GABAergic neurons of the cerebellar cortex comprise projection PCs and local INs. All these cells derive from progenitors pools (PCPs and PIPs, respectively) located in different VZ subregions (see “Glutamatergic Phenotypes” section).
Purkinje Cell Migration (T. Miyata)
Morphologically, PCPs span the ventricular (apical) and pial (basal) surfaces of the cerebellar primordium, taking an elongated (neuroepithelial or radial glial) shape . How PCPs behave is not well understood: whether they divide symmetrically or asymmetrically, how they undergo interkinetic nuclear migration, and the morphology of newly generated PCs in the VZ all remain unknown. Also, whether (and how) a single class of PCPs changes in a temporally regulated manner to generate the different PC subclasses or whether PCPs are inherently heterogeneous within VZ needs to be studied.
Nascent PCs from E14 in mice form a layer several cells thick called the PC plate [128, 129]. This transient structure normally spreads through the presumptive cerebellar cortex until late embryonic days, followed by PC-monolayer formation during the early postnatal stage triggered by RELN signaling (see “Development of Cerebellar Compartmentation” section). The supply of PCs to the cortex occurs sequentially [129, 130]. PCs generated at E10, especially ones born at the posterior VZ close to the RL, are the first to finish their migration and form the plate [131, 132]. This may be because the distance between their birthplace and the goal is the least among the entire PC population (no more than ~200 μm). PCs generated at E11 join the plate from E15, followed by E12-born PCs from E16. These later-born PCs (as well as PCs generated anteriorly at E10) need to migrate over much longer distances than the posteriorly E10-born PCs (up to 700–800 μm in mice).
The early/posterior-born PCs take a tangential migratory route (which is parallel to the pial surface) until E13. They are characterized by long leading processes (100 μm or longer), which morphologically and molecularly resemble axons, and a much shorter trailing process containing the Golgi apparatus. This tangential migration crosses radial glial fibers and is reminiscent of that exhibited by RL-derived cells. The initial departure of this early/posterior-born population from the VZ and the subsequent tangential migration are normal in the cerebellum of reeler mice. The early/posterior-born PCs then change orientation by sending the original Golgi-rich trailing process into the cortical region that intensely expresses RELN (produced at E13 by RL-derived tangentially migrating cells of the nuclear transitory zone, NTZ) . This switchback-like, tangential-to-radial orientation change between E13 and E14 is strictly dependent on RELN. These observations suggest a short-range action of RELN. This model is supported by co-culture experiments showing that PCs (both normal and reeler-derived) will align in vitro along an artificial RELN-rich zone . However, transgenic reeler mice artificially expressing RELN under the control of the nestin promoter showed an apparently normal (rescued) arrangement of PCs , suggesting that RELN may regulate PC behavior in a context-dependent manner.
In contrast to the posterior-born PCs, PCs born more anteriorly exhibit radially oriented somatal morphologies during migration. The close spatial relationship between these PC somata and radial glial fibers supports the prevailing model that PC migration is guided by radial glial fibers [129, 135]. These radially oriented PCs at E12 or E13 have axon-like fibers that ascend towards the pia , while post-migratory (late embryonic) PCs forming the plate have downward-directing axons [129, 130, 136]. Several points remain to be elucidated: how these anterior-born PCs proceed the mid-embryonic migration processes, whether these cells are also affected by dynamic changes of cellular orientation or polarity, and how RELN contributes to histogenesis by these radially oriented PCs.
Development of Cerebellar Compartmentation (M. Arancillo, R. Hawkes, R. V. Sillitoe)
The fundamental architecture of the cerebellum is an elaborate pattern of transverse zones and parasagittal stripes [137, 138] that is highly reproducible between individuals and conserved across birds [139, 140] and mammals (reviewed in [141, 142]). Compartmentation is revealed by intrinsic differences between subsets of PCs (e.g., zebrin II/aldolase C (ZII ; phospholipase C (PLC) β3/4 ; HSP25 ), the restriction of INs (reviewed in ), patterns of pathological PC death (reviewed in ), the phenotypes of multiple cerebellar mutants (e.g., lurcher (Grid
Lc)—; rostral cerebellar malformation (Unc5c
rcm)—; weaver (Kcnj6
wv)—; cerebellar deficient folia (Ctnna2
cdf)—), and the topography of afferent and efferent projections (reviewed in ).
Cerebellar compartmentation appears to start at ~E10 in the VZ of the fourth ventricle but not earlier, e.g., [153–156]. PC subtype specification likely occurs when PCs undergo terminal mitosis between E10 and E13  in the Ptf1a-expressing progenitor domain of the VZ ([23, 34, 49]: see “Glutamatergic Phenotypes” section). Birthdating studies have identified two distinct PC populations: an early-born cohort (E10–E11.5) destined to become zebrin II (ZII)+ and a late-born cohort (E11.5–E13) destined to become ZII−. A direct correlation is also found between PC birthdates and their adult stripe location, suggesting that both subtype specification (e.g., ZII+ vs. ZII−) and positional information (which zone or stripe) are acquired at this time, e.g., [30, 157–160]; both are cerebellum-intrinsic and not activity- or afferent-dependent, e.g., [161–164]. However, there is no reason to believe that individual PC stripes have a clonal origin. During the same interval, the cerebellar anlage undergoes a 90° rotation, which converts the embryonic rostrocaudal axis into the mediolateral axis of the cerebellar primordium . This suggests the possibility that the adult mediolateral stripe array derives from the anteroposterior patterning of dorsal r1.
A Ptf1a-Neurogenin 1/2 (Neurog1/2)-Early B-cell factor 2 (Ebf2) regulatory network is implicated in PC subtype specification . By this model, the early-born PC cohort expresses neither Neurog1/2 nor Ebf2 and therefore expresses the ZII+ phenotype in the adult. Soon after E11, Neurog1/2 is upregulated by Ptf1a in the later-born PC progenitors (e.g., ). In this context, neurogenin 2 regulates cell-cycle progression, neuronal output, and early dendritogenesis of PC progenitors , but neither Neurog1 nor Neurog 2 deletions affect cortical patterning (Hawkes, unpublished observation). In turn, Neurog1/2
+ precursors express EBF2, which represses the ZII+ phenotype ([132, 159]: Ebf2 deletion results in transdifferentiated PCs that express markers characteristic of both the ZII+ and ZII− subtypes—the only manipulation known to alter a PC subtype phenotype). In addition, Ebf2 plays an anti-apoptotic role in ZII− PCs by locally regulating Igf1 gene expression . As a result of these events, early-born PCs become ZII+ in the adult and late-born PCs adopt the ZII− phenotype.
Postmitotic PCs migrate dorsally from the VZ, in part along radial glia processes (: also see Section 4.1), and stack in the cerebellar plate with the earliest-born (Ebf2−) PCs located most dorsally. Starting at ~E14, the cerebellar plate undergoes a complex redisposition, such that by E18 a reproducible array of clusters of multiple molecular phenotypes is present on each side of the midline ([166–170], etc.: reviewed in [171, 172]) . The cellular processes that guide cluster formation are not understood but grafts of dissociated PCs also organize into discrete ZII+/− compartments , pointing to cell-cell adhesion molecules as possible organizers: cadherins are strong candidates reviewed in . Over 50 distinct clusters have been identified [175, 176]. The mapping between embryonic clusters and adult stripes is complex: in some cases, one cluster yields a single stripe, e.g., [175, 177, 178], but in others single stripes derive from the fusion of several clusters, e.g., , or single clusters split into multiple stripes, e.g., .
The embryonic cluster architecture is the scaffold around which other cerebellar elements are organized. First, clusters likely restrict the distribution of cerebellar INs and their processes (GCs—; Golgi cells—; stellate/basket cells—; UBCs—: reviewed in : also glia—e.g., [183, 184]). Secondly, most afferent projections also enter the cerebellum between E14 and E18 and target specific PC clusters (the so-called matching hypothesis—reviewed in ). There are two major sources of sensory input to the cerebellum: CFs from the inferior olive, and mossy fibers from a number of brain and spinal cord regions. Both afferent systems invade the cerebellum at around E 13/14 in mouse [186, 187], and thereafter they terminate with a spatial organization that parallels the pattern of PC stripes [137, 188] (Fig. 6a, b). PC subtype organization is thought to play a key role in instructing circuit wiring into topographic maps. Spontaneous and engineered mouse mutants that display disrupted PC stripes have equivalent alterations in the spatial arrangement of mossy fiber and CF terminals [189–191]. But, what molecular mechanisms trigger cluster dispersal and wiring?
From ~E18, the embryonic PC clusters disperse, triggered by RELN secreted by the EGL [192–194]: recently thoroughly reviewed in . RELN is critical as its deletion in the reeler mouse (Reln
rl: [192, 196] blocks dispersal, e.g., [128, 197]). One model is that secreted RELN binds two surface receptors on PCs—Apolipoprotein E receptor 2 and the Very Low Density Lipoprotein Receptor. Deletion of either receptor (Lrp8
tm2Her or Vldlr
tm1Her) produces a partial, stripe-specific disruption of cluster dispersal : deletion of both receptors blocks dispersal . In turn, RELN binding induces receptor clustering  and activates two Src-family kinases—Fyn and Src [201, 202]—which tyrosine phosphorylate [203, 204] the intracellular adaptor docking protein Disabled1 (DAB1—[205–209]: phosphorylation is essential and key tyrosine point mutations phenocopy reeler—). DAB1-phosphorylation results, at least in the neocortex, in Repressor activator protein 1 (Rap1)-mediated homophilic cadherin 2 cell-cell interactions that promote neuronal migration [211, 212]. The upshot is that by ~P20, as cerebellar lobulation matures, the PC clusters string along the rostrocaudal axis into the adult array of long parasagittal stripes. However, one critical question is how do these dynamically changing stripes acquire their functional properties during development?
During postnatal development, chemo- and activity-dependent mechanisms may play important (and possibly distinct) roles in establishing the afferent topographical map. There is a long-standing hypothesis that first a “crude” topographic map is established by genetic cues. And one of the most compelling hypotheses postulates that cues in the afferent source domains match up with cues in the PC map . The molecular cues that would mediate this mechanism have not been resolved, although it may involve a chemoaffinity mechanism mediated by eph/ephrin and cadherin signaling. But, one has to also consider how the circuit is then sculpted into a “fine” map, and does the mechanism for refining the map also involve PC patterning? Alternatively, are there non-genetic mechanisms that also contribute? Indeed, it was recently shown that the striped patterning of PCs is disrupted when neurotransmission in the PCs themselves is selectively silenced . Interestingly, the patterning of spinocerebellar mossy fiber terminals into distinct stripes was also altered in the absence of PC activity. The sharp stripe boundaries that are typically observed were severely compromised, although the basic features of the topographic map were left intact. These data argue that neuronal activity may play an important role in fine-tuning the cerebellar map into topographic domains. But maybe PCs are not the only players. Perhaps inhibitory INs in the developing molecular layer (ML) or the millions of excitatory GCs also influence circuit topography by modulating the levels of PC activity. INs, after all, are also organized into stripes  and they integrate into the PC microcircuit at a time when they could have a powerful affect on how the circuit is firing when it is wiring .
Which features of neuronal firing need to be examined in order to understand how circuits assemble? PCs provide the best starting point since these cells exhibit two distinct types of action potential that are experimentally tractable with in vitro and in vivo paradigms. The first are complex spikes that are triggered by CF inputs, and the second are simple spikes generated intrinsically within the PCs but modulated by mossy fiber inputs (Fig. 6c–cʺ, d–dʺ). Recent work in anesthetized and awake mice used in vivo electrophysiology to record PC activity in postnatal mice . The study found that the rate of complex spike firing increased sharply at 3 weeks of age, whereas the rate of simple spike firing gradually increased until 4 weeks of age. They also found that compared to adult, the pattern of simple spike firing during development was more irregular as the cells tended to fire in “bursts” that were interrupted by long pauses (Fig. 6c, d). The regularity in simple spike firing only reached maturity at 4 weeks of age. These data show that PC activity is dynamically sculpted throughout postnatal development, traversing several critical events that are required for circuit formation. Importantly, the establishment of PC firing properties seems to overlap with the final stages of stripe maturation . However, there are also data that support an alternate view which suggests that activity-dependent mechanisms may not be involved in topographic wiring. Surgically lesioning the neonatal spinocerebellar afferent tracts did not induce a competitive sprouting of the adjacent cuneocerebellar pathway  and the regression of supernumerary CFs appears spatially and temporally unlinked to the formation of stripe patterns—both processes may be strictly dependent on molecular cues . Therefore, it is interesting to speculate that compared to molecular cues, the rate and pattern of spikes, at particular ages, could shape a more subtle level of topography by refining the connectivity within stripes rather than between individual stripes. If this was to be the case, then chemical tags might define the fundamental patterns of stripes irrespective of sensory experience, and perhaps only later during postnatal development is when activity might tune and complete the existing map.
GABAergic Interneurons (K. Leto, E. Parmigiani, K. Schilling, A. Wefers)
The term “GABAergic interneuron” is traditionally used to refer to a diverse set of neurons that, in the healthy adult cerebellum, result in local inhibition (Fig. 7). Actually, it is somewhat of a misnomer since in CN and the GL, the majority of GABAergic INs also use glycine as a co-transmitter, and a few are strictly glycinergic [217, 218]. Distinct differences in their afferent and efferent wiring, their morphology, and the differential expression of a set of molecular markers (inter alia mGluR2 and neurogranin) allow us to distinguish, in the GL, at least four discrete sets of Golgi cells sensu stricto as well as Lugaro and globular cells. While these clearly distinct inhibitory INs of the GL are sometimes collectively referred to as Golgi cells, INs of the ML are traditionally classified as basket or stellate cells. Today, the weight of the evidence rather supports the view that these two terms describe two exemplary variants of one population that shows gradual morphological  and molecular  differences which may be secondary to their position in the ML (discussed in ). Lastly, the still rather enigmatic candelabrum cell  may be yet another member of the basket/stellate class (see  for a broader discussion).
GABAergic INs are produced from late embryonic life to the second postnatal week, according to a precise inside-out sequence (Fig. 7). The proliferation of inhibitory INs peaks during the first postnatal week and exclusively occurs in the PWM, a postnatal niche containing heterogeneous cell types at different maturation stages [29, 32, 223–227]. CN INs are the first to be born during embryonic and early postnatal life, followed by GL INs (Golgi and Lugaro cells) and, finally, by those of the ML (basket and stellate cells; [29, 31, 225, 228]).
The first studies of cerebellar neurogenesis postulated that ML INs derive from the EGL, the only germinal layer known at that time to be active during postnatal development [229, 230]. Later, analysis of chick-quail chimeras, transplantation experiments, and retroviral injections demonstrated that the EGL exclusively generates GCs and indicated that all the GABAergic neurons, including the ML INs, derive from the VZ [154, 231–234]. While these diverse cells all originate from the VZ, there is little evidence that this germinal layer is pre-patterned, say by differential gene expression, such as to presage the diversity of cerebellar inhibitory INs in the adult. One possible exception may be the expression of Neurog1, which may distinguish the lineages leading to cerebellar cortical and CN inhibitory INs ([235, 236]: the interpretation of these results is ambiguous because the BAC used may not faithfully recapitulate cognate Ngn1 expression, specifically in CN ). In contrast, there is compelling experimental evidence that PIPs maintain a high degree of plasticity and acquire their definitive fate only as they migrate through the deep cerebella mass (i.e., the PWM) on their way into the CN or the nascent cerebellar cortex . This is particularly striking as individual subsets of cerebellar cortical inhibitory INs withdraw from the cell cycle over an extended period from E13 through the second postnatal week. However, cells collected at any point through this long generative phase are capable, when heterochronously transplanted, of acquiring a fate temporally appropriate to the host tissue . Significantly, fate determination by transplanted cells is critically dependent on migration through the recipient’s PWM. These findings have led to the recognition of the nascent white matter as an instructive niche critical for the maturation and diversification of cerebellar inhibitory INs . Unfortunately, we are still quite ignorant as to how this instruction is realized on the molecular level. There is experimental evidence that PC-derived SHH regulates proliferation of precursors of cerebellar inhibitory INs in and near the ventricular epithelium [33, 238]; yet, whether PCs also influence the diversification of these cells remains unclear.
The dividing intermediate progenitors that are responsible for the extensive amplification of the inhibitory interneuron populations strongly express Ptf1a . At later stages, the precursors of inhibitory INs are seen as a population of Pax-2+ cells that appears in the VZ around E12.5 and later moves into the cerebellar parenchyma [32, 228].
The precise lineage relationships linking Ptf1a+ interneuron progenitors to the other precursor pools in the PWM (namely, progenitors of astrocytes and oligodendrocytes) remained obscure for many years. The existence of multipotent progenitors in the postnatal cerebellum has been proposed by two independent studies [239, 240], in which the isolated progenitors were able to form neurospheres and to differentiate into neurons, astrocytes, and oligodendrocytes both in vitro and in vivo after grafting into newborn mice [239, 240].
A series of more recent studies have clarified the properties of PWM progenitors and the relationships between the different lineages. For instance, it has been shown that the postnatal cerebella of mice lacking the proneural gene Ascl-1 have fewer PAX-2+ INs and increased numbers of SOX9+ astrocytes compared to controls . Conversely, overexpression of ASCl-1 in the VZ results in more INs at the expense of astrocytes , suggesting the existence of lineage relationships between these cell types throughout cerebellar development. It is not clear whether similar relationships between INs and astrocytes are also present at earlier embryonic ages. Indeed, previous fate mapping analyses tagging embryonic progenitors producing PCs or INs only rarely generated astrocytes [34, 50], suggesting that the bulk of cerebellar astrocytes do not derive from these progenitors. Subsequent lineage analysis during embryonic and postnatal development showed that both INs and astrocytes derive from ASCL-1+ precursors . Similarly, fate-mapping studies of progenitors with astroglial traits—such as the expression of hGFAP , TenascinC , and GLAST —reveal that the progeny of such progenitors often comprise a mixed population of astrocytes and GABAergic INs. In particular, in the study of Fleming et al. , a population of primary CD133+TenC+ astroglial progenitors was identified in the PWM as the putative source of both PTF1-a+ intermediate IN progenitors and CD15+ intermediate astrocyte precursors. The existence of these bipotent progenitors has been confirmed by a recent study in which mixed clones of INs and astrocytes were derived from PWM astroglial progenitors both in vitro and in vivo .
Finally, PCs are critical for the terminal differentiation and morphogenesis of cerebellar INs. Specifically, the complexity of basket/stellate cell axonal arborizations and their positioning on PCs is critically dependent on neurofascin [243, 244] and also Semaphorin (Sema3a)/neuropilin-1-mediated signaling between PCs and differentiating ML INs . Further, the preferential orientation of basket/stellate cell axons in the sagittal plane may be due to PC guidance, as they extend in the ML . Conversely, dendritic differentiation of basket/stellate cells appears primarily sensitive to GC-derived input, including BDNF  and signaling through the GluD1 receptor. In fact, ablation of this receptor, which is highly expressed in ML INs and concentrated at their synapses with PFs, results in reduced survival and stunted growth of early post-migratory ML INs .
Development of the Cerebellar Nuclei (R. J. T. Wingate)
The description of cerebellar development above is focused on the cerebellar cortex. The CN develop in parallel, using much of the same molecular machinery. The CN dictate the participation of the cerebellum in a range of circuits by providing an almost exclusive efferent connectivity via axon pathways to more caudal structures (from medial nuclei) and more rostral structures projections (from lateral nuclei; ). Their largely spontaneously active output can be excitatory or inhibitory and nuclei contain several different local interneuron types [249–253]. Nuclei receive collateral input from afferents to the cerebellar cortex (CFs and mossy fibers) in addition to inhibitory input from PCs. Decoding both the development and integrative function of nuclear circuits is likely to be critical for understanding the broader function of the cerebellum. Despite their functional significance, our knowledge of CN development is incomplete. Furthermore, the number of CN varies between tetrapod species suggesting that the developmental mechanisms responsible for their patterning are a key locus of evolutionary adaptation .
The perhaps surprising absence of a detailed description of CN specification and maturation can to some extent be explained by a major conceptual revision of their development over the last 10 years following insights from a series of studies of gene expression [51, 60] and genetic fate-maps [25, 26, 34, 255]. Prior to this, CN, which condense initially in a “nuclear transitory zone” at the margin of the cerebellar anlage, were assessed by birthdate to be derived exclusively from the VZ . This assumption was overturned by genetic lineage maps using an Atoh1 reporter showing that excitatory CN neurons arise by tangential migration from the RL prior to GCP production [25, 26]. More recent mapping using a Ptf1a reporter  reveals that the VZ gives rise only to inhibitory neurons of the CN. Hence, just as for the cerebellar cortex, the assembly of CN is defined as the coordinated integration of PTF1a+ and ATHO1+ lineages in local circuits. However, patterns of temporal specification in either lineage suggest important differences compared to the cerebellar cortex in how these lineages interact.
Ptf1a-derived CN neurons are generated as the first of a sequence of inhibitory neurons destined for progressively more superficial fates . To some extent, this reflects a progressive dorsal expansion of a Gsx1+/Ptf1a domain that gives rise to INs, which occupy the Olig2/Ptf1a domain and give rise to PCs (see “Specification of Cerebellar Progenitors” section). However, fate determination in the former Pax-2
+ lineage is notably a product of local microenvironmental factors. Migrating PIPs persist in the white matter, continuing to contribute to the CN ([29, 36]: see “GABAergic Interneurons” section) well beyond the production of the first, inhibitory projection neurons (Fig. 8). By contrast, cell fate in the ATHO1+ CN derivatives of the RL not only correlates with birthdate [25, 26, 257] but appears determined at the RL . This raises the possibility that the temporal pattern of RL derivatives establishes a template around which GABAergic neurons are organized.
The allocation of a temporal framework of RL-derived CN components is accompanied by a characterized sequence of transcriptional maturation [51, 60] that results in a first born LHX9+ lateral nucleus (projecting to midbrain and thalamus), followed by a TBR1+ medial (fastigial) group, which sends axons to the hindbrain via the fasciculus uncinatus, or hook bundle . The progressive deposition of cells in more dorsal (ultimately medial) positions reflects a decreasing sensitivity to netrin signaling from ventral midline in migrating cells [259, 260]. Netrin receptors are also responsible for determining the laterality of the projections of CN axons , which extend seamlessly from the leading processes of migrating cells . Target selection itself (rostral or caudal CNS) appears to be a property of LHX9  versus TBR1  expression.
If this process of RL cell fate specification provides a template for nucleus assembly, it places a special emphasis on poorly understood events at the NTZ. It is here that neurons segregate into a series of mediolaterally distributed nuclei as they are subducted under the rapidly expanded cortex by either passive displacement, or possibly active translocation . To understand whether lineage interactions are part of this process, let alone underlying mechanisms of nucleogenesis, will rely on a better description of neurons subtypes. For example, it is unclear whether only more lateral nuclei contain inhibitory projection neurons [252, 262] or indeed how much of the repertoire of neurons seen in the lateral nucleus is recruited to other nuclei . Similarly, the embedded nature of nuclei within a series of re-entrant loops that include the inferior olive, in addition to pontine neurons and PCs, has implications for later developmental events. How does the convergence of afferents onto nuclei influence the organization of INs into precise, geometric functional units ? These elements of fine-grained developmental detail, which are so significant for the function of the cerebellar circuit, are almost completely unexplored.
Gliogenesis in the Cerebellum (A. Buffo)
In contrast to other CNS areas in which gliogenesis follows neurogenesis, in the cerebellum the generation of glia parallels the generation of GCs and INs. What triggers the activation of gliogenesis and regulates its course in this territory is still poorly understood.
In the mature mammalian cerebellum, four astroglial subtypes are classically distinguished, including fibrous astrocytes in the white matter (WM), stellate multipolar astrocytes with profuse tiny processes (velate astrocytes) or more slender morphologies (protoplasmic astrocytes) in the GL, and neuroepithelial cells displaying radial BG basal processes spanning from the cells bodies in the PC layer through the entire ML, up to the subpial basement membrane (Fig. 9a, [130, 229, 264]). In the future, this classification may expand to comprise more astroglial subtypes based on neurochemical, topographical, and morphological criteria, as shown by a recent detailed investigation on the human cerebellum .
Through the comparative analysis of different mammalian species, Ramón y Cajal  proposed that cerebellar astroglia derive from the VZ. Cajal’s interpretation has been fully proved by means of fate mapping analyses in mice in which radial glial progenitors (RG) at the ventricle were targeted based on the expression of stem-cell markers [266–268], or VZ-restricted tags . A small contribution of the RL to cerebellar astrogliogenesis has also been proposed  but so far remains controversial (see  and references therein). As showed by anatomical investigations and functional manipulation of regulatory pathways, in a first astrogliogenic wave (up to about E14 in mouse), RG detach from the ventricle and displace the cell body towards the nascent PWM, transforming into progenitors of BG ([229, 266, 271, 272]: the relationship between astrocytes and GABAergic INs is reviewed in “GABAergic Interneurons” section). A subset of these precursors appears to be already postmitotic at the moment of translocation  and readily differentiates into BG. Conversely, other precursors form a proliferative layer that expands [130, 266, 271] up to the first postnatal week in parallel with the tangential expansion of the cerebellar surface. At later stages of embryonic development, a second wave of astroglial-like progenitors lacking the basal process delaminates from the VN into the cerebellar PWM where they proliferate, forming astrocytes populating the prospective GL and white matter [130, 271]. Whether PWM astroglial progenitors also produce some BG or whether proliferating BG contribute astrocytes to other cortical layers remains to be established. Similarly, the dynamics and timing of the amplification of astroglial progenitors are largely unknown.
How the specialization of the astroglial subsets is achieved is only partially clarified. BG morphogenesis requires a tight and timely regulated interaction with the surrounding cerebellar microenvironment (basement membrane, PCs, migratory and proliferative GCs, see Fig. 9b). Notably, impairment of these regulatory mechanisms results in BG malpositioning and/or the acquisition of a stellate morphology, which may thus represent a default differentiation pathway for cerebellar astroglial precursors. Yet, it is likely that the refinement of the variety of multipolar morphologies in the GL and WM is instructed by local cues. Further, few intrinsic determinants are known that take part in the establishment of distinct astroglial cerebellar phenotypes (see ).
In contrast to astrocytes, so far no evidence clearly demonstrates the derivation of oligodendrocytes from the cerebellar VZ . Rather, a minor oligodendroglial fraction appears to derive from progenitors in subventricular positions, likely residing in the PWM [31, 273]. Alternatively, mouse transplantation experiments indicate an extracerebellar origin for the majority of oligodendrocytes . In line with these data, experiments in chick-quail chimeras and in ovo transplants in the chick brain demonstrated that cerebellar oligodendroglia are generated in the chick mesencephalic neuroepithelium and only subsequently invade the cerebellum via the velum medullare . A similar extracerebellar source in the mammalian brain remains to be identified. Further, fate-mapping analyses  support the hypothesis that the majority of cerebellar oligodendrocytes have no lineage relationships with cerebellar astrocytes (and neurons). However, ex vivo experiments  and functional deletion of the polycomb group protein Bmi1  pointed to the existence of bipotent gliogenic precursors, whose identity remains to be established.
Once settled in the cerebellar primordium, oligodendrocyte progenitors first surround the CN and gradually invade the nascent cortical lobules, progressing in a centrifugal direction. The same pattern is reflected in the course of both differentiation and myelination that proceed from the inner cerebellar portions to the lobule apices [276–279].
Despite the fact that most intrinsic and extrinsic mechanisms regulating oligodendrocyte differentiation appear to be common to multiple CNS sites, including the cerebellum [280, 281], a particular role in the regulation of the maturation of cerebellar oligodendroglia is exerted by thyroid hormones (TH; L-triiodothyronine, T3; L-tetraiodothyronine, thyroxine, T4; see “The Role of Thyroid Hormone in Cerebellar Development” section) and PC-derived signals. In particular, PCs secrete SHH, which stimulates oligodendrocyte progenitor proliferation at early postnatal stages, whereas by the end of the first postnatal week, they start producing vitronectin, which drives oligodendrocyte maturation and myelin formation .
Extrinsic Regulators of Cerebellar Development: The Role of SHH (C. Chiang)
The SHH pathway has been extensively studied in the context of GCP proliferation. However, more recent studies have revealed additional roles for this important pathway during different phases of cerebellar development. The common theme emerging from these studies is that SHH is a key mitogen for the expansion of functionally diverse neuronal and glial cell types from spatially and temporally restricted precursors. However, the mechanisms by which SHH stimulates proliferation of these precursors appear to be distinct, involving both cerebellar and extracerebellar strategies (Fig. 10).
In the cerebellum, Shh expression is restricted to PCs starting at E16.5 and continuing throughout adulthood [42, 283]. The early phase of Shh expression is critical for rapid clonal expansion of GCPs as blockade of Shh expression in PCs leads to drastic reduction of GCP number in the EGL [62, 283–285]. The expansion of GCPs requires the cell surface proteins BOC and GAS1 , which synergistically promote SHH binding to its receptors PTCH1/2. In the absence of Shh, PTCH1 functions as a negative regulator of SHH signaling by suppressing the activity and localization of a seven-pass transmembrane protein, smoothened (SMO) to the primary cilium, a slim, microtubule-based non-motile structure that projects from the surface of nearly all vertebrate cells [287, 288]. Therefore, SHH binding to PTCH1 relieves SMO from inhibition, triggering SHH signaling and subsequently activating downstream target gene expression mediated by the GLI family of transcription factors . Accordingly, mutations in ciliary components that disrupt SMO localization to the tip of the cilium all lead to altered SHH signaling and reduced GCP proliferation [290, 291]. Among the three GLI proteins, GLI2 acts as the primary transcriptional activator in GCPs . In addition to canonical GLI target genes Gli1, Hhip, and Ptch1, several others including MycN and CcndD1 involved in cell-cycle regulation of GCPs have been reported [293, 294]. Notably, cerebellar phenotype in the absence of MycN function resembles that of Shh mutants . Recent studies show that activation of GLI-dependent target genes is facilitated by JMJD3, a H3k27me3 demethylase involved in epigenetic conversion of inactive to active chromatin state , highlighting the importance of chromatin modification in enabling GCPs to respond to SHH during cerebellar development.
While Shh expression persists in PCs, GCPs eventually exit the cell cycle and differentiate to GCs. This process, as shown by recent studies, appears to be promoted by the transcriptional repressor BCL6 through recruitment of BCOR co-repressor and SIRT1 deacetylase to the Gli1 and Gli2 promoter regions, thus epigenetically silencing their expressions . Indeed, loss of BCL6 impedes the differentiation of GCPs, which however do not continue to proliferate, likely due to p53-mediated cell death. Removal of p53 rescues GCPs from cell death and restores their proliferation. However, it is unclear how BCL6 is activated in the immature GCs .
In addition to GCPs, SHH signaling is also required for the expansion of GABAergic INs by regulating precursors in two different neurogenic niches [33, 42]. In the VZ niche, SHH signaling is activated in multipotent radial glial cells after E12.5. Defective SHH signaling in VZ severely impairs the proliferation of radial glial cells and their ability to generate GABAergic interneuron progenitors during the embryonic period . Conversely, persistent activation of SHH signaling greatly expands their numbers. The source of SHH signal acting on VZ radial glial cells appears to be extracerebellar as SHH expression is not yet established in the emerging PC population. Indeed, SHH is prominently expressed in the hindbrain choroid plexus epithelium (hCPe), a secretory organ whose development is in close apposition with the cerebellar VZ (Fig. 10a). The presence of SHH protein in the circulating embryonic cerebrospinal fluid suggests that SHH is actively secreted from the hCPe and delivered to the adjacent VZ via a transventricular mechanism. Further support for this model comes from the observation that VZ progenitor proliferation is compromised in mice with reduced Shh expression in the hCPe .
The cerebellar VZ is also the source for stem-like astroglial cells of the secondary neurogenic niche residing in the PWM during late embryonic and postnatal period [33, 239, 242]. These astroglial cells transiently respond to PC-derived SHH and express cell surface marker CD133 (also referred to as Prominin) as well as extracellular matrix glycoprotein Tenascin-C (Tnc) . Lineage analysis reveals that the PWM astroglial cells generate intermediate progenitors of both astrocytes and GABAergic INs as marked by the expression of CD15 and PTF1a, respectively (Fig. 10b) . Furthermore, attenuation of SHH signaling in astroglia during the neonatal period leads to a significant reduction of both intermediate progenitor classes, underscoring the importance of SHH signaling in the maintenance of the PWM niche. Surprisingly, PTF1a-expressing GABAergic progenitors represent an additional population that responds to SHH in PWM [33, 238]. In contrast to cerebellar VZ , these PTF1a-expressing progenitors are proliferative and likely responsible for rapid expansion of the late-born GABAergic INs during the first week of the postnatal period .
Cerebellar Foliation (A. L. Joyner)
The most striking morphological feature of the cerebellum of birds, mammals, and some fish is its foliation pattern, or subdivision into lobes, lobules, and sublobules that are separated by a series of fissures [130, 171, 297, 298] (Fig. 11a). In most species, the foliation pattern is symmetrical with respect to the midline, and the fissures run perpendicular to the anterior-posterior (AP) axis in the medial cerebellum (vermis) of most species. The mammalian cerebellum is further subdivided into two lateral hemispheres and adjacent flocculi/paraflocculi, each with distinct foliation patterns with fissures that vary in their orientation. Nevertheless, the lobules of the hemispheres are continuous with lobules in the vermis (Fig. 11b). Larsell proposed a unified scheme for naming the lobules in the vermis of birds and mammals with roman numerals I–X from anterior to posterior [299, 300]. To account for the variation in foliation pattern between species and the fact that the cerebellum has more than ten lobules in many species, lobules were subdivided into sublobules separated by shallower fissures (e.g., VIa and VIb, Fig. 11c). The vermis of mice has eight or nine lobules, as lobules I/II are not separated in some strains and IV/V are fused, and the hemispheres have four lobules [301, 302] that extend laterally from lobules VI and VII (Fig. 11b, c). The basic pattern of vermis foliation is conserved throughout mammals, but foliation in the hemispheres is more variable than in the vermis and is very complex in primates (see discussion in Chapter 1 of , ). As the cerebellum modulates the functions of all areas of the neocortex , by extrapolation the development of the two brain regions should have co-evolved. Indeed, the entire spinocerebellar tract projects only to the medial cerebellum, thus the hemispheres are enriched for connections to the neocortex. Furthermore, although the volume of the cerebellum as a percentage of the total brain is constant across species [305, 306], the greatest proportional increase in brain regions has occurred in the cerebellum and neocortex , and the ratio of the number of neurons in the cerebellum to the neocortex is remarkably constant across mammalian species . It is therefore tempting to speculate that during evolution particular lobules in the hemispheres and folds (gyri) in the neocortex of gyrencephalic mammals that house interconnected neural circuits have arisen and expanded in unison . Defining the circuits between the cerebellum and neocortex in primates as well as rodents is a major challenge and high priority for future cerebellar research, but dependent on development of effective tools for tracing across multiple synapses. It will be exciting to trace the axon pathways from a parasagittal stripe of PCs in one region of a particular lobule through to the neocortex and back to the cerebellum. One hypothesis is that this will reveal an elaborate spatial organization of neurons, with ones dedicated to similar functions being housed in distinct lobules in the cerebellum and gyri in the neocortex, and with new circuits discovered specific to humans.
The cerebellum undergoes its major growth in the third trimester and infant stage in humans, and the first 2 weeks after birth in mice, primarily due to expansion of GCPs [130, 309, 310]. The surface area of the cerebellum increases during development much more than its volume due to the formation of lobules [311–314]. The lobules thus serve to house a large number of neurons in a layered cytoarchitecture in a small area. Foliation begins at E16.5 in the mouse with the sequential formation of the base of each fissure, which we have termed anchoring centers . The GCPs, PCs, and BG within anchoring centers have distinct characteristics, and the lobules grow out away from them [314, 315]. The first sign of formation of an anchoring center is an inward thickening of the EGL that is followed by formation of an indentation of the outer surface of the cerebellum and elongation of the bodies of the GCPs (Fig. 11d, ). The underlying PC layer indents and then the surrounding fibers of BG project to the base of the fissure. In the mouse vermis, four initial anchoring centers form that defines five initial lobes, which are further subdivided. Based on mutant analysis, the timing of formation and position of two adjacent anchoring centers define the morphology of the intervening lobule and thus the allocation of cells available for distinct long-range circuits [315–317]. Importantly, the homeobox engrailed genes (En1/2) are fundamental to the patterning process as they determine when particular anchoring centers form [315–320]. A number of theories have been proposed for how foliation is regulated [130, 315, 321–323]. Fundamental questions remain, such as whether one cell type initiates formation of anchoring centers and how are they positioned.
Clonal analysis previously uncovered that GCPs divide symmetrically to expand each clone and then differentiate en masse . Surprisingly, from a clonal analysis of GCPs, we recently found that the anchoring centers act as lineage restrictions that prevent GCPs from moving between lobules . One possibility is that the restriction in GCP movement produces a mechanical force driving the lobules outward. Furthermore, the number of cells per clone in long lobules is almost twice that in short ones. In terms of clone geometry, the length is greater in the AP than medial-lateral axis in all EGL clones, especially in long lobules, accounting for the tremendous AP expansion of the cerebellum (Fig. 11e). Moreover, in En1/2 mutants with smaller lobules, the size and geometry of clones is similar to wild-type clones in short lobules . Thus, the dynamics of GCP expansion is differentially regulated in lobules with different shapes/sizes. Whether this is a cell intrinsic property of GCPs that form different lobules that is established before anchoring centers are formed remains to be determined. What forces within and outside the cerebellum impact on the foliation process is an additional critical question to address. A further question is how scaling of all cell types in the cerebellum is regulated to ensure that the correct proportions of all neurons/glia are allocated to each lobule. One interesting possibility is that SHH secreted by PCs [285, 292] determines the expansion of INs and astrocyte/BG progenitors in unison with regulating GCP proliferation [33, 64, 285, 326].
The cerebellum arose in gnathostomes, but the emergence of a stable transient amplifying population of GCPs in an EGL structure that is stimulated by SHH seems to date to the transition to amniotes [326–332]. While the production of the enormous number of granule neurons (>50 % of all neurons in mouse and human [333, 334]) is thought to drive foliation  and has been experimentally linked to foliation, e.g., [64, 335], it is important to point out that some sharks have extensively foliated cerebella . Interestingly, the degree of foliation seems to correlate better with the ecological environment and/or complexity of prey behavior within and between clades rather than with phylogeny [306, 336]. An important question to resolve is whether the ratio of GCs to PCs is increased in sharks with highly foliated cerebella. Although it is not known whether SHH stimulates neurogenesis and of what cell types in sharks, it will be interesting to determine whether the SHH pathway has been co-opted to increase GC production in sharks since Shh has been detected in PCs of a shark but not in zebrafish . Furthermore, given the different cytoarchitecture of the shark cerebellum, the mechanisms underlying foliation could be distinct from that in amniotes. Determining the similarities and differences in the foliation processes between sharks and amniotes and their implications for circuit allocation will be valuable, and likely to have implications for formation of folds in the neocortex of gyrencephalic mammals.
Refinement of the Climbing Fiber Afferents (M. Kano, N. Uesaka)
PCs in the adult cerebellum receive two distinctive excitatory synaptic inputs—from PFs, the axons of GCs, and from CFs arising from the inferior olivary nuclei in the medulla oblongata. Each PC receives functionally weak but numerous (~100,000 in mice) PF synapses on spines of its distal dendrites. In contrast, most PCs are innervated by a single but functionally very strong CF on the stubby spines of their proximal dendrites. However, in the neonatal cerebellum, each PC is innervated on the soma by multiple CFs . How is the adult one to one relationship between a CF and a PC established during postnatal development?
Immature olivocerebellar axons extensively ramify in the white matter and the GC layer, and give rise to many collaterals around PCs (creeper stage) . Since immature PCs have no large primary dendrites, CFs terminate on perisomatic protrusions and thorns emerging from the PC somata. By P2–P3, several individual CFs form multiple synapses with relatively similar synaptic strengths on a single PC (Fig. 12). During the first postnatal week, a single CF is selectively strengthened on the soma of each PC (termed “functional differentiation”). Mice deficient in Cav2.1, the α-subunit of the P/Q-type voltage-dependent Ca2+ channel (VDCC), show impairment in the selective strengthening of a single CF, suggesting that activity-dependent Ca2+ influx through VDCCs is crucial for establishing a single “winner” CF on each PC [339, 340]. Next, the strongest CF extends its innervation territory from the soma to the dendrites (“CF translocation”: Fig. 12). As mentioned above, CFs initially establish synaptic contacts on the fine processes emerging from the soma and form a plexus (“pericellular nest” stage) . As the dendrites of the PCs start to grow into the ML, from around P6, multiple CFs continue to innervate the PC somata until P9. After the functional differentiation of CFs, only the “winner” CF extends its innervation territory from the soma to the stem dendrites from P9 (“capuchon” stage) . In the “dendritic” stage , CF synapses undergo progressive translocation to the growing PC dendrites. In contrast, the “loser” CFs remain around the soma and are eventually eliminated in two distinct phases (the “early and late phases of CF elimination”) mediated by distinct mechanisms [216, 339, 340]. The early phase of CF synapse elimination starts at around P7 soon after the functional differentiation is completed. Unlike the late phase of CF synapse elimination, the early phase is not dependent on the proper generation of GCs and PF-PC synapses. Several lines of evidence suggest that neuronal activity is crucial for this event [339, 340].
The late phase of CF synapse elimination starts at around P12 [216, 339, 340]. This process is critically dependent on the proper formation of excitatory PF synapses and inhibitory basket cell synapses on PCs. In mice deficient in mGluR1 or any of its downstream signaling molecules (Gαq, PLCβ4, PKCγ), the late phase of CF elimination is severely impaired. The immediate early gene Arc/Arg3.1, the neurotrophin receptor TrkB, and insulin-like growth factor 1 are also involved in CF synapse elimination [339, 340]. A recent study has revealed that postsynaptic Sema7A, a GPI linked subtype of Semaphorin, and its receptors (ItgB1 and PlxnC1) on CFs are involved in the cascade downstream of mGluR1 . In contrast, Sema3A, a secreted class of Semaphorin, and its receptors (PlxnA4) on CFs maintain both weak and strong CFs from P8 to P18, and therefore oppose synapse elimination . Thus, semaphorins mediate retrograde signals from PCs to CFs that regulate multiple processes of CF synapse elimination.
Dendritic Differentiation of Purkinje Cells (I. Dusart)
The PC stands as a neuronal model to study dendritic differentiation. In addition to being beautiful, PCs are also popular because of their convenience in terms of ease of immunohistochemical detection and genetic manipulation. Proteins such as calbindin or IP3R (inositol phosphate 3 receptor) are specifically abundantly expressed in PCs and label their dendritic tree, soma, and axon. Grafting experiments have provided major progress in our understanding of the biology of neurons . The L7/pcp2 promoter drives gene expression specifically in PCs and retinal rod bipolar neurons . Combining the L7 promoter and an inducible CRE/loxP system with in utero electroporation allows the specific regulation of gene expression in PCs in a temporally controlled manner . Furthermore, different viral vectors target PCs specifically  for review. Last but not least, for the study of the development of the dendritic tree, PCs can develop in organotypic cultures. By using this technique, the morphology of individual PCs can be studied since they are isolated from their neighbors [346, 347].
Adult PCs are highly recognizable by their large dendritic tree with prototypical morphological characteristics (Fig. 13a). One characteristic of the PC dendritic tree is that its extension and ramification occurs in the sagittal plane, resembling an espaliered fruit tree . This highly stereotyped and simple architecture in a two-dimensional plane is likely at the origin of the PC’s popularity: many neurobiologists, among them Ferdinando Rossi, have been fascinated by their beauty, e.g., . The development of this spectacular dendritic tree occurs during the first three postnatal weeks of the mouse life. Interestingly, it is not a linear process as there is a clear discontinuity at the end of the first postnatal week. During the first postnatal week, successions of growth and retraction have been described [349, 350]. Immature PCs present a panel of very different morphological forms (Fig. 13b,c). In the absence of time-lapse analyses, the relations between these different morphological forms are not yet well understood. It is only from the second postnatal week on that PCs develop their characteristic dendritic trees (Fig. 13d). At the beginning of the second postnatal week, the PCs have a single stem segment at the apical pole that already presages the form of the mature dendritic tree. During the second postnatal week, and up to the end of the third postnatal week, the dendritic tree grows first wider and then taller . By using virus-mediated gene transfer followed by three-dimensional reconstruction of confocal images of labeled PCs, Kaneko et al.  demonstrated that PCs achieve their monoplanar configuration by the dynamic remodeling of an initially irregular arrangement extended in multiple sagittal planes during the third postnatal week in mice.
The transition between these two morphological developmental phases—a first phase of intense remodeling and a second phase of continuous development of the mature dendritic tree—occurs in parallel with profound functional transitions [353, 354]. As they occur in parallel with a circulating peak of TH and the acquisition of the ability to walk outside the mother nest, we have proposed that these transitions are reminiscent of amphibian metamorphosis [354, 355].
Numerous intrinsic or environmental factors regulate the dendritic development of PCs (for reviews see [342, 345, 356, 357]) and indeed, understanding the development and the maintenance of the dendritic tree is far from being accomplished. Recent studies of PCs have shed light on new mechanisms. For example, the PC is one of the rare types of neurons in which factors involved in the maintenance of dendritic tree have been identified . In mammals, mechanisms of dendritic self-avoidance, a critical process in patterning neural circuits during development, has also been reported in [359, 360]. Last, in parallel to the classic neurotrophic theory for axons, developing PC dendrites compete for limiting amounts of Neurotrophin-3 (NT3) and require anterograde NT3 from their presynaptic partners in order to grow .
Neurodevelopmental Disorders of the Cerebellum
Developmental Malformations (W.B. Dobyns, P. Haldipur, K. J. Millen)
Numerous cerebellar malformations have been described in humans, primarily classified by MRI studies, and can occur in isolation or as part of a broader malformation syndrome involving multiple systems. Most cause cognitive in addition to motor and sensory integration deficits [362, 363]. Cerebellar developmental mechanisms are well conserved between humans and rodents, making studies in mice highly informative towards defining pathogenic mechanisms. Notably, however, cerebellar development in humans begins around the ninth gestational week and continues beyond birth. This protracted developmental timeline makes the human cerebellum particularly vulnerable to insult, especially during 24–40 weeks of gestation, when massive neurogenesis in the EGL causes a fivefold increase in size of the cerebellum . Thus, while several malformations have a genetic basis, inflammation, fetal hemorrhage, and prematurity are often contributing factors. Here, we discuss some of the common and best understood human cerebellar malformations and their causes.
Dandy Walker Malformation (DWM) is the most common human cerebellar malformation with an estimated incidence of 1/3000 live births [365, 366]. DWM is an imaging diagnosis characterized by an enlarged posterior fossa, cerebellar vermis hypoplasia, and an enlarged fourth ventricle (Fig. 14a). DWM can occur in association with agenesis of the corpus callosum, but more often occurs as an isolated finding on MRI scans. DWM clinical features are variable. Patients may exhibit symptoms ranging from intellectual disability to autism or they may be completely unaware of any deficits until diagnosed as adults for unrelated reasons [367–369]. The genetic causes of DWM remain largely unknown. However, recent studies indicate that deletions in FOXC1 and ZIC1/4 are responsible for a small subset of DWM cases [370, 371]. Prenatal cerebellar hemorrhage however can also cause DWM , which may also be associated with genetic risk factors; however, these have yet to be determined. Research in animal models has led to the hypothesis that disruptions of posterior fossa signaling from the mesenchyme surrounding the brain to the underlying embryonic cerebellum are key. Signaling disruptions cause dramatic reductions in cerebellar anlage neuronal progenitor proliferation, as well as abnormal migration of both RL- and VZ-derived cells. This ultimately leads to foliation and lamination defects [370, 373].
Joubert syndrome and related disorders (JSRD) is a group of disorders with an incidence of 1 in 80,000–100,000 live births [374, 375]. JSRD is characterized by cerebellar vermis hypoplasia, thick and abnormally oriented superior cerebellar peduncles, and a deep interpeduncular fossa, all of which give it the pathognomonic Molar Tooth Sign (MTS) seen in axial brain scan images (, Fig. 14d). Patients with JSRD exhibit variable neurological symptoms such as ataxia, developmental delay, abnormal eye movements, and altered breathing patterns. To date, ~23 genes have been identified as causative for JSRD [377, 378]. Most have been linked to the primary cilia and its function, bringing JSRD under the umbrella of a highly heterogeneous group of disorders called ciliopathies. Studies in animal models as well as human fetal tissue from JSRD patients indicate reduced GC proliferation suggesting impaired SHH signaling ([290, 291, 379]: see “Extrinsic Regulators of Cerebellar Development: The Role of Sonic Hedgehog” section). Additionally, the primary cilia also plays a role in the mediation of signaling pathways involving WNT and platelet-derived growth factor which can impact cerebellar anlage fusion earlier in fetal development [380, 381].
Cerebellar hypoplasia (CH) refers to underdevelopment of the cerebellum. This category of cerebellar malformation is distinct from DWM, as it does not involve a concomitant enlargement of the posterior fossa. CH is also an extremely heterogeneous group of disorders, and often, other CNS abnormalities are observed, including lissencephaly, microcephaly, and cortical heterotopia. CH may be unilateral, global, vermian, or ponto-cerebellar, where in addition to the cerebellum, the volume of the pons is also reduced likely reflecting the common developmental origin of GCs and pontine nuclei neurons in the cerebellar RL ([20, 382], Fig. 14b, e). In contrast to DWM, almost all individuals exhibit cognitive and motor impairments. Several genes have been associated with CH including mutations in CASK, DAB1, OPHN1, RELN, CHD7, several tubulin genes, and several TSEN genes [209, 383–392]. Each causes developmental defects in a multitude of cerebellar developmental programs, including progenitor proliferation and neuronal migration and even developmental cell survival. Notably, CH can also occur due to a variety of non-genetic causes such as perinatal cytomegalovirus infection and perinatal exposure to alcohol and drugs such as cocaine [393–397].
Cerebellar agenesis is an extremely rare anomaly distinguished by a complete or near-complete absence of the cerebellum (; Fig. 14c). Individuals show a number of neurological deficits particularly related to movement and speech, but can be otherwise surprisingly unaffected . Homozygous mutations in PTF1A have been associated cerebellar agenesis in humans . In mice, Ptf1a is required for the generation of all VZ-derived GABAergic cerebellar neurons. Failure to generate these neurons means that RL-derived cells have no trophic support and these too are therefore lost, resulting in cerebellar agenesis in neonates . Fetal hemorrhages that completely disrupt the early cerebellar anlage have also been predicted to cause cerebellar agenesis .
Recent developments in neuropathology and neuroimaging have tremendously improved the diagnosis of developmental disorders of the cerebellum. Several genes responsible for these heterogeneous malformations have been identified and animal models have revealed novel developmental mechanisms of interest to both clinical and basic science. A deeper appreciation of the cellular and signaling mechanisms responsible for these malformations will enable improved diagnosis and potential treatment of these disorders.
The Role of Thyroid Hormone in Cerebellar Development (N. Koibuchi)
The importance of T3 and T4 in brain development has been well documented [402, 403]. Deficiency of TH during fetal and early postnatal period results in severe mental retardation, known as cretinism in humans . Since there is a distinct “critical period” of TH action in brain development, replacement of TH should be started as early as possible to prevent irreversible neurological disorders.
T4 enters the brain through the blood–brain barrier (BBB) more easily than T3, an active form of TH . After crossing the BBB, T4 is taken up by astrocytes and deiodinated to produce T3 by type 2 iodothyronine deiodinase . T3 is then transferred to neurons or oligodendrocytes, possibly via monocarboxylate transporter 8 . TH effects are mainly exerted through the nuclear TH receptor (TR; TRα1, TRβ1, and TRβ2), a ligand-dependent transcription factor . TR binds to specific DNA enhancer sequences known as the TH-response elements located in the promoter region of target genes . The rodent cerebellum is a good model to investigate the TH action. This is partly because the rodent cerebellar development occurs largely postnatally, allowing cerebellar TH status to be precisely manipulated at various developmental stages .
Perinatal hypothyroidism dramatically affects morphogenesis [402, 403]: the growth, dendrite arborization, and dendrite spine formation of PCs are all markedly decreased; synaptogenesis between PCs and PFs is dramatically repressed; the disappearance of the EGL is postponed as a result of the delayed proliferation and migration of GCs into the GL . TRs are expressed in the most subsets of cells in the developing cerebellum . TRα1 is abundant in GCs, whereas TRβ1 is mainly expressed in PCs. The expression of many cerebellar genes is altered by perinatal hypothyroidism . Representative TH-responsive genes in the cerebellum include neurotrophins such as nerve growth factor, BDNF, NT3, and NT-4/5, and receptors such as the inositol trisphosphate 3 receptor, and retinoic acid receptor-related orphan receptor α, hairless, and myelin basic protein [411, 412]. The expression of many of these genes is regulated by TH only during a limited period of development. (It should be noted, however, that although TH action in the brain is greater during development, TR levels are greater in the adult brain: .) Thus, TH sensitivity may be controlled by other unknown epigenetic mechanisms such as DNA methylation and histone modification.
Various animal models have been used to study TH in cerebellar development . Interestingly, TRα knock-out mice, TRβ knock-out mice and TRα/TRβ double knock-out mice do not display obvious cerebellar defects, suggesting that most of the consequences of congenital hypothyroidism in the brain are due to the detrimental activity of unliganded TR. This hypothesis is supported by studies of transgenic animals expressing mutant TR, which show severe neurodevelopmental defects [415, 416].
Although these animal models have contributed greatly to our understanding on the role of TR in cerebellar development, these may not sufficiently address the mechanisms of direct TH action. Since TH acts not only in the brain but also in the peripheral organs, brain development may be affected by peripheral metabolic changes. Thus, cell or organ-specific inhibition/activation of TH action is required. For such purpose, Fauquier et al.  used a L7/Pcp2 promoter to generate transgenic mice that express a mutant TRα1 specifically in PCs after P8. Probably because the timing of transgene expression is slightly after the critical period, this mouse showed only limited alterations in cerebellar morphogenesis. On the other hand, by Ptf1a-Cre recombination, mutant TRα1 was expressed in PCs and GABAergic INs from prenatal stages, showing the alteration of PC morphogenesis . We have also generated a transgenic mouse using mutant human TRβ1 with L7/Pcp2 promoter (Fig. 15; ). Expression of mutant TRβ1 was observed as early as P2. This mouse showed decreased PC dendritic arborization and lower levels of expression of TH-regulated genes in PCs. To our surprise, GC migration was also retarded and the expression of TH-regulated genes in GCs and oligodendrocytes was also decreased. As a possible consequence, this mouse shows cerebellar ataxia. These studies indicate that TH may mainly act through TR in the PC to regulate the whole cerebellar development. Additional factors that transmit TH signaling in PCs to other subsets of cells are required. Furthermore, although it is usually considered that the critical period for TH action in the rodent cerebellum is the first two postnatal weeks, the actual critical period may be earlier. Disruption of TH action by environmental factors during the critical period may produce adverse effects .
Abnormal Purkinje Cell Development and Cerebellar Ataxia (E.B.E. Becker)
The cerebellar ataxias comprise a heterogeneous group of neurological disorders characterized by gait disturbances, motor incoordination and imbalance, dysarthria, and oculomotor deficits [420, 421]. The etiology of cerebellar ataxia is complex and includes acquired causes as well as a steadily growing number of inherited conditions [421–423]. The genetic ataxias are usually progressive. For many of these disorders, pathologic changes in PCs and a substantial loss of these neurons resulting in cerebellar atrophy are thought to cause the symptoms of the disease. However, accumulating evidence from cell- and animal-based models of cerebellar ataxia suggest that abnormal PC development and related early changes in PC physiology might contribute to the disease, thus challenging our view of cerebellar ataxias as pure neurodegenerative disorders. Here, I briefly review the emerging concept that PC developmental abnormalities might be contributing to disease pathogenesis in cerebellar ataxia.
Spinocerebellar ataxia type 1 (SCA1) is caused by a CAG repeat expansion in the Ataxin1 (ATXN1) gene and is one of the most intensely studied dominant ataxias. Numerous mouse and other animal models have been generated for SCA1 that recapitulate different aspects of the human disease. For example, both transgenic mice overexpressing expanded ATXN1[82Q] as well as knock-in mice (SCA1
154Q/2Q) exhibit motor impairments and PC degeneration . Interestingly, transgenic mice in which expression of the expanded transgene is delayed until well after the cerebellum has matured display a much reduced disease phenotype, suggesting that mutant ATXN1 interacts with a pathway involved in PC development . Indeed, the same study demonstrated a key interaction of ATXN1 with retinoic acid-related orphan nuclear receptor α (RORα), a transcription factor critical for cerebellar development. Moreover, RORα expression levels were found to be reduced in the ATXN1[82Q] model . Taken together, the results of this landmark study provided the first functional genetic evidence that compromising PC development contributes to the severity of neurodegeneration. Subsequent studies have shown that SCA1 transgenic mice display abnormalities in PC development, including a reduction of CF translocation along the developing dendritic tree and decreased pruning of CF terminals from the PC soma [426, 427]. Similarly, profound impairments in PC dendritogenesis, spine development, and synaptogenesis have been described in the staggerer mouse [428, 429], which harbors an autosomal recessive mutation in the Rora gene encoding RORα, and is viewed by some as an extreme model of SCA1 .
Other studies have demonstrated similar PC developmental abnormalities in different mouse models of degenerative cerebellar ataxias. A PC-specific transgenic mouse model that expresses a truncated form of expanded human Ataxin-3, the disease protein causing SCA3, displays disarrangement of PCs and poor PC dendritic arborization [430, 431]. As described above for SCA1, the PCs in this SCA3 model express decreased levels of the developmental transcription factor RORα , suggesting a potential molecular link between the observed developmental abnormalities in both models.
Impaired dendritic arborization of PCs as well as abnormal spine morphogenesis have also been described in mice lacking β-III spectrin, a model of SCA5 . Similarly, cultured PCs overexpressing PKCγ with SCA14-causing mutations after adenoviral infection display a decreased dendritic arbor as well as decreased spine density . In vivo lentiviral-mediated expression of mutant PKCγ in PCs led to impaired pruning of CF synapses from developing PCs, although no dendritic abnormalities were observed . However, the recently reported transgenic SCA14 mouse model shows abnormal dendritic development of PCs both in vivo as well as in organotypic slice cultures .
Besides models of the human SCAs, several other genetic mouse mutants have highlighted the relationship between abnormal PC development and ataxia. For example, the ataxic Moonwalker (Mwk) mouse harbors a dominant gain-of-function mutation in the TRPC3 ion channel, resulting in adult-onset PC loss but also impairments in PC dendritic arborization during cerebellar development . TRPC3 is a key player in the mGluR1 signaling pathway vital for PC function [437, 438]. Interestingly, impaired mGluR1 signaling has been demonstrated in a number of the mouse models described above including SCA1 [439, 440], SCA3 , and SCA5  models. Future research should help to clarify whether there is a causal relationship between impaired mGluR1 signaling at PC-PF synapses and developmental PC abnormalities in these and other models of ataxia.
In summary, PC developmental abnormalities are clearly evident in a wide range of ataxic mouse mutants including models of the degenerative SCAs. The observed PC developmental defects commonly include impaired dendritic arborization, resulting in synaptic deficits affecting CF and PF connections and ultimately altering PC physiology. It will be important to better understand the underlying—likely common—molecular mechanisms by which mutations in distinct genes cause abnormal PC development and function. These could offer attractive future therapeutic targets to alleviate motor dysfunction in cerebellar ataxia.
Deregulated Developmental Pathways in Medulloblastoma (S. Marino, T.O. Millner)
MB is the most common pediatric brain tumor and is the most common cause of pediatric death from cancer. Histologically, cases are classified into classic, nodular/desmoplastic, and large cell/anaplastic subtype , and prognosis is performed by combining histological subtype, clinical markers, namely, age, metastatic stage, and level of resection, as well as selected molecular markers. Morphologically, MB cells closely resemble GCs and GCPs; hence, it has been long postulated that a link exists between these tumors and the normal development of the cerebellum. Evidence gained from candidate gene approaches in mouse models and more recently “-omics” screening of large tumor series has shown deregulation of specific developmental pathways in subgroups of these tumors.
The current consensus is that MB can be sub-classified based on genetic, epigenetic, and transcriptomic characteristics [444, 445] into four distinct subgroups (Fig. 16): WNT, SHH, Group 3, and Group 4 . The WNT and SHH subgroups have been associated with constitutive activation of the WNT/β-catenin and SHH pathways, respectively, whereas Group 3 and Group 4 MBs are less well characterized. Each of these subgroups have defining demographic, clinical, genetic, and epigenetic profiles, and emerging evidence links their origin to different cerebellar progenitor cells at different developmental time points. Here, we will summarize the key features of the current molecular stratification of MB from a developmental oncobiology perspective.
GCPs and SHH-MB
GCPs are the main cell of origin of SHH-MB, as shown in mouse models in which Ptch1 is conditionally inactivated in GCPs. Importantly, constitutive activation of SHH signaling induces neoplastic transformation of more undifferentiated progenitor cells only upon commitment toward a GC lineage . Pre-neoplastic lesions expressed Atho1, a marker of the GC lineage, and showed activation of Gli1, Cyclin D1, and MycN, which are SHH target genes. Their gene expression profiles were more similar (differing by 34 genes) to tumor cells than GCPs (differing by 75 genes) . SHH-MBs have also been shown to originate from cells located in the cochlear nuclei of the brainstem .
MRI studies have shown differences in the location of MB subgroups, and MB location (as well as enhancement pattern) can predict the molecular subgroup of pediatric MBs. SHH tumors are mainly detected within the cerebellar hemispheres [449–451] consistent with a GCP cell of origin. Moreover, in a series of 63 human MBs that were morphologically nodular/desmoplastic, a histological subtype likely to be typical of the SHH subgroup , 33 % had very close contact to the cochlear nuclei on MRI . Human Shh MBs  have a 1:1 male-to-female ratio and a bimodal age distribution (very frequently seen in infants and adults), with a good prognosis in infants but an intermediate prognosis in other age groups. All histological nodular/desmoplastic MBs are likely SHH-MBs , but 50 % of SHH-MBs are of other morphology.
Embryonic Dorsal Brain Stem Precursors and WNT-MB
Mouse models have shown that WNT-MBs, characterized by activating mutations in the Wnt pathway effector CTNNB1, arise from cells outside the cerebellum, in the embryonic dorsal brainstem . These studies also showed that the genes characterizing human WNT-MBs are more often expressed in the lower RL and embryonic dorsal brainstem than in the upper RL of the developing cerebellum. In addition, transcriptome analysis showed that the MBs arising in these mice matched human WNT-MBs.
MRI studies in patients have shown that WNT tumors are often found within the fourth ventricle (cerebellar peduncle/cerebellopontine angle cistern) and infiltrated the dorsal brainstem [449, 450], with the majority of them being continuous with the cuneate nucleus , a region that corresponds to the origin defined for the murine WNT-MBs described above. Human WNT-MBs  have a 1:1 male-to-female ratio and occur at all ages (uncommon in infants). They have a very good long-term prognosis in comparison to the other subgroups of MB (survival rate likely exceeds 90 % with current treatment). The large majority of WNT-MBs investigated so far have classic histology.
Cerebellar Stem Cells and Their Role as MB Cell of Origin
Cerebellar stem cells are a third likely cell of origin for human MBs. Models using Rb
Tp53 mutant mice were generated where MBs developed from neural stem cells in the cerebellar white matter [453, 454]. These tumors resembled human Group 3 MBs histologically while expressing high levels of neural stem cell markers (Nestin, Sox2, and Sox9). A further model using Myc
Tp53 mutant mice developed MBs from cerebellar neural stem cells and from GCPs, although the tumors that formed from GCPs lost their lineage specific markers first . This model expressed high levels of neural stem cell genes and also resembled human Group 3 tumors at a histological and molecular level.
In the MRI studies to date, Group 3 and Group 4 MBs are characterized by brainstem contact with most of the tumors growing within the vermis. Most also contact both the cochlear and cuneate nuclei and always infiltrate the fourth ventricle [450, 451]. Group 3 MBs  are seen more frequently in males than females and almost never in adults, have a poor prognosis, and are frequently metastatic. Group 3 MBs most frequently have a classic morphology, but the Group 3 subgroup contains the majority of the large cell/anaplastic tumors.
Group 4 comprises 30 % of MBs and it is the least well characterized molecularly. At present, it is also unclear from which cells these tumors originate. This subgroup has a male-to-female ratio of 2:1 and has an intermediate prognosis, similar to the SHH subgroup . Group 3 and Group 4 MBs share some molecular features: amplification of the OTX2 oncogene, which is not seen in the other subgroups, and isochromosome 17q (26 % in Group 3 and 66 % in Group 4 MBs ). However, there are also important differences. Group 3 MBs show high levels of MYC, and often gain of chromosome 1q and/or loss of chromosome 5q and 10q, whereas Group 4 MBs have low levels of MYC and MYCN and loss of the X chromosome in 80 % of females within the subgroup .
In summary, “-omics” datasets on large series of MB combined with the results of ontogeny studies performed in mouse models aiming at characterizing the cell of origin of the various subgroups together with magnetic resonance imaging (MRI) studies in patients lend additional support to the notion that MB is in fact a disease arising from deregulated cellular and molecular mechanisms involved in the development and homeostasis of the cerebellum.