Synaptic input as a directional cue for migrating interneuron precursors

During development of the vertebrate nervous system, neuronal precursors migrate extensively before they reach their adult positions and terminally differentiate (Valiente and Marín, 2010). Both in the developing forebrain and cerebellum, projection neurons destined to form cortical structures originate in the ventricular epithelium, from where they follow a radial route to reach their adult positions. Precursors of inhibitory interneurons, on the other hand, follow more varied routes to reach their target positions, which generally include extensive tangential relocations (Goldowitz and Hamre, 1998; Letinic et al., 2002; Tan et al., 1998). The eventual establishment of efficiently wired CNS circuits is dependent on a precise coordination of these morphogenetic migrations (Marín et al., 2010). In fact, increasing numbers of neurologic and psychiatric conditions are being associated with neuronal migration deficits (Barkovich et al., 2012; Peñagarikano et al., 2011; Rivière et al., 2012).

In contrast to signals and mechanisms subserving cortical projection neuron migration (Jossin and Cooper, 2011; Rice and Curran, 2001), our mechanistic understanding of interneuronal pathfinding, dispersal and eventual matching with projection neurons is still rather fragmentary (Marín et al., 2010). Cortical inhibitory interneurons are either generated late in ontogenesis or have to migrate for extended periods to their final positions (Leto et al., 2009; Rymar and Sadikot, 2007). Consequently, they have to navigate the nascent lattice formed by earlier settled projection neurons to reach their specific target areas. A key question is whether and how the latter may direct interneuron migration and maturation to assure the generation of properly functioning CNS microcircuits. Intriguingly, an increasing body of evidence indicates that the mobility of neuronal precursors, including immature interneurons, is sensitive to neurotransmitters (Bortone and Polleux, 2009; de Lima et al., 2009; Komuro and Rakic, 1993; Manent and Represa, 2007).

Here, we follow up on this lead and ask how neural communication might affect corticogenesis, and in particular directed interneuron migration in the cerebellar cortex. This cortical structure seems uniquely suited to address such a general question: compared with the forebrain cortex, the cerebellar cortex is evolutionarily rather conserved (Bell et al., 2008). It is constituted by a limited number of genetically well characterized cell types with distinct developmental histories (Goldowitz and Hamre, 1998; Schilling et al., 2008). Its inhibitory interneurons, and particularly those in the molecular layer (ML), arguably form one of the largest and best-characterized groups of inhibitory interneurons in the vertebrate CNS (e.g. Markram et al., 2004; Kepecs and Fishell, 2014; Carter and Regehr, 2002; Jörntell et al., 2010; Kim et al., 2014).

Among the first cerebellar neurons formed are Purkinje cells, inhibitory interneurons of the deep nuclei, and inhibitory interneurons of the future granule cell layer (GL). In the mouse, these go through their last mitosis from around embryonic day (E) 10.5-13.5, E10.5-11.5, and E13.5 to postnatal day (P) 1, respectively (e.g. Sudarov et al., 2011; for a recent review, see Leto et al., 2016), and subsequently migrate radially to form the cerebellar anlage. From about E12.5-16, Math1 (Atoh1)⁺ cells migrate from the rhombic lip (Ben-Arie et al., 1997) underneath the pial surface of the cerebellar anlage to form the external granule cell layer (EGL), where they proliferate extensively up to about P15. Math1⁺ cells give rise to all excitatory cerebellar neurons, including granule cells, unipolar brush cells and excitatory cells in the deep nuclei (Machold and Fishell, 2005). Both afferent climbing fibres (Kita et al., 2015; Wassef et al., 1992) and mossy fibres reach the cerebellar anlage before or around birth (Ashwell and Zhang, 1992; Grishkat and Eisenman, 1995; Nunes and Sotelo, 1985).

Lastly, inhibitory interneurons resident in the ML are formed over a rather protracted period, from about E15 to P7 (Leto et al., 2009; Sudarov et al., 2011). They have to traverse the nascent white matter (WM), the emerging internal granule cell layer (IGL), and eventually the Purkinje cell layer (PCL) to reach the ML. There, they initially accumulate at the border of the EGL before they take up their adult positions (Leto et al., 2009; Weisheit et al., 2006; Zhang and Goldman, 1996; Simat et al., 2007; Maricich and Herrup, 1999). According to current thought, it is only then that they become synaptically integrated with granule cell afferents and Purkinje neurons, the sole projection neurons of the cerebellar cortex (Schilling, 2013).

In striking contrast to this detailed descriptive understanding, the mechanisms that direct cerebellar interneuron migration and dispersal remain elusive. To address this issue, we combined two-photon time-lapse imaging and patch-clamp recordings to characterize genetically tagged, migrating precursors of cerebellar inhibitory interneurons. Unexpectedly, we found that these precursors in transit are synaptically innervated. Elimination of synaptic exocytosis reduced the speed and, conspicuously, the directionality of migration. These findings establish a hitherto unknown, early synaptic innervation of developing interneurons. Further, they document that such synapses implement a mechanism that allows these cells to properly navigate the nascent cerebellar cortex.

In acute cerebellar slices of 7- to 9-day-old mice expressing EGFP from the locus of the paired-box gene Pax2 (Pfeffer et al., 2002; Weisheit et al., 2006), precursors of ML inhibitory interneurons can be recognized as strongly EGFP⁺, slender, fusiform cells. They may be found within the prospective WM, the IGL, and the nascent ML, through which they consecutively migrate to reach their adult position. Their morphology and mobility towards the developing ML (Figs 1 and 2, Movies 1-3) allow them to be readily distinguished from the larger, roundish, and already settled inhibitory interneurons of the IGL (ʻGolgi cells'), which also express Pax2 (Maricich and Herrup, 1999). Within the ML, newly arrived Pax2-EGFP cells could be readily distinguished from more mature ML interneurons by their strong EGFP signal, their preferential orientation perpendicular to the PCL, and their mobility. After entering the ML, these cells first translocate towards the border of the ML and the EGL (Simat et al., 2007; Weisheit et al., 2006). Upon their subsequent embedding in the growing ML, they rapidly downregulate Pax2 and Pax2-EGFP (Glassmann et al., 2009; Weisheit et al., 2006), and they can be recognized as dimly stained, immobile cells with their long axis preferentially parallel to the PCL (Fig. 2B).

Strongly Pax2-EGFP⁺, immature precursors of ML inhibitory interneurons (henceforth referred to as ʻPax2 cells') were characterized by a high input resistance (R_in>1 GΩ in all but 6/176 cells) and a capacitance of 9.4±0.4 pF (mean±s.e.m.). Their whole-cell current pattern (Fig. 3A) was made up by transient and delayed rectifier K⁺ currents and voltage-gated, TTX-sensitive Na⁺ currents (Fig. S1). We did not observe spontaneous action potentials in these cells. Upon current injection (Fig. 3B), Pax2 cells in the prospective WM (n=7), the nascent IGL (n=8) and the ML (n=10) generated spikelets, but no full action potentials. Their Na⁺ current density was 0.7±0.1 pA/µm².

This clearly distinguishes Pax2 cells from Pax2-EGFP⁺ inhibitory interneurons of the GL, which are characterized by a much larger membrane capacitance (47.8±5.3 pF, n=13). This latter value is in excellent agreement with data reported previously (Dieudonné, 1995; Dieudonné, 1998; Elisabetta Cesana et al., 2006). The electrophysiological properties of Pax2 cells are also clearly distinct from those of mature inhibitory ML interneurons, which show spontaneous action potentials of up to 35 Hz and Na⁺ current densities exceeding 30 pA/µm² (Ruigrok et al., 2011; Southan and Robertson, 1998).

Application of GABA (100 µM) or the GABA_A receptor agonist muscimol (300 µM) to Pax2 cells held at −70 mV consistently resulted in inward currents [0.03±0.01 pA/µm² (n=14) and 0.1±0.01 pA/µm² (n=13), respectively] that were blocked by bicuculline (100 µM) by about 85% [to 0.001±0.001 pA/µm² (n=4) and 0.02±0.01 pA/µm² (n=4) for GABA and muscimol, respectively; Fig. 3C]. The reversal potential at the maximal slope conductance was −0±2 mV (n=12), confirming a Cl⁻-mediated current (Fig. S2A-A″).

Further, these Pax2 cells in transit also express ionotropic receptors of the AMPA/kainate (KA) type (Fig. 3D, Fig. S2B-B″). We noted that Pax2 cells migrating through the WM display significantly smaller KA (300 µM)-induced receptor currents (0.3±0.06 pA/µm², n=14; V_h=−70 mV) than those in the ML (0.95±0.13 pA, n=14; P<0.001 for WM versus ML) indicating developmental changes of receptor density or gating properties (Smith et al., 2000), e.g. due to a switch in flip/flop splicing (Monyer et al., 1991). Cyclothiazide (100 µM), an allosteric modulator of the flip versions of AMPA receptors (Partin et al., 1993), potentiated KA-induced responses in Pax2 cells in the WM (by 86±13% at –70 mV, n=12; P<0.01) but not in the ML (15±6%, n=9; P=0.3). NBQX (20 µM) completely blocked KA-induced currents (n=8). Our data are consistent with the reported relative increase of AMPA-receptor flop variants (Monyer et al., 1991) and/or increased hetero-oligomerization (Liu and Cull-Candy, 2002) with ongoing development.

Unexpectedly, we observed spontaneous postsynaptic currents (sPSCs) in Pax2 cells, suggesting that these immature cells in transit receive synaptic input (Fig. 3E-G). sPSCs could be recorded from 15 out of 17 Pax2 cells in the ML and 5/6 cells transiting the nascent IGL. We could not detect sPSCs in Pax2 cells transiting the prospective WM (n=11).

Next, we wanted to confirm that Pax2 cells receive synaptic input while actually migrating. Over a period of several hours these cells travel substantial distances and may traverse the entire cerebellar cortex (Movie 1). We used high-resolution time-lapse video microscopy and selected cells that had moved at least 3 µm (n=35; median 11.2 µm, range 3-43 µm) for subsequent patch-clamp recording. Wash-out of EGFP during recording confirmed that the visually selected cell was impaled by the patch pipette.

Consistent with the above data for cells selected on the basis of their position and morphology alone, Pax2 cells that had migrated immediately before patch-clamping received synaptic input, depending on their position. Eight out of 13 cells in the IGL, 15/19 in the ML, but 0/3 cells in the prospective WM showed sPSCs. The synaptic origin of this activity in Pax2 cells traversing the IGL and ML was supported by the increased rate of miniature PSCs (mPSCs) following application of the Ca²⁺ ionophore ionomycin (3 µM; Fig. 3L-N). This agent increases the intracellular Ca²⁺ concentration and thus enhances release from the readily releasable vesicular pool in synaptic terminals. In the WM, in the absence of TTX, no sPSCs were observed in Pax2 cells, even after application of ionomycin.

Analysis of current kinetics distinguished two types of sPSCs with decay time constants of <1 ms and about 12 ms, respectively (Fig. 3E,F,H,I,K upper panel; Fig. 3E,G,K lower panel). Further analysis showed that the former were sensitive to NBQX and APV (i.e. were mediated by ionotropic glutamate receptors; Fig. 3J), whereas the latter were abolished by bicuculline (i.e. mediated by GABA_A receptors; Fig. 3H). As expected for AMPA/KA and GABA_A receptors, the reversal potentials of the fast and slow sPSCs were 0 and –40 mV, respectively (Fig. S3). The rise times of these sPSCs were 0.37±0.01 ms for fast-decaying sPSCs (n=1843, 20 cells) and 0.46±0.01 ms for slowly decaying sPSCs (n=384, 17 cells).

All cells with sPSCs (n=43) received glutamatergic input, whereas GABAergic input was less abundant (n=35/43). There was no indication of layer-specific differences in GABAergic or glutamatergic innervation.

On their way into the ML, Pax2 cells migrate sequentially through the WM, the nascent IGL, the PCL, and in the ML up to the EGL (for an overview, see Fig. 4A). To assess the density of presynaptic elements that may contact Pax2 cells en route, we quantified synaptophysin-immunoreactive puncta in these layers. In line with the layer-specific frequency of sPSCs, the density of synaptophysin⁺, putative presynaptic elements increased from the WM to the ML (Fig. 4B-D). Ultrastructural analysis of Pax2 cells in the lower ML confirmed the occurrence of synaptic structures on these cells (Fig. 5, Fig. S4). This particular location was chosen since expression of Pax2-EGFP allowed the reliable selection of interneuron precursors migrating towards more superficial positions over those that have already settled in the ML (Weisheit et al., 2006). This ultrastructural evidence, together with the fast rise times of the sPSCs, suggests that these sPSCs originate from direct synaptic innervation rather than from transmitter spill-over.

To probe the functional significance of this surprising and hitherto unknown synaptic input, we used tetanus toxin (TeNT) and Co²⁺ to abrogate presynaptic vesicular release. Of these agents, TeNT is more efficient as it inhibits exocytosis by cleaving synaptobrevin. We verified the efficacy of these agents in our preparation by monitoring their effect on the frequency and amplitudes of sPSCs recorded from Purkinje neurons. These cells, rather than sparsely innervated Pax2 cells, were chosen because their already strong innervation allows a dependable assessment of synaptic activity. Taking cell-wise sPSC average amplitudes multiplied by the number of events occurring within 5 min as a proxy of synaptic innervation, we observed that TeNT eliminated 99.7±0.2% (mean±s.e.m.) of all synaptic input. Co²⁺ reduced synaptic input by 80.1±3.7% (five cells, from five slices from two animals, measured before and after Co²⁺ wash-in). Note that remaining sPSCs under Co²⁺ represent essentially spontaneous, action potential-independent release (mPSCs).

We analysed the effects of TeNT on cell motility in three slices obtained from three animals. Migration of 917 Pax2 cells was monitored prior to treatment (exemplary slice shown in Movie 2) and, within the very same volume elements, another 651 cells were monitored following 120 min incubation with TeNT (100 µg/ml). TeNT resulted in robust reductions of Pax2 cell displacement, (average) speed and, intriguingly, the ratio of displacement and total path length, which provides a measure of directionality (Batschelet, 1981; Benhamou, 2004) (Fig. 6; Fig. S5 for methodological details). Sham incubation (without TeNT) did not bring about such changes (Fig. 6, Fig. S5D). Exemplary tracks of control cells and cells after incubation with TeNT are documented in Movie 3.

Mechanistically, the nonspecific antagonist of voltage-gated Ca²⁺ channels, Co²⁺ (used at 2 mM, 20 min wash-in; three slices from three animals), prevents secretion upstream of the site of action of TeNT. It also avoids the protracted incubation needed with TeNT. Furthermore, any potential nonspecific side effects of these two agents may be predicted to be distinct. Intriguingly, the effects of Co²⁺-reduced synaptic release on cell motility and directionality were virtually identical to those observed following TeNT blockade of exocytosis (Fig. 6).

To probe the loss of directionality following abrogation of presynaptic vesicular release directly and at the high temporal resolution afforded by our sampling rate, we analysed turning angles formed by the track segments (ʻsteps') joining cell positions recorded at subsequent time points, which were separated by ∼42 s (see Fig. S6 for a scheme). These angles provide a measure of directional persistence: for cells with high directional persistence, they are centred about zero degrees. By contrast, tracks of cells with low directional persistence are characterized by highly variable inter-step angles. Under control conditions (Fig. 7A-C), angles between subsequent steps (Δφ) were concentrated about 0° in the coronal and sagittal planes, indicating preferentially directional translocation. Following incubation with TeNT (Fig. 7A′,A″) or wash-in of Co²⁺ (Fig. 7C′,C″), the frequency of Δφ centred about 0° was conspicuously reduced, and larger Δφ values occurred more frequently (Fig. 7A′,C′). This indicates that, following synaptic silencing, directional persistence of migrating Pax2 cells was reduced. This should not be confused with an overall loss of directed migration. However, the loss of directional persistence makes directed migration much less efficient.

Another way to measure directional persistence is by quantifying the turning angles without reference to anatomical planes (Fig. S6, right). This reduces the dimensionality of the data to one and conveniently allows their statistical analysis with Kuiper's test. This confirmed the visual impression conveyed by the density plots that TeNT (Fig. 7A″) and Co²⁺ (Fig. 7C″) had significant effects on directional persistence (P<0.001), whereas sham incubation (Fig. 7B″) did not (P=0.221).

An intriguing question is whether the effects of synaptic activity on Pax2 cell motility are related to the localized nature of synaptic signalling. Conversely, one might invoke the contribution of synaptic activity to ambient transmitter concentrations to explain the effects observed. To address this question, we blocked ionotropic and metabotropic receptors of glutamate and GABA, but also receptors for purines (P2R) and several cytokines and for neurotrophins [TrkA/B/C (NTRK1-3), p75 (NGFR); see the Materials and Methods for an overview of the receptors targeted by the blockers used], which have previously been shown to impinge on neuronal translocation. In contrast to the elimination of synaptic vesicular release by TeNT or Co²⁺, global elimination of transmitter action by receptor blockade resulted, at best, in minor and highly variable effects on all motility parameters assessed (Fig. 6).

Together, these morphological and functional findings document that precursors of ML interneurons in transit through the cerebellar cortex receive GABAergic and glutamatergic synaptic input, and that synaptic input modulates their migratory behaviour. To the best of our knowledge, synapses have so far not been described on migratory neuronal precursors.

The fast rise times of the postsynaptic currents observed in Pax2 cells are also compatible with ectopic release, i.e. release from presynaptic elements in close contact with cells that lack typical postsynaptic morphology (Matsui and Jahr, 2004). Yet, they rule out spill-over transmission. We prefer the term synaptic, as our ultrastructural data document the existence of morphologically bona fide synapses on Pax2 cells. We note that ectopic release, should it be involved, would still constitute a highly localized signal, functionally equivalent to synapses (as discussed by Matsui and Jahr, 2004).

Although it is well established that neurotransmitters or blockade of their receptors affects neuronal migration (Bolteus and Bordey, 2004; Bortone and Polleux, 2009; Cuzon et al., 2006; de Lima et al., 2009; Komuro and Rakic, 1993; Manent and Represa, 2007; Manent et al., 2005), and notably speed, the present data also unveil a hitherto unknown effect of neurotransmitter signalling on the directionality of migration. This effect was specifically observed following ablation of presynaptic release. Thus, presynaptic elements may function as activity-modulated guideposts for migrating interneuronal precursors that navigate the nascent cerebellar cortex. This conclusion is underscored by the outstanding specificity of tetanus toxin for targeting synaptic release. In particular, this agent does not interfere with neurite outgrowth (Verderio et al., 1999), which might interfere with migration, and which has been observed when (synaptic) release was eliminated by ablation of Munc18-1 (Stxbp1) and Munc13 (Broeke et al., 2010).

At first sight, the distinct effects of ablation of vesicular release and receptor blockade might seem counterintuitive. Yet, if one contrasts how TeNT (or Co²⁺) and receptor blockade act on transmitter-mediated signals perceived by migrating cells (Fig. 8), these differences provide a first clue as to how the hitherto unknown synaptic input to migrating neurons may mechanistically control their directionality and locomotion.

Traditionally, effects of transmitters on the motility of neuronal precursors have been surmised to be exclusively due to ambient, paracrine-acting transmitters (Bolteus and Bordey, 2004; Bortone and Polleux, 2009; Cuzon et al., 2006; de Lima et al., 2009; Komuro and Rakic, 1993; Manent and Represa, 2007; Manent et al., 2005), i.e. a non-localized, or isotropic, signal. By contrast, synaptic input, as identified here, results in highly anisotropic signals. Notably, synaptic input onto Pax2 cells does not generate action potentials and thus results in a localized, rather than a propagating, signal. This conclusion may be reached considering that no action potentials, not even spikelets, could be elicited from Pax2 cells by current injections up to 40 pA (see first five sweeps in Fig. 3B, counting from bottom to top), i.e. well beyond the amplitudes of sPSCs recorded from these cells.

Suppression of synaptic release abrogates this anisotropic input to migrating cells. It will eventually also dampen global, isotropic (paracrine) input due to its effect on ambient transmitter concentrations. Yet, as the latter are likely to be regulated in the developing brain primarily by non-synaptic mechanisms (Luhmann et al., 2015; Manent et al., 2005), such as transmitter-permeant anion channels (e.g. Liu et al., 2009), neurotransmitter transporters (possibly working in reverse; see Raiteri and Raiteri, 2015), or insufficient transmitter uptake (by glial cells; e.g. Komuro and Rakic, 1993), synaptic silencing will not effectively shut down isotropic transmitter effects (Fig. 8, middle). By contrast, receptor antagonists eliminate signalling completely and irrespective of its source (Fig. 8, right).

Thus, the distinct effects of blockade of vesicular release and receptors may be rationalized by a model that combines a permissive global (isotropic, paracrine) input with a local (anisotropic, synaptic) instructive signal to regulate neuronal migration and pathfinding. The permissive signal is required to keep the locomotor apparatus of the cell malleable and able to react to a localized (guiding) signal. In the specific absence of localized signals (i.e. following incubation with TeNT or Co²⁺), cell movements become erratic and disoriented. In the absence of all external signals (i.e. following receptor blockade), cell locomotion initially persists due to the internal inertia of its locomotor system; eventually, though, its motility will also deteriorate, as the decay of its locomotor system is no longer balanced by reorganization dependent on external input.

This model is parsimonious and explains all of the observed effects without the need to invoke some enigmatic, synaptically released agent directing Pax2 cell migration. It provides a consistent and parsimonious explanation for both the relatively high variability of the effects seen following receptor blockade, and the observation that receptor blockade does eventually change cell mobility, when observed over more protracted time periods (but also at much sparser sampling rates) as we did here (e.g. Bolteus and Bordey, 2004; Bortone and Polleux, 2009; Cameron et al., 2009; Cuzon et al., 2006; de Lima et al., 2009; Komuro and Rakic, 1993; Manent and Represa, 2007; Manent et al., 2005). Further, the model is conceptually equivalent to the well-established context-dependence of axon guidance signalling, where the attractive or repulsive interpretation of one signal may be switched by a second, permissive signal (or internal state) of the receptive cell (see Naoki et al., 2016 and references therein). Third, this model may also be seen as an instance of a more general model in which contrasting local and global effects are integrated to allow cells to steer in shallow gradients and to avoid adaption (Hecht et al., 2011; Xiong et al., 2010). These conceptual parallels might also help to conceive future research to further identify the molecular mechanisms underpinning the observed migratory behaviour.

To date, the mechanisms that direct the fate, differentiation and spatial dispersal of cerebellar inhibitory interneurons remain elusive. Recent experimental evidence (Leto et al., 2009) revealed that postmitotic precursors of these cells retain a considerable degree of plasticity while in transit from the ventricular zone to their adult positions, and are sensitive to, and dependent on, instructive signals received en route. The present results document that migration of inhibitory interneuron precursors is sensitive to presynaptic vesicular release; more significantly, we show that early synaptic activity provides a directional cue for these cells.

This suggests a novel function for released neurotransmitters beyond those established for neural precursor proliferation (e.g. Liu et al., 2005; LoTurco et al., 1995), motility (e.g. Bouzigues et al., 2007; Manent et al., 2005) and network integration (e.g. Ge et al., 2006; for a review see Spitzer, 2006). We demonstrate that transmitters act on developing cells in transit in a highly localized manner, indicating a role for synaptic innervation in early neural histogenesis.

Abrogation of synaptic release onto Pax2 cells reduced their displacement by some 40% (Fig. 6). This should give an idea of the size of the effect that synaptic activity may have on the migration and dispersal of basket/stellate cells in vivo. Further, the spatiotemporal resolution of the present data allowed us to identify changes in cellular directionality, rather than speed, as the major determinant of displacement. As alluded to above, the robust effect of synaptic input on directionality in terms of directional persistence as described here for Pax2 cells is in keeping with models (Hecht et al., 2011; Xiong et al., 2010) that explain how localized input enables cells to navigate molecular gradients generated by paracrine release and to avoid adaptation.

The migrating Pax2 cells that we analysed (in the P7-9 cerebellar anlage) will eventually settle in the upper part of the ML (Leto et al., 2009, 2011), to which their dendrites will also remain confined (e.g. Rakic, 1972; Sultan and Bower, 1998). There, they will form excitatory synapses with granule cell-derived parallel fibres and inhibitory synapses with each other (for a review and details on the synaptic wiring of ML interneurons, see Schilling, 2013).

Potential candidates providing afferent input to Pax2 cells en route include inhibitory interneurons of the GL and parallel fibres in the lower ML. The latter will not synapse on Pax2 cells analysed here once they have reached the upper ML. However, we consider parallel fibres an unlikely source of developmental synaptic input to migrating Pax2 cells, given that there is no experimental evidence supporting spontaneous or triggered granule cell activity during the critical developmental period analysed here (for a review, see Sassoe-Pognetto and Patrizi, 2013).

GABAergic Purkinje cells, glutamatergic mossy and/or climbing fibres or developmental, intermediate forms of the latter two (Mason and Gregory, 1984) seem more likely candidates to provide (transient) afferents to mobile Pax2 cells. First, they are known to be active during the early postnatal period of cerebellar development (e.g. Puro and Woodward, 1977; Shimono et al., 1976; Sokoloff et al., 2015). Second, at least mossy fibres and climbing fibres are notorious for establishing developmentally transient synapses (e.g. Kalinovsky et al., 2011; Mason and Gregory, 1984; Watanabe and Kano, 2011).

The finding that Pax2⁺ precursors of (upper) ML interneurons are synaptically innervated in areas not targeted by the afferents of their adult forms implies that they experience a switch in synaptic input as they migrate to their final destinations. This switch would differ qualitatively from the well-known developmental synaptic remodelling that occurs on more mature postmigratory cells to assure adequate matching of pre- and post-synapses (Gianola et al., 2003; Kalinovsky et al., 2011; Watanabe and Kano, 2011; Watt et al., 2009).

As the reduced preparation used here preserves only intracerebellar sources of mossy fibres, i.e. those originating in deep nuclei (Batini et al., 1992), the impact of the mossy fibre system on Pax2 cell migration has probably been underestimated.

Our results identify a novel mechanism that allows synaptic activity in long-distance projections to fine-tune local circuit formation by impinging on interneuron positioning during development.

Mice expressing EGFP from the Pax2 locus (Pax2-EGFP; line BAC#30; Pfeffer et al., 2002; Weisheit et al., 2006) were kept as heterozygotes on a C57BL/6 background. All animal handling was performed in accordance with local governmental and institutional animal care regulations.

Mice of either sex (P7-9) were anaesthetised and decapitated. Parasagittal slices (200 or 300 µm) were cut from the cerebellar vermis using a vibratome (VT 1000 S or 1200 S, Leica) and collected in ice-cold buffer comprising 87 mM NaCl, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 7 mM MgCl₂, 0.5 mM CaCl₂, 25 mM NaHCO₃, 25 mM glucose, 75 mM sucrose (347 mOsm) gassed with carbogen. The sections were stored for 30 min at 32°C, cooled to room temperature (RT; ∼24°C) over a period of 30 min, and transferred to artificial cerebrospinal fluid (ACSF, RT) containing 126 mM NaCl, 3 mM KCl, 1.25 mM NaH₂PO₄, 2 mM MgSO₄, 2 mM CaCl₂, 26 mM NaHCO₃, mM 10 glucose.

Slices were viewed with infrared DIC optics. Pax2 cells in lobule IV were identified by their expression of EGFP (using epifluorescence, with a Zeiss Axioskop FS2 or Leica DM 6000 CES microscope) and their distinct morphology. Whereas stationary interneurons of the GL, which also express Pax2, are rounded with a diameter of about 12-15 µm, Pax2 cells are slender with a perikaryon measuring about 12×5 µm. They were typically oriented perpendicular to the PCL while in transit through the cerebellar cortex. Lastly, we note that mature interneurons of the GL, including Golgi cells, showed a characteristic electrophysiological signature and were able to generate action potentials, which unequivocally distinguished them from Pax2 cells. Cells were recorded in the whole-cell mode. During recording, slices were continuously perfused with ACSF, unless stated otherwise. Patch pipettes (borosilicate capillaries) had resistances of 3-4.5 MΩ and were filled with 130 mM KCl, 3 mM Na₂-ATP, 2 mM MgCl₂, 0.5 mM CaCl₂, 5 mM BAPTA, 10 mM HEPES; or with 130 mM CsCl, 3 mM Na₂-ATP, 2 mM MgCl₂, 0.5 mM CaCl₂, 5 mM BAPTA, 10 mM HEPES; or with 125 mM K-gluconate, 2 mM Na₂-ATP, 2 mM MgCl₂, 0.5 mM EGTA, 10 mM HEPES, 20 mM KCl, 3 mM NaCl; or with 125 mM Cs-gluconate, 2 mM Na₂-ATP, 2 mM MgCl₂, 0.5 mM EGTA, 10 mM HEPES, 20 mM CsCl, 3 mM NaCl (all solutions were pH 7.25-7.3; liquid junction potentials corrected for gluconate solutions). EGFP washout was used to confirm that recordings were actually obtained from Pax2 cells. To allow visualization of cells after EGFP washout, internal solutions also contained 0.05% dextran-conjugated TRITC or 0.1% dextran-conjugated Texas Red (both MW 3000, Molecular Probes). Data were recorded with EPC 8, 9 or 800 amplifiers, filtered at 1-3 kHz, sampled at 0.1-30 kHz, and digitized with DAAD converters (ITC 16 or LIH 1600) controlled by TIDA software (HEKA Elektronik). Test pulses were elicited between −160 and +70 mV (before and after adding TTX, 1 µM), subsequent to applying conditioning pre-pulses (to −110 and −10 mV) to remove current inactivation. AMPA/KA receptors were analysed in solutions containing bicuculline (10 μM) or picrotoxin (50 μM); GABA_A receptors were analysed in solutions containing NBQX (10-20 μM) or CNQX (10 μM); both solutions also contained 10 mM BaCl₂, 4 mM 4-AP, 30 µM CdCl₂, 1 µM TTX, 50 µM DL-APV or 20-25 µM D-APV, 135 mM NaCl, 5 mM KCl, 2 mM MgCl₂, 2 mM CaCl₂, 10 mM glucose, 10 mM HEPES (Fig. 3C,D, Fig. S2). Agonists were bath applied. Data were analysed with Igor Pro 6 (WaveMetrics) and MiniAnalysis 6 (Synaptosoft) or PClamp 10 (Molecular Devices). PSCs with amplitudes less than twice the noise level were excluded. All electrophysiological recordings were obtained at RT. Slices for these analyses were obtained from a total of ∼100 animals, and all measurements reported are based on values obtained from at least three slices obtained from animals taken from distinct litters. More typically, data were obtained from 8-12 slices from 5-7 animals from distinct litters. All data recorded were used for analysis.

Migration of Pax2 cells was observed in lobules IV and V of a total of 14 acute slices (300 µm) prepared from 14 animals from distinct litters. Slices were kept in modified, 35°C ACSF containing 132 mM NaCl, 3 mM KCl, 1.25 mM NaH₂PO₄, 2 mM MgCl₂, 2 mM CaCl₂, 20 mM NaHCO₃, 10 mM glucose. We blocked presynaptic release by adding either TeNT (100 µg/ml; three slices) or Co²⁺ (2 mM; three slices) to ACSF. One slice was used for sham incubation, and one for long-term observations (up to 4 h) to check for run-down of our preparation. To block postsynaptic receptors, we combined antagonists against ionotropic glutamate receptors of the AMPA/KA (NBQX, 20 µM) and NMDA (D-APV, 25 µM) subtypes, metabotropic glutamate receptors (group I-III: LY341495, 5 µM; mGluR5: MPEP, 10 µM; mGluR1: 3-MATIDA, 40 µM), ionotropic GABA_A receptors (gabazine, 20 µM), metabotropic GABA_B receptors (CGP55845, 4 µM), ionotropic purinergic receptors (P2X_1-6: PPADS, 200 µM), metabotropic purinergic receptors (P2Y_1-12: suramin, 100 µM; P2Y_{1,2,4,6,11,12}: MRS2279, 0.5 µM), cytokine receptors including TrkA/B/C, p75 and others (NGF inhibitor: Ro 08-2750, 5 µM; pan-Trk inhibitor: GNF5837, 0.5 µM; JAK2, FLT3 and TrkA inhibitor: lestaurtinib, 0.5 µM). For these pharmacological studies, we analysed cells from a total of six slices from six animals from distinct litters.

Time-lapse recordings were performed either in x,y,t (3D) with a Zeiss confocal LSM5 Pascal [argon laser (Lasos), excitation at 488 nm], or in x,y,z,t (4D) with a Leica LSM TCS-SP5 and two-photon excitation [pulsed IR laser (Mai Tai BB, Newport/Spectra Physics), excitation at 980 nm]. For 3D time-lapse recording, images (217×217 µm²) were taken every 10 s for up to 4.5 h from an individual Pax2 cell, which was manually kept in the focal plane. In 4D time-lapse recording, volumes of 456×456×30 µm³ were imaged every 42 s up to 4 h. If the z-drift was >4 µm, z-settings were readjusted. Offline xyz-drift correction was based on the stationary Golgi cells. For quantitative analysis of Pax2 cell migration, centres of mass calculated from isosurfaces were tracked over time (Imaris 7.5.1, Bitplane). Only cells that were visible during at least 80% of the observation time were included. Speed was calculated from the distance the cells travelled between two successive observation points. Displacement represents the straight line distance between the start and end points of a cell trajectory, and directionality is the ratio of displacement and length of the entire trajectory.

Angles between subsequent migratory steps were measured in the sagittal plane (i.e. the plane of slice sectioning) and in the coronal plane. For calculation of probability density plots, we accounted for the fact that angular changes measured in the sagittal plane are continuous at ±π radians, and those in the coronal plane at ±0.5 π radians. Data-based smoothing kernels were determined as described by Jones et al. (1996) and Sheather and Jones (1991).

Transgenic mice expressing EGFP under the control of the Pax2 promoter were anesthetized and perfusion fixed transcardially with 0.01 M PBS (pH 7.4) containing 4% paraformaldehyde (PFA) and 0.05% glutaraldehyde at P8. Cerebella were sectioned using a vibratome (50 µm). Pre-embedding immune-peroxidase staining and electron microscopy were performed as previously described (Boulland et al., 2003). Briefly, sections were treated with 1 M ethanolamine-HCl (pH 7.4) and 1% H₂O₂, followed by incubation overnight with rabbit anti-GFP antibodies (1:1000; Rockland Immunochemicals, 600-401-215) in 10% newborn calf serum in 0.1 M Tris/HCl (pH 7.4) containing 0.3 M NaCl. The sections were then incubated with biotinylated anti-rabbit antibodies, streptavidin-peroxidase complexes, and finally with diaminobenzidine and 0.1% H₂O₂. Samples of stained tissue were dissected out, treated with 1% OsO₄ in phosphate buffer, dehydrated in a graded ethanol series and propylene oxide and embedded in Durcupan ACM (Sigma). Subsequently, ultrathin sections were cut and lightly contrasted with 2% uranyl acetate and 0.3% lead citrate. The samples were examined with a Tecnai CM10 electron microscope (FEI).

Alternatively, the DAB reaction product from immune labelling for GFP was intensified with a modified gold-substituted silver peroxidase technique (van den Pol and Gorcs, 1986). After washing DAB-reacted slices with 0.1 M Na cacodylate buffer (CB; pH 7.6), they were fixed in 0.2% glutaraldehyde in CB, and then osmicated in CB with 1% OsO₄ and 1.5% potassium hexacyanoferrate for 15 min. After dehydration in a graded ethanol series, they were flat-embedded in Epon. Semi-thin sections (1 μm) were cut on an LKB Ultratome, stained with Toluidine Blue and used for orientation. Ultrathin sections (50-100 nm) were collected on Formvar-coated grids, counterstained with uranyl acetate and lead citrate, and examined with a Philips CM 100 transmission electron microscope (FEI).

PFA-fixed slices (50 µm) were blocked with 2% normal goat serum in PBS for 2 h and incubated at 4°C with antibody G95 against synaptophysin (1:500) (Ikin et al., 1996) or against calbindin D28K (1:4000; Swant, #300) overnight. Alexa-546 goat anti-rabbit or goat anti-mouse secondary antibodies (1:500; Molecular Probes) were applied for 2 h. Confocal images were acquired using a Leica TCS SP2 system (excitation, 488 nm/543 nm; z-width, 115 nm). To obtain a measure of the density of presynaptic elements in cerebellar layers, we measured the fraction of pixels indicating synaptophysin immunoreactivity separately for each layer. To assess whether Pax2 cells might be contacted by synaptophysin⁺ terminals, we first generated a mask based on the EGFP signal to delineate Pax2 cells. Next, colocalization of EGFP and synaptophysin immunoreactivity was quantified as described by Costes et al. (2004). Computation was performed with Imaris.

NBQX, D-APV, LY341495, lestaurtinib, GNF5837, Ro 08-2750, PPADS, suramin, CGP55845, MRS2279, 3-MATIDA, MPEP, gabazine and ionomycin were from Tocris; CNQX, bicuculline and muscimol from Ascent Scientific; DL-APV, KA, cyclothiazide, GABA and picrotoxin from Sigma; TTX from Alomone Labs; and tetanus toxin from Calbiochem.

We thank R. Fimmers for advice on statistics and D. Krauss and N. Haias for animal husbandry.