,24 . NeuroD subfamily proteins were also shown to regulate neurite stratification of inhibitory amacrine cells in the retina 25 and peptidergic differentiation of inhibitory neurons in the dorsal spinal cord, downstream of Ptf1 26 . Since expression of the Neurod2 gene was observed in MLIs 24,27 , members of the NeuroD subfamily represent plausible candidate transcription factors to regulate excitatory and inhibitory circuit formation in the cerebellum.
Here, we show that deletion of Neurod2 in mice affects granule cell survival during a critical postnatal period, but is not essential for the assembly of granule cell circuitry. In contrast, NeuroD2 deficiency not only decreases MLI survival, but also impedes basket cell axogenesis and synapse formation onto Purkinje cells. We therefore hypothesize that an imbalance of excitatory and inhibitory neurotransmission in the cerebellar cortex could contribute to impaired motor learning in Neurod2 mutants.

Results
Generation of viable Neurod2 null mutant mice. Neurod2 expression was previously reported in the ML of the cerebellar cortex, most likely derived from MLIs 24 . Several co-expressed NeuroD family members often serve partly overlapping functions in CNS neurons 28 . Therefore, we determined expression patterns of NeuroD1, NeuroD2, and NeuroD6 using sensitive reporter mouse lines. X-gal histochemistry combined with immunostaining for parvalbumin (PV) in Neurod2-lacZ 'knock in' reporter mice 24 confirmed Neurod2 expression in MLIs at P20. In addition, Neurod2 was prominently expressed in granule cells, but absent from Purkinje cells (Fig. 1A). A corresponding analysis of Neurod1-lacZ 'knock in' reporter mice 29 revealed similar expression of Neurod1 in granule cells and MLIs, which was absent from Purkinje cells (Fig. 1A). In Nex-Cre 'knock in' mice 30 harboring a red fluorescent Cre reporter 31 , scattered Neurod6 expression was present in granule cells, whereas Neurod6 expression was absent from Purkinje cells and MLIs (Fig. 1A). These findings identify distinct expression profiles of NeuroD family members in granule cells and MLIs, providing excitatory and inhibitory input, respectively, to NeuroD family-negative Purkinje cells (Fig. 1B).
These expression profiles suggest partially overlapping NeuroD1 and NeuroD2 functions in granule cells and MLI, which could be investigated by a loss-of-function approach in mice. We recently generated Neurod2 mouse mutants 28 (Fig. 1C), which showed pronounced growth retardation after birth on a mixed C57Bl/6 × 129/ Sv genetic background (not shown). Most Neurod2 null mutants (Neurod2 −/− ) died before the age of 4 weeks during breeding to a C57Bl/6 genetic background, similar to Neurod2-lacZ mice 24 . Therefore, we backcrossed Neurod2 −/− mutants into a 129/SvJ genetic background, a strategy previously used to circumvent neonatal lethality of Neurod1 null mutants 32 . Premature death of Neurod2 −/− mutants was no longer observed after backcrossing into a 129/SvJ genetic background, i.e. breedings of heterozygous (Neurod2 +/− ) with homozygous Neurod2 −/− mutants produced 47% (53 out of 113 pups) Neurod2 −/− mutants at two months of age. Quantitative real time-PCR (Fig. 1C) and Northern blotting (not shown) confirmed absence of Neurod2 transcripts from the cerebellum of adult Neurod2 −/− mutants. A transient delay in weight gain persisted even after 4 backcrosses, but Neurod2 −/− mutants caught up after 4 weeks of age (Fig. 1D). Growth retardation of Neurod2-lacZ mutants in a mixed C57Bl/6 × 129/Sv background was associated with reduced serum levels of thyroid hormones 33 . However, analysis of thyroid hormones in Neurod2 −/− mutants in a 129/SvJ background revealed unaltered serum levels of T3 at P7 (Fig. 1E) and T3/T4 at P20 (data not shown). Disturbed organization of radial Bergmann glia is a hallmark of thyroid hormone deficiency 34 . Consistent with a normal thyroid hormone status, we observed no changes in radial Bergmann glia formation in the cerebellum of Neurod2 −/− mutants (Fig. 1F). Thus, backcrossing to a 129/SvJ genetic background restored a normal thyroid hormone status and provided a strategy to investigate the outcome of Neurod2 deficiency during postnatal and adult stages.
NeuroD2-deficient granule cells terminally differentiate and functionally integrate into the cerebellar circuit. Total cerebellar size was reduced by ~30% in Neurod2 −/− mutants at P30 compared to heterozygous Neurod2 +/− mice (in the following referred to as controls) when measured on parasagittal sections of the cerebellar vermis ( Fig. 2A,B). A quantification of total area sizes of lobules 3 and 5, as well as individual layers within lobule 5 (molecular layer, ML; Purkinje cell layer, PL; granule cell layer, GL; white matter, WM), suggested that size reduction in Neurod2 −/− mutants affected all lobules tested and individual layers within those lobules to a similar extent (Fig. 2B). As a result, the percentage of lobule 5 area covered by individual layers was unaltered in Neurod2 −/− mutants. Fluorescent immunostaining at P15 demonstrated similar numbers of PCNA-positive cells in the EGL of Neurod2 −/− mutants and controls (control, 166 ± 26.43; mutant, 145.6 ± 11.29; n = 4 each; p = 0.5049; Fig. 2C), suggesting no changes in the proliferation of granule cell progenitors in the EGL or a delay in radial granule cell migration from the EGL to the IGL in Neurod2 −/− mutants. However, in agreement with previous findings 24 , TUNEL staining revealed an increased granule cell apoptosis in the IGL of Neurod2 −/− mutants at P5 (control, 10.67 ± 2.404; mutant, 20.33 ± 1.453; n = 3 each; p = 0.0263; Fig. 2D), suggesting that size reduction of the GL in adult mutants results from the loss of postmitotic and postmigratory granule cells in the IGL. When quantified in a dorsal region of lobule 5 (marked by dashed boxes in Fig. 2A), granule cell density was not significantly reduced in Neurod2 −/− mutants at P30 (Fig. 2E). Even when measured in dorsal lobule 5 at 1 year of age, granule cell density in Neurod2 −/− mutants was not significantly changed (control, 417 ± 36/0.02 mm 2 ; mutant, 401 ± 18/0.02 mm 2 ; n = 3 each; p = 0.99), and we found no evidence for granule cell apoptosis or secondary granule cell degeneration in Neurod2 −/− mutants after P20 (Fig. 3F,G). Interestingly, the total number of NeuroD2-negative Purkinje cells was also reduced in lobule 5 of Neurod2 −/− mutants at P30 (control, 192 ± 6; mutant 149 ± 6; n = 3 each; p < 0.05). However, combined with reduced PL 'length' (measured in mm), the number of Purkinje cells per mm PL was not significantly reduced (Fig. 2E). These data demonstrate that NeuroD2 is required for the survival of granule cells during a critical postnatal period, but suggest that NeuroD2 is not essential for the long-term maintenance of granule cells.  Next, we tested the impact of NeuroD2-deficiency on the functional integration of glutamatergic cerebellar inputs. Golgi staining at P25 identified granule cell dendrites with claw-like telodendria, a feature of differentiated granule cells, in Neurod2 −/− mutants. Similarly, we observed no changes in projections of granule cell axons towards the ML in Neurod2 −/− mutants (Fig. 3A), suggesting intact PF trajectories and Purkinje cell innervation. Further support for normal postsynaptic development came from immunostaining for the α6 subunit of the GABA A receptor, a marker of terminally differentiated granule cells, which revealed normal expression at  ). Consistent with anatomical data, PF-Purkinje cell synapses were functional in Neurod2 −/− mutants, and PF-EPSCs facilitated to a similar extent as in controls (EPSC2/EPSC1; control, 1.51 ± 0.13; mutant, 1.58 ± 0.03; n = 5 each; p = 0.61; Fig. 3C), showing functional integration of NeuroD2-deficient granule cells into the cerebellar circuit. We also evaluated PF-EPSC kinetics to assess potential changes in transmitter release (PF: control rise, 0.50 +/− 0.06 ms, n = 5, mutant rise 0.57 +/− 0.6 ms, n = 6, p = 0.38; control decay: 3.2 +/− 0.7 ms, mutant decay: 3.0 +/− 0.3 ms, p = 0.73). The similar kinetics of control and mutant PF-EPCSs suggest that deletion of NeuroD2 does not significantly alter the time course of transmitter release at the PF-PC synapse. The similar EPSC kinetics are also consistent with the absence of gross changes in Purkinje cell morphology of NeuroD2 mutants. To test for differences in PF density we also stimulated PF inputs with increasing stimulus intensities. The slopes of the input-output curves of PF-EPSCs derived from 4 stimulus intensities (0, 3, 6 and 9 mV) for control and mutant were 0.11 and 0.12, respectively, and were not significantly different from each other (p = 0.91, n = 4 each), suggesting the PF density is similar in mutant and control. Purkinje cells also receive powerful excitatory input from CFs originating in the inferior olive. Postnatal elimination (~P14) of transiently established multiple CF inputs marks Purkinje cell maturation and requires normal PF development 35,36 . To test for multiple CF inputs, CF-EPSCs in Purkinje cells were evoked with increasing stimulus intensities.
Neurod2 expression is maintained in granule cells during adult stages 38,39 . Thus, we next addressed the long-term survival of granule cells and their stable integration into the cerebellar circuit in adult Neurod2 −/− mutants. Immunostaining for calretinin and vesicular glutamate transporter 1 (VGLUT1) to label granule cell axons and their terminals, respectively, and PV to mark Purkinje cell dendrites, showed unaltered PF trajectory and PF terminals along Purkinje cell dendrites in Neurod2 −/− mutants at 1 year of age ( Fig. 3E; please note differences in interneuron-associated PV staining in the ML that are further addressed below). These findings are supported by electron microscopy. We observed normally structured nuclear heterochromatin and no signs of heterochromatic clumping (that precedes cell death) in granule cells from 1 year old Neurod2 −/− mutants (Fig. 3F). Moreover, electron microscopy revealed a similar density of PF-Purkinje cell spine synapses in Neurod2 −/− mutants (control, 37.8 +/− 4.8 synapses/100 µm 2 ; mutant, 31.27 +/− 0.5 synapses/100 µm 2 ; n = 3 each; p = 0.25), in line with the finding of largely unaltered granule cell and Purkinje cell densities in Neurod2 −/− mutants. In summary, these results show that NeuroD2 is not essential for the morphological differentiation of granule cells and the long-term maintenance of glutamatergic inputs to the cerebellar cortex.
NeuroD1 is closely related to NeuroD2 and prominently expressed in granule cells suggesting that it functionally compensates for the loss of NeuroD2. Therefore, we generated Neurod1/2 compound mutants by breeding Neurod2 −/− mutants to conditional Neurod1 mutants 30 , in which recombination occurs in postmigratory granule cells using GABA Aα6-Cre driver mice 40 . At three weeks of age we observed no granule cell apoptosis in conditional Neurod1 null mutants with one wildtype copy of Neurod2 (Neurod2 +/− ; Neurod1 fl/− ; Cre) or Neurod2 null mutants with one functional copy of Neurod1 (Neurod2 −/− ; Neurod1 fl/− ; Fig. 3G). In contrast, Neurod1/2 double mutants (Neurod2 −/− ; Neurod1 fl/− ; Cre) were severely ataxic, and histological analysis revealed reduced cerebellar size and increased granule cell apoptosis at 3 weeks of age compared to single mutant controls (Fig. 3G). These findings identify highly overlapping survival functions of NeuroD1 and NeuroD2 in postmigratory granule cells.

NeuroD2 promotes survival and terminal differentiation of ML interneurons.
In addition to glutamatergic granule cells, NeuroD2 is expressed by GABAergic MLIs 24 (Fig. 1A), but its role in these cells after P25 is unknown. Immunostaining at both P15 and P25 demonstrated that PV-expressing MLIs were present in control mice, but virtually absent from the ML of Neurod2 −/− mutants (Fig. 4A,D). Next, we asked whether absence of PV marker expression resulted from a loss of MLI in Neurod2 −/− mutants. As determined by TUNEL staining, the number of apoptotic cells was not altered at P15 (data not shown), and cell density in the ML was even increased in Neurod2 −/− mutants (Fig. 4B,C). In stark contrast, hematoxylin and eosin (H&E) histology, as well as immunostaining for glutamate decarboxylase 1 (GAD67) demonstrated that MLI density was strongly reduced (to ~30%) in Neurod2 −/− mutants at P25 (Fig. 4E,F). Thus, we conclude that NeuroD2 is required for MLI survival at late postnatal stages. Since loss of PV expression preceded the reduction of MLIs, and the surviving population of GAD67 + MLIs lacked PV expression (Fig. 4F), we furthermore hypothesized that NeuroD2 acts during the terminal differentiation of these cells upstream of PV expression, but downstream of the expression of more basal GABAergic markers, such as GAD67.
To further test this hypothesis, we performed additional immunostainings at P25 and found that the remaining population of GAD67 + /Parv − MLIs in Neurod2 −/− mutants maintained expression of the GABA Aα1 receptor (Fig. 5A). We verified these findings by in situ hybridization and found persistent expression of tha GABA Aα1 receptor (Fig. 5B), GAD67, and the vesicular GABA transporter (VGAT) (data not shown) in surviving MLIs, but a dramatic loss of PV transcripts in Neurod2 −/− mutants (Fig. 5B). Finally, we examined the expression of Pax2, a transcription factor that marks the transition from proliferative precursor cells to terminally differentiating postmitotic MLIs 41,42 . At P5, we observed a similar number and distribution of Pax2-positive interneuron precursors in the prospective white matter (lobule 5; control, 358.3 +/− 32.7, n = 4 sections from 2 mutants; mutant, 391.4 +/− 24.22, n = 6 sections from 3 mutants; p = 0.54), IGL, and ML of Neurod2 −/− mutants when compared to controls (Fig. 5C), suggesting that NeuroD2 is not required for the production and initial migration of interneuron precursors. At P10, the loss of Pax2 expression in the ML of controls (following an inside-out gradient) was paralleled by GAD67 accumulation. In contrast, Pax2 expression was maintained and the acquisition of GAD67 immunoreactivity delayed in the ML of Neurod2 −/− mutants (Fig. 5D). Pax2 was completely absent from MLIs in adult controls, but persisted in the majority of NeuroD2-deficient MLIs at P50 (data not shown) and even at 6 months of age (Fig. 5E). This suggests that surviving mutant MLIs retain an immature developmental state. By contrast, we found no evidence that Golgi cells, the GABAergic interneurons of the GL, were affected in Neurod2 −/− mutants. In situ hybridization for GAD67 ( Immunostaining for Pax2 identified a normal number of Pax2 + Golgi cells in the GCL of Neurod2 −/− mutants compared to controls at 6 months of age (Fig. 5G). Since Golgi cells provide both feedforward and feedback inhibition to granule cells 43 , these findings further support an intact functional integration of granule cells into the cerebellar network. In summary, we posit that NeuroD2 is not necessary to instruct a GABAergic phenotype in surviving MLIs, but is required for terminal differentiation of GABAergic interneurons in the ML of the cerebellum.

Irregular basket cell axon growth, impaired formation of basket cell terminals, and decreased synaptic inhibition of Purkinje cells in Neurod2 −/− mutants.
To elucidate specific NeuroD2 controlled functions in MLI development, we analyzed morphological differentiation of MLIs in more detail. Golgi staining and immunostaining for the dendritically expressed microtubule-associated protein MAP2 at P25 indicated that stellate cell morphology, including dendritic arborization in the outer ML, was not grossly altered in Neurod2 −/− mutants (Fig. 6A). Immunostaining for GAD65 and PV in the upper ML, where most presynaptic terminals are derived from stellate cells 11,12 , showed GABAergic puncta aligned along Purkinje cell dendrites (Fig. 6B). Together, these data suggest that surviving stellate cells in Neurod2 −/− mutants develop dendrites, axons, and synapses onto Purkinje cells. However, the unequivocal identification and quantification of stellate cell/Purkinje cell synapses will require further studies, including higher-magnification microscopic images of pre-and postsynaptic structures. As in the case of stellate cells, mutant basket cells in the inner ML developed normal dendritic arbors, as revealed by Golgi silver staining (Fig. 6C) and 3D reconstruction of MAP2 labeled dendrites at P25 (Fig. 6D) (total dendritic length: control, 351 ± 18 µm, n = 20 cells; mutant, 421.8 ± 31.3 µm; n = 30 cells; p = 0.09). To determine their growth trajectories, we labeled basket cell axons at P25 by fluorescent immunostaining for the neurofilament heavy chain (NFH) and performed 3D reconstructions of individual axons (Fig. 6E). As expected from the lower number of basket cells, the number of NFH labeled basket cell axons was prominently reduced in the ML of Neurod2 −/− mutants. Strikingly, while axons in control mice grew along the translobular plane and extended collaterals towards the PL, basket cell axons in Neurod2 −/− mutants followed random trajectories, even towards the pial surface. Consequently, very few of the basket cell axon collaterals descended towards the PL in Neurod2 −/− mutants to form contacts with Purkinje cells (Fig. 6E).
To test whether aberrant basket cell axon growth in Neurod2 −/− mutants compromises the formation of basket cell terminals at Purkinje cells, including the characteristic basket cell pinceau at the axon initial segment (AIS), we performed immunostainings for two specific markers: the hyperpolarization activated cyclic nucleotide-gated potassium channel 1 HCN1 44 and postsynaptic density protein 95 (PSD95) 45 . Both markers were absent from the perisomatic compartment and AIS of Purkinje cells in Neurod2 −/− mutants at P25 (Fig. 7A,B). Since postsynaptic targeting of ankyrin 3 (ANK3) to the AIS is required for the pinceau assembly 46 , its presence in Purkinje cells of Neurod2 −/− mutants (Fig. 7B) indicates normal AIS formation and suggests that disrupted pinceau formation is a basket cell-intrinsic defect. Furthermore, immunostaining for NFH and PSD95 at P75 demonstrated that PL-directed basket cell axon growth was permanently disrupted and not only developmentally delayed in Neurod2 −/− mutants (Fig. 7C). We conclude that NeuroD2 controls a basket cell-intrinsic transcriptional program that regulates targeted axon growth and presynaptic differentiation, but is not essential for MLI dendritogenesis. We next addressed whether the reduction in the number of MLIs and the morphologically abnormal basket cell axons result in decreased synaptic inhibition to Purkinje cells. To test whether the number of functional inhibitory synapses is diminished in Neurod2 −/− mutants, we examined the frequency of miniature inhibitory postsynaptic currents (mIPSCs) in Purkinje cells of acute cerebellar slices (Fig. 7D). In the presence of sodium channel, glutamate receptor, GABA B receptor, and glycine receptor blockers, we found that mIPSC frequency was significantly reduced in Neurod2 −/− mutants ( Fig. 7E; left: control, 26.9 ± 1.7 Hz, n = 15; mutant, 15.4 ± 1.4 Hz, n = 16, p < 0.0001; middle: cumulative probability histogram of inter-event intervals, p < 0.0001, Kolmogorov-Smirnov test), whereas mIPSC amplitude was unaffected ( Fig. 7E; right: control, 225.5 ± 9.0 pA, n = 15; mutant, 230.8±17.3 pA, n = 16, p = 0.8; cumulative probability histogram, p = 1.0, Kolmogorov-Smirnov test), as well as mIPSC kinetics (control rise time 0.47 ± 0.07 ms, n = 8, mutant rise time, 0.46 ± 0.06 ms, n = 7, p = 0.43, t-test; control decay time 4.0 ± 0.4 ms, n = 8, mutant decay time, 3.5 ± 0.2 ms, n = 7, p = 0.1, t-test), suggesting no major differences in synaptic release of GABA from remaining MLIs. However, we found that input resistance (Ri) of Purkinje cells was increased in Neurod2 −/− mutants (control, 23.6 ± 3.8 MOhm, n = 8; mutant, 51.1 ± 8.5 MOhm, n = 7, p < 0.05), consistent with decreased GABAergic input 47 . To test whether the differences in Ri between control and mutant are due to intrinsic Purkinje cell properties or GABAergic input, we recorded spontaneous Purkinje cell firing in the presence of synaptic blockers. The average firing frequencies of Purkinje cells were similar in mutants and controls and thus suggest that Purkinje cell excitability was not . Together, these data indicate that GABAergic inputs onto Purkinje cells are reduced in number, but that GABAergic transmission at remaining inputs is intact in Neurod2 −/− mutants. MLIs also form reciprocal GABAergic synapses 48 . We therefore tested whether inhibition was also impaired in MLIs of Neurod2 −/− mutants.
To preferentially target stellate cells, MLIs from the outer two thirds of the ML were recorded (Fig. 7D). Indeed, mIPSC frequency was decreased in MLIs from Neurod2 −/− mutants compared to controls ( 504 ± 185 MΩ, n = 5, p = 0.7, t-test), suggesting that decreased inhibitory input onto mutant MLIs does not alter intrinsic electrical properties of surviving MLIs or affect MLI excitability. Consistent with intact glutamatergic input and unaltered mIPSC amplitudes of the remaining GABAergic inputs, we observed no morphological abnormalities in the somatodendritic compartment of Purkinje cells in Neurod2 −/− mutants (Figs 3E,F and 5A). In addition, somatic recordings of spontaneous firing rates in Purkinje cells in the presence of synaptic blockers showed no differences between controls and Neurod2 −/− mutants (control, 30.8 ± 4.0 Hz, n = 12; mutant, 27.3 ± 3.0 Hz, n = 17; p = 0.49). This indicates that action potential generation is not compromised and further suggests that the somatodendritic compartment of Purkinje cells in Neurod2 −/− mutants is normal. Nevertheless, the identification of minor structural and functional changes in Purkinje cell dendrites of Neurod2 −/− mutants will require further investigation. In contrast to dendrites, we frequently observed swellings of Purkinje cell axons proximal to the (basket cell terminal-lacking) AIS in Neurod2 −/− mutants (Fig. 7G). In conclusion, loss of basket cell contacts has no major effects on the somatodendritic compartment of Purkinje cells, but could result in axonal pathology.

Discussion
NeuroD2 deficiency in a mixed C57Bl/6 × 129/Sv background is associated with reduced serum TH levels 33 , severe growth retardation and postnatal death 24 . Genetic background heavily affects phenotype expressivity in many mouse mutants 49 . Accordingly, backcrossing of our Neurod2 mutants to a 129/SvJ background restored normal blood TH levels and viability, and allowed us to examine NeuroD2 functions in the cerebellar cortex in the absence of confounding TH deficits. We found that ~70% of granule cells are preserved in adult Neurod2 −/− mutants, demonstrating that NeuroD2 is required for granule cell survival only during a critical postnatal period (until ~P20). Timely and complete CF elimination as well as normal short-term plasticity of PF and CF inputs in Purkinje cells of Neurod2 −/− mutants indicated that mutant granule cells structurally and functionally integrate into the cerebellar circuit. Thus, our findings suggest that NeuroD2 is not essential for the formation and functional integration of glutamatergic inputs in the cerebellum. NeuroD1 and NeuroD2 are co-expressed in granule cells 39 . Granule cell apoptosis was strikingly increased in Neurod1/2 double mutants, in which Neurod1 disruption occurs in postmigratory granule cells. We conclude that NeuroD1 and NeuroD2 serve redundant survival functions in postmigratory granule cells. We observed markedly reduced MLI numbers (~30% of controls) in Neurod2 −/− mutants at P25, but not at P15. This suggests that NeuroD2 is required for MLI survival during a critical postnatal period, i.e. after their arrival in the nascent ML. Loss of MLIs may result from a failure to properly execute a terminal differentiation program 50 . In support of this hypothesis, surviving NeuroD2-deficient MLIs maintain Pax2 expression, a marker of immature interneurons 41,42 . Concomitantly, mutant MLIs lack expression of differentiation markers, such as PV and HCN1 ion channels. We speculate that these factors are integral components of a NeuroD2-controlled MLI terminal differentiation program. However, whether Pvalb and Hcn1 are bona fide NeuroD2 target genes requires further investigation. The abnormal expression status of MLIs is associated with a specific impairment in the morphological differentiation of basket cells, as basket cell axon branching directed to the PL and inhibitory terminal formation, but not dendritic growth, are severely affected in Neurod2 −/− mutants. Irregular basket cell axon growth could be the consequence of detrimental effects of persistent Pax2 expression on basket cell differentiation. A previous study indicates that continuous Pax2 expression in the renal epithelium of transgenic mice causes kidney abnormalities, possibly by instilling an immature state 51 . Furthermore, NeuroD2 deficiency may impair the expression of axonal signaling receptors, similar to the situation in Neurod2/6 double mutants, in which disrupted callosal axon tract formation is associated with the loss of contactin-2 expression 28 . A candidate receptor in basket cell axons is Neuropilin-1, which mediates SEMA3A-induced axon guidance signaling during basket cell axon branching 52 .
Consistent with a significant reduction in MLI number and undirected basket cell axon growth, we found that inhibitory terminals are diminished in the ML of Neurod2 −/− mutants. Although the density of GABAergic synapses is reduced throughout the entire ML, the most striking consequence of Neurod2 deletion remains the absence of functional GABAergic basket cell terminals. A previous study suggested that basket cells mediate large-amplitude calcium store-dependent mIPSCs in Purkinje cells 53 . However, mIPSC amplitude distributions of control and Neurod2 −/− mutants were similar, indicating that stellate cells may also generate such large mIPSCs. In addition to intact excitation, CF elimination from Purkinje cell dendrites also requires GABAergic inhibition 54 . Despite the finding that mIPSC frequency is decreased by >50% in Neurod2 −/− mutants, we found that CFs are pruned by P19 in Neurod2 −/− mutants. We conclude that residual Purkinje cell inhibition by stellate cells (and possibly Purkinje cell collaterals) is sufficient for the efficient elimination of CF synapses in Neurod2 −/− mutants.
Our data indicate that Purkinje cell dendrites and the AIS develop properly in the absence of perisomatic basket cell innervation. Since they appeared anatomically normal, intrinsic functional properties of Purkinje cells were not rigorously tested in electrophysiological experiments. To clarify this issue in detail, more experiments in current clamp would be required, e.g. step current injections with varying amplitudes. However, spontaneous firing frequencies of PCs measured in the presence of synaptic blockers were similar, suggesting that intrinsic properties of PCs were not dramatically altered in mutants. Input resistance in mutants was assessed with GABAergic transmission intact. The apparent increase in input resistance in mutants is consistent with decreased inhibition, as previously shown 55 . Futher experiments could test how partial loss of inhibtion in the cerbellar cortex alters its output, e.g. in vivo recordings from deep cerebellar nuclei (DCN). Our data suggest that basket cell terminals could serve a role in maintaining axonal integrity as we frequently observed swellings of Purkinje cell axons proximal to the AIS in Neurod2 −/− mutants. Axonal swellings ('torpedoes') are a common feature of various neurodegenerative diseases with cerebellar involvement, but also occur in the developing mouse cerebellum and peak at P11 56,57 . Axonal swellings in Neurod2 −/− mutants are not merely a consequence of abolished inhibition, as the removal of GABA A receptor-mediated synaptic inhibition in Purkinje cells causes no corresponding axonal pathology 50 . Instead, we suggest that the formation of an intact pinceau is required for maintaining the long-term integrity of Purkinje cell axons. Interestingly, it has recently been questioned whether the pinceau serves GABAergic signaling at all 51 . Instead, it may generate high resistivity in the AIS environment, serving the electrical inhibition of the proximal axon 52 . We thus speculate that the pinceau assumes an axoprotective function by controlling AIS excitability, and despite normal spontaneous firing rates in the Purkinje cell soma, it remains to be determined whether action potential propagation along the Purkinje cell axons is affected in Neurod2 −/− mutants.
Purkinje cells control motor behavior via inhibitory projections to deep cerebellar and vestibular nuclei 58 . Neurod2 −/− mutants showed normal basic locomotor behavior but impaired balancing and advanced motor learning, similar to the genetic blockade of all synaptic inhibition in Purkinje cells, which affects vestibulo-cerebellar motor learning, but not baseline motor performance 59 . Feed-forward inhibition by MLIs temporally modulates rate and regularity of Purkinje cell firing 15,16,18,19,58 . Thus, MLI-mediated inhibition of Purkinje cells may play an important role in motor learning. However, due to the global nature of the mutation, we cannot exclude that changes in granule cells or other brain regions, e.g. motor cortex, DCN, and basal ganglia, contribute to or even underlie motor learning deficits. However, we hypothesize that diminished inhibition of Purkinje cells in Neurod2 −/− mutants together with normal PF-and CF-mediated excitation results in an imbalance of excitation and inhibition in Purkinje cells, ultimately contributing to altered Purkinje cell firing and deficits in motor learning.

Materials and Methods
Mouse mutants. For  TUNEL assay. Fragmented DNA of apoptotic cells in paraffin tissue sections was end-labeled using the DeadEnd colorimetric TUNEL system (Promega) according to manufacturer's instructions.
Golgi impregnation. Golgi impregnation of cryostat sections (50 µm) embedded with Eukitt (Sigma) was performed using a FD Rapid GolgiStain Kit (FD Neuro Technologies) according to manufacturer's instructions.
Morphometric analysis. Digital images of chromogenic and fluorescent immunostainings were generated using a Zeiss Axiophot microscope, a Leica DM RXA microscope, and a LSM510 Meta Zeiss confocal laser-scanning microscope. Cerebellar size, lobule 3/5 area sizes, lobule 5 layer sizes, and MLI numbers (in cerebellar lobule 5) were quantified on sagittal sections (total cerebellar area: 2 sections/mouse; lobules 3/5: 6-7 sections/mouse; n = 3 per genotype) using ImageJ software. Apoptotic cells (as identified by TUNEL assay in the IGL) were quantified on 2 sagittal sections per mouse (n = 3). Granule cell density and MLI density was quantified on six toluidine blue stained 0.5 μm semithin sagittal sections per mouse in 0.02 mm 2 regions of the dorsal GL in lobule 5 (n = 3 per genotype). Purkinje cell density was quantified on four toluidine blue and two methylen-blue stained 0.5 μm semithin sagittal sections per mouse in lobule 5 (n = 3 per genotype). PF to Purkinje cell synapses were identified by their typical ultrastructural morphology 60 , and their density was determined on 50 nm thick ultrathin sections in 100 μm 2 regions of the ML in lobule 5. For the reconstruction of individual basket cells neurites, 12 μm deep z-stacks were collected in the translobular plain (40 images per stack) and analyzed with Neuromantic V.1.6.3 software for 3D reconstructions.
Scientific RepoRts | (2019) 9:1448 | https://doi.org/10.1038/s41598-018-37850-7 a variable amplitude sliding template was used. Detection threshold was 2-3 SDs from noise, and detected events were visually inspected after detection for accuracy. Corrupt events were rejected from further analysis. Although clustered events were uncommon (with the exception for mIPSC recordings in control Purkinje cells), the first event in a cluster was accepted for events that were separated less than 5 ms, all events that were more than 5 ms apart were included in the analysis.
Behavioral analysis. Motor function was investigated on a rotating rod (diameter 3 cm). Endurance of male mice (n = 4-5 per genotype) was tested at constant rotating speed (5 rpm). Motor learning of male mutants (n = 24) and controls (n = 32) was assessed with the following protocol: On day 1 (habituation) mice were placed three times on the rod (with 10 min intertrial resting intervals). Rotating speed was successively increased from 3.3-12 rpm within 2 min. On day 2 (challenge) mice were placed 5 times on a rod (with 10 min intertrial resting interval). Rotating speed was successively increased from 4.1-41 rpm within 6 min. To detect subtle motor coordination defects mutants (n = 8) and controls (n = 4) were placed on a grid (30 × 30 cm, 1,5 cm grid hole size) and allowed to freely explore. Numbers of front and hind limb slips were quantified in 2 consecutive trials (3 min each; intertrial intervals, 10 min). Balancing performance was tested by placing male mice (controls, n = 6; mutants, n = 8) three times (10 min intertrial resting intervals) on a horizontal bar (diameter 1.8 cm) for a maximum of 1 min and measuring the time before mice fell off.

Statistical analysis.
Reported values for electrophysiology and morphometry are the mean ± SEM., all other numerical values are shown as the mean ± SD. Unless stated otherwise, p values were determined by the Student's t test of two-tailed uncoupled samples. In Figs 1D and 8B we used multiple t tests and corrected for multiple comparisons using Holm-Sidak method. Statistical differences were considered to be significant when p < 0.05 (*p < 0.05, **p < 0.01, ***p < 0.001). All statistical analysis was performed using the software Statistica 10.0 (StatSoft, Tulsa, USA), GraphPad (Prism) and MS Excel.